UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Parental work and child-care in Canadian families Gagne, Lynda Giselle 2002

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

Item Metadata

Download

Media
831-ubc_2002-731634.pdf [ 8.57MB ]
Metadata
JSON: 831-1.0090537.json
JSON-LD: 831-1.0090537-ld.json
RDF/XML (Pretty): 831-1.0090537-rdf.xml
RDF/JSON: 831-1.0090537-rdf.json
Turtle: 831-1.0090537-turtle.txt
N-Triples: 831-1.0090537-rdf-ntriples.txt
Original Record: 831-1.0090537-source.json
Full Text
831-1.0090537-fulltext.txt
Citation
831-1.0090537.ris

Full Text

P A R E N T A L WORK AND CHILD-CARE IN CANADIAN FAMILIES by L Y N D A GISELE G A G N E B.A., Simon Fraser University, 1987 M.A. , Simon Fraser University, 1994 A THESIS SUBMITTED IN PARTIAL F U L F I L M E N T OF T H E REQUIREMENTS FOR T H E DEGREE OF DOCTOR OF PHILOSOPHY i n T H E F A C U L T Y OF G R A D U A T E STUDIES (Department of Economics) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA December 2001 copyright Lynda Gagne, 2001 In p resent ing t h i s t h e s i s i n p a r t i a l f u l f i l m e n t of the requirements f o r an advanced degree at the U n i v e r s i t y of B r i t i s h Columbia, I agree that the L i b r a r y s h a l l make i t f r e e l y a v a i l a b l e f o r reference and study. I fu r ther agree that permission f o r extensive copying of t h i s t h e s i s f o r s c h o l a r l y purposes may be granted by the head of my department or by h i s or her representa t ives . It i s understood that copying or p u b l i c a t i o n of t h i s thes is f o r f i n a n c i a l ga in s h a l l not be al lowed without my wr i t ten permiss ion . Department of The U n i v e r s i t y of B r i t i s h Columbia Vancouver, Canada ABSTRACT In 2000, 79 percent of married Canadian women between the ages of 25 and 44 were in the labour force and 75 percent were employed.1 Many Canadian families with working parents use costly child-care, and many of these families take advantage of the child-care expense deduction (CCED): in 1998, 71 percent of families with pre-school children used child-care services to work or study at a given point in time,2 and 868,460 taxfilers reported nearly $2.4 billion in child-care expenditures on 1,390,200 children.3 In this thesis, I examine the effects of parental labour supply and child-care use on children, the impacts that child-care costs have on the labour supply of married mothers, and the fairness of the tax system with respect to child-care costs. Chapters I, and V are introductory and concluding chapters, respectively. In chapter II, I consider the question of whether parental labour supply and child-care use affect child cognitive and behavioural outcomes. Parental labour supply reduces the amount of time parents have for their children. On the other hand, parents can replace their own time with child-care services and can also purchase more market goods with additional income earned at work. I examine this question using the first three cycles of the National Longitudinal Survey of Children and Youth (NLSCY), which provide both a large sample size and a rich source of data, including controls for parenting skills. The possible joint detennination of labour supply and child outcomes is also tested. In chapter III, I estimate the impact of child-care costs on the return to work of married Canadian women with children under three, using data from the 1988 Canadian National Child-care Survey (CNCCS) and Labour Market Activity Survey (LMAS). Data from the 1995 Canadian General Social Survey indicate that Canadian mothers have split views on the issue of 1 Statistics Canada: http:/www.statcan.ca/english/Pgdb/People/Labour/labor24c.htm, February 10, 2001. ii whether parental labour supply has deleterious effects on child outcomes. Furthermore, women's views on these issues tend to be consistent with their labour supply, suggesting their views may affect whether they choose to work or not. If women's preferences for work are based on then-views and are correlated with other explanatory variables such as education and cost of care, the estimated coefficients on these explanatory variables will be biased. In order to allow for these potential differences in responsiveness to childcare costs, I estimate separate models where current or previous occupation and weeks worked in the previous 12 months are used as control variables in the estimation to account for heterogeneity of preferences. In chapter IV of the thesis, I use data from the CNCCS and LMAS to examine the vertical and horizontal equity of the CCED. Vertical equity is evaluated by comparing CCED benefit rates for different family levels of earnings. This is done for dual earner families with childcare costs and similar characteristics. Horizontal equity is examined by investigating whether the existence of the CCED increases or decreases the difference between effective tax rates of families with similar earnings but different labour supplies. I use measures of actual and potential earnings to evaluate both vertical and horizontal equity. 2 National Longitudinal Survey of Children and Youth microdata file, Cycle 3. 3 Source: http://www.ccra-adrc.gc.ca/tax/m^ July 14, 2001. iii TABLE OF CONTENTS ABSTRACT II TABLE OF CONTENTS IV LIST OF TABLES VI LIST OF FIGURES VII ACKNOWLEDGEMENTS IX CHAPTER I OVERVIEW A N D S U M M A R Y 1 1.1 INTRODUCTION 1 1.2 CHAPTER II- THE EFFECT OF PARENTAL WORK AND CHILD-CARE USE ON CHILD OUTCOMES ' 2 1.3 CHAPTER III- CHILD-CARE COSTS AND CANADIAN WOMEN'S MARKET WORK: HETEROGENOUS PREFERENCES 3 1.4 CHAPTER IV-POTENTIAL INCOME AND THE EQUITY OF THE CHILD-CARE EXPENSE DEDUCTION 4 CHAPTER II THE EFFECT OF P A R E N T A L W O R K A N D CHILD-CARE USE ON CHILD OUTCOMES 5 2.1 INTRODUCTION : 5 2.2 LITERATURE SURVEY 8 2.3 MODEL AND ESTIMATION ISSUES 13 2.4 DATA AND PRELIMINARY STATISTICS 26 2.5 ESTIMA TION RESULTS AND DISCUSSION 32 2.6 CONCLUSION 66 CHAPTER III CHILD-CARE COSTS A N D C A N A D I A N W O M E N ' S M A R K E T WORK: HETEROGENEOUS PREFERENCES 69 3.1 INTRODUCTION 69 3.2 LITERATURE SURVEY ; 70 3.3 DATA AND PRELIMINARY STATISTICS 75 3.4 MODEL AND ESTIMATION PROCEDURE 81 3.5 ESTIMATION RESULTS AND DISCUSSION 84 3.6 CONCLUSION 104 APPENDIX 3.1 105 CHAPTER TV POTENTIAL INCOME A N D THE EQUITY OF THE CHILD-C A R E EXPENSE DEDUCTION 107 4.1 INTRODUCTION 107 4.2 THE CANADIAN CHILD-CARE EXPENSE DEDUCTION _ 108 4.3 VERTICAL AND HORIZONTAL EQUITY CONCEPTS AND RELATED LITERATURE 113 4.4 CCED CEILINGS AND CHILD-CARE PAYMENTS BY INCOME GROUPS, 1987/1988 122 4.5 DATA AND ESTIMATION PROCEDURE 124 iv 4.6 ANALYSIS 129 4.7 CONCLUSION 146 APPRENDIX 4.1 ; 148 APPENDIX4.2 _ 149 APPENDIX 4.3 150 APPENDIX 4.4 151 APPENDIX 4.5 152 CHAPTER V CONCLUDING COMMENTS 154 5.1 CONCLUDING COMMENTS 154 BIBLIOGRAPHY 156 v L I S T O F T A B L E S T A B L E 2.1 Means, Ranges and Standard Deviations of Variables 27 T A B L E 2.2 OLS Regression Coefficients - Standard PPVT Scores 35 T A B L E 2.3 Fixed Effects Regression Coefficients - Standard PPVT Scores 44 T A B L E 2.4 IV Estimator/ Hausman Test (OV) - Standard PPVT Scores 49 T A B L E 2.5 OLS Regression Coefficients - Hyperactivity Score, 4-5 Years Old 51 T A B L E 2.6 OLS Regression Coefficients - Other Behavioural Scores, 4-5 Years Old Cross-Sectional Data - Specification II 58 T A B L E 2.7 OLS Regression Coefficients - Other Behavioural Scores, 4-5 Years Old Longitudinal Data - Specification II 61 T A B L E 2.8 OLS Regression Coefficients - Behavioural Scores, 2-3 Years Old Cross-Sectional Data - Specification II 64 T A B L E 3.1 Means of Dependent and Independent Variables 79 T A B L E 3.2 Reduced-form Work Probit and Log-Wage Equation Coefficients _ 86 T A B L E 3.3 Bivariate Work and Pay for Care Probit and Cost Equation Coefficients 89 T A B L E 3.4 Child Care Costs and Wage Elasticities 93 T A B L E 3.5 A Structural Probability of Being at Work Probit Coefficients Occupation 101 T A B L E 3.5B Structural Probability of Being at Work Probit Coefficients Weeks Worked 102 T A B L E 4.1 The Widower and Nanny Get Married 115 T A B L E 4.2 Horizontal and Vertical Equity and the CCED 138 T A B L E 4A1.1 CCED Limits 1972-1999 148 T A B L E 4A3.1 Log-Wage Equation Coefficients 150 T A B L E 4A4.1 Preordered Inequity Index 151 vi L I S T O F F I G U R E S FIGURE 2.1 PPVT Scores, Labour Supply and Education: High Parenting Skills.... 39 FIGURE 2.2 PPVT Scores, Labour Supply and Education: Average Parenting Skills 40 FIGURE 2.3 PPVT Scores, Labour Supply and Education: Low Parenting Skills.... 41 FIGURE 2.4 Hyperactivity Scores, Labour Supply & Parenting Skills: Average Parental Education Levels Specification III(X) 54 FIGURE 2.5 Hyperactivity Scores, Labour Supply & Parenting Skills: Average Parental Education Levels Specification III(L-l) 55 FIGURE 2.6 Hyperactivity Scores, Labour Supply & Parenting Skills: Average Parental Education Levels Specification III(L-2) 56 FIGURE 3.1 1988 Employment Probability (Occupations -1): Married Mothers with Children < 3, by Age of Youngest Child in Months 96 FIGURE 3.2 1988 Probability of Being at Work (Occupations -1): Married Mothers with Children < 3, by Hourly Cost of Care 97 FIGURE 3.3 1988 Probability of Being at Work (Occupations - II): Married Mothers with Children < 3, by Hourly Cost of Care 98 FIGURE 3.4 1988 Probability of Being at Work (Weeks Worked - II): Married Mothers with Children < 3, by Hourly Cost of Care 100 FIGURE 4.1 Number of CCED Claimants , 110 FIGURE 4.2 Reported and Allowed Claimant Child Care Costs, ('000 of 1999 $): 1972-1996 I l l FIGURE 4.3 Percentage of Constrained Returns... 112 FIGURE 4.4 Percentage of Taxable Claimants Affected by Family Limits: 1987 & 1988 by Income Groups 123 FIGURE 4.5 Reported Child Care Payments as Percentage of Income: 1987 & 1988 by Income Groups 124 FIGURE 4.6 1988 Average Family Tax Rates as a Proportion of Predicted Family Earnings: Two Parent Families with Children < 6 129 FIGURE 4.7 1999 Average Family Tax Rates as a Proportion of Predicted Family Earnings: Two Parent Families with Children < 6 130 FIGURE 4.8 1988 Average Family Tax Rates as a Proportion of Potential Family Earnings: Two Parent Familes with children <6 131 FIGURE 4.9 1999 Average Family Tax Rates as a Proportion of Potential Family Earnings: Two Parent Familes with children < 6 132 FIGURE 4.10 Potential CCED Benefits: Two Parent Family 134 FIGURE 4.11 Potential CCED Benefits: Single Parent Family 135 FIGURE 4.12 Potential CCED Benefit as a Percentage of Earnings: Two Children < 7 136 FIGURE 4.13 Predicted Benefit as a Proportion of Predicted Family Earnings: Two Parent Families with One Child < 7, 1988 Dual Earner Families with Child Care Costs 140 FIGURE 4.14 Predicted Benefit as a Proportion of Potential Family Earnings: Two Parent Families with One Child < 7, 1988 Dual Earner Families with Child Care Costs 141 vn FIGURE 4.15 Predicted Benefit as a Proportion of Predicted Family Earnings: Two Parent Families with One Child < 7, Dual Earner Families with Child Care Costs - 1999 Taxes '. 142 FIGURE 4.16 Predicted Benefit as a Proportion of Potential Family Earnings: Two Parent Families with One Child < 7, Dual Earner Families with Child Care Costs - 1999 Taxes 143 FIGURE 4.17 Predicted CCED Benefit as Percentage of Earnings by Earnings Deciles: Two Parent Families with One Child < 7, 1988 Dual Earner Families with Child Care Costs 144 FIGURE 4.18 Predicted CCED Benefit as Percentage of Earnings by Earnings Deciles: Two Parent Families with One Child < 7, Dual Earner Families with Child Care Costs - 1999 Taxes 145 FIGURE 4A5.1 Kakwani Index - 1999 Potential Earnings 153 viii ACKNOWLEDGEMENTS I would like to thank my supervisor Craig Riddell who provided me with invaluable on-going moral and academic support and guidance, opportunities for learning and development, and assistance in accessing the resources needed for my work. I would also like to thank the other members of my supervising committee, David Green, who always had thoughtful advice and kind words, and who asked the right questions, and Jon Kesselman who was always so prompt and considerate. I would like to thank faculty and students in the Economics Department for attending lunchtime workshops and giving me an opportunity to present my research to a critical and knowledgeable audience. I would like to thank the staff in the Economics Department for their on-going and able assistance. I would like to thank the staff at Statistics Canada for their kindness and efficiency in responding to my data and facilities needs. I would also like to thank my children, Danielle and Robin, for their patience. ix CHAPTER I OVERVIEW AND SUMMARY 1.1 I N T R O D U C T I O N The increase in women's labour force participation in the last four decades, and in particular that of married women, has been phenomenal. The overall participation rate for women in Canada went from 24 percent in 1951 to 60 percent in 2000; by 2000, 79 percent of married women between the ages of 25 to 44 were in the labour force and 75 percent were employed.4 The proportion of pre-school children whose parents were both working has been increasing as a result, and most of these children are in substitute care. In 1996, both parents were working for 56 percent of pre-school children, compared to 52 percent in 1991 and 38 percent in 1981.5 By 1998,71 percent of pre-school children with parents who were working or studying were in substitute child-care.6 The labour force participation of mothers raises a number of issues. One of these issues is whether parental labour supply and the use of substitute child-care have an impact on the cognitive and behavioural development of children. Another issue is whether and how much child-care costs act as a deterrent to the labour supply of mothers. A third issue is whether the child-care expense deduction available in Canada adequately addresses questions of equity between families with stay-at-home parents and families without. In this dissertation, I address these three issues using Canadian data on child outcomes, child-care use and costs, and parental labour supply. This dissertation is composed of five chapters. In the continuation of this chapter, I briefly describe the research undertaken in the three subsequent chapters, as well as the main results. Chapters II to IV explore each of the issues just described. In Chapter V, I offer concluding comments. 4 Statistics Canada: http:/www.statcan.ca/english/Pgdb/People/Labour/labor24c.htm, February 10, 2001. 5 Statistics Canada, The Daily, Statistics Canada Catalogue no. 11-001E, June 9, 1998. 6 Tabulations from cycle 3 of the National Longitudinal Survey of Children and Youth. 1 1.2 CHAPTER H - THE EFFECT OF PARENTAL WORK AND CfflLD-CARE USE ON CHILD OUTCOMES Chapter II examines the relationship between parental labour supply, child-care use, and pre-school children's cognitive and behavioural development using a sample of children aged two to five from the 1994, 1996 and 1998 National Longitudinal Survey of Children and Youth (NLSCY). Child outcomes include Peabody Picture Vocabulary Test (PPVT) for three and a half to five year old children, conduct disorder, property offenses and indirect aggression scores for four and five year old children, hyperactivity, emotional disorder and prosocial behaviour scores for children two to five years old, and physical aggression and separation anxiety scores for two and three year old children. The focus is on children in two parent families. Assuming that maternal labour supply, income and child-care use are exogenous variables, I find that maternal labour supply has little effect on the various scores but is associated with somewhat worse PPVT scores for children of mothers with above average education and parenting skills. For the average child, parental labour supply has no effect on PPVT scores. Maternal labour supply is associated with reduced hyperactivity for four and five year olds when parenting skills are better than average. The evidence also indicates that parental labour supply has no effect on conduct disorder and indirect aggression scores for four and five year olds and on hyperactivity, prosocial behaviour, and separation anxiety scores for two and three year olds. Four and five year old children of mothers who do not work have better property offense scores, but those of parents who work part-time have worse scores. Four and five year old children of fathers who work part-time have better emotional disorder and prosocial behaviour scores. Two and three year old children of fathers who do not work have worse aggression and emotional disorder scores. Hours in care have no overall effect on PPVT scores, but the evidence indicates that along this dimension, quality of care is increasing in income. Hours in care are also associated with higher 2 J hyperactivity and prosocial behaviour scores for all children, with higher emotional disorder scores for four and five year olds, and with higher aggression scores for two and three year olds. Overall, I find that children's behavioural scores are largely explained by parenting skills and family functioning, and by family and child characteristics. I find limited evidence that suggests maternal labour supply may be endogenous. This evidence indicates that maternal labour supply may have a significant negative impact on PPVT scores. 1.3 C H A P T E R H I - C H I L D - C A R E C O S T S A N D C A N A D I A N W O M E N ' S M A R K E T W O R K : H E T E R O G E N O U S P R E F E R E N C E S Chapter III presents estimates of the impact child-care costs have on the probability that Canadian women with young children will have returned to work. The estimation is performed on a sample of married women from the 1988 Canadian National Child Care Survey (CNCCS), who have relatively recent work experience (within the last five years) and children between the ages of 5 and 35 months. The sample is restricted to women whose children are older than four months to allow for the statutory maternity leave provisions in effect at the time to elapse. Prior/current occupation ranking and weeks worked in the last twelve months are used as additional control variables in two separate procedures to account for heterogeneity of preferences and habit formation. The results indicate that the labour supply of professional women is not strongly responsive to predicted current child-care costs, but that child-care costs have a significant impact on the labour supply of women belonging to other occupational ranks. Results also indicate that when recent labour supply is controlled for, there is little left to explain. The results are quite sensitive to whether women who take care of some of their children while at work are included or excluded from the sample. In particular, when these 3 women are excluded from the sample, the labour supply responsiveness to child-care costs of low skilled women increases dramatically. This indicates that low skilled women have fewer child-care choices than the more skilled. 