Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Residential schooling and sources of Aboriginal disparity in Canada Feir, Donna Leanne 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2013_fall_feir_donna.pdf [ 1.6MB ]
Metadata
JSON: 24-1.0165638.json
JSON-LD: 24-1.0165638-ld.json
RDF/XML (Pretty): 24-1.0165638-rdf.xml
RDF/JSON: 24-1.0165638-rdf.json
Turtle: 24-1.0165638-turtle.txt
N-Triples: 24-1.0165638-rdf-ntriples.txt
Original Record: 24-1.0165638-source.json
Full Text
24-1.0165638-fulltext.txt
Citation
24-1.0165638.ris

Full Text

Residential Schooling and Sources ofAboriginal Disparity in CanadabyDonna Leanne FeirB.A., The University of Calgary, 2006M.A., The University of Calgary, 2007A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Economics)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)October 2013c? Donna Leanne Feir 2013AbstractIndigenous peoples throughout the world live in more difficult socio-economic circumstancesthan their non-indigenous counterparts. This dissertation investigates the long run economicand cultural consequences of one of the most infamous policies to affect Indigenous peoples:the forcible removal of children from their homes and their placement in boarding schools(also known as residential schools). These sorts of policies were instituted in numerouscountries throughout the world, including the United States, Canada, and Australia, andthey have been heavily criticized. The policies have often had the stated goal of culturalassimilation, are generally perceived to have been educational failures and to have harmedIndigenous peoples both directly and intergenerationally. I investigate these possibilitiesusing several sources of variation. I find that attendance at a residential school is associatedwith both economic and cultural assimilation but I do not find strong evidence that thisis transmitted intergenerationally. The results suggest that while residential schooling hadsignificant impacts, it is likely not the predominate cause of the economic disparity observedtoday. To get a sense of plausible alternative explanations of Indigenous disparity in theCanada context, I finish by establishing some of the basic patterns in earnings differencesbetween Indigenous groups and their non-indigenous counterparts.iiPrefaceThis dissertation is original work by Donna Leanne Feir.Another version of chapter 5 has been accepted and is forthcoming in Canadian PublicPolicy under the title Size, Structure, and Change: Exploring the Sources of AboriginalEarnings Gaps in 1995 and 2005, Volume 39, Number 2. June 2013.Chapter 3 has been approved by the UBC Human Ethics Research Board under projectnumber H11-03492 under the name The Impact of Indian Residential Schools on FirstNations Peoples and Their Communities: A Statistical Analysis.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 A History of Residential Schools . . . . . . . . . . . . . . . . . . . . . . . . 62.1 The Intentions of the System and School Location . . . . . . . . . . . . . . 142.2 Previous Literature on the Effects of Residential School . . . . . . . . . . . 162.3 The Environments in Residential Schools and Their Alternatives . . . . . . 182.4 The Ultimate Closure of the System . . . . . . . . . . . . . . . . . . . . . . 222.5 Public Revelation of Abuse . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.6 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 The Long Term Effects of Cultural Separation and Assimilative Educationon Skill Accumulation: The Case of Indian Boarding Schools . . . . . . 353.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.2 Brief Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.3 Data Source, Basic Patterns and the Identification Problem . . . . . . . . . 463.4 The Empirical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51ivTable of Contents3.4.1 Historical Plausibility . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5 Additional Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.6.1 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.6.2 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 763.8 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 774 The Intergenerational Legacy of Residential School . . . . . . . . . . . . . 894.1 Data and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . 914.2 Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1004.3.1 Accounting for Grandparents Attendance . . . . . . . . . . . . . . . 1054.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1074.5 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1095 The Size, Structure and Change in Aboriginal Earnings Gaps in 1995 and2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1225.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1225.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1255.3 Methodology and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1305.3.1 Methodology: Decomposing Differences in Earnings into Differencesin Weeks Worked and Weekly Earnings . . . . . . . . . . . . . . . . 1305.3.2 Methodology: Decomposing Weekly Earnings in Differences in Char-acteristics and Differences in the Wage Structure By Year . . . . . . 1315.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1345.4 Sensitivity of the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1385.4.1 Censoring and the Proportion On-Reserve . . . . . . . . . . . . . . . 1385.4.2 Adjusting For Legal Changes, Ethnic Drift and Other Factors . . . . 1395.5 The Role of Taxes and Transfers . . . . . . . . . . . . . . . . . . . . . . . . 141vTable of Contents5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1435.7 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1466 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1546.1 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158AppendicesA More Detail On The Decisions of Aboriginal Families . . . . . . . . . . . 176B Appendix Tables For Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . 181C The Log-Likelihood Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 185D Mapping Individuals to Communities . . . . . . . . . . . . . . . . . . . . . 187E Appendix Tables For Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . 189F Appendix Tables For Chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . . . 190viList of Tables2.1 Number and Religion of Total Residential Schools that Ever Existed . . . . . 342.2 School Schedule (Gresko 1986, 93) . . . . . . . . . . . . . . . . . . . . . . . . 343.1 Descriptive Statistics By Residential School Attendance . . . . . . . . . . . . 773.2 What Determines Whether a School is Open When an Individual is of School-ing Age? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783.3 Residential School Attendance and High School Graduation: Coefficient Es-timates from the Bivariate Probit Model . . . . . . . . . . . . . . . . . . . . 793.4 The Impact of Residential School on Economic Outcomes . . . . . . . . . . . 803.5 Estimates of the Intent to Treat From the 1991 Census . . . . . . . . . . . . 813.6 Estimates of the Intent to Treat For Those Who Are Ineligible: The Effect ofHaving the Closest School Open . . . . . . . . . . . . . . . . . . . . . . . . . 823.7 On and Off Reserve: Bivariate Probit Results . . . . . . . . . . . . . . . . . 833.8 Effect on High School Graduation Conditional on Obtaining At Least Grade10: A Lower Bound on the Impact of Residential School on High SchoolGraduation Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 843.9 Being Away from Home: The Effect of Residential School . . . . . . . . . . . 853.10 Does the Effect Vary by Era of Attendance? . . . . . . . . . . . . . . . . . . 863.11 Descriptive Statistics Abuse . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.12 Heterogeneity in the Effect of Residential Schooling: Religion and Abuse . . 884.1 Descriptive Statistics: APS 2001 . . . . . . . . . . . . . . . . . . . . . . . . . 1104.2 First Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1114.3 The Effect of Parental Residential School Attendance: Economic Outcomes . 1124.4 The Effect of Parental Residential School Attendance: Social Outcomes . . . 113viiList of Tables4.5 The Effect of Residential School Attendance By Parental Attendance: Eco-nomic Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1154.6 The Effect of Residential School Attendance By Parental Attendance: SocialOutcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1164.7 Intergenerational Persistence? The Effect of Parental Residential School At-tendance on Own Attendance . . . . . . . . . . . . . . . . . . . . . . . . . . 1174.8 Correlation of Siblings Residential School Attendance With Support Factors 1184.9 The Intergenerational Effects of Residential School Conditional On Grand-parent's Attendance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1194.10 The Intergenerational Effects of Residential School Conditional On Grand-parent's Attendance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1205.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1465.2 The Log Earnings Gap, the Log Weekly Earnings Gap and the Oaxaca De-composition of Log Weekly Earnings . . . . . . . . . . . . . . . . . . . . . . 1485.3 The Log Annual and Weekly Earnings Gaps with a Constant Fraction thathave Positive Earnings/On-Reserve Over Time . . . . . . . . . . . . . . . . . 1505.4 The Log Earnings and Log Weekly Earnings Gaps Adjusted to Have the SameComposition as in 1995 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1525.5 The Role of Taxes and Transfers in Reducing Earnings Differences: 2005 . . 1536.1 Comparing the Effect of Residential School in the APS 1991, APS 2001 andthe APS 2001 Limited To Only Those Who Knew Whether Their CousinAttended Residential School . . . . . . . . . . . . . . . . . . . . . . . . . . . 157B.1 Linear Probability Model First Stage Results: The Effect of the Instrumentson Residential School Attendance . . . . . . . . . . . . . . . . . . . . . . . . 182B.2 The Impact Residential School on Economic Outcomes: Province Fixed Ef-fects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183B.3 The Effect of Residential School Attendance on Sociological Outcomes . . . 184viiiList of TablesE.1 The Correlation of an Indicator for Missing Parental or Own ResidentialSchool Attendance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189F.1 Results of the Oaxaca Decomposition for Log Weekly Earnings using 1995Years of Education and Potential Experience . . . . . . . . . . . . . . . . . . 191F.2 Results of the Detailed Oaxaca Decomposition for Log Weekly Earnings . . . 193ixList of Figures2.1 Location of Residential Schools in 1930 and Aboriginal Settlements . . . . . 302.2 Opening Of Residential Schools Across the Country . . . . . . . . . . . . . . 312.3 The Percentage of Enrollment and Attendance Accounted for By ResidentialSchools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.4 Composition of Enrollment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.5 Cumulative Probability of School Opening and Closing . . . . . . . . . . . . 33xAcknowledgmentsI would like to thank Nicole Fortin and Siwan Anderson for their for invaluable guidanceand Gordon Dahl, Mukesh Eswaran, Catherine Douglas, Nancy Gallini, Morley Gunderson,Vadim Marmer, Kevin Milligan, Anji Redish, Marit Rehavi, Craig Riddell, and all the par-ticipants in the UBC Empirical Workshop and CEAs for their for insightful comments andvaluable discussions. The excellent research assistance of Ethel Newton and Jiayun Yao (An-gela) was also greatly appreciated. I would also like to thank my classmates, and husbandfor making these years some of the best of my life. Finally, I wanted to thank David Green.Words cannot express what your guidance, patience and encouragement has meant.xiChapter 1IntroductionThere are hundreds of millions of Indigenous people throughout the world (Coates 2004)and while there is substantial heterogeneity, they often live in more challenging economicand social circumstances than non-indigenous people. In the United States, for example,American Indians on reservation make 39 percent less than the average American (Cornelland Kalt 2010), while in Canada registered Indians make 45 percent less than average (Cooke,Beavon and McHardy 2004). Similar disparities exists in Australia (Altman, Biddle andHunter 2008), New Zealand (Maani 2004), Latin America (Hall and Patrinos 2006), and inthe Arctic and Northern Europe (Andersen, Kruse and Poppel 2002). Yet, there is littlework in the mainstream economics literature addressing the causes of this disparity.1Thisdissertation makes a contribution to this small but growing research agenda in two ways.The first is by examining the long run consequences of a policy often attributed with largenegative economic consequences for Indigenous people across the world - specifically, theforcible removal of children from their homes and their placement in boarding schools. InCanada, these boarding schools are known as Indian Residential Schools. The second waythis dissertation contributes is by establishing some of the basic facts about Indigenousdisparity in Canada and how it has changed over time.For decades, numerous governments throughout the world implemented residential school-1The literature that exists includes the work by Cornell and Kalt (2000) who examineinstitutional formation, legitimacy, and economic growth; Evans and Topoleski (2002) whostudy the long run effects of Casinos on Indian Reservations; and Dippel (2012) who examinesthe long run consequences of forced integration of American Indians on Indian reservations.Sunde, Jorgensen, and Akee (2012) address the effect of parliamentary versus presidentialsystems on economic development while Akee (2009) argues that high transaction costs arecould be a significant source of the lack of development on Indian reservations. Kuhn andSweetman (2002) frame Aboriginal people in Canada as unwilling immigrants and arguethat measures of social integration are linked with economic integration.1Chapter 1. Introductioning policies. These policies aimed not only to educate Indigenous children, but to immersethem in a European way of life (Miller 1996; Milloy 1999; Smith 2009; Glenn 2011; Dawson2012). Even after the official stance on the negative influence of the Aboriginal home wasabandoned, these institutions still played an important role in government education andwelfare policy (Glenn 2011). In Canada, the United States and Australia these policies haveresulted in apologies from federal governments (Rudd 2008; Harper 2008; Gover 2000) andcalls for compensation (BSHP 2008-2011; Cassidy 2009). Indian residential schools in Canadaand child removal policies in Australia have also been subject to scathing public inquiries(RCAP 1991, Commonwealth of Australia 1997). In Canada, residential schools have beencalled the most damaging of the many elements of Canada's colonization, (Milloy 1999,xiv) and have resulted in the largest class action settlement in Canadian history (Reimer2010). However, despite the wide spread nature of these institutions and the fact thereare numerous individual accounts of the negative impact of residential schools (Haig-Brown1991; Fournier and Crey 1997), there currently does not exist any causal statistical researchon their long run and intergenerational consequences (Nuu-Chah-Nulth Tribal Council 1996;The Cariboo Tribal Council 1991; Commonwealth of Australia 1997).2The first three chapters of this dissertation are dedicated to addressing this research gapby leveraging the unique data and circumstances in the Canadian case in order to makebroader conclusions about the effects of residential schooling. Chapter 2 briefly discussesthe history of residential schooling focusing on the details that will be important in laterchapters. Chapter 3 examines the effect of residential schooling on a set of labour market2There are a handful of statistical studies that do not address the issue of selection. Forexample, in a study of their members, the Cariboo Tribal Council finds significant negativeimpacts on mental health from attendance at residential schools and abuse, but not onfinal educational outcomes or employment status (1991). Similar findings were reported bythe Nuu-Chah-Nulth Tribal Council (1996). The Australian Bringing Them Home Report(1997) on the impact of removal of Aboriginal children from their homes found that therewas no improvement in schooling or employment for the group taken away from their homesrelative to the group that remained. In addition, those taken from their homes were threetimes as likely to have acquired a police record and were twice as likely to use illicit drugs.However, they also earned higher incomes due to a greater degree of urbanization and thusgreater access to welfare payments. Not all literature has focused on the negative impacts ofresidential schools. Recent work has made the case that there have been benefits politically,such as the creation of a Pan-Indian identity which motivated Aboriginal leadership (Glenn2011).2Chapter 1. Introductionand cultural outcomes. This chapter uses the confidential 1991 Aboriginal Peoples Survey(APS) and exploits differential church resistance to residential school closure to identify theeffects of residential schools. Chapter 4 focuses on a similar set of outcomes but uses the 2001APS in order to study the intergenerational effects of residential schooling. This chapter isunable to use the same identification strategy as the last and instead exploits the extensiveinformation in the 2001 APS on family residential school experience and geography.The history of Canadian Indian residential schooling and Aboriginal education in generalis complex and fundamentally entwined with issues of forced cultural integration, the removalof children from their families, neglect and abuse (Miller 1996; Milloy 1999; Smith 2009;Glenn 2011; Dawson 2012). The brief historical account given here does not cover many ofthe details of the system and is only intended to provide a context for the following chapters.In addition, when assessing the results that follow it is important to keep in mind that allthe reported effects of residential school say nothing about the desirability of the system.Rather, the results speak to the effectiveness of the system in achieving particular measurableoutcomes relative to pre-existing alternatives. The findings here do not speak to individualexperiences nor to they minimize the suffering many have experienced as a consequence oftheir attendance at residential schools.In chapter 3, I find evidence that attendance at a residential school results in accumulationof market skills at the expense of traditional cultural skills, unless the residential schoolattended was highly abusive. If a residential school was attended in a highly abusive decade,there is no accumulation of market skills and still a loss of traditional skill. I measurethe accumulation of market skills by formal high school graduation, employment and notreceiving of government transfers. I measure the accumulation of traditional skills by whetheran individual lives on-reserve, speaks an Aboriginal language at home, or participates intraditional activities. These findings are not mechanical. Even among a population ofAboriginal children who had access to high school, those that attended a residential schoolwere still more likely to graduate from high school and even those living on-reserve are morelikely to be employed. However, I find substantial heterogeneity in the effect among schools.Those individuals that attended schools in decades that could be categorized as exteremely3Chapter 1. Introductionabusive were, if anything, less likely to engage in the market activities and, at the same time,less likely to participate in cultural activities.Although the literature on intergenerational mobility and childhood investment suggeststhat child and parent economic outcomes are positively correlated, it is not obvious thatthe direct effects of residential schooling found in the last chapter will be equivalent to theintergenerational effects.3In chapter 4, I present evidence on the intergenerational effects ofresidential school.The above work complements the existing literature on residential school in a number ofways. First, it offers a different perspective on the residential schooling experience. Second,it provides further evidence that Aboriginal children were systematically selected on theirpotential market and cultural outcomes and that the residential schooling system decreasedtraditional cultural capital within Aboriginal communities. Third, it offers a different viewfrom that generally taken in the existing literature on residential schooling. It does this byexamining its statistical impact on market and traditional skills, examining the heterogenietyin the effect of residential school attendance and investigating its intergenerational effects.Chapter 5 is an attempt to better understand this disparity. In this chapter, I examine dif-ferences in earnings among Canadian Aboriginal and non-aboriginal groups, the compositionof these differences and how they have changed from 1995 to 2005. In order to understandhow these labour market differences may translate into differences in consumption, I brieflyexamine the role of the Canadian taxes and transfer system in minimizing final income dif-ferences. This chapter confirms the large unconditional earnings disparities between differentAboriginal and non-aboriginal people in previous work and demonstrates that a notable frac-tion is due to differences in weeks worked out of the year. This chapter also demonstratesthat a non-negligible fraction of the wage differences between Aboriginal and non-aboriginalgroups is not explained by differences in observables. However, the largest unexplaineddifferences are actually between North American Indians living on-reserve and North Amer-ican Indians living off-reserve. Essentially none of the difference in market wages can be3It should be noted that there is substantial debate about the nature of this correlationand its extent.4Chapter 1. Introductionexplained by differences in observables between these groups. Additionally, although theearnings differences between Aboriginal groups living off-reserve and non-aboriginal peoplehas declined, the difference between the Aboriginal population on-reserve and the populationliving off-reserve has increased. This increase is not explained by changes in the observablecomposition of either population. It may however be explained by unobservable changes inthe population driven by the large changes in employment and the growing proportion ofindividuals identifying as Aboriginal off-reserve. Finally, I find differences in earnings aregenerally not eliminated once taxes and transfers are accounted for.5Chapter 2A History of Residential SchoolsIn 2006 approximately 4 percent of the Canadian population self-identified as Aboriginal(Statistics Canada 2008a). Aboriginal people are the youngest and fastest growing ethnicgroup in Canada and have notably lower annual earnings than the average Canadian.4Thereare three commonly recognized groups of Aboriginal peoples in Canada: First Nations (orNorth American Indians), M?tis,5and Inuit.6While all three populations had experienceswith the residential school system, the experiences of North American Indian children aremost well-known and will be focused on here.74Estimates of how much lower varies substantially depending on the sample, gender andhow an Aboriginal person is defined. Estimates of the earnings gap vary from 7 to 63percent for men and 2 to 15 percent for women (Mueller 2004). Samples that include onlyfull-time/full-year workers such as DeSilva (1999) and George and Kuhn (1994) producemuch smaller estimates of the earnings gap then samples that include all workers (such asPendakur and Pendakur 2002). In 2006, 40 percent of Aboriginal people lived in governmentdefined communities known as reservations or reserves, which have varying degrees of self-governance. Those eligible to live in these areas are known as North American Indians orFirst Nations. Those First Nations members who live on-reserve typically have significantlylower earnings than their off-reserve counterparts. Drost and Richards (2003) estimate themedian on-reserve income penalty to be approximately 40 percent in 1995. George and Kuhn(1994) have estimated the mean on-reserve earnings penalty to be about 14 percent for menusing the 1986 Census. They find the on-reserve penalty is only about 8 percent for women.5R v Powley (2003), the primary and landmark case in defining M?tis rights under the1982 Constitution, specifies the term M?tis as distinctive peoples who, in addition to havingmixed Indian and European ancestry, developed their own customs, and recognizable groupidentity separate from their Indian or Inuit and European forebears More generally, theM?tis are understood to be of mixed American Indian and European ancestry that are noteligible for registered Indian Status. For a discussion of the definition of M?tis see Sawchuk(2001).6The Inuit are Indigenous people who live in the Northern, Artic in Canada and the UnitedStates.7Although M?tis children attended residential schools, the system was not designed forthem. The explicit policy outlining admission of M?tis students was created in a 1911contract between the Federal government and the churches. Clause 4(b) of the contractstated that M?tis children were not to be admitted unless Indian children did not fill theresidential school authorized admission level. If this was the cause, the SuperintendentGeneral could provide authorization for a M?tis child to be admitted, but was not allowed tofund his education in any dimension. This policy was maintained throughout the rest of the6Chapter 2. A History of Residential SchoolsResidential schooling is a phenomenon that has touched a large fraction of the First Na-tions population and weighs heavily in the popular conscience. It is estimated that about150,000 Aboriginal children passed through these schools, with 80,000 former students livingtoday (TRC 2012). During the system's peak from 1930 to 1945, 50 percent of registeredIndian students attended a residential school and in 2001, over 44 percent of North Amer-ican Indian children had a family member who attended residential school. Although 139residential schools are officially recognized, many more may have existed (TRC 2012).8Residential schools operated in every territory and province with the exception of New-foundland and Prince Edward Island. Figure 2.1 shows the distribution of residential schoolsacross the country during the peak of the system in 1930. The flags represent the locations ofresidential schools and the dots indicate the centroid of Aboriginal communities included inthe 1991 Census. It can be seen from this figure that residential schools existed throughoutthe country, but they were more prevalent in the western provinces. The reasons for thepattern are a matter of historical debate and discussed further in Section 2.1.The system involved three main actors: the missionaries (who opened, closed and oper-ated the schools for most of their history); the federal government (who funded, regulatedand enforced attendance); and the Aboriginal families (whose children could be compelledto attend these schools).Missionary involvement with residential schooling predates any federal government in-volvement in Aboriginal education. Missionaries were creating both residential and dayschools as both a conversion and humanitarian effort in the new world as early as the 1500s.The first residential schools were formed by missionaries in New France in the 1600s butthey quickly disappeared, partly because the Aboriginal peoples had no interest and partlyhistory of the system (RCAP, 1996). The Inuit had very little contact with formal schoolingin general and were subject to residential school policy much later than most of Canada. Fora discussion of the M?tis' experience with Indian Residential schools see Chartrand, Loganand Daniels (2006) and for a discussion of the Inuit's experience see King (2006) and Milloy(1999, 239-259).8The government did not embrace the residential school model as formal policy until the1880s when more centralized accounting began to take place (Miller 1996). In addition,the Indian Residential School Settlement Agreement does not cover any residential schoolthat operated without federal government support. Religious organizations and provincialgovernments have often operated residential schools independently of the federal government.7Chapter 2. A History of Residential Schoolsbecause powerful forces on the French side saw little purpose in them (Miller 2004, 222).The later experiments in British Canada took a stronger hold with the creation of one ofthe first and longest lasting residential institutions in 1834: the Mohawk institute (GeneralSynod Archives, 2008; Miller 1996).9Table 2.1 describes the religious patterns among 139 schools included in the Indian Res-idential School Settlement. Approximately 50 percent or more of the schools were operatedby the Catholic Church. Only much later did non-denominational schools begin to open bythe federal and provincial governments: primarily as hostels and in the Northern areas ofthe country. The Anglican Church was the second biggest religious player, followed by theUnited Church. Fifteen of the total 70 Catholic orders were involved in the operation andcreation of the residential school system (Canadian Conference of Catholic Bishops 2012).The federal government's formal involvement in the system began with confederation in1867. Confederation brought with it self-proclaimed fedal government responsibility for thelives of the Aboriginal and in 1868 legislation authorized funds to be used for their education.This spurred the growth of the residential school system. The federal government assumedthe funding of some 57 schools already in existence. Only two were residential schools. By1879 that number was four and then it grew to 71 in 1923. Part of the government's decisionto pursue the residential school model was inspired by a complimentary report publishedby Nicholas F. Davin in 1879. After visiting industrial schools in the United States, Davinpromise in the institutions and recommended the Canadian government to support the model(The Davin Report 1879).The federal government provided per capita grants and funds to establish schools, but theChurches were the ones to propose the schools location and their willingness was essentialin determining whether a residential school was constructed. Milloy (1999) reports that thepattern of rapid growth of residential schools was not based on any Department strategy,9Arguably, the success of the British in establishing schools rather than the French was dueto the shift in the economic motivations of the colonists. Early relationships established withAboriginal people were based on the fur trade and military alliances. Later, the colonists'interests shifted to settlement. The period where it was profitable for the Aboriginal peopleto continue with their traditional lifestyles was coming to a close. The close of the War of1912 also began to limit the use of military alliances with the Aboriginal people (Miller 2004,226).8Chapter 2. A History of Residential Schoolsbut by federal reactions to missionary requests and persistent lobbying by the Churches.The Church's perceptions were often paternalistic: some seemed to believe that they weremore capable of determining a community's needs than the community itself. For example,the Anglican Church believed that the Church represents in many of these developing areasthe appropriate voice of peoples slowly emerging into community consciousness, (AnglicanChurch of Canada, Joint Committee, 1960, 796). These sorts of attitudes and the lack ofinvolvement and control of the Aboriginal population in their education system suggest thatthe existence of residential schools was due to supply, rather than demand concerns. Thegovernment showed a surprising lack of direction and control in the construction and locationof residential schools and rarely ever rejected a missionary's request for school funding (Milloy1999, 56-58). The locations of these schools seemed to be organized mostly due to personalcauses,10the connections of missionaries, and competition among Churches for souls (Miller2001, King 1964).Analysis of confidential government records indicates requests for residential schools wererarely rejected.11Figure 2.2 shows the opening of these schools visually. Many of the lastopenings occurred in the in the north and northern western provinces in align with the latertime of settlement. Schools continued to open into the 40s, 50s and even into the late 60salthough the number of openings were much smaller as government policy shifted away fromresidential schools. However, Church political influence and passion extended the system farpast the date the government believed it was optimal policy and were actively attemptingto close the residential schools.10An example includes the opening of Carcross Residential schools in the Yukon. In thelate 1890s a pioneer Anglican Bishop was becoming increasingly disturbed at the turmoil andsocial upheavals caused the immigrants flooding to his headquarters in Dawson as a resultof the gold rush and he began looking for a more secluded, yet still convenient headquartersto continue his missionary work. Bishop Bompass had already been in the Yukon for overthirty years. He had come from England as a young minister and had become a Bishop as aresult of his dedicated service in the north. For a dozen or more years he had maintained asort of school and orphan care in his own home in Dawson, and had envisioned a permanentresidential school for Indian children but had never found the right location. He had neverbefore been to Carcross and Lake Bennett. As soon as he made the trip the town becamehis See City as well as the location of the school of which he had dreamed so long, (King1964, 40).11There is some evidence that government did show a slight preference for residential schoolscloser to urban settlements if two denominations proposed schools for the same band (Milloy1999, 56-58).9Chapter 2. A History of Residential SchoolsFor most of the system's history, the government had very little direct involvement inits operation and relied on missionary passion, knowledge and involvement. The federalgovernment provided per capita grants and funds to establish schools, but the churches werethe ones to propose the schools location and their willingness was essential in determiningwhether a residential school was constructed. The government played no role in the man-agement of the system until 1911 when it set out a regulatory framework for the churches(Milloy 1999, 52). The government increased its role in the system over time, tighteningregulations regarding the operation of the schools in the 1950s and eventually forcing thechurches out of the official operation of the schools in 1969. However, the role of the churchwas never fully eliminated: until the end of the system's life many of the school's employeeswere originally hired by the churches.In addition to providing funding, and eventually regulations, for the residential schoolingsystem, the federal government also enforced attendance at residential schools.12In the earlyhistory of the system attendance at school was voluntary and many of the students that at-tended residential schools were orphans or came from families that were unable to care forthem (Sealey 1980; Barman 1986). Principals of residential schools were told orphans andchildren without any persons to look after them should be first selected (Milloy 1999, 31).But in 1920, an amendment to the Indian Act made school attendance mandatory for allIndian children between the ages of seven and fifteen. Section A10(1) of the 1920 IndianAct states that every Indian child between the ages of seven and fifteen years who is phys-ically able shall attend such day, industrial or boarding school as may be designated by theSuperintendent General...Provided, however, that such school shall be the nearest availableschool of the kind required , (Indian Act 1920, Emphasis added). Section 10 also outlinesthe mechanics of enforcement: truant officers, and, on summary conviction, penalties offines or imprisonment for non-compliance.It should be noticed that the Act did not clearly define what determined the type of12It should be noted that educational provision for all children is generally part of a gov-ernment's mandate and Aboriginal children are not unique in having their children subjectto mandatory school attendance laws. However, Aboriginal people are unique in havingthe children subject to mandatory schooling laws that allowed the government to seperatechildren from their parents for years at a time.10Chapter 2. A History of Residential Schoolsschool that is of the kind required and left a substantial amount of discretion to theSuperintendent General for student selection. This discretion resulted in residential schoolsbeing operated for orphan children, children from broken homes and those who because ofisolation or the migratory way of life of their families, are unable to attend day schools,(The Administration of Indian Affairs 1964, 44).13After 1951, the welfare role of the residential schools became even more predominant asprovinces began to provide child welfare services on reserves and to use residential schoolsas a resource (Milloy 1999).14This uptake of welfare services by the provinces led to whatis known as the sixties sweep: a large increase in rate of children being removed fromtheir homes and placed into foster care and non-aboriginal homes. Much of this massiveincrease was due cultural differences between provincial welfare authorities and Aboriginalfamilies (Johnston 1983, 23; Jacobs and White 1992). It was not uncommon for childrento be removed if their homes did not abide by typical western family structure (Jacobsand White 1992). For example, Indian Affairs had to send out notice that parents beingseparated, divorced, or unmarried is not sufficient grounds for addition of the child to astudent residence (Indian Affairs 1969). Welfare agents also assumed children to be livingin poverty when they were merely living in accordance to traditional values: while no foodin the house implied hunger in the average Canadian home, it may have only have impliedthere was food outside in an Aboriginal one.If no other schooling options were available, children were forced to attend residentialschools in order to comply with federal legislation. How strictly this legislation was enforced13This selection of children also predominated before the 1960s. For example, in 1923the deputy minister of Indian Affairs told an Oblate bishop that Ottawa was directing itsagents on reservations to give preference to children from neglectful or destitute homes whenvacancies occurred at residential schools (Miller 1996, 313). Even before the 1920s those whocould not attend school on a regular basis were often shipped to residential schools (Sealy1980). In addition, the only residential school to ever exist in Atlantic Canada was openedin 1930 explicitly to service underprivileged Aboriginal children (Miller 1996, 313-314).14A survey completed in 1953 reported that approximately 40 percent of the students fellinto the category of neglected. A 1961 analysis of residential schools in British Columbia cal-culated that 50 percent of the students enrolled were from undesirable homes and the reportsuggested that this was representative of national circumstances. By 1966, a confidential re-port suggested that 75 percent of the students were in this category. By the mid-1970s, theestimates from a number of schools suggest that more than 80 percent of those in attendanceat residential schools were there because of perceived neglect (Milloy 1999).11Chapter 2. A History of Residential Schoolscame down to the discretion of the government agent on reserve (the Indian Agent). If theIndian Agent desired to enforce the law to its full extent, children could be forcefully removedfrom their home by truancy officers and their parents subject to fines or imprisonment(Section 10, Indian Act 1920). Some attempted to fight the system but were punished orthreatened into submission (Haig-Brown 1991, 95-96; Haig-Brown 1991, 109). In addition,after 1945, parents would not receive the Family Allowance15if they did not comply withprovincially legislated schooling ages. If Aboriginal children wanted to attend high school,the only option was often to attend a residential school or to board in a residential schoolwhile attending public school in a non-aboriginal community.16The federal government's enforcement of attendance at residential schools varied overtime. In 1945, government policy began to shift away from the residential schooling systemafter harsh critiques presented by Aboriginal peoples and members of the Indian Affairsdepartment before the Re-establishment and Reconstruction Commission, (Leslie 2002).17The federal government began to view residential schooling as a relic of the past.18Figure2.3 shows this sharp change in government enforcement of attendance at residential schools.While residential schools accounted for over 50 percent of enrollments in schools in 1945, theyaccounted for less than 20 percent by 1965. Later, the federal government began to tightenthe grounds for admission to residential schools and to put more stringent specificationson when removal of children from their homes was warranted. This further decreased theproportion of children attending residential schools (Milloy 1999, 219).In 1969 the director of the Indian Affairs Operations Branch, W.E. Armstrong sentout clarification to Chiefs, Band Councils, Regional Directors and all Superintendents ofIndian Agencies regarding the admission of Aboriginal children in to residential schools.It stated only students who met 1 of 6 requirements could be admitted to a residential15The Family Allowance was introduced in 1945 as a monthly income supplement issued toparents by the federal government on various conditions including compliance with provincialeducation regulations (Milloy 1999, 205).16For a more detailed discussion of the exact content of the admission regulations, pleasesee the chapter 2.17The Re-establishment and Reconstruction Commission was established to review thegeneral state of Canadian affairs after the Second World War.18The government promoted the system's closure and started integrating children intopublic day schools with the assistance of the provinces.12Chapter 2. A History of Residential Schoolsschool: 1) Home is isolated and removed from day school services; 2) Parents or guardiansare migratory; 3) Problems in the home; 4) The handicapped student who has a chroniccondition, but can live in a student residence and obtain regular medical follow-up whichwould be difficult to obtain in the home area; 5) Students who require a period of adjustmentto urban living through living in a residence with peers who share his culture - that is, astudent who requires a gradual orientation to urban living before he can manage in a privateboarding home in the community; 6) No suitable private boarding home is available in thearea in which the appropriate school is located. Categories 1, 2, 3 and 4 applied to studentsup to 14 years of age while categories 3, 4, 5 and 6 applied to students 15 year of age andover. The sixties sweep came as the department began to firmly lay out the conditions forwhich a child could be placed in a residential school (Indian Affairs 1969). In addition, theremay be cases where a family is so large that the parents are unable to provide a proper homewhere the children can continue school. In such cases, consideration can be made to haveolder children admitted to student residences, (Indian Affairs 1969).Not until the late 1960s and early 1970s did Aboriginal parents have any active choicein the education of their children. Before then, Indians took no part in the processes ofeducation, (Hawthorn 1967, 40).19However, even before the official attempt at inclusionAboriginal families were another important actor in the residential schooling system  largelyone of resistance. Although there is significant evidence that Aboriginal people desirededucation, demanding it in the form of residential schools was not the common pattern.20Parents are frequently described as resistant to the residential schooling system, attemptingto prevent their children from attending these schools both indirectly and overtly (Furniss1995). Some parents attempted to fight the system. A Shuswap mother reports attemptingto send her children to public school in order to keep her children out of residential schoolbut was prohibited by the Department of Indian Affairs (Haig-Brown 1991, 95-96).19Although by 1967 the Indian Affairs Department was making an effort to employ andtrain Aboriginal teachers and develop Indian Home and School Associations and IndianSchool Committees these initiatives were only embryonic (Hawthorn 1967).20For example, when First Nations negotiated treaties with the federal government theydemanded clauses obligating the federal government to provide education to some extentfor their children. When the treaties demanded schools, they required them to be built onreserve.132.1. The Intentions of the System and School Location2.1 The Intentions of the System and School LocationIt is the prevailing view that the primary intention of the residential schooling systemwas cultural assimilation and education. Perhaps the most politically and morally neutraldescription of the residential schooling system was given by King (1964). He describes theresidential schooling as serving a unique social purpose. It serves as the institutionalizedmeans by which a dominant society seeks to transmit a body of information, including bothformalized subject matter content and social norms (King 1964, 1).Three motivations are frequently cited for the creation of residential schooling system.The first are feelings of moral obligation to the Aboriginal people. Missionaries and thegovernment felt it was their duty to improve the quality of life of the Indigenous populationthrough Christianity, literacy, skill building and modern health practices (Miller 1996). Res-idential schools also existed partially as welfare institutions for orphans, destitute childrenand children that were from homes that were perceived as neglectful.However, the second motivation often suggested is less benevolent. Numerous authorsargue the Residential schooling system was an attempt by the government to eradicate theIndian way of life and possibility even the Indian himself. Chrisjohn, Young and Maraun(2006) argue that residential schools were one of many attempts at the genocide of the Abo-riginal Peoples, (2006, 21). The term survivor is commonly used dialogue about residentialschools (Aboriginal Healing Foundation 2010). Some academics have concluded that termslike cultural genocide and ethnocide are appropriate in the Canadian case (Hudson andMacDonald 2012). Hudson and MacDonald assert that the essence of what the IRS systemwas about was the attempted destruction of Aboriginal languages, religions, and culturesin Canada (Hudson and Donald 2012, 4). The Assembly of First Nations asserts that all thecharacteristics of the IRS system meet the UN convention of cultural genocide (Assembly ofFirst Nations 2002).The third motive, less frequently cited, was to help prevent rebellion. On more than oneoccasion there was violence between the Aboriginal population and the colonists and thebloodshed that characterized the American experience was undesirable. The potential for142.1. The Intentions of the System and School Locationit to erupt increased with the beginning of settlement in the west, the diminution of thebuffalo, the return of the smallpox epidemic, and the Riel rebellion in 1869-70. The Indianswere seen as a threat to be neutralized through education and assimilation (Milloy 1999).However, portraying the system has having uniform objectives throughout its history maybe overly simplistic. Initially, the goal of the residential school system was to prepare chil-dren to integrate into western society. However, this policy ended in 1910 and was replacedwith the intention of preparing children for a civilized life on-reserve (Barman 1986, 120).Policy seemed to fall half way between assuming Indians were incapable of being integratedinto the majority and that it was undesirable to have them integrate. Either way the civ-ilization effort continued through residential schools attempting to endow a knowledge ofEnglish, western culture and manual labour and homemaking skills (Glenn 2011, 29). Policyagain shifted after the Second World War when integration into public schooling becamethe preferred policy. During this last shift in policy, the residential schooling system waslargely seen as undesirable, but a necessary institution in order to smooth the transition tofull integration and to care for children who had undesirable home conditions.The motivations above for the residential schooling system provide some insight to thereason residential schooling was more prevalent in the western provinces. The first plausibleexplanation is that western Aboriginal people were viewed as less socially advanced, (Sealey1980; Miller 2004, 245). For example, as early as 1869 and 1884 the Indian Act and theIndian Advancement Act allowed the federal government to grant a reasonable level of self-government for the more progressive bands. However, by 1946 with very few exceptions,no bands in Western Canada were granted this privilege. On the other hand, practically allthe bands in Ontario, Quebec, and the Maritime Provinces were under the elective system(MacInnes 1946, 392-394).Another plausible explaination for residential schools being more prominent in the westernprovinces regards the simulatenous increase in funding for residential schools, the decline ofthe buo and the treaty making process (Glenn 2011). Each of the treaties signed in the1870s contained some clause regarding a government obligation to provide schooling. Eachtreaty differed in the exact wording of the education provision and thus brought slightly152.2. Previous Literature on the Effects of Residential Schooldifferent obligations. The most frequently cited example is Treaty One, 1871 which requiredthe government to maintain a school on each reserve `whenever the Indians of the reserveshould desire it' (Glenn 2011, 29-39). However, most of the treaties' wording allows theGovernment to circumvent the wishes of the band by the provision that a school wouldbe provided whenever the band desires it and be it advisable to her Majesty. TreatiesSeven through Eleven have much different wording and do not require schools on reserveor at the behest of the community. Carr-Stewart (2001) suggests that the government didnot meet its obligations to establish schools on reserves when the band desired it. On theother hand, church willingness to establish schools was a convenient method through whichthe government could partially meet their treaty obligations with little active engagement(Milloy 1999). Since there was more funding for the residential school system than beforechurches may have been relatively more willing to establish residential schools than dayschools. The decline of the buo and the increased reliance on the government on the partof the western First Nations may have resulted in less resistance.2.2 Previous Literature on the Effects of Residential SchoolMany authors have suggested that the residential schooling system was unsuccessful asan academic institution and as a care facility (Milloy 1999, Miller 1996; RCAP 1996; AFN2002; Nuu-chah-nulth Tribal Council 1996; Cariboo Tribal Council 1991). Residential schoolshave been accused of failing to protect and provide for the children that attended (Claes andCliffton 1998) and of damaging families (Stonefish 2007).Through the history of the system children who attended these schools, their familiesand Aboriginal and non-aboriginal commentators alike described the children who cameback home as trapped between two worlds. An unnamed mother of children who at-tended residential school recalled: When they came home I was so happy, but they actedso strange...Before they would always hunt, set rabbit snares, and haul wood, but afterthey didn't want to listen and they'd call me names like 'Old Indian,', (Aboriginal HealingFoundation 2002, 26-27). Reporting to government officials, the Pas band stated a that162.2. Previous Literature on the Effects of Residential Schoolchild who returns from a residential school at the age of 16 or 17 is inenviably unable to fitinto the life of the reserve...while inadequate training in the schools and racist attitudes inEuro-Canadian society made it impossible for them to find jobs (SJC Minutes 1947, BensonDifferent Visions, The Pas band brief 32-33).These effects are suspected to persist in generations unaffected directly by residentialschools. The Canadian Truth and Reconciliation Commission on Indian Residential School'sasserts that ...