1.4 CHAPTER IV-POTENTIAL INCOME AND THE EQUITY OF THE CfflLD-CARE EXPENSE DEDUCTION7 Critics of the child-care expense deduction (CCED) contend that it violates principles of horizontal and vertical equity. In chapter IV I find, however, that when equity is evaluated using potential rather than actual earnings as the measure of ability to pay, this contention does not generally hold. Potential earnings measure earnings that parents would earn if they worked for pay full-time at their predicted wage rate, while actual earnings are estimated using parents' usual hours of work. I argue that potential earnings are a superior measure of ability to pay because unlike actual earnings, they recognize the market value of leisure and home production. I use data from the 1988 Canadian National Child-care Survey and Labour Market Activity Survey to examine these issues. The data are used with 1988 tax rules and are converted into 1999 dollars in order to also be used with 1999 tax rules to evaluate equity. Horizontal equity is examined by investigating whether the existence of the CCED increases or decreases the difference between effective tax rates of families with similar earnings but different labour supplies. Vertical equity is evaluated by comparing CCED benefit rates of dual earner families with childcare costs and similar characteristics at different family levels of earnings. The analysis adds new perspectives to the Canadian debate on child-care related expenditures and challenges some widespread perceptions about the equity of the CCED. 7 Chapter IV, with minor differences, has been published in the Canadian Tax Journal volume 49, number 3, 2001. The title of the article is the same title as used for Chapter IV. 4 CHAPTER II THE EFFECT OF PARENTAL WORK AND CHILD-CARE USE ON CHILD OUTCOMES 2.1 I N T R O D U C T I O N According to the 1995 General Social Survey, a number of Canadians are concerned about the effect that maternal labour supply has on children. Zukewich Ghalam (1997) reports that while the majority of Canadian men (59 percent) and women (67 percent) agreed or strongly agreed that an employed mother can establish just as warm and secure a relationship with her children, over half of the men (59 percent) and women (51 percent) also agreed or strongly agreed that a pre-school child is likely to suffer if both parents are employed.8 Parental labour supply can adversely affect young children if substitute care is of lower quality than parental care or if parental care deteriorates because of work outside the home. Income spent on child-care services is not available for other uses. On the other hand, working parents who do not use substitute care are putting in a double shift and may have a lot less time to devote exclusively to their children. Finally, the additional family income that is available to working parents relative to non-working parents can offset potentially adverse effects of reduced parental time or lower quality substitute care. While parents worry about the welfare and development of their children, policy makers look for the right responses. Are increases in funded maternity or parental leaves beneficial to children? Should funding to child-care be increased (or decreased) and if so, should it be targeted to low-income families, users of licensed child-care, or should this funding be allocated to all child-care users? Are there other equally or more important factors than parental labour supply and child-care, which have an effect on children outcomes and which could be influenced via government policy? In this chapter, I attempt to find answers to some of these questions by examining the effect of parental (maternal) labour supply and child-care use on child cognitive and behavioural outcomes. 5 To perform the analysis, I use data from cycle 1 (1994), cycle 2 (1996) and cycle 3 (1998) of the NLSCY's master files. The cognitive dependent variable is the Peabody Picture Vocabulary Test - Revised (PPVT-R) for children three and a half to five years old.9 The behavioural dependent variables include hyperactivity, prosocial behaviour, emotional disorder, physical aggression and separation anxiety scores for children aged two and three, and hyperactivity, prosocial behaviour, emotional disorder, conduct disorder, indirect aggression, and property offences scores for children aged four and five.10 One of the difficulties that arises in estimating the relationship between employment, income, child-care use and child outcomes is that the employment, income and child-care use variables may be correlated with unmeasured child or family characteristics that also affect child outcomes. For example, parents who choose to stay at home with their children may be relatively more skilled at child-care than parents who choose market work. Alternatively, workers or high-income workers may have certain abilities that non-workers or lower-income workers do not have and that are not fully captured by education. The inclusion of a comprehensive set of explanatory variables that are relevant to children's outcomes and likely correlated with other explanatory variables may address this issue. For example, variables such as relationship and parenting skills are likely to have an influence on children outcomes and on labour supply and income.11 However, additional characteristics, such as ability, which may be partially but not fully proxied by education and relationship skills, are not available. If these relevant omitted characteristics are associated with the parents and can be considered to be fixed over time, then the availability of siblings data allows the use of family fixed-effects models.12 8 Zukewich Ghalam (1997), p. 16 9 From now on, the test is just referred to as PPVT. 1 0 Questions that make up the scores vary by age-groups. 1 1 Goleman (1998) presents evidence that interpersonal skills (or "emotional intelligence") have a considerable effect on labour market success. 1 2 It is reasonable to assume that in addition to labour supply and income, child-care choice depends on family characteristics: child-care is used primarily to meet parents needs for substitute care. While the choice of care is likely dependent on the child's age, this variable is observed. 6 Endogeneity can also be a problem if parental (maternal) labour supply is affected by child outcomes. For example, a mother might decide not to work if her child exhibits poor cognitive or behavioural outcomes. However, if this problem is a health issue rather than a cognitive or behavioural issue, it can be solved by including a measure of the child's health as an explanatory variable. If child cognitive and behavioural outcomes themselves are determinants of labour supply, the appropriate solution is an instrumental variables estimator provided appropriate instruments are available. For this analysis, data from cycle 1, cycle 2 and cycle 3 of the NLSCY are pooled to construct large cross-sectional samples. To allow parental employment to have differential effects for parents with different skill levels, education and parenting skills variables are interacted with measures of parental employment. The data sets are also merged longitudinally to allow the use of a related lagged behavioural score of children when they were two and three years old as an additional explanatory variable for their behavioural outcomes at ages four and five. The lagged behavioural score is considered a stock variable, which captures the effect of prior parental investments in the child. No prior related score is available for PPVT scores. To address potential omitted variables bias and endogeneity issues, I use the following strategy: 1) I include a comprehensive set of explanatory variables. In addition to various family and child characteristics, parenting styles, family functioning and parental time investments in their children's development via reading activities are included as explanatory variables. 2) I estimate a family fixed-effect model for PPVT scores. Because families are followed over time, two and three year old siblings of four and five year old children in cycle 1 are four and five years old in cycle 2, and two and three year old siblings of four and five year old children in cycle 2 are four and five years old in cycle 3. 7 The use of a fixed-effect model eliminates bias from unobserved family heterogeneity such as parental ability, which may be poorly proxied by parental education. 3) I use an instrumental variables estimator to estimate PPVT scores. Maternal labour supply is instrumented. The instrumental variables estimator corrects for problems of simultaneous determination of maternal labour supply and child outcomes. 2.2 L I T E R A T U R E S U R V E Y While there is a considerable body of literature on the effects of maternal employment and/or child-care on child outcomes using US data, findings are mixed. Much of the research examines one issue or the other: estimates from studies that examine the effect of labour supply without controlling for child-care embody both labour supply and child-care effects. A lot of the earlier work on child-care effects failed to appropriately control for differences between families. Some research uses data that provide measures of child-care quality to examine how important child-care quality is to child outcomes. The Canadian literature on these issues is just emerging as data from the NLSCY become available. To my knowledge, this is the only Canadian study that uses all three cycles of the NLSCY to examine the impact of both parental labour supply and child-care use on outcomes of pre-school children and that tests for endogeneity of maternal labour supply. Labour supply and child-care effects may differ from US to Canadian children for a variety of reasons. In particular, American and Canadian child-care services may not be comparable,13 and American and Canadian parents may have different levels of parenting skills. Hanushek (1992) uses four-year data (1971-1975) from the Gary Income Maintenance Experiment and estimates achievement growth models for school children and preschool achievement In Canada, centre care is delivered in the majority of cases by non-profit organizations, while in the US it is delivered primarily by for-profit organizations; in addition, the US has a large pool of illegal immigrants that can provide child-care services in people's homes. 8 for pre-school children. Al l the families in the sample are low-income blacks, some of which received payments under alternative negative income tax schemes, while others were part of the experimental control group. The achievement variables are the results of the Iowa reading comprehension and vocabulary tests. Hanushek investigates the trade-offs between number of children and their scholastic performance and finds that being early in the birth order is an advantage presumably because of the longer time spent in a small family. He finds no apparent impact of market work by mothers on test scores of either school children or preschoolers. Blau and Grossberg (1992) use a sample of three- and four-year old children from the 1986 NLSY to investigate the effect of maternal employment on children's PPVT scores. Independent variables include a measure of mother's verbal ability in 1979 as well as parental education at the time of the child's birth. They use a Hausman test for the heterogeneity of working versus non-working mothers, and find no evidence of heterogeneity.14 The IV estimator replaces labour supply variables with their fitted values. Their main findings are that maternal employment has a negative impact when it occurs in the first year of a child's life with a potentially offsetting positive effect when it occurs during the second and subsequent years. O'Brien Caughy, DiPietro, and Strobino (1994) use a sample of five- and six-year old children from the 1986 NLSY to examine the impact of day-care participation during the first 3 years of life on the cognitive functioning of school-aged children. They interact day-care participation variables with family income variables, and find that initiation of day-care before the first birthday is associated with higher reading recognition scores for children from impoverished home environments and with lower scores for children from more optimal environments. Ruhm (2000) uses a sample of three- to six-year old children from multiple years of the NLSY to investigate the effect of parental employment on PPVT scores of three- and four-year old children, and on the reading and math achievements of five- and six-year old children. To control for potential unobserved heterogeneity between working and non-working mothers, Rhum uses maternal 9 employment prior to and after the birth of the child, as well as a variety of observed family characteristics, including child-care use. He also estimates fixed-effect siblings' models. The investigation suggests that maternal employment during the first three years of life has a small negative effect on the verbal ability of three- and four-year olds, and a substantial negative impact on the older children's reading and math achievement. In addition, paternal and maternal employment are found to have qualitatively similar effects. Greenstein (1995) examines the impact of maternal employment on children's PPVT scores using children bom to mothers who participated in the 1979 NLSY and were between the ages of 14 and 21 at the time. The children were between 48 and 83 months of age at either the 1986, 1988, or 1990 NLSY interview. Greenstein's primary concern is to determine whether there are any differential effects of maternal employment on child cognitive outcomes for families differing in resource level. The hypothesis is tested using interaction variables for family income and maternal labour supply in OLS regression models. Separate regressions are estimated for Hispanics, blacks and others. No evidence is found for differential effects. Hill and O'Neill (1994) use 1986 and 1988 NLSY data to examine the impact of family endowments on children's PPVT scores. They find large and significant positive effects of mother's AFQT scores, her schooling, and the grandparents' schooling. They also find that mothers' labour supply and welfare participation both have significant negative effects on the score. Selection models are used to account for potential unobserved heterogeneity with respect to fertility,15 labour supply and welfare participation. Mott (1991) uses 1986 NLSY data to examine how child gender and health relate to developmental effects of infant care. Developmental measures include scores at ages 1-4 on Memory for Location, Motor and Social Development, and PPVT measures. Mott finds that the average 1 4 Maternal labour supply coefficients are very imprecisely estimated in the IV equation. 1 5 In the N L S Y , the base unit of analysis is the mother; mothers with higher AFQT test scores are less likely to have children early in life, so that the children included in the sample are more likely to come from mothers with lower AFQT scores. 10 young child's abilities are not sensitive to the generic nature of his or her child-care arrangements, but that healthy infant girls benefit from being taken care of extensively by caretakers other than the mother and that infant boys with health problems benefit from spending more time with their mothers. Baydar and Brooks-Gunn (1991) use 1986 NLSY data to examine the effect of maternal employment and child-care arrangements on preschool children's cognitive and behavioural outcomes. They find that maternal employment in the 1st year has detrimental effects on cognitive and behavioural scores. They also find that grandmother care is the most beneficial arrangement for cognitive development of children in poverty and that maternal care is most beneficial to boys behavioural development, while baby-sitter care is most beneficial to girls. Phillips, Brooks-Gunn et al (1998) examine the claim by Herrnstein and Murray (1994) that contemporary racial differences in socioeconomic status explain about a third of the gap on standardized test scores between African-Americans and white Americans. They use data on children from the NLSY to estimate the gap with an expanded set of explanatory variables such as mothers' high school quality, parenting practices, grandparents' educational attainment, children's birth weight and children's household size. They find that a broader index of family environment may explain up to two thirds of the gap. Lefebvre and Merrigan (1998) use data on four- to eleven-year old children from cycle 1 of the NLSCY in various OLS specifications to examine the effect of income, maternal employment and family background on child development outcomes. The dependent variables include the PPVT scores for four- and five-year old children, and hyperactivity, prosocial behaviour, emotional disorder and physical aggression/conduct disorder scores for children aged 4-11. They find that with an all-inclusive specification, maternal work in the previous year has a weak negative impact on PPVT scores and does not have an impact on behavioural scores. Lefebvre and Merrigan (2000) use cycle 1 of the NLSCY to examine the effect of child-care arrangements on children's development outcomes. The dependent variables are PPVT scores for 11 four- and five-year old children, and Motor and Social Development (MSD) scores for children aged 0-47 months. A mother fixed-effect model is estimated to control for unobserved family characteristics, along with various OLS specifications. They find that when family and child characteristics are controlled for, infant-toddler non-parental arrangements have insignificant impacts on PPVT and MSD scores. Lipps and Yiptong-Avila (1999) use cycles 1 and 2 of the NLSCY to examine the effect of non-parental care on four and five year old children's subsequent school achievements. They find that children who were in non-parental care arrangements two years earlier, are more likely to have top scores in mathematics than children who were not.16 Currie and Thomas (1995) use data from the NLSC and the National Longitudinal Survey's Child-Mother files (NLSCM) to examine the impact of Head Start on school