[residential school's] impact has been transmitted from grandparents to par-ents to children. This legacy from one generation to the next has contributed to socialproblems, poor health, and low educational success rates in Aboriginal communities today,(TRC 2010, 1).However, some academics have argued that the residential schooling system generatedan educated elite with a strong Aboriginal identity that spent their careers fighting forIndigenous legal rights and cultural preservation (Glenn 2011; Gresko 1979; Szaz 2006; Miller1996; Reyhner and Eder 2004), rather than generating a culturally stranded, uneducatedpopulation.At this point the statistical literature on the effects of residential school is sparse and doesnot rely on what most economists would understand as causal methodology. The CaribooTribal Council's study on their members finds significant negative impacts of attendance atIRS and abuse on mental health, but not final educational outcomes or employment status. Astudy by the Nuu-Chah-Nulth Tribal Council finds similar results. The Australian BringingThem Home Report (1997) conducted on the impact of removal of Aboriginal children fromtheir homes found that there were no improvements in schooling or employment for thegroup taken away from their homes relative to the group that remained. In addition, thosetaken from their homes were three times as likely to have acquired a police record and weretwice as likely to use illicit drugs. They also earned higher incomes due to a greater degreeof urbanization and thus greater access to welfare payments. The only study I am awareof on the intergenerational effects of residential school is by Bougie and Sen?cal (2000) whodemonstrate a negative association between parental residential school attendance and childschool performance. The negative association the authors find is completely accounted for172.3. The Environments in Residential Schools and Their Alternativesby family income and other factors.Although the existing literature on the effect of residential school is informative, economistswould generally not consider the methodolgoy to be informative regarding the causal effectsof residential school. Given the above description of student selection and the geographicdistribution of residential schools, a comparison of mean outcomes between those that attendresidential school and those that do not, or whose parents attended and those whose parentsdid not, will not be informative regarding causal effects. Chapters 3 and 4 focus on resolvingthis problem.2.3 The Environments in Residential Schools and Their AlternativesResidential schools were located both within Aboriginal communities and as far as hun-dreds of kilometers away. Although children were permitted to return home for summervacation starting in 1920, children were often taken extraordinary distances to attend a res-idential school and many didn't see their family for years (Miller 1996, 311-312; AboriginalHealing Foundation, 2002; McFarlane 1999). Miller stated that the sheer difficulty andexpense of sending children home for summer holidays often accounted for the protractedseparation from family for which residential schools were notorious, (Miller 1996, 311).Unlike schools attended only during the day (day schools) the residential schooling systemoperated on a half day system for much of its history. Half the day was spent in academicsand religion and the other half in skills such as shoe-making and other trades. However, by1910, the half day system did not involve as much instruction in trades as it did manuallabour (Gresko 1986, 94). Although the half day system officially ended in 1951 (Milloy1999, 227) it was abolished in Anglican schools in the mid-1940s and it continued in someschools until the late 1950s (Miller 1996, 530). Until that time, student labour was used tomake up for school budgetary short falls. Regimes at these schools tended to be much moreregulated than a student's life at home (Gresko 1986, 33). Gresko (1986, 92-93) gives an ideaof the weekday schedule of Qu'Appelle Industrial School during the earliest era in Table 2.2.Schooling also involved cultural learning such as ethics, music, differences between white182.3. The Environments in Residential Schools and Their Alternativesand Indian ways of life, and gender roles. Children were only permitted to speak Englishand were either punished for speaking their native language or rewarded for not. Some ofthese punishments were reported to have been severe. Examples of such severe punishmentinclude being beaten to the point of permanent scarring (Crey and Fournier 1998, 62) andhaving needles being inserted into one's tongue (Aboriginal Healing Foundation 2002, 6).However, not all principals and teachers submitted to government preferences regardingEnglish language usage and the restriction of parental visits (Barman 1986; Gresko 1986).Gresko (1986) finds that the principal of Qu'Appelle from 1884 to 1917, Father Hugonnard,failed to inform the department about the Cree and Sioux catechism classes he held. Healso neglected to mention in his reports to the department that he asked the sisters  someof whom had learned to speak Cree  to teach new pupils first in Cree, than in English.Hugonnard also did not yield to the department in its desire to keep children away fromtheir parents. He tended to promote parental visits in a conversion effort (Gresko 1986, 93).Over and above child labour and removing children from their families, residential schoolswere notorious for their early health conditions. One of the most infamous publications on theconditions in residential schools was a pamphlet published in 1922 by Dr. Peter Bryce, theformer Chief Medical Officer of the Department of Indian Affairs. Dr. Bryce was instructedby the Indian Affairs department to conduct a special investigation of 35 residential schoolsin the Prairie Provinces. In his report he highlighted the high death rates among childrenwho had attended these schools. One of the most extreme cases he described was the FileHills reserve's school  75 percent of all individuals to attend the school in its 16 year lifespan were dead at the time of Bryce's survey (Bryce 1922, 4).21The inspection of the schools in 1907 and 1909 was repeated by Dr. F.A. Corbett in1920 and 1922. The Doctor found as many as 75 percent of children were infected withtuberculosis. What concerned Dr. Corbett even further were the signs of neglect. Thediscovery that 60 percent of students at Old Sun in Alberta had scabies or itch...in anaggregated form which was easily remedied and commonly found in children in crowded21The report does not discuss the death rates among individuals who did not attend theschool in the community.192.3. The Environments in Residential Schools and Their Alternativesand unhygienic living conditions (Milloy 1999, 99).Miller states that regulations that were supposed to ensure a healthy student bodywere tightened following the overhaul of schooling arrangements in 1911, with more precisedirections from Ottawa and more supervision. However, neither the provision of medical carenor the enforcement of regulations improved until the more auent days of the later 1950sand they were never comprehensive and effective in their application, (Miller 1996, 301-302). The epidemics of influenza and tuberculosis that ravaged the Aboriginal populationwere not well managed by the schools (Miller 1996, 301-305; Milloy 1999; Bryce 1922) andthe death toll was relatively high even in the 1957 influenza outbreak (Miller 1996, 304).Miller discusses the lack of commitment of the government to give adequate provisions forstudents until after the 1950s. For example, in 1946 students were reduced to brushing theirteeth with soap (Milloy 1996, 307).However, while residential schools were inadequate in many respects, the situations inmany students' home communities were often less than ideal. Over-crowding, poor medicalcare, poor sanitation, poor diet and inadequate clothing were present in both the schoolsand in many of the students' homes (Lux 2001, 107). In addition, regular contact betweencommunity members and the schools in some regions may have facilitated the spread ofdisease and made the health conditions in the two impossible to separate (Stoops, 2006).On top of forced labour, isolation from families, and poor health conditions, the qualityof education received at the residential schools has also been repeatedly questioned over thehistory of the system. However, the small amount of academic literature on early federal dayschools suggests education received at these schools was also poor (Hamilton 1986, 17-18).In fact, day schools in Aboriginal communities were often operated by the same religiousdenominations that ran the residential schools and suffered many of the same challenges.Both types of institutions suffered from significant staff turn-over, employed unqualifiedteachers, and were chronically underfunded (The Department of Citizenship and Immigration1965).2222Funding arrangements also differed between the two types of schools. Until 1957 theresidential schools operated on a per-capita funding basis from the government. After 1957the schools were funded according to a cost based scheme (RCAP 1996). Additional costs had202.3. The Environments in Residential Schools and Their AlternativesFinally, Residential schools are now infamous for the abuses children suffered when at-tending (RCAP 1991; AFN 2002; Nuu-Chah-Nulth Tribal Council 1996; The Cariboo TribalCouncil 1991; Milloy 1999). However, whether the treatment at day schools was far superioris questionable: students who attended day schools are currently launching a lawsuit for thetreatment they received while attending school (CBC, 2012). The issue of abuse is discussedfurther in Section 2.5.In the early 1950s the Department of Indian Affairs (called the Department of Citizenshipand Immigration at the time) began to integrate Aboriginal children into the public schoolsystem. The rise of public schooling gave Aboriginal children a third schooling option toresidential schooling or attending a community day school. At the time of integration,Aboriginal children received no special treatment in public schools and there was no separateprogram to attend to their needs. After an expensive review of Aboriginal education it wasdetermined that this was problematic: the child on entry and the teacher do not implicitlyshare as many values and expectations as do the teacher and the typical middle-class whitechild, (Hawthorn 1967, 13). Socio-economic differences between middle-class white childrenand Aboriginal children made keeping up in public schools difficult. Children were oftensent home because they were dirty or improperly dressed. This was difficult to remedysince many Aboriginal homes did not provide bathing facilities common in non-aboriginalhomes. On top of the fact children were also confounded with a lack of places to study athome (Hawthorn 1967, 19), many children upon entering school still did not speak Englishor French as their first language. This made keeping up at school even more difficult.Children who attended a public school from home could face as long as a two hour busto be made up either by the Church's own funds or by student labour (The Department ofCitizenship and Immigration 1965; Milloy 1996). On the other hand, the federal governmenttook responsibility of teacher's salaries directly in day schools. Prior to September 1, 1949the government did not pay the salaries of teachers in any Indian residential school. On thatdate, the government took over payment of teacher's salaries in three residential schools.In 1961 the government owned 61 of the residential schools and entered into operatingagreements with the religious authorities (expect for one which is operated directly by thefederal government). In 1965 there were four Church-owned residential schools to whichper-capita grants were still given: Albany (Fort Albany, Ont), Fort George (Fort George,P.Q), Notre Dame (Norway House, Manitoba), Holy Angels (Fort Chipewyan, Alta.), andChristie (Kakawis, B.C.) (The Department of Citizenship and Immigration 1965, 32-33).212.4. The Ultimate Closure of the Systemcommutes each way (Educational Task Force 1975, 33). Not only did this increase the incen-tive to not attend classes, but those that did were unable to participate in extra-curricularactivities or receive additional tutoring because of tight bus schedules (Educational TaskForce 1975, 33). On the other hand, in residential schools, some children could engage inextra-curricular activities such as brass bands, Cubs, Brownies, Scouts, Girl Guides andCadets (Miller 2004, 246; Persson 1986). If students were too far from any school to com-mute on a daily basis and if they did not attend a residential school then they would stay inprivate, predominately white, boarding homes. These boarding homes became the prevailingoption for small isolated Aboriginal communities after the closure of the residential schools.In Ontario in 1975, parents had no choice in the boarding home their child was assigned tooand an evaluation of these homes found they were often over-crowded, under-supervised withpoor communication between boarding home and native parents (Educational Task Force1975, 31).Most interactions between the parents and the school board at this time involved theschool board informing parents of the inadequacy of their children. In addition, despiteefforts by the department, textbooks continued to include material about Aboriginal peoplethat was inaccurate, overgeneralized and even insulting (Hawthron 1967, 19). In addition,because Indians did not pay school taxes due to their special status, they were not permittedto be elected on school boards. As a consequence, Indians feel completely dissociatedfrom decisions taken, and too often consider them harmful to the welfare of their children(Hawthron 1967, 69). On the other hand, beginning in the 1970s residential schools hadschool committees Aboriginal parents could be involved in - although evidence seems tosuggest the school committees were dysfunctional (Hawthron 1967, 82).2.4 The Ultimate Closure of the SystemDue to the creation of the Family Allowance in 1945 and the rapidly increasing Aboriginalpopulation (which was rebounding from the tuberculosis epidemic), there was a dramaticrise in the enrollment rates of Aboriginal children in school from 1945 to 1955 (Indian222.4. The Ultimate Closure of the SystemAffairs Department 1945-1955). Ottawa projected that in order to accommodate the growingnumbers of students they needed 60 new classes a year by 1960 (General Synod Archives2008).Rising enrolment rates coincided with changing political attitudes toward racially segre-gated education and the rapid aging of residential school buildings. All these factors wouldcontribute to the decision to end the residential schooling system. The political motivation toend the system was established starting in 1943 during the hearings of the Re-establishmentand Reconstruction Commission. The Commission was formed to review the state of Cana-dian affairs during the SecondWorld War and provided a stage for the critics of the residentialschooling system to voice their concerns. The commission ultimately recommended a reviewof the Indian Act and their findings eventually lead to the closure of the residential schoolingsystem (Leslie 2002).23However, the desire for closure was not enough to lead to a speedy end of the system. Theprocess of closure was long and difficult. It took over forty years for the system to fully shutdown. In fact, the rapidly increasing size of the Aboriginal population, educational demands,and the persistent lobbying of the Catholic Church required a mild expansion of the systembefore the desires for closure could be fully realized. However, integration into provincialpublic schools became the preferred policy and was to be conducted as quickly as possiblewith the residential schools subsequently closed.24Other schools where built in order to23Instrumental to the change was the Indian Affairs Superintendent of Welfare and TrainingRobert Hoey. He brought to the attention of the Committee the increasing costs of theresidential school system and his serious doubts about its efficacy in general. Hoey hadbeen shocked by a visit he made to the Mount Elgin School in 1942 when he discovered theseverely neglected sanitary conditions at the school. A report he commissioned afterward(the Simes report) found similar conditions at a number of other schools across the country(Milloy 1999, 192).24There was also an effort to build more day schools where children did not have access tothem before. In 1945 the Indian Affairs Departmental reports show there were 76 residentialschools in operation with 8,865 on roll with average attendance of 8,006. The majorityof schools were in Alberta with nineteen. In both residential and day schools, there wereno mention of any students past grade 9. In total there were 7,480 students roll, with anaverage attendance of 5,092. The largest number of day schools was in Ontario, with thesmallest number in Alberta with one. There were only six integrated schools in the countrywith only 67 students in Quebec, Ontario, Manitoba and Saskatchewan (Indian Affairs 1945,190-192). Within 10 years, after the end of the war, these numbers looked dramaticallydifferent. Alberta still had 18 residential schools, but now had 30 day schools. BC stillhad 13 residential schools, but now 65 day schools, Manitoba had the same nine residentialschools and 64 day schools. Ontario had 11 residential schools but now 97 day schools.232.4. The Ultimate Closure of the Systemprovide access to education for the new large population of Aboriginal school aged children.Dilapidated schools that did not meet current regulations of health or safety standards wereclosed where possible and students integrated into the provincial system (Milloy 1999).When the decision to shut down the residential school system was taken, Indian Affairsstarted a substantial drive to integrate Aboriginal children into public schools. The firstjoint agreement was signed in 1950 (South Indian Lake, Le Pas) in Manitoba (Hawthorn1967, 67). By 1952 there were fourteen more completed and on the basis of that experiencea set of procedures was drawn up for the Negotiated Integrated School Programs and JointAgreements. The government tried to induce local school boards and provincial governmentsto admit students by offering to cover the costs of educating Indian children in provincialschools by paying a per capita amount based on the general operating costs when educationservices already exist. Where new buildings have to be built, the sharing of expenses isbased on the relative size of the Indian student population compared with the total studentpopulation. If the proportion is a quarter or a third or a half, the federal government will paythe same proportion of the total construction costs, (Hawthorne 1967, 68). The willingnessof the federal government to support local public school boards in integrating Aboriginalchildren into their schools did not mean that agreements to take in new Aboriginal studentswere easily reached. The integration period took over twenty years due to political disputes,consultation and disagreements on what was best for the Aboriginal population.There was major opposition to public integration and the closure of residential schoolsfrom the Catholic Church in western Canada. In Ontario and Quebec, where children couldbe easily integrated into Catholic day-schools, there was no opposition mounted (Milloy1990, 220). After extensively reviewing confidential department of Indian Affairs files, Mil-loy concluded that it was not study, nor quiet rational consideration and discussion, thatdominated the discourse on the western schools over the next decade but political strugglesover the fate of each school...the Department saw the church's hand behind every incident ofopposition and the fight took on a greater character of who would control Indian communi-ties (Milloy 1999, 231). There was even a suspicion that the various religious denominationsPatterns in other provinces looked similar.242.4. The Ultimate Closure of the Systemand sympathetic officials were admitting children who were not neglected in any sense toresidential schools simply to keep the schools open (Milloy 1991, 219).Where schooling agreements could be made, children were integrated into public schoolsimmediately. However, in areas where integration was delayed, the residential schoolingsystem was used to facilitate children's integration into public schools. Children from isolatedcommunities who did not yet have schools in their communities and were too far to betransported to the closest public school, resided at the closest residential school in order toattend public school during the day. In fact, many residential schools ceased to be schoolsand eventually acted as hostels. By 1967, public school integration had practically reachedits saturation point in Manitoba and Saskatchewan (and largely Alberta).25By the late 60sand early 70s, the residential schools remaining primarily acted as hostels and increasinglyas welfare institutions (Presbyterian Archives 2010).A turning point in the entire Indian education system took place after the publicationof the Federal government's White Paper which promoted the end of all legal distinc-tion between Aboriginal and non-aboriginal people and the abolition of the Department ofIndian Affairs (The Government of Canada, 1969). The white paper induced an up-roarfrom Aboriginal leaders throughout the country. When the government attempted to closeBlue Quill's Residential School in Alberta in 1970 (Persson 1986), the people of the SaddleLake-Athabaska district rose up and took control of the school, protesting and physicallycontrolling the building. Eventual transfer of the school to the band was made markingthe beginning of the transition of the last remaining residential schools to the control of thebands (Royal Commission of Aboriginal People 1995, 325). This event is taken as one of thelargest signs of resistance to federal government policy and marking the beginning of IndianControl of Indian Education (National Indian Brotherhood, 1972). Although 23 proposalsfor transfers of residential schools to bands existed, ultimately only 5 followed Blue Quills, allin Saskatchewan (Milloy 1999, 237). By the 1980s only a dozen residential schools operatedby bands were left in existence with one school operated by government at band request.25Hawthorn (1967. 66) asserts that the public schools in these regions were superior to thegovernment's own and thus they felt integration was justified.252.4. The Ultimate Closure of the SystemThe last government-run school closed in 1996, and the last band-run in 1998 (AboriginalHealing Foundation 2009, 176).26Although it took nearly 40 years, the residential system was fully shut down. I investigatehow the composition of education changed as a result of this shift in government policy atthe end of World War Two by compiling data from the Canada Year Book 1941-1970 andthe Department of Indian Affairs Annual reports. Figure 2.3 shows that residential schoolsaccounted for a large fraction of enrollment from 1928 to 1945 and an even greater fractionof attendance. Then, as policy shifted and children were integrated into public schools orday schools were built in their communities, residential schools rapidly began to play a muchsmaller role.Figure 2.4 shows the proportion of total and high school enrollment that federal dayschools accounted for since 1958. We can see from this figure that public integration con-tinued systematically over this period. In addition, most federal day schools were not atthe high school level. Those that were, were residential schools. This indicates that if anindividual attended high school, they likely either attended a public school or residentialschool.Figure 2.5 demonstrates regional variation within provinces and time also exists. Noticethe eastern provinces have a systematically different pattern than western Canada of bothopenings and closures. This may be the result of the different views of the federal governmenttoward the western Aboriginal people and the timing of settlement. Notice in terms ofclosure times, Manitoba and British Columbia are extraordinary similar. This may partially26The final year of closure of the last residential school is a matter of debate. The AboriginalHealing Foundation has used the dates 1892 to 1969 to designate the time period of theofficial residential schooling system (King, Napier and Kechego 2004). Although residentialschools existed long before then, it was not until then a formal order-in-council established thefederal government joint partnership with the missionaries and a set of regulations governingthe schools. The 1969 date is used as a date of closure because it marks the formal endof Church involvement in residential schools and in government's formal residential schoolsystem. However, it should be noted that the government continued to run a number of theschools until 1996 with emphasis on band involvement and ultimate control. Some of theseschools we directly transferred to the band for operation with no government involvement(King, Napier and Kechego 2004). Miller (1996) on the other hand claims the system endedin the 1970s, while Milloy (1999) asserts 1986. The Truth and Reconciliation Commission,uses the date of closure of Gordon Residential in 1996 as the formal date of closure (TRC2012).262.5. Public Revelation of Abusebe due to the fact that both provinces reached early agreements with the federal governmentregarding child welfare provision. This is also true for Ontario, but the patterns there do notcomply with the other two provinces. Figure 2.5 gives the conditional probability distributionof each school being open by each western province. This picture demonstrates that the timesof closure varied across many cohorts in the west. Taken together these figures imply thereis substantial variation in the type of schooling received by Indian students over this timeperiod and that the timing of residential school closure varies significantly across the countryover time.Detailed academic work on the reasons for the timing of residential school closure isrelatively limited. The most in-depth discussions are given by Miller (1996) and Milloy(1999). After going through public documentation, Miller found very limited evidence ofany pattern in closure, calling the process ad hoc. Milloy's description is more thoroughbecause of his access to Northern and Indian Affairs confidential records. His account givesseveral reasons for the differential times of closure across the country, but he largely attributesit to differential strength of vested interests across the country such as the Catholic Church(predominately the Oblates of Mary Immaculate) as described above. If this variation inschool opening and closure was relatively random it can be used as a source of exogenousvariation with which to identify the effect of Indian residential schools on long term outcomes.If Milloy was right, and the Catholic church based its decision to fight to keep residentialschools open because of their relative ability to keep the souls of the Aboriginal childrenin different locations and the timing of school closure should be exogenous to differences inoutcomes between cohorts of Aboriginal children over time. In combination with detailedband and cohort fixed effects, and changes in government policy this type of variation is acredible way to identify variation the effect of residential school. This will be used as themain source of identifying variation in chapter 3.2.5 Public Revelation of AbuseIn 1990, Phil Fontaine, then head of the Association of Manitoba Chiefs, later Grand272.5. Public Revelation of AbuseChief of the Assembly of First Nations, came forward with his experiences of sexual abuseat the residential school in Fort Alexander, Manitoba. He claimed physical, sexual and psy-chological abuse was commonplace in the residential school system and called for a nationalinquiry to what occurred at the residential schools.Although allegations of abuse had existed since the origins of the system, until Fontainespoke out, the criticisms were largely ignored (TRC 2012). In 1994 the Assembly for FirstNations released the report Breaking the Silence: An Interpretive Study of ResidentialSchool Impact and Healing as Illustrated by the Stories of First Nation Individuals. In 1991the Cariboo Tribal Council published their study on the impact of residential schools ontheir members and in 1994 the Nuu-Chah-Nulth Tribal Council did the same. All three ofthe studies recounted stories of deprivation, neglect and abuse. The 1991 Royal Commissionon Aboriginal people concluded, after hearing professional historical testimony and speakingwith former students, that despite department efforts to establish guidelines for punishmentin federal schools, abusive treatment existed through the history of the system.27Lawsuitsand criminal charges began to occur throughout the country, captured international attention(Economist 2000), and eventually ended in the largest class action lawsuit in Canadianhistory: The Indian Residential Schools Settlement Agreement (Reimer, 2010).The settlement agreement consisted of four main components. The first is the CommonExperience Payment. This payment is given to every individual who attended a residentialschool. Individuals are compensated ten thousand dollars for their first year of attendance,and three thousand for every year after. Approximately 80,000 children who attended theseschools received compensation. The Independent Assessment Process was the second com-ponent of the settlement, which was designed to provide additional compensation to studentswho suffered physical, sexual assaults or any other wrongful acts committed by adult em-ployees of the government or church while attending a residential school. Currently, the27The directives regarding strapping, which first were implemented in 1949 and were tight-ened in 1953 and 1962, did not prohibit other forms of punishment such as confinement,public beatings, and deprivation of food, all of which continued to be commonly applied wellpast 1950 (Report of the Royal Commission of Aboriginal People 1996, p.275).282.5. Public Revelation of Abuseassessment process expects 29,000 people to apply (Curry 2012). The third component ofthe settlement required the establishment of a Truth and Reconciliation Commission inorder to acknowledge residential school experiences, the impacts and the consequences of theresidential schooling system. The commission was also required to promote public aware-ness about the Indian residential schooling system and to create a as complete as possiblehistorical record of the system. Finally, it set aside funds for the Aboriginal Healing Foun-dation whose mandate is to assist in healing from the residential school experience (IndianResidential Schools Settlement Agreement, 2006). Students who attended day schools arenow launching their own lawsuit for the treatment they received (CBC, 2012). In the nextchapter, I attempt to deal empirically with the issue of abuse in residential schools.292.6. Tables and Figures2.6 Tables and FiguresBelow are the tables and figures for Chapter 2.Figure 2.1: Location of Residential Schools in 1930 and Aboriginal SettlementsNotes: Data on Aboriginal settlements and positions of residential school locations compiled from geographic sources cited inthe geographic references section. Data on resdiential schools compiled from Where are the Children, by the Legacy for HopeFoundation. This source can be found at http://www.wherearethechildren.ca/. Last Accessed September 28, 2012.302.6. Tables and FiguresFigure 2.2: Opening Of Residential Schools Across the CountryNotes: These figures were generated using the times of government involvement from Aboriginal Affairs and Northern Devel-opment.312.6. Tables and FiguresFigure 2.3: The Percentage of Enrollment and Attendance Accounted for By Residential SchoolsNotes: These calculations were made using the 1941 to 1980 Indian Affairs Reports.Figure 2.4: Composition of EnrollmentNotes: These data were compiled from the 1960-1970 Canada Year Book.322.6.TablesandFiguresFigure 2.5: Cumulative Probability of School Opening and ClosingNotes: The calculations for this figure were made using the information available from www.wherearethechildren.ca.332.6. Tables and FiguresTable 2.1: Number and Religion of Total Residential Schools that Ever ExistedReligion Number Percent of TotalAnglican 30 21.58Catholic 69 49.64Mennonite 3 2.16Non-Denominational 20 14.39Protestant 4 2.88United Church 13 9.35Total 139 100Notes: These calculations were done from the information provided by the Truth and Reconciliation Commission of Canadaand can be found at www.trc.ca.Table 2.2: School Schedule (Gresko 1986, 93)Time Activity5:30 Rise6:00 Chapel6:30-7:15 Bed making, milking and pumping7:15-7:30 Inspection to see children are well7:30 Breakfast7:30-8:00 Fatigue [Chores] for small boys8:00 Trade boys at work9:00-12:00 School, with a 15 minute morning recess12:40 Dinner12:40-2:00 Recreation2:00-4:00 School and trades for older pupils4:45-6:00 Fatigue [Chores], sweeping, pumping, and so forth6:00-6:10 Preparing for supper6:10-6:40 Supper6:40-8:00 Recreation8:00 Prayer and retire34Chapter 3The Long Term Effects of Cultural Separation and Assimilative Education onSkill Accumulation: The Case of Indian Boarding Schools3.1 IntroductionThere are 370 million Indigenous people throughout the world (UN 2009). Despite significantdiversity in this population, indicators of social and economic well-being frequently suggestthey live in more difficult circumstances than the average individuals in their countries. Inthe United States, for example, American Indians on reservation make 39 percent less thanthe average American (Cornell and Kalt 2010), while in Canada registered Indians make 45percent less than average (Cooke, Beavon and McHardy 2004). Similar disparities exists inAustralia (Altman, Biddle and Hunter 2008), New Zealand (Maani 2004), Latin America(Hall and Patrinos 2006), and in the Arctic and Northern Europe (Andersen, Kruse andPoppel 2002). Yet, only a hand full of papers in mainstream economics literature addressthe causes of this disparity.28This paper makes a contribution to this important and growingresearch area by examining the long run consequences of a policy often attributed with largenegative consequences for Indigenous people across the world - specifically, the forcible re-moval of children from their homes and their placement in boarding schools. These boarding28The literature that exists includes the work by Cornell and Kalt (2000) who examineinstitutional formation, legitimacy, and economic growth; Evans and Topoleski (2002) whostudy the long run effects of Casinos on Indian Reservations; and Dippel (2012) who examinesthe long run consequences of forced integration of American Indians on Indian reservations.Sunde, Jorgensen, and Akee (2012) address the effect of parliamentary versus presidentialsystems on economic development while Akee (2009) argues that high transaction costs arecould be a significant source of the lack of development on Indian reservations. Kuhn andSweetman (2002) frame Aboriginal people in Canada as unwilling immigrants and arguethat measures of social integration are linked with economic integration.353.1. Introductionschools are known as Indian Boarding or Residential Schools. They have captured interna-tional attention (Smith 2009; The Economist 2000) and are now illegal under internationallaw (United Nations Declariation of the Rights of Indigenous people 2007).For decades, numerous governments throughout the world implemented residential school-ing policies with the help of various religious organizations. These policies aimed not onlyto educate Indigenous children, but to immerse them in a European way of life (Miller 1996;Milloy 1999; Smith 2009; Glenn 2011; Dawson 2012). Even after the official stance on thenegative influence of the Aboriginal home was abandoned, these institutions played an impor-tant role in government education and welfare policy (Glenn 2011). Hunderds of thousandsof children in the United States, Canada and Australia have been affected by child removaland residential schools and in all three countries there have been calls for compensationand federal government apologies (Rudd 2008; Harper 2008; Gover 2000; BSHP 2008-2011;Cassidy 2009). In Canada and Australia these policies have been subject to scathing publicinquiries (RCAP 1991, Commonwealth of Australia 1997). In Canada, residential schoolshave been called the most damaging of the elements of Canada's colonization, (Milloy 1999)and have resulted in the largest class action settlement in Canadian history (Reimer 2010).However, despite the wide spread nature of these institutions and the fact there are numerousindividual accounts of the negative impact of residential schools (Haig-Brown 1991; Fournierand Crey 1997), there does not exist any statistical research on the school's long run con-sequences economists would recognize as causal (Nuu-Chah-Nulth Tribal Council 1996; TheCariboo Tribal Council 1991; Commonwealth of Australia 1997).29This is the first causal29There are a handful of statistical studies that do not address the issue of selection. Forexample, in a study of their members, the Cariboo Tribal Council finds significant negativeimpacts on mental health from attendance at residential schools and abuse, but not onfinal educational outcomes or employment status (1991). Similar findings were reported bythe Nuu-Chah-Nulth Tribal Council (1996). The Australian Bringing Them Home Report(1997) on the impact of removal of Aboriginal children from their homes found that therewas no improvement in schooling or employment for the group taken away from their homesrelative to the group that remained. In addition, those taken from their homes were threetimes as likely to have acquired a police record and were twice as likely to use illicit drugs.However, they also earned higher incomes due to a greater degree of urbanization and thusgreater access to welfare payments. Not all literature has focused on the negative impacts ofresidential schools. Recent work has made the case that there have been benefits politically,such as the creation of a Pan-Indian identity which motivated Aboriginal leadership (Glenn2011) and careful study on the psychological and cultural consequences have yielded no effect363.1. Introductionstudy exploring the long-term consequences of these institutions.In this chapter, I assess the effect of residential schools on a set of educational, labormarket, and cultural outcomes using a novel, restricted-assess Canadian data set. I overcomethe fact that the federal government systematically selected children to attend these schoolsby leveraging the conflicting objectives of the Catholic Church, the federal government andthe Aboriginal people. Specifically, when the federal government started to shut down theresidential schooling system, the Catholic Church differentially resisted the schools beingclosed based on the local availability of alternative religious infrastructure and religiouscompetition. Hence in areas where there was more or less intense religious resistance toresidential school closure, residential schools were more or less likely effect a given cohort ofchildren in a random fashion. I take advantage of this variation with communities and cohortsby interacting the initial non-indigenous religious concentration surrounding an Aboriginalcommunity in 1941 with the national trend in residential school enrollment. Consequentially,identification is based on the assumption that the interaction between the initial regionalvariation in the non-indigenous religious composition and national changes in residentialschool enrollment is not correlated with changes in outcomes within cohorts and Aboriginalcommunities.30Using this variation in conjunction with confidential data, I assess the long-run effects of residential schooling on both market and traditional skill accumulation.Although a large body of literature suggests that accumulation of formal education is adriver of economic development,31research on the economic conditions of American Indianreservations suggests that communities perform best when formal institutions are rooted intraditional culture (Dippel 2011; Cornell and Kalt 2000). Indian residential schools wereof boarding school attendance within both Canadian and American communities (Walls andWhitbeck 2000).30Initial regional variation interacted with national trends is widely used in the work onlocal labor markets and was first proposed by Bartik (1993) and Blanchard and Katz (1992).I also use opening, closure and proximity to school as identifying variation in additionalspecifications. This sort of variation follows in the spirit of Duflo (2004). Geographicalproximity to educational institutions is used by Card (1995), Tyler (1994), and Neal (1997)as exogenous variation.31This literature is extensive, but citations include Lucas (1988); Mankiw, Romer, andWeil (1992); Benhabib and Spiegel (1994); Hanushek and Kimko (2000); and Aghion et. al.(2005).373.1. Introductiondesigned to simultaneously educate and assimilate the Indigenous population.32Whetheror not these schools achieved their objectives becomes an important question for both thecurrent labour market performance of Indigenous people and the long term development ofAboriginal communities.I find evidence that attendance at a residential school results in accumulation of marketskills at the expense of traditional cultural skills, unless the residential school attended washighly abusive. If a residential school was attended in a highly abusive decade, there isno accumulation of market skills, but still a loss of traditional skill. Accounting for theselection problem turns out to be important for reaching these conclusions. Once selectioninto residential school attendance is accounted for, I find the increase in economic assimilationis substantial and not mechanical: even conditional on reaching high school, residentialschooling increases the likelihood of graduation by at most 22 percent. Market outcomesalso seem to improve: the likelihood of relying on government transfers is reduced by 16percent and the probability of being employed is increased by approximately 15 percentand wages are raised. On the other hand, the loss in cultural connections is significant.Individuals are 16 percent more likely to live outside reservations,3310 percent less likely toparticipate in traditional activities and 8 percent less likely to speak an Aboriginal languagein the home if they attend a residential school. These effects are substantial: less than 20percent of Aboriginal people in the sample speak an Aboriginal language at home.While Indian boarding schools are an extreme example, public educational institutionsgenerally play an assimilative role (Justman and Gradstein 2008) and my results highlightthe potential cultural consequences of these policies. Perhaps the most notably, even thoughResidential schools actively, and at times aggressively, attempted to eliminate cultural con-nection, the extreme assimilation policy within the schools did not drive cultural loss. Rather,I find suggestive evidence that removal from the home environment drove the loss of cultureand segregating Indigenous children from non-indigenous children helped preserve cultural32See Miller (1996), Milloy (1999), Smith (2009), Glen (2010), and Dawson (2012) forexamples.33Reservations are plots of land legally set aside for the exclusive use of specific Aboriginalpeoples.383.1. Introductionconnectedness into adulthood. I show this using information on where a person lived whileattending school: a residential school, a non-aboriginal home, with their own or anotherAboriginal family, or somewhere else. When I compare individuals who stayed in a non-aboriginal home or somewhere else (i.e. not with their family) during school to children whoattended residential schools, I find residential schooling is (if anything) negatively associatedwith economic outcomes and positively with cultural retention.Residential schools are notorious for the abuses children suffered while attending. I ad-dress this issue by examining whether the effects of attending a residential school differsbased on the abusiveness of the environment. To measure the abusiveness of the environ-ment, I construct a ratio of filed abuse claims to the number of children that attended agiven residential school over a given decade. I also examine whether the effect of residentialschooling differs by the religion of the school. While I find the religion of the school matterslittle, relatively extreme ratios of abuse claims to enrollments are negatively related to eco-nomic and social outcomes. In fact, attending a school in a decade with a large proportion ofabuse claims completely eliminates any increase in employment residential school attendancewas associated with.Not only do my results contribute to understanding of the current economic and culturalconditions in Aboriginal communities, but they also contribute directly to the discussion onresidential schools for Indigenous people in the United States and Canada, and Aboriginalchild removal policies in Australia.34My work complements this existing literature by offeringa different perspective on the residential schooling experience, provides further evidence34In a study of their members, the Cariboo Tribal Council finds significant negative im-pacts on mental health from attendance at residential schools and abuse, but not on finaleducational outcomes or employment status. Similar findings were reported by the Nuu-Chah-Nulth Tribal Council. The Australian Bringing Them Home Report (1997) on theimpact of removal of Aboriginal children from their homes found that there was no improve-ment in schooling or employment for the group taken away from their homes relative to thegroup that remained. In addition, those taken from their homes were three times as likelyto have acquired a police record and were twice as likely to use illicit drugs. However, theyalso earned higher incomes due to a greater degree of urbanization and thus greater accessto welfare payments. Not all literature has focused on the negative impacts of residentialschools. Recent work has made the case that there have been benefits politically, such asthe creation of a Pan-Indian identity which motivated Aboriginal leadership (Glenn 2011)and careful study on the psychological and cultural consequences have yielded no effect ofboarding school attendance within both Canadian and American communities (Walls andWhitbeck 2000).393.2. Brief Backgroundthat Aboriginal children were systematically selected on their potential market and culturaloutcomes and that the residential schooling system decreased traditional cultural connectionlwithin Aboriginal communities.Section 3.2 gives a brief description of residential schooling, its alternatives and the mainactors in the system in order to put it into context. Section 3.3 discusses the main datasource used, the basic patterns in the data and explains how the selective process on thepart of the federal government results in an identification problem. Section 3.4 formalizes theintuition given in Section 3.3 by laying out an empirical model of the residential schoolingsystem in order to be precise about the nature of the identification problem. This section alsoclarifies how leveraging the disjoint objectives of the missionaries, the federal governmentand the Aboriginal people provides a solution. Section 3.5 describes the additional data usedto estimate the model presented in Section 3.4 and Section 3.6 presents the main results andextensions.3.2 Brief BackgroundThe history of Aboriginal education in Canada is not a simple one: it is often muddled withother broader and sensitive issues. I discuss here only the elements of the system necessaryfor understanding what follows. For more on the historical details of the system, see chapter2.A total of 139 residential schools existed and operated in every province and territoryexcept Newfoundland and Prince Edward Island. Approximately 150,000 Aboriginal childrenattended these schools with more than half of these former students still living today (TRC2012). Figure 2.1 shows the distribution of residential schools across the country during thepeak of the system in 1930. The flags represent the locations of residential schools and thedots indicate the centroid of Aboriginal communities included in the 1991 Census.The system involved three main actors: the missionaries (who opened, closed and oper-ated the schools for most of their history); the federal government (who funded, regulated403.2. Brief Backgroundand enforced attendance); and the Aboriginal families (whose children could be compelled toattend these schools). The federal government's formal involvement began shortly after con-federation in 1867 and was inspired by a complementary report by Nicholas F. Davin, whoafter visiting industrial schools in the United States, saw promise in the residential schoolmodel (The Davin Report 1879). For most of the system's history, the federal governmenthad very little direct involvement in the operation of the residential schools themselves andrelied on missionary participation. The federal government provided per capita grants andfunds to establish schools, but the missionaries were the ones to propose the schools' locationand their desire to offer a school was essential in determining whether a residential schoolwas constructed. This religious involvement with residential schooling would continue until1969 when the government and various religious denominations would end their partnership(Milloy 1999).In addition to providing funding, and eventually regulations, for the residential schoolingsystem, the federal government also enforced attendance at residential schools.35In 1920,an amendment to the Indian Act made school attendance mandatory for all Indian childrenbetween the ages of seven and fifteen. Section A10(1) of the 1920 Indian Act states thatevery Indian child between the ages of seven and fifteen years who is physically able shallattend such day, industrial or boarding school as may be designated by the SuperintendentGeneral...Provided, however, that such school shall be the nearest available school of thekind required , (Indian Act 1920, Emphasis added). The Act did not clearly define whatdetermines the type of school that is of the kind required and left a substantial amount ofdiscretion to the Superintendent General for student selection. This discretion resulted inresidential schools being operated for orphan children, children from broken homes and thosewho because of isolation or the migratory way of life of their families, are unable to attendday schools, (The Administration of Indian Affairs 1964, 44).36After 1951, the welfare35It should be noted that educational provision for all children is generally part of a gov-ernment's mandate and Aboriginal children are not unique in having their children subjectto mandatory school attendance laws. They were however unique in the sense their childrencould be forcibly removed for years at a time.36This selection of children also predominated before the 1960s. For example, in 1923the deputy minister of Indian Affairs told an Oblate bishop that Ottawa was directing itsagents on reservations to give preference to children from neglectful or destitute homes when413.2. Brief Backgroundrole of the residential schools became even more predominant as provinces began to providechild welfare services on reserves and to use residential schools as a resource (Milloy 1999).37It should be noted that many children that were perceived to be neglected may actuallyhave been well cared for and their removal was a consequence of cultural misunderstanding(Johnston 1983; Jacobs and White 1992).If no other schooling options were available, children were forced to attend residentialschools in order to comply with federal legislation. How strictly this legislation was enforcedcame down to the discretion of the government agent on reserve (the Indian Agent). If theIndian Agent desired to enforce the law to its full extent, children could be forcefully removedfrom their home by truancy officers and their parents subject to fines or imprisonment(Indian Act 1920). Some attempted to fight the system but were punished or threatenedinto submission (Haig-Brown 1991, 95-96; Haig-Brown 1991, 109). In addition, after 1945,parents would not receive the Family Allowance38if they did not comply with provinciallylegislated schooling ages. If Aboriginal children wanted to attend high school, the only optionwas often to attend a residential school or to board in a residential school while attendingpublic school in a non-aboriginal community.39The federal government's enforcement of attendance at residential schools varied overtime. In 1945, government policy began to shift away from the residential schooling systemafter harsh critiques presented by Aboriginal peoples and members of the Indian Affairs de-partment before the Re-establishment and Reconstruction Commission, (Leslie 2002).40Thevacancies occurred at residential schools (Miller 1996, 313). Even before the 1920s those whocould not attend school on a regular basis were often shipped to residential schools (Sealy1980).37A survey completed in 1953 disclosed that approximately 40 percent of the students fellinto the category of neglected. A 1961 analysis of residential schools in British Columbia cal-culated that 50 percent of the students enrolled were from undesirable homes and the reportsuggested that this was representative of national circumstances. By 1966, a confidential re-port suggested that 75 percent of the students were in this category. By the mid-1970s, theestimates from a number of schools suggest that more than 80 percent of those in attendanceat residential schools were there because of neglect (Milloy 1999).38The Family Allowance was introduced in 1945 as a monthly income supplement issued toparents by the federal government on various conditions including compliance with provincialeducation regulations (Milloy 1999, 205).39For a more detailed discussion of the exact content of the admission regulations, pleasesee chapter 2.40The Re-establishment and Reconstruction Commission was established to review thegeneral state of Canadian affairs after the Second World War.423.2. Brief Backgroundfederal government began to view residential schooling as a relic of the past. The govern-ment promoted the system's closure and started integrating children into public day schoolswith the assistance of the provinces. Figure 2.3 shows this sharp change in governmentenforcement of attendance at residential schools. While residential schools accounted forover 50 percent of enrollments in schools in 1945, they accounted for less than 20 percentby 1965. Later, the federal government began to tighten the grounds for admission to res-idential schools and to put more stringent specifications on when removal of children fromtheir homes was warranted. This further decreased the proportion of children attendingresidential schools (Milloy 1999, 219).Aboriginal families were another important actor in the residential schooling system.There is significant evidence that Aboriginal people desired education, though demandingeducation in the form of residential schools was not the common pattern.41Parents arefrequently described as resistant to the residential schooling system, attempting to preventtheir children from attending these schools both indirectly and overtly (Furniss 1995). Notuntil the late 1960s and early 1970s did Aboriginal parents have any active choice in theeducation of their children. Before then, Indians took no part in the processes of education,(Hawthorn 1967, 40).42Residential schools were located both within Aboriginal communities and as far as hun-dreds of kilometers away. Although children were permitted to return home for summervacation, children were often taken extraordinary distances to attend a residential schooland many didn't see their family for years (Miller 1996, 311-312; Aboriginal Healing Foun-dation, 2002; McFarlane 1999). Unlike schools attended only during the day (day schools)the residential schooling system operated on a half day system for much of its history. Halfthe day was spent in academics and religion and the other half in skills such as shoe-makingand other trades. However, by 1910, the half day system did not involve as much instruction41For example, when the Aboriginal people negotiated treaties with the federal governmentthey demanded clauses obligating the federal government to provide education to some extentfor their children. When the treaties demanded schools, they required them to be built onreserve.42Although by 1967 the Indian Affairs Department was making an effort to employ andtrain Aboriginal teachers and develop Indian Home and School Associations and IndianSchool Committees these initiatives were only embryonic (Hawthorn 1967).433.2. Brief Backgroundin trades as it did manual labour (Gresko 1986, 94). Although the half day system officiallyended in 1951 (Milloy 1999, 227) it was abolished in Anglican schools in the mid-1940s andit continued in some schools until the late 1950s (Miller 1996, 530). Regimes at these schoolstended to be much more regulated than a student's life at home (Gresko 1986, 33). Schoolingalso involved cultural learning such as ethics, music, differences between white and Indianways of life, and gender roles. Children were only permitted to speak English and were eitherpunished for speaking their native language or rewarded for not. Some of these punishmentswere severe. Examples of severe punishment include beatings to the point of permanentscarring (Crey and Fournier 1998, 62) and the insertion of needles into tongues (AboriginalHealing Foundation 2002, 6).Not all principals and teachers submitted to government preferences regarding Englishlanguage usage and the restriction of parental visits (Barman 1986; Gresko 1986). Althoughthe quality of education received at the residential schools has been questioned repeatedlyover the history of the system, the small amount of academic literature on early federal dayschools suggests education received at these schools was also poor (Hamilton 1986,17-18).In fact, day schools in Aboriginal communities were often operated by the same religiousdenominations that ran the residential schools and suffered many of the same challenges.Both types of institutions suffered from significant staff turn-over, were chronically under-funded, and employed unqualified teachers (The Department of Citizenship and Immigration1965).43Residential schools are now notorious for the abuses children suffered when attending(RCAP 1991; AFN 2002; Nuu-Chah-Nulth Tribal Council 1996; The Cariboo Tribal Council1991; Milloy 1999). However, whether the treatment at day schools was superior is ques-tionable: students who attended day schools are currently launching their own lawsuit forthe treatment they received while attending school (CBC, 2012).43Funding arrangements also differed between the two types of schools. Residential schoolsoperated on a per-capita basis until 1957, unlike day schools where the government tookdirect responsibly for teachers' salaries and costs (The Department of Citizenship and Im-migration 1965; Milloy 1996). In 1957 both went on a cost based funding system.443.2. Brief BackgroundAs the decades progressed children were integrated into public schools. By 1967, approx-imately 50 percent of Indian children attended public schools (Hawthorn 1967). Whetherpublic schools offered a better learning environment than residential schools is not obvious.Socioeconomic and cultural misunderstandings between parents and teachers were prevalent.Children were often sent home because they were dirty or improperly dressed which wasdifficult to remedy due to the lack of bathing facilities in many Aboriginal homes common tomany non-aboriginal ones. In addition, many Aboriginal students did not speak English orFrench as their first language and this resulted in communication barriers with their teachersat public schools (Hawthorn 1967). In addition, those children that could attend a publicschool from home could face bus commutes as long as two hours each way (Educational TaskForce 1975, 33).44Some children were simply too far from public schools to commute, sothey would have to leave their homes to attend high school. If they did not stay in a res-idential school with other Aboriginal children, they stayed in private, predominately whiteboarding homes. These became a prevailing option after the closure of residential schools(Educational Task Force 1975).The final fundamental actor in the residential schooling system was the missionary or-ganizations that ran, organized and lobbied for the system. Of the total denominationalresidential schools established approximately 60 percent were Catholic, 30 percent Anglicanwith the remainder divided between various other Protestant groups (AANDC 2012). Mis-sionaries had been setting up residential schools as education and conversion effort since the1600s in New France, but only started to establish long lasting institutions in the 1830s inBritish Canada with the financial assistance of the Federal government (Miller 1996). Mis-sionary organizations proposed the school's location and their willingness and initiative wasessential in determining whether a residential school was constructed (Milloy 1999). Themissionaries' perceptions were often paternalistic: some seemed to believe that they weremore capable of determine an Aboriginal community's needs than the community itself. For44Not only did this increase the incentive not to attend classes, but children were unableto participate in extra-curricular activities or receive additional tutoring because of tightbus schedules (Educational Task Force 1975, 33). On the other hand, in residential schools,some children could engage in extra-curricular activities such as brass bands, Cubs, Brownies,Scouts, Girl Guides and Cadets (Miller 2004, 246; Persson 1986).453.3. Data Source, Basic Patterns and the Identification Problemexample, the Anglican Church believed that the Church represents in many of these devel-oping areas the appropriate voice of peoples slowly emerging into community consciousness,(Anglican Church of Canada, Joint Committee, 1960, 796). These sorts of attitudes and thelack of involvement and control of Aboriginal people suggest that the existence of residentialschools was due to supply, rather than demand concerns. The government showed surprisinglack of direction and control in the construction and location of residential schools and rarelyever rejected a missionary's request for school funding (Milloy 1999, 56-58). The battle forthe souls between various religious denominations often led to the quick establishment ofresidential schools and played a pivotal role in the operation of the system (Miller 2001).In fact, church political influence and passion extended the system far past the date thegovernment believed it was optimal policy. The federal government at times faced fiercereligious opposition to residential school closure, most notably on the part of the CatholicChurch in Western Canada (Hawthorn 1967, Milloy 1999).Interestingly, it is this struggle and the conflicting goals of the Church, the federal gov-ernment, and Aboriginal families that allows for identification of the impact of residentialschooling on long term outcomes. The next section presents the main source of data used inthis chapter and presents the basic patterns. I argue that these patterns may be explainedby the selective process that sorted children into schools, rather than the effect of residentialschool attendance itself.3.3 Data Source, Basic Patterns and the Identification ProblemThe main body of this analysis uses the confidential 1991 Aboriginal Peoples Survey (APS)Adult Retrieval file. The APS sample was derived from the census population that answeredthe long form questionnaire45and claimed Aboriginal ancestry and/or individuals who wereregistered under the Indian Act. Those in the APS sample were further required to iden-45The long form was given to 20 percent of households off reservation and 100 percent ofpeople on reservation.463.3. Data Source, Basic Patterns and the Identification Problemtify with their Aboriginal origins.46The data was collected by personal interviews withrespondents, and in 17 percent of cases the survey was conducted through another house-hold member on the behalf of the respondent if they were absent at the time of interview.The data was collected in June 1991 and response to the survey was voluntary. The responserate was 79 percent. Contact could not be made with 14 percent of the sample and 7 percentrefused to be interviewed.47It is important to note that the survey does not include the institutional population(such as those in prisons) nor does it include the homeless. To the extent that residentialschooling increases incarceration rates or homelessness, the results here will be biased. Toget a sense of how important this could be, I form an estimate of the Canadian Aboriginalhomeless population. According to the 2011 Vancouver Homeless Count (2012) there wereapproximately 2,650 people either visibly homeless or in shelters.48Twenty-seven percent ofthese individuals self-identified as Aboriginal. Extrapolating this number to the other fourcities over one million people in Canada and weighting by their population sizes (StatisticsCanada 2013) yields an estimate of approximate 5,000 Aboriginal homeless. Adding thenumber of federally incarcerated Aboriginal people yields a final number of approximately8,400 Aboriginal people not observed.49An estimated 150,000 Aboriginal people attendedresidential school. If all of these 8,400 individuals attended residential school, it implies theywould make up 5.6 percent of the residential schooling population. It should also be keptin mind that this research inherently looks individuals who are still living. Many of thechildren who attended residential school did not live until adulthood (Milloy 1999). To theextent that this fraction is higher than for children that did not attend residential school the46The identification question was: With Which Aboriginal group do you identify? NorthAmerican Indian, Inuit, M?tis, Another Aboriginal group? If they didn't identify withan Aboriginal group they asked if they were a registered Indian under the Indian Act ofCanada? If they said no, they were asked one final question and were then excluded from thesurvey. If residential schools were extraordinarily effective at integration and out-marriage,then people may not identify as Aboriginal and as a consequence my estimates will be alower bound on the assimilative impact of residential schools.47If an individual did not answer a question used to construct a specific dependent variableor if they are missing any of the primary dependent variables they are excluded from thatpart of the analysis. This is not an issue as long as the missing data is random in nature.48I focus on Vancouver because it has a high proportion of homeless and a relatively largeAboriginal population.49The federal Aboriginal inmate population is approximately 3,400 (CBC 2013).473.3. Data Source, Basic Patterns and the Identification Problemresults will be biased.The APS 1991 has two main advantages over the 2001 and 2006 waves of the survey. Thefirst advantage is that those who were between the ages of seven and fifteen during the peakof the residential schooling system were largely still living and of working age.50Second, the2001 confidential file only samples the 130 largest reserve communities and the 2006 files donot include anyone living on reservations, while the 1991 files samples all communities thatresponded to the census.51A notable disadvantage to the 1991 APS is that separate residential schooling questionswere asked to those between the ages of 50 and 64 and for those between 15 and 49. Com-parability of the responses for these questions may be an issue. The question asked to thosebetween 50 and 64 was Did you ever attend a residential school?. The question to thoseless than the age of 49 asked first whether an individual attended a single elementary schoolor multiple elementary schools. Then they asked subsequently where did you live whileattending school: a) lived with family while at school; b) lived with a non-aboriginal familywhile at school c) lived at a residential school d) lived somewhere else. This process was thenrepeated for high school education if attendants ever made it to high school. All of thesesub-questions are used to create a single indicator of whether an individual ever attended aresidential school.52Anyone over the age of 65 was not asked any questions regarding theireducation.I restrict my sample to include those individuals who are registered under the IndianAct, are members of an identified band, live in the western provinces (British Columbia,Alberta, Manitoba and Saskatchewan) and are aged 20 to 65. I limit my sample to those50The oldest people who answered the residential schooling question were 65 years of age,thus would have been born in 1926 and been of the mandatory schooling age by 1933.51Note here that the way I am defining communities is based on census subdivisions whichare municipalities or areas that are deemed to be equivalent to a municipality for statis-tical reporting purposes such as an Indian reserve (Statistics Canada). The reader shouldnote that there are technically many more reservations than Census subdivisions. Thereare 2,675 reserves Canada wide, but only 615 bands. Over half of these reservations are inBritish Columbia (http://www.gov.bc.ca/arr/reports/facts/overview.html). Census Subdi-visions often aggregate many smaller settlements into larger statistical areas.52Another disadvantage of the APS 1991 is that it does not provide intergenerational in-formation on residential schooling. The APS 2001 and the APS 2006 ask very detailedquestions regarding family history of residential schooling experience. Issues regarding theintergenerational effects of residential schooling are left to chapter 4.483.3. Data Source, Basic Patterns and the Identification Problemwho are registered under the Indian Act and are members of a band since these are theindividuals that the residential schooling system was designed for. Although the M?tis wereaffected by the residential schooling system, their experiences are unique and I do not includethem here due to the risk of over-generalizing.53I restrict the location of residence to thewestern provinces because residential schooling serviced a greater proportion of individualsthan in the eastern provinces, these bands are more uniform in their pre-settlement contactand educational alternatives and this avoids dealing with the unique circumstances of theInuit.54The sample is restricted to those older than 20 in order to ensure individuals aregiven a chance to complete their high school education, and it is restricted to those youngerthan 65 since anyone older was not asked schooling questions in the 1991 APS.Table 3.1 shows summary statistics of the dependent and independent variables of interestby whether or not a child attended a residential school. This includes all people indepen-dent of the time their closest residential school ceased operation. Individuals who attendedresidential schools are more likely to be female, tend to have solely Aboriginal ethnic origins53The explicit policy outlining admission of M?tis students was drawn up in a 1911 contractbetween the Federal government and the churches. Clause 4(b) of the contract stated thatM?tis children were not to be admitted unless Indian children did not fill the residentialschool authorized admission level. If this was the cause, even then the SuperintendentGeneral could provide authorization for the child to be admitted, but was not allowed tofund his education in any dimension. This policy was maintained throughout the rest of thehistory of the system (RCAP, 1996). For a discussion of the M?tis and Indian Residentialschools see Chartrand, Logan and Daniels (2006).54The Inuit had very little contact with formal schooling in general and were subject toIndian Residential Schools much later than most of Canada. For a discussion of the Inuitexperience see King (2006). Aboriginal peoples in the West faced substantially differentcircumstances than those in the East in ways that might violate the assumptions of theframework outlined in this chapter. The first, western Aboriginal people were viewed asless socially advanced, than those in the eastern provinces (Sealey 1980; Miller 2004, 245);as early as 1869 and 1884 the federal government began granting a considerable levels ofself-government for the more progressive bands, and by 1946 with very few exceptions,all bands in Ontario Quebec, and the Maritime Provinces were under the elective systemwhile no bands in Western Canada were (MacInnes 1946, 392-394). In addition westernCanada did not have long standing day schools like in the eastern provinces  by the early1900s there were 28 Aboriginal day schools in the Atlantic provinces alone (Hamilton 1986)and over one hundred in the Eastern provinces in total. Residential schools were also moreprominent in the western provinces as a result of the treaty making process during the 1870sin an attempt to avoid violence and the decline of the buffalo (Glenn 2011). All of thesefacts could heavily influence the ability of Aboriginal communities to drive the opening andclosure of residential schools and results in substantially different time patterns than in theeastern. Restricting the sample to the western provinces also clarifies the counterfactualenvironment faced by the Indigenous population.493.3. Data Source, Basic Patterns and the Identification Problemand are on average ten years older than those who did not attend residential school. Theyare also located closer to their nearest major city.Interestingly, the means comparison also reveals that children who attended a residentialschool are equally likely to have graduated high school as those who attended a day school.While many have claimed residential schooling was of low quality (Milloy 1999; RCAP 1996),these results suggest that they were no less successful than the alternative options for highschool graduation. Aboriginal people who attended a residential school are more likely toreceive government transfers and less likely to be employed. This falls in line with earliersuggestions that the system did not economically benefit the children who attended residen-tial schools (Miller 1996). On the other hand, Aboriginal peoples who attended residentialschools are more likely to speak an Aboriginal language at home and to participate in tra-ditional activities. This is contrary to the claim these schools were culturally destructive.These patterns could be explained in the following way: the harsh enviroments in resi-dential schools harmed children's ability to accumulate skills that were valued in the marketand thus prevented those who attended from engaging in it. As a consequence, individualssubstituted away from market activities to cultural activities. While feasible, there are sev-eral entirely different explanations for this pattern that have nothing to do with the effects ofresidential schooling. One very plausible explanation that follows clearly from the historicalbackground is the selective process on the part of the federal government. People who wereorphans or came from broken/neglectful homes, may preform worse in the labour marketregardless of whether they attended a residential school. If residential schooling actuallyimproved educational outcomes, children from homes deemed to be neglectful could be lesslikely to be employed as adults. In addition, children with parents who participated intraditional lifestyles may themselves be more likely to participate in traditional activities,independent of their residential schooling experience.The enforcement of residential school attendance by the federal government varied overtime as a consequence of shifting public opinion and culminated in a sudden change ingovernment policy as discussed above and as demonstrated in Figure 2.3. This figure explainswhy those who attended a residential school are much older than those who did not attend a503.4. The Empirical Frameworkresidential school. This sharp change in policy coincided with a rising demand for educationin Canada in general (Milloy 1999). At the same time, Aboriginal languages were in decline.As a consequence, subsequent generations are more likely have formal education and lesslikely to participate in cultural activities than the older generation are due to trends unrelatedto residential school attendance. In this environment, it is clearly important to account forcohort effects.Even accounting for differences in cohorts may not be sufficient to account for differencesbetween these two groups. As mentioned previously, the likelihood of attending a residentialschool would be affected by treaty obligations on the part of the government. If differencesbetween treaties and/or bands matter for outcomes independent of residential schooling,then accounting for them is also important.What is left is the variation within communities over time, net of national cohort trends.While this may seem like a promising source of variation, it is not robust to the first iden-tification problem outlined: the selective enforcement of residential school attendance. Forexample, if a community had slower than average economic integration, then the federalgovernment (based on the historical discussion above) would likely reduce pressure to attendresidential schools more slowly than in areas with faster than average economic integration.If this is related to outcome differences between cohorts within communities, this source ofvariation will also be invalid.3.4 The Empirical FrameworkGiven the above discussion, it is useful to be more formal about the actions and decisions ofthe government, Aboriginal families, and the missionaries. In this section I lay out a simpleframework based on the historical accounts of the residential schooling system. I model thefederal government as an enforcement agent who desires to assimilate and educate Aboriginalchildren. Aboriginal families are concerned with their children's well-being and choose howmuch to resist their children being taken to residential school. Together, the decision rules513.4. The Empirical Frameworkof the government and Aboriginal families determine the demand for residential schools.Missionary organizations are assumed to care only about converting Aboriginal childrenand choose the supply and location of residential schools on that basis. This frameworkis obviously a highly stylized, but its simplicity allows me to be precise about the natureof the endogeneity concerns and assists me in clarifying my identification strategy and itsplausibility.First, assume the government attempts to educate and assimilate Aboriginal children byselectively enforcing the provision that allows the state to remove children from their homes.The level of enforcement the government chooses to implement is child specific and givenby Eijt, where i indexes children, j a given child's community and t their cohort.55 Second,assume that a person's well-being is determined by their adult skills, specifically, their skillsthat are valued in the market and skills that are valued their traditional community. Forthe sake of comparability with the rest of the economics literature I will call skills thatare valued in the labour market human capital and skills that are valued in a traditionalcommunity cultural capital. Aboriginal families care about whether their child attends aresidential school or a day school because it has different consequences for their child's finalaccumulation of these skills. Aboriginal families choose some level of resistance regardingtheir child being taken to residential school based on these preferences and the other optionsavailable to their children. Parent's optimal amount of resistance is given by ??ijt.56A childattends a residential school if the amount of resistance chosen by their parent is less thangovernment enforcement, Eijt > ??ijt. This can be represented as:Aijt =?????10if Eijt > ??ijtif otherwise, (3.1)where Aijt indicates attendance at a residential school.55Enforcement should be thought of as the cost the government imposes on parents if thechild is not sent to residential school.56Examples of parental resistance include hiding their children, physically resisting theIndian Agent, or paying fines.523.4. The Empirical FrameworkWe saw in section 2 that government enforcement varied along several dimensions. First,government enforcement varies by cohort due to changes in policy over time (demonstratedin Figure 2.3). The cohort-dependent level of enforcement is given by Ct. Enforcement alsovaries by community: the community specific level of enforcement is given by Bj in themodel.57This allows government preferences to vary over fixed community characteristicssuch as treaty region or proximity to the closest city. The government enforcement level alsodepends on the cost of sending a child to a residential school. The cost of enforcement for acohort and community depends on the local supply of residential schools, given by e(zjt, ?j)where ?j is the distance of the closest residential school to community j, and zjt indicateswhether the school is open when cohort t is of schooling age.58Government enforcement is assumed to also depend on unobservable, individual specific,idiosyncratic endowments of market and cultural skill (which are given by hijt and ?ijt re-spectively). The initial idiosyncratic endowments of cultural and market skill children receivewill also determine their adult stocks of these skills independent of parental decisions. Thegovernment is assumed to care about these endowments since families with more adherenceto traditional cultural norms were targeted historically.59Thus, the effect of cultural capitalon the enforcement level, ??, is positive while the effect of human capital, ?h, is negative.I also allow the selection of individuals to depend on their gender and whether they havenon-aboriginal ancestry, which is given by the vector xijt. There is also another unobserv-able idiosyncratic term that varies by cohort, community and individual, ?ijt.60 Thus theenforcement level for each individual is given by:57I use Bj to represent band which is not the same necessarily the same as geographiccommunity. However, frequently they are and a set of fixed effects can be used for band orcommunities with no consequential effects on the empirical results.58This can also be thought of as representing an existing contract with a religious group tooperate the school for a specific band or group of bands.59Adherence to traditional cultural norms is assumed to be correlated with fewer marketskills because of the naturally limited time available to dedicate to each of these activities.It is not to say that they were not at times complementary such as with the accumulationof wealth for Potlatch ceremonies Salish communities (Lutz 2008).60This is meant to represent Indian Agent specific preferences for residential school atten-dance for a particular child. Since Indian agents vary by time and place, ? also indexed byj and t.533.4. The Empirical FrameworkEijt = ?x?ijt + Ct +Bj ? e(zjt, ?j) + ijt. (3.2)where ? is a parameter vector and ijt = ???ijt + ?hhijt + ?ijt.Aboriginal parents know that the government will enforce attendance according to Equa-tion 3.2, but do not observe ?ijt. Let ??ijt represent total cultural capital and h?ijt total humancapital. Human capital accumulates for each child according to the total amount of timethey spend in school, given by ?s, where s indexes the type of school (s = d for day schooland s = b for residential school)61, multiplied by the quality of the schooling given by q. Theamount of cultural capital accumulated is given by the amount of time a child spends withtheir family over the course of their schooling years. This is given by ?? ? ?s where ?? is thetotal time available during their schooling years. The accumulation rate of cultural capital isgiven by ?. The human and cultural capital accumulation equations are given respectively ash?ijt = q?s+hijt and ??ijt = ?(????s)+?ijt. Parent's utility is assumed to be given by the somelinear combination of h?ijt and ??ijt and parents choose their optimal level of resistance, ??,accordingly. To solve the parental decision problem I assume that ijt is normally distributedwith mean zero and variance equal to one.To construct the outcome equations assume there exists a set of cultural outcomes, eachgiven by ?ijtk and market outcomes, eijtm, whose return is given by ?k??ijt+B?jk+C?tk+??k2xijtand ?mh?ijt + B?jm + C?tm +??m2xijt respectively.62The subscript m indexes market outcomesand k indexes cultural outcomes. The factors B?jk, C?tk, and ??k2xijt, allow the return to a givencultural activity k to vary by community, birth cohort and a set of individual characteristicssuch as gender and ethnic origins. Parameters specific to market activities are definedsimilarly. Whether an individual chooses to engage in each type of activity will depend ontheir return to that activity. If the return to that activity is positive they will engage in it,and if negative they won't. Substituting for h?ijt in the return to market activity m will givethe decision rule for engaging in market activities and substituting in for ??ijt in the returns61Note that the time in boardings school will be greater than the time spent in a day schoolso ?b > ?d.62Given that most of the outcomes I have access to are binary I focus on zero/one outcomeshere, but a similar intuition follows for continuous variables.543.4. The Empirical Frameworkto cultural activity k gives the cultural activity decision rule. Solving the parent's decisionproblem yields63Aijt =?????10if ?1 + ??2xijt ? e(zjt, ?j) +Bj + Ct + ijt > 0if otherwiseeijtm =?????10if ?m1 + ??m2xijt + ?m3Aijt + B?jm + C?tm + ?mijt > 0,if otherwise?ijtk =?????10if ?k1 + ??k2xijt + ?k3Aijt + B?jk + C?tk + ?kijt > 0,if otherwise?kijt, ?mijt, ijt ? N (?, ?) , ? =?????000?????, ? =?????1 0 ?10 1 ?2?1 ?2 1?????(3.3)where ?ijtk represents a set of k cultural outcomes and eijtm a set of m market outcomes,?kijt = ?k?ijt, ?kijt = ?k?ijt, i = 1, ...N, j = 1, ..., J, t = 1, ..., T , and ?1 = ?k?k and?2 = ?m?h.64 The parameters B?jk, C?tk, and ??k2xijt, allow the return to a given culturalactivity k to vary by community, birth cohort and a set of individual characteristics such asgender. Parameters specific to market activities are defined similarly.As a result of government selection being based on children's initial unobservable endow-ments of human and cultural capital the outcome and attendance equations are correlatedthrough their error terms. To evaluate the causal effect of residential school attendance on63The parents decision problem and the outcome equations is repeated and expanded onin appendix A.64Given that most of the outcomes I have access to are binary I focus on zero/one outcomeshere, but a similar intuition follows for continuous variables.553.4. The Empirical Frameworkoutcomes, an additional parameter - the correlation of the errors terms - must be estimated.65For the model above to be identified independent of functional form restrictions there mustbe at least one variable that varies over both cohorts and communities and affects residentialschool attendance but not adult outcomes.Note that in this framework, the cost of government enforcement depends on the timevarying local supply of residential schools. This supply depended heavily (and does exclu-sively in this framework) on the decisions of the missionaries. If the geographic supply ofresidential schools is driven primarily by religious objectives rather than the selective processof the federal government, one could imagine using a community's distance from the closestresidential school, ?j, and the process of the school's opening and closing, zjt, as exogenousvariation (i.e. I could impose that e(zjt, ?j) = ?3zjt + ?4zjt?j in estimation).However, thinking about the missionary's decision problem makes clear the strong re-strictions required to use distance and school opening and closing as exogenous variation.Consider the decision of a missionary that is distance ? from community j. The missionarygets utility from educating and converting Aboriginal children. The missionary's indirectutility function can be given by:v?j = zjt?1jtE[?Ntji=1Aijt]+ (1? zjt)?2jtNjt,where Aijt is equal to one when a child in community j and time t attends a residentialschool and zero otherwise. The expected number of children who will attend the residentialschool is given by E[?Ntji=1Aijt]and the proportion of those children who will be convertedand educated is given by ?1jt per dollar spent.66The per dollar fraction of the communitythat will be educated and converted if the missionary does not open a residential schoolgiven by ?2jtNjt where Njt is the number of Aboriginal people within a community.67Thevariable zjt = 1 if the missionary opens the school and zjt = 0 otherwise. A similar intuition65Note that ?kijt and ?mijt do not have to be uncorrelated for the results to be consistentsince this restriction is not imposed in estimation.66The expectation is taken over the sum of the attendance equation in 3.3 for all individualsin a cohort and community.67Which may depend on whether there is alternative Catholic school available for thesechildren.563.4. The Empirical Frameworkfollows if the school is already open at a given t and j and the missionary must decide tokeep the school open or close it (although ?1jt and ?2jt will be different). Thus missionarydecisions to open or close a residential will depend on E[?Ntji=1A?ijt], Ntj, ?2jt and ?1jt (theaverage attendance at a residential school, size of the Aboriginal population, and the relativecost effectiveness of conversion).To use distance and the opening and closing of schools as exogenous variation, variationsin E[?Ntji=1A?ijt], Ntj, ?2jt and ?1jt from their community and cohort averages must vary in-dependently from human and cultural capital endowments ?ijt and hijt. In other words, ifopenings and closings are to be used directly as a source of exogenous variation, missionariesmust only make decisions based on persistent impressions of a given community's residentialschool attendance given the selection process. This restriction is rather strong. For example,it assumes that missionaries cannot have rational expectations regarding student residentialschool attendance.68In this framework, if one cohort in a community has relatively lowmarket skill endowments then the government will select many children from this cohort toattend a nearby residential school. Consequentially, nearby missionaries may choose to opena school and keep it open. However, if successive cohorts in this community have highermarket skill endowments, fewer children will successively be selected to attend a residentialschool. As a consequence, missionaries nearby may choose to close their school. If this storywas systematically true, then the opening and closing of schools may be correlated with stu-dent outcomes through their human capital endowments and could not be used as exogenousvariation.Fortunately, this framework also suggests capitalizing on a cleaner source of variation.What the framework makes clear is that there is at least one factor that influences the supplyof residential schools that does not enter demand. Specifically, the likelihood of conversionper dollar which enters the objective function of the missionaries (who control the supply ofresidential schools) but does not enter the decisions of the Aboriginal people or the federalgovernment (whose joint decisions determine the demand for residential schools). Thus, the68Missionary decisions are permitted to vary by unobservable community-specific cohortvarying factors that are unrelated to individual outcomes such as plausibly Njt and evenmore plausibly?1jt/?2jt.573.4. The Empirical Frameworkperceived effectiveness of religious alternatives available to the Aboriginal people could beused to directly determine the effect of residential school attendance. In the model, the sizeof ?1jt relative to ?2jt represents the per dollar likelihood of conversion through residentialschooling relative to the likelihood of conversion using the existing religious infrastructure.Differing amounts of religious infrastructure imply differing returns from a residential schooland thus differing levels of resistance to school closure on the part of the missionaries whenthe federal government attempted to reduce enrollments rates. If there were fewer optionsfor non-secular education - or more competition from other religious organizations - mis-sionaries would be less likely to close their residential school, which would reduce the costof enforcement and thus kept enrollment rates higher for longer periods than in areas wherethere were more options for non-secular education or less competition from other religiousorganizations. Conditional on fixed geographic characteristics (such as distance from theclosest major city), this sort of community-cohort variation in residential school attendanceis a useful source of variation since it is plausibly exogenous from changes within communitiesand cohorts in adult outcomes.The general religiosity of the non-aboriginal population surrounding an Aboriginal com-munity should be highly correlated with the presence of a non-indigenous denominationalschool nearby and general religious infrastructure. Thus, historic variation in geographicreligious composition should influence the trends in enrollments within communities overtime. To capture this, I multiply the 1941 religious composition (specifically the Catholicproportion) surrounding an Aboriginal community with the deviation in national enrollmentrates from their peak in the 1930s.69Let ?2jt/?1jt = ?twjt=1941 where wjt=1941 indicates the69Given that I control for cohort effects and geographic fixed effects, I need the following as-sumption for identification: how the historic non-aboriginal religious composition in a givenarea interacts with the overall government-determined trend in residential school enrollmentis conditionally independent of unobserved changes in outcomes between cohorts within com-munities. If this assumption holds and the variable just described is significantly correlatedwith attendance at a residential school, then the model is identified independent of functionalform restrictions. An example of a phenomenon that might violate this restriction would beif the proportion of Catholic individuals (of the non-aboriginal population) influenced rateof change in discrimination against the Aboriginal population in proportion to the change inresidential school attendance at the national level. So, for example, if the opportunities foremployment of Aboriginal peoples increase when there is less discrimination, and in areaswith a higher proportion of Catholic people decrease discrimination more quickly than thosewith a lower proportion of Catholic people, this estimation strategy would be biased toward583.4. The Empirical FrameworkCatholic proportion in a census division surrounding an Aboriginal community in 1941 and?t is the proportion of children nationally that attended a residential school in each cohort.Since ?twjt=1941 is plausibly independent of variations in Njt and hjt and ?tj, this proposedsource of variation will not suffer from the same challenges as using zjt and ?jt directly.Given the above framework, the missionaries' choice of ?j is a function of ?1jt, ?2jt,E[?Ntji=1A?ijt], and Njt. As a consequence, e(zjt, ?j) = e(?1jt, ?2jt,E[?Ntji=1A?ijt], Njt) whereE[?Ntji=1A?ijt]may depend on Ntj,hjt, and ?tj. If ?2jt/?1jt is an additively separable, linearcomponent of this function, e(zjt, ?j) = ??3?twjt=1941 + f?(Ntj, hijt, ?ijt), it implies that theerror term, ijt, in Equation A.4 becomes ?ijt = f?