performance, cognitive attainment, preventive medical care, and health and nutritional status.17 Participant children are compared to their non-participant siblings to control for unobserved heterogeneity across families. A similar procedure is also used to compare siblings with other pre-school participation to siblings without any pre-school participation. The effects of Head Start are compared with pre-school effects. Currie and Thomas find that Head Start is associated with large and significant gains in PPVT scores among whites and African-Americans. However, the gains are quickly lost for African-Americans. In a subsequent paper (1997d), Currie and Thomas find that African-American children tend to attend worse schools than white children, which can partly explain why the Head Start effects appear to fade out for the former group. After a review of the literature on the effect of child-care on children outcomes, Burchinal (1999) reaches the following weak conclusion: "After 30 years of research into the relation between 1 6 It is not clear from the report whether appropriate controls were used in the estimation method. Since child-care use is highly correlated with income, labour supply, and education, the lack of controls can result in an overestimate of the benefits of non-parental care. 1 7 Head Start is a US "program that aims to improve the learning skills, social skills, and health status of poor children so that they can begin schooling on an equal footing with their more advantaged peers." (Currie and Thomas, 1995, p. 341) 12 child-care experiences and child development, it appears that some aspects of child-care experiences are related to some developmental outcomes for at least some children." Burchinal also states that "studies of early intervention for children from families living in poverty suggest that high-quality child-care, beginning in infancy, can have large long-term effects on cognitive development." Finally, Burchinal also finds that "there is some evidence that extensive care may be modestly negatively related to social outcomes and that center care may be modestly positively related to cognitive outcomes."18 2.3 MODEL AND ESTIMATION ISSUES The analysis is done in the context of Becker's model of household utility maximization, where utility is derived from the consumption of activities that require market and time inputs. One such activity includes rearing children. Utility is increasing in child quantity and quality, where quality may be evaluated on the basis of a variety of child outcomes, including verbal and mathematical ability as well as behaviour. Parents select their labour supply and consumption of goods and services, including child quantity and quality subject to child outcomes production functions and a time constraint. I estimate child outcomes production functions. Child outcomes represent various aspects of child quality and depend on a variety of parental inputs and on other factors that affect parental inputs. Parents can affect child quality by investing time and market goods in their children. A stay-at-home parent will invest her own time in teaching children a variety of skills. In families where all parents work, parental schedules may be staggered, or child-care services may be purchased. Beneficial goods may include books and educational toys, as well as the provision of a generally healthy and pleasant living environment. Since highly skilled parents can transmit more skills and wealthier parents can purchase more goods, parental education, parenting skills,19 and money income 1 8 P. 89. 1 9 While it may be argued that causation runs from child to parent rather than from parent to child, it is assumed here that the direction of causation is from parent to child. 13 are all expected to contribute favourably to child outcomes. Parental education is expected to be strongly associated with cognitive scores, while parenting skills are expected to have a strong impact on behavioural scores. The amount of time a parent spends reading to/with her child is a time investment that will also likely have a strong impact on cognitive scores. Since more children are competing for resources in larger families, the number of children is expected to adversely affect cognitive outcomes. Similarly, since the older children in a family tend to have lived in a smaller family longer than the younger children in the same family, birth order is expected to result in poorer cognitive outcomes. If behavioural outcomes are less highly dependent on the level of parental attention, a higher birth order will not necessarily entail poorer behavioural outcomes: older children in the family will have spent more time with less experienced and possibly less tolerant parents, and parents may exhibit favoritism towards younger children and have higher expectations and be more critical of older ones. Today's families are relatively small. A pre-school child who stays at home with a parent is therefore likely to receive more one-on-one attention from the parent than she would receive in a child-care setting: child/staff ratios for children aged two to six years of age in Canadian full-day centres range from 4 tol for two year olds to 15 to 1 for 6 year olds.20 Hence, just from the adult/child ratio perspective, assuming that parents are as good teachers as child-care providers, a child who is taken care of by a stay-at-home parent should have better cognitive outcomes than a child who spends several hours in a child-care setting, ceteris paribus.21 On the other hand, working parents contribute additional income, which may compensate for the loss of parental time. In addition, working parents who do not use substitute care services may not have as much time to spend one-on-one with their children than stay-at-home parents. Some working parents stagger their work schedules to avoid child-care expenses. This may benefit the child if the parents are still able to provide more positive attention to their children than 2 0 Childcare Resource and Research Unit (2000). 2 1 There could be offsetting beneficial spillover effects among children in care, however. 14 the children would receive in a child-care setting. On the other hand, this may be detrimental to the child relative to the use of substitute care if such parents are (more) overworked (than child-care workers) and have little quality time left to offer their children. It is therefore an empirical question whether children of working parents are better off if their parents use child-care or stagger their work schedules. If child-care has a negative impact on outcomes, omitting child-care variables may result in more., negative coefficient estimates for labour supply as most working parents of preschoolers use child-care.22 When measures of labour supply, hours in care, and household income are included, the labour supply variable should pick up the effect of parental time, the hours in care variable should pick up the substitute care effect net of parental time savings and pecuniary costs, and the household income effect should pick up the effect of income, before child-care costs and taxes. Adult/child ratios may be less important for behavioural outcomes, particularly for non-infants. Furthermore, children may leam valuable social skills in a group environment. Hence, substitute care may not have any adverse impact on child behaviour. The reduced adult attention may be offset by the attention received from and given to other children. If children benefit from being taken care of by parents because of the additional adult-intensive time they can enjoy, it is also possible that the importance of this effect will depend on parental education and parenting skills. To investigate this possibility, I test for interactions between parental time, and education and parenting skills. Another question of interest is the existence and extent of variability in child-care quality. While the quantity (hours) of child-care used is available in the data, no measures of child-care quality are available. To investigate this, I test for interactions between household income and hours in care. 2 2 Tabulations for the PPVT sample show that 73 percent of children in families where parents work full-time are in some form of substitute care and that 45 percent of children in dual-parent families where the mother works part-time are in substitute care; the corresponding percentage for children single parent families is 53 percent. 15 To illustrate the model, consider the following linear child outcome function: (2.1) Cq i k t = a 0 + aiCVi + 0 2 X k + ajKiu + a»W ik t where: = outcome q for child i in family k at time t, X k = a vector of (relatively) fixed family demographic variables and endowments such as region, parental education and parenting skills, Ki k = a vector of fixed child demographic variables such as gender and parental age at birth, and W i k t = a vector of family k investments in child i (flow variables) such as parental time and income, and other period-specific variables at time t.23 In 2.1, child quality in period t is a function of child quality in the prior period, of current parental investments and other period-specific variables, and of fixed demographic variables and endowments. The coefficient on the lagged child quality variable (ai) captures the effect of prior parental time investments and other period-specific variables, coefficients on fixed variables such as education and parenting skills and child gender (ai and 03) capture the current or short-term effect of these variables, and coefficients on current flow variables such as parental time and income (04) capture the effect of current investments.24 In the absence of prior information on child quality, 2.1 can be solved recursively, yielding the following function of current and lagged period-specific explanatory variables and fixed demographic variables in a three period model:25 (2.2) Cqikt=(ao + a 0 a, + a 0 a,2) + (az + a 2 a, + <Xi2 )X k + (a3 + a 3 a i +a 3ai 2)Ki k + 0 4 ai 2 W ik,.2 + CL , ai W iUt., + 04W i k t An econometric specification for (2.2) in the context of the classical linear regression model would be as follows: 2 3 Note that while some of these investments are child-specific (reading time), other investments such as income and time (labour supply) are family-specific. 2 4 Hanushek (1992) uses this framework. 16 (2.2)' C q i k t=(ao + aoa, + a 0 a, 2) + (Oz + a, + Oz a,2 )Xk + (a3 + a 3 aj +a 3 a 1 2 )K i k + 0 4 ai 2 W i k , . 2 + 0 4 0 1 W i k M + ctjWjk, + eikt Assuming that the error term 8 i k t ~ N(0,cr2) and 8 i k t is homoskedastic and not correlated with any of the regressors, and that except for possible correlations between error terms from the same family, the error terms are not correlated with each other, OLS with robust variance estimation on (2.2)' will yield unbiased and efficient estimates.26 In (2.2)', coefficients on period-specific variables capture current and prior investment effects on current outcomes, while coefficients on fixed variables capture the long-term effect of these variables. The intercept is also larger in the absence of a lagged outcome. When working with purely cross-sectional data, prior information on child quality and lagged period-specific variables is not available. Coefficients on fixed variables capture the long-term effects of fixed variables if fixed variables are not correlated with lagged period-specific variables and additional effects if fixed variables are correlated with past values of period-specific variables. Coefficients on current period-specific variables capture only current effects if current period-specific variables are not correlated with their past values, but also capture past effects if current and past values are correlated. In particular, if current and past labour supply are positively correlated, the coefficient on current labour supply will be larger (in absolute terms) than it would have been in the presence of past labour supply.27 Ruhm (2001) and Blau and Grossberg (1992) use this framework. 2 6 In the OLS estimations, robust variance estimation is used. The procedure further allows error terms for observations from the same family to be correlated, while error terms for observations from different families are not. This procedure does not affect coefficient estimates. For a discussion of the Huber-White robust variance estimator and the relaxation of error term independence for certain groups of observations, see Stata User's Guide Release 7, pp. 254-258. All estimations are weighted. 2 7 While the use of a full longitudinal set of variables appears warranted, the longitudinal sample sizes for the various age groups are small and the number of potential additional explanatory variables is large. Furthermore, the survey is conducted every two years so that even with longitudinal data, half of the lagged variables are not available. Consistently with the OLS results that use the cross-sectional data and are presented in the next section, estimates using OLS and the small longitudinal sample for PPVT scores did not yield significant labour supply and child care effects. 17 A standard concern in models that look at the effect of parental labour supply on child outcomes is that there could be unobserved heterogeneity between workers and non-workers. In other words, some variable in the vector X of (2.2)' is unobserved and thus included in the error term eikt, and this variable may be correlated with (one of) the Ws (or Xs or Ks). For example, non-workers may be relatively more skilled at childcare than at market work, and/or non-workers may be staying home because of poor health or a child's poor health.28 With a simple OLS regression, unobserved skill differences between workers and non-workers gives rise to omitted variables bias: estimated coefficients are biased because labour supply is contemporaneously correlated with the error term. In this instance, the omitted variable is related to the family. A similar concern arises if mothers of sick children decide to quit work and care for their children. Sick children will likely have worse scores than healthy children. If child health is unobserved, estimated coefficients are biased.29 In this instance, the omitted variable is related to the child: a variable in the vector K of (2.2)' is unobserved and thus included in the error term 8jkt, and this variable is correlated with (one of) the Ws. Another concern that arises is that maternal labour supply may be endogenous. For example, mothers of children with poor cognitive or behavioural outcomes may decide to quit work and spend more time with their children. As long as the child outcome directly causes the parent to not work, labour supply is endogenous. An endogenous labour supply would be contemporaneously correlated with the error term in (2.2)' leading to biased OLS estimates. To address potential omitted variables bias and endogeneity issues, I follow a three-step approach. The inclusion of a comprehensive set of explanatory variables such as child health, a variety of measures of parental investments such as reading time, parenting skills and family functioning, which are arguably excellent proxies for more general social skills, is likely to capture a substantial portion of heterogeneity and is the first of these steps. For the majority of this chapter, the 2 8 A variable indicating if the child is in poor health is included in the analysis. 2 9 Child health is observed in the NLSCY, and is controlled for in the estimations. 18 determinants of child outcomes are estimated with the assumption that any heterogeneity is captured by this expanded set of observed explanatory variables, and that child outcomes do not directly affect maternal (parental) labour supply. To allow for the possibility that heterogeneity is not fully captured by the comprehensive set of explanatory variables, but that heterogeneity is parent-related rather than child-related, and that child outcomes do not directly affect parental labour supply, I estimate a family fixed effects estimator. Rewriting equation (2.2)' for siblings i and j , and subtracting j 's equation from i's equation yields: (2.3) C q i k t - C q j k . = + (a3 + a3 a, + a3 a,2 )(K,k- KJk)+ a* aj2 (Wik t.2 -Wjkt.2) + 04 aj (WiUt., -WJkt. 1) + a,(Wikt -Wjkt) + (E i k t - ejkt) Assuming the error term (s^ - Sjkt) ~ ^0,2(0^- 0(8^,8^)) and homoskedastic and not correlated with itself or any of the regressors, OLS on (2.3) will yield unbiased and efficient estimates. The error variance follows from the error term correlation assumptions made in (2.2)' regarding observations from the same family. In equation 2.3, all family fixed elements disappear, leaving only period-specific differences and child differences. As long as the unobserved heterogeneity is a family fixed-effect and child outcomes do not directly affect parental (maternal) labour supply, estimating (2.3) with OLS will yield unbiased coefficient estimates. If the unobserved heterogeneity is child-related, that is if the missing variables are part of the K vector of variables rather than part of the X vector of variables, their effect will be included in (Sjkt - 8jkt) and the fixed-effect estimator will also suffer from omitted variables bias, as long as the missing variables are correlated with other regressors.30 If parental (maternal) labour supply is partially and directly determined by child outcomes, the simultaneity is not resolved in (2.3) and the endogenous labour supply will be contemporaneously correlated with the If the missing variables are orthogonal to the regressors, coefficient estimates will not be biased, but variance estimates will still be. See Greene (1993), 246-47, Tor a discussion of omitted variables bias. 19 error term. In a purely cross-sectional estimation, labour supply effects will likely capture prior labour supply effects as current and past labour supply are probably correlated. Now suppose that a child outcome directly affects maternal labour supply.31 As discussed previously, the mother of a child with poor outcomes may choose to reduce her labour supply. If this is the case, negative effects of maternal labour supply will be underestimated, or alternatively, its positive effects will be overestimated. To illustrate, consider the following modification of (2.2)' into a simultaneous equations model: (2.4) C \ t = Po + p i C V , + kXu + PjKjk + p4Z,kt+ P J H K , + £ I K T (2.5) H k t = (po + <PiRk + <fcSik + (p 3C q i k t + p k , 3 2 where: Zjk, = a vector of family k investments in child i (flow variables) such as income, and other period-specific variables at time t, H k t = maternal hours of work in family k a time t, R k = a vector of (relatively) fixed family demographic variables and endowments such as region, parental education and parenting skills, and S i k = a vector of fixed child demographic variables such as gender. Assume the error term S j k t ~ N(0,aE2) and is homoskedastic and not correlated with any of the regressors, and except for possible correlations between error terms from the same family, the error terms are not correlated with each other. Also assume the error term u k t ~ N(0,aM2) and p k t is homoskedastic and not correlated with any of the regressors and except for possible correlations between error terms from the same family, the error terms are not correlated with each other. That is, parental labour supply at time t is correlated with parental labour supply at other times.33 3 1 Alternatively, there may be unobserved heterogeneity at the child level, that affects child outcomes and is correlated with maternal labour supply. 