(Ntj, hijt, ?ijt)+?hhijt+???ijt+?ijt. Assum-ing ?ijt is normally distributed with mean zero and variance one, both these models can beestimated by quasi-maximum likelihood (using a bivariate probit). The model's likelihoodsare given in appendix C.The results of estimating the specifications given by Equation 3.3 can be found in section3.6.1. This model allows me to estimate the causal effect of attending a residential schoolfor all children (also known as the average treatment effect or ATE) and, perhaps moreplausibly, the effect of attending a residential school for the children who actually attended(also known as the average treatment effect on the treated or ATET).3.4.1 Historical PlausibilityThe importance of religious entitlement in the operation and persistence of the residentialschooling system cannot be underestimated nor can the bitterness of the sectarian environ-ment in which educational and evangelical experiment in Indian schooling was initiated,(Miller 2001, p.141). The sense of the religious competition between various denominationscan even be felt entrenched in the legislation of the 1920 Indian Act: . . . no Protestantchild shall be assigned to a Roman Catholic school or a school conducted under RomanCatholic auspices, and no Roman Catholic child shall be assigned to a Protestant schoolor a school conducted under Protestant auspices, (Indian Act 1920, Section 10(1)). Whenfinding positive economic effects of residential schooling.593.4. The Empirical Frameworkthe Act allowed for this section to be over-ridden by written directive of the parent, theAnglican Church argued that it would be wiser to omit this clause from the legislation thusdemonstrating possessiveness over the souls of Aboriginal children that superseded the valueplaced on parental rights (Anglican Church of Canada, Joint Committee, 1960, 800).When the federal government began attempting to shut down the residential schools infavor of integrating children into the public education system the churches would have theirsay and would make their influence felt, (Milloy 1999, 219). Especially intense feelings wereexpressed on the part of the Catholic Church. The Roman Catholic Church regarded itselfas the `one true faith' and did very little to disguise its view that Protestantism. . . was atbest an error and at worst dangerous heresy, (Miller 2001, 141) and non-secular educationwas looked at with particular disdain. For example, in 1965, an Oblate Father said thisto his parishioners: Satan and his legion, making a review of their positions came to theconclusion that they were losing ground the world over and the Indian population was notexempt; therefore, they changed their strategy...What is this strategy? Or, to put in modernwords, what is this policy? To them religion must be done away with in all schools. A formulamust be found to lure away the Indians from denominational schools. . . (Hawthorn 1967,57).Although the differential resistance by the Catholic Church within western Canada toresidential school closure to this point has not been quantitatively documented, the anecdotalevidence is suggestive. Specifically, there was major opposition to public integration and theclosure of residential schools from the Catholic Church in western Canada where the provincesdid not provide for public Catholic schools, but in Ontario and Quebec, where children couldbe easily integrated into Catholic day-schools, there was no opposition mounted (Milloy1990, 220). Certain areas in western Canada were hives of evangelism  fractionalizedamongst various religious groups and were political hot spots where religious control ofeducation was particularly contentious and arguably there was an incentive for the Catholicchurch to hold as much ground as possible with the Aboriginal population (Miller 2001, 143).After extensively reviewing Indian Affairs department files acquired through the Freedomof Information Act, the preeminent historian on residential schooling in Canada concluded603.4. The Empirical Frameworkthat it was not study, nor quiet rational consideration and discussion, that dominated thediscourse on the western schools over the next decade but political struggles over the fate ofeach school...the Department saw the church's hand behind every incident of opposition andthe fight took on a greater character of who would control Indian communities; there waseven a suspicion in the Indian Affairs department that the various religious denominationsand sympathetic officials were admitting children to residential schools who were not evenperceived to be neglected in any sense in order to simply to keep the schools open (Milloy1999, 231, 219).The religious battle for souls and the political strength of the Catholic church is cited asone of the major reasons the residential schooling system took over 40 years to shut downonce it was determined the system was no longer desirable (Milloy 1999). Perhaps this viewis best expressed in one of the most extensive reviews of Aboriginal affairs in Canadianhistory:An examination of the attitudes of the denominational groups throws a light onthe opposition experienced by the Indian Affairs Branch in its search for viablesolutions. These attitudes act as a brake on the development of Indian educationthrough the stress they place on their own privileges and on the dangers whichschool integration presents to faith and morals, (Hawthorn 1967, 62).Although none of this historical discussion clearly demonstrates the mechanism for identifi-cation I propose, I view it as highly suggestive. After discussing the additional data collectedto identify the effect of residential schooling, I go on to present empirical evidence for thischannel of identification.613.5. Additional Data3.5 Additional DataIn addition to the APS, I use several other data sources to construct the variables dis-cussed in Section 3.4. The variables I construct include the distance of a community tothe closest residential school, ?j, an indicator of whether the closest residential school wasopen when an individual was of schooling age, zjt, the proportion of Catholics surroundinga community in 1941, wjt=1941, and the national residential school enrollment rates ?t. Iuse information from the Aboriginal Healing Foundation on the dates of closure, openingand location of different residential schools across the country.70At most, sixty-two schoolsare included in the analysis. There are a number of reasons why this number of schools isused. First, a significant number of the schools that existed were in the territories and afew in the eastern provinces. Second, many of the schools closed before the time the indi-viduals in my sample were of schooling age. Third, many schools often listed separately arein fact geographic and religious continuations of each other and thus I do not count themas different entities. Finally, at times schools of different religious affiliations existed in thesame area and I choose only the closest school. This sample of schools is further restrictedbeyond the sixty-two schools because I limit attention to those communities with residentialschools that closed before 1965. Restricting the analysis to these schools makes the empir-ical analysis more consistent with the theoretical framework discussed above and does nothave a qualitative impact on the results. After 1965 the federal government began to take70These dates and locations can be found at http://wherearethechildren.ca/en/about/ahf.html.Last Retrieved September 29, 2012. If the school was transferred to the band or group ofbands before the school was ultimately closed, the date of transfer is given instead of thedate of closure. Transfers started to occur in 1970 with Blue Quills residential school inAlberta. Although 23 proposals for transfers of residential schools to bands were made,ultimately only 5 followed Blue Quills, all in Saskatchewan (Milloy 1999, 237). By the1980s only a dozen residential schools operated by bands were left in existence with oneschool operated by government at band request; gradually only a few remain, with the lastgovernment-run school closing in 1996, and the last band-run in 1998 ( Aboriginal HealingFoundation 2009, p.176). These are certainly not the only dates of opening and closure ofresidential schools that could be used. To obtain actual dates of closure of the schools I userecords compiled by the General Synod Archives of the Anglican Church. I have also runspecifications which use the dates of federal government involvement used in the ResidentialSchools Settlement Agreement. These dates are not used as the main specifications becausemany of the dates extend well past the time when the schools resembled the historicalresidential schooling system.623.5. Additional Dataover residential schools from the churches and the Aboriginal people began to acquire moreauthority in the education of their children. In addition, around this time residential schoolsbegan to become more ambiguous in nature with some acting solely as hostels rather thanschools. Restricting the analysis to schools that closed pre-1966 makes understanding whatit means to attend residential school more straight forward. In addition in specificationsthat include band (community) fixed effects, all bands with sample sizes less than 40 areexcluded to ensure credible estimation.I use census sub-divisions (CSD) as the geographic definition of a community. Combiningdata on the coordinates of CSDs provided by the Environmental Systems Institute withseveral provincial data sets from the Canadian Atlas Map Bundle on Canadian cities andtowns allows schools to be matched with communities. Residential schools are matched tocities/towns and then ARC GIS is used to locate the closest residential school to a givencommunity.71Then, the closest residential school to each CSD is chosen using as the crowflies distance from the center of the CSD.72This distance is used as the main distancemeasure ?j. By construction, all communities are tied to some residential school.I construct zjt  the indicator whether a community's closest residential school was openwhen a given cohort was of schooling age  in the following way. For each cohort in eachcommunity zjt = 1 if a school was open when a given cohort would have been affected bythe compulsory school attendance laws. Otherwise, zjt = 0. Before 1945, the mandatoryschool attendance ages for Aboriginal children were defined through the Indian Act.73To beeligible for the Family Allowance implemented in 1945, parents had to comply with provincialschooling laws. Thus, mandatory ages are defined to comply with both federal and provinciallegislation after 1945. The provincial schooling ages and their changes over time after 1945is taken from Riddell and Song (2011).7471The only schools included in the match are those that existed in 1928 or later since it isthe meaningful time frame for my sample. Using these files, the latitude and distance fromthe closest city are also calculated.72Census sub-divisions in the context of the Aboriginal population include Indian reserves,Indian settlements and unorganized territories (Statistics Canada 2003).73In 1920, the mandatory ages for school attendance were seven to fifteen. In 1930, therewas an amendment to the Indian Act to extend the mandatory ages to sixteen.74Riddell and Song (2011) expand upon the initial data collected by Oreopoulos (2006).633.5. Additional DataIt is important to understand what the closure of the closest residential school implies.If the closest school to a community closes - zjt changes from one to zero  it impliesthat the cost of enforcement has increased for the federal government. This implies thegovernment will have a weaker incentive to enforce attendance at residential school and thusfewer children will attend. However, their still may be a subset of children who are forcedto attend residential school despite its increased cost to the government. These children willattend a residential school further away. In addition, if the closest residential school wasCatholic and a child's parents were Anglican, the closure of the closest residential schoolwould not affect their attendance. As a consequence, the closure of the closest residentialschool does not induce attendance to drop to zero.75To construct a measure of historic religious infrastructure surrounding an Aboriginalcommunity, I visually match 1941 census divisions and sub-divisions to their 1991 coun-terparts using the division maps from the 1941 and 1991 Census. At times this involvesreconstructing the 1991 divisions using subdivisions in the 1941 data. Once comparablegeographic regions are constructed, I use the 1941 census population counts to constructthe proportion of non-aboriginal people in a division that are Catholic in 1941. This giveswjt=1941. To estimate changes in national policy, ?t, I use the national downward trend inresidential school attendance from 1928 to 1966 from the Canada Year Book (1940-1970)and past 1966 I construct this measure using the proportion in each cohort that attendeda residential school in the 1991 APS.76Using historic geographical variation combined withnational trends as exogenous variation is reminiscent of strategies used in the local labourmarket literature, specifically based on the work of Bartik (1991) and Blanchard and Katz(1992).The process of mapping individuals outside of Aboriginal communities to their origincommunity is more involved. Although the APS does not specify where an individual was75Tests for a structural break in residential school attendance suggests the decrease inattendance after closure is 21.4 percent with a standard error of 6 percent. Tests indicatethat there is no statistical trend in attendance before school closure, with the pre-closureslope coefficient of attendance equal to -0.01 with a standard error of 0.01.76Specifically for each year I use the proportion of children who would have been seven inthat year minus the proportion who attended a residential school at the system's peak in1934.643.6. Resultsborn, it does specify what band an individual belongs to. More than half of these bandshave a legally defined land base. A large fraction of these land bases link uniquely to oneor two CSDs. Using Aboriginal Affairs and Northern Development Canada's (AANDC)legal-linkage files of bands to CSDs, I can reconstruct individuals' origin communities. Ifa band is linked to more than one possible sub-division, I use the 1991 Confidential LongForm Census files to estimate the probability of being from each of these CSDs. I thenmatch individuals who currently reside outside one of the previously specified CSDs to therelevant area using the estimated probability distributions. A more thorough discussion ofthis process is given in appendix D.Although the APS is intended to be a representative sample of all Aboriginal communities,it will inherently have more noise for small communities than large ones. The ConfidentialLong Form 1991 Census files partially overcomes this problem and includes non-aboriginalpeople in the sample. It is not used for the main analysis since it does not have informationon residential school attendance. Use of this sample allows me to compute the intent to treatfor a larger set of Aboriginal communities and also allows me to construct a falsification testdiscussed in Section 3.6.1.3.6 Results3.6.1 Main ResultsOne standard piece of evidence often provided in favor of random assignment is to showbalance in observable covariates. Assuming observable variables are randomly drawn from asample of total characteristics, observable variables being uncorrelated with the instrumentis suggestive evidence that unobservable variables are also uncorrelated with the instrumentand thus assignment to treatment is effectively random. Since the argument presented hererequires conditional exogeneity, I regress the variables used as exogenous variation on a setof observable characteristics conditional on cohort and band fixed effects. The results fromthis exercise can be found in Table 3.2. While there is some evidence that the observables arecorrelated with opening and closure, there is far less evidence that they are correlated with653.6. Resultsthe Bartik style instrument once band fixed effects are accounted for. This provides somesuggestive evidence that the 1941 Proportion of Catholic multiplied by the national trend inresidential school attendance, ?twjt=1941, is not significantly correlated with unobservables.Of course, this is far from proof of the exclusion restriction, but it is at least reassuring.In addition, it should be noted that the proportion of Catholics in 1941 multiplied by thenational trend in attendance enters the probability of being open (zjt) negatively as predictedby the model.77Table 3.3 reports the results from the bivariate probit model described by Equation 3.3using high school graduation as the outcome equation. The first panel uses the indicator ofthe school being open when the individual was of schooling age (zjt and ztj?j) as the exclusionrestriction. The second panel uses the Bartik style variation, ?twjt=1941, as the exclusion re-striction. The upper panel reports the estimated coefficients from the high school graduationdetermination equation (the ??s in Equation 3.3) and the lower panel contains the resultsfrom the residential school attendance equation (the ??s in Equation 3.3). The specificationsdiffer in the geographical controls included in the sample.78First, the effect of residentialschool attendance on high school graduation is large, positive, and statistically significant inall specifications. Being female is positively associated with high school graduation, whiledistance from a major city and latitude are negatively correlated. The second panel presentsthe results from the attendance equation. Conditional on distance to the closest major city,latitude, solely Aboriginal ethnic origin, gender, birth cohort, and geographic fixed effects,the excluded instruments are significantly correlated with residential school attendance. Ap-pendix B, table B.1 reports the first stage in a linear probability model. The F-statisticsfor the instruments are all above ten controlling for band fixed effects.79This is suggestive77The intuition is the following: if there is a higher proportion of Catholic individuals in aregion, there will be more religious infrastructure and thus the value of a residential schoolto the Catholic Church is lower and thus the school is more likely to close.78Column (1) is the full sample with provincial effects, while Column (2) includes all bandsover sample size 40 with 108 band fixed effects 108. The reason for these restrictions withthe various geographic controls is due to convergence of the likelihood function and credibleestimation of the time invariant effects. The incidental parameters problem is not encoun-tered in this context since the asymptotics are taken to be with respect to large N ratherthan large J or T (Neyman and Scott 1948).79If I include the full sample rather than just communities whose schools closed before 1965,then the F-statistics meets the more stringent criteria of Stock and Yogo (2003). However,663.6. Resultsevidence that the model is well identified in the absence of functional form restrictions.Table 3.4 reports the average treatment effect (ATE) and effect of the treatment on thetreated (ATET) for a set of human and cultural capital outcomes and contains the mainresults of this chapter. The treatment is whether or not an individual attended a residentialschool. I calculate the ATE by first predicting the probability each individual in the samplewould have a particular outcome if they were and were not to attend a residential school. Ithen average the predicted outcome probabilities if individuals were to attend a residentialschool and if they were to not attend a residential school. Finally I subtract one from theother which yields the ATE. The ATET is calculated similarly, but the sample is limited toonly those individuals who actually attended a residential school.All estimations include the same control variables as in Table 3.3 and band fixed effects.Table B.2 contains the results with province fixed effects and the full sample with similarresults. The first panel contains the results from the specification that uses the Bartik styleinstrument as exogenous variation, while the second panel uses the open indicator and itsinteraction with distance. The first column reports an estimate of the ATE that does notaccount for selection into residential school attendance (i.e. assuming ? = 0).80 The secondcolumn reports an estimate of the ATE that takes into account selection into residentialschool attendance. The third column reports the ATET and the final column reports thecorrelation between residential school attendance and the outcome listed on the left mostcolumn (?).The first thing to note is that estimates of the ATE in the first and second columnare notably different. In fact the estimate of the ATE is actually of the opposite signin the two columns. This implies accounting for the selection of children into residentialschool attendance is of substantial importance. If one were not to account for selection intoattendance (as in the first column), one would conclude that residential schooling had little orpositive effects on cultural outcomes and negative effects on economic outcomes even thoughit should be noted that the conditions for valid estimate in a bivariate probit specificationare not necessarily the same as in linear IV regression.80This is estimated using the marginal effects of a univariate probit for each outcomespecified.673.6. Resultsit increased high school graduation rates. The correlation coefficients reported in the fourthrow in each panel clearly demonstrate that the selection of children into residential schoolis positively correlated with cultural outcomes, but negatively correlated with economicoutcomes. In addition, the effects of residential school attendance are large. The moreconservative specification suggests residential school decreased the likelihood of receivinggovernment transfers by nearly 15 percent for those that attended. Given the proportionof individuals receiving government transfers in this group, this implies residential schoolsdecreased government transfer receipts by 40 percent. Attendance at a residential schoolalso dramatically increases the likelihood of high school graduation and employment.However, this table also highlights the cultural implications of residential schooling. Con-servative estimates suggest the the percentage of people who speak an Aboriginal languageat home was reduced by nearly 30 percent due to residential schools.81The reduction in par-ticipation in traditional activities due to residential schooling was about 12 percent for thosethat attended. This implies that the proportion of people in this group who participated intraditional activities would have been nearly double were it not for residential school. Theimpacts on sociological outcomes are less clear and given in appendix B in Table B.3.While the results above are interesting, there are a few concerns about their interpreta-tion. First, individuals may not honestly report whether they attended a residential school,and, even if they do, the residential schooling question is inconsistent between cohorts whichmay bias estimates of the ATE and ATET. Second, although the APS was designed bea representative sample of all Aboriginal communities, the cost of running this extensivesurvey in remote areas may result in their under-representation. Third, spill-overs from in-dividuals attending residential schools will bias the estimates of the ATE and the ATET. Toovercome these problems I use the 1991 Confidential Long Form Census files and estimatethe effect of the intention to treat. Specially, I measure the effect of having an individual'sclosest residential school open during his or her schooling years. I then adjust this effect81I arrive at this number since approximately 20 percent of individuals who attended aresidential school currently speak an Aboriginal language at home. The most conservativeestimates show this would have been at least 27.6 percent according to estimates of theATET in Table 3.4.683.6. Resultsfor how far away this residential school is. This larger data set includes more Aboriginalcommunities and this methodology captures any spill-over effects that residential schoolingmay have while avoiding problems with non-random measurement error in residential schoolattendance.Table 3.5 presents the results of this exercise. The row labeled open contains my mea-sure of the intention to treat for each outcome listed in the top row.82The row labeled open? distance shows the decrease in this effect for every ten kilometers of distance between theresidential school and the center of the community of interest (zjt?j). The impact on highschool graduation and employment is positive and significant and impact on the likelihoodof receiving government transfers is negative and significant, as before. There is also a 14percent log point increase in the average log wage, but the impact on total weeks workedis insignificant. On the other hand, there is a negative impact on both the likelihood ofliving on reserve or speaking an Aboriginal language at home. Information on participationin traditional activities are unavailable in the census. To obtain an estimate of the ATEfrom the intention to treat, one merely divides the intention to treat by the percentage ofindividuals induced to attend residential school via the instrument (the compliers). Thisis approximately 20 percent. Doing this yields a larger estimate of the ATE than thosecalculated in Table 3.4 which suggests one of several things. First, there may be spill-overeffects from individuals attending a residential school. Second, smaller Aboriginal commu-nities missed in the APS may be more intensely impacted by residential schools. Finally,there may be misreporting of residential school attendance which biases the estimated effecttoward zero. These results reinforce the findings of Table 3.4: residential schools increasedeconomic integration at the expense of cultural connection.To offer further support of the exclusion restriction we can perform this same exercise forgeographically adjacent non-indigenous cohorts. If opening and closing are not correlatedwith changes in local conditions reflected in the outcomes of non-indigenous people (suchas employment conditions or educational opportunities) then we should observe no effect of82Which is measured by the effect of a community's closest residential school being openwhen an individual was between the legally mandated schooling ages (zjt).693.6. Resultsan intention to treat as measured above on non-indigenous people. Table 3.6 estimatesthe intent to treat for non-aboriginal people who lived in the census divisions where theAboriginal communities of interests are located. In nearly all specifications, estimates of theintent to treat are small and insignificant. In the one case it is significant the effect is of theopposite sign to that in Table 3.5. This supports the conclusion that, to the extent changingconditions among the non-aboriginal and Aboriginal people are correlated, changing localconditions are not driving the findings in Table 3.5.Another concern regarding the results above regards the matching procedure. Thoseoff reserve had to be matched back to their origin communities using information onband membership and the geographical association of bands. There is necessarily errorin this matching process due to changes in band names over time and misreporting of bandinformation. Consequently, both of the instruments may be more weakly associated with theoff reserve population than the on-reserve population. To address this Table 3.7 splits thesample by those who had to be matched back to an origin community and those who alreadylived in one. Table 3.7 demonstrates that the effect of residential schools is similar both onand off reserve although the estimates for the off reserve population are less precise. Thisimplies the economic effects of residential schools were not driven by leaving the reservation:an individual can still retain traditional cultural connections (as represented by living onreserve) and engage in the labour market. This suggests that although residential schoolingdid result in migration off reservations, communities could still potentially benefit from theformal educational consequences of the schools.3.6.2 MechanismsThe effects identified above are a bundled effect of several treatments. For example, theeffect of attending a residential school is relative to a number of counterfactuals: attendinga federal Indian day school, a provinical non-Indian school, not attending school, and livingat home versus not living at home. As a consequence it is difficult to draw general economiclessons from the estimates above. To narrow down the set of plausible counterfactuals I limitmy sample in a number of ways. First, I limit my sample to individuals who have a grade703.6. Resultsten education at minimum. By definition, all members of this subgroup had access to highschool and by focusing on them it is possible to identify whether the large increase in highschool graduation associated with residential schooling was solely due to access. Table 3.8demonstrates that residential schools' impact on high school graduation was not throughthe purely mechanical channel of accessibility. Panel (1) includes the full sample, all thecontrol variables as in previous specifications and provincial fixed effects. Panel (2) includesall individuals in bands that have a sample size greater than 40, all the control variablesin previous specifications and band fixed effects. It becomes immediately apparent that theeffect of residential school attendance on those that attended is not due to solely to access.In the preferred specification (with full band fixed effects and the Bartik style instrument),none of the effect of residential schooling is accounted for through this channel.Since the effect of residential schooling doesn't seem to be acting solely though access,what is the effect driven by? Is it due to isolating children from their communities, or becauseof some other characteristics of the system? I address this question by using another sub-sample from the APS. Those under the age of forty-nine were asked where they lived duringschool, specifically, whether they lived at a residential school, at home, with an Aboriginalfamily other than their own, a non-aboriginal family, or somewhere else.83I limit thesample to only those people who attended school while staying away from home (with anon-aboriginal family, somewhere else, or a residential school) and re-estimate the model.I then compare these results to the results obtained using all individuals in the under-49sample. If the type of education children received did not vary between the two samples,I could then infer that the difference in the results was not due to differences in quality ofschooling, but rather due to home/community environment. Note that if a child attendedhigh school during this time period from home, it was likely a public school. Figure 2.4shows the proportion of total and high school enrollment that federal schools accounted forsince 1958. We can see from this figure that public integration continued systematically83Given the historical context somewhere else would have likely been a foster home orplausibly another form of de-segregated boarding school. Children often lived with non-aboriginal families at private boarding homes while attending schools or with foster parentsif they had been removed from their homes.713.6. Resultsover this period by the downward trend in the Federal School line. It also shows thatmost federal schools that serviced high school students were residential schools. The vastmajority of high school students who were not attending a residential school attended apublic school. This is indicated by the dotted Federal High School line almost overlappingthe Residential High School line. This demonstrates that the children who stayed at homewere likely receiving the same high school education as children who went away. If therewere not extreme differences in the ease of access to high schools between these children,the differences in the main results and these results can be interpreted as being due to thehome/community environment. Table 3.9 shows the results from estimating the model withinthis sub-sample. In this sub-sample, residential schools played a relatively minor role in highschool graduation rates, increased the likelihood of receiving government transfers, increasedthe likelihood of living on reserve and increased the likelihood of speaking an Aboriginallanguage. These results suggest that residential schools' primary effect was through removalfrom the home/community environment.84Manual labour, Religion and AbuseIn the model discussed in Section 3.4, time away from the family is equivalent to timespent in human capital accumulation. However, if there is a wedge between time spent inhuman capital accumulation and time away from the family, the model would predict lesshuman capital accumulation and equivalent cultural accumulation. Before 1951, residentialschool students spent half the day in manual rather than academic activities. The modelpredicts children who attended residential school before 1951 would accumulate less humancapital than those that attended later, while they should have equivalent cultural outcomes.The first half of the prediction is borne out in Table 3.10. The effect of attending a residen-tial school on high school graduation is lower before 1951, but contrary to the prediction ofthe model, cultural outcomes also differ. For example, participating in traditional activities84This result should be interpreted with caution. It does not imply there was inherentlysomething wrong with or inadequate about Aboriginal homes. This effect may be due topoverty (which could also occur in a non-aboriginal home) or due to the parent's relationshipwith the school relative to the boarding home.723.6. Resultsand Aboriginal language use is significantly more affected by residential schooling before1951. This may suggest that policies designed to eliminate traditional cultural connectionswere more intense during this period. This finding highlights the fact that if there is het-erogeneity in student treatment over time and schooling that influences the rate of humanand cultural capital accumulation (q and ?), the effect of residential schooling will differ.I explore two other avenues that may significantly impact the rate of human and culturalcapital accumulation: the religion of the school and whether the school had an abusiveenvironment.The history of residential schools is tightly connected with emotional, physical and sexualabuse85, but how it varied among schools, time periods and affected individual's outcomes isunexplored. To explore these issues, I have obtained data from Aboriginal Affairs and North-ern Development Canada on the number of individuals who attended each residential schoolin each decade and data from the Indian Residential School Adjudication Secretariat on thenumber of approved abuse claims by school and decade under the Independent AssessmentProcess. This allows me to construct the proportion of individuals who were abused in eachschool in each decade. If a school in a given decade had five or fewer cases of abuse, theIndependent Assessment Process has censored the count to be zero. I assume all decades andschools not in the sample had five individuals who were abused to account for this censoring.Then I scale up the number of abuse cases reported in each school-decade to account for thefact they only represent approximately 30 percent of the total expected. This gives results inan estimate of the expected proportion of children abused by school and decade. The numberof approved cases to date is 8,960, with a total of 29,000 expected to apply. Although thelevels of abuse calculated here are obviously an underestimate, the Adjudication Secretariatbelieves this sample is representative of the final distribution of abuse cases.However, one feasible scenario that may lead to the sample distribution inaccuratelyrepresenting the final distribution would be abuse claims occurring in a cascade. Specifically,if as soon as some threshold number of victims in a cohort and community make an abuse85A non-exhaustive list of references include Report of the Royal Commission of AboriginalPeople 1996; 1991 the Cariboo Tribal Council 1991; The Nuu-Chah-Nulth Tribal Council1994; Miller 1996; Milloy 1999; Smith 2009; and Dawson 2012.733.6. Resultsclaim, the remaining victims also do, the sample distribution would be unrepresentative.However, as long as reaching the threshold number of abuse cases is not correlated withthe socio-economic outcomes of interest then this would only result in attenuation bias. Ifan effect of abuse is found, then we can assume it is a lower bound. It should be notedthat the process of filing an abuse claim is completely confidential and all claims undergo anexamination process regarding their plausibility. Thus, it is feasible that individual abuseclaims are not inherently correlated with each other within a cohort or community. Inaddition the process of needing to validate claims acts as a disincentive to filing a falseclaim. If anything, the application process and extensive examination of the claims wouldresult in under-reporting.Two things should be noted regarding the possible cohort patterns in abuse reporting.First, one might suspect that older generations would be less likely to report any abuse (oreven recognize some forms of physical maltreatment as abuse  such as being hit with aleather strap) even if there was more abuse present during the era they attended school.In the data, there does appear to be a correlation between the number of abuse cases andthe decade of attendance: individuals who attended residential schools in later decades aremore likely to claim severe forms of sexual abuse. However, any cohort trends that maybe correlated with the socioeconomic outcomes of interest are taken into account by cohortspecific fixed effects in all of the specifications.Second, the main results presented in the last sections were only for a sub-sample ofschools that closed relatively early. To the extent that staff in schools that closed before1965 date were more abusive than staff in schools that closed after, the estimates in the lastsection will under-estimate the positive economic effects of residential school. In this section,I include all schools that I have abuse information for.Table 3.11 contains descriptive statistics for the proportion of students abused, calculatedfrom the data sources described above. The first column contains the proportion of studentswho have successfully filed an abuse claim over the total proportion of students who attendedthat school in that decade. The second reports the scaled estimates. The table reportsthe mean, median, 95th percentile, and maximum proportion of individuals who have filed743.6. Resultssuccessful abuse claim or who are expected to do so. The table indicates that abuse outcomesare highly skewed. The vast majority of schools in most decades have five or fewer cases ofabuse. However, the results from some decades are dramatic, with the greatest proportionof children who have filed successful abuse claims reaching 44 percent.I link this information to schools and decades in the 1991 APS. This provides a measureof the abusiveness of an environment in each school and decade. Schools are linked toindividuals through their proximity to communities and the decade an individual was sevenyears of age. Merging the data on abuse with the APS results in some loss of information.I do not include all schools in all decades for the reasons discussed in the data section.In addition, in the original analysis I included schools that were not covered under theIndependent Assessment Process.86I then construct two indicators of whether a school hadan abusive environment in a given decade. I count a school-decade as having an abusiveenvironment if the school-decade was in the 95th or 99th percentile of school-decade abuseproportions. This corresponds to proportions between 8 and 14 percent.I explore heterogeneity in the effect of residential school attendance in Table 3.12. Table3.12 reports the marginal effects of residential school attendance (given in the row labeledattendance), whether the school had an abusive environment (the two different percentilesinteracted with attendance) and the school's religion (Anglican, Methodist or Presbyterian:Catholic schools are used as the comparison group). Opening, closing and distance are usedas exclusion restrictions and all specifications include latitude, gender, distance from closestcity, an indicator for only Aboriginal ancestry, birth cohort and provincial fixed effects. Allprevious conclusions regarding the effects of residential school attendance are unchanged.Presbyterian schools perform worse than Catholic schools in economic outcomes, but aremore likely to have their former students marry. Methodists and Anglican schools do notseem to have different effects than Catholic schools. The effect of attending a school in the95th percentile of school-decade abuse proportions is insignificant in most cases. On theother hand, the impact of attending a residential school in a decade with abuse proportions86Not all residential schools were included under the Independent Assessment Process. Onlythose decades in which the federal government had involvement with a school are coveredunder the Residential Schools Settlement Agreement.753.7. Conclusionin the 99th percentile has substantial effects on outcomes. The likelihood of marriage andemployment is substantially decreased while the likelihood receiving government transfersincreases substantially. Attending a school with a relatively abusive environment completelywipes out any positive economic effect of residential schooling. It should be noted that thereported specification only includes provincial fixed effects. Estimations of a full set of bandfixed effects creates difficulties with convergence, so these results should be kept with thisqualification in mind. However, it becomes clear from Table 3.12 that abusive environmentsand the religion of the school are not main drivers of the results.3.7 ConclusionThis is the first analysis to offer causal evidence on the long term consequences of forciblechild removal and residential schools. To circumvent data limitations and the non-randomselection of children into residential schools, I focus on a Canadian case study and usetwo connected sources of exogenous variation. The first source is geographic variation inthe times of school opening, closing and changes in school distance. The second sourcesuses Bartik style variation that exploits policy-driven national changes in residential schoolattendance and its interaction with historic regional variation in the non-indigenous religiouscomposition.I find that accounting for selection into residential schools is extremely important. Myresults suggest that attendance at an Indian residential school improved economic and ed-ucational outcomes but reduced the likelihood of participating in traditional activities andspeaking an Aboriginal language at home. The evidence suggests that residential schoolsimpacted these outcomes primarily by removing Aboriginal children from their communities.Thus, the analysis may be informative regarding child removal policies more generally.However, it is important to understand all the results presented here are conditional onresidential schools already existing for a number of generations. The intergenerational effectsof residential schooling are addressed in the following chapter.763.8. Tables and Figures3.8 Tables and FiguresBelow are the tables for Chapter 3.Table 3.1: Descriptive Statistics By Residential School AttendanceAttended Did Not AttendFemale 0.547 0.518(0.012) (0.009)Single Ethnicity 0.926 0.705(0.007) (0.009)Latitude 52.393 52.631(0.060) (0.043)Age 40.139 30.759(0.286) (0.131)Distance to City (KM) 109.19 116.85(0.205) 0.151High School Graduate 0.477 0.487(0.012) (0.009)At Least Grade 5 0.695 0.766(0.011) (0.006)Receive Government Transfers 0.372 0.274(0.012) (0.007)Employed 0.394 0.492(0.012) (0.009)In Aboriginal Community 0.559 0.348(0.013) (0.006)Participate in Traditions 0.147 0.068(0.009) (0.004)Number of Observations 5460 16999Number of Schools 62Notes: Standard errors are reported in parentheses. The variable labeled Single Ethnicity is a oneif an individual claimed they only has Aboriginal ancestry and zero otherwise. The variable labeledParticipate in Traditions equals one if an individual saw a traditional healer or participated inhunting, gathering, dancing and other traditional recreational and religious activities in the pastyear.773.8.TablesandFiguresTable 3.2: What Determines Whether a School is Open When an Individual is of Schooling Age?Dependent VariableOpen Open ?Distance 1941 Prop Catholic ? TrendCovariates (1) (2) (1) (2) (1) (2)Female 0.126 -0.034 -0.741* -0.280 -0.003* 0.001(0.074) (0.071) (0.312) (0.236) (0.001) (0.001)Single Ethnicity -0.025 0.395** 0.104 -0.148 -0.004* -0.003(0.125) (0.151) (0.565) (0.526) (0.002) (0.001)non-aboriginal Emp Rate -0.168 0.480* 1.585** 1.091** 0.002 0.002(0.144) (0.196) (0.544) (0.397) (0.002) (0.001)Latitude -0.157*** -0.460*** 2.388*** 3.175*** 0.002*** 0.000(0.016) (0.132) (0.086) (0.476) (0.000) (0.001)KM (10) to City 0.038*** 0.016 0.296*** 0.203** 0.001*** 0.001***(0.004) (0.024) (0.026) (0.064) (0.000) (0.000)1941 Prop Catholic ? Trend 0.232 -3.756*** -51.060*** -13.015*** - -(0.717) (0.688) (3.859) (3.886) - -N 11460 10271 11460 10271 11460 10271Notes: Standard errors are reported in parentheses and are clustered by band and three year cohort.. Panel (1) is the full sample with province effects, while Panel (2) includesall bands with at least 40 people. The total number of bands in this specification is 108. Both specifications include a full set of cohort fixed effects. The reason for theserestrictions with the various geographic controls regards convergence of the likelihood function and credible estimation of the time invariant effects. The dependent variable isindicator variable called open which is equal to one for an individual if the closest residential school to a community was open when they of the legally mandated schoolingage (which depend on federal and provincial legislation). The asterisks indicate the level of significance: * p<0.10, ** p<0.05, *** p<0.01.783.8. Tables and FiguresTable 3.3: Residential School Attendance and High School Graduation: Coefficient Estimates fromthe Bivariate Probit ModelSchool Open and Distance 1941 Prop Catholic ? Trend(1) (2) (1) (2)High School GraduationAttendance 0.704* 0.604*** 0.447 0.573***(0.365) (0.191) (0.428) (0.207)Female 0.257*** 0.207*** 0.269*** 0.207***(0.079) (0.055) (0.079) (0.057)Latitude -0.043*** -0.130* -0.043** -0.128*(0.016) (0.067) (0.017) (0.068)KM (10) to City -0.014*** 0.006 -0.015*** 0.006(0.005) (0.011) (0.005) (0.012)Single Ethnicity -0.762*** -0.733*** -0.738*** -0.730***(0.128) (0.126) (0.136) (0.131)AttendanceOpen 0.374*** 0.684***(0.102) (0.105)Open?Distance -0.036*** -0.041***(0.007) (0.007)1941 Prop Catholic ? Trend -1.573 -4.229***(1.051) (1.011)Female 0.140** 0.045 0.147** 0.064(0.064) (0.047) (0.065) (0.049)Single Ethnicity 0.521*** 0.525*** 0.493*** 0.522***(0.145) (0.126) (0.150) (0.128)Latitude 0.011 -0.297*** 0.001 -0.249***(0.016) (0.077) (0.017) (0.077)KM (10) to City 0.001 -0.029** -0.005 -0.034***(0.005) (0.019) (0.005) (0.012)Birth Cohort Fixed Effects X X X XProvincial Fixed Effects X XBand Fixed Effects X XCorrelation -0.254 -0.234** -0.097 -0.222*(0.237) (0.115) (0.256) (0.125)N 11460 10271 11460 10271Notes: Standard errors are reported in parentheses and are clustered by band and three year cohort. Columns labeled (1)include the full sample with province effects. Columns labeled (2) includes all bands over sample size 40 with band fixed effectswith a total of 108 bands. The reason for this restriction with the band fixed effects regards convergence of the likelihoodfunction and credible estimation of the time invariant effects. All regressions include latitude, gender, distance from closestcity, an only Aboriginal ancestry indicator, birth cohort fixed effects, the geographical effects specified. The first set of panelsincludes the open indicator and the distances to the school in the attendance equation while the second set of panels includes1941 Proportion Catholic in individual's subdivision ? (average attendance in that individual's cohort - average attendance incohort at peak 1934). The row titled correlation contain the estimate the correlation of the error terms between the highschool graduation equation and the residential school attendance equation. It can be understood as a summary statistic for theextent of unobservable selection bias. The asterisks indicate the level of significance: * p<0.10, ** p<0.05, *** p<0.01.793.8.TablesandFiguresTable 3.4: The Impact of Residential School on Economic Outcomes(1) (2)Outcomes ATE (? = 0) ATE (? 6= 0) ATET (? 6= 0) Correlation (?) ATE (? 6= 0) ATET (? 6= 0) Correlation (?)HS Graduation 0.070*** 0.181*** 0.179*** -0.222* 0.191*** 0.188*** -0.243**(0.023) (0.044) (0.065) (0.125) (0.044) (0.066) (0.115)Government Transfers 0.001 -0.158*** -0.210*** 0.410*** -0.117*** -0.153*** 0.289**(0.057) (0.042) (0.052) (0.134) (0.036) (0.045) (0.134)Employed -0.045** 0.121*** 0.145*** -0.346** 0.140*** 0.172*** -0.398**(0.023) (0.043) (0.065) (0.175) (0.047) (0.070) (0.159)In Aboriginal Community -0.021 -0.187*** -0.253*** 0.555*** -0.085*** -0.113 0.207(0.019) 0.062 (0.132) (0.196) (0.038) (0.073) (0.141)Participate Traditional 0.008 -0.078 -0.129* 0.552*** -0.057 -0.092* 0.412***(0.020) (0.064) (0.082) (0.147) (0.050) (0.066) (0.151)Aboriginal Language -0.021*** -0.053** -0.076** 0.141 -0.067** -0.095** 0.208*(0.007) (0.030) (0.044) (0.136) (0.031) (0.045) (0.123)Source of Variation 1941 Prop Catholic ? Trend (?twjt=1941) School Open (zjt) and Distance (zjt?j)N ~10271Notes: The columns titled ATE (? = 0) contain the univariate probit marginal effects. The columns titled correlation contain the estimate of the correlation of the errorterms between the outcome equations, whose dependent variable is listed on the left hand side, and the residential school attendance equation. Standard errors are reported inparentheses and are estimated by the Huber Sandwich Estimator. The columns titled ATE and ATET contain estimates of the average treatment effect and the effect ofthe treatment on the treated respectively. It can be understood as a summary statistic for the extent of unobservable selection bias. Both of their standard errors are calculatedusing the delta method, are reported in parentheses and are based off standard errors clustered at the band-and three year cohort level. The first panel used 1941 ProportionCatholic in an individual's census division ? (average attendance in that individual's cohort - average attendance in cohort at peak 1934) in the attendance equation. Thesecond panel includes the open indicator and the distances to the school in the attendance equation as the exclusion restriction. All regressions include latitude, gender, distancefrom closest city, an only Aboriginal ancestry indicator, birth cohort fixed effects, and band fixed effects with 108 bands in total. All bands included have a sample size of 40.The sample size varies by the dependent variable and thus the number of observations, N , is approximate. The asterisks indicate the level of significance: * p<0.10, ** p<0.05,*** p<0.01.803.8. Tables and FiguresTable 3.5: Estimates of the Intent to Treat From the 1991 CensusOutcome High School Bachelor's Degree Government Transfers EmployedOpen 0.087*** 0.008 -0.048*** 0.052**(0.050) (0.005) (0.022) (0.024)Open X Distance -0.018*** -0.009 0.007** -0.006**(0.004) (0.009) (0.003) (0.003)N 35574 35574 35440 35568Outcome Total Ln Wages Total Weeks Worked On reserve Aboriginal LanguageOpen 0.140*** 0.001 -0.022 -0.042**(0.065) (0.037) (0.034) (0.011)Open X Distance -0.008* -0.004 -0.005 0.012***(0.004) (0.003) (0.004) (0.004)N 19657 20400 34959 34787Notes: The estimates reported are the probit marginal effects. Standard errors are reported in parentheses and clustered at theband-cohort level. All specifications include latitude, gender, distance from closest city, an only Aboriginal ancestry indicator,birth cohort fixed effects and band fixed effects with 458 bands in total. The asterisks indicate the level of significance: *p<0.10, ** p<0.05, *** p<0.01.813.8.TablesandFiguresTable 3.6: Estimates of the Intent to Treat For Those Who Are Ineligible: The Effect of Having the Closest School OpenOutcome High School Bachelor's Degree Gov't Transfers Employed Total Ln Wages Total Weeks WorkedOpen -0.059* -0.047 -0.011 -0.003 0.011 -0.003(0.031) (0.035) (0.040) (0.032) (0.025) (0.011)Open X Distance -0.003 -0.003 0.005 0.004 -0.007*** 0.001(0.003) (0.004) (0.004) (0.003) (0.002) (0.001)N 41320 41320 59748 59740 46779 51033Notes: The estimates reported are the probit marginal effects. Standard errors are reported in parentheses and clustered at the band-cohort level. All regressions includelatitude,gender, distance from closest city, an only Aboriginal ancestry indicator, birth cohort fixed effects, and census division fixed effects. The indicator open is equal to onefor an individual if the closest residential school to a community was open when they of the legally mandated schooling age (which depend on federal and provincial legislation).It is zero otherwise. Open ?Distance is this indicator times the distance from the closest residential school. The asterisks indicate the level of significance: * p<0.10, **p<0.05, *** p<0.01.823.8.TablesandFiguresTable 3.7: On and Off Reserve: Bivariate Probit Resultson reserve off reserve(1) (2) (1) (2)Outcomes ATE ATET ? ATE ATET ? ATE ATET ? ATE ATET ?HS Graduation 0.171*** 0.167*** -0.125 0.183*** 0.179*** -0.149 0.148*** 0.177** -0.221 0.168*** 0.199*** -0.265(0.028) (0.047) (0.135) (0.027) (0.047) (0.133) (0.055) (0.081) (0.254) (0.056) (0.083) (0.247)Gov't Transfers -0.084*** -0.085*** 0.133 -0.060*** -0.060*** 0.090 -0.155 -0.221** 0.647** -0.147* -0.207** 0.680***(0.028) (0.031) (0.159) (0.027) (0.030) (0.152) (0.123) (0.134) (0.262) (0.114) (0.126) (0.255)Employed 0.203*** 0.201*** -0.335 0.171*** 0.171*** -0.278 0.032 0.041 -0.399 0.053 0.066 -0.444(0.050) (0.070) (0.410) (0.040) (0.054) (0.313) (0.083) (0.101) (0.298) (0.089) (0.110) (0.283)Traditional -0.163* -0.205** 0.776*** -0.032 -0.037 0.149 -0.043 -0.071 0.612** -0.069 -0.118 0.396**(0.102) (0.113) (0.295) (0.037) (0.048) (0.201) (0.066) (0.096) (0.260) (0.108) (0.137) (0.201)Aborig Language 0.004 0.004 -0.143* -0.071*** -0.073** 0.062 0.008 0.012 0.011 -0.009 -0.014 -0.154(0.030) (0.043) (0.097) (0.027) (0.040) (0.114) (0.040) (0.070) (0.249) (0.038) (0.058) (0.196)Source of Variation ?twjt=1941 zjt and zjt?j ?twjt=1941 zjt and zjt?jBirth Cohort FE X X X X X X X X X X X XProvince FE X X X X X XCensus Division FE X X X X X XF-Stat in First 34.71 25.24 4.50 0.63N 8789 2671Notes: The columns titled ATE and ATET contain estimates of the average treatment effect and the effect of the treatment on the treated respectively. Both of theirstandard errors are calculated using the delta method, are clustered at the birth cohort-year level and are reported in parentheses. The columns titled ? contain the estimatethe correlation of the error terms between the outcome equations, whose dependent variable is listed on the left hand side, and the residential school attendance equation. Itcan be understood as a summary statistic for the extent of unobservable selection bias. All regressions include latitude, gender, distance from closest city, an only Aboriginalancestry indicator, birth cohort fixed effects, provincial fixed effects specified. The first set of panels that contain  ?twjt=1941 use the Bartik style variation for identification,while the panels that containzjt and zjt?j use the open indicator and the distances to the school in the attendance equation as the exclusion restriction. The asterisks indicatethe level of significance: * p<0.10, ** p<0.05, *** p<0.01.833.8.TablesandFiguresTable 3.8: Effect on High School Graduation Conditional on Obtaining At Least Grade 10: A Lower Bound on the Impact of ResidentialSchool on High School Graduation RatesSample Restriction: Conditional on Getting Grade 10Open and School Distance 1941 Prop Catholic ? Trend(1)Coefficient ATE (? 6= 0) ATET (? 6= 0) Correlation (?) Coefficient ATE (? 6= 0) ATET (? 6= 0) Correlation (?)HS Graduation 2.111*** 0.135*** 0.226*** -0.952*** 1.871*** 0.099** 0.164*** -0.592*(0.187) (0.024) (0.034) (0.228) (0.373) (0.059) (0.051) (0.334)(2)HS Graduation 2.055*** 0.112 0.185* -0.846*** 2.021*** 0.108 0.179** 0.890(0.299) (0.141) (0.119) (0.366) (0.299) (0.114) (0.102) (0.600)Notes: Standard errors are reported in parentheses and and clustered at the band-three year cohort level for the last set of panels and at the band-year cohort level for thefirst set of panels. The columns titled ATE and ATET contain estimates of the average treatment effect and the effect of the treatment on the treated respectively. Both oftheir standard errors are calculated using the delta method based off the standard errors clustered at the aforementioned levels above. The columns titled correlation containthe estimate the correlation of the error terms between the outcome equations, whose dependent variable is listed on the left hand side, and the residential school attendanceequation. It can be understood as a summary statistic for the extent of unobservable selection bias. Panel (1) is the full sample with province effects. Panel (2) includes allbands over sample size 40 with fixed effects for band. The total number of bands in this specification is 108. The reason for these restrictions with the various geographiccontrols regards convergence of the likelihood function and credible estimation of the time invariant effects. All regressions include latitude, gender, distance from closest city,an only Aboriginal ancestry indicator, birth cohort fixed effects, the geographical effects specified. All specifications include includes opening, closure and school distance inthe attendance equation in the left hand panel and the 1941 Proportion Catholic in individual's census division ? (average attendance in that individual's cohort - averageattendance in cohort at peak 1934). The asterisks indicate the level of significance: * p<0.10, ** p<0.05, *** p<0.01.843.8.TablesandFiguresTable 3.9: Being Away from Home: The Effect of Residential SchoolSample Restriction<49 < 49 & AwayOutcome ATE (? 6= 0) ATET (? 6= 0) Correlation (?) ATE (? 6= 0) ATET (? 6= 0) Correlation (?)HS Graduation 0.225*** 0.247*** -0.986*** -0.181* -0.199* 0.308*(0.007) (0.018) (0.765) (0.090) (0.101) (0.189)Government Transfers -0.229*** -0.281*** 0.779*** 0.089* 0.092 -0.009(0.001) (0.004) (0.158) (0.063) (0.079) (0.127)Employed 0.148*** 0.169*** -0.487*** -0.129* -0.134* -0.049(0.009) (0.023) (0.163) (0.083) (0.091) (0.222)Participate Traditional -0.079*** -0.109*** 0.575*** -0.114** -0.048* 0.305(0.006) (0.014) (0.181) (0.068) (0.033) (0.235)Aboriginal Language -0.218*** -0.309*** 0.875*** 0.167*** 0.177*** -0.450*(0.001) (0.002) (0.157) (0.010) (0.020) (0.234)In Aboriginal Community -0.345*** -0.430*** 0.845** 0.034*** 0.038*** 0.145(0.001) (0.001) (0.242) (0.011) (0.016) (0.237)N 9769 3899Notes: The columns titled ATE and ATET contain estimates of the average treatment effect and the effect of the treatment on the treated respectively. Standard errorsare reported in parentheses and are clustered at the birth cohort-band level and calculated using the delta method. The columns titled correlation contain the estimate of thecorrelation of the error terms between the outcome equations, whose dependent variable is listed on the left hand side, and the residential school attendance equation. It canbe understood as a summary statistic for the extent of unobservable selection bias. The panel labeled <49 includes all people who were asked there they lived while attendingschools, which is the less than 49 age group. The panel labeled <49 and Away includes only individuals who went to residential school, lived with a non-aboriginal family,or somewhere else. All specifications include latitude, gender, distance from closest city, an only Aboriginal ancestry indicator, birth cohort fixed effects, provincial fixedeffects. All specifications include includes open indicator and distance ? open as excluded instruments. The asterisks indicate the level of significance: * p<0.10, ** p<0.05,*** p<0.01.853.8.TablesandFiguresTable 3.10: Does the Effect Vary by Era of Attendance?OutcomeHigh School Government Transfers Employed Participate Traditional Aboriginal Language On ReserveATE 1950-1965 0.219*** -0.184*** 0.147*** -0.075*** -0.232*** -0.384***(0.013) (0.006) (0.015) (0.013) (0.002) (0.001)ATE Pre-1950 0.137*** -0.258*** 0.142*** -0.142*** -0.416*** -0.443***(0.033) (0.007) (0.041) (0.014) (0.002) (0.001)Notes: Standard errors are reported in parentheses and are clustered at the birth cohort-band level. All regressions include latitude, gender, distance from closest city, an onlyAboriginal ancestry indicator, birth cohort fixed effects, the provincial fixed effects. The attendance equation includes the open indicator and the distances to the school as theexclusion restriction. The columns titled ATE 1950-1965 and ATE Pre-1950 contain estimates of the average treatment effect for those who were aged seven after 1950 andthose who were seven before, and the effect of the treatment on the treated, respectively. They were estimated using a bivariate probit which allowed the effect of residentialschool attendance to vary between these two eras. Both of their standard errors are calculated using the delta method. The asterisks indicate the level of significance: *p<0.10, ** p<0.05, *** p<0.01.863.8. Tables and FiguresTable 3.11: Descriptive Statistics AbuseStatistic Using Sample of Abused Cases Scaled Sample of Abused CasesMean 0.03 5%Median 0.02 3%95th Percentile 0.09 15%99th Percentile 0.21 36%Max 0.47 78%N 434Notes: The proportion abused is calculated from data provided by Aboriginal Affairs and Northern Development on the numberof individuals who attended each residential school in each decade. In addition, the Indian Residential School AdjudicationSecretariat has provided the number of approved individuals who have applied for compensation for abuse under the IndependentAssessment Process by school and by decade. If a school in a decade has five or less cases of abuse, the school is counted ashaving zero cases of abuse. The number of approved cases I have been given access to at this point include 8,960 cases, withtotal number expected to apply at 29,000. The Adjudication Secretariat predicts that this sample will be representative ofthe distribution of abuse cases among schools and decades, but the level of abuse is obviously an underestimate. To obtaina reasonable estimate of the level of abuse, I assume all decades and schools where abuse was not reported in my sample tohave 5 individuals who were abused. Then I scale up all decades by 70 percent to arrive at the final estimates reported. Thisprocess will obviously entail error, but its ultimate effect on the estimates should not be affected since the measurements usedin estimation are based on rank in the upper tail of the distribution rather than scale. The first column is the proportion abusedby decade within the sample and the second column uses the scaled up numbers to construct the proportion.873.8.TablesandFiguresTable 3.12: Heterogeneity in the Effect of Residential Schooling: Religion and AbuseOutcomeATET of Interest HS Graduation Government Transfers Employed on reserve Participate Traditional Aboriginal Language MarriedAttendance 0.234*** -0.184*** 0.114*** -0.383*** -0.095* -0.356*** 0.074***(0.054) (0.044) (0.05) (0.163) (0.065) (0.286) (0.029)Abuse 95 Perc -0.033 0.005 0.219*** 0.120 -0.022 0.024 -0.097***(0.046) (0.065) (0.054) (0.152) (0.079) (0.431) (0.031)Abuse 99 Perc -0.03 0.251*** -0.482*** -0.145 0.101 -0.038 -0.29***(0.066) (0.077) (0.056) (0.253) (0.149) (0.652) (0.062)Anglican -0.017 0.006 0.043 0.028 0.009 0.027 -0.035(0.044) (0.06) (0.050) (0.153) (0.084) (0.414) (0.032)Methodist -0.073 0.024 -0.011 0.026 0.004 0.011 -0.032(0.043) (0.062) (0.05) (0.154) (0.082) (0.403) (0.032)Presbyterian -0.33*** 0.007 -0.168*** 0.176 0.041 0.195 0.093**(0.062) (0.07) (0.062) (0.17) (0.101) (0.526) (0.041)Notes: The marginal effects are reported on residential school attendance (given in the row labeled attendance), whether the school had an abusive environment interactedwith attendance, and whether the school was Anglican or Methodist, interacted with attendance. Standard errors are reported in parentheses, estimated by the delta methodand clustered at the band-cohort level. All specification include latitude, gender, distance from closest city, an only Aboriginal ancestry indicator, birth cohort fixed effects andprovincial fixed effects specified. The open indicator and distance ? open are included in the attendance equation as excluded instruments and their interaction with abuse andthe Anglican, Presbyterian and Methodist indicators. The asterisks indicate the level of significance: * p<0.10, ** p<0.05, *** p<0.01.88Chapter 4The Intergenerational Legacy of Residential SchoolNumerous Canadian authors suggest that residential schooling has led to a plethora of socialdysfunctions in First Nations87families and communities (Aboriginal Healing Foundation1999; Milloy 1999; Stonefish 2007; Chrisjohn, Young and Maraun 2006). The CanadianTruth and Reconciliation Commission on Indian Residential School has stated that ...[resi-dential school's] impact has been transmitted from grandparents to parents to children. Thislegacy from one generation to the next has contributed to social problems, poor health, andlow educational success rates in Aboriginal communities today, (Truth and ReconciliationCommission (TRC) 2010, 1). The policies of forcible child removal from Indigenous familiesin Australia have been perceived similarly: [f]or individuals, their removal as children andthe abuse they experienced at the hands of the authorities or their delegates have perma-nently scarred their lives. The harm continues in later generations, affecting their childrenand grandchildren, (Human Rights and Equal Opportunity Commission1997, 4).While the anecdotal evidence on the intergenerational effects of residential schoolingand Aboriginal child removal are generally consistent with this view (Ing 2000; Ing 1999;Brow, Rodger, and Fraehlich 2009; Claes and Clifton 2998; Haig-Brown 1998; Chrisjohn1991, Human Rights and Equal Opportunity Commission 1997), there is only one empiricalstudy regarding the consequences of parental residential school attendance I am aware of.Bougie and Sen?cal (2000) demonstrate a negative association with parental residential schoolattendance and child school performance. While it is possible this relationship is causal, thefindings may be subject to selection bias since the historical evidence and the results of87The term First Nations is often used to refer to a specific group Indigenous North Amer-icans. Typically, First Nations people are eligible to live on reservations and apply formembership in Indian bands.89Chapter 4. The Intergenerational Legacy of Residential Schoolthe previous chapter suggest that parents who attended residential school are not ex-antecomparable to parents that did not attend.In this chapter I assess whether parents attending a residential school affects their chil-dren's adult outcomes. The outcomes I look at include high school graduation, employment,social assistance use, speaking an Aboriginal language at home, divorce and the presence ofsocial support networks. I also investigate whether the direct effects of residential schoolingvary with family residential schooling history. I deal with non-random selection into res-idential school attendance by exploiting historical evidence on the nature of selection andsuggesting that it can be accounted for using the unique nature of Indigenous family struc-tures. I argue the first source of bias is some unobserved family endowment of skills that maybe correlated with parental residential school attendance and child outcomes. The secondsource of bias is assumed to take the form of some unobserved contextual effect that influ-enced both life outcomes and residential school attendance. I argue that I can account forboth of these unobservable fixed factors by leveraging the extensive information available inthe 2001 Aboriginal Peoples Survey regarding family history with residential schooling usingthe results of Chamberlain (1977) and Carneiro, Hansen, and Heckman (2003). If siblings'and cousins' residential school attendance are noisy measures of problematic unobservablefamily and contextual fixed factors, I can identify the intergenerational effects of residentialschooling. If this is true, the predicted value of cousins' residential school attendance basedon siblings' attendance will yield an unbiased estimate of the family and contextual fixedfactor. This predicted value can then be added as a control variable in the outcome equationand solve the omitted variable bias problem. Intuitively, if a respondent, their siblings andtheir cousins experience similar family and community environments, the predicted value ofcousins' residential school attendance essentially acts as a time-varying environment fixedeffect. Indentification in this context is not dissimilar to exploiting variation between sib-lings, twins or adoptees. The assumption that the respondent and their siblings and cousinsexperience similar family and community environments is most likely to hold if cousins andsiblings live in close proximity to each other and the respondent. Given the typical im-portance of extended family in Aboriginal communities and the tendency to live in multiple904.1. Data and Descriptive Statisticsfamily dwellings, this assumption is much more plausible among Indigenous populations thanin general (Harrison et al 1990; Red Horse 1997; Staples and Mirande 1980).88While I find little evidence that on average parent's attendance at residential school isstatistically associated with their children's adult outcomes, I find substantial heterogeneityin the effects of own residential school attendance based on whether or not one's parentsattended residential school. Given the use of detailed geographic and cohort fixed effects,the controls for unobserved heterogeneity, and the fact parental and child attendance isconditionally uncorrelated, this finding is arguably not the result of selection into residentialschooling or some unobservable quality of the relevant residential school. I find that ifa family has had no previous experience with residential school, the effect of attendinghas negative implications for employment, increases social assistance use and increases thelikelihood of divorce. Even if there is some bias not accounted for by the identificationstrategy, the results on the direction of selection from the previous chapter would at leastsuggest that any increase in market skills are most predominant for children whose familieshave already been exposed to the system. Taken at face value, these results may suggestthat residential schooling negatively impacted parent ability (or preference) to invest in theirchildren's market skills.The rest of this chapter is structured as follows. Section 4.1 describes the data and themain patterns present. Section 4.2 presents the methodology used in this chapter and section4.3 presents the main results. The final section concludes.4.1 Data and Descriptive StatisticsI study the intergenerational effects of residential schooling using the confidential 2001 Abo-riginal Peoples Survey (APS). Like the 1991 APS, the 2001 APS is a cross-sectional, post-census survey and includes information on labour market outcomes, education, cultural88I am unable to use the strategy taken in the previous chapter because I do not know theage of an individual's parents or the individual's (and thus their parents') band affiliation.However, a restricted version of the strategy was tested here and it finds similar results.914.1. Data and Descriptive Statisticsactivities, language use, marriage, and health. However, unlike the 1991 APS, the 2001 APSincludes information on family history with residential schooling.89The 2001 target popu-lation differs from the 1991 population in two important ways. First, the 2001 populationincludes the Aboriginal ancestry population and not merely the identity population as inthe 1991 population. However, the response rate for individuals with Aboriginal ancestrywho did not identify as Aboriginal is much lower than for individuals of Aboriginal identity(68.6 percent compared to approximately 85 percent). Second, the way Aboriginal commu-nities were selected to participate in the survey differed. The 1991 APS was designed to berepresentative of all Aboriginal communities, while the 2001 APS surveyed only 123 of thelargest First Nations communities (reserves). The survey also focused on 52 Inuit commu-nities, 38 communities with a concentration of 40 percent or more Aboriginal peoples (28of these communities are predominately M?tis) and five additional communities with largenumbers of Aboriginal peoples (Prince Albert, North Battleford, Wood Buffalo, Yellowknifeand Whitehorse). While in most provinces these communities cover between 50 to 55 per-cent of the on-reserve population, there is notably less coverage of those living on reserve inBritish Columbia due to the large number of small reserves and the high cost of sampling.The data on residential school attendance available in the 2001 APS includes informa-tion on own, mother's, father's, siblings', grandmother's, grandfather's, cousins', aunts' anduncles' or other family's attendance. However, this is the only family information containedin the survey. In particular, I have no information on parental income, education and orother relevant data. In addition, I do not have the necessary information to construct acomplete family tree. For example, although I know if someone's aunt or uncle went toresidential school, I do not know if it was their mother's or father's sibling. This limita-tion also applies to grandparents. The outcomes I look at include high school graduation,employment, receipt of government transfers, Aboriginal language use in the home, living89The 2001 APS also includes questions than can be used to construct an index of theextent of social support an individual has. Specifically, it includes a question on whether anindividual has someone they can confide in, someone they can count on, to listen to them,love them, enjoy spending time with, take them to the doctor, or someone to have a goodtime with. I construct this index using a set of indicators that equal one if an individual hadsomeone for each activity pretty much all of the time and zero otherwise. I then sum allthe indicators to create the index.924.1. Data and Descriptive Statisticson-reserve, divorce, and the extent of social support.For consistency with the previous chapter, I restrict the population to those between 20and 65, the western provinces and to band member, Status Indians. Here I also restrictthe analysis to individuals with non-missing information on their own, their parents' andcousins' residential school attendance. There was a significant proportion of people whowere missing information on their cousins' residential school attendance and the attendanceof at least one their parents. This is approximately 26 percent of the sample. Another4 percent was excluded due to missing own schooling information. Table E.1 presents theresults of a regression of all the covariates regressed on an indicator of whether the respondentwas missing parental, cousin or own residential schooling information. This is done in orderto get a sense of whether and how much this may bias the results. While some factors areassociated with missing information on own and parental residential schooling information,such as receiving government transfers, or speaking an Aboriginal language at home, beingdivorced, or an individual's grandmother attending a residential school, these associationsare relatively small (for example the likelihood of missing information is increased between1.5 to 2 percent by any of these factors). However, there is a substantial association betweenother missing information on family residential school attendance. For example, if someoneis missing information on their siblings' attendance at residential school, they are 50% morelikely to be missing parental or own information on residential schooling.Table 4.1 presents some basic descriptive statistics broken down by whether or not atleast one parent had attended residential school. The first row of this table shows a highcorrelation between parental and own residential school attendance. If both parents attendeda residential school, an individual was 2.5 times more likely to attend themselves. However,this could be due to unobserved family and contextual factors discussed in the methodologysection. We also see that mothers were more likely to attend residential school than fathers(or at least be reported as having attended). However, if at least one parent attendedresidential school, there is a 46 percent chance that both parents did.90If at least one parent90This high correlation could be due to couples forming at school. It could also arise asa consequence of within community-cohort matching even if couples are not formed duringtheir schooling years.934.2. Frameworkhas attended a residential school, respondents are more likely to report one of their siblingsattending a residential school than to have reported themselves having attended. This is notnecessarily an indicator of biased reporting since individuals can have multiple siblings.91This explanation is also consistent with the fact that it is even more likely that individualshad at least one cousin reporting to have attended residential school.The average age and gender of individuals in each group is roughly equal, while those thathad at least one parent attend residential school are more likely to live in Alberta, Manitobaor Saskatchewan. This is due primarily to differences in treaty obligations on the part of thefederal government for education provision. Individuals are equally likely to graduate highschool in the two groups, but they are more likely to receive government transfers and live onreserve if at least one parent attended residential school. Interestingly, they are equally likelyto speak an Aboriginal language at home. While the likelihood of divorce is equal betweenthe two groups, those that had parents attend residential school have a slightly lower indexof social support. A discussion of these patterns and its relation to own residential schoolattendance will be discussed in more detail section 4.3.The message from this table is similar to that described in the the previous chapter:if anything, parental residential schooling is negatively correlated with economic outcomes.However, the results of previous chapter gives us reason to question the causal nature ofthis correlation. The following section discusses the methodology used to deal with thesystematic selection of individuals into residential schooling.4.2 FrameworkLet the respondent's generation in each family be given by g. Their parent's generation willbe indexed by g ? 1 and their grandparent's generation by g ? 2. Let Ag denote whethersomeone attends a residential school and hg index the individual's final human capital stock.9291The probability of at least one of sibling attending residential school is an additive func-tion of the probability that any given sibling attended.92As noted elsewhere, the term human capital should not be understood normatively.Rather, human capital refers to skills that in a return in the dominate market economy.944.2. FrameworkDifferences in human capital accumulation translate into differences in economic and socialoutcomes. Each of these outcomes can be given by Ygm where m indexes the outcome. Agiven generation's human capital will influence the next generation's skills by a factor ? andinfluences the next generation's likelihood of attending a residential school by a factor ?. Thelogic for including parental human capital, hg?1 in the likelihood of an individual attendingresidential school is given in section 3.2: if an individual had a high level of human capital,they were less likely to be destitute, and thus were less likely to have their child taken toresidential school.I also allow for unobserved contemporaneous factors to influence whether a child attendsa residential school and their human capital accumulation, which are given by ?g and ??grespectively. Contemporaneous factors that affect whether a child attends a residentialschool, ?g, may include whether or not a residential school was open near an individualscommunity when they were of schooling age, national policy at the time of school age, orthe presence of a particular Indian agent with a strong preference for residential schools.Contemporaneous factors that influence an individual's human capital accumulation, ??g,may include a particularly good teacher or improved access to study areas. Let ?g and ?grepresent idiosyncratic, mean zero shocks to the likelihood of residential school attendanceand skills respectively. Given this set-up, the outcomes of each generation can be given bythe following system:Ag = ?hg?1 + ?g + ?ghg = ?Ag + ?hg?1 + ??g + ?g(4.1)Ag?1 = ?hg?2 + ?(g?1) + ?g?1hg?1 = ?Ag?1 + ?hg?2 + ??(g?1) + ?g?1(4.2)The coefficient ? is the effect of residential schooling on human capital accumulation(which the previous chapter argued was positive). Here hg?2 is the human capital endowmentof the observed individual's grandmother. This work is not primarily interested in ?, theWhile one may debate about the appropriateness of the terminology, it is standard in theeconomics literature and thus is used here.954.2. Frameworkaffect of parental human capital on outcomes, but rather the effect of parental residentialschooling attendance on child outcomes given by ??. In the model above, parental residentialschooling only matters for child outcomes to the extent it influences parental human capital(which affects child outcomes through ?). To derive the reduced form model, substitute infor hg?1 in 4.1. This yields the following specification:Ag = ??Ag?1 + ??hg?2 + ???(g?1) + ??g?1 + ?g + ?ghg = ?Ag + ??Ag?1 + ?2hg?2 + ???(g?1) + ??g?1 + ??g + ?g(4.3)The difficulty in estimation is two fold. First, we do not observe the family endowmentof human capital handed down to the parents' generation (hg?2). This unobserved variableenters the residential school attendance of all generations and thus it is not possible toseparately identify the effect of own or parental residential school attendance from someunobserved family factor. Second, the unobservable contextual factors that affects parent'shuman capital accumulation (??(g?1)) and unobservable contextual factor that affects childresidential school attendance and human capital (?g) also results in omitted variable bias.The only case in which the human capital equation is identified is if ? = 0, i.e. parents donot transmit human capital to their children. Thus the model will not be identified withoutfurther restrictions.My proposed solution to use the approach proposed by Chamberlain (1977) and gener-alized by Carneiro, Hansen and Heckman (2003).93Their approach uses the fact that if youhave two measurement equations of the problematic unobserved factors, the equations aboveare identified. In other words if you have two imperfect measurements for the problematicunobserved factors, you can get around the standard errors-in-variables problem by instru-menting one measurement with the other. In my case, I argue that I can use informationon sibling and cousin residential school attendance in order to identify the intergenerationalconsequences of residential schooling. Let an individual's cousin be represented by the indexg? and their aunt/uncle by g? ? 1. Their outcome equations can be written as93The connection between these two pieces of work has been made by Green, Gallipoli,Foley (2010).964.2. FrameworkAg??1 = ?hg??2 + ?(g??1) + ?g??1 (4.4)Ag? = ??Ag??1 + ??hg??2 + ???(g??1) + ??g??1 + ?g? + ?g?hg? = ?Ag? + ??Ag??1 + ?2hg??2 + ???(g??1) + ??g??1 + ??g? + ?g?. (4.5)Equation 4.4 gives the residential school attendance of the observed generation's aunt/unclewhile Equation 4.5 gives the system of equations for their cousins'. Now, it is not unreason-able to assume that the parents of your parents are also the parents of your aunt or uncle.In the present context this assumption is non-trivial because I do not know complete familyhistories. To partly deal with this, the main specification will use at least one parent attend-ing residential school as the main variable of interest. This will cover the possibility thataunts referred to in the question are more likely to be related to the parent of interest. Sec-ond, I also assume that aunts and uncles experience the same cohort specific shocks to theirhuman capital accumulation as parents do. However the assumption could be weakened. Solong as it is true that ?hg?2 = hg??2 and ??(g??1) = ???(g?1), and if one is willing to assumethat the contemporaneous factors influencing cousins' residential school attendance and hu-man capital accumulation are linearly related to those influencing the observed individual's(?g? = ??g, and ??g? = ???g), then we can re-write cousin's outcome equations as1?(Ag? ? ??Ag??1 ? ??g??1 ? ?g?) = ??hg?2 + ???(g?1) + ?gand1?(?Ag? ? ??Ag??1 ? ??g??1 ? ?g? + hg?) = ?2hg?2 + ???(g?1) + ??g.The assumption of cousins' attendance accurately reflecting the same family driven hu-man capital and contextual environment as the respondent is most plausible if they live closetogether. This assumption is arguably more reasonable in Aboriginal communities than it isin the public at large since extended family plays an extremely important role in Aboriginal974.2. Frameworklife, culture, and child rearing (Harrison et al 1990; Red Horse 1997; Staples and Mirande1980). Statistically, Aboriginal peoples, especially on-reserve, are significantly more likely tolive in multiple family dwellings than non-aboriginal people (Statistics Canada 2008) and,although out-marriage is not infrequent, marriage within bands is common and incentivizedby band membership and registration regulation (Clathworthy 2003, 2005).Continuing on, I substitute the above restricted version of the cousins' equations into theobserved individual's residential school attendance equation after multiplying by ? and ?.This results inAg = ??Ag?1 + 1?Ag? ? ??1?Ag??1 + ?(?g?1 ?1??g??1) + (?g ?1??g?)hg = ?Ag + ??Ag?1 ? ? 1?Ag? ? ??1?Ag??1 ? ?(?g?1 ?1??g??1) + (?g ?1??g?) + ?hg?. (4.6)Note that a new problem arises. The new human capital equation (the second line of 4.6)is determined by Ag? and hg?. Since hg? is determined partially by Ag?, the standard assumptionsnecessarily to estimate 4.6 do not hold. Thus I follow Chamberlain (1977) and use a secondmeasurement equation, siblings' attendance at residential school, to instrument for cousins'attendance. If sibling and cousin residential school attendance are only correlated throughgrandparent human capital accumulation and contemporaneous factors, predicting cousinresidential school attendance with sibling attendance will provide a consistent estimate of?hg?2 + ???(g?1) + ?g which is orthogonal to hg?. Consequentially, we can use this predictedvalue to control for unobserved factors. To see this, write the siblings' residential schoolattendance as:Ag? = ??Ag?1 + ??hg?2 + ???(g?1) + ??g?1 + ?g + ?g?.Let the predicted value of cousins' residential school attendance from a regression of Ag?on Ag?, Ag?1and Ag??1 be given by A?g?. Thus the final specification 4.3 is given by984.2. FrameworkAg = ??Ag?1 + 1?A?g? ? ??1?Ag??1 + ?(?g?1 ?1??g??1) + (?g ?1??g?)hg = ?Ag + ??Ag?1 ? ? 1?A?g? ? ?1??Ag??1 ? ?(?g?1 ?1??g??1) + (?g ?1??g?) + hg?. (4.7)Now I want to clarify a number of assumptions made implicitly in the above. First, I needto assume that any accounted for contemporaneous changes or family factors correlated withresidential school attendance or human capital accumulation are captured by the predictedvalue of cousins' residential school attendance.94Second, I need to assume residential schoolattendance and human capital accumulation does not depend on cousin residential schoolattendance directly. Finally, for both the attendance and the human capital equations to beidentified, sibling residential school attendance must not affect human capital accumulationor residential school attendance directly.95This rules out any human capital spillovers fromsibling residential school attendance and any strategic choices on the part of the parent to letone child attend and another not based on their children's relative expected life outcomes.This last assumption is similar to that made in research involving twins.9694This is fundamentally tied to siblings' and cousins' residential school attendance beingreliable measures of these factors.95It should be noted I do not need to instrument for aunt's residential school attendanceor parental residential school attendance as long as the random shocks to human capitalaccumulation, ?g?1 and ?g??1, are uncorrelated with the shocks to the likelihood of attendingresidential school. Also note that if one is willing to assume that ? = 1, then the abovespecification in Equation 4.7 inherently has over-identifying restrictions that can be tested.The first is that in the residential school attendance equation, the predicted value of cousin'sresidential school attendance should have a coefficient equal to one. In addition, the coef-ficient on cousin's residential school attendance should equal the negative of the coefficienton own attendance and the coefficient on aunt's attendance should be equal to the negativeparental attendance. It should also be noted that that using sibling's attendance as the mainmeasure of the unobservable factors will not result in the same estimating equations as onethat uses cousin's attendance above. Specifically, if the order of the measurement equationswas reversed, the resulting specification would beAg = (??? ??)Ag?1 + Ag? + (?g ? ?g?)hg = ?Ag + (??? ??)Ag?1 ? ?Ag? + ?(?g?1 ? ?g?1) + (?g ? ?g?) + hg?.96This includes but is not limited to Berham and Rosenzweig (2002), Antonovics andGolderger (2005), Sacerdote (2002, 2007), Plug (2004), and Bj?rklund, Lindahl and Plug(2004, 2006).994.3. ResultsThis last assumption may be the most unpalatable of three. It is possible to imag-ine parental resistance to one of their children attending residential school may depend onwhether they have other children that have been removed. However, even if sibling at-tendance directly affects own residential school attendance, the human capital estimatingequations described above will still yield consistent estimates so long as own attendance iscontrolled for.97However, even if one believes that the identification strategy outlined abovedoes not fully account for all of the bias, the estimates here will still be useful to bound theeffect of parental attendance on long run outcomes.4.3 ResultsThe first stage results which predict cousins' residential school attendance with siblings' at-tendance are reported in Table B.1. Both columns in Table B.1 include a full set of birthcohort and census division fixed effects in addition to the controls shown. The differencebetween the two columns is an interaction between own and parent attendance and is in-cluded in the second model but not in the first. The correlation between sibling and cousinattendance is large: if a sibling attended residential school, the likelihood your cousin didincreases by approximately 30 percent. It can be seen that the first stage between cousinsand siblings is strong enough to pass any weak instruments test and inferences made will beexpected to have normal size.Table 4.3 contains the first set of main results: the intergenerational effect of residentialschool attendance for a set of economic variables closely matched to those available in theprevious chapter. Each panel represents a different dependent variable which is listed as aheading. The first and second columns do not account for unobservable family or contem-poraneous changes. The third controls for cousins attendance. The fourth and fifth use the97Another possible, more mechanical source of bias may come from the fact that if anindividual has/had no siblings, they could not have attended residential school. If this iscorrelated with parent specific human capital shocks, this would be problematic if it was notcaptured by a similar source of bias in cousins attendance.1004.3. Resultspredicted value of cousins attendance as a control. The final column includes an interactionbetween own residential school attendance and parents attendance. All columns conditionon gender, latitude, distance to closest major city,98census division and birth cohort fixedeffects and selection. The results do not change if more detailed geographic fixed effects(reserve/census sub-division) are used. Note that dropping all individuals with missing in-formation gives similar results. The results are also similar if a standard Heckman correctionis applied.99The same is true for all following estimation results.The first panel reports the results for the probability of high school graduation. Inthe first columns there is no evidence for any positive association between own or parentresidential school attendance on high school graduation. However, once cousins' residentialschool attendance is added and the predicted value is used, the effect of residential schoolattendance increases and is positive, although not statistically significant. Predicted cousins'residential school attendance enters with the opposite sign as would be predicted by thedirection of selection in the previous chapter. However, there appears to be virtually noeffect of parental residential school attendance on child outcomes. On the other hand, ifthe effect of residential school attendance is allowed to vary by parental attendance, we findthat the positive effect of residential school attendance on high school graduation is entirelyisolated to children whose parents attended a residential school. If both parents attendinga residential school is used as a measure, rather than at least one attending, the effect isstatistically significant.In the next panel, the dependent variable is receiving social assistance in the previousyear. If one does not account for unobservables, both parental attendance and own atten-dance are positively associated with receiving social assistance. However, once selection isaccounted for, the effect of own residential school attendance on social assistance disappearsand the effect of parental attendance becomes insignificant. Allowing for an interaction effect98Note that for those off reserve, I do not have information on the latitude and distancefrom the closest major city of the center of their subdivision. Thus, I use the average ofthe latitude and distance in their census division. Since I control for census division fixedeffects however, all this variation is captured. Thus it is the variation within census divisionsgenerated by those on reserve that identifies these coefficients.99The additional variation used in the Heckman correction is an indicator for missing grand-parental information.1014.3. Resultsbetween own and parental residential school attendance suggests that if an individual didnot attend a residential school while their parents did they are actually more likely to receivesocial assistance. In addition, if both parents and children attended a residential school, thenegative effect of own attendance is no longer present. Similar findings exist in the nextpanel with respect to employment. Residential schooling of either the parent or the childappears to be negatively associated with the likelihood of employment unless both parentand child attended residential school. These results are extremely important for understand-ing and contextualizing the effects in the previous chapter. It should be noted that thesample used in the previous chapter included many smaller, more isolated communities andan older generation whose parents may have been more likely to attend residential school.As a consequence, this effect may have dominated the results. Alternatively, if children fromsmaller communities were the ones to do better if they attended residential school, then theseanalyses are not comparable and we should not expect the same results as in the previouschapter.Table 4.4 performs the same exercise as Table 4.3, but examines the effect of residentialschooling on likelihood of speaking an Aboriginal language at home, living on-reserve, beingdivorced and the index of social support. While there is little evidence for any intergen-erational effect of residential school attendance on Aboriginal language use in the home, ifparents attended a residential school their children are more likely to live on-reserve.The next table further demonstrates the heterogeneity in the effect of residential schoolattendance by splitting the sample by the number of parents that attended a residentialschool. The first column of Table 4.5 replicates the second column of Table 4.3 for compa-rability's sake. The second column restricts the sample to individuals who have at least oneparent that attended residential school. The third column restricts the sample to people whohad both parents attend a residential school while the fourth column focuses on individualsfor whom neither parent attended. All specifications condition on predicted cousins' residen-tial school attendance. Comparing columns one to four suggests substantial heterogeneityin the impact of residential schooling. The results contained in the first panel suggest thatall the positive effect of residential schooling on high school graduation rates is found among1024.3. Resultschildren whose parents attended residential school themselves. The results also suggest ifan individual's parents did not have experience with residential schooling, attending a res-idential school themselves is associated with a reduced the likelihood of employment laterin life and an increased likelihood of divorce. While not statistically significant, residentialschool attendance is additionally associated with having a lower score on the social supportindex. These results are informative regarding the claim that residential schools have dam-aged family formation and Aboriginal communities. Residential schooling may have had adirect negative effect on family formation and social support networks, but only if the familyhad not been previously exposed to residential schooling.Taken together, these results suggest that even if the predicted value of cousins' atten-dance doesn't fully account for the problematic unobserved factors, as long as selection issymmetric along the lines of parental residential school attendance, those whose parentshave attended residential school are not harmed in the same way by residential schoolingas those whose parents did not. This may suggest that attendance at a residential schoolchanges either parenting ability, or preference for investing in their children's human capitalaccumulation. In the cases where parenting has been affected by residential schools, childrenare less harmed (or may even benefit) from being removed from the home and put into aresidential school themselves.Although it is not possible to directly test the assumption that selection is symmetricbetween children whose parents attended residential school and those whose parents did not,I can examine whether parental attendance is associated with own residential school atten-dance. If there is differential selection in the type of children taken to residential school thatdepends on parental attendance, we would also expect that parental attendance would beassociated with a change in the probability of their children attending. Table 4.7 examineswhether this is the case. The first column of Table 4.7 does not account for unobservable fam-ily fixed effects or contemporaneous changes, while the second column controls for cousin'sattendance and the third uses the predicted value of cousin's attendance. If my estimateof the unobserved bias is not controlled for (as in the first and second column), parentalresidential school attendance is still positively correlated with child attendance even once1034.3. Resultschild birth cohort, geography and gender are accounted for. However, once predicted cousins'residential school attendance added to the specification the correlation between parental res-idential school attendance and child attendance disappears. While this does not necessarilyimply symmetric selection among individuals with different family residential histories, itis supportive of the notion. Although not included here, one could also interact parents'residential school attendance with predicted cousins' attendance to attempt to capture dif-ferential selection. When this is done, the qualitative nature of the results still holds. Onemight also suspect that because the nature of residential schooling changed over time, thedifferential effects found between individuals whose parents attended residential school andthose whose parents did not attend may simply be due to the changing nature of the coun-terfactual between different generations, some of whom are more likely to have parents whoattended residential school than others. I test this hypothesis by interacting own residentialschool attendance with birth cohort to capture these changes over time. Again, I find thatthis does not influence the conclusions drawn.Table 4.8 examines another assumption made in the above discussion. Specifically itwas assumed that siblings' residential school attendance does not affect an individual's lifeoutcomes directly. While testing this explicitly is not possible, it could be argued thatsiblings' attendance at residential school may affect a respondants ability to get adviceand support later in life. This support and advice may matter for how one does in thelabour market. Table 4.8 attempts to provide some evidence that siblings' residential schoolattendance does not affect an individual's human capital through this plausible channel.The APS 2001 includes questions on whether an individual has someone they can confide in,someone they can count on, someone to listen to them, love them, enjoy spending time withthem, take them to the doctor, or have a good time with. Table 4.8 reports the estimatesfrom a regression of these measures on siblings' residential school attendance along withseveral other control variables. We can see that the indicator of siblings' attendance islargely uncorrelated with all of the support factors mentioned above. These variables werealso added as controls in the previous models as a robustness check and the results werelargely unaffected. Obviously there are problems with this (given they are likely endogenous1044.3. Resultsand their sum was used as a dependent variable earlier on), but it provides at least somereassurance that these factors are not substantially influencing the results.However, there are a number of other reasons why this identification strategy may bequestionable. There is a recent literature that claims there are spillover effects from siblingeducation (Qureshi, J. A. 2011; Shrestha 2011; Begum, Islam, and Smyth 2012) and if ulti-mately this newly developing literature establishes these spillover effects, the identificationstrategy used here may be invalid. Another possible criticism regards the possibility thatfamilies may resist one child attending a residential school more than for the others based onsome unobserved characteristics of the child. If this is true, the identification strategy wouldalso be biased. However, this same criticism can be made of the literature that uses familyfixed effects, twins or adoptees for the identification. Additionally, the results presented inthe last section suggest that conditioning on predicted cousins' attendance does seem toeliminate some of the bias. While this does not imply that siblings are validly excluded fromthe equations, it does suggest that controlling for unobserved family and contextual fixedeffects may be important. In a sense, the estimates presented here admit a lower boundon the severity of the bias and present a conservative perspective regarding the effect ofresidential school attendance if one believes the results from the previous chapter are valid.4.3.1 Accounting for Grandparents AttendanceThe finding that own residential school attendance interacts with parental attendance raisesthe issue of controlling for grand-parental residential school attendance. If the above resultsare consistent over time, the findings would suggest that the effect of parental attendancewould depend on whether an individual's grandparents attended residential school. Theresults so far will allow us to make predictions regarding the patterns we should expect tofind if we were to condition on grand-parental attendance.First assume that parents only affect their children through their parenting human cap-ital rather than their market human capital. Also ignore the issue of unobservable family1054.3. Resultsand environment fixed factors for the moment.100Let p represent parenting capital stock,R index residential school parenting, and g ? 1 index the parent's generation. Now recon-sider the human capital equation of the respondent, but allow the effect of residential schoolattendance to vary with parent's parenting human capital :hg = ?(AgpR + (1? Ag)pg?1) + ?g,where ? is assumed to be greater than zero.In addition, assume that parents acquire their parenting ability during adolescence. Ifthey remain with their own parents, they acquire the ability of their parents (the respon-dent's grandparents). If they attend a residential school they acquire the parenting skillsendowed by the school. As before, let g ? 2 index the grandparent's generation. With theseassumptions we can write parent's parenting stock aspg?1 = Ag?1pR + (1? Ag?1)pg?2.Substituting this equation into the first and re-arranging yields:hg = pg?2 + ?Ag + ?(pR ? pg?2)Ag + (pR ? pg?2)Ag?1 ? ?(pR ? pg?2)AgAg?1 + ?g. (4.8)This framework, which is suggested by the above findings, implies a number of restrictionsthat depend on grand-parental residential schooling. Let ?1 = pg?2, ?2 = ?(pR ? pg?2),?3 = (pR ? pg?2), ?4 = ??