3 2 Note that the direct parental time investment in the child is not observed. Parental hours of work are observed and are necessarily the same for each child in the family. 3 3 The IV estimator for PPVT scores uses the cross-sectional data. Hence, parental hours of work are for one period only. 20 Furthermore, also assume that E(ejktukt) = E(8; k tp k t) = p. Finally, assume that H k t is censored, that is, H k t = max(0, H k t*). Here H k t is a censored endogenous explanatory variable. This is a variant of the model due to Nelson and Olsen (1978), which is described in Maddala (1983) chapter 8.8 and in Greene (1998) section 28.4.34 Suppose there is a negative random shock for e^. This causes a decrease in C q i k t , which in turn causes an increase in Hkt. The variable H k t is contemporaneously correlated with the error term Sik, and OLS on (2.4) yields biased estimates. If sufficient and suitable instruments can be found for H k t , an instrumental variables estimator can be used to estimate (2.4). A suitable instrument will be contemporaneously uncorrelated with 8 j k t , and should be highly correlated with maternal labour supply. In other words, a good instrument would explain labour supply but not child outcomes. If maternal hours of work were not censored and good instruments were found, three-stage least squares could be used to estimate (2.4). However, because three-stage least squares does not account for censoring in (2.5), the Nelson-Olsen estimator is used instead.35 The procedure for the Nelson-Olsen estimator is described in the next section. For the majority of this chapter, exogeneity of right hand side variables is assumed. The basic equation that is estimated is as follows: (2.6) C q t= a + (pCqt., +p,LFPm+p2LFPf + p3EDm+p4EDf + p5PS + p6EDm* LFPm+ p 7ED f* LFPf + p8PS*LFPPMIC+ p9Y + p10Y*THCC + p nTHCC + 6X + 8, where: C q t = child outcome q at time t, 3 4 Except for the possible correlation between error terms from the same family, the model is identical to Nelson and Olsen's model. 3 5 1 estimated a three-stage least square model for (2.4) and (2.5) also, ignoring the censoring issue. The coefficient on maternal hours of work had the same (negative) sign as in the Nelson-Olsen estimator, was statistically significant but was more than twice as large as the Nelson-Olsen estimator coefficient which is reported in the next section. 21 LFPm= mother's labour supply, represented by dummy variables or hours of work LFP f= father's labour supply, represented by dummy variables or hours of work LFP p m k = person-most-knowledgeable's labour supply, E D m = mother's education, E D f = father's education, E D p m i ^ person-most-knowledgeable's education, PS = parenting skills, Y = family income, X = a vector of child and family characteristics THCC = total hours in care, and e ~ N^cr 2 ) and e is homoskedastic and not correlated with any of the regressors, and except for possible correlations between error terms from the same family, the error terms are not correlated with each other. In specification I, p5, p6, P7, Ps, and P10 are restricted to be equal to zero. This is a basic equation, where parenting skills and reading activities are assumed to have no effect on the score, and where no interactions between parental labour supply and education are allowed. In addition, no interaction between hours in child-care and household income is allowed. Finally, 9 is also restricted to be equal to zero since specification I is estimated for PPVT scores only and no prior scores are available. In specification II, p6, P7, Ps, and p ) 0 are restricted to be equal to zero. This is an enhanced basic equation, where no interactions between parental labour supply, and education and parenting skills are allowed. In addition, no interaction between hours in child-care and household income is allowed. Finally, 9 is also restricted to be equal to zero when strictly cross-sectional data are used. However, parental skills effects are allowed as 3 6 Hours of work are only used in equations that test whether labour supply is endogenous. 22 well as the effect of reading to/with the child. Specification II is the main specification and is estimated for all scores. In specification III there are fewer restrictions. Here, only Pio is restricted to be equal to zero. Hence, no interaction between hours in child-care and household income is allowed. The cross-sectional data restriction is also imposed when cross-sectional data is used. Specification III is estimated for PPVT and hyperactivity scores. Specification IV has no restrictions, except for the cross-sectional restriction when applicable. Hence, interactions are allowed for labour supply with education and parenting skills, and for hours in care and household income. Specification IV is estimated for PPVT 37 scores. The cross-sectional restriction is always imposed for PPVT scores as no prior score is available for the vast majority of children. For the same reason, it is also imposed for two-and three year old children. This restriction is relaxed when prior scores are available. This is applicable for behavioural scores of some of the four and five year old children in cycle 2 and cycle 3. Some of these children have a prior related score in the previous cycle, when they were two and three years old. With longitudinal identifiers, the records can be linked. When the cross-sectional restriction is imposed, coefficient estimates for period-specific variables are likely to include effects of the related lagged variables. I estimated a variety of alternative specifications with the PPVT scores before choosing the general specification represented by equation (2.6). I estimated a specification to test whether the effect of income from the husband differed from the effect of income from the wife. To do this, I included a linear and a square term for wife and husband income and an interaction term for husband and wife income. The equation I estimated can be written as follows: 3 7 Specifications III and IVwere estimated for all behavioural scores but are not presented here. There were no systematic/significant interactions for behavioural scores except for hyperactivity scores. 23 (2.7) C q = a + piLFPm +p2LFP f + p 3 ED m +p4ED f + p5PS + p 6 Y f + p 7 Y m + p 8 Y f 2 + p 9 Y m 2 + p 1 0 Y f Y m + p„*THCC + 5X + e, Equation (2.7) is a restricted version of equation (2.6) in terms of labour supply, and education and parenting skills interaction, but it relaxes the assumption that the effect of household income is independent of whether the father or mother earns the income. If maternal and paternal income have the same effects, then (i) P6 = Pv (ii) p8 = p9, and (iii) p 1 0 = 2p9 = 2p8 I found a p-value of .29 for the joint test of these hypotheses, and thus did not reject the null hypothesis that the effects of male and female income on the PPVT test scores are equal. From then on I used the combined measure of income. I tried specifications with income and its square and specifications with income dummy variables. The latter indicate that income effects are not quite as smooth as a quadratic specification would suggest38 and I opted for the latter specification. I estimated specifications with hours of work 3 9 rather than dummy variables for labour supply. In some of these specifications I included square terms for hours of work and mother-father hours of work interaction terms. Except for a negative coefficient on the latter variable, none of the labour supply coefficients were significant. I estimated these equations with and without interaction terms for labour supply, education and parenting skills. None of these specifications performed better than the ones reported in this chapter. I also estimated equations that used dummy variables to represent parental education, rather than continuous variables. Continuous variables performed better with R-squared used as the decision criterion. 3 8 Chi ldren i n families with incomes o f $50,000 - 59,999 had the best P P V T scores. 24 The specification presented here performed only marginally better than other specifications. Findings are consistent across specifications. I prefer the labour supply dummy variable model because it allows non-linearities in the labour supply interactions with education and parenting skills. I V estimators use hours o f work rather than labour supply dummy variables. 25 2.4 DATA AND PRELIMINARY STATISTICS Table 2.1 shows the weighted means, ranges and standard deviation of the dependent variables and explanatory variables for the cross-sectional observations.40 These are divided into two subgroups, one for children in dual parent families (dp) and one for children in single parent families (sp). Many of the reported means are for dummy variables, in which case the range column is left blank. While the first column (PPVT Group) coincides with the sample used in the PPVT OLS regressions, the other columns include all of the cross-sectional observations in the particular age categories.41 Cross-sectional observations are children selected from particular age groups in each one of the cycles of the survey. These observations are then pooled to form a larger sample size. The household income for observations from cycle 1 and cycle 2 is adjusted to 1997 levels using the consumer price index. The data used in the estimation includes all children within the particular age categories, except for children in grade 1, as long as the score is available for the child. 4 2 The PPVT score is a standardized score ranging from 46-160. The behavioural scores were derived from weighting categorical responses to a series of questions asked of the PMK about the child. Except for the prosocial behaviour score, all other behavioural scores are increasing in undesirable behaviour. Means and regression estimates are weighted. 4 1 Some of these observations are not included in the OLS regressions because either the dependent variable or the lagged related score variables (when applicable) is missing. The means are for all observations in the age category. 4 2 While some explanatory variables are missing, the observations were kept, but dummy variables were included in the regressions to identify these observations. The missing explanatory variable is coded as a zero, which allows the dummy variable coefficient to pick up the average full effect of the missing explanatory variable for these observations. I chose this methodology rather than dropping the observations or substituting means of non-missing explanatory variables because of the potential for selection bias: for example, I found that average hours of work are more than twice as much for mothers whose education is missing than for mothers whose education is available. Hence, since labour supply is highly correlated with education, mothers whose education is not available are likely considerably more educated than the average. 26 TABLE 2.1 Means, Ranges and Standard Deviations of Variables PPVT 4-5 YEAR OLDS (X)2-3 YEAR OLDS DP SP | DP SP | DP SP RANGE STD. DEV.* N 11254 1981 12836 2071 8733 1350 blank=dummy blank=dummy Dependent Variables PPVT Score Hyperactivity - inattention Prosocial behaviour Emotional disorder -anxiety Physical aggression & opposition Separation anxiety Conduct disorder Indirect aggression Property offences Employment and Education Father years of education Did not work Worked part-time Worked full-time Conditional/Interacted Parenting Skills 15.32 3.31/2.93 4.06/2.80 2.15/1.40 3.01 2.01 1.86 1.29 1.28 Mother/PMK did not work 0.29 0.37 0.30 0.38 0.29 0.37 Mother/PMK worked part-time 0.32 0.23 0.31 0.23 0.33 0.25 Mother/PMK worked full-time 0.39 0.40 0.39 0.39 0.39 0.38 Father did not work 0.05 0.06 0.05 Father worked part-time 0.09 iH 0.09 0.08 Father worked full-time 0.86 0.85 0.87 Mother/PMK years of education Did not work 12.27 11.01 12.25 11.01 12.24 10.66 0-20 2.20/2.27 Worked part-time 13.01 12.09 12.98 12.14 13.13 11.97 0-20 2.13/2.18 Worked full-time 13.17 12.24 13.17 12.33 13.40 12.20 0-20 2.14/2.17 2.61/2.67 2.54/2.42 2.46/2.45 Mother/PMK did not work 14.60 13.58 -8.84 -9.26 -9.05 -9.50 0-20/-25-0 3.34/3.73** Mother/PMK worked part-time 15.13 14.56 -9.13 -9.74 -9.01 -9.10 0-20/-25-0 3.12/3.43** Mother/PMK worked full-time 14.85 14.13 -8.68 -9.45 -8.85 -9.01 0-20/-25-0 3.30/3.67** Child and Family Characteristics Age in months/five/three years old 59.75 60.01 4.50 4.52 2.50 2.50 Boy 0.50 0.53 0.51 0.55 0.51 0.51 Speaks no English or French 0.02 0.02 0.03 0.02 0.04 0.02 Immigrant parents 0.10 0.04 0.11 0.05 0.12 0.06 Poor health 0.03 0.04 0.03 0.05 0.02 0.04 In kindergarden 0.62 0.60 0.63 0.63 wm Mother (parent) age at birth 29.30 26.93 29.30 27.04 29.68 26.41 1 + 5.74/5.78 Number of children < 17 2.38 1.96 2.38 1.95 2.21 1.95 1 + 0.98/1.02 Number of older children 0.84 0.66 0.83 0.64 0.88 0.72 0 + 0.96/1.00 Reads or is read to daily 0.58 0.50 0.58 0.50 0.57 0.52 " " several times a day 0.09 0.06 0.09 0.06 0.14 0.14 PMK depression score 4.07 7.46 4.02 7.29 4.42 8.36 0-35 4.62/4.89 Ineffective parenting score 8.92 9.48 8.86 9.45 8.96 9.22 0-25 3.62/3.70 Punitive score 8.79 9.08 8.74 9.03 9.02 9.11 0-20 2.03/2.28 Consistency score 14.87 14.02 14.86 13.95 14.49 13.72 0-20 3.25/3.36 Positive interaction score 14.59 14.35 14.57 14.42 16.27 16.13 0-20 2.68/2.54 Family dysfunction score 7.63 9.62 7.69 9.68 7.92 10.56 0-36 5.05/5.24 continued 21 T A B L E 2.1 (continued) PPVT 4-5 YEAR OLDS (X)2-3 YEAR OLDS DP SP DP SP DP SP RANGE STD. DEV.* Family Income Less than $20,000 0.05 0.58 0.06 0.59 0.06 0.63 $20,000 - $29,999 0.09 0.19 0.09 0.18 0.09 0.18 $30,000 - $39,999 0.13 0.12 0.14 0.11 0.15 0.10 $40,000 - $49,999 0.15 0.05 0.15 0.05 0.15 0.05 $50,000 - $59,999 0.14 0.03 0.14 0.03 0.14 0.02 $60,000 - $69,999 0.11 0.01 0.11 0.01 0.12 0.01 $70,000 - $79,999 0.10 0.01 0.09 0.01 0.08 0.01 $80,000 + 0.22 0.01 0.22 0.01 0.21 0.01 Location Newfoundland 0.02 0.02 0.02 0.02 0.02 0.02 Prince Edward Island 0.00 0.00 0.00 0.00 0.00 0.00 Nova Scotia 0.03 0.04 0.03 0.04 0.03 0.04 New-Brunswick 0.03 0.02 0.02 0.02 0.02 0.03 Quebec 0.25 0.22 0.24 0.23 0.24 0.22 Ontario 0.39 0.38 0.39 0.36 0.39 0.40 Manitoba 0.04 0.04 0.04 0.04 0.04 0.04 Saskatchewan 0.04 0.03 0.04 0.03 0.03 0.04 Alberta 0.10 0.07 0.11 0.08 0.10 0.09 British Columbia 0.11 0.17 0.11 0.17 0.12 0.13 Urban - 500,000 + 0.45 0.47 0.45 0.48 0.47 0.45 Urban - 100,000-499,000 0.18 0.19 0.18 0.18 0.18 0.21 Urban < 100,000 0.21 0.24 0.21 0.24 0.20 0.24 Child Care Hours Total hours 11.41 13.82 11.57 13.72 14.73 16.85 0-168 17.61/19.54 Average hours for those in care 25.32 31.86 25.74 31.91 30.96 36.68 0-168 18.02/17.32 Child Care Users (proportion of children in type of care by type of care) Girl - non-relative outside 0.09 0.08 0.10 0.08 0.09 0.08 Boy - non-relative outside 0.10 0.07 0.10 0.07 0.11 0.08 Girl - non-relative inside 0.04 0.05 0.04 0.04 0.04 0.03 Boy - non-relative inside 0.04 0.03 0.05 0.04 0.04 0.03 Girl - relative 0.04 0.03 0.04 0.02 0.05 0.05 Boy - relative 0.04 0.04 0.04 0.04 0.05 0.06 Girl - daycare/after sc. 0.05 0.07 0.05 0.07 0.05 0.08 Boy - daycare/after sc. 0.06 0.09 0.06 0.09 0.06 0.09 Girl - other 0.01 0.01 0.01 0.01 0.01 0.01 Boy - other 0.01 0.01 0.01 0.01 0.01 0.01 Average hours in primary care by primary care Girl - non-relative outside 21.58 28.21 21.61 28.18 28.50 30.75 0-168 12.61/13.13 Boy - non-relative outside 22.12 25.93 22.34 25.59 27.64 34.08 0-168 13.00/13.73 Girl - non-relative inside 22.77 23.27 22.97 25.02 24.58 22.42 0-168 14.36/14.07 Boy - non-relative inside 22.93 23.55 22.81 25.60 25.36 31.91 0-168 16.95/15.54 Girl - relative 18.45 23.98 18.26 24.53 25.74 31.22 0-168 13.48/14.62 Boy - relative 18.83 23.34 22.54 23.66 23.90 30.20 0-168 23.93/14.66 Girl - daycare/after sc. 23.89 26.33 23.71 26.86 29.72 33.06 0-168 13.40/12.40 Boy - daycare/after sc. 25.03 28.05 24.99 27.50 30.76 36.51 0-168 13.16/12.97 Girl - other 22.46 12.14 23.62 12.14 19.65 4.57 0-168 18.36/16.18 Boy - other 13.16 23.81 14.13 23.95 19.87 30.51 0-168 14.98/14.89 * Except for PPVT scores, standard deviations are for children in dual-parent families for the 4-5 year olds (x) and the 2-3 year olds. ** The first standard deviation is for the PPVT group (consistency score), while the second is for the 4-5 year olds (x). 28 Employment dummy variables combine work and study. For example, a full-time student is treated as fully employed. A part-time student who does not work is treated as employed part-time, and a part-time student who also works part-time is treated as employed full-time.43 In addition, a full-time worker has worked an estimated 30 hours per week or more in the last year, and a part-time worker has worked some hours but an average of less than 30 hours per week in the last year. Although estimations for children in single parent families are only reported for PPVT scores, single parents are included in Table 2.1 to provide an idea of the differences in outcomes and explanatory variables for children in these families. PPVT and behavioural scores indicate that children in single parent families have worse outcomes than children in two parent families, except for prosocial behaviour where two and three year old children of single parents have better scores than those in two parent families. Most children live in families where the mother or single parent works or studies. However, children of single parents are more likely to have a stay-at-home parent (37-38 percent) than children of married mothers (29-30 percent) are likely to have a stay-at-home mother. Workers are more educated than non-workers and married mothers are more educated than single parents. There are six parenting skills and family functioning measures: the PMK depression, positive interaction, ineffective parenting, consistency, punitive style and family dysfunction scores. These scores are the average results of four to seven questions asked of the PMK. The PMK depression score is based on questions of the PMK about her appetite, moods, happiness, concentration, energy, sleep and perceived acceptance by others. Positive interactions include such activities as playing, talking or laughing with the child, or praising or doing something special with them. Ineffective parenting includes such behaviours as expressing annoyance, putting down, disapproving, and angrily punishing the child, as well as overall management problems with the child. Consistency basically 4 3 The variable attempts to measure average annual labour supply but does so imperfectly: current hours of work are multiplied with work weeks in the last 12 months to arrive at average hours of work in the last 12 months; furthermore, the school participation variables are also current variables. 29 covers discipline: does the parent set and follow-up on limits? Punitive style includes such parental behaviours as yelling at and physical punishment of the child. Family functioning is based on questions that assess the degree of cohesiveness, positive communication, and support in the family. It is increasing in dysfunction and is sometimes referred to here as family dysfunction. The interacted parenting skills dummies have been constructed as follows: for the PPVT score, the interacted parenting is the consistency score in most specifications.44 For behavioural scores, the labour supply dummy variables are interacted with the negative of the ineffective parenting score, which has the highest correlation with all but one of the behavioural scores. It is not particularly highly correlated with prosocial behaviour but it still works best as the interacted variable. The positive interaction score has the highest correlation with prosocial behaviour scores. Of the parenting behaviours other than the PMK depression score, ineffective parenting has the highest coefficient of variation, which may explain its strength as an explanatory variable. Table 2.1 shows the means of the labour supply/parenting skills dummy variables for PPVT and behavioural scores. Single parents have generally worse scores on these measures than married parents. Single parents score much worse on the depression score which shows the highest divergence; this is followed by the family dysfunction score, with differences of around 75 percent and 38 percent of a standard deviation between married and single PMKs respectively. Children of workers fare better in terms of the interacted parenting variable for PPVT scores (consistency) than children of non-workers. Children of workers also fare better in terms of the interacted parenting variable for behavioural scores, except for four and five year old children in two-parent families where mothers work part-time. Other variables of interest include family size and birth order, the number of other adults present in the household, reading frequency, family income, child age, child language and immigrant One of the reported fixed-effect specifications uses the negative of family functioning as the interacted variable. Of the parenting and relationship skills scores, consistency and family functioning have the highest correlation with PPVT scores. 30 status, the age of the mother (or father if mother is unavailable) at birth, the child's gender, as well as location variables. Single parent families have fewer children; however, single parent families also have a lower adult/child ratio. Single parents are less likely to read to their children on a daily basis. While the smaller family size should work to improve outcomes, the reading frequency is likely to hurt them. Children of immigrants are more likely to be in two parent families. Immigrant and language status variables are included in the estimation equations as children with these characteristics are not likely to perform as well on official language tests, although some immigrants may come from other English or French speaking countries. Family income dummy variable means show that around 60 percent of the children of single parents lived in a family with an income of less than $20,000 in the prior year.45 Conversely, 22 percent of children in two parent families lived in a family with an income in excess of $80,000. Two and three year old children tend to live in families that have a lower income than four and five year old children. Provinces and urban location dummy variables have been included as explanatory variables because regional differences may contribute to outcomes. Approximately 59-64 percent of the children in the samples live in Ontario and Quebec, while 63-67 percent live in cities with more than 100,000 people. Child-care variables are sorted into three groups. Child-care hours show the mean total hours and the mean total hours for child-care users. Child-care users shows the proportion of children who are in a particular type of care. The average hours for children in a particular type of primary care are shown below the probabilities. Average hours tend to be higher for two and three year olds, who do not attend kindergarden and thus may require longer child-care hours. Average hours are also higher for children of single parents. Children in single parent families who use child-care are most likely to be in a centre or with a non-relative outside the home, while children in two parent families who use 31 child-care are almost twice as likely to be in care with a non-relative outside the home than in a centre. 2.5 ESTIMATION RESULTS AND DISCUSSION Specifications I, II, III and IV are reported for standardized PPVT scores for two parent families and single parents in table 2.2. Specifications II and III modified for the fixed-effect estimator are reported for standardized PPVT in table 2.3. Results of IV estimations for PPVT scores are presented in table 2.4. Specifications U and III are reported for hyperactivity for 4 and 5 year olds using the cross-sectional data and the longitudinal data in table 2.5. Specification II is reported for 4 and 5 year olds using the cross-sectional and the longitudinal data in tables 2.6 and 2.7. Specification II is reported for 2 and 3 year olds using cross-sectional data in table 2.8. (a) PPVT OLS equation Table 2.2 shows OLS estimates for specifications I, II, III and IV on the pooled cycle 1 to 3 observations for standardized PPVT scores for children in two-parent and single-parent families. In general, PPVT tests were administered to four and five year old children although a few children took the test when they were between 42 to 47 months old in cycles 2 and 3.46 Parents in the reference family for labour supply work/study full-time. The results for specification I indicate that parental labour supply and child-care hours have no effect on PPVT scores. Variables that have an impact include parental education, immigration and language status, health of the child, the mother's age at birth, the number of older children, and family income. The impacts are in the expected direction. Specification I does not allow any interactions between labour supply and education. Furthermore, parenting skills and reading time coefficients have been constrained to equal zero. As these variables are added into specification II, the positive parental education and the negative child-4 5 Family income is for the calendar year preceding the survey and has been restated in 1997 dollars. 4 6 Since some of the children can come from the same household, or indeed can even be children who took the test at 3 and then at 5 years old, the estimation procedure specifies that the observations are independent across 32 care hours effects decline. Labour supply variables are not significant in either of these two basic specifications. Because of the labour supply interaction variables, the results shown for specifications D3 and IV in table 2.2 are somewhat difficult to interpret. These specifications test whether education and parenting skills (consistency) effects are dependent on maternal (PMK for parenting skills) labour supply. An interaction dummy variable for maternal and paternal education and for PMK parenting skills is created for each labour supply level. This allows the coefficients on education and parenting skills to vary for each of the labour supply levels. P-values for tests of whether the maternal education and PMK parenting skills coefficients are equal across the different labour supply levels are shown at the bottom of the second page of the table. These results indicate that while the hypothesis that maternal education coefficients are equal across labour supply levels cannot be rejected, the hypothesis that parenting skills coefficients are equal across the labour supply levels is rejected. In particular, the parenting skills coefficient for married PMKs who do not work is greater than the parenting skills coefficient for married PMKs who work full-time. For ease of exposition, from now on I will assume that the mother is the PMK. 4 7 households, but not necessarily independithin households. The procedure affects the estimated standard errors and variance-covariance matrix of the estimators but not the estimated coefficients. 4 7 About 92-94% of PMK's in two-parent families are women. 33 34 i B <u u o o t/3 PM PL, tj Ui 03 •O a es I H S3 o U a © WD cu CZ3 O CN W -J a tu >-a es Ov r\) —- (\ OS (N, 00 ^ OS c\( -1 «_> OS ~_ rf orj VI t \ CN ««, rf VI VO rt ^ rt c> v\) Is. 3 * ro ro od OS ro oo ro od OS t-; ro 00 fN VO in <N 00 CN as ro ro © CN r r OS r r 00 § 1 2 S 5 >-O es a es u S •a S _2 l a "3. go S .2 a cu i_ es CM o OS CN O d f- VO v i r r O rt CO O 00 o SO o OS rt d d CN VI SO rt rt 0\ r r rt rt ^ CO OO OS M wi i n O v© r f VI r f OS p r r p CN r f CO v i oo co O SO Tf rt 1/1 VI rt cN VI d d CN v i co O CN rt o d d oo m » O CN VI d ^ d CN VI CO O CN rt o d d oo CN d d r--OS VI ro oq r r o d VI CN CO r r rt d 00 sq oo OS co o o r r p SO co O ro p 00 v> r f oo oo o ro oo VO od *—1 1 d d CN i r f ro d CO ro CO o i d VI r r r r co OS r f OS CN O 00 r f OS so r f ro t-~ O co r f VI co SO CN o r ~ oo • d • d • • -1 d d d d d d d r r oq CN SO CN CN o CO OS t~-CN vo so r f r f O oo VI co so CN O CN CN ro CN od 1 d d • r f ~* ro d ro ro co CN d vo oq os CN CO VO VI OS OS r f r - r f ro VI O VI r f SO ro v£> CN r -o v i 00 VO i d d i 1 d d d d d d d d CO oo CN co VI co oo CN o VO as sq 00 VI OS d d • d 1 d r r ro CN o oq SO CN OS CO VI r f O o v i CN CO 00 d d • 1 d d d © r f CO vo CN r f co p CN d d d O rf Ov SO rf CN rf CN d o — ! © so co v i rf rf c cu I N es a a C3 cu 3 a ••c a o C J ^ - ^ CN c4 5 5 VO i n OS CO 00 CN VO 00 i n © OS © r-i n vq i n CN OS © SO r -i n r r CN rT © CN r r r r r r 00 r t © © i d • i r f © © CN CN i d • rt © CN i d d © i d • d • d • © • d 1 © i n o CO oo vq r - p- VO CN VO vq CO 00 oq m i n oo © c o © © rT CO c o © oo © CO CO CN OS i n d i d i CO 1 CN —' d CN i © rt © © i d • © © i 1 d i 1 © 1 CN 1 i n i n 00 >n CN r- CN m co 00 CN i n 00 p~ r r VO 00 © CN vq r r CN r -© i n © rT CN © r r rT CN i n co VO oo SO d • d • i rT 1 © d CN CN • © • ^ d CN 1 © i © • © © i © 1 d d 1 d d i n o oo t~ CO CO i n vo CN r r vq CO i n oq CO i n oo o © © rT SO © VO r r OS r r CN VO oq d • d co i CN 1—1 —' d CN © i —' © © • © 1 d • d • d d d CN 1 r f c o r r i n OS OS CN VO CN CN oo vo vq CO 00 OS OS i n CN OS r r VO CN VO r r © r r SO r ~ m OS i n VO d • © '7 r r 1 © © CN CN 1 © i © 1—t 1 d © © © 1 d d • © © © >/-> o 00 m i n m OS i n CO OS p~ CN VO 00 CN 00 O O CO SO oo © © VO rT oo © CN OS oq CN r r d • © i co • CN © © CN 1 © 1 —1 © © 1 © • d • © 1 © 1 © i © 1 CO • co c o CN OS VO i n i n OS SO OS VO © r -r r Ov oq 00 © rT co i n CO CN VO oo c o oo i n OS d i d i r f i © CN CN • © 1 © • d • © • © i © i © • © d oo O CN VO 00 i n © CN VO 00 r r VO CN OS vq i n 00 © © CO r r CN VO © 00 CN oq r r d • d • r f i co © — , d CN i © 1 © i 1 d • i © i CN • CO r f r -so co co r-; r - p-CO VO i n r r c o c o v q c o i n i n © co 00 OS © vq m CN Os vq 00 c o c o vq O O r r VO oo rT c o oo CN d i 1 co 1 vb i r f i d CN i CN • i n r f 1 CN i CN 1 CO 1 CO • CN i CO • 1 © © i r r o vo r r vq co oo r r r r CO i n SO © © i n so co SO © r r Ov © © CN r r i n oq - i n r r vo CN m VO 00 CN d • © 1 od • 00 1 vo i d © © i i CN i n d • © 1 © 1 © i n • co • CN i CO • 1 © • © 1 VO VO 00 CO r r i n co i n v q oo c o OS © i n 00 O N © OS vq CN OS VO CO vq © i n r - m CO m r r m d • i co i VO 1 r f i © CN • CN i i n r f i CN i d c o • co i CN • c o 1 © i © r r o vo VO CN CN CN © r r r r CO vo VO t~ OS os t-; VO co VO © r r CN © © r r r r oq CO VO rT i n s o rT © i n d • d 00 1 od • vo • © © © 1 © CN i n © i d • © i i n 1 CO CN i CO 1 • © © vo vo vo CN 00 vq OS 00 CN c o OS r r co OS i n r r r r © r-~ oq 00 OS oo OS OS 00 i n r ~ VO i n CN r r SO CO co d • i co VO r f d CN i CN i i n r f d 1 CN i © 1 CO i c o • CN CO 1 © 1 © r r © CN VO VO t-; p-CN CO r -OS CO CO co vo Os OS oq 00 co m © rT CN © co r r c> 00 CN rT rT CN OS i n c o d • © i 00 1 od • vd • © © © i © 1 CN i n d • © 1 © I i n • r f i CN i CO i i © i © i/-> 00 VO s q r -© CN co Os co i n p~ r r VO CO r r °) CN r r oq r» CN s q O O r t CN CN CO r r vo d i r f 1 c o i d CN 1 CN i 1 CO 1 CO • CN i CO • © 1 © co r r i n CN r r - H m CN © © © CO CN 00 r r rT © rt o rt © CN CN © rT o 1 © v d • r f • CN i CO t 1 d • d Pi © © © vo t— oo 6<j m a eu u CS a S3 )--a <u S a X! a o w rj CN 5 oo N O ro N O oo ro o ro CN O N ro O N CN o oo oo CN O N —1 CN -7 d d • —' d d d d o O N oo oo N O m ro O N oo m p i n N O o i n CN i n CN CO T T ' CN 1 CN —1 d ro d d • CN N O W> o T T CN co f -O N CO O N CN co CN O O N i n O N —1 rt CN • rt d d • —1 d d d d o o O O O N N O oo m p O N i n - t-m CN O ro o ro T T CN 1 CN CN d ro d d ~* rf r - CN i n - N O N O CN 00 O N o CN O N O i n CN CN O oo 00 CN O N —1 CN 1 rt d d • —' d d d d o O N O N 00 O N V) O N oq CN i n p ro T T o CN i n CN ro CO T T CN i CN CN d i ro d d O fN o o CN ro CN p r- oo o ro CN O N C N CN CN 1 i—1 d d CN d d d O O o CN t~- CN ro O N ro CN TT oq o co m N O oo CN i n i n I T ) CN CN d co d rt fN CN o O N C O CN N O O O N p oo CN O N i n CN CN CN ro CN —' o 1 —1 I CN CN d CN i d d i CN O i n oo O N CN V O O N O N O CN N O T T 00 o CN —1 o 1 i CN —' CN o i d d • N O ro m o o O N i n N O O N oq o p ro CN N O CN p CN i n CN d CN CN d CN i d d r-co N O o O O oo ro N O CN O CN O N CN *—< oo p ro rt d 1 i CN 1 T—1 CN d • 1 d d • i n O ro CN O N O N t-; ro CN O N CN O N O CO CN N O O N oo o r -CN —1 d • —1 I CN d CN d 1 I d d • O CN O O oo p m O N O CN O O N O CN r -ro i n o d i — 1 I CN d rt CN d • • d d • O O N CO CN O CO t-; co O N p O N O N N O o i n p CN i n CN —1 d CN 1 ~* — 1 —• CN d CN i d d 00 00 m co O O N CO O N O N r -CN CN ro CN i n 00 o CO d CN 1 d CN d • d d I O N C N C N C N i n o d d O N N O O O C N N O i n o d d i T J " C O m C N o d d ro oo oo ro ^ f o d d 73 s a" d o I o CO S P S3 ^ O H II <D ^ s i O H T 3 -a t-H I « .5 ^ CO ^ 5 N O ro o d d —i N O i n r - C N ro N O o d d 00 .—i O o d d O N C N — i — i o o d d C H % | S I s S3 1 i O H J3 t-. I II I S s CD s •5 II S o I a I s co According to specification III, for children in two parent families with a father who works full-time and identical other characteristics, the difference between the predicted score for a child whose mother is not working and one whose mother is working and using 40 hours of child-care is: 4 8 (2.8) -6.17 + (.99 - .77) * mother's years of education + (.36 - .07) * PMK parenting skills - 40*(-.02) = -5.37 + .22*mother's years of education + .29*PMK parenting skills Given average education levels and parenting skills this difference is very small, but is sensitive to parenting skills. 4 9 The same exercise can be repeated for a child with a mother who does not work and a child with a mother who works but does not use child-care. The difference is: (2.9) -6.17 + .22*mother's years of education + .29*PMK parenting skills This is almost identical to (2.8) because the child-care effect is small and (statistically) insignificant. At average parenting skills and education for working mothers, (2.8) equals 1.83 and (2.9) equals 1.03, which are rather negligible. To get a better idea of whether and how much education and parenting skills matter, I have constructed a series of graphs that plot predicted scores against parental education for three parenting skills levels. These are presented in figures 2.1 - 2.3. The base case is for a healthy four and a half year old girl who lives in a metropolitan B.C. area. She speaks English and is not an immigrant. She has one older sibling and was born when her mother was 30 years old. The family's income category is "$80,000 +". Other characteristics are average unless otherwise specified With the different labour supplies there is likely to also be an income effect, but assuming incomes of more than $80,000 for both groups reduces this effect to zero. Children in that income group do not perform significantly differently than children in the $50,000 - $59,999 income group. 4 9 There is also sensitivity to education, although the p-value testing whether the two education coefficients are equal is 34.5 percent. 38 FIGURE 2.1 PPVT Scores, Labour Supply and Education: High Parenting Skills 8 9 10 11 12 13 14 15 16 17 18 Education Figure 2.1 shows predicted PPVT scores for such children whose parents have top parenting skills - a consistency score of 20. This is the top score and 6.8 percent of two parent observations have this score. Figure 2.2 uses the median consistency score of 15, and Figure 2.3 uses a consistency score of 9 - 3.5 percent of two parent observations have a score lower than 9. Figure 2.1 to 2.3 show that for parents with average education and consistency maternal labour supply does not matter very much. Children of highly educated and consistent mothers do somewhat better when their mothers do not work, while children of poorly educated and inconsistent mothers do slightly worse when their mothers do not work. At education of 18 and consistency of 20, equation (4) equals 4.3, and at education of 8 and consistency of 9, it equals —1.1. 