(pR ? pg?2) and re-write 4.8 ashg = ?1 + ?2Ag + ?3Ag?1 + ?4AgAg?1 + ?g.First, if an individual's grandparents attended a residential school, pg?2 = pR, we wouldpredict, ?2 = ?3 = ?4 = 0. However, if grandparent parenting ability was superior to theparenting ability endowed by the residential school, we would predict, ?2, ?3 > 0, while100These unobservable fixed factors will be accounted for in all the estimations that follow.1064.4. Conclusion?4 < 0 and that ?4 = ??3. These patterns are not predicted by some obvious story ofselection and thus finding them may be supportive of the notion that residential schoolingharmed parent's ability to invest in their child's human capital.I investigate whether these expected patterns are realized in Tables 4.9 and 4.10. The firstcolumn of Table 4.9 and 4.10 replicates the results from Table 4.3 and 4.10 for comparability.The second restricts the sample only to individuals for whom no grandparents attendedor were missing information on grand-parental attendance. Note that it does not matterwhether those with missing grand-parental information are excluded from this column, theresults are the same. The third column restricts the sample to be only people for whom atleast one grandmother and grandfather attended residential school. We find that of the total49 restrictions imposed by the framework above, 37 are not rejected. While this is far fromproof against asymmetric bias, it lends further credibility to the notion that residential schoolattendance harms parenting ability which induces heterogeneity in the effect of residentialschooling.4.4 ConclusionAlthough the previous chapter provided evidence that residential school attendance increasedthe likelihood of graduating high school and being employed, this chapter demonstrates thatthere is little evidence that parental residential school attendance on average increased themarket engagement or graduation rates of their children. In addition, this chapter uncoverssubstantial heterogeneity in the effect of residential schooling based on family history withresidential schooling. If children attended residential school when neither of their parentsdid, they are less likely to be employed and more likely to be divorced. On the otherhand, if a child attends residential school and so did their parents, residential schooling mayimprove the likelihood of graduating high school. These results may suggest that residentialschooling changes parent's ability and/or willingness to invest in their children's marketskills. These findings are interesting in part because they may suggest that the effects of1074.4. Conclusionresidential schooling policy are not generation invariant.1084.5. Tables and Figures4.5 Tables and FiguresBelow are the tables for Chapter 4.1094.5. Tables and FiguresTable 4.1: Descriptive Statistics: APS 2001Control VariablesNeither Parent Attended At Least One Parent Attended P-Value (Difference)Attended 0.104 0.261 0.003(0.093) (0.193)Mother Attended 0.824(0.145)Father Attended 0.64(0.23)Both Parents Attended 0.463(0.249)Brother Attended 0.143 0.352 0.001(0.123) (0.228)Aunt or Uncle Attended 0.159 0.823 0.000(0.134) (0.146)Cousin Attended 0.174 0.531 0.000(0.144) (0.249)Age 38.77 38.341 0.989(0.357) (0.245)Female 0.567 0.587 0.433(0.246) (0.242)Live in AB, MN, or SK* 0.492 0.633 0.113(0.25) (0.232)Dependent VariablesHigh School Graduation 0.624 0.642 0.434(0.235) (0.230)Social Assistance 0.229 0.330 0.104(0.177) (0.221)Employment 0.699 0.659 0.664(0.21) (0.225)Proportion on Reserve 0.193 0.322 0.036(0.156) (0.218)Aboriginal Language At Home 0.037 0.042 0.042(0.036) (0.04)Divorced 0.086 0.085 0.532(0.079) (0.078)Number of Support Factors 3.666 3.500 0.035(0.072) (0.057)Sample Size ~5000 ~6000Notes: Standard deviations are reported in parentheses. Sample Sizes are approximate because they vary slightly byeach dependent variable. * This is not used as a control variable. Rather census division fixed effects are used in themain regressions.1104.5. Tables and FiguresTable 4.2: First StageDependent Variable: Cousin Attended(1) (2)Sibling Attended 0.295*** 0.296***(0.025) (0.025)Attended 0.151*** 0.202***(0.029) (0.042)At Least One Parent Attended 0.008 0.022(0.018) (0.019)Attended X At Least One Parent -0.078*(0.044)Missing Sibling 0.246*** 0.246***(0.072) (0.072)Aunt Attended 0.412*** 0.406***(0.017) (0.020)Missing Aunt 0.256*** 0.252***(0.035) (0.036)Female 0.002 0.001(0.015) (0.015)Latitude 0.020 0.020(0.028) (0.029)Distance to City 4.930 0.045(7.566) (0.075)Census Division FE Yes YesBirth Cohort FE Yes YesF-Statistics on Sibling Attend 138.77 144.01R-Squared 0.482 0.483N ~10,000 ~10,000Notes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate thelevel of significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this tableincludes 58 census division and 47 cohort fixed effects. Sample sizes are approximate.1114.5. Tables and FiguresTable 4.3: The Effect of Parental Residential School Attendance: Economic OutcomesOLS 1 OLS 2 OLS 3 IV 1 IV 2Dependent Variable: High School GraduationAttended 0.015 0.010 0.015 0.049 0.016(0.019) (0.019) (0.02) (0.035) (0.047)At Least One Parent Attended 0.025 -0.001 0.000 -0.009(0.018) (0.022) (0.022) (0.024)Attended X At Least One Parent 0.052(0.039)Cousin Attended -0.032(0.021)?Cousin Attended -0.137 -0.140(0.09) (0.09)Dependent Variable: Social AssistanceAttended 0.109*** 0.098*** 0.074*** 0.015 0.075(0.020) (0.02) (0.021) (0.035) (0.049)At Least One Parent Attended 0.053*** 0.033* 0.030 0.047**(0.016) (0.018) (0.019) (0.021)Attended X At Least One Parent -0.093**(0.039)Cousin Attended 0.057***(0.02)?Cousin Attended 0.239** 0.244***(0.093) (0.093)Dependent Variable: EmploymentAttended -0.059*** -0.058*** -0.054** -0.037 -0.147***(0.022) (0.022) (0.023) (0.038) (0.051)At Least One Parent Attended -0.003 -0.009 -0.008 -0.039*(0.017) (0.021) (0.021) (0.022)Attended X At Least One Parent 0.171***(0.044)Cousin Attended -0.017(0.023)?Cousin Attended -0.075 -0.084(0.087) (0.085)Census Division FE Yes Yes Yes Yes YesBirth Cohort FE Yes Yes Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate the levelof significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this table includes 58census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing information for uncleor aunts attendance, uncle or aunts attendance, missing information for siblings attendance, latitude and distance to the closestcity. The first and second columns do not account for unobservable family or contemporaneous changes. The third controls forcousin's attendance. The fourth and fifth use the predicted value of cousin's attendance as a control.1124.5. Tables and FiguresTable 4.4: The Effect of Parental Residential School Attendance: Social OutcomesOLS 1 OLS 2 OLS 3 IV 1 IV 2Dependent Variable: Speak an Aboriginal Language At HomeAttended 0.014* 0.016** 0.018** 0.022** 0.028*(0.007) (0.007) (0.007) (0.011) (0.016)At Least One Parent Attended -0.007 -0.006 -0.006 -0.004(0.005) (0.006) (0.006) (0.006)Attended X At Least One Parent -0.010(0.015)Cousin Attended -0.005(0.006)?Cousin Attended -0.017 -0.017(0.024) (0.024)Dependent Variable: On-ReserveAttended 0.107*** 0.084*** 0.043** -0.028 -0.011(0.016) (0.016) (0.017) (0.026) (0.04)At Least One Parent Attended 0.114*** 0.057*** 0.054*** 0.059***(0.011) (0.013) (0.014) (0.015)Attended X At Least One Parent -0.027(0.034)Cousin Attended 0.086***(0.015)?Cousin Attended 0.311*** 0.313***(0.06) (0.06)1134.5. Tables and FiguresTable 4.4: The Effect of Parental Residential School Attendance: Social Outcomes (Continued)OLS 1 OLS 2 OLS 3 IV 1 IV 2Dependent Variable: DivorcedAttended 0.011 0.012 0.013 0.033 0.089*(0.016) (0.016) (0.016) (0.028) (0.047)At Least One Parent Attended -0.007 0.004 0.006 0.021(0.012) (0.016) (0.016) (0.017)Attended X At Least One Parent -0.088**(0.041)Cousin Attended 0.008(0.014)?Cousin Attended -0.05 -0.045(0.057) (0.055)Dependent Variable: Number of Support FactorsAttended -0.246** -0.236* -0.13 -0.294 -0.453(0.124) (0.126) (0.133) (0.221) (0.326)At Least One Parent Attended -0.049 -0.038 -0.043 -0.086(0.106) (0.129) (0.132) (0.145)Attended X At Least One Parent 0.248(0.269)Cousin Attended -0.337**(0.134)?Cousin Attended 0.191 0.177(0.561) (0.555)Census Division FE Yes Yes Yes Yes YesBirth Cohort FE Yes Yes Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate the level of significance: *p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this table includes 58 census division and 47 cohort fixedeffects. Other variables included are an indicator for gender, missing information for uncle or aunts attendance, missing information for siblingsattendance, uncle or aunts attendance, latitude and distance to the closest city. The first and second columns do not account for unobservablefamily or contemporaneous changes. The third controls for cousin's attendance. The fourth and fifth use the predicted value of cousin's attendanceas a control.1144.5. Tables and FiguresTable 4.5: The Effect of Residential School Attendance By Parental Attendance: Economic Out-comesAll At Least One Both NeitherHigh School GraduationAttended 0.049 0.049 0.071** 0.024(0.035) (0.035) (0.035) (0.035)At Least One Parent Attended 0.000(0.022)?Cousin Attended -0.137 -0.171* -0.162* -0.109(0.090) (0.090) (0.088) (0.085)Social AssistanceAttended 0.015 -0.011 -0.049 0.076(0.035) (0.035) (0.032) (0.037)At Least One Parent Attended 0.030(0.019)?Cousin Attended 0.239** 0.283*** 0.273*** 0.17(0.093) (0.093) (0.083) (0.089)EmployedAttended -0.037 0.031 0.038 -0.188***(0.038) (0.038) (0.037) (0.039)At Least One Parent Attended -0.008(0.021)?Cousin Attended -0.075 -0.232** -0.166* 0.120(0.087) (0.087) (0.094) (0.093)Census Division FE Yes Yes Yes YesBirth Cohort FE Yes Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate thelevel of significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this tableincludes 58 census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing informationfor uncle or aunts attendance, uncle or aunts attendance, missing information for siblings attendance, latitude and distanceto the closest city. The row labeled cousin attended are the coefficients on the predicted value for unobservable family andcorrelated fixed effects.1154.5. Tables and FiguresTable 4.6: The Effect of Residential School Attendance By Parental Attendance: Social OutcomesAll At Least One Both NeitherOn ReserveAttended -0.047 -0.031 0.020 -0.016(0.037) (0.037) (0.027) (0.034)At Least One Parent Attended 0.036(0.023)?Cousin Attended 0.159* 0.329*** 0.113 0.263***(0.094) (0.094) (0.069) (0.074)Speaking An Aboriginal Language At HomeAttended -0.028 0.025** 0.035* 0.018(0.026) (0.026) (0.013) (0.019)At Least One Parent Attended 0.054***(0.014)?Cousin Attended 0.311*** -0.023 -0.098** -0.014(0.06) (0.06) (0.029) (0.043)Table 4.6: The Effect of Residential School Attendance By Parental Attendance: Social Outcomes(Continued)All At Least One Both NeitherDivorcedAttended 0.033 -0.011 0.008 0.133**(0.028) (0.028) (0.022) (0.022)At Least One Parent Attended 0.006(0.016)?Cousin Attended -0.05 0.061 0.073 -0.217***(0.057) (0.057) (0.051) (0.046)Number of Support FactorsAttended -0.294 -0.108 -0.117 -0.594(0.221) (0.221) (0.188) (0.216)At Least One Parent Attended -0.043(0.132)?Cousin Attended 0.191 -0.404 -0.244 0.808(0.561) (0.561) (0.475) (0.506)Census Division FE Yes Yes Yes YesBirth Cohort FE Yes Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate the levelof significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this table includes58 census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing information foruncle or aunts attendance, uncle or aunts attendance, missing information for siblings attendance, latitude and distance to theclosest city. The row labeled cousin attended are the coefficients on the predicted value for unobservable family and correlatedfixed effects.1164.5. Tables and FiguresTable 4.7: Intergenerational Persistence? The Effect of Parental Residential School Attendance onOwn AttendanceDependent Variable: Attended A Residential SchoolOLS 1 OLS 2 IV 1At Least One Parent Attended 0.125*** 0.045*** 0.011(0.013) (0.016) (0.026)Female -0.008 -0.008 -0.009(0.012) (0.012) (0.021)Latitude -0.01 -0.012 -0.031(0.017) (0.02) (0.042)KM to Closest City (10) 0.174*** 0.137** 0.022(0.065) (0.061) (0.097)Cousin Attended 0.276***(0.023)?Cousin Attended 1.326***(0.099)Aunt or Uncle Attended -0.009 -0.516***(0.019) (0.053)Missing Aunt 0.024 -0.310***(0.025) (0.065)Missing Sibling -0.317***(0.101)R-squared 0.2 0.283Census Division FE Yes Yes YesBirth Cohort FE Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate the levelof significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this table includes58 census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing information foruncle or aunts attendance, uncle or aunts attendance, missing information for siblings attendance, latitude and distance to theclosest city. The first column does not account for unobservable family or contemporaneous changes. The second controls forcousin's attendance. The third uses the predicted value of cousin's attendance as a control. The third uses the predicted valueof the cousin's attendance as a control but restricts the sample to just the on-reserve population. This is done to demonstratethat none of the results hing fundamentally on the on-reserve status of the individual. The final column restrict instruments forown and parent attendance at a residential school using the school presence indicator and the distance to the closest residentialschool and the bartik style instrument respectively.1174.5. Tables and FiguresTable 4.8: Correlation of Siblings Residential School Attendance With Support FactorsDependent Variable: Sibling AttendanceSomeone to Count on 0.004(0.026)Someone to Listen -0.033(0.028)Someone to Confide in -0.008(0.023)Someone to Take you to Doctor -0.005(0.02)Someone to Love -0.008(0.019)Someone to Have a Good time 0.037*(0.021)Someone be Together With 0.030(0.022)Attended 0.496***(0.026)At Least One Parent Attended 0.030*(0.018)Female 0.005(0.015)Latitude -0.005(0.015)KM to Closest City (100) 0.004(0.052)Cousin Attended 0.247***(0.023)Census Division FE YesBirth Cohort FE YesR-squared 0.491Notes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate thelevel of significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this tableincludes 58 census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing informationfor uncle or aunts attendance, missing information for siblings attendance, latitude and distance to the closest city.1184.5. Tables and FiguresTable 4.9: The Intergenerational Effects of Residential School Conditional On Grandparent's At-tendanceAll Neither or Missing Grandparent Grandmother & GrandfatherDependent Variable of Interest: High School GraduationAttended 0.016 0.001 0.267**(0.047) (0.049) (0.104)At Least One Parent Attended -0.009 -0.005 0.019(0.024) (0.028) (0.048)Attended X At Least One Parent 0.052 0.057 -0.241***(0.039) (0.044) (0.088)?Cousin Attended -0.140 -0.098 -0.243*(0.09) (0.1) (0.143)Dependent Variable of Interest: Social AssistanceAttended 0.075 0.084* -0.026(0.049) (0.05) (0.119)At Least One Parent Attended 0.047** 0.058*** 0.007(0.021) (0.023) (0.049)Attended X At Least One Parent -0.093** -0.086* -0.030(0.039) (0.044) (0.104)?Cousin Attended 0.244*** 0.209** 0.350**(0.093) (0.099) (0.15)Dependent Variable of Interest: EmploymentAttended -0.147*** -0.159*** 0.045(0.051) (0.056) (0.121)At Least One Parent Attended -0.039* -0.048* 0.000(0.022) (0.026) (0.05)Attended X At Least One Parent 0.171*** 0.193*** -0.002(0.044) (0.05) (0.109)?Cousin Attended -0.084 -0.049 -0.272*(0.085) (0.094) (0.151)Census Division FE Yes Yes YesBirth Cohort FE Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate the levelof significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this table includes 58census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing information for uncleor aunts attendance, uncle or aunts attendance, missing information for siblings attendance, latitude and distance to the closestcity. All columns use predicted cousin's attendance to account for observed selection. The first column includes the wholesample, while the second only includes respondents that had no grandparents attend residential school or were missing grand-parental information. In the third column only includes individuals who had at least one grandmother and one grandfatherattend residential school.1194.5. Tables and FiguresTable 4.10: The Intergenerational Effects of Residential School Conditional On Grandparent's At-tendanceAll Neither or Missing Grandparent Grandmother & GrandfatherDependent Variable: Speak an Aboriginal Language At HomeAttended 0.028* 0.026 -0.025(0.016) (0.017) (0.03)At Least One Parent Attended -0.004 -0.008 0.01(0.006) (0.007) (0.012)Attended X At Least One Parent -0.010 -0.013 0.049**(0.015) (0.017) (0.023)?Cousin Attended -0.017 -0.017 -0.004(0.024) (0.026) (0.054)Dependent Variable: Live on ReservationAttended -0.011 -0.011 -0.075(0.04) (0.044) (0.095)At Least One Parent Attended 0.059*** 0.088*** -0.065*(0.015) (0.018) (0.035)Attended X At Least One Parent -0.027 -0.054 0.061(0.034) (0.039) (0.091)?Cousin Attended 0.313*** 0.347*** 0.289***(0.06) (0.071) (0.095)1204.5. Tables and FiguresTable 4.10: The Intergenerational Effects of Residential School Conditional On Grandparent'sAttendance (Continued)All Neither or Missing Grandparent Grandmother & GrandfatherDependent Variable: DivorcedAttended 0.089* 0.123** -0.039(0.047) (0.053) (0.058)At Least One Parent Attended 0.021 0.005 0.049(0.017) (0.016) (0.034)Attended X At Least One Parent -0.088** -0.092* 0.032(0.041) (0.047) (0.055)?Cousin Attended -0.045 -0.138** 0.071(0.055) (0.056) (0.074)Dependent Variable: Number of Support FactorsAttended -0.453 -0.426 -0.796(0.326) (0.335) (0.725)At Least One Parent Attended -0.086 0.028 -0.665**(0.145) (0.158) (0.28)Attended X At Least One Parent 0.248 0.106 0.599(0.269) (0.287) (0.698)?Cousin Attended 0.177 0.163 -0.003(0.555) (0.598) (0.804)Census Division FE Yes Yes YesBirth Cohort FE Yes Yes YesNotes: Standard errors clustered at the cohort-census division level are reported in parentheses. The asterisks indicate the levelof significance: * p<0.10, ** p<0.05, *** p<0.01. The coefficients are in the first row. The specification in this table includes 58census division and 47 cohort fixed effects. Other variables included are an indicator for gender, missing information for uncleor aunts attendance, uncle or aunts attendance, missing information for siblings attendance, latitude and distance to the closestcity. All columns use predicted cousin's attendance to account for observed selection. The first column includes the wholesample, while the second only includes respondents that had no grandparents attend residential school or were missing grand-parental information. In the third column only includes individuals who had at least one grandmother and one grandfatherattend residential school.121Chapter 5The Size, Structure and Change in Aboriginal Earnings Gaps in 1995 and 20055.1 IntroductionIn 2006 approximately 4 percent of the Canadian population self-identified as Aboriginal(Statistics Canada 2008a). Aboriginal people are the youngest and fastest growing ethnicgroup in Canada and have notably lower annual earnings than the average Canadian. Esti-mates of how much lower varies substantially depending on the sample, gender and how anAboriginal person is defined. Estimates of the earnings gap vary from 7 to 63 percent formen and 2 to 15 percent for women (Mueller 2004). Samples that include only full-time/full-year workers such as DeSilva (1999) and George and Kuhn (1994) produce much smallerestimates of the earnings gap then samples that include all workers (such as Pendakur andPendakur 2002). Forty percent of Aboriginal people live in government defined communitiesknown as reservations or reserves, which have varying degrees of self-governance. Thoseeligible to live in these areas are known as North American Indians (NAIs) or First Na-tions. Those NAIs who live on-reserve typically have significantly lower earnings than theiroff-reserve counterparts. Drost and Richards (2003) estimate the median on-reserve incomepenalty to be approximately 40 percent in 1995.101Differences in labour market earningsbetween Aboriginal and non-minority Canadians may indicate social exclusion and discrimi-nation, while differences between Aboriginal people on- and off-reserve speak to the relativeeconomic state of NAI communities.101George and Kuhn (1994) have estimated the mean on-reserve earnings penalty to beabout 14 percent for men using the 1986 Census. They find the on-reserve penalty is onlyabout 8 percent for women.1225.1. IntroductionIn this chapter I compare the log of annual earnings of the M?tis and NAIs living off-reserve to non-minority Canadians and study the earnings penalty associated with livingon-reserve in 1995 and 2005 using the confidential Census Long Form files. I then considerthree possible sources of these earnings gaps and their changes over time: differences in i)weeks worked, ii) weekly earnings due to observable characteristics, and iii) the returns tothese characteristics. I compare the log of these annual earnings differences to their weeklycounterparts to better understand how the number of weeks worked in a year contributes tothe levels and changes of these earning differences. To understand the remaining differencesand their changes over time, I decompose the log weekly earnings gap into factors attributableto differences in characteristics, both individual and job, and returns to these characteristicsusing the Oaxaca (1973) and Blinder (1973) decomposition (onward referred to as the Oaxacadecomposition).102Finally, I use newly available tax information in the 2006 Census toanalyze the role of taxes and transfers in reducing the log earnings gap.I find that differences in weeks worked plays an important role in the earnings gap andthe on-reserve earnings penalty. My results suggest that eliminating differences in weeksworked throughout the year would reduce the on-reserve annual earnings penalty and theAboriginal/non-minority earnings gap by as much as half. Most previous work focuses on theearnings of full-time/full-year workers103and broadening the scope of the analysis allows fora more comprehensive assessment of disparity. Not only are Aboriginal people more likelyto engage in seasonal work, but unemployment is higher for all Aboriginal people acrossCanada (Drost 1994; Hall 2001; Menlson 2004; White, Spence and Downie 2007; Walters,White and Maxim 2004). Higher amounts of seasonal work and unemployment imply thatAboriginal people are more likely to work fewer weeks out of the year than non-aboriginalpeople, and as my results demonstrate, this is a major source of the log earnings gap.102Previous literature that decomposes that Aboriginal earnings gap includes DeSilva (1999),George and Kuhn (1994), Mueller (2004), and Patrionos and Sakellariou (1992). For moreliterature on Aboriginal earnings and income disparity see Drost and Richards (2003), Kuhnand Sweetman (2002), Macdonald and Wilson (2010), and Pendakur and Pendakur (2010,2007, 2002, 1998).103Work that falls into this category includes DeSilva (1999); George and Kuhn (1994);Mueller (2004); Patrionos and Sakellariou (1992).1235.1. IntroductionThe remainder of the gap is attributable to log weekly earnings. I build on previousliterature on the structure of wage differentials between Aboriginal and non-aboriginal people(DeSilva 1999; George and Kuhn 1994; Mueller 2004; Patrionos and Sakellariou 1992) and theon-reserve earnings penalty (Drost and Richards 2003; George and Kuhn 1994) by examininga larger and more recent data set and provide the first decomposition that distinguishesbetween the M?tis and NAIs. Recent work by Pendakur and Pendakur (2010) has clearlydemonstrated the importance of separating NAIs and the M?tis when considering log earningsand income differences. My results from the Oaxaca decomposition attributes between 45to 60 percent of the log weekly earnings gap between Aboriginal and non-minorities todifferences in observable characteristics for both the M?tis and NAIs, in line with earlierresults. On the other hand, nearly all of the on-reserve log weekly earnings penalty is dueto differences in returns to job characteristics.I find that earnings differences have generally declined between M?tis and non-aboriginalpeople for both men and women over this time period, while NAIs kept pace with non-aboriginal people. However, the log on/off-reserve weekly earnings penalty has increasedsignificantly for men. An increase in the relative number of weeks worked by the M?tiscan explain a substantial amount of the decline in the M?tis/non-aboriginal earnings gapbetween 1995 and 2005. The results also suggest that improvements in education and jobtype explain 50 percent of the catch-up in weekly earnings for M?tis men, while it explainsall of the catch-up for M?tis women. All of the increase in the on-reserve log weekly earningspenalty is attributed to differences in returns, specifically to returns for the types of jobsperformed on-reserve.Some of these results should be interpreted with caution. Specifically, estimates usingLee's (2009) bounding method suggests that the increase in the on-reserve penalty couldbe explained by changes in the proportion of NAIs with positive earnings living off-reserve.However, changing proportions of people with positive earnings are not significant factorsin the other results. DiNardo, Fortin and Lemieux's (1996) re-weighting method suggeststhat changes in the composition of the Aboriginal population in terms of registered Indianstatus, band membership, and ethnic composition do not account for any of the trends in1245.2. Datathe log earnings gaps changes once endowments, such as education, are accounted for.While taxes and transfers significantly reduce the earnings gap, they do not eliminate it inmost cases. While the on-reserve earnings penalty for men is reduced, it is completely elimi-nated for women. In fact, taxes and transfers reduce the earnings gap more for women thanmen in general. Sharpe et al. (2009) estimated that the Aboriginal population received 6.2billion dollars more in transfer payments and other social services than all other Canadiansdue to substantial labour market and social inequities. Despite this and Canada's progres-sive income tax system, my results suggest the transfers and redistributive tax policies areinsufficient to eliminate total income differences in most cases.This chapter is structured as follows. Section 5.2 describes the data used while Section5.3 discusses the methodology and the main results: namely the size, structure and changeof the Aboriginal earnings gap and on-reserve wage penalty from 1995 to 2005. Section5.4 elabourates on the main results by providing a sensitivity analysis. Finally, Section 5.5investigates the effect of taxes and transfers on observed earnings difference in 2005. Thechapter concludes with a discussion of the results.5.2 DataI use the 1996 and 2006 confidential Canadian Long Form Census files to assess patterns inthe log annual earnings gap between Aboriginal and non-minority Canadians.104This dataset overcomes a number of limitations faced when using public use data. First, the confiden-tial Long Form Census data includes a 20 percent sample of the entire Canadian populationand includes the largest sample of Aboriginal workers. In fact, 100 percent of householdsliving on-reserves that participate in the Census complete the Long form Census. This avoidsthe small sample size problems sometimes faced in the Census Public Use Microdata File(PUMF) or the Survey of Labour and Income Dynamics (SLID). In addition, the confiden-tial Long Form Census data allows on-reserve populations to be identified separately from104That the all income variables are given for the previous year, hence why all income resultsare for 1995 and 2005.1255.2. Dataother Aboriginal groups unlike with publicly available data. The final advantage of theseconfidential files is that they are the only recent files that consistently define the Aboriginalpopulation over a ten year period.I restrict the entire sample to individuals between 25 and 55 years of age to avoid peo-ple who are still in school or considering retirement and to individuals whose main sourcesof earnings were from salaries or wages over the past calendar year.105,106Since I am ulti-mately interested in comparing Aboriginal people to the majority population, I limit thenon-aboriginal population to non-minority, non-immigrant Canadians. The dependent vari-able is gross weekly earnings and salaries before deductions. This includes allowances, tips,commissions and cash bonuses as well as all types of casual earnings in the previous cal-endar year. Information on the annual hours of work is limited in the Census, so I followBoudarbat, Lemieux and Riddell (2010) and use weekly earnings as the principal measureof wages. For clarity, I will use the term weekly earnings instead of wages. Finally, thewage incomes of those who receive less than 100 dollars of annual wages and salaries are setto zero to avoid issues with outliers.107I make two adjustments to the on-reserve population for the purpose of valid histori-cal comparisons. First, I include only those reserve communities that participated in bothCensus years. In 1996, 77 reserves refused to participate, while in 2006 only 22 refused.The 2006 Census provides an indicator for those areas that should be dropped when makingcomparisons between these years. This adjustment excludes 63 reserves in total. The sec-ond adjustment is to the on-reserve classification. The Census uses the Aboriginal Affairsand Northern Development's definition of on-reserve which changed between these years.Again, the 2006 Census provides an indicator that allows for a definition of reserve compara-105These individuals are classified as paid workers working for salaries, weekly earnings,commission or tips in the job they held last week before the Census week for which theyworked the most hours. If they were not employed, this was for the job of the longestduration. This excludes currently paid workers that were formally self-employed.106It is important to note that this restriction is non-trivial. There is a difference in the self-employment rates of Aboriginal people and non-aboriginal people but the difference is small.For example, results from the labour force survey suggest that approximately 11 percent ofAboriginal people off-reserve are self-employed while 15 percent of non-aboriginal people are(Usalcas 2011). This difference may be significantly more for those living on-reserve107This restriction has no significant effect on the results.1265.2. Datable to 1996.108I use these indicators in combinations with Census subdivision codes in orderto construct a consistent definition of on-reserve in both years. To my knowledge, this isthe first chapter to make these adjustments. If these adjustments are not made, the observedchanges in the on-reserve earnings penalty could be due to changes in how the on-reservepopulation is defined.109Consistency issues also arise between the two years in how income sources are measured.In the 2006 Census respondents could grant permission for their income information to betaken directly from their tax records. Slightly over 80 percent of the population chose thisoption. Since this analysis focuses on earnings differences between groups, this change willnot matter as long as it affects Aboriginal, non-minority and on-/off-reserve populations inthe same way. Statistics Canada (2008b) reports that this change affects very low incomeearners the most significantly, but this is far less of an issue for full-time, full-year workers.This analysis has been repeated using only this subgroup with similar results.Finally, unlike the 1996 Census, the 2006 Census does not allow one to construct yearsof schooling for people with less than a high school education. For this reason, I will use fiveeducation categories (Boudarbat, Lemieux and Riddell 2010) for most of the analysis. Thefive categories are: less than a high school diploma, high school diploma, post-secondary de-gree or diploma below a university bachelor's degree (including trade certificates), universitybachelor's degree, and postgraduate degree. While this may have minimal impact for theCanadian population in general, it could matter substantially for the Aboriginal population.I show the robustness of the results in 1995 by, including a decomposition of weekly earnings108Specifically, I use the variable B06_adjustedBase_IR_PP_19962006 which identifiesthe set of consistently included reserves in 1996 and 2006. To adjust for the changingdefinition of reserve I use the variable B06_Area_Residence_9606 which identifies thecommunities which were counted as reserve communities in one year and not the other.However, the results are not sensitive to this adjustment or to adjustments in the reservedefinition.109It is important to note that all the results presented here are averages for a specificsubset of reserves defined consistently over this ten year period. It does not mean thatthe on-reserve earnings penalty has increased for reserves not included in this analysis orfor specific subsets of these reserves. It also does not imply that certain types of reservesare faring worse relative to their off-reserve counterparts. O'Sullivan and McHardy (2007)demonstrate that over 1981 to 2001 there were significantly different rates of improvementin the Community Well-Being Index in different communities. This could also be true forearnings.1275.2. Datathat accounts for years of education and potential experience110in the appendix.111I includeage as a proxy for experience in the main specifications.The Census defines the Aboriginal population of Canada in two ways: Aboriginal ancestryand Aboriginal identity. The first is derived from the ethnic origin question in the Census.The second is the response to an explicit question regarding whether the respondent is NAI,M?tis or Inuit. I use the concept of Aboriginal identity rather than Aboriginal ancestry asit more closely relates to the population of interest.112I exclude the Inuit and individualsliving in the Territories because of small sample sizes and the fact their unique economiccircumstances are beyond the scope of this chapter.I divide the Aboriginal identity population into three groups: M?tis, NAIs off-reserve andNAIs on-reserve. I exclude individuals with multiple Aboriginal identities from the analysisin order to simplify the discussion. This is only a small percentage (approximately 3 percent)of the identity population and does not make a substantial difference in the results. The on-reserve earnings penalty is constructed by comparing NAIs off- and on-reserve. A robustnesstest was conducted limiting the population to registered Indians with band membership.This more homogeneous population allows for a more explicit focus on the on-/off-reservecomparison, but I find the definition can be expanded without affecting any of the mainfindings. Final weighted sample sizes for each group in each year are given in Table 5.1.113The summary statistics presented in Table 5.1 suggest Aboriginal people are much morelikely to work in government than non-minority individuals. Aboriginal people also engagein more part-time and/or part-year work than non-minority Canadians. Aboriginal peopleare on average less likely to have a bachelor's degree and are more likely to live in ruralareas and in the Western provinces. They are also more likely to be disabled, especially110Potential experience is defined by economists to as age - years of education - 6 and ismeant to proxy for actual experience.111The results in appendix table F.1 suggest that although the importance of education maybe underestimated using only five broad categories, the nature of the results are not affected.112Specifically, I use the question that asks whether the individual in the household Is thisperson an Aboriginal person, that is, North American Indian, M?tis or Inuit (Eskimo)? If Yes, mark ? the circle(s) that best describe(s) this person now. The options are No,Yes, North American Indian, Yes, M?tis, Yes, Inuit (Eskimo)113The weighted sample sizes are given because use of the confidential files prohibits publi-cation of unweighted frequencies. The weights used are the population weights given by theCensus.1285.2. Datathose living on-reserve. Although they are less likely to be married, they have more children.Comparing individuals who identify as M?tis and NAI to non-minority Canadians confirmsthe standard findings: earnings differences between the M?tis and non-minority Canadiansare smaller than those between non-minorities and NAIs. On the other hand, NAIs livingoff-reserve have higher earnings than those on-reserve.Table 5.1 also previews one of the first results of this chapter. Real annual earningsdifferences are in general much higher than differences in real weekly earnings. Consistentwith broader patterns in the literature on earnings inequality, there is less disparity betweenwomen of different groups (Pendakur and Pendakur (2007)). In addition, both the M?tis andNAI off-reserve made progress in closing the Aboriginal/non-minority earnings gap between1995 and 2005. Notably, the M?tis earnings gap reduced by nearly half over this time period.While NAIs have been gaining ground off-reserve relative to the non-minority population,those remaining on-reserve are falling behind. Table 5.1 also suggests that both these phe-nomena are at least partially explained by patterns in weeks worked. The fifth row of Table5.1 shows that Aboriginal groups and those on-reserve work fewer weeks out of the year thannon-aboriginal and off-reserve individuals.Table 5.1 makes it clear that real earnings differences exist and have been changingsignificantly over time. However, it is not clear how important weeks worked are in theseearnings differences. Neither is it clear how the substantial differences in characteristicsreported in Table 5.1 translate to differences in weekly earnings. The next section lays outa framework to investigate the relative importance of weeks worked in the earnings gap andthe importance of differences in characteristics in the weekly earnings gap.1295.3. Methodology and Results5.3 Methodology and Results5.3.1 Methodology: Decomposing Differences in Earnings into Differences inWeeks Worked and Weekly EarningsTo understand the importance of weeks worked in generating annual earnings differences Istart by considering the composition of an individuals earnings.Consideregi = wgi ? wksgiwhere i indexes individuals and g is an indicator variable for the group of interest. Annualearnings is given by ei, weekly earnings is represented by wi, and wksi represents weeksworked during the year. When comparing M?tis (NAI) to non-minority people, g = 1 ifan individual is M?tis (NAI) and zero otherwise. When calculating the on-reserve penalty,g = 1 if an individual is on-reserve and zero other wise.For comparability with the literature on earnings and wages I take the natural logarithmof this equation which yieldslnegi = lnwgi + lnwksgi . (5.1)Taking expectations and then subtracting Equation 5.1 for group one from the expectedvalue of the equation for group zero results in:(E(lneg=0i )? E(lneg=1j )) = (E(ln wg=0i )? E(ln wg=1j )) + (E(lnwksg=0i )? E(lnwksg=1j ))(E(lnwksg=0i )? E(lnwksg=1j )) = (E(lneg=0i )? E(ln wg=0i ))? (E(lneg=1i )? E(ln wg=1j )).(5.2)The first component in line 1 of Equation 5.2 - (E(lneg=0i )?E(lneg=1j )) - is known as the logearnings gap and is what is commonly estimated in the literature.114The second component114Note that for small numbers, the log earnings differences can be interpreted as percentagedifferences in earnings. However, Halvorsen and Palmquist (1980) show that this interpreta-1305.3. Methodology and Resultsis the log wage (log weekly earnings gap). Re-arranging the equation demonstrates thatcomparing the log earnings and log weekly earnings gap is equivalent to comparing the logweeks worked gap which is done in the first rows of Table F.2 in Section 5.3.3.5.3.2 Methodology: Decomposing Weekly Earnings in Differences inCharacteristics and Differences in the Wage Structure By YearA substantial fraction of the Aboriginal earnings gap/on-reserve penalty is due to differencesin weekly earnings. To understand the basic factors which contribute to these differences Iuse the Oaxaca decomposition.This decomposition accounts for two factors that contribute to differences in weeklyearnings: differences in the returns to certain characteristics, and differences in the quantitiesof these characteristics. For example, Aboriginal people may have lower weekly earningsthan non-aboriginal people because they have less education, or because they are paid lessfor each year of education. The Oaxaca decomposition separates the two possibilities. I basemy discussion of the decomposition on Benjamin, Gunderson, Lemieux and Riddell (2007,367).Individual weekly earnings are set according to the standard human capital model givenin Equation 5.3.lnwg=0i = ?0X?0i + ?0i and ln wig=1 = ?1X?1i + ?1i (5.3)where X1i and X0i are Aboriginal and non-aboriginal's productive characteristics, ?1 and?0 are the rates of return for these characteristics, and ?0i and ?1j are error terms with anexpected value of zero. Taking the difference between Equation 5.3 for each group and takingexpectations results in the expression:tion is not strictly true for dummy variables and is only approximately true for coefficientsclose to zero. The further from zero our estimates, the more of an approximation this be-comes. Since some of the estimates are rather large, all differences are assumed to be in logpoint differences.1315.3. Methodology and ResultsE(lnwg=0i )? E(lnwg=1i ) = lnw0 ? lnw1 = ?0X?0 ? ?1X?1,where X0 and X1 are the expected values of X1i and X0i. By adding and subtracting theterm ?0X1 and rearranging, it is easy to show thatlnw0 ? lnw1 = ?0(X0 ?X1)?? ?? ?"Differences in Characteristics"+ (?0 ? ?1)X?0? ?? ?"Differences in Returns". (5.4)In the absence of differences in labour market returns (or unmeasured characteristics) logweekly earnings differences should arise only from differences in X0 and X1. The component?0(X0 ?X1) is often called the explained component of the differential or the part of thedifferential due to pre-market characteristics. The component (?0 ? ?1)X1 is often calledthe unexplained component (or the differences in labour market returns).115The variables that make up Xo and X1 in the decomposition include:1161. Experience: Age; age squared.2. Education: Five education categories (less than a high school diploma; high schooldiploma; post-secondary degree or diploma below a university bachelor's degree (in-cluding trade certificates); university bachelor's degree; postgraduate degree).3. Disability: An individual reported a long term disability or handicap, or reported atleast one activity limitation either at home, school or work.4. Household Composition: An indicator if an individual is married; the presence of chil-dren in the household; the number of children present.115The estimates computed here are the estimates of the differences in measured pricesand skills. If reserves suffer from brain-drain (high skilled people leave the reserve) andaccounting for differences in education and other characteristics does not fully account forthis, then the calculated returns to moving off-reserve will be biased. In other words,we would interpret differences in unmeasured skills between people on- and off-reserve asdifferences in prices.116When I decompose the on-reserve weekly earnings penalty, I also include indicators forband membership and registered Indian status under the heading Legal.1325.3. Methodology and Results5. Language: Can speak English and an indicator of whether their first language was anofficial language.6. Geography: Whether an individual lives in a major metropolitan region; whether theindividual lives in a rural area; the population of the community; province indicators.7. Work: Indicator of part-time; four categories of weeks worked out of the year; nineindustry dummies.117If a significant fraction of the low-skilled population does not have positive earnings and this isdifferent between Aboriginal and non-minority Canadians - once all the observable variablesare accounted for - the estimates will still be biased.118Hull (2001) provides evidence thisis not a concern if differences in education are accounted for and if the missing data is fromdifferences in participation rates. To deal with categorical variables I follow Gardeazabaland Ugidos (2004) and Yun (2005).119It should be noted that the Oaxaca decomposition is typically not used when the group ofinterest (g) is a choice variable. Obviously, whether or not to live on-reserve is a choice andit may be affected by observable and unobservable characteristics. However, this is only anissue when it comes to interpretation of the results. For example, the decomposition will beunable to distinguish between differences in returns being due to wage structure differenceson- and off-reserve and differences in unobservable skills on- and off-reserve. This is discussedin more detail in Fortin, Lemieux and Firpo (2011) and should be kept in mind during theremainder of this chapter.117This analysis has also been run with 16 industry categories, but only 9 are used in orderto provide a more consistent definition of industries between 1996 and 2006. In 1996 onlythe 1980 Standard Industrial Classification is used to categorize industries, while in 2006 the2002 North American Industry Classification System (NAICS) codes are used. I attemptto provide as a consistent definition as possible using the correspondence tables availablefrom Statistics Canada at http://www.statcan.gc.ca/concepts/concordances-classifications-eng.htm (retrieved June 6, 2012).118More technically, the interpretation of the Oaxaca decomposition above is only valid ifthe conditional distribution of unobservable skills for Aboriginal and non-minorities is thesame.119This adjusts the contribution of the categorical variables so it is invariant to the basecategory by restricting the coefficients of to sum zero and expresses the effects as deviationsfrom a grand mean.1335.3. Methodology and Results5.3.3 ResultsThe log annual and log weekly earnings gaps, given by the first and second elements ofthe first line of Equation 5.2, are reported in the first two rows of Table 5.2. These resultsconfirm what was observed in levels in Table 5.1, namely, the annual and weekly earnings gapsare larger for NAIs than for M?tis and the on-reserve earnings penalty is significant.120Inaddition, female and male M?tis earnings gaps are statistically equivalent, yet the differencebetween NAI and non-minority women is smaller than between NAI and non-minority men.The female on-reserve penalty is less than half the size of the male penalty.The weeks worked gap, given in the last line of Equation 5.2, can be directly inferredby subtracting the two estimates in the first two rows of Table 5.2. These results suggestdifferences in the weeks worked gap account at least a third of the log earnings differences inall cases and account for more than half of the log earnings gap between non-minority andM?tis men.Second, the log annual earnings gap for M?tis men and women has decreased by half be-tween 1995 and 2005. The log weekly earnings gap has decreased, though not as significantlysuggesting that part of the catch-up has been in weeks worked.121The change in log weeklyearnings and log annual earnings is significant at all conventional significance levels. The logannual earnings gap has fallen between male NAIs and non-minorities while the log weeklyearnings gap has remained constant.122This suggests that all of the decline in the annualearnings gap was due to weeks worked.The on-reserve earnings penalty is a different story. Both the log annual and weeklyearnings penalty have increased, with the on-reserve weekly earnings penalty increasing bynearly 50 percent for men.123The log annual earnings gap has remained relatively constant120The p-value for the tests of the claim the M?tis earnings gap is smaller than the NAIs'for men is 0.113 in 1995 and less than 0.001 in 2005. The p-value for the test of whether theon-reserve penality is equal to zero is less than 0.001 in both years.121The test statistic men's for the change in the log earnings and log weekly earnings gapsare 12.25, 2.15 respectively. Both yield a p-values less than 0.0005.122The test statistics for whether there was a change in the gap between the two years are-1.36 for log weekly earnings gap and 3.98 for the log annual earnings gap.123The increases are statistically significant with p-value of 0.002 for log annual earningsand less than 0.0001 for log weekly earnings.1345.3. Methodology and Resultsfor women, while the log weekly earnings has increased significantly.124The next two rows of Table 5.2 show the decomposition of the log weekly earnings gapinto differences in characteristics (the third row in Table 5.2 and given by the first elementof Equation 5.4) and returns to those characteristics (the fourth row in Table 5.2 and givenby the second element of Equation 5.4). In both years, approximately 50 to 60 percentof the differences in log weekly earnings for male M?tis and NAIs are due to differences incharacteristics, mirroring the results of George and Kuhn (1994), DeSilva (1999), and Mueller(2004).125While both differences in characteristics and returns declined for M?tis, differencesin characteristics declined at a slightly faster rate, while characteristics and returns stayedrelatively constant for NAIs off-reserve. On the other hand, the on-reserve wage penalty islargely explained by differences in returns to various endowments. Nearly 80 percent of thelog weekly earnings penalty is accounted for by differences in returns in 1995, while in 2005they account for 90 percent.126Appendix breaks down characteristics and returns into the variable categories listedabove. For both M?tis and off-reserve NAIs, education is the single largest contributingfactor to log weekly earnings differences. The fact that both these groups are younger thanthe non-minority population is also important. Both M?tis and NAIs are more likely tohave some form of disability than other Canadians. Although its total contribution to thelog weekly earnings differences is small, it is still statistically significant. Differences inhousehold structure are also associated with weekly earnings inequality. Although this re-lationship is in part due to lower wage earners being less likely to form families, there is alarge economic literature that argues family structure has a significant causal effect on weeklyearnings (see Ribar (2004) for a review). For men, differences in job type are important for124The test statistics for at test of whether there has been a change in the log earnings gapfor women is -0.35, while the test statistic for the test of whether the log of weekly earningshas changed is 2.25.125George and Kuhn (1994) find that endowment differences can explain about fifty per-cent of the gap. DeSilva (1999) and Mueller (2004), who use the 1991 and 1996 Censusrespectively, find that endowment differences explain a slightly larger proportion of the logearnings gap. Mueller (2004) uses the 1996 PUMF and finds the endowments may explainas much as 64 percent of the gap. DeSilva (1999) uses the 1991 PUMF and finds endowmentdifferences may explain as much as 84 percent of the gap.126The associated test statistic is 2.14 with an associated p-value of 0.03.1355.3. Methodology and Resultsmale NAIs off-reserve and M?tis women. Differences in geography actually play a relativelyminor role in weekly earnings differences as the previous literature on community well-beingand community location would suggest (White and Maxim 2007).There are also important differences in the returns of various characteristics. First, lowerreturns to age is an important contributing factor for female M?tis and NAIs. This may bea consequence of differences in work experience per year relative to non-minority women.There is some evidence that the same is true for M?tis men in 2005 and the on-/off-reserveearnings gap in 1995. One of the most robust patterns for M?tis men seems to be differencesin the coefficients related to family formation. M?tis seem to have lower returns to familyformation than the average Canadian. Interestingly, on-reserve NAI women have greaterreturns to family formation than those off-reserve. This may be due to those remainingon-reserve being able to form better social support systems (White and Maxim 2003). BothM?tis and NAIs off-reserve have the same return to education as the non-minority population,confirming previous results in the literature (Pendakur and Pendakur 2010; Walters, Whiteand Maxim. 