39 FIGURE 2.2 PPVT Scores, Labour Supply and Education: Average Parenting Skills H > CL. EL. •a •a 15 I 10 105 100 95 90 H 85 ^ ^ ^ ^ ^ ^ ^ ^ - • - Mother (PMK) not at work -*- Mother (PMK) wft - no daycare ! -* - Mother (PMK) wft - 40 hrs care 110.9 108.1 10 11 12 13 14 Education 15 16 17 Specification IV allows the effect of hours in care to vary with the family income level for children in two-parent families. Results indicate that in terms of fostering children's cognitive abilities, quality of care is increasing in income for these families. The break-even income level where hours in care have no effect occurs at around $94,000. This effect would not necessarily be expected for single parents, since many low-income single parents qualify for child-care subsidies which allows them to purchase higher quality care or requires that they purchase licensed care services. 40 FIGURE 2.3 115 T PPVT Scores, Labour Supply and Education: Low Parenting Skills Education Table 2.2 also indicates that children of older mothers score slightly better, and that children who are higher in the birth order and children in larger families tend to score worse. In single parent families, the family size effect is quite important, while the birth order coefficient is insignificant; the parental age coefficient is also more than twice as large as the two parent families parental age coefficient. Children in two-parent families who read or are read to daily score around 19 percent of a standard deviation more than children who do not, and those who read or are read to several times daily score around 36 percent of a standard deviation more. Although the PMK depression score does not significantly affect children's score in two parent families, it does in single parent families. Family dysfunction has also an important negative effect on children's score, although the coefficients are not significant in single parent families. 41 Specifications III and IV also show an important labour supply interaction effect with fathers' education. In particular, the education effect of fathers who work part-time is significant and large. In single parent families, parental education has a strong positive effect, but the consistency score does not. Children in two-parent families with income of $50,000 - $59,999 have the highest scores. At low paternal income levels, maternal labour supply can have a positive impact on children via the income effect. For children in single-parent families, family income does not have a significant impact on PPVT scores, although scores are higher at high-income levels. Table 2.2 and figures 2.1 - 2.3 indicate that for the average family, labour supply and child-care have little or no effect on children's cognitive scores, that children of highly skilled mothers tend to do somewhat better if their mothers do not work, that children of poorly skilled mothers tend to do slightly better if their mothers work, and that higher income is associated with higher quality care in two-parent families. (b) PPVT Fixed Effects Equation I estimate a family fixed effect equation for PPVT scores. If an unobserved variable that affects child outcomes is correlated with income, labour supply, and/or possibly also child-care use, the OLS coefficient estimated will be biased. A family fixed effect model assumes that the unobserved heterogeneity is family-based, as opposed to child-based. For example, if stay-at-home mothers are better at child rearing than working mothers and mothers' child-rearing skills are unobserved, then children in families with a stay-at-home mother would have better scores, not because of the mother's decision to stay home, but because of her relative skills level. However, the inclusion of parenting skills variables in the OLS equation likely adequately addresses this concern and missing measures of other types of ability may be the concern. Alternatively, if mothers who work outside the home are more driven to success and pay greater attention to their children's literacy and other cognitive achievements, or have greater cognitive ability than mothers who stay home, then their children might be expected to fare better in terms of cognitive achievements for reasons other 42 than their labour supply. But if these differences are unobserved, the labour supply variable in an OLS regression would also capture these effects and either overestimate the positive effects of mothers' labour supply or underestimate its negative effects. While education can capture some of these differences, education is not a perfect proxy for cognitive ability or the desire for one's children to succeed. Table 2.3 shows the result of family fixed-effect estimations. As shown in equation (2.3) family variables disappear in the differencing; these include parental education, immigrant status, and location. Since the family fixed effect model exploits siblings' differences, the number of available observations left for the estimation is substantially reduced. With such a small number of observations, coefficients on the explanatory variables are generally not significant. Specification II does not interact parenting skills with labour supply. In specification III, I use either consistency or family functioning50 as an interaction variable with labour supply. In the fixed effect model, family dysfunction and mother's age at birth are the only significant effects. In fact, the coefficient on family dysfunction is more than 2.5 times larger here than in Table 2.2. Hence a decrease in family functioning within a family is associated with poorer cognitive outcomes for the children. Family dysfunction scores range between zero and thirty-six with a mean of 7.6 and a standard deviation of 5.1. An increase in family dysfunction of one standard deviation would result in a decrease in the PPVT score of around 1.8 to 1.9. In the fixed effect model, family functioning is more significant as an explanatory variable than consistency. However, the effect of consistency varies more with labour supply than the effect of family functioning. 43 TABLE 2.3 Fixed Effects Regression Coefficients - Standard PPVT Scores (t-ratio in parenthesis) II III Consistency Interacted Family Functioning Interacted Number of siblings pairs* Adjusted R squared .5704 1,688 .5718 .5680 Constant 114.78 9.84 114.71 9.77 114.35 10.03 Parental Employment and Parenting Skills and Parenting Skills (Mother works full-time) Mother did not work 1.11 0.38 5.32 0.92 3.14 0.83 Mother worked part-time -0.14 -0.05 -2.44 -0.53 0.54 0.14 Parenting skills P M K did not work -0.19 -0.50 0.47 1.61 P M K worked part-time 0.25 0.70 0.26 0.92 P M K worked full-time 0.09 0.27 0.18 0.61 Child and Family Characteristics Child age in months 0.00 0.02 0.00 0.01 -0.01 -0.08 Boy -0.38 -0.26 -0.31 -0.22 -0.32 -0.22 Speaks no English or French -0.35 -0.05 -0.74 -0.11 -1.01 -0.14 Poor health -1.32 -0.33 -1.33 -0.33 -1.04 -0.25 In kindergarden 1.73 0.85 1.72 0.84 1.81 0.90 Mother (parent) age at birth -0.39 -1.78 -0.40 -1.72 -0.39 -1.84 Number of children < 17 -1.07 -0.41 -1.10 -0.43 -1.19 -0.47 Number of older children -0.65 -0.51 -0.69 -0.54 -0.70 -0.55 Reads or is read to daily 0.64 0.32 0.64 0.33 0.71 0.35 " " several times a day 3.24 0.99 3.03 0.94 3.28 0.99 P M K depression score 0.23 1.15 0.24 1.17 0.21 1.03 Consistency score 0.06 0.21 0.06 0.21 Family dysfunction score -0.36 -1.74 -0.37 -1.76 Family Income Less than $20,000 -4.42 -0.87 -4.66 -0.92 -4.74 -0.95 $20,000 - $29,999 -2.82 -0.72 -2.87 -0.72 -3.33 -0.84 $30,000 - $39,999 -3.41 -0.99 -3.43 -0.98 -3.67 -1.07 $40,000 - $49,999 -0.23 -0.09 -0.27 -0.10 -0.33 -0.12 $60,000 - $69,999 -0.75 -0.27 -0.78 -0.28 -0.82 -0.28 $70,000 - $79,999 0.61 0.20 0.67 0.21 0.66 0.21 $80,000 + 1.28 0.40 1.41 0.44 1.24 0.38 Child Care Hours Total hours -0.05 -0.78 -0.05 -0.75 -0.05 -0.79 *Only children in two-parent families are included here In the fixed-effect model, the mother's age at birth variable is likely picking up a birth order effect here: if the mother is older for one child versus the other, then the first child 44 was born after the second. This is in contrast with the OLS result where maternal age has a positive effect on the PPVT scores. The signs of the coefficients on the child-care variables are negative as in the OLS estimation in specification I, but not significant. The magnitudes are more than twice as large. The signs and magnitudes of the income variable coefficients are of the same order and magnitude as in Table 2, for all three specifications. Father's labour supply and parental education are not included in this specification as there would be too little variation in these variables to include them.51 (c) IV estimator for PPVT scores Consider the following two-equation system: (2.10) y, = yy2 + pYX, + s, (2.11) y 2 = cpy, + pVX 2 + s2 Assume a bivariate normal distribution with zero means. Also assume that y 2 is censored, that is, y 2 = max(0,y2*). In this context, yi represents a child outcome while y 2 represents maternal hours of work.52 Here, the assumption that maternal labour supply is an exogenous variable is relaxed. This model is due to Nelson and Olsen (1978) and is described in Maddala (1983) chapter 8.8 and in Greene (1998) section 28.4. The estimation procedure is as follows: (1) Estimate a reduced-form equation for (PPVT scores) and obtain its variance, CT2!. (2) Estimate a reduced-form tobit equation for y2 (maternal hours of work) and obtain predicted hours and variance-covariance matrix VO and a 2 2. (3) To obtain the correlation between the two equations (pi2), I follow Greene (1998) section 28.4: first run a reduced-form probit on maternal labour supply; retrieve the inverse Mills Parental labour supply can vary for different siblings i f siblings are from different survey years but paternal labour supply is much more constant over years than maternal labour supply. 5 2 Equations (2.10) and (2.11) correspond to equations (2.4) and (2.5) respectively. 45 ratio, append to a reduced-form PPVT score equation and estimate with OLS; the coefficient on the inverse Mills ratio is equal to p^oi. (4) Estimate a structural equation for PPVT scores using OLS by substituting the predicted value for maternal hours of work for the actual value. (5) Calculate the variance-covariance matrix for the structural equation, which equals: V , v = (cr2, - 2ypi2a1CT2)*[Z,Z] + y 2[Z ,Z]*M*[Z ,Z]; where, M = [X'Z]'*V0*[X'Z] and, X = matrix of exogenous variables for the system and, Z = matrix of explanatory variables for PPVT with maternal hours of work replaced by predicted maternal hours of work. Since maternal income and child care hours depend on maternal labour supply, these variables will be endogenous if maternal labour supply is endogenous. In this section, maternal income and child-care use are not included as explanatory variables and maternal labour supply is tested for endogeneity. The measured effect of maternal labour supply thus incorporate related income and child-care effects.53 The instrument for this procedure is the unemployment rate by province and year of survey: observations originate from three separate years (1994, 1996, 1998) and unemployment rates for the prior years are used. While maternal labour supply is expected to vary with current local economic conditions, there is no theoretical reason to believe that PPVT scores would be directly affected.54 The PPVT equation coefficients for parts 1, 2 and 4 above, as well as the corresponding OLS coefficients, are presented in Table 2.4. The standard errors calculated in part 5 are so large that there There are not enough instruments to allow for two additional endogenous variables. 5 4 The unemployment rate varies by region, but dummy province variables are included to capture regional differences. 46 are no significant t-statistics for the IV estimator.55 The Hausman test based on part 5 results in accepting the null hypothesis that maternal hours of work are exogenous.561 have also included the results for an additional OLS regression of the PPVT equation, which adds predicted maternal hours of work to Xj. Kennedy (1992) suggests using this omitted variables (OV) variation of the Hausman test when the number of endogenous regressors is small relative to the number of exogenous regressors. In this OV variation of the Hausman test, the variable "mother annual hours of work" is considered endogenous if the coefficient on "mother predicted annual hours of work" is significant. The rationale behind the test is as follows. Without the instrument the regression would produce residuals 8 . If the instrument is correlated with 8 then the coefficient on the instrument will not be equal to zero. Using the OV version of the Hausman test, the null hypothesis of exogeneity of maternal labour supply is rejected. The tobit coefficients and t-statistics for the hours of work equation are shown in the first two columns of table 2.4. These indicate that average maternal hours of work decrease with increases in the unemployment rate. At the same time, the reduced-form PPVT coefficients shown in the following column indicate that PPVT scores for children increase with the unemployment rate. These relationships explain the TV estimator's negative coefficient on maternal hours. Using the IV coefficient on maternal labour supply shown in Table 2.4, the predicted PPVT score when the mother works 37.5 hours per week is more than 10 points lower than when the mother does not work. This is two-thirds of a standard deviation. The IV estimator coefficients on child and family characteristics are similar to the OLS estimator coefficients. However the IV estimator low-income effects are not as strong, the maternal education coefficient is markedly higher, and children in eastern provinces have worse results. Since PPVT scores are positively correlated with the unemployment rate, controlling for the unemployment rate through maternal labour supply in the IV The V matrix shown in part 5 is very large. The test statistics (»0) equals (a lv-aOLS)'*[V lv-VOLS]"1*(aIV-aOLS), and is distributed chi-square(37). 47 estimator results in lower coefficients for provinces with higher unemployment rates relative to their OLS estimator coefficients. With only one instrument, it is hard to draw strong conclusions about the possible endogeneity of maternal labour supply. There could be an explanation other than maternal labour supply (and also other than what has been controlled for here) for the positive relationship between unemployment rates and PPVT scores. However, given the results presented in this section, it is important that the question of the potential endogenity of maternal labour supply be kept in mind in any future research. 48 T A B L E 2.4 IV Estimator/ Hausman Test (OV) - Standard PPVT Scores Maternal Hours PPVT Score N= 11,214 Tobit Reduced-Form OLS IV OVTest Constant 1.99 0.48 75.69 15.36 80.17 18.58 76.25 76.06 16.13 Unemployment rate -1.50 -6.86 0.43 1.80 Parental Employment and Education Mother pred. weekly hrs o f work -0.28 -0.30 -1.89 Mother weekly hrs o f work 0.03 1.51 0.03 1.58 Father weekly hrs o f work 0.06 3.64 0.02 0.84 0.02 0.76 0.04 0.04 1.56 Mother yrs o f education 2.29 17.64 0.81 4.90 0.76 4.45 1.46 1.45 3.69 Father yrs o f education -0.36 -3.08 0.47 3.46 0.48 3.57 0.37 0.38 2.69 Father's Income Less than $20,000 3.03 3.18 -3.02 -2.86 -3.09 -2.93 -2.16 -2.17 -1.82 $20,000 - $29,999 4.86 5.57 -2.25 -2.23 -2.31 -2.30 -0.87 -0.90 -0.74 $30,000 - $39,999 2.99 3.60 -1.64 -1.71 -1.72 -1.80 -0.79 -0.80 -0.74 $40,000 - $49,999 3.84 4.55 -0.17 -0.18 -0.25 -0.26 0.92 0.89 0.78 $60,000 - $69,999 1.14 1.10 -0.98 -0.76 -1.03 -0.80 -0.66 -0.65 -0.51 $70,000 - $79,999 -5.08 -3.74 1.31 1.07 1.44 1.18 -0.13 -0.10 -0.07 $80,000 + -7.67 -6.98 0.99 0.69 1.06 0.73 -1.19 -1.13 -0.59 Child and Family Characteristics C h i l d age in months 0.02 0.40 -0.05 -0.89 -0.05 -0.92 -0.05 -0.05 -0.81 B o y 0.49 1.07 -0.53 -1.07 -0.54 -1.08 -0.40 -0.40 -0.79 Speaks no Eng. or Fr. -5.97 -3.82 -8.70 -3.52 -8.74 -3.59 -10.39 -10.36 -4.13 Immigrant parents -5.70 -6.99 -8.46 -6.72 -8.43 -6.75 -10.08 -10.05 -6.93 Poor health -2.59 -1.75 -6.58 -4.23 -6.46 -4.13 -7.32 -7.30 -4.53 In kindergarden 0.04 0.06 0.46 0.56 0.45 0.55 0.47 0.47 0.58 Mother's age 0.21 3.78 0.23 3.23 0.23 3.13 0.29 0.29 3.88 Number o f children < 17 -3.70 -13.65 -1.74 -4.95 -1.68 -4.67 -2.79 -2.77 -4.15 Number same or younger -4.77 -11.29 1.21 2.74 1.28 2.94 -0.15 -0.12 -0.13 Reads or is read to daily -2.27 -4.19 2.76 5.03 2.77 5.07 2.12 2.14 3.37 " " several times a day -4.30 -4.73 5.17 4.75 5.25 4.77 3.95 3.99 2.95 P M K depression score -0.31 -5.69 -0.06 -0.99 -0.05 -0.87 -0.14 -0.14 -1.86 Family dysfunction 0.10 2.10 -0.15 -2.83 -0.15 -2.94 -0.12 -0.12 -2.15 Consistency score 0.06 0.74 0.21 2.47 0.20 2.44 0.22 0.22 2.66 Location Newfoundland 8.27 2.94 -3.06 -1.20 1.14 1.12 -0.71 -0.67 -0.47 Prince Edward Island 10.29 2.91 -2.96 -1.52 -0.12 -0.11 -0.04 -0.00 -0.00 Nova Scotia -0.07 -0.04 0.19 0.14 1.75 1.74 0.17 0.20 0.15 New-Brunswick 0.38 0.23 -2.69 -2.61 -1.48 -1.84 -2.58 -2.55 -2.57 Quebec 0.33 0.38 0.82 0.83 2.01 2.56 0.92 0.94 0.98 Manitoba -1.86 -1.41 1.54 1.41 0.85 0.84 1.02 1.04 1.02 Saskatchewan -0.33 -0.24 2.23 2.18 1.15 1.35 2.14 2.13 2.15 Alberta -4.61 -4.99 2.69 2.92 2.08 2.48 1.38 1.41 1.56 Br i t i sh Columbia -2.43 -3.02 -0.23 -0.21 -0.42 -0.38 -0.92 -0.90 -0.79 Urban - 500,000 + 4.02 5.54 -1.23 -1.74 -1.35 -1.92 -0.09 -0.12 -0.13 Urban - 100 ,000-499 ,000 3.40 4.17 0.16 0.25 0.02 0.03 1.13 1.10 1.28 Urban < 100,000 3.00 3.90 0.03 0.05 -0.13 -0.25 0.88 0.86 1.17 Rho 0.68 R-square 0.1638 0.1638 0.1638 0.1647 49 (d) Hyperactivity scores for four and five year olds Table 2.5 shows OLS estimates for specifications II and III for hyperactivity scores of 4 and 5 year old children in two parent families.57 A higher score indicates that the child is more hyperactive. The dependent variable has been transformed so that it is equal to the percentage difference between the child's score and the mean score. Hence, coefficients on explanatory variables represent a percentage increase or decrease from the mean score. Mean current scores in the cross-sectional sample are 4.67 with a standard deviation of 3.3. For the longitudinal sample, the mean is 4.81 with a standard deviation of 3.29, and the mean past score is 4.24 with a standard deviation of 2.9. Current scores range from zero to sixteen, and prior scores range from zero to fourteen. Specifications II and III are each estimated for the pooled cross-sectional data (X), and the longitudinal58 data ((L-l) and (L-2)). L - l is estimated to determine whether variations in results between X and L - l are due to changes in the sample or to changes in the equation. Results for single parent families have very few significant coefficients. 5 8 The only longitudinal feature of this data is that it contains behavioural scores from two years before; the data also has cross-sectional features in that it contains records from cycle 2 with cycle 1 behavioural score information and records from cycle 3 with cycle 2 behavioural score information. 