2004).The largest contributing factor to unexplained earnings differences between those that liveon- and off-reserve is the return to the industry one works in. In 1995, there are significantdifferences in the returns to education for women, which are eliminated in 2005. George andKuhn (1994) demonstrate that in 1985, there was little, if any return to education on-reservewhile there was a positive return for Aboriginal people off-reserve. This indicates therewere differences in the returns to education on- and off-reserve in 1985 as well. This changein 2005 may be due to the recent efforts to improve Aboriginal education outcomes andquality (White, Spence and Maxim 2005) or because of the changing nature of the returnsto education over this period (Boudarbat, Lemieux and Riddell 2010).The entire source of the on-reserve earnings penalty is due to unexplained differences inreturns on- and off-reserve raises important questions for the sustainability of reserves andtheir economic development. Two possible explanations for these differences are: 1) thatNAIs living off-reserve have more unmeasured skills than those that live on-reserve; or 2)differences in returns are due to differences in the economic environment. The economic1365.3. Methodology and Resultsdevelopment literature contends that most of the income differences between countries arenot due to differences in classical endowments, but differences in cultural and political envi-ronments (Olson 1996). This may also be true for the on-reserve earnings gap. Individualson- and off-reserve are operating in economies with different governance structures (Graham2010) and social capital (Maxim and White 2003). It is possible that differences in these fac-tors are fundamental to understanding differences in the returns between on- and off-reserveNAI groups.The large increase in unexplained variation in weekly earnings is attributed to job typefor women and both geography and job type for men. These increases may have occurred fora number of reasons. One possibility is that the relative barriers to the markets have beenincreasing on-reserve. The other possibility is that those off-reserve are more likely to haveaccess to high paying occupations within industries in 2005, while those on-reserve are not.Another related possibility is that differences in returns to particular jobs and geographyare acting as proxies for an increase in the unmeasured skills of the off-reserve population.It is also feasible that reserves are unable to provide the same monetary compensation forcertain jobs because of weak economic foundations. Finally, if the new group of individualsidentifying as NAIs is more westernized than the old group, we may expect these samepatterns (Kuhn and Sweetman 2002).127127Some clarification is required when interpreting the results in this section and the follow-ing results in Section 5.4. First, all exercises conducted are inherently partial equilibrium.Second, this chapter decomposes log earnings differences into differences in log weeks workedand log weekly earnings and also decomposes differences in log weekly earnings into differ-ences due to characteristics and returns. Therefore, it does not consider the importance ofdifferences in endowments to differences in weeks worked and thus should not be taken asthe total effect of differences in endowments on log annual earnings. The literature on Abo-riginal labour market outcomes have concluded that differences in education are related todifferences in employment and likely weeks worked as well (Drost (1994); Hall (2001); Menl-son (2004); Walters, White and Maxim 2004; White, Spence and Maxim 2005; and White,Spence and Downie 2007). However, this is done implicitly by the re-weighting method usedby DiNardo, Fortin and Lemieux (1996) in Section 5.4.1375.4. Sensitivity of the Results5.4 Sensitivity of the Results5.4.1 Censoring and the Proportion On-ReserveTo determine whether it is possible that a change in the proportion of people with positiveannual earnings may be driving the observed trends, I use a bounding method proposed byLee (2009). The method he proposes suggests trimming observations from one tail of thewage distribution in order to equate the percentage of individuals with positive earningsbetween the two years to bound the effect it could have on the results. For example, if Iobserve a larger proportion of non-minority people with positive earnings in 1995, I trim fromthe upper tail of the non-minority population's earnings distribution in that year. Trimmingthe distribution in this way reduces non-minority log earnings in 1995 and makes both the1995 M?tis/non-minority and NAI off-reserve/non-minority log earnings gap smaller. Thisminimizes the decline in the log earnings gap between 1995 and 2005. For M?tis and NAIoff-reserve I trim from the upper tail of the wage distribution in 2005 in order to minimizethe trend observed in 2005 if the proportion without positive earnings is greater in 1995.I also examine the extent to which the increase in the off-reserve population could ac-count for the increase in the on-reserve log weekly earnings penalty. The idea is similarto the bounding method discussed above, but rather than eliminating excess off-reserveindividuals, I reclassify them as on-reserve. This is the maximum partial-equilibrium effectthat brain-drain or an increasing off-reserve population could have on the results.Table 5.3 demonstrates that changes in censoring for the M?tis/NAIs relative to non-minority Canadians do not affect the general trends. In all cases, the changes are stillstatistically significant.128Thus changes in the proportion of people with positive earningsbetween the years does not drive changes in the earnings gap. This not is true for the on-reserve earnings penalty. This is driven by the large increase in the censoring rate on-reserve.128The test statistic for whether the change in log annual earnings is equal zero, is 14.39for the M?tis and 4.34 for NAIs. Both of these have a corresponding p-value of less than0.001. The corresponding test statistics for the change in log weekly earnings being equal tozero, are 12.09 (p-value less than 0.001) for the M?tis and 3.02 (p-value< 0.005) for NAIs.The test statistic for the change in the on-reserve log weekly earnings penalty is -4.86 whichcorresponds to p-values of 0.064 and less than 0.001.1385.4. Sensitivity of the ResultsThe direct effect of brain-drain or an increased proportion of people off-reserve could alsosubstantially impact the estimates. When the highest earning off-reserve group is assumedto move on-reserve and keep their current earning levels, the on-reserve earnings gap declinesand actually becomes an on-reserve premium. However, this counterfactual is too unrealistic.Only a portion of NAIs living off-reserve are actually eligible to live on-reserve. The thirdpanel of Table 5.3 restricts the analysis to only eligible individuals. Although there is nolonger a statistically significant log weekly earnings premium, there is still a decline in theon-reserve penalty, rather than an increase. Taking these results together implies that theincreased number of people living off-reserve relative to living on may explain the entiretrend in the log annual and weekly earnings gap.129The above suggests that brain-drain could possibly explain the change in the earningspenalty, but previous literature suggests otherwise (Cooke, Clatworthy and Norris 2007;Mendelson 2004; and Statistics Canada 2008a). A quick check of the percentage of peoplewho were on-reserve five years ago (using mobility variables in the Census) and were notin 2006, confirms the increased percentage of Aboriginal people off-reserve is not due tooff-reserve migration. In all likelihood, the main source of the increasing proportion of NAIsliving off-reserve is due to an increase in the number of people self-identifying as Aboriginal.If the increasing on-reserve penalty is merely an artifact of an increasingly high wage/highskilled group of people becoming eligible to live on-reserve or identifying as NAI, it representsan opportunity for reserves to improve their economic situation by attracting people fromthis group to live on-reserve.5.4.2 Adjusting For Legal Changes, Ethnic Drift and Other FactorsThe number of people with Aboriginal identity and ancestry grew dramatically over aten year period; much more than changes in birth and death rates can explain (Statistics129It is important to note that this exercise calculates the maximum impact by assumingNAIs off-reserve can take their jobs with them if they were to move on-reserve. In a similarexercise that moves the appropriate number of individuals from off-reserve to on-reserve, letsthem keep their skills (such as experience and education), but gives them the reserve wagestructure, the effect would be much more modest. The true importance of this phenomenonlikely lies between these two scenarios.1395.4. Sensitivity of the ResultsCanada 2008; Guimond 2003). Legal changes in 1986 allowed many more NAIs with non-aboriginal backgrounds to claim Indian status (Clatworthy 2007), while increased awarenessregarding the existence of the M?tis (Royal Commission on Aboriginal People 1996) andlegal advancements in their hunting rights (R. v. Powley, [2003] SCC) may have inducedmore westernized individuals to self-identify as Aboriginal.To deal with both these possibilities, I use the semi-parametric re-weighting methodproposed by DiNardo, Fortin and Lemieux (1996). Their re-weighting procedure allows meto address the question: What would the 2005 earnings gap be if the M?tis and NAIson- and off-reserve had the same distribution of endowments, legal status and ethnicity asthey did in 1995?. The theoretical background of this procedure is described in detail inDiNardo, Fortin and Lemieux (1996) and Fortin, Lemieux and Firpo (2011). I estimatethe relevant weights for this procedure using a linear probability model130including all thevariables discussed in Section 5.2 plus other variables related to ethnic drift and changes inlegal structure.131Table 5.4 presents the results of this exercise. The first row of the first and second panelreproduce the log of actual earnings differences. The third row of coefficients in each ofthese panels reproduces differences in log weekly earnings from Section 5.3. The row labeledRe-Wgted Log Annual Earnings (No Legal or Ethnic) weights the population such thatthe distribution of the endowments discussed in the data section are the same as in 1995.The final row, labeled Re-Wgted Log Annual Earnings (Legal or Ethnic), re-weights thepopulation for the same endowments as the previous row, but also accounts for changes inAboriginality using information on legal status and ethnic background.In the first row of Table 5.4, we can see that adjusting the M?tis in 2005 to have thesame characteristics as they did in 1995 reduces the increase in the earnings gap and elim-130If the probability predicted is greater or equal to one, I replace the probability with 0.99.If the probability is less than or equal to zero, I replace the probability to 0.01.131These additional variables include multiple ethnic identification (Aboriginal and otherethnic origins, such as British), an indicator of whether an individual's mother tongue is anAboriginal language, registration status and band membership. In addition, I allow changesin registration, band membership and ethnic identification to vary by education, province,age, and disability. I also allow the relationship of registration, band membership and ethnicidentification to vary.1405.5. The Role of Taxes and Transfersinates the decline in the weekly earnings gap.132The same adjustment for off-reserve NAIseliminates the decline in the earnings gap. For women, re-weighting the M?tis to have thesame endowments as in 1995 completely eliminates any gains made. The gap for NAI womenwould have been even larger if they had the same endowments as previously. This suggeststhat increases in endowments of NAI women off-reserve have allowed them to keep pace withnon-minority Canadians.The last three columns of Table 5.4 adjust NAIs on- and off-reserve to have the samecharacteristics and Aboriginality. The first of these columns, titled NAI Off - NAI On (1),re-weights only the on-reserve populations to have its 1995 distribution of characteristics.The next column, NAI Off - NAI On (2) re-weights only the off-reserve population. Thefinal column re-weights both of the populations.Changing composition of the on-/off-reserve population can explain the increase in thelog annual earnings penalty, but not the male weekly earnings gap. By comparing panelsentitled, NAI Off - NAI On (1), and NAI Off - NAI On (2), it becomes clear that thisdecline is due to positive changes in endowments of those off-reserve. This implies that theincrease in the earnings gap was partly due to greater increases in the skills of the off-reservepopulation relative to the on-reserve population. Comparing rows two and three in thesecolumns suggests that re-weighting for ethnic and legal characteristics does nothing to reducethe earnings gap once all other endowments are accounted for. In summary, this subsectionillustrates that part of the reason for the decline in the Aboriginal/non-minority earningsgap and the increase in the on-reserve weekly earnings penalty is the changing compositionof the Aboriginal population, as the Oaxaca decomposition results suggest, but that changesin legal and ethnic compositions are not the driving forces once other factors are considered.5.5 The Role of Taxes and TransfersMany individuals on-reserve may not have to pay some forms of taxes and receive formsof government assistance that those off-reserve do not. It is possible that people move on-132The test statistic against the null hypothesis of the change in annual earnings equal tozero for the M?tis is 6.62.1415.5. The Role of Taxes and Transfersand off-reserve to minimize their tax burden and the observed earnings gap is merely atax wedge. NAIs off-reserve and the M?tis also receive additional benefits not available tothe general population which may fully compensate for their lower relative labour marketearnings.Although data on after-tax income is not available in the 1996 Census files, data on taxespaid and transfers received in the previous year are available in the 2006 Census. Using this2005 income information, I conduct a brief analysis of the impact of taxes and transfers onthe log earnings gap.In addition to the variable used in most of this analysis (income from wages and salaries), Iwill be using measurements of total income, total income from government transfer paymentsand after-tax income. Total income includes income from all possible sources includinginvestments and government transfers.133After-tax income refers to total income minusfederal, provincial and territorial income taxes paid in that year. If the respondent givesconsent, all income data from 2005 is collected from Canada Revenue Agency. Approximately80 percent of people gave permission for their tax records to be used. The rest of the taxinformation is obtained by a direct question to the respondents on taxes paid. As in thepreviously, I set all individual's income from these various sources to 0 if they have less than100 dollars of income from these sources.To analyze how taxes and transfers affect the log earnings gap, I use the same approachas in Table 5.2 to estimate earning differences.134I then repeat this for log total income, logtotal income minus government transfers, and log total income after taxes. Note that the logof total income minus government transfers is equivalent to the log of employment incomeplus income from investments and other non-government sources. By comparing the logdifferences of the various earnings measures, it is possible to see how government transfers,taxes, and other income augment the earnings gap. The first row of Table 5.5 gives the133Government transfer payments include payments from Old Age Security, Guaranteed In-come Supplement Benefits, Canada and Quebec Pension Plan Benefits, Employment Insur-ance, Child Benefits and all other transfers from received from federal, provincial, territorialor municipal programs.134This is given by the first component of Equation 5.2 as before.1425.6. Conclusionlog annual earnings gap reported in Section 5.3, the second row the gap in the log of totalincome minus government transfers, the third the log total total income gap, and finally thelast row provides the gap in the log total income after-tax.To see how other income changes the earnings gap, I compare the gaps in the first row tothe gaps in the second. In general, adding other sources of income leaves the earnings gaprelatively unchanged. This suggests that inequalities in other income sources are minor. Theimportance of transfers in reducing the log annual earnings gap can be inferred by comparingthe third row to the second. The third row differs from the second only in the addition ofgovernment transfers. In all cases the log total income gap is smaller than than log totalearnings gap. For women, the log earnings gap is reduced by approximately 0.10 log pointsfor most comparisons.135This is roughly a third of the earnings gap. The reductions inthe log earnings gap due to transfers is much smaller for men; approximately only 0.05 logpoints. Transfers may matter more for women because of the assistance available for motherswith dependent children.Comparing the last row to the third makes it clear that post-tax earnings differences areeven smaller than post-transfer differences alone. However, differences in taxes play a muchsmaller role than transfer payments in reducing the log earnings gap. In general however,both sets of fiscal policies are not enough to eliminate the total income gap for most groups.The exception is that transfers and taxes statistically eliminate the on-reserve total incomepenalty for women.5.6 ConclusionIn this chapter I analyze the size, structure and change in the log earnings gap andthe on-reserve log earnings penalty. I consider three possible sources of these log earningsdifferences: differences in i) weeks worked, ii) weekly earnings due to characteristics, and iii)the returns to these characteristics.135The reduction in the earnings gaps are statistically significant with p-values less than0.05 in all cases. The fact the total income gap is lower replicates the finding of Pendukarand Pendukar (2010). The standard errors are larger in these estimates because earnings arerounded to the nearest 100 dollar for confidentially reasons.1435.6. ConclusionI find that there has been an increase in the log weekly earnings gap between thoseliving on- and off-reserve and a decline in the log earnings gap between M?tis and non-minorities. These patterns are consistent between men and women, but women on averageexperience smaller earnings differences. I find that the differences in weeks worked explaina large portion of the log earnings gap for all groups and identify it as a large reason for thedecrease in the log earnings gap between M?tis and non-minorities.The Oaxaca decomposition results suggest that about 50 percent of the difference in logweekly earnings between the off-reserve Aboriginal and non-minority groups can be explainedby differences in observable characteristics. On the other hand, the vast majority of themale on-reserve earnings penalty is due to unexplained differences in returns. Unexplaineddifferences in returns are also the primary reason for the increase in the on-reserve log weeklyearnings penalty. This highlights the need for research into the sources of this penalty.Whether the observed difference in the log weekly earnings structure is due to unmeasureddifferences in productivity, social capital, supply and demand, or institutional structures on-and off-reserve is a productive area for future research.The results also suggest that changes in the composition of the M?tis population fullyaccount for changes in the weekly earnings gap for women, but not for men. Allowing fora changing degree of Aboriginality and legal status matters little once other factors such asage and education are accounted for. This suggests that the decreasing gap for the M?tisand the increasing gap between on- and off-reserve NAIs is not due to changes in ethnicityand legal factors all else equal. These sensitivity checks also indicate that positive changesin the composition of the off-reserve population may be very important. The increase in theon-reserve annual earnings penalty can be explained by changes in the composition and/orproportion of people with positive incomes living on- and off-reserve. Finally, my findingssuggest that while taxes and transfers reduce the log earnings gap, they neither eliminate thedisparity between non-minority Canadians and Aboriginal groups, nor the male on-reserveearnings penalty. On the other hand, they statistically eliminate the on-reserve penalty forwomen.1445.6. ConclusionThere are a number of conclusions may be drawn from these findings. First, earningsdifferences between off-reserve Aboriginal groups and non-minority individuals are decliningand it is not entirely due to the observable composition of the Aboriginal population for theM?tis. Second, increasing the number of weeks worked by Aboriginals so it is the same asthe average non-minority person would decrease the earnings gap by as much as 50 percentholding all else constant. Third, eliminating the differences in endowments (such as educa-tion) between off-reserve Aboriginal people and non-minority Canadians would decrease thelog weekly earnings gap by approximately half. Forth, eliminating endowment differencesbetween on- and off-reserve NAI would not have a substantial effect on the on-reserve logweekly earnings penalty holding all else equal. The gap is fully generated by either currentlyunexplained differences in the earnings structure on-reserve or earnings potential. Finally,the progressive taxation system in Canada and the current structure of government trans-fers does not eliminate total income differences generated by differences in labour marketearnings.1455.7. Tables and Figures5.7 Tables and FiguresBelow are the tables for Chapter 5.Table 5.1: Summary StatisticsMenNot Aboriginal M?tis NAI Off-Reserve NAI On-Reserve1995 2005 1995 2005 1995 2005 1995 20051995 Real Earnings 38527 45856 22424 36642 22056 31551 14096 19437(37.07) (80.16) (232.31) (283.9) (219.34) (267.76) (77.26) (118.55)1995 Real Wages 858 983 604 834 566 750 449 574(1.21) (1.94) (14.83) (9.16) (6.54) (11.7) (3.46) (4.83)Real Earnings Gap 16103 9214 16471 14305 7960 12115(134.69) (182.03) (128.2) (173.96) (148.3) (193.15)Real Wage Gap 253.15 148.69 291.32 233.64 117.4 175.39(8.02) (5.55) (3.87) (6.82) (5) (8.26)Weeks Worked 45.70 47.30 40.14 45.12 39.36 43.40 32.77 37.22(0.014) (0.012) (0.212) (0.120) (0.208) (0.150) (0.109) (0.101)Age 38.889 40.438 37.283 39.158 36.432 38.605 37.227 39.089(0.01) (0.011) (0.117) (0.089) (0.109) (0.096) (0.051) (0.053)Government Job 0.067 0.084 0.084 0.095 0.097 0.107 0.301 0.251(0.00003) (0.00003) (0.004) (0.003) (0.004) (0.003) (0.003) (0.003)BA or Higher 0.187 0.213 0.053 0.084 0.07 0.08 0.022 0.033(0.001) (0.001) (0.003) (0.003) (0.004) (0.003) (0.001) (0.001)Part Time Work 0.057 0.044 0.093 0.061 0.107 0.085 0.129 0.121(0.00005) (0.00005) (0.004) (0.003) (0.004) (0.003) (0.002) (0.002)No Children 0.443 0.483 0.475 0.513 0.485 0.503 0.343 0.37(0.001) (0.001) (0.007) (0.005) (0.007) (0.006) (0.003) (0.003)Married 0.604 0.523 0.471 0.46 0.426 0.38 0.466 0.398(0.001) (0.001) (0.007) (0.005) (0.007) (0.006) (0.003) (0.003)Disabled 0.047 0.104 0.085 0.175 0.09 0.167 0.062 0.107(0.00003) (0.00004) (0.004) (0.004) (0.004) (0.004) (0.002) (0.002)Western Province 0.161 0.175 0.64 0.533 0.316 0.333 0.44 0.417(0.001)* (0.001) (0.007) (0.005) (0.006) (0.005) (0.003) (0.003)Rural Area 0.215 0.201 0.294 0.276 0.231 0.211 0.934 0.876(0.001) (0.001) (0.006) (0.005) (0.005) (0.005) (0.002) (0.003)Weighted Sample Size 4217730 4191070 26910 59700 31700 49310 26590 285701465.7. Tables and FiguresTable 5.1 : Summary Statistics (Continued)WomenNot Aboriginal Metis NAI Off-Reserve NAI On-Reserve1995 2005 1995 2005 1995 2005 1995 20051995 Real Earnings 24368 28878 14527 23206 15362 22140 12943 18451(21.5) (25.76) (176.63) (175.9) (162.07) (186) (73.6) (90.76)1995 Real Wages 566 652 381 551 406 558 356 488(0.87) (0.96) (8.26) (6.08) (6.54) (11.09) (2.99) (4.24)Real Earnings Gap 9840 5671 9006 6737 2419 3689(99.07) (100.83) (91.79) (105.88) (117.84) (138.38)Real Wage Gap 185.08 100.68 160.84 94.05 49.97 70.24(4.57) (3.52) (3.7) (6.03) (4.76) (7.67)Weeks Worked 43.96 45.50 39.58 43.19 39.07 42.18 37.04 40.64(0.016) (0.015) (0.238) (0.142) (0.218) (0.155) (0.124) (0.097)Age 38.915 40.66 37.325 39.452 37.205 38.994 37.195 39.23(0.01) (0.011) (0.122) (0.087) (0.106) (0.09) (0.057) (0.053)Government Job 0.047 0.077 0.068 0.099 0.093 0.118 0.284 0.257(0.00003) (0.00003) (0.004) (0.003) (0.004) (0.003) (0.003) (0.003)BA or Higher 0.195 0.258 0.081 0.129 0.095 0.146 0.067 0.092(0.001) (0.001) (0.004) (0.003) (0.004) (0.004) (0.002) (0.002)Part Time Work 0.256 0.2 0.275 0.231 0.259 0.208 0.192 0.149(0.001) (0.001) (0.007) (0.004) (0.006) (0.004) (0.003) (0.002)No Children 0.395 0.418 0.36 0.408 0.356 0.38 0.224 0.237(0.001) (0.001) (0.007) (0.005) (0.006) (0.005) (0.003) (0.003)Married 0.606 0.533 0.473 0.461 0.433 0.365 0.495 0.415(0.001) (0.001) (0.008) (0.005) (0.007) (0.005) (0.004) (0.003)Disabled 0.043 0.107 0.089 0.18 0.088 0.179 0.07 0.123(0.00003) (0.00004) (0.005) (0.004) (0.004) (0.004) (0.002) (0.002)Western Province 0.164 0.171 0.632 0.53 0.322 0.339 0.429 0.408(0.001) (0.001) (0.007) (0.005) (0.006) (0.005) (0.004) (0.003)Rural Area 0.207 0.198 0.284 0.26 0.213 0.201 0.93 0.873(0.001) (0.001) (0.007) (0.004) (0.005) (0.004) (0.002) (0.002)Weighted Sample Size 3950960 4187370 24560 60320 32720 56410 20150 26920The means of each variable are listed first with the standard deviations given in parenthesis. Real earnings and real weekly wages are in 1995dollars using the national cpi deflator. NAI is shorthand for North American Indian.Source: Authors calculations using Confidential Census of Population Long Form Files 1996, 2006.1475.7.TablesandFiguresTable 5.2: The Log Earnings Gap, the Log Weekly Earnings Gap and the Oaxaca Decomposition of Log Weekly EarningsMenNon-minority - Metis Non-minority - NAI Off NAI Off - NAI On1995 2005 1995 2005 1995 2005Comparison of Log Annual Earning Gap and Log Weekly Earnings GapLn Annual Earnings Gap 0.447*** 0.220*** 0.519*** 0.439*** 0.473*** 0.546***(0.040) (0.031) (0.044) (0.042) (0.057) (0.058)Ln Weekly Earnings Gap 0.250*** 0.151*** 0.282*** 0.304*** 0.209*** 0.311***(0.033) (0.032) (0.03) (0.033) (0.038) (0.039)Log Weekly Earnings: Oaxaca DecompositionDifferences In Endowments 0.150*** 0.086** 0.152*** 0.142*** 0.045 0.032(0.023) (0.028) (0.02) (0.023) (0.04) (0.038)Differences In Returns 0.100*** 0.065*** 0.13*** 0.162*** 0.164*** 0.279***(0.018) (0.011) (0.018) (0.016) (0.038) (0.038)1485.7.TablesandFiguresTable 5.2: The Log Earnings Gap, the Log Weekly Earnings Gap and the Oaxaca Decomposition of Log Weekly Earnings (Continued)WomenNon-minority - M?tis Non-minority - NAI Off NAI Off - NAI On1995 2005 1995 2005 1995 2005Comparison of Log Annual Earning Gap and Log Weekly Earnings GapLn Annual Earnings Gap 0.432*** 0.267*** 0.414*** 0.357*** 0.144* 0.177***(0.039) (0.031) (0.035) (0.026) (0.053) (0.036)Ln Weekly Earnings Gap 0.264*** 0.184*** 0.213*** 0.232*** 0.056 0.113***(0.025) (0.025) (0.02) (0.023) (0.036) (0.024)Log Weekly Earnings: Oaxaca DecompositionDifferences In Endowments 0.17*** 0.121*** 0.109*** 0.119*** -0.008 0.018(0.025) (0.021) (0.019) (0.017) (0.044) (0.035)Differences In Returns 0.093*** 0.063*** 0.103*** 0.114*** 0.065 0.095*(0.017) (0.011) (0.014) (0.013) (0.038) (0.041)The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients are attached to the statistic: * for p<.05, ** for p<.01, and *** for p<.001.Standard errors have been clustered by province?geographic region where there are 3 geographic regions in total: rural, non-cma, cma. NAI is shorthand for North American Indian. The shorthand Offis used to denote off-reserve and On to denote on-reserve. The weighted number of observations is given in Table 5.1.Source: Authors calculations using Confidential Census of Population Long Form Files 1996, 2006.1495.7.TablesandFiguresTable 5.3: The Log Annual and Weekly Earnings Gaps with a Constant Fraction that have Positive Earnings/On-Reserve Over TimeMenNon-minority - M?tis Non-minority - NAI Off NAI Off - NAI On1995 2005 1995 2005 1995 2005Estimates with a Constant Proportion of People with Positive EarningsLn Annual Earnings 0.411*** 0.225*** 0.486*** 0.445*** 0.616*** 0.538***(0.018) (0.012) (0.017) (0.013) (0.034) (0.022)Ln Weekly Earnings 0.216*** 0.158*** 0.250*** 0.316*** 0.326*** 0.301***(0.015) (0.012) (0.014) (0.013) (0.026) (0.018)Estimates with a Constant Proportion of People with Positive Earnings and On-ReserveLn Annual Earnings 0.616*** 0.224***(0.034) 0.055Ln Weekly Earnings 0.326*** 0.314***(0.026) 0.043Estimates with a Constant Proportion of People with Positive Earnings and On-Reserve: Restricted to Band Memnber, Registered Indians (NAI)Ln Annual Earnings 0.536*** 0.055(0.028) (0.044)Ln Weekly Earnings -0.298*** 0.173***(0.015) (0.030)Tail Trimmed for Censoring lower - lower - lower (off) lower (on)% Trimmed for Censoring <1% - 1% - 1% 9%Tail Moved On-Reserve - - - - - upper (off)% Moved On-Reserve - - - - - 20%1505.7.TablesandFiguresTable 5.3: The Log Annual and Weekly Earnings Gaps with a Constant Fraction that have Positive Earnings/On-Reserve Over Time(Continued)WomenNon-minority - M?tis Non-minority - NAI Off NAI Off - NAI On1995 2005 1995 2005 1995 2005Estimates with a Constant Proportion of People with Positive EarningsLn Annual Earnings 0.409 *** 0.291*** 0.393*** 0.377*** 0.226*** 0.158***(0.018) (0.015) (0.014) (0.016) (0.030) (0.022)Ln Weekly Earnings 0.240*** 0.211*** 0.190*** 0.260*** 0.131*** 0.088***(0.013) (0.013) (0.012) (0.016) (0.020) (0.019)Estimates with a Constant Proportion of People with Positive Earnings and On-ReserveLn Annual Earnings 0.226*** -0.428***(0.030) (0.051)Ln Weekly Earnings 0.131*** -0.400***(0.020) (0.039)Estimates with a Constant Proportion of People with Positive Earnings and On-Reserve: Restricted to Band Memnber, Registered Indians (NAI)Ln Annual Earnings 0.128*** 0.089*(0.030) (0.035)Ln Weekly Earnings 0.052* 0.074***(0.026) (0.019)Tail Trimmed for Censoring lower - lower - lower (off) lower (on)% Trimmed for Censoring 2% - 1% - 1% 5%Tail Moved On-Reserve - - - - - upper (off)% Moved On-Reserve - - - - - 16%The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients are attached to the statistic: * for p<.05, ** for p<.01, and *** for p<.001.NAI is shorthand for North American Indian. The shorthand Off is used to denote off-reserve and On to denote on-reserve. The stanard errors are calculated using a non-parameteric bootstrappingprocedure with sample draws with replacement the size of the sample, 100 times. The weighted number of observations is given in Table 1. The estimates in each cells only differ from those in Table 1 byexcluding observations to make the proportion with positive earnings and on-reserve the same in each year .Note that the percentage trimmed from the off-reserve and on-reserve group varies from thiswhen I am just considering just registered, band member NAIs. Note that for non-aboriginal men, the upper tail of the earnings or wage distribution in 1996 is trimmed 2% for men and 1% for women.The weighted number of observations is given in Table 5.1.Source: Authors calculations using Confidential Census of Population Long Form Files 1996, 2006.1515.7.TablesandFiguresTable 5.4: The Log Earnings and Log Weekly Earnings Gaps Adjusted to Have the Same Composition as in 1995Non-Min- M?tis Non-Min - NAI Off NAI Off - NAI On (1) NAI Off - NAI On (2) NAI Off - NAI On (3)1995 2005 1995 2005 1995 2005 1995 2005 1995 2005MenActual Ln Annual Earnings 0.447*** 0.220*** 0.519*** 0.439*** 0.473*** 0.546*** 0.473*** 0.546*** 0.473*** 0.546***(0.040) (0.031) (0.044) (0.042) (0.057) (0.058) (0.057) (0.058) (0.057) (0.058)Re-Wgted Ln Annual Earnings 0.327*** 0.504*** 0.492*** 0.367*** 0.671***(No Legal or Ethnic) (0.01) (0.012) (0.015) (0.015) (0.015)Re-Wgted Ln Annual Earnings 0.329*** 0.508*** 0.491*** 0.366*** 0.671***(Legal or Ethnic) (0.01) (0.012) (0.015) (0.015) (0.015)Actual Ln Weekly Earnings 0.250*** 0.151*** 0.282*** 0.304*** 0.209*** 0.311*** 0.209*** 0.311*** 0.209*** 0.311***(0.033) (0.032) (0.03) (0.033) (0.038) (0.039) (0.038) (0.039) (0.038) (0.039)Re-Wgted Ln Weekly Earnings 0.23*** 0.349*** 0.278*** 0.273*** 0.316***(No Legal or Ethnic) (0.008) (0.01) (0.012) (0.012) (0.012)Re-Wgted Ln Weekly Earnings 0.215*** 0.347*** 0.276*** 0.27*** 0.317***(Legal or Ethnic) (0.008) (0.01) (0.012) (0.012) (0.012)WomenActual Ln Annual Earnings 0.432*** 0.267*** 0.414*** 0.357*** 0.144* 0.177*** 0.144* 0.177*** 0.144* 0.177***(0.039) (0.031) (0.035) (0.026) (0.053) (0.036) (0.053) (0.036) (0.053) (0.036)Re-Wgted Ln Annual Earnings 0.438*** 0.48*** 0.134*** -0.02 0.332***(No Legal or Ethnic) (0.011) (0.013) (0.013) (0.013) (0.013)Re-Wgted Ln Annual Earnings 0.436*** 0.488*** 0.132*** -0.022 0.332***(Legal or Ethnic) (0.011) (0.013) (0.013) (0.013) (0.013)Actual Ln Weekly Earnings 0.264*** 0.184*** 0.213*** 0.232*** 0.056 0.113*** 0.056 0.113*** 0.056 0.113***(0.025) (0.025) (0.02) (0.023) (0.036) (0.024) (0.036) (0.024) (0.036) (0.024)Re-Wgted Ln Weekly Earnings 0.302*** 0.326*** 0.085*** 0.034** 0.163***(No Legal or Ethnic) (0.008) (0.01) (0.011) (0.011) (0.011)Re-Wgted Ln Weekly Earnings 0.291*** 0.32*** 0.083*** 0.032** 0.164***(Legal or Ethnic) (0.008) (0.01) (0.011) (0.011) (0.011)Re-Wgting done for NAI On-Reserve - ? ?Re-Wgting done for NAI Off-Reserve - ? ? ?The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients be attached to the statistic: * for p<.05, ** for p<.01, and *** for p<.001.They are calculated using a non-parameteric bootstrapping procedure with sample draws with replacement the size of the sample, 100 times. NAI is shorthand for North American Indian and Re-Wgtedis shorthand for re-weighted. The shorthand Off is used to denote off-reserve and On to denote on-reserve. The weighted number of observations is given in Table 5.1.Source: Authors calculations using Confidential Census of Population Long Form Files 1996, 2006.1525.7.TablesandFiguresTable 5.5: The Role of Taxes and Transfers in Reducing Earnings Differences: 2005Men WomenNon-Min - M?tis Non-Min - NAI Off NAI Off - NAI On Non-Min - M?tis Non-Min - NAI Off NAI Off - NAI OnLn Annual Earnings 0.220*** 0.439*** 0.546*** 0.268*** 0.358*** 0.177***(0.031) (0.042) (0.058) (0.031) (0.026) (0.036)Ln Total Income 0.230*** 0.453*** 0.561*** 0.280*** 0.370*** 0.206***Net Transfers (0.031) (0.044) (0.059) (0.032) (0.028) (0.035)Ln Total Income 0.183*** 0.373*** 0.521*** 0.196*** 0.229*** 0.118***(0.030) (0.037) (0.048) (0.025) (0.018) (0.029)Ln After Tax Income 0.158*** 0.316*** 0.409*** 0.173*** 0.186*** 0.044(0.030) (0.034) (0.046) (0.022) (0.016) (0.028)The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients be attached to the statistic: * for p<.05, ** for p<.01, and *** for p<.001.Standard errors have been clustered by province?geographic region where there are 3 geographic regions in total: rural, non-cma, cma. NAI is shorthand for North American Indian. The shorthand Offis used to denote off-reserve and On to denote on-reserve. The Post-Tax income variable is is collected from Revenue Canada as to avoid recall error and is after-tax income. The calculations in thistable were all preformed with the dollar variables round to the nearest 100 dollar for confidentially reasons. The approximate weighted number of observations is given in Table 5.1.Source: Authors calculations using Confidential Census of Population Long Form Files 1996, 2006.153Chapter 6ConclusionThis dissertation contributes to the literature on Aboriginal disparity in two ways. Thefirst is by exploring the patterns of Aboriginal earnings disparity in Canada, demonstratingthey are not corrected by the Canadian tax and transfer system and that most of the on-reserve earnings penalty is not explained by observed factors. The second is by analyzing thelong-term consequences of Indian residential schools on a set of labour market and culturaloutcomes accounting for the non-random selection of children.While residential schooling is an extreme example, the results from the first chaptershighlight the conflict assimilative educational policies place between cultural connection andeconomic participation. Public educational institutions generally play an integrative rolein society (Justman and Gradstein 2008) and my results highlight their potential culturalconsequences for minority populations. The role of assimilative education policy in socialevolution is an new area of research interest where understanding the effects of residentialschooling may be instructive. There is recent work in the development literature that triesto disentangle the possible role education plays in democratization, economic empowermentand ethnic identification (Kremer, Miguel, and Thornton 2011). Some of the strategies laidout in this dissertation could be used to examine the role of assimilative education policieson voting behavior, the adoption of self-government, and property rights. Understandinghow residential schooling influences these factors would be useful in shedding some light onhow assimilative education policies could influence political mobilization in minority groupsand may be useful for understanding the Canadian circumstance more generally.The work conducted here also raises questions regarding the economic development ofAboriginal communities. Although a large body of literature suggests that education is a154Chapter 6. Conclusiondriver of economic development, research on the economic conditions of American Indianreservations suggests that communities perform best when formal institutions are rooted ina strong traditional foundation (Dippel 2011; Cornell and Kalt 2000). The results in the firstchapters suggest that if residential schools damaged Aboriginal communities economically, itwas likely though cultural depreciation and possibly perverse intergenerational transmissionrather than any direct negative effect on formal education. This finding complements theobservation in the final chapter that the main source of the earnings penalty for those livingon-reserve is largely unexplained. Obvious educational differences do not appear to be asatisfactory explanation for Aboriginal people facing inequality in the labour market.This work also raises the question of what drives individuals to identify as an Aborig-inal person. As mentioned previously, and emphasized in the last chapter, the populationidentifying as North American Indian has grown dramatically over the past two decades anddifferential selection into identification could be the lone source of the increasing wage gapbetween those living on-reserve and those living off-reserve. Little is understood about ethnicidentification and how government policy and economic incentives influence it. Investigat-ing these issues and their interplay with cultural survival would be a productive avenue forfuture research. What increasing ethnic identification means for the economic developmentof Aboriginal communities is also an interesting question.However, at the end of this work we are left with a bit of a puzzle. Chapter 3 seems tosuggest that residential schools shifted adult outcomes toward those associated with marketbased skills and away from traditional ones. Yet we don't seem to find similarly strongevidence for this on average in Chapter 4. Although it is beyond the scope of this workto fully reconcile the discrepancy, one reasonable explanation for these findings regards thechange in sample between the 3rd and the 4th chapters. Explicitly, in order to use theidentification strategy suggested in Chapter 4, anyone who was missing information regardingtheir cousins' residential school attendance had to be dropped from the sample. In Chapter 3,I presented suggestive evidence that residential schools had their assimilative effects throughremoval from the home/community. Arguably, those that were removed from their homesand never returned (those that likely experienced the most assimilative effects) would also be1556.1. Tables and Figuresthe most likely to not be aware of whether or not any of their cousins attended a residentialschool. For that matter, they may not be aware of any of their family member's residentialschooling history. As discussed in Chapter 4, missing information on cousins' residentialschool attendance results in nearly 30 percent of the sample being excluded from the analysis.While there wasn't overly apparent correlation with missing information and observablesit is possible that there is selection on unobservable characteristics. Sample limitationsmay inherently imply we are excluding the individuals for which residential schooling hadassimilative effects.136I demonstrate the possible importance of sample selection in Table 6.1 by comparing theestimates of the treatment effect of residential school attendance in both papers. I presentthe estimates without controlling for selection to in order to avoid any differences due todifferences in the identification strategy in order to highlight what I believe may be the mainissue: a fundamentally different sample of people considered in the two papers. The firstcolumn presents the estimates from the 1991 APS, the second column presents the resultsfrom the APS 2001 including those that did not know whether any of their cousins attendeda residential school while the final column presents these results excluding those that didnot report whether or not their cousins attended a residential school.Comparing the first and second column, we can see that the estimates from the twosamples are relatively similar while the estimates in the last column are notably different withregards to high school graduation. This difference may suggest that the necessary samplerestrictions in Chapter 4 may not be innocuous. Thus all the results of the intergenerationaleffects in this paper should be thought of as conditional on this select sample. This is animportant limitation to be aware of when understanding the results of Chapter 4 and futurework should investigate this issue further.6.1 Tables and FiguresBelow are the tables for Chapter 6.136It may also be true that those whose cousin's had very traumatic experiences at residentialschool were less likely to discuss it.1566.1.TablesandFiguresTable 6.1: Comparing the Effect of Residential School in the APS 1991, APS 2001 and the APS 2001 Limited To Only Those Who KnewWhether Their Cousin Attended Residential SchoolDependent Variable APS 1991 APS 2001 APS 2001 RestrictedHigh School Graduate 0.070*** 0.030* 0.015(0.023) (0.017) (0.019)Employed -0.045** -0.037* -0.059**(0.023) (0.017) (0.0.022)Receives Social Assistance 0.001 0.064*** 0.109***(0.057) (0.016) (0.020)Uses Aboriginal Language in Home -0.021*** 0.018** 0.014*(0.007) (0.008) (0.007)The coefficients are the first listed and the standard errors are in parenthesis. The coefficient represents the effect of attending a residentialschool on outcomes without controlling for selection into residential school. Note that the estimates contained in cells in this column areestimated using a probit while the others are estimated using a linear probability model. However, the results are similar when the firstcolumn results are estimated using a linear probability model. The first column presents the estimates from the 1991 APS, the secondcolumn presents the results from the APS 2001 including the those that did not know whether their cousin attended a residential schoolwhile the final column presents these results excluding those that did not report whether or not their cousin's attended a residentialschool. Each dependent variable listed is binary and equal to one if the statement is true and zero otherwise.157Bibliography[1] Aboriginal Healing Foundation. The Indian Residential Schools Settlement Agree-ment's Common Experience Payment and Healing: A Qualitative Study Exploringthe Impacts on Recipients. Aboriginal Healing Foundation, Ontario: Ottawa: 2010.http://www.ahf.ca/downloads/cep-2010-healing.pdf. Last retrieved 14 May 2013.[2] Aboriginal Healing Foundation. Response, Responsibility, and Renewal Canada'sTruth and Reconciliation Journey. Aboriginal Healing Foundation, Ottawa: 2009.http://www.ahf.ca/downloads/trc2.pdf. Last retrieved 23 July 2012.[3] Aboriginal Healing Foundation. The Healing Has Begun: An Operational Update fromthe Aboriginal Healing Foundation. Aboriginal Healing Foundation, Ottawa: 2002.http://www.ahf.ca/downloads/the-healing-has-begun.pdf. Last retrieved 14 May 2013.[4] Adams, G., X. Chen , D. Hoyt, and L. B. Whitbeck. 2004. Conceptualizing and Mea-suring Historical Trauma Among American Indian People, American Journal of Com-munity Psychology , 33(3/4)[5] Aghion, P., Boustan, L., Hoxby, C., and Vandenbussche, J. 2009. The Causal Impactof Education on Economic Growth: Evidence from the United States. Unpublished,Harvard University.[6] Akee R., M Jorgensen, and U. Sunde. 2012. Constitutions and Economic Development:Evidence from the American Indian Nations, Unpublished.[7] Akee, R. 2009. Checkerboards and Coase: The Effect of Property Institutions onEfficiency in Housing Markets. Journal of Law and Economics , 52(2): 395-410.158Bibliography[8] Altman, Jon C. The Economic Status of Indigenous Australians. Canberra: Centrefor Aboriginal Economic Policy Research, 2000.[9] Andersen, T., J. Kruse and B.r Poppel. 2002. Survey of Living Conditions in theArctic: Inuit, Saami and the Indigenous Peoples of Chukotka (SLICA) Artic, 55(3):310-315[10] Angrist, J. D., G.W. Imbens and, D. B. Rubin. 1996. Identification of causal effectsusing instrumental variables. Journal of the American Statistical Association, 91:434.[11] Antonovics, K. L., and Goldberger, A. S. 2005. Does Increasing Women's SchoolingRaise the Schooling of the Next Generation? Comment. American Economic Review ,1738-1744.[12] Barman, J. Separate and Unequal: Indian and White Girls at All Hallows School,1884-1920 Indian Education in Canada Volume 1: The Legacy. Ed. Jean Barman,Yvonne Hebert and Don McCaskill. Nakoda Institute Occasional Paper No. 2. Vancou-ver: University of British Columbia Press, 1986.[13] Bartik, T. 1993. Who Benefits from Local Job Growth: Migrants or the OriginalResidents?, Regional Studies, 27(4): 297311.[14] Bauer, T., G. S. Epstein and I. N. Gang. 2005. Enclaves, Language, and the LocationChoice of Migrants, Journal of Population Economics. 18(4): 649-662.[15] Begum, L., Islam, A., and Smyth, R. 2012. Girls' Education, Stipend Programs andthe Effects on Younger Siblings' Education. IZA Working Paper.[16] Behrman, J. R., and Rosenzweig, M. R. 2002. Does Increasing Women's SchoolingRaise the Schooling of the Next Generation? The American Economic Review , 92(1):323-334.[17] Benhabib, J, and M. Spiegel. 1994. "The Role of Human Capital in Economic Develop-ment Evidence from Aggregate Cross-Country Data." Journal of Monetary Economics,34(2): 143-173.159Bibliography[18] Benjamin, Gunderson, et al. Labour Market Economics: Theory, Evidence, and Policyin Canada. Sixth Edition. McGraw-Hill Ryerson: Toronto. 2007. Print.[19] Bj?rklund, Anders, Mikael Lindahl and Erik Plug. 2004. Intergenerational Effects inSweden: What Can We Learn from Adoption Data? IZA Discussion Paper 1194.[20] Bj?rklund, Anders, Mikael Lindahl and Erik Plug. 2006. The Origins of Intergen-erational Associations: Lessons from Swedish Adoption Data, Quarterly Journal ofEconomics, 121(3): 999- 1028.[21] Blanchard, O. and L. F. Katz. 1992. Regional Evolutions. Brookings Papers on Eco-nomic Activity, 23(1): 1-76.[22] Blinder, A.S., 1973. Wage Discrimination: Reduced Form and Structural Estimates.Journal of Human Resources, 8(4): 436455.[23] Boudarbat, B., T. Lemieux and W. C. Riddell. 2010. "The Evolution of the Returnsto Human Capital in Canada, 1980-2005." Canadian Public Policy, 36(1): 63-89.[24] Bougie, E. and S. Sen?cal. 2010. "Registered Indian Children's School Success andIntergenerational Effects of Residential Schooling in Canada." The International In-digenous Policy Journal , 1:1(5).[25] Bourdieu, Pierre. Masculine Domination. Stanford: Stanford University Press. 2001.[26] Brasfield, C. 2001. Residential School Syndrome. BC Medical Journal , 43(2):78-81.[27] Canadian Conference of Catholic Bishops. "Apology on Residential Schools bythe Catholic Church" http://www.cccb.ca/site/eng/media-room/files/2630-apology-on-residential-schools-by-the-catholic-church%20Access%2025%20March%202012 Lastretrieved 25 March 2012.[28] Canada Department of Mines and Resources. (1945-1965). Canada De-partment of Mines and Resources Report of Indian Affairs Branch. Ot-tawa. http://www.collectionscanada.gc.ca/databases/indianaffairs/001074-119.03-160Bibliographye.php?page_id_nbr=33668&PHPSESSID=albnefemvhkcbdt3di3v523ek7. Last re-trieved 4 November 2012.[29] Canada, The Education Division, The Department of Citizenship and Immigration.The Education of Indian Children in Canada. A Symposium Written by Members ofIndian Affairs Education Division, with Comments by the Indian Peoples. The Cana-dian Superintendent. Toronto: The Ryerson Press, 1965.[30] Canada. Royal Commission on Aboriginal Peoples. Report of the Royal Commission onAboriginal Peoples. Vol. 1: Looking Forward Looking Back. Part Two: False Assump-tions and a Failed Relationship. Chapter 10: Residential Schools [CD-ROM]. Ottawa:Libraxus, 1997.[31] Card, D. 1995. "Using Geographic Variation in College Proximity to Estimate theReturn to Schooling," In Aspects of labour Market Behavior: Essays in Honour of JohnVanderkamp, ed, by Louis N. Christofides, E. Kenneth Grant, and Robert Swidinsky.Toronto: University of Toronto Press, 201-222.[32] Cariboo Tribal Council. The Impact of the Residential School . Williams Lake: CaribooTribal Council, 1991.[33] Carino, J. "Poverty and Well-being. State of the World's Indigenous Peoples. TheSecretariat of the United Nations Permanent Forum on Indigenous Issues. 14-49. 2010.[34] Carneiro, P., K. T. Hansen, and J. J. Heckman. 2003. Estimating Distributions ofTreatment Effects with an Application to the Returns to Schooling and Measurementof the Effects of Uncertainty on College Choice," International Economic Review , 44(2): 361-422.[35] Carr-Stewart S. 2001. A Treaty Right to Education. Canadian Journal of Education,26(2): 125:143.161Bibliography[36] Cassidy, J. 2009. The Canadian Response to Indian/Aboriginal Residential Schools:Lessons for Australia and the United States? Murdoch University Electronic Journalof Law , 16(2): 38-71.[37] CBC News. 2012. Residential School Day Scholars Launch Class-Action Lawsuit: DayStudents Excluded from 2006 Residential School Settlement. CBC News . August 16,2012.[38] CBC News. 2013. "Aboriginal Corrections Report Finds Systemic Discrimination,CBC News . March 7, 2013.[39] Chamberlain, G. 1977 Education, Income and Ability Revisited," Journal of Econo-metrics, 5 (2):241-257.[40] Chiswick, B. R and P.W. Miller. 1996. Ethnic Networks and Language ProficiencyAmong Immigrants. Journal of Population Economics, 9(1): 19-35.[41] Chrisjohn, R. D., S. L. Young S. L and M. Maraun. The Circle Game: Shadows andSubstance in the Indian Residential School Experience in Canada. Penticton: TheytusBooks Ltd, 2006.[42] Chriswick, B. R. 1978. The Effect of Americanization on the Earnings of Foreign BornMen, Journal of Political Economy , 86(5): 897-921.[43] Clatworthy S., M. J. Norris, and M. Cooke. 2007. Aboriginal Mobility and MigrationPatterns and the Policy Implications. In Aboriginal Conditions: Research as a Foun-dation for Public Policy ed. J. White, P. Maxim and D. Beavon, 108-130. Vancouver:UBCPress.[44] Clatworthy, S. 2007. Impacts of the 1985 Amendments to the Indian Act on FirstNations Populations. In Aboriginal Conditions: Research as a Foundation for PublicPolicy ed. J. White, P. Maxim and D. Beavon, 63-90. Vancouver: UBCPress.162Bibliography[45] Cole E., R. Barnes, and N. Josefowitz. 2006. Residential Schools Impact on Abo-riginal Students' Academic and Cognitive Development. Canadian Journal of SchoolPsychology. 21(1/2): 18-32.[46] Coleman, J. 1988. Social Capital in the Creation of Human Capital. American Journalof Sociology, 94: S95-S120.[47] Cornell, S and J. P. Kalt. 2000. Where's the Glue? Institutional and Cultural Foun-dations of American Indian Economic Development. The Journal of Socio-Economics,29(5): 443-470.[48] Cornell, S., and J. P. Kalt. American Indian Self-Determination: The Political Econ-omy of a Policy that Works. John F. Kennedy School of Government, Harvard Uni-versity, 2010.[49] Corrado, R, and I Cohen. 2003. Mental Health Profiles for a Sample of BritishColumbia's Aboriginal Survivors of the Canadian Residential School System. TheAboriginal Healing Foundation. Accessed at http://www.ahf.ca/downloads/mental-health.pdf. Last retrieved 23 March 2012.