50 T A B L E 2.5 OLS Regression Coefficients - Hyperactivity Score, 4-5 Years Old (t-ratios in parenthesis) II m X L - l L-2 X L - l L -2 N 12,731 4,579 4,579 12,731 4,579 4,579 R - Squared 0.24 0.26 0.35 0.24 0.27 0.36 Mean 4.67 3.31 4.81 3.29 4.81 3.29 4.67 3.31 4.81 3.29 4.81 3.29 Constant 15.63 -0.99 -11.85 -0.53 -44.71 -2.14 -24.20 -1.41 -26.34 -1.06 -63.04 -2.72 Pr ior score 7.41 15.36 7.40 15.49 Parental Employment, Education and Parenting Skills (Both parents work full-time) Mother did not work -2.35 -0.77 1.32 0.28 0.70 0.16 0.63 0.05 24.27 1.03 33.94 1.46 Mother worked part-time -2.33 -0.89 -2.98 -0.85 -2.88 -0.86 18.61 1.22 31.52 1.54 31.07 1.59 Father did not work 10.29 2.21 12.57 1.54 10.86 1.40 16.14 1.40 34.80 1.43 37.67 1.65 Father worked part-time -0.57 -0.15 -4.64 -0.84 -5.47 -1.00 -7.63 -0.48 -32.07 -1.40 -24.33 -1.13 Mother yrs of education -1.67 -2.94 -1.47 -1.74 -0.96 -1.21 Did not work -1.07 -1.35 -1.07 -0.75 -1.23 -0.90 Worked part-time -2.44 -2.74 -2.03 -1.80 -1.32 -1.32 Worked full-time -1.45 -1.83 -0.76 -0.67 -0.10 -0.09 Father years of education -0.60 -1.22 -0.79 -1.08 -0.62 -0.87 Did not work -1.05 -1.05 -3.29 -1.73 -3.38 -1.89 Worked part-time 0.01 0.01 1.25 0.74 0.89 0.55 Worked full-time -0.68 -1.27 -1.24 -1.63 -0.91 -1.25 Interacted parenting skills 6.59 16.52 6.46 10.80 5.35 8.71 Did not work -6.24 -11.14 -5.52 -5.69 -4.41 -4.74 Worked part-time -6.18 -9.78 -5.57 -6.20 -4.46 -4.71 Worked full-time -7.15 -15.33 -7.68 -11.62 -6.59 -9.88 Chi ld and Family Characteristics Five years old -6.23 -2.28 -2.65 -0.64 -2.57 -0.66 -5.96 -2.21 -2.13 -0.54 -2.14 -0.58 Boy 16.12 7.82 17.09 5.88 14.74 5.37 16.22 7.93 17.53 6.16 15.17 5.68 Immigrant parents -6.81 -1.74 4.35 0.81 2.56 0.52 -6.83 -1.75 5.01 0.96 3.03 0.64 Poor health 25.74 3.66 29.48 2.39 26.11 2.24 25.54 3.61 29.25 2.36 26.06 2.21 In kindergarden 3.92 1.34 4.73 1.17 4.99 1.32 3.85 1.32 4.38 1.11 4.65 1.26 Mother/parent age at birth -0.5.7 -3.42 -0.91 -3.57 -0.62 -2.50 -0.55 -3.34 -0.91 -3.65 -0.62 -2.53 Number of children < 17 -5.24 -2.91 -6.42 -2.31 -5.62 -2.00 -5.23 -2.93 -6.27 -2.30 -5.51 -2.00 Number of older children 3.55 1.99 6.92 2.39 6.08 2.12 3.46 1.94 6.78 2.35 5.96 2.09 Reads or is read to daily -6.16 -2.50 -5.44 -1.49 -2.80 -0.84 -6.13 -2.51 -5.42 -1.54 -2.79 -0.87 " " several times a day -4.99 -1.25 -9.84 -1.72 -7.17 -1.24 -4.50 -1.14 -8.27 -1.49 -5.73 -1.02 P M K depression score 1.29 5.21 0.94 2.59 0.82 2.30 1.29 5.25 0.98 2.69 0.87 2.41 Punitive score 2.36 3.83 2.75 2.88 2.16 2.41 2.41 3.92 2.77 2.97 2.18 2.50 Consistency score -1.62 -4.53 -1.66 -3.17 -1.31 -2.63 -1.60 -4.54 -1.71 -3.49 -1.36 -2.89 Postive interaction score 0.59 1.44 0.64 1.01 0.40 0.68 0.60 1.47 0.60 0.96 0.38 0.66 Family dysfunction score -0.14 -0.67 -0.26 -0.86 -0.55 -1.89 -0.15 -0.72 -0.27 -0.90 -0.56 -1.95 continued 51 TABLE 2.5 (continued) (t-ratios in parenthesis) II III X L - l L-2 X L - l L-2 Family Income Less than $20,000 2.40 0.44 0.54 0.07 -1.50 -0.20 2.51 0.47 1.64 0.21 -0.72 -0.10 $20,000 - $29,999 -4.57 -1.05 -4.37 -0.72 -3.98 -0.65 -4.64 -1.06 -5.87 -0.95 -5.61 -0.91 $30,000 - $39,999 -0.81 -0.19 3.38 0.53 0.64 0.10 -0.89 -0.21 2.48 0.39 -0.12 -0.02 $40,000 - $49,999 -2.27 -0.57 -4.06 -0.66 -2.83 -0.46 -2.20 -0.55 -3.77 -0.62 -2.49 -0.41 $60,000 - $69,999 2.92 0.74 2.15 0.38 2.30 0.41 2.70 0.69 2.12 0.37 2.26 0.41 $70,000 - $79,999 -6.53 -1.47 -5.63 -0.88 -11.41 -1.82 -6.56 -1.48 -5.38 -0.85 -11.19 -1.80 $80,000 + -1.86 -0.49 -4.34 -0.74 -3.86 -0.67 -1.64 -0.43 -3.92 -0.68 -3.75 -0.66 Child Care Hours Total hours 0.17 2.12 0.26 2.47 0.21 2.06 0.16 2.10 0.24 2.43 0.19 2.00 Location Newfoundland -7.45 -1.67 -6.59 -0.97 -6.46 -1.03 -7.59 -1.70 -6.80 -0.99 -7.25 -1.14 Prince Edward Island 2.12 0.40 -2.43 -0.31 -1.61 -0.22 2.02 0.38 -1.83 -0.23 -1.31 -0.18 Nova Scotia 0.19 0.05 -0.27 -0.04 -1.47 -0.27 0.34 0.08 -0.25 -0.04 -1.70 -0.31 New-Brunswick 2.09 0.59 -2.24 -0.38 1.11 0.20 1.96 0.56 -2.33 -0.39 0.85 0.16 Quebec 6.85 1.91 5.46 1.08 3.78 0.81 7.19 2.02 6.22 1.25 4.31 0.94 Manitoba -0.57 -0.14 -2.89 -0.49 -5.76 -1.07 -0.70 -0.17 -3.00 -0.51 -6.09 -1.13 Saskatchewan -3.70 -1.10 -4.51 -0.90 -6.48 -1.39 -3.47 -1.04 -4.27 -0.86 -6.25 -1.35 Alberta -1.71 -0.52 -7.52 -1.56 -9.97 -2.18 -1.28 -0.40 -6.59 -1.40 -9.15 -2.05 British Columbia 4.39 1.18 0.62 0.11 -0.17 -0.03 4.58 1.24 1.05 0.19 0.26 0.05 Urban - 500,000 + -4.84 -1.64 -1.26 -0.28 1.39 0.33 -4.67 -1.58 -0.69 -0.16 1.99 0.47 Urban - 100,000-499,001 0.06 0.02 1.21 0.30 0.61 0.17 0.27 0.10 1.87 0.46. 1.15 0.31 Urban < 100,000 0.47 0.19 2.87 0.76 1.61 0.46 0.68 0.27 3.22 0.84 1.92 0.54 P-Values for tests of equality between: Mother/single parent years of education Did not work = worked part-time 0.19 0.57 0.95 Worked part-time = worked full-time 0.36 0.34 0.33 Did not work = worked full-time 0.65 0.85 0.49 Parenting skills Did not work = worked part-time 0.78 0.97 0.97 Worked part-time = worked full-time 0.10 0.01 0.01 Did not work = worked full-time 0.04 0.04 0.03 |*Only children in two-parent families are included here Using past behaviour as an explanatory variable considerably increases the explanatory power of the model. Results indicate that prior scores, the ineffective parenting score, the gender of the child and his or her health, other family functioning variables including whether the child reads or is read to daily, family size and birth order all have significant impacts on hyperacticity scores. Boys scores significantly higher than girls. Children in larger families are less hyperactive and younger children are more hyperactive. Punitive parenting behaviour and PMK depression increase the scores,59 while reading time and consistency reduce the score. Parents who punish may be parents who perceive more infractions; the child may respond adversely to punishing behaviour, thus reinforcing the cycle; alternatively, parents who punish may face more infractions; 52 Results indicate that interacted parenting skills (the negative of the ineffective parenting score) have a larger impact when the mother works full-time, which is contrary to expectations. P-values for tests of equality between the interacted labour supply, education and parenting skills variables are shown at the bottom of page 2 of the table. I expected the effects for non-working mothers to be stronger because they spend more time with their children. However, it may be that effective parenting skills are more important for working mothers for this score because the disruption and stress of having a mother who leaves for and returns daily from work is compensated by more effective parenting skills. The maternal education interacted variables coefficients have the expected sign in the cross-sectional estimations but they are not always significant. For mothers working full-time, the sign on the interacted education variable coefficient changes from negative to almost zero once the lagged score is included in the estimation, although looking at X , L - l , and L-2 shows that the change is largely associated with the change in the sample. A comparison of the coefficients on the interacted parenting skills variables in X and L - l , and L-2 indicates that the coefficients for purely cross-sectional equations (X and L-l) likely pick up long-term effects of parenting skills. Once the lagged variable is included, these long-term effects are captured by the lagged variable and the parenting skills coefficients decline. This pattern is consistent with the majority of the explanatory variables coefficients in the child and family characteristics section and with the hours in care coefficient. Figure 2.4, which is based on specification III (X) helps in assessing the differential impacts of parenting skills on mothers with different labour supply/child-care choices. The base case for the graphs in this section is the same as described in the PPVT section, except that the child is four years old. Children in care tend to be more hyperactive. For example, a child in 40 hours of care would have a score 7 percent higher than the mean score or 10 percent of a standard deviation higher than here I assume that all parents have access to other tools to control child behaviour, but choose particular parenting behaviours that in turn affect child behaviour. 53 the mean score. Children of stay-at-home mothers tend to be the least hyperactive except at higher parenting skills levels where children of working mothers who do not use daycare are the least hyperactive. At average parenting skill levels, children of stay-at-home mothers and children or working mothers who do not use daycare are equally hyperactive. FIGURE 2.4 Hyperactivity Scores, Labour Supply & Parenting Skills: Average Parental Education Levels Specification III(X) 8 0 Parenting Skills Similar graphs for high and low education levels show that hyperactivity decreases with education levels and that the lines cross at slightly lower levels of parenting skills when parental education is higher. Figures 2.5 and 2.6 based on specification III, L-l and L-2, respectively, show that in the longitudinal sample children of stay-at-home mothers score worse than children of working mothers at high parenting skills level, but better at low parenting skills levels. Although this crossing also occurs in the cross-sectional sample, it only occurs with the scores for children of mothers who work full-time but do not use daycare. 54 FIGURE 2.5 Hyperactivity Scores, Labour Supply & Parenting Skills: Average Parental Education Levels Specification III(L-l) 80 Parenting Skills These two figures also show that the results are robust to the inclusion of the lagged variable, but that sensitivity to (current) parenting skills is slightly diminished when prior scores are included. It is worth noting that although prior parenting skills may have had an effect on current child behaviour, the effect of current parenting skills is still very important. This suggests that improved parenting skills can have an immediate positive impact on children's behaviour, regardless of past parental practices. 55 FIGURE 2.6 Hyperactivity Scores, Labour Supply & Parenting Skills: Average Parental Education Levels Specification III(L-2) 80 Parenting Skills (e) Other behavioural scores for four and five year olds using pooled cross-sectional data Table 2.6 shows OLS estimates for specification II on the pooled cross-sectional observations for other behavioural scores of four and five year olds.60 The emotional disorder scores range from zero to sixteen, the conduct disorder scores from zero to twelve, the indirect aggression scores from zero to ten, the property offenses scores from zero to twelve, and the prosocial behaviour scores from zero to twenty. Very few observations lie at the negative extremes of these distributions, so the averages are quite low for the scores that are increasing in undesirable outcomes. The interacted parenting skills variable is the (negative of) ineffective parenting which ranges from zero to twenty-five. 56 The only significant maternal labour supply effect is for property offenses. Here, children of mother who do not work have the lowest scores, while children of mothers who work part-time have the highest scores. Children of fathers who work part-time score higher on property offenses but also score higher on prosocial behaviour. Parental education has no significant effect on these scores. Ineffective parenting skills have a major impact on all scores, but less so for prosocial behaviour, which is more influenced by positive interaction between the child and the parent. Child-care hours have only a statistically significant effect for the emotional disorder scores, which are worse when the child is in care. Specification II is not included because there are very few significant interactions. 57 TABLE 2.6 OLS Regression Coefficients - Other Behavioural Scores, 4-5 Years Old Cross-Sectional Data - Specification H (t-ratios in parenthesis) Conduct Property Emotional Prosocial Indirect Disorder Offenses Disorder Behaviour Aggression 7Y 12,769 12,809 12,788 11,997 12,295 R - Squared .2288 .2205 .1739 .1762 .1160 Mean score 1.56 1.86 0.98 1.28 2.01 2.15 11.84 4.06 0.71 1.29 Constant 173.3 -7.29 -72.38 -2.65 -102.7 -4.49 -5.98 -0.75 -41.35 -0.88 Parental Employment, Education and Parenting Skills (Both parents work full-time) Mother did not work -6.83 -1.22 -9.52 -1.87 -0.92 -0.19 -2.17 -1.24 -10.69 -1.20 Mother worked part-time 6.95 1.43 8.78 1.91 2.11 0.53 -0.20 -0.15 13.36 1.46 Father did not work 10.06 0.94 16.47 1.40 11.69 1.26 -2.13 -0.94 16.79 1.09 Father worked part-time -2.62 -0.46 12.55 1.90 3.15 0.53 4.44 2.35 3.00 0.25 Mother years of education 0.60 0.65 -0.54 -0.53 0.54 0.65 -0.46 -1.42 -0.94 -0.53 Father years of education -0.63 -0.71 -1.02 -1.13 -0.19 -0.23 -0.24 -0.84 -0.36 -0.21 Child Care - Hours in Care Total hours -0.12 -1.13 -0.01 -0.12 0.19 1.74 0.05 1.22 -0.16 -0.58 Child and Family Characteristics Five years old -13.13 -3.18 -4.32 -0.92 9.21 2.04 6.10 4.42 30.89 3.90 Boy 20.27 6.06 24.75 6.89 -0.84 -0.26 -10.50 -9.79 -19.61 -2.97 Immigrant parents -11.86 -1.95 -5.83 -0.84 -1.31 -0.21 -0.04 -0.02 25.65 1.68 Poor health 15.38 1.62 16.23 1.51 23.64 2.48 -1.55 -0.41 -1.45 -0.07 In kindergarden -1.00 -0.22 -2.01 -0.39 5.03 1.05 6.15 4.02 16.11 2.10 Mother (parent) age at birtl -0.31 -0.97 -1.01 -2.99 -0.23 -0.85 -0.29 -2.77 -0.95 -1.94 Number of children < 17 10.00 3.85 -4.05 -1.48 10.11 4.38 3.42 4.03 -2.25 -0.38 Number of older children 5.88 2.14 10.39 3.45 -20.33 -8.07 -4.50 -4.86 17.74 3.09 Reads or is read to daily -3.03 -0.78 -10.94 -2.57 0.26 0.07 0.40 0.32 8.02 1.10 " " several times a day 11.40 1.39 -12.36 -1.60 11.00 1.39 2.28 0.97 22.56 1.57 P M K depression score 1.98 4.25 2.56 4.62 3.69 8.43 0.27 2.01 2.46 2.07 Ineffective parenting skills 12.32 17.33 11.22 13.98 7.81 13.28 -0.48 -2.26 10.98 6.10 Punitive score 2.80 2.43 3.75 3.15 1.78 1.80 -1.48 -4.35 1.12 0.60 Consistency score -0.31 -0.48 -2.46 -3.64 -0.74 -1.27 0.38 2.00 -2.79 -2.24 Positive interaction score 1.48 2.16 0.42 0.53 -0.10 -0.13 2.45 10.82 -0.61 -0.46 Family dysfunction score 0.54 1.45 0.23 0.57 0.44 1.27 -0.76 -6.35 0.73 1.16 continued 58 T A B L E 2.6 (continued) (t-ratios in parenthesis) Conduct Property Emotional Prosocial Indirect Disorder Offenses Disorder Behaviour Aggression Family Income Less than $20,000 -10.89 -1.15 8.92 0.74 1.37 0.16 1.24 0.48 -37.22 -2.30 $20,000 - $29,999 -16.66 -2.33 -3.80 -0.43 -6.84 -1.15 -1.82 -0.81 -39.30 -2.61 $30,000 - $39,999 -7.84 -1.12 -5.15 -0.61 -2.87 -0.49 -0.31 -0.15 -16.24 -0.93 $40,000 - $49,999 -12.03 -1.85 -11.97 -1.66 -8.75 -1.51 -2.62 -1.29 -27.68 -1.83 $60,000 - $69,999 -15.24 -2.23 -10.86 -1.43 -9.22 -1.57 -0.71 -0.36 -33.96 -2.23 $70,000 - $79,999 -18.47 -2.41 -19.58 -2.38 -9.09 -1.36 -3.70 -1.44 -39.79 -2.66 $80,000 + -5.75 -0.86 -7.03 -0.96 -6.99 -1.22 -1.36 -0.66 -21.29 -1.39 Location Newfoundland -13.95 -2.16 -20.92 -2.99 -30.74 -5.24 -9.08 -4.44 -0.37 -0.03 Prince Edward Island 6.87 0.79 -15.98 -1.86 -1.48 -0.17 -5.33 -1.99 10.10 0.66 Nova Scotia 6.46 0.96 -2.39 -0.30 -12.88 -2.16 2.46 1.14 5.54 0.51 New-Brunswick 13.87 2.29 5.05 0.77 -7.06 -1.42 -1.80 -1.02 4.00 0.38 Quebec 13.55 2.48 24.30 3.97 10.38 1.99 -8.58 -4.84 6.22 0.54 Manitoba 11.10 1.55 4.62 0.59 0.20 0.03 -2.37 -1.16 -17.88 -1.29 Saskatchewan 15.02 2.43 2.04 0.30 0.86 0.16 -0.83 -0.49 -1.81 -0.17 Alberta 10.04 1.47 7.41 1.12 -8.50 -1.56 -2.07 -1.15 -10.50 -0.91 Bri t ish Columbia 11.95 1.87 8.10 1.25 6.56 1.12 -0.59 -0.30 -3.62 -0.35 Urban - 500,000 + -8.71 -1.73 1.19 0.23 -1.43 -0.33 0.64 0.42 7.26 0.74 Urban - 100,000-499,000 -9.60 -2.13 10.12 2.09 -2.45 -0.59 3.65 2.66 -4.42 -0.58 Urban < 100,000 -3.51 -0.80 6.09 1.30 -1.71 -0.46 2.21 1.82 -3.47 -0.50 Base case* -4.7 -14.6 -9.0 -0.4 -15.6 Mother does not work/no care -6.8 -23.5 -17.4 -4.4 -20.1 Ineffective parenting down 1 SD -51.4 -64.1 -45.7 2.2 -59.8 Mother works full time -44.6 -54.6 -44.7 4.4 -49.1 Daycare to 40 hrs -49.3 -55.2 -37.3 6.2 -55.4 *The base case is a four year old girl whose parents both work full-time. The girl is in daycare for 40 hours per week. The parents both have 12 years o f education and are not immigrants. The family lives in B . C . and earns between $80,000 or more per year. The family has two children and the other child is older. The child is healthy and reads or is read to daily. The mother was 30 years old when the child was born. The family has otherwise average characteristics. Subsequent cases build on the immediately preceding one. For prosocial behaviour it is the positive interaction score that has been improved by one standard deviation rather than the parenting skills score. ' 'Only children in two-parent families are included here Older children fare worse on the indirect aggression and emotional disorder scores and better on conduct disorder and prosocial behaviour scores. Boys have worse conduct disorder, property offenses and prosocial behaviour scores, but have better indirect aggression scores. Children of irnmigrant parents score better on conduct disorder but worse on indirect aggression. Poor health is associated with higher emotional disorder scores. Children in kindergarden have better prosocial behaviour scores but worse indirect aggression scores. 59 Children of younger mothers have worse property offense scores and indirect aggression scores, but better prosocial behaviour scores. Family size has a negative influence on conduct and emotional disorders, but a positive effect on prosocial behaviour. Birth order has a negative influence on conduct disorder, property offenses, prosocial behaviour and indirect aggression, but a positive influence on emotional disorders. Reading is associated with reduced property offenses. The PMK depression score has a significant negative influence all scores. The punitive score has a significant negative influence on all scores except indirect aggression. Consistency has a significant positive influence on property offenses, prosocial behaviour and indirect aggression. Positive interaction has a significant positive influence on prosocial behaviour, but a negative one on conduct disorder. Family dysfunction has a significant negative effect on prosocial behaviour. Family income effects are not monotonic. The only relatively remarkable finding here is that children in the reference income category and in the $30,000 - $39,999 tend to have the lowest indirect aggression score. Overall, these findings do not support any significant and widespread effects of labour supply and child-care on children's behaviour scores. However, the findings do show that reported parenting behaviour is the most important determinant of the scores.61 (f) Behavioural scores for four and five year olds using longitudinal information Table 2.7 shows OLS estimates for specification II for other behavioural scores of 4 and 5 year olds using the longitudinal data. Results are similar to those in table 2.6, although the only significant labour supply effect now is a lower emotional disorder scores for children of fathers working part-time. As in table 2.6, there is a significant positive coefficient on hours in care for emotional disorder, but here there is also a significant positive coefficient on hours in care for prosocial behaviour. Children in care for 40 6 1 Parents answer the parenting behaviour and child behaviour questions. Relationships between the two sets of answers may be exaggerated by parents' optimism or pessimism. 