[50] Craig, R. W. and Song, X. 2011. "The Impact of Education on Unemployment Inci-dence and Re-Employment Success: Evidence from the U.S. labour Market." labourEconomics, 18(4): 453-463.[51] Crey E. and S. Fournier. Stolen from Our Embrace: The Abduction of First NationsChildren and the Restoration of Aboriginal Communities. David Neel D&M PublishersInc. 1998.[52] Curry, B. 2012. Cost to Redress Native Residential School AbuseSet to Pass $5-Billion. The Globe and Mail . September 6, 2012.http://m.theglobeandmail.com/news/politics/cost-to-redress-native-residential-school-abuse-set-to-pass-5-billion/article2242226/?service=mobile. Last retrieved28 September 2012.163Bibliography[53] Davin, N. F. 1879. Report on Industrial Schools for Indians and Half-Breeds[The Davin Report]. Sir John A. Macdonald Papers. 91(NAC MG 26A): 35428-35445. http://www.canadianshakespeares.ca/multimedia/pdf/davin_report.pdf. Lastretrieved 31 March 2012.[54] Dawson, A. 2012. Histories and Memories of the Indian Boarding Schools in Mexico,Canada, and the United States. Latin American Perspectives , 39(5): 80-99.[55] DeSilva, A. 1999. Wage Discrimination against Natives. Canadian Public Policy,25(1): 65-85.[56] DiNardo, J., N. Fortin and T. Lemieux. 1996. "labour Market Institutions and theDistribution of weekly earnings: A Semi-Parametric Approach. Econometrica, 64(5):1001-1044.[57] Dippel, C. 2011. Forced Coexistence and Economic Development: Evidence from Na-tive American Reservations. January 29: Working Paper[58] Dlugokinski E and L. Kramer. 1974. A System of Neglect: Indian Boarding Schools.American Journal of Psychiatry , 131(6):670-3.[59] Dong, Y., A.L. Lewbel and T. T. Yang. 2012. Comparing Features of ConvenientEstimators for Binary Choice Models with Endogenous Regressors. May. WorkingPaper.[60] Doyle, J. 2007. Child Protection and Child Outcomes: Measuring the Effects of FosterCare. American Economic Review, 97(5): 15831610.[61] Doyle, J. 2008. Child Protection and Adult Crime: Using Investigator Assignment toEstimate Causal Effects of Foster Care. Journal of Political Economy, 116(4): 746770.[62] Drost, H. 1994. Schooling, Vocational Training and Unemployment: The Case ofCanadian Aboriginals. Canadian Public Policy, 20(1): 52 - 65.[63] Drost, H. and J. Richards. 2003. Income On- and Off-reserve: How Aboriginals areFaring. C.D. Howe Institute Commentary 175. Toronto: The C.D Howe Institute.164Bibliography[64] Duflo, E. 2004. "The Medium Run Effects of Educational Expansion: Evidence from aLarge School Construction Program in Indonesia." Journal of Development Economics ,74(1): 163-197.[65] Economist. "Tales Out of School," The Economist . Oct 26th 2000. Last Accessed athttp://www.economist.com/node/404059. Last retrieved 1 October 2012.[66] Evans, W. N., and J. H. Topoleski. The Social and Economic Impact of Native Amer-ican Casinos. National Bureau of Economic Research, No. W9198. 2002.[67] First Nations Health Commission. Breaking the Silence: An Interpretive Study of Res-idential School Impact and Healing as Illustrated by the Stories of First Nations Indi-viduals . Ottawa: Assembly of First Nations. 1994[68] Foley, K, D. Green, and G. Gallipoli. 2012. Ability, Parental Valua-tion of Education and the High School Dropout Decision, Working Paper.http://www.economics.ubc.ca/files/2013/05/pdf_paper_giovanni-gallipoli-ability-parental-valuation-education.pdf[69] Fortin, N., T. Lemieux and S. Firpo. 2011. Decomposition Methods in Economics.In Handbook of labour Economics Volume 4 Part 1, ed O. Ashenfelter and D. Card.Amsterdam: Elsevier.[70] Friedman, W., Kremer, M., Miguel, E., & Thornton, R. Education as Liberation?National Bureau of Economic Research, No. W16939. 2011.[71] Gardeazabal, J., and A. Ugidos. 2004. More on Identification in Detailed Wage De-compositions. The Review of Economics and Statistics, 86 (4): 10341036.[72] Gauthier, M. 2010. The Impact of the Residential School, Child Welfare System andIntergenerational Trauma Upon the Incarceration of Aboriginals. Masters of Educationin Faculty of Education, Queen`s University.[73] George, P. and P. Kuhn. 1994. The Size and Structure of Native-White Wage Differ-entials in Canada. The Canadian Journal of Economics, 27(1): 20-42.165Bibliography[74] Glenn, C. American Indian/First Nations Schooling: From the Colonial Period toPresent. New York: Palgrave MacMillian, 2011.[75] Gould, E. D., Lavy, V., and Paserman, M. D. 2009. Sixty years After the MagicCarpet Ride: The Long-run Effect of the Early Childhood Environment on Social andEconomic Outcomes National Bureau of Economic Research Working Paper 14884.[76] Gover, K. 2000. "Remarks of Kevin Gover, Assistant Secretary Indian Affairs: Addressto Tribal Leaders." Journal of the American Indian Education, 39(2): 4-6.[77] Gradstein, M. and M. Justman. 2002. Education, Social Cohesion, and EconomicGrowth. American Economic Review , 92(4): 1192-1204.[78] Graham, J. 2010. The First Nation Governance System: A Brake on Closing the Com-munity Well-being Gap. Policy Brief No. 36. The Institute on Governance, Ottawa.[79] Grant, A. No End of Grief: Indian Residential Schools in Canada. Winnipeg: Pemmi-can Publications. 1996.[80] Gresko, J., White `Rites' and Indian `Rites': Indian Education and Native Responsesin the West, in Shaping the Schools of the Canadian West , Ed. David C. Jones, NancyM. Sheehan, and Robert M. Stamp, Calgary: Detselig, 1979.[81] Guimond, E. 2007. Changing Ethnicity. In Aboriginal Conditions: Research as aFoundation for Public Policy , ed. J. White, P. Maxim and D. Beavon, 91-107. Vancou-ver: UBCPress.[82] Haig-Brown, C. Resistance and Renewal: Surviving the Indian Residential School . TheSecwepemc Cultural Education Society. Vancouver: Tillacum Library, 1988.[83] Halvorsen, R. and R. Palmquist. 1980. The Interpretation of Dummy Variables inSemi-logarithmic Equations. The American Economic Review, 70 (3): 474-475.[84] Hamilton, W.D. The Federal Indian Day Schools of the Maritimes . Fredericton: TheMicmac-Maliseet Institute, 1986.166Bibliography[85] Hanushek, E. A., and D. D. Kimko. 2000. "Schooling, labour-Force Quality, and theGrowth of Nations." American Economic Review , 90(5): 1184-1208.[86] Harper, Stephen. "Prime Minister Harper offers full apology on behalf of Canadians forthe Indian Residential Schools System." Stephen Harper: Prime Minister of Canada.2008.[87] Harrison, A. O., Wilson, M. N., Pine, C. J., Chan, S. Q., and Buriel, R. 1990. FamilyEcologies of Ethnic Minority Children. Child Development , 61(2): 347-362.[88] Hawthorn, H. B. A Survey of the Contemporary Indians of Canada: Economic, Po-litical, Educational Needs and Policies . Indian Affairs Branch, Ottawa. Volume III,1967.[89] Heckman, J.J. and B. Singer. 1984. A Method for Minimising the Impact of Distribu-tional Assumptions in Econometric Models for Duration Data, Econometrica. 22(2):271-320.[90] Helliwell, J. F. and R. Putnam "Education and Social Capital," NBER Working Paper7121.[91] Hudson G. and D. B. MacDonald. 2012. The Genocide Question and Indian ResidentialSchools in Canada. Canadian Journal of Political Science 45(2): 427-449.[92] Hull, J. 2005. Post-Secondary Education and Labour Market Outcomes Canada, 2001.Catalogue No. R2-399/2001E-PDF. Services Canada.[93] Imbens, G. W and J. M. Wooldridge. 2009. Recent Developments in the Econometricsof Program Evaluation. Journal of Economic Literature, 47(1): 5-85.[94] Indian Affairs and Northern Development. 1969. Statement of the Government ofCanada on Indian Policy, 1969 . Ottawa, Ontario: Queen's Printer[95] Indian Residential Schools Resolution Canada [IRSRC]. 2006. Indian ResidentialSchools Settlement Agreement . http://www.residentialschoolsettlement.ca/settlement.html. Last retrieved 3 December 2008.167Bibliography[96] Katz L. F., J. R. Kling, and J. B. Liebman. 2007. Experimental Analysis of Neighbor-hood Effects on Youth. Econometrica, 1(1): 83-119.[97] Katz, L. F, J. R. Kling, and J. R, J. Ludwig. 2005. Neighborhood Effects on Crime forFemale and Male Youth: Evidence from a Randomized Housing Voucher Experiment.The Quarterly Journal of Economics , 120(1): 87-130.[98] King, A. R. A Case Study of an Indian Residential School. M.A Thesis. Stanford:Stanford University School of Education, 1964.[99] Krieken, R. 2004. Rethinking Cultural Genocide: Aboriginal Child Removal andSettler-Colonial State Formation. Oceania, 75(2): 125-151.[100] Kuhn, P. and A. Sweetman, 2002. "Aboriginals as Unwilling Immigrants: Contact,Assimilation and labour market Outcomes." Journal of Population Economics, 15(2):331-355.[101] Lancaster, T. 2000. The Incidental Parameter Problem Since 1948. Journal of Econo-metrics , 95: 391-413.[102] Lee, D. S. 2009. Training, weekly earnings, and Sample Selection: Estimating SharpBounds on Treatment Effects. Review of Economic Studies, 76 (3): 10711102.[103] Leslie, J. 2002. "The Indian Act: An Historical Perspective, Canadian ParliamentaryReview , 25(2): 23-27.[104] Lucas, R. E. 1988. "On the Mechanics of Economic Development." Journal of MonetaryEconomics , 22(1): 3-42.[105] Lutz, J. S. Makuk: A New History of Aboriginal-White Relations. UBC Press, 2008.[106] Macdonald, D. and D. Wilson. 2010. The Income Gap Between Aboriginal Peoplesand the Rest of Canada. Canadian Centre for Policy Alternatives, Ottawa.168Bibliography[107] MacInnes, T. R. L. 1946. History of Indian Administration in Canada. The CanadianJournal of Economics and Political Science, 12(3): 387-394.[108] Mankiw G., D. Romer, and D. Weil. 1992. "A Contribution to the Empirics of EconomicGrowth." The Quarterly Journal of Economics , 107(2): 407-437.[109] McFarlane, K. A. Educating First-Nation Children in Canada: The Rise and Fallof Residential Schooling. M.A. Thesis. Kingston: Queen's University Department ofGeography, 1999.[110] McHardy, M. and E. O'Sullivan. 2007. The Community Well-Being Index (CWB):Well-Being in First Nations Communities, Present, Past and Future. In AboriginalWell-Being: Canada's Continuing Challenge, ed. J. White, D. Beavon and N. Spence.,111-148. Toronto, ON: Thompson Educational Publishing Inc.[111] Meseyton H. Daughters of Indian Residential School Survivors: Healing Stories. Mas-ters of Social Work Thesis. Vancouver: University of British Columbia. Faculty ofGraduate Studies. 2005.[112] Miller, J.R. 1996. Shingwauk's Vision: A History of Native Residential Schools.Toronto: University Toronto Press.[113] Miller, J.R. Lethal Legacy: Current Native Controversies in Canada. Toronto: McClel-land and Steward Ltd. 2004.[114] Milloy, J. 1999. A National Crime The Canadian Government and the ResidentialSchool System, 1879 to 1986 . Winnipeg: Manitoba Studies in Native History XI TheUniversity of Manitoba Press.[115] Mueller, R. 2004. The Relative Earnings Position of Canadian Aboriginal in the1990s. Canadian Journal of Native Studies , 14(1): 37-63.[116] Neal, D. 1997. The Effects of Catholic Secondary Schooling on Educational Attain-ment. Journal of labour Economics , 15(1): 98-123.169Bibliography[117] Neyman, J., and E. L. Scott. 1948. "Consistent Estimates Based on Partially ConsistentObservations." Econometrica: Journal of the Econometric Society , 1-32.[118] Nuu-Chah-Nulth Tribal Council. Indian Residential Schools: The Nuu-chah-nulth Ex-perience. Port Alberni: Nuu-chah-nulth Tribal Council, 1996.[119] Oaxaca, R., 1973. MaleFemale Wage Differentials in Urban labour Markets. Inter-national Economic Review , 14(3): 693709.[120] O'Hara, J. and P. Treble. Residential Church School Scandal. Maclean's, June 26,2000. Print.[121] Olson, M. 1996. Big Bills Left on the Sidewalk: Why Some Nations are Rich, andOthers Poor. Journal of Economic Perspectives , 10(2): 3-24.[122] Oreopoulos, P. 2006. "Estimating Average and Local Average Treatment Effects of Edu-cation when Compulsory Schooling Laws Really Matter." American Economic Review ,96(1): 152-175.[123] Pendakur, K and R. Pendakur. 2002. Colour my World: Have Earnings Gaps forCanadian-Born Ethic Minorities Changed Over Time? Canadian Public Policy, 28(4): 489 - 512.[124] Pendakur, K. and R. Pendakur. 1998. "The Colour of Money: Earnings DifferentialsAmong Ethnic Groups in Canada." Canadian Journal of Economics , 31(3): 518-548.[125] Pendakur, K. and R. Pendakur. 2007. "Minority Earnings Disparity Across the Distri-bution." Canadian Public Policy, 33 (1): 41-62.[126] Pendakur, K., & Pendakur, R. 2011. Aboriginal Income Disparity in Canada. Cana-dian Public Policy , 37(1): 61-83.[127] Persson, Diane. The Changing Experience of Indian Residential Schooling: BlueQuills, 1931-1970. Indian Education in Canada Volume 1: The Legacy. Ed. Jean Bar-man, Yvonne Hebert and Don McCaskill. Nakoda Institute Occasional Paper No. 2.Vancouver: University of British Columbia Press, 1986.170Bibliography[128] Plug, E., & Vijverberg, W. 2003. Schooling, Family Background, and Adoption: Is itNature or is it Nurture? Journal of Political Economy, 111(3): 611-641.[129] Plug, E. 2004. Estimating the Effect of Mother's Schooling on Children's SchoolingUsing a Sample of Adoptees. American Economic Review, 94(1): 358-368.[130] Psacharopoulos, G., and H. A. Patrinos. Indigenous People and Poverty in Latin Amer-ica: An Empirical Analysis . World bank, 1994.[131] Presbyterian Archives. "The Presbyterian Church In CanadaArchives: Guide To Records Relating To The Residential Schools"Toronto: 2010. http://www.presbyterianarchives.ca/RS%20-%20FA-Residential%20Schools%20Thematic%20Guide.pdf. Last retrieved 14 May 2013.[132] R. v. Powley, [2003] 2 S.C.R. 207. 2003 SCC 43.[133] Red Horse, J. 1997. Traditional American Indian Family Systems. Families, Systems,& Health, 15(3): 243-250.[134] Reimer, G. 2010. The Indian Residential Schools Settlement Agreement'sCommon Experience Payment and Healing: A Qualitative Study Explor-ing Impacts on Recipients. Aboriginal Healing Foundation. Ottawa: Ontario.http://Indigenouspeoplesissues.com/attachments/article/4655/4655_cep-healing.pdfLast retrieved 27 March 2011.[135] Reyhner, J., and J, Eder. American Indian Education: A History, Norman: Universityof Oklahoma Press , 2004.[136] Ribar, D. 2004. What Do Social Scientists Know About the Benefits of Marriage? AReview of Quantitative Methodologies. IZA Discussion Paper No. 998.[137] Rudd, Kevin. "Apology to the Stolen Generations of Australia. Part 1." ABC 1: 13-22.171Bibliography[138] Sacerdote, B. 2007. How Large Are the Effects from Changes in Family Environment?A Study of Korean American Adoptees. The Quarterly Journal of Economics , 122(1):119-157.[139] Sacerdote, B. 2002. The Nature and Nurture of Economic Outcomes. American Eco-nomic Review (Papers and Proceedings), 92(2): 344-348.[140] Sawchuk, J. 2001. Negotiating an Identity: Metis Political Organizations, the Cana-dian Government, and Competing Concepts of Aboriginality. American Indian Quar-terly , 25(1): 73.[141] Sealey, D. B. The Education of Native Peoples in Manitoba. Monographs in EducationIII. Ed. Alexander Gregor and Keith Wilson. Winnipeg: University of Manitoba, 1980[142] Sharpe, A., J. F. Arsenault, S. Lapointe and F. Cowan. 2009. The Effect of IncreasingAboriginal Educational Attainment on the Labour Force, Output and the Fiscal Bal-ance. CSLS Research Report 2009-3. The Centre for the Study of Living Standards,Ottawa.[143] Smith, A. Indigenous Peoples and Boarding Schools: A Comparative Study . PermanentForum on Indigenous Issues Eighth session. New York, 18-29 May 2009.[144] Staples, R., and Mirande, A. 1980. Racial and Cultural Variations Among AmericanFamilies: A Decennial Review of the Literature on Minority Families. Journal ofMarriage and the Family , 42(4): 887-903.[145] Statistics Canada. 2013. "Population and Dwelling Counts, forPopulation Centres, 2011 and 2006 Censuses," Statistics Canada.http://www12.statcan.gc.ca/census-recensement/2011/dp-pd/hlt-fst/pd-pl/Table-Tableau.cfm?LANG=Eng&T=801&PR=0&RPP=9999&SR=1&S=3&O=D Lastretrieved 2013 June 6.[146] Statistics Canada. 2008a. Aboriginal Peoples in Canada in 2006: Inuit, M?tis and FirstNations . Catalogue No. 97-558-XIE. Statistics Canada: Ottawa.172Bibliography[147] Statistics Canada. 2008b. Income and Earnings Reference Guide, 2006 Census . Cata-logue no. 97-563-GWE2006003. Statistics Canada: Ottawa.[148] Statistics Canada. 2003. Aboriginal Peoples Survey 2001  Initial findings: Well-Being of the Non-Reserve Aboriginal Population. Catalogue no. 89-589-XIE. StatisticsCanada: Ottawa.[149] Stock, J. H., and Yogo, M. 2002. Testing for Weak Instruments in Linear IV Regres-sion, National Bureau of Economic Research Working Paper. No. 0284.[150] Stonefish, B. 2007. Moving Beyond: Understanding the Impacts of Residential School.Owen Sound: Ningwakwe Learning Press.[151] Stoops, M. "Health Conditions at Norway House Residential School, 1900 -1946." M.A. Thesis. Hamilton: McMaster University Department of Anthropology.2006. http://digitalcommons.mcmaster.ca/opendissertations/5470. Last retrieved 29September 2012.[152] Szaz, M C. Through a Wide Angle Lens: Acquiring and Maintaining Power, Posi-tion, and Knowledge through Boarding Schools, in Boarding School Blues: RevisitingAmerican Indian Educational Experiences. Omaha: U of Nebraska Press, 2006.[153] The Assembly of First Nations. Human Rights Report to Non-Governmental Organi-zations: Redress for Cultural Genocide: Canadian Residential Schools. November 21,2002. http://www.turtleisland.org/news/afnrezschools.pdf. Last retrieved 6 May 2012.[154] The Boarding School Healing Project. Reparations and Amer-ican Indian Boarding Schools: A Critical Appraisal. Ac-cessed at http://www.boardingschoolhealingproject.org/files/A_Critical_Appraisal_of_Reparations_final.pdf. Last retrieved on 15 April 2012.[155] The Truth and Reconciliation Commission of Canada. 2012. They Came for the Chil-dren: Canada, Aboriginal Peoples, and Residential Schools. Manitoba: Winnipeg.173Bibliography[156] Three Ways To Home. 2012. "One Step Forward...Results of the 2011 Metro VancouverHomeless Count," Regional Steering Committee on Homelessness. Metro Vancouver.http://www.metrovancouver.org/planning/homelessness/ResourcesPage/2011MetroVancouverHomelessCountFinalReport.pdfLast retrieved 2013 June 6.[157] Throsby, D. 1999. "Cultural Capital." Journal of Cultural Economics , 23(1): 3-12.[158] Tyler, Brian S., An Analysis of Public and Catholic Secondary Education and theEarnings of Men. University of Chicago Ph.D. Thesis. 1994.[159] UN Secretariat. Permanent Forum on Indigenous Issues, Divison for Economic andSocial Affairs and Division for Solicy Policy and Development. State of the World'sIndigenous Peoples. New York, 2009. (ST/ESA/328). (U.N. sales No. 09.VI.13).[160] UN General Assembly. United Nations Declaration on the Rights of Indigenous Peoples: Resolution / Adopted by the General Assembly. New York, 2007. (A/RES/61/295).http://www.refworld.org/docid/471355a82.html. Last retrieved 26 September 2013.[161] Usalcas, J. 2011. "Aboriginal People and the Labour Market: Estimates from theLabour Force Survey, 2008-2010," The Aboriginal Labour Force Analysis Series. Cata-logue no. 71-588-X, no. 3. Labour Statistics Division. Statistics Canada: Ottawa.[162] Walters, D., J. White, and P. Maxim. 2004. Does Postsecondary Education BenefitAboriginal Canadians? An Examination of Earnings and Employment Outcomes forRecent Aboriginal Graduates. Canadian Public Policy , 30(3): 283 - 302.[163] White, J. and P. Maxim. 2003. Social Capital, Social Cohesion and Population Out-comes in Canada's First Nations Communities. In Aboriginal Conditions: Research asa Foundation for Public Policy , ed. J. White, P. Maxim and D. Beavon, 7-34 Vancouver,BC: UBCPress.[164] White, J. and P. Maxim. 2007. Community Well-Being: A Comparable CommunitiesAnalysis. In Aboriginal Well-Being: Canada's Continuing Challenge, ed. J. White, D.Beavon and N. Spence, 173-184. Toronto, ON: Thompson Educational Publishing Inc.174[165] White, J., P. J. Spence, D. Nicolas and R. Downie. 2007. Aboriginal Labour MarketParticipation in Canada, Australia, New Zealand and the United States. TechnicalReport. APRC (I) 2007-02. London, ON: University of Western Ontario.[166] White, J., Spence, N., and Maxim, P. 2005. Impacts of Social Capital on EducationalAttainment in Aboriginal Communities: Lessons from Australia, Canada, and NewZealand. Social Capital in Action, 66.[167] Yun, M. S. 2005. A Simple Solution to the Identification Problem in Detailed WageDecompositions. Economic Inquiry , 43(4): 766-772.Geographic Data Sources1. ESRI ArcCanada 3.1. csd_1991.shp [1991 Census Subdivisions - Canada] [computerfile]. (2005) Redlands, CA: Environmental Systems Research Institute.2. Canadian Atlas Map Bundle. BCmdppn.shp [British Columbia mid-sized populationpoints] [computer file]. 2010. Markham, ON: DMTI Spatial. Available:3. Canadian Atlas Map Bundle. BCmjppn.shp [British Columbia major population points][computer file]. 2010. Markham, ON: DMTI Spatial. Available: UBC4. Canadian Atlas Map Bundle. BCmnppn.shp [British Columbia minor populationpoints] [computer file]. 2010. Markham, ON: DMTI Spatial. Available: UBCThe last three sources are the exact same expect British Columbia is replaced with allother provinces and BC with the appropriate provincial abbreviations.175Appendix AMore Detail On The Decisions of Aboriginal FamiliesHere I further elabourate on decision problem of Aboriginal parents and how it interactswith government enforcement to generate the demand for residential schools. I also furtherdiscuss the outcome equations.Remember that government enforcement is given byEijt = ??xijt +Bj + Ct ? e(zjt, ?j) + ???ijt + ?hhijt + ?ijt. (A.1)where ? is a parameter vector.Substituting in for Eijtin Equation 3.1 gives:Aijt =?????10if ??xijt +Bj + Ct ? e(zjt, ?j)? ??ijt > ?ijtif otherwise(A.2)where Aijt is an indicator of whether a child attends a residential school or not andijt = ???ijt + ?hhijt + ?ijt.Aboriginal parents know that the government will enforce attendance according to Equa-tion A.1, but do not observe ijt. They choose their level of resistance, ??ijt, accordingly. Iassume that parents care about two things for their children: cultural capital and humancapital. Let ??ijt represent total cultural capital and h?ijt total human capital. Human cap-ital accumulates for each individual according to the total amount of time they spend inschool, given by ?s, where s indexes the type of school (s = d for day school and s = b forresidential school)137, multiplied by the quality of the schooling given by q. The individual137Note that the time in boardings school will be greater than the time spent in a day schoolso ?b > ?d.176Appendix A. More Detail On The Decisions of Aboriginal Familiesalso inherits an idiosyncratic level of human capital given by hijt. The amount of culturalcapital accumulated is given by the amount of time a child spends with their family overthe course of their schooling years, which is assumed to be the amount of time they are notin school. This is given by ?? ? ?s where ?? is the total time available during their schoolingyears. The accumulation rate of cultural capital is given by ?. Again, students inherit somelevel of cultural capital, ?ijt. The human and cultural capital accumulation equations aregiven respectively as:h?ijt = q?s + hijt, and ??ijt = ?(?? ? ?s) + ?ijt.The utility of the parents for their child attending each type of schooling is given by:Day School: udijt = ?(?? ? ?d) + q?d + ?ijt + hijtResidential School: ubijt = ?(?? ? ?b) + q?b + ?ijt + hijt.(A.3)Parents choose the human and cultural capital of their child indirectly by choosing howstrongly to resist their child being taken to a residential school. With knowledge of EquationA.2 parents choose their optimal amount of resistance by solving:max?ijt{?(?ijt)(ubijt) + (1? ?(?ijt))(udijt)? ?ijt}subject to Equation A.2, where ? is the probability that a parents' resistance will surpassgovernment enforcement.138An important assumption is that parents can exert negative re-sistance. Resistance surpassing enforcement is only probabilistic because parents cannotobserve the idiosyncratic enforcement level for their child until resistance is exerted. Resis-tance is assumed to be costly to parents and thus enters their utility function negatively.139To get further I assume that ijt in Equation A.2 is normally distributed with mean zero138I assume the distance to a day school is zero. Differences in travel time to day schoolswould translate into differences in time with family and time in school. It could also showup in the enforcement equation of the government where distance from a day school wouldbe subtracted from distance to a residential school. In the actual estimations, distance tothe closest city is intended to absorb distance to the closest day school.139This also could be modeled as the parents valuing some consumption good and resistancebeing financially costly.177Appendix A. More Detail On The Decisions of Aboriginal Familiesand variance equal to one140and explicitly solve for the optimal choice of parental resistance??ijt.141Substituting in for ub, ud, and ??ijt into A.2, yieldsAijt =?????10if ?1 + ??2xijt ? e(zjt, ?j) +Bj + Ct + ijt > 0if otherwise(A.4)where ?1 =?ln( 2?2pi) + ln((q ? ?)(?b ? ?d)),142 and ?2 = ?. Equation A.4 summarizesthe selection of children into residential school.In the model, parents care whether their children attend a residential school or a dayschool because it will influence their child's final human and cultural capital accumulation.Parents care about both human capital and cultural capital because they contribute dif-ferently into their child's later life outcomes. Children who grow up with high amounts oftraditional cultural capital will receive higher returns from living with those who share theirculture and thus will be more likely to live in Aboriginal communities. Since the cost willbe lower for those individuals who have high cultural capital they will also be more likely touse their Aboriginal language in their homes and participate in traditional activities. Thosewith high human capital will be more likely to graduate high school, receive high returns tomarket activities and thus be employed, be less likely to receive government transfers, andreceive higher income.Assume there exists a set of cultural outcomes, each given by ?ijtk and market outcomes,eijtm, whose return is given by ?k??ijt + B?jk + C?tk +??k2xijt and ?mh?ijt + B?jm + C?tm +??m2xijtrespectively.143The subscript m indexes market outcomes and k indexes cultural outcomes.The parameters B?jk, C?tk, and ??k2xijt, allow the return to a given cultural activity k to varyby community, birth cohort and a set of individual characteristics such as gender and ethnicorigins. Parameters specific to market activities are defined similarly. Whether an individual140A sufficient condition for this is that ?hhijt, ???ijt and ?ijt are mean zero and theirvariances sum to one.141This yields ??ijt =?ln( 2?2pi) + ln((q ? ?)(?b ? ?d)) + ?xijt +Bj + Ct ? e(zjt, ?j).142Note that for ?1 to be a real number, (q ? ?)(?b ? ?d) must be positive. This will be trueas long as the rate of accumulation of human capital is greater than of cultural capital.143Given that most of the outcomes I have access to are binary I focus on zero/one outcomeshere, but a similar intuition follows for continuous variables.178Appendix A. More Detail On The Decisions of Aboriginal Familieschooses to engage in each type of activity will depend on their return to that activity. Ifthe return to that activity is positive they will engage in it, and if negative they won't.Substituting for h?ijt in the return to market activity m implies the decision to engage in thatactivity can be given byeijtm=?????10if ?m1 + ??m2xijt + ?m3Aijt + B?jm + C?tm + ?mijt > 0if otherwisewhere, ?m1 = ?mq?d, ?m3 = ?mq(?d ? ?b), and ?mijt = ?mhijt. On the other hand,substituting in for ??ijt in the returns to cultural activity k implies the decision to engage ina particular cultural activity is given by?ijtk=?????10if ?k1 + ??k2xijt + ?k3Aijt + B?jk + C?tk + ?kijt > 0if otherwisewhere ?k1 = ?k?(?? ? ?d), ?k3 = ?k?(?d ? ?b), and ?mijt = ?k?ijt.Note that the error between the outcome and residential school attendance equationsare correlated as a result of government selection on unobservable endowments of humanand cultural capital: ?kijt and ijt by ?k?? and ?hijt and ijt by ?e?h . This implies theoutcome equation cannot be estimated consistently without jointly estimating the attendanceequation. To evaluate the causal effect of residential school attendance on outcomes, anadditional parameter - the correlation of the errors terms - must be estimated. For the modelabove to be identified independent of functional form restriction long term life outcomesand which varies over both cohorts and communities.144The cost of enforcement for thegovernment depends on how far the closest residential school is from a community whichvaries over time and place via the opening and closing of residential schools. This can beexcluded from the outcome equations as it determines residential school attendance, but nothuman and cultural capital directly. It is important to note that this is conditional on a144The variable must vary over both cohorts and communities or it will be collinear withthe cohort or community fixed effects.179Appendix A. More Detail On The Decisions of Aboriginal Familiestime invariant set of community characteristics such as distance to the closest major city.Whether or not this is reasonable depends on the nature of the missionaries' decisions.180Appendix BAppendix Tables For Chapter 3181Appendix B. Appendix Tables For Chapter 3Table B.1: Linear Probability Model First Stage Results: The Effect of the Instruments on Resi-dential School AttendanceResidential School Attendance(1) (2)(a) (b) (a) (b)Open 0.114*** 0.182***(0.035) (0.034)Open?Distance -0.012*** -0.012***(0.002) (0.002)1941 Prop Catholic ? Trend -0.517** -1.222***(0.252) (0.306)Female 0.035** 0.014 0.039** 0.019(0.016) (0.014) (0.017) (0.014)Single Ethnicity 0.123*** 0.111*** 0.121*** 0.110***(0.028) (0.026) (0.029) (0.026)Latitude 0.002 0.066*** -0.001 0.055***(0.004) (0.017) (0.004) (0.017)KM (10) to City 0.001 -0.006* -0.002 -0.008**(0.001) (0.003) (0.001) (0.004)Birth Cohort FE X X X XProvincial FE X XBand FE X XF-Statistic 12.60 15.93 4.22 17.69Hansen J Statistic 8.998 0.362 - -N 11460 10271 11460 10271Notes: Standard errors are reported in parentheses and are clustered at the three year birth cohort-band level. Panel (a) is thefull sample with province effects, while Panel (b) includes all bands with at least 40 people. The total number of bands in thisspecification is 108. The reason for these restrictions with the various geographic controls regards convergence of the likelihoodfunction and credible estimation of the time invariant effects. The indicator open is equal to one for an individual if theclosest residential school to a community was open when they were of the legally mandated schooling age (which depends onfederal and provincial legislation). It is zero otherwise. Open ?Distance is this indicator times the distance from the closestresidential school. The row labeled 1941 Prop Catholic ?Trend is the coefficients on 1941 Proportion Catholic in individual'scensus division ? (average attendance in that individual's cohort - average attendance in cohort at peak 1934). The F-statisticon the excluded instruments is given in the row labeled F-Statistic. The row labeled Hansen J Statistic is the test statisticfrom the Sargan (1958) and Hansen (1982) tests of overidentifying restrictions. The asterisks indicate the level of significance:* p<0.10, ** p<0.05, *** p<0.01.182AppendixB.AppendixTablesForChapter3Table B.2: The Impact Residential School on Economic Outcomes: Province Fixed Effects(1) (2)Outcomes ATE (? = 0) ATE (? 6= 0) ATET (? 6= 0) Correlation (?) ATE (? 6= 0) ATET (? 6= 0) Correlation (?)HS Graduation 0.116*** 0.140*** 0.151*** -0.097 0.222*** 0.236*** -0.254(0.023) (0.041) (0.058) (0.256) (0.043) (0.065) (0.237)Government Transfers 0.001 -0.249*** -0.302*** 0.785*** -0.200*** -0.239*** 0.585***(0.011) (0.115) (0.096) (0.191) (0.070) (0.060) (0.162)Employed -0.103*** 0.222*** 0.251*** -0.653*** 0.157*** 0.176*** -0.488***(0.056) (0.096) (0.091) (0.185) (0.064) (0.064) (0.182)In Aboriginal Community 0.030* -0.371 -0.455 0.902*** -0.362 -0.442 0.895(0.025) (1.090) (1.356) (0.405) (0.638) (0.797) (1.021)Participate Traditional 0.045* -0.101 -0.144 0.641*** -0.076 -0.106 0.495***(0.021) (0.093) (0.096) (0.188) (0.063) (0.071) (0.161)Aboriginal Language -0.021** -0.202* -0.278** 0.830** -0.242 -0.336 0.987***(0.012) (0.139) (0.128) (0.361) (0.497) (0.451) (0.170)Source of Variation 1941 Prop Catholic ? Trend (?twjt=1941) School Open (zjt) and Distance (zjt?j)N ~11460Notes: The columns titled ATE (? = 0) contain the univariate probit marginal effects. The columns titled correlation contain the estimate of the correlation of the errorterms between the outcome equations whose dependent variable is listed on the left hand side, and the residential school attendance equation. Standard errors are reported inparentheses and are estimated by the Huber Sandwich Estimator. The columns titled ATE and ATET contain estimates of the average treatment effect and the effect ofthe treatment on the treated respectively. It can be understood as a summary statistic for the extent of unobservable selection bias. Both of their standard errors are calculatedusing the delta method, clustered at the birth cohort-year level and are in parentheses. Panel (1) includes 1941 Proportion Catholic in individual's census division ? (averageattendance in that individual's cohort - average attendance in cohort at peak 1934) in the attendance equation. Panel (2) includes the open indicator and the distances to theschool in the attendance equation as the exclusion restriction. All regressions include latitude, gender, distance from closest city, an only Aboriginal ancestry indicator, birthcohort fixed effects, band fixed effects, with 108 bands in total. All bands included have a sample size of 40. The sample size varies by the dependent variable and thus thenumber of observations N is approximate. The asterisks indicate the level of significance: * p<0.10, ** p<0.05, *** p<0.01.183AppendixB.AppendixTablesForChapter3Table B.3: The Effect of Residential School Attendance on Sociological Outcomes(1) (2)ATE (? 6= 0) ATET (? 6= 0) Correlation (?) ATE (? 6= 0) ATET (? 6= 0) Correlation (?)Married -0.006 -0.007 -0.001 -0.056** -0.075** 0.123(0.035) (0.044) (0.156) (0.033) (0.044) (0.136)Daily Smoker -0.093 -0.112 0.239 -0.071** -0.085** 0.190*(0.036) (0.048) (0.163) (0.035) (0.047) (0.147)Seek Parents Support -0.059* -0.078** 0.191 -0.072** -0.095** 0.228*(0.038) (0.047) (0.164) (0.038) (0.047) (0.144)Source of Variation 1941 Prop Catholic ? Trend (?twjt=1941) School Open (zjt) and Distance (zjt?j)N ~10271Notes: The columns titled ATE and ATET contain estimates of the average treatment effect and the effect of the treatment on the treated respectively. Both of theirstandard errors are calculated using the delta method, reported in parentheses and are clustered at the birth cohort-band level. The columns titled correlation contain theestimate of the correlation of the error terms between the outcome equations, whose dependent variable is listed on the left hand side, and the residential school attendanceequation. It can be understood as a summary statistic for the extent of unobservable selection bias. Panel (1) includes 1941 Proportion Catholic in individual's census division? (average attendance in that individual's cohort - average attendance in cohort at peak 1934) in the attendance equation. Panel (2) includes the open indicator and thedistances to the school in the attendance equation as the exclusion restriction. The specifications include all bands with at least 40 people. The total number of band fixedeffects in this specification is 108. The reason for these restrictions with the various geographic controls regards convergence of the likelihood function and credible estimation ofthe time invariant effects. All regressions include latitude, gender, distance from closest city, an only Aboriginal ancestry indicator, birth cohort fixed effects, the geographicaleffects specified. The asterisks indicate the level of significance: * p<0.10, ** p<0.05, *** p<0.01.184Appendix CThe Log-Likelihood FunctionThe notation and set up below is from Greene (2002) p.710-711. LetX2 = [1, xijt, zjt, ?j, bt, cj]and X1 = [1, ai, bt, cj, xijt], where bt is a vector of birth cohort dummies and cj is a vector ofcommunity dummies. The bivariate normal cdf is given by145Prob(X1 < x1, X2 < x2) =x2???x1????2(z1, z2, ?)dz1dz2,which I will denote ?2(x1, x2, ?). The density is?2(z1, z2, ?) =e?1/2(x21+x22?2?x1x2)/(1??2)2pi(1? ?2)1/2.To construct the log-likelihood, let qijt1 = 2?kijt ? 1 and qijt2 = 2Aijt ? 1. Now letzi1 = ?m1+??m2xijt+?m3Aijt+??jcj+??tbt and zi2 = ?1 + ??2xijt + ?3zjt + ?4zjt?j + ?tbt + ?jcjand wi1 = qi1zi1 and wi2 = qi2zi2 and ?i? = qi1qi2?.Then the log likelihood function is given bylog L =?ni=1 ln Prob(Y1 = ?kijt, Y2 = Aijt|x1, x2) =n?i=1ln ?2(wi1, wi2, ?).Which I estimate using maximum likelihood. The marginal effects of each independentfactor on final outcomes can be computed as follows. Let gi1 = ?(wi1)?[wi2??i?wi1?1??2i?]. Note145The only away in which the likelihood differs when wjt is used as exogenous variation isthe omission of zjt and ?j by wjt.185Appendix C. The Log-Likelihood Functionthat there are several marginal effects one might want to evaluate in the bivariate probitmodel (See Green 1996b). For convenience in evaluating them, we will define a vectorx1 = x1 ? x2 and let x?1?1 = x??1. Thus, ?1 contains all the nonzero elements of ?1andpossibility some zeros in the positions of variables in x that appear on in the other equation;?2 is defined likewise. The bivariate probability isProb[y1 = 1, y2 = 1|x] = ?2(x??1, x??2, ?).The marginal effects of changes in x on this probability are given by??2?x= g1?1 + g2?2.186Appendix DMapping Individuals to CommunitiesThe communities that are matched are all areas defined as Indian reserve, settlements orunorganized regions that contain Aboriginal communities that can be linked to a band. Insome cases, villages or towns are included if they are associated with a particular Aboriginalband identified in the 1991 Census.Once the sample is weighted using the population weights, nearly 50 percent of registeredIndians who are in my sample do not currently live in one of the specified Aboriginal com-munities that have been linked to a school. Unweighted, this proportion of the populationis a much less important part of the sample. Although the APS does not specify where anindividual was born, it does specify what band an individual belongs to. More than halfof these bands have a legally defined land base. A large fraction of these land bases linkuniquely to only one or two CSDs. Using Aboriginal Affairs and Northern DevelopmentCanada's (AANDC) legal-linkage files of bands to CSDs, I can reconstruct an individual'sorigin communities. The draw-back of these files are that they define bands only by their2006 names and their 2006 CSD. These differ substantially in some cases from their 1991names and definitions. I convert the 2006 CSD using the correspondence tables proved byStatistics Canada to link the 2006 CSD to the 2001 CSD, then the 2001 divisions to the 1996CSD and finally back to the 1991 CSD. Codes are aggregated when necessary to producereasonably consistent geographic regions.Neither Statistics Canada nor AANDC provides a correspondence table between the 2006band definitions and the 1991 band definitions. I construct a correspondence using sourcessuch as the Canadian encyclopedia or band websites which often provide band histories. Atotal of 420 out of the 660 bands in 2006 either had the same name in both years or were187Appendix D. Mapping Individuals to Communitiesincorrectly spelled in 1991 and thus were straight forward. A total of 196 bands experiencesname changes. Approximately 44 of the bands were difficult to match either because theyhad been dissolved, reformed, or had no legal land base and needed to be matched based ontheir traditional locations.Another limitation of the band listings in the 1991 APS is that some individuals did notlist their band but instead listed their tribal council or the ethnic group they belong to (forexample instead of saying Bigstone Cree or Chapleau Cree, they would just list Cree). Inthese cases, I link the tribal council or ethnic group to a large subset of possible CSD.If a band is linked to more than one possible sub-division, I use the 1991 ConfidentialLong Form Census files to estimate the probability of being from each of these divisions,given each band. I then match individuals who currently reside outside one of the previouslyspecified Aboriginal communities to one of these divisions using these estimated probabilitydistributions. If a community has no legal land base, or if there are no individuals inthe communities predicted given their band's legal land base, I estimate the probabilitydistribution of their location based on where they actually are.Note that using the 1991 probabilities rather than the probabilities at the time theindividuals were in school is a matter of practicality. I have been unable to find a statisticalresources that would allow me to calculate these probabilities for earlier time periods. Whatdo exist are the number of Indians in a particular census division in 1921, 1931, 1941 and1951 and every five years thereafter. Theoretically, these broader regions could be linked tobands using the 1991 geographic distribution of First Nations. However, given the relianceon the 1991 distributions, this route would add little over what aggregation of the 1991distributions would provide.188Appendix EAppendix Tables For Chapter 4Table E.1: The Correlation of an Indicator for Missing Parental or Own Residential School Atten-danceDependent Variables Coef. Std. Err. t P>|t|Female 0.005 (0.009) 0.53 0.596Attended -0.022 (0.015) -1.54 0.129Grandmom Attended 0.037 (0.020) 1.85 0.070Missing Grandmom Attendance 0.114 (0.034) 3.34 0.001Grandfather Attended 0.009 (0.018) 0.53 0.600Missing Grandfather Attendance 0.127 (0.029) 4.37 0.000Sibling Attended 0.029 (0.012) 2.42 0.019Missing Sibling Attendance 0.513 (0.028) 18.14 0.000Aunt or Uncle Attended 0.007 (0.011) 0.65 0.518Missing Aunt or Uncle Attendance 0.149 (0.020) 7.40 0.000High School Graduation -0.009 (0.009) -1.03 0.307Government Transfers 0.037 (0.015) 2.58 0.013Aboriginal Lang. at Home 0.029 (0.015) 1.94 0.058Employed -0.016 (0.010) -1.69 0.096Hunted 0.016 (0.014) 1.15 0.254No History Ab. Lang. -0.019 (0.010) -1.87 0.066Missing Language History 0.018 (0.030) 0.60 0.549In Good Health -0.038 (0.014) -2.73 0.008Obese -0.005 (0.012) -0.39 0.701Daily Smoker -0.004 (0.011) -0.33 0.744Participate in Traditions -0.021 (0.012) -1.81 0.076Live on Reserve 0.017 (0.014) 1.18 0.241Divorced -0.04 (0.017) -2.38 0.021Married -0.015 (0.011) -1.35 0.183Census Division FE YesBirth Cohort FE YesN 12757Adjusted R Squared 0.4384Notes: Standard errors clustered at the cohort-census division level are reported in parentheses. The coefficients are in the firstrow. The specification in this table includes 58 census division and 47 cohort fixed effects.189Appendix FAppendix Tables For Chapter 5190AppendixF.AppendixTablesForChapter5Table F.1: Results of the Oaxaca Decomposition for Log Weekly Earnings using 1995 Years of Education and Potential Experience1995MenNon-Minority - M?tis Non-Minority - NAI off NAI off-NAI on Non-Minority - M?tis Non-Minority - NAI off NAI off-NAI onLn Weekly Earnings 0.250*** 0.282*** 0.209*** 0.250*** 0.282*** 0.209***(0.033) (0.030) (0.038) (0.033) (0.030) (0.038)Differences In Endowments Differences In ReturnsExperience -0.004 0.017*** -0.05*** 0.058 -0.057 0.246***(0.004) (0.004) (0.007) (0.058) (0.054) (0.061)Education 0.111*** 0.089*** 0.095*** -0.146 -0.005 0.353**(0.01) (0.008) (0.014) (0.189) (0.108) (0.102)Household 0.02** 0.039*** 0.038* 0.067** 0.038 -0.088(0.006) (0.007) (0.017) (0.022) (0.033) (0.046)Geography 0.015 -0.019 -0.027 0.072** -0.004 0.006(0.021) (0.016) (0.033) (0.022) (0.023) (0.072)Legal 0.007 0.054(0.007) (0.029)Work Variables 0.009 0.021** -0.002 0.023 -0.019 0.176***(0.008) (0.007) (0.014) (0.029) (0.024) (0.029)Disabled 0.008*** 0.009*** -0.007** -0.014 -0.015 0.047(0.001) (0.001) (0.002) (0.027) (0.025) (0.034)Total 0.159*** 0.156*** 0.054 0.091*** 0.126*** 0.156***(0.023) (0.02) (0.04) (0.019) (0.018) (0.039)The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients are attached to the statistic: * for p<.05, **for p<.01, and *** for p<.001. Standard errors have been clustered by province?geographic region where there are 3 geographic regions in total: rural, non-cma, cma. NAI isshorthand for North American Indian. The shorthand Off is used to denote off-reserve and On to denote on-reserve. The weighted number of observations is given in Table5.1. The definitions for each of the categories can be found at the end of Section 5.3. Source: Authors calculations using Confidential Census of Population Long Form Files1996.191AppendixF.AppendixTablesForChapter5Results of the Oaxaca Decomposition for Log Weekly Earnings using 1995 Years of Education and Potential Experience (Continued)1995WomenNon-Minority - M?tis Non-Minority - NAI off NAI off-NAI on Non-Minority - M?tis Non-Minority - NAI off NAI off-NAI onLn Weekly Earnings 0.213*** 0.213*** 0.056 0.213*** 0.213*** 0.056(0.02) (0.02) (0.036) (0.02) (0.02) (0.036)Differences In Endowments Differences In ReturnsExperience 0.004 0.004 -0.017*** 0.078* 0.078* 0.102*(0.003) (0.003) (0.003) (0.034) (0.034) (0.05)Education 0.071*** 0.071*** 0.087*** -0.283 -0.283 0.528**(0.007) (0.007) (0.016) (0.18) (0.18) (0.2)Household 0.027*** 0.027*** 0.003 -0.011 -0.011 -0.079(0.008) (0.008) (0.013) (0.032) (0.032) (0.043)Geography -0.001 -0.001 0.029 -0.007 -0.007 0.023(0.015) (0.015) (0.037) (0.024) (0.024) (0.049)Legal -0.001 -0.051(0.008) (0.055)Work Variables 0.009 0.009 -0.087*** 0.002 0.002 0.068**(0.009) (0.009) (0.017) (0.016) (0.016) (0.025)Disabled 0.007*** 0.007*** -0.004* -0.032* -0.032* 0.063***(0.001) (0.001) (0.002) (0.015) (0.015) (0.016)Total 0.117*** 0.117*** 0.009 0.082*** 0.096*** 0.047(0.019) (0.019) (0.044) (0.017) (0.015) (0.038)The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients are attached to the statistic: * for p<.05, **for p<.01, and *** for p<.001. Standard errors have been clustered by province?geographic region where there are 3 geographic regions in total: rural, non-cma, cma. NAI isshorthand for North American Indian. The shorthand Off is used to denote off-reserve and On to denote on-reserve. The weighted number of observations is given in Table5.1. The definitions for each of the categories can be found at the end of Section 5.3. Source: Authors calculations using Confidential Census of Population Long Form Files1996.192AppendixF.AppendixTablesForChapter5Table F.2: Results of the Detailed Oaxaca Decomposition for Log Weekly EarningsMen WomenNon-min- M?tis Non-min- NAI off NAI off - NAI on Non-min - M?tis Non-mino- NAI off NAI off - NAI on1995 2005 1995 2005 1995 2005 1995 2005 1995 2005 1995 2005Ln Weekly Earnings 0.25*** 0.151*** 0.282*** 0.304*** 0.209*** 0.311*** 0.264*** 0.184*** 0.213*** 0.232*** 0.056 0.113***(0.033) (0.032) (0.03) (0.033) (0.038) (0.039) (0.025) (0.025) (0.02) (0.023) (0.036) (0.024)Differences In CharacteristicsExperience 0.026*** 0.018*** 0.04*** 0.024*** -0.016** -0.008 0.015*** 0.014*** 0.016*** 0.018*** 1? 10?3 -0.004(0.003) (0.003) (0.003) (0.004) (0.005) (0.005) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003)Education 0.073*** 0.063*** 0.062*** 0.071*** 0.031*** 0.036*** 0.063*** 0.064*** 0.053*** 0.062*** 0.032*** 0.056***(0.006) (0.005) (0.005) (0.005) (0.005) (0.006) (0.005) (0.005) (0.006) (0.005) (0.008) (0.007)Household 0.018*** 0.011 0.025*** 0.02*** -0.005 0 0.009*** 0.004** 0.013*** 0.011*** 0.005 0.028***(0.004) (0.006) (0.004) (0.005) (0.008) (0.009) (0.002) (0.001) (0.003) (0.002) (0.007) (0.006)Geography 0.014 -0.009 -0.021 -0.016 -0.017 -0.02 0.036 0.007 -0.003 -0.00001 0.044 0.027(0.022) (0.026) (0.017) (0.017) (0.034) (0.03) (0.024) (0.022) (0.017) (0.016) (0.038) (0.028)Legal 0.007 0.015* -0.002 0.013**(0.007) (0.007) (0.008) (0.004)Work Variables 0.011 0.001 0.024** 0.027*** -0.003 -0.034* 0.037** 0.027** 0.009 0.01 -0.09*** -0.123***(0.008) (0.007) (0.007) (0.007) (0.014) (0.015) (0.011) (0.009) (0.009) (0.008) (0.017) (0.012)Disabled 0.008*** 0.012*** 0.009*** 0.01*** -0.007** -0.011*** 0.008*** 0.01*** 0.007*** 0.01*** -0.005* -0.006***(0.001) (0.002) (0.001) (0.001) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002)Language 0.00003 -0.009 0.013 0.004 0.055** 0.055*** 0.002 -0.005 0.014 0.007 0.007 0.027*(0.007) (0.005) (0.008) (0.007) (0.017) (0.014) (0.006) (0.006) (0.008) (0.007) (0.014) (0.012)Total 0.15*** 0.086** 0.152*** 0.142*** 0.045 0.032 0.17*** 0.121*** 0.109*** 0.119*** -0.008 0.018(0.023) (0.028) (0.02) (0.023) (0.04) (0.038) (0.025) (0.021) (0.019) (0.017) (0.044) (0.035)193AppendixF.AppendixTablesForChapter5Results of the Detailed Oaxaca Decomposition for Log Weekly Earnings (Continued)Men WomenNon-mino - M?tis Non-min - NAI off NAI off - NAI on Non-min- M?tis Non-min - NAI off NAI off - NAI on1995 2005 1995 2005 1995 2005 1995 2005 1995 2005 1995 2005Differences In ReturnsExperience 0.146 0.352* -0.257 -0.139 0.859*** 0.066 0.662* 0.376* 0.65** 0.337 -0.019 -0.058(0.267) (0.165) (0.234) (0.16) (0.239) (0.193) (0.298) (0.148) (0.191) (0.267) (0.255) (0.307)Education -0.022 0.004 0.004 -0.012 0.034 0.012 -0.001 0.01 0.022* 0.008 0.039** 0.019(0.017) (0.009) (0.014) (0.015) (0.022) (0.021) (0.017) (0.007) (0.011) (0.006) (0.014) (0.01)Household 0.044*** 0.034*** 0.03 0.066*** -0.014 -0.038 -0.001 0.015 -0.019 0.05*** -0.019 -0.086***(0.012) (0.009) (0.016) (0.017) (0.021) (0.025) (0.02) (0.009) (0.015) (0.012) (0.023) (0.02)Geography 0.078*** -0.01 -0.003 0.021 0.001 0.120* -0.174* 0.021 -0.008 0.021 0.005 0.013(0.021) (0.02) (0.023) (0.021) (0.072) (0.05) (0.074) (0.018) (0.024) (0.014) (0.048) (0.063)Legal 0.053 0.047 -0.051 -0.084*(0.029) (0.034) (0.056) (0.038)Work Variables 0.021 0.062* -0.018 0.044 0.177*** 0.245*** 0.033 0.031 0.004 0.039* 0.066** 0.095***(0.028) (0.03) (0.024) (0.023) (0.028) (0.023) (0.018) (0.017) (0.015) (0.016) (0.024) (0.024)Disabled -0.014 -0.010 -0.014 -0.004 0.05 0.059*** -0.002 0.0000 -0.033* 0.012** 0.067*** 0.023*(0.028) (0.008) (0.025) (0.009) (0.035) (0.01) (0.024) (0.006) (0.016) (0.004) (0.017) (0.009)Language 0.02 0.045* 0.017 0.043** -0.082* -0.09*** -0.014 0.031 0.02 0.037 -0.059 -0.059*(0.022) (0.019) (0.026) (0.015) (0.038) (0.021) (0.02) (0.025) (0.024) (0.02) (0.03) (0.024)Constant -0.171 -0.412* 0.37 0.142 -0.913*** -0.143 -0.41 -0.42** -0.531** -0.39 0.034 0.231(0.273) (0.18) (0.243) (0.168) (0.229) (0.204) (0.31) (0.153) (0.198) (0.262) (0.261) (0.293)Total 0.1*** 0.065*** 0.13*** 0.162*** 0.164*** 0.279*** 0.093*** 0.063*** 0.103*** 0.114*** 0.065 0.095*(0.018) (0.011) (0.018) (0.016) (0.038) (0.038) (0.017) (0.011) (0.014) (0.013) (0.038) (0.041)The coefficients are the first listed and the standard errors are in parenthesis. Stars denoting the significance of the coefficients are attached to the statistic: * for p<.05, **for p<.01, and *** for p<.001. Standard Errors have been clustered by province?geographic region where there are 3 geographic regions in total: rural, non-cma, cma. NAI isshorthand for North American Indian. The shorthand Off is used to denote off-reserve and On to denote on-reserve. The weighted number of observations is given in Table5.1. The definitions for each of the categories can be found at the end of Section 5.3.Source: Authors calculations using Confidential Census of Population Long Form Files 1996, 2006.194

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0165638/manifest

Comment

Related Items