60 hours per week have predicted scores that are higher by approximately 13 percent of a standard deviation for emotional disorder and 21 percent for prosocial behaviour. TABLE 2.7 OLS Regression Coefficients - Other Behavioural Scores, 4-5 Years Old Longitudinal Data - Specification II (t-ratios i n parenthesis) Conduct Property Emotional Prosocial Indirect Disorder Offenses Disorder Behaviour Aggression N 4,555 4,561 4,624 3,940 4,384 R - Squared .3162 .2519 .2315 .2639 .1805 Mean 1.66 -1.90 1.06 1.35 2.04 2.16 11.79 3.98 0.71 -1.31 Constant 193.58 -5.16 -91.71 -2.07 -144.6 -4.16 -23.51 -1.93 -128.83 P r i o r Score** 9.08 9.41 5.30 4.88 18.53 9.38 4.07 11.9 6.35 3.01 Parental Employment, Education and Parenting Ski l l s (Both parents work full-time) Mother did not work -7.46 -0.87 -8.88 -1.15 -6.40 -0.90 -0.35 -0.13 -13.09 •0.92 Mother worked part-time 7.39 1.12 8.30 1.22 -0.01 -0.00 0.42 0.20 13.29 1.03 Father did not work -4.58 -0.24 6.67 0.25 5.48 0.30 -2.99 -0.54 55.50 1.15 Father worked part-time -11.96 -1.48 2.41 0.28 -12.03 -1.75 0.47 0.20 10.94 0.57 Mother yrs of education 1.39 1.04 0.33 0.20 1.81 1.39 -1.01 -2.00 0.34 0.15 Father years of education -1.90 -1.34 -1.36 -0.94 -0.65 -0.49 0.40 0.98 -2.58 -0.97 C h i l d Care - Hours i n Care Total hours 0.04 0.30 0.20 1.00 0.35 2.23 0.18 2.98 0.01 0.01 C h i l d and Fami ly Characteristics Five years old -13.36 -2.37 -7.39 -0.94 13.14 1.80 1.42 0.69 53.60 3.94 Boy 20.05 4.40 25.56 4.69 6.23 1.26 -6.70 -4.02 -23.92 -2.23 Immigrant parents -3.54 -0.34 15.84 1.41 1.27 0.14 4.95 1.31 47.68 1.95 Poor health 6.86 0.46 11.35 0.68 11.68 0.72 -4.70 -0.67 -34.99 -1.82 In kindergarden 0.47 0.08 -2.74 -0.36 -1.29 -0.19 2.31 1.10 5.86 0.52 Mother/parent age at birth -0.33 -0.79 -0.90 -1.86 -0.21 -0.52 -0.11 -0.72 -0.68 -1.02 Number of children < 17 8.04 2.06 -9.29 -2.16 10.66 2.82 -0.09 -0.07 -4.32 -0.53 Number of older children 9.51 2.36 15.58 3.30 -13.69 -3.39 -0.21 -0.16 25.65 2.71 Reads or is read to daily -0.46 -0.09 -9.58 -1.49 1.55 0.30 -0.45 -0.24 21.72 2.10 " " several times a day 16.97 1.57 -12.46 -1.00 11.13 0.92 1.49 0.42 39.05 1.73 PMK. depression score 1.36 2.41 1.20 1.66 2.70 4.66 0.05 0.22 0.82 0.49 Ineffective parenting skills 11.04 10.24 10.70 8.66 7.90 8.90 -0.43 -1.25 12.34 4.11 Punitive score 1.16 0.72 3.70 1.92 2.17 1.43 -0.82 -1.45 3.46 1.18 Consistency score 0.42 0.42 -2.02 -1.98 0.45 0.51 -0.13 -0.46 -1.02 -0.50 Positive interaction score 1.38 1.33 0.19 0.15 -0.01 -0.01 2.39 6.55 -0.73 -0.34 Family dysfunction score 0.60 1.15 -0.01 -0.02 -0.82 -1.62 -0.66 -3.61 0.46 0.52 continued 61 T A B L E 2.7(continued) (t-ratios in parenthesis) Conduct Property Emotional Prosocial Indirect Disorder Offenses Disorder Behaviour Aggression Family Income Less than $20,000 -0.88 -0.07 15.99 0.86 6 44 0.53 -2 04 -0.45 -72 60 -2.45 $20,000 - $29,999 -18.13 -1.90 1.40 0.12 2 11 0.24 -5 07 -1.17 -28 84 -1.13 $30,000 - $39,999 -12.54 -1.41 -9.06 -0.72 3 36 0.38 -3 21 -0.92 -24 18 -0.83 $40,000 - $49,999 -16.11 -1.84 -16.37 -1.60 -14 13 -1.76 -7 60 -2.03 -27 61 -1.08 $60,000 - $69,999 -18.19 -2.00 -12.85 -1.21 -12 70 -1.51 -1 82 -0.56 -42 39 -1.69 $70,000 - $79,999 -38.65 -3.78 -19.37 -1.49 -16 66 -1.69 -5 14 -1.29 -53 76 -2.01 $80,000 + -8.46 -0.89 -9.44 -0.83 -9 99 -1.12 -5 18 -1.45 -33 07 -1.24 Location Newfoundland -8.68 -0.89 -16.93 -1.43 -25 25 -2.74 -2 83 -0.86 3 02 0.13 Prince Edward Island 1.03 0.08 -18.47 -1.43 -7 57 -0.59 -12 14 -2.96 -15 37 -0.62 Nova Scotia 5.08 0.56 -2.37 -0.20 -17 56 -1.96 0 79 0.23 2 55 0.15 New-Brunswick 27.54 2.47 15.31 1.24 -4 72 -0.60 -4 35 -1.47 17 94 0.97 Quebec 7.43 0.99 25.89 2.72 6 44 0.84 -4 93 -1.83 18 20 1.08 Manitoba -11.15 -1.23 5.59 0.49 -8 90 -0.84 -4 97 -1.64 -37 47 -1.87 Saskatchewan 7.09 0.83 -7.23 -0.77 -9 75 -1.25 -2 24 -0.81 -18 11 -1.13 Alberta 1.07 0.11 6.57 0.63 -21 68 -2.46 -0 37 -0.14 -30 08 -1.81 British Columbia -2.83 -0.29 0.72 0.07 -3 34 -0.36 -2 14 -0.75 -1 95 -0.12 Urban - 500,000 + -3.16 -0.46 -8.10 -1.06 0 32 0.05 0 36 0.15 14 80 0.98 Urban - 100,000 - 499,00C -16.89 -2.58 7.44 1.01 -9 20 -1.36 2 82 1.33 -4 26 -0.35 Urban < 100,000 0.41 0.07 2.83 0.42 -4 22 -0.74 4 09 2.36 -1 10 -0.10 Base case* -104.4 -116.0 -76.9 7.3 14.9 Mother does not work/no care -113.7 -132.7 -97.3 -0.1 1.6 Ineffective parenting down 1 SD -154.2 -172.0 -126.3 6.0 -5.4 Mother works full time -146.8 -163.1 -119.9 6.4 7.7 Daycare to 40 hrs -145.0 -155.3 -105.9 13.4 7.9 * The base case is a four year old girl whose parents both work full-time. The girl is in daycare for 40 hours per week. The parents both have 12 years of education and are not immigrants. The family lives in B.C. and earns between $80,000 or more per year. The family has two children and the other child is older. The child is healthy and reads or is read to daily. The mother was 30 years old when the child was born. The family has otherwise average characteristics. Subsequent cases build on the immediately preceding one. For prosocial behaviour it is the positive interaction score that has been improved by one standard deviation rather than the parenting skills score, and for indirect aggression, it is the punitive score that has been improved. **The prior score is the matching score for emotional disorder and prosocial behaviour, and it is the physical aggression and opposition score for the other bahavioural measures. ***Only children in two-parent families are included here The effect of parenting skills is quite important, as in table 2.6. The significant point estimates for child and family characteristics are remarkably similar to those in table 2.6 although there are some exceptions. The coefficient for five year olds is no longer significantly positive for prosocial behaviour. For children of irnrnigrant parents, the negative conduct disorder coefficient is no longer significant. The poor health coefficient is no longer significantly positive for emotional disorder, but is now significantly negative for indirect aggression. The kindergarden coefficients are 62 no longer significantly positive for prosocial behaviour and indirect aggression. The mother's age at birth coefficients are no longer significantly negative for prosocial behaviour and indirect aggression. The family size coefficient is no longer significantly positive for prosocial behaviour but it is significantly negative for property offenses. The birth order coefficient is no longer significantly negative for prosocial behaviour. The daily reading coefficient for property offenses is no longer significantly negative but frequent reading coefficients are now significantly positive for indirect aggression. The PMK depression score coefficients are no longer significantly positive for prosocial behaviour. The ineffective parenting score coefficient is no longer significantly negative for prosocial behaviour. The punitive score coefficient is no longer significantly positive for emotional disorder and negative for prosocial behaviour. The consistency score coefficients are no longer significant for prosocial behaviour and indirect aggression. Some of the changes in the magnitude and significance of coefficients are due to the smaller sample, but also because the prior child behavioural score now captures part of child and family characteristics that may be long-term and have long-term impacts. (g) Behavioural scores for two and three year olds using pooled cross-sectional datasets Table 2.8 shows OLS estimates for specification II, on the pooled observations for two and three year old children behavioural scores. The scores include hyperactivity-inattention scores, ranging from zero to fourteen, emotional disorder-anxiety scores, ranging from zero to twelve, physical aggression and opposition scores, ranging from zero to sixteen, separation anxiety scores, ranging from zero to ten, and prosocial behaviour scores, ranging from zero to ten. The interacted parenting skills variable is (the negative of) ineffective parenting. The dependent variable is the percentage difference between the score and the mean score. 63 TABLE 2.8 OLS Regression Coefficients - Behavioural Scores, 2-3 Years Old Cross-Sectional Data - Specification IT (t-ratios i n parenthesis) Hyper Emot iona l P r o s o c i a l Separat ion act iv i ty Aggress ion Di so rde r Behaviour Anxie ty 7Y** 9356 9289 9385 8627 9422 R - Squared .2196 .3420 .1367 .1864 .1477 Mean 4.13 2.92 4.93 3.01 1.07 1.40 5.60 2.80 2.74 2.01 Constant 24.69 -1.71 -70.75 -5.99 -64.97 -2.47 -30.07 -2.53 18.72 1.12 Parenta l Employment , Educa t ion and Pa ren t ing S k i l l s (Both parents work full-time) Mother did not work 0.26 0.09 -0.01 -0.00 6.52 1.14 -0.99 -0.42 5.40 1.56 Mother worked part-time -3.19 -1.24 -1.86 -0.98 -2.60 -0.55 1.95 0.90 0.58 0.20 Father did not work -5.17 -1.02 -8.32 -2.17 -20.87 -2.52 -1.51 -0.42 -0.41 -0.08 Father worked part-time -3.32 -0.96 1.39 0.48 -2.65 -0.40 1.49 0.54 -2.13 -0.45 Mother yrs o f education -1.76 -3.38 0.43 1.04 -1.07 -0.97 0.51 1.13 -0.86 -1.33 Father years o f education -0.88 -1.74 -0.37 -0.97 -0.01 -0.01 -0.83 -1.95 -1.16 -1.86 C h i l d C a re - H o u r s i n Care Total hours 0.18 3.01 0.12 2.24 0.00 0.02 0.12 2.20 0.13 1.49 C h i l d and F a m i l y Charac ter i s t ics Three years o ld -2.42 -1.20 -5.25 -3.36 16.00 3.91 16.51 10.52 2.75 1.24 B o y 8.10 4.22 4.20 2.79 6.74 1.71 -15.03 -9.76 3.25 1.51 Immigrant parents -0.89 -0.25 -10.96 -3.75 16.31 2.15 -3.15 -0.93 14.69 3.07 Poor health 23.04 2.49 12.86 1.99 67.15 3.86 -30.35 -4.21 35.96 3.73 Mother/parent age at birtl -0.31 -1.55 -0.25 -1.59 -0.29 -1.02 -0.09 -0.60 0.02 0.11 Number o f children < 17 -0.89 -0.45 0.84 0.53 19.80 4.85 7.63 5.40 -1.92 -0.97 Number of older children -1.94 -0.95 6.36 3.90 -34.92 -8.74 -7.57 -4.84 -0.53 -0.25 Reads or is read to daily -5.00 -2.00 -2.83 -1.46 2.53 0.51 3.47 1.72 -2.26 -0.83 " " several times a day -6.84 -2.01 0.38 0.14 10.49 1.64 11.23 4.02 0.60 0.15 P M K depression score 1.15 5.14 0.81 4.36 2.62 6.03 0.74 4.35 1.47 5.45 Ineffective parenting scor 6.90 17.95 8.18 30.64 7.93 12.17 0.42 1.48 4.38 9.69 Punitive score 1.95 3.65 1.28 3.10 -0.01 -0.01 -2.76 -6.10 -0.63 -0.98 Consistency score -1.23 -3.78 -1.31 -5.13 -0.95 -1.62 0.98 3.91 -3.07 -7.52 Positive interaction score 0.74 1.65 0.53 1.51 -0.43 -0.53 2.36 6.85 0.48 1.01 Family dysfunction score -0.66 -3.08 -0.26 -1.63 0.22 0.50 -0.65 -3.44 0.11 0.45 continued The only significant labour supply coefficients are negative coefficients on the aggression and emotional disorder scores for children of non-working fathers. There are no significant maternal labour supply effects. Hours in care are associated with higher hyperactivity and aggression scores, but also with higher prosocial behaviour scores. A child in 40 hours of care would be expected to have hyperactivity and prosocial behaviour scores of 10 percent of a standard deviation above the average, and an aggression score 8 percent of a standard deviation above the average. 64 T A B L E 2.8 (continued) (t-ratios in parenthesis) Hyper- Emotional Prosocial Separation activity Aggression Disorder Behaviour Anxiety Family Income Less than $20,000 10.67 1.96 7.37 1.63 8.47 0.82 2.41 0.59 11.38 1.70 $20,000 - $29,999 6.70 1.53 1.00 0.32 10.97 1.36 3.43 1.06 4.31 0.84 $30,000 - $39,999 0.03 0.01 4.91 1.83 1.76 0.24 3.03 1.03 10.46 2.25 $40,000 - $49,999 1.09 0.32 1.28 0.46 0.06 0.01 -1.98 -0.72 3.10 0.77 $60,000 - $69,999 2.34 0.62 -2.36 -0.83 -11.57 -1.69 -2.17 -0.66 -2.83 -0.66 $70,000 - $79,999 0.47 0.10 -0.98 -0:29 -6.23 -0.79 -5.24 -1.61 -11.29 -2.44 $80,000+ -1.07 -0.29 3.02 1.13 3.43 0.47 3.80 1.29 1.33 0.30 Location Newfoundland -1.33 -0.30 -12.31 -3.04 -14.67 -1.72 -7.54 -2.54 -4.21 -0.94 Prince Edward Island 6.46 1.30 3.70 0.95 1.30 0.13 -2.62 -0.66 -7.02 -1.24 NovaScot ia -3.53 -0.94 -3.41 -1.11 -6.81 -0.76 -1.95 -0.66 -1.98 -0.46 New-Brunswick -8.71 -2.60 -6.58 -2.54 -10.48 -1.55 -9.42 -3.83 -6.52 -1.81 Quebec -2.51 -0.80 -9.97 -4.18 -5.43 -0.98 -20.00 -8.09 -7.19 -2.12 Manitoba -0:49 -0.13 0.58 0.18 5.64 0.77 -0.79 -0.27 -4.07 -1.03 Saskatchewan -6.63 -1.83 -5.07 -1.78 6.37 0.85 -3.27 -1.30 3.47 0.93 Alberta -2.97 -0.88 -4.28 -1.61 -9.95 -1.52 -3.08 -1.18 -12.14 -3.21 Br i t i sh Columbia -2.30 -0.68 -1.40 -0.53 -3.35 -0.52 1.63 0.58 -2.76 -0.66 Urban - 500,000 + -6.46 -2.32 -4.29 -1.96 -7.26 -1.36 -3.86 -1.84 0.72 0.23 Urban- 100,000-499,00 0.71 0.26 0.39 0.18 -1.60 -0.30 0.26 0.14 -0.03 -0.01 Urban< 100,000 0.61 0.25 0,70 0.35 -3.10 -0.67 -2.27 -1.32 0.68 0.26 Base case* 32.3 Mother does not work/no care 25.2 Ineffective parenting down 1 SD -0.3 Mother works full time -0.6 Daycare to 40 hrs 6.8 2.5 -2.3 -32.5 -32.5 -27.7 -16.4 -10.0 -39.3 -45.8 -45.7 3.3 -2.4 3.5 4.5 9.3 11.5 11.9 -4.3 -9.7 -4.7 * The base case is a two year old girl whose parents both work full-time. The girl is in daycare for 40 hours per week. The parents both have 12 years o f education and are not immigrants. The family lives in B . C . and earns between $80,000 or more per year. The family has two children and the other child is older. The chi ld is healthy is read to daily. The mother was 30 years o ld when the chi ld was born. The family has otherwise average characteristics. Subsequent cases build on the immediately preceding one. For prosocial behaviour it is the positive interaction score that has been improved by one standard deviation rather than the parenting skills score. '*Only children in two-parent families are included here Older children have lower aggression, and higher emotional disorder and prosocial behaviour scores. Boys have worse scores for all the measures except separation anxiety. Children of irnmigrants have lower aggression scores, and higher emotional disorder and separation anxiety scores. Children with poor health score worse on all measures. Children from larger families have worse emotional disorder scores, but better prosocial behaviour scores. Children who are higher in the birth order have a higher physical aggression score, and lower emotional disorder and prosocial behaviour scores. Coefficients on the other measures of parenting skills and family functioning 65 generally have the expected signs, and many are significant. However children of depressed PMKs have higher prosocial behaviour scores and family dysfunction is associated with decreased hyperactivity scores. Children of depressed PMKs could be more sociable in an attempt to cheer up their depressed parent and family dysfunction may well be a deterrent to a child's conspicuous behaviour. Daily reading decreases hyperactivity and increases prosocial behaviour. Al l scores are somewhat decreasing in income, although most income effects are not significant. Overall, this group of children is not largely affected by labour supply and but is somewhat affected by child-care. Again, the behaviour of these children is largely dependent on family functioning, parenting skills, health and other family or child characteristics. 2.6 CONCLUSION Assuming that maternal labour supply, income, and child-care use are exogenous variables, the above analysis indicates that maternal labour supply alone does not have any detrimental effect on pre-school children's cognitive scores for parents with average education and parenting skills.63 Findings for standardized PPVT scores are also consistent with the hypothesis that children of parents with better parenting skills tend to benefit more in terms of cognitive skills from additional parental time than children of less skilled parents although the effects are small. Although hours in care have no overall effect on PPVT scores, there is evidence that quality of care in terms of its impact on cognitive outcomes increases with income. Fixed effect models indicate that within families, increased family dysfunction has adverse effects on children's cognitive scores. Maternal labour supply is associated with reduced hyperactivity scores for four and five year olds in two-parent families with better than average parenting skills,64 but with possibly increased hyperactivity scores when parenting skills are worse than average. This arises because parenting 6 2 Positive interaction is associated with higher hyperactivity as for 4-5 year olds. It is also associated with higher conduct disorder. 6 3 Here parenting skills refer to consistency scores. 6 4 Here parenting skills refer to the negative of the ineffective parenting scores. 66 skills coefficients for working mothers are larger than those for non-working mothers. Hence, in two-parent families where both parents work, good parenting skills are particularly important for children hyperactivity scores. Furthermore, hours in care also increase hyperactivity levels, so that children of mothers who work and use child-care tend to score worse than children whose mothers do not work, unless the mothers have exceptionally good parenting skills. The evidence also indicates that parental labour supply has no effect on conduct disorder and indirect aggression scores for four and five year olds and on hyperactivity, prosocial behaviour, and separation anxiety scores for two and three year olds. Four and five year old children of mothers who do not work have better property offenses scores, but those of parents who work part-time have worse scores. Four and five year old children of fathers who work part-time have better emotional disorder and prosocial behaviour scores. Two and three year old children of fathers who do not work have worse aggression and emotional disorder scores. Hours in care are also associated with higher prosocial behaviour scores for all children, with higher emotional disorder scores for four and five year olds, and with higher aggression scores for two and three year old children. While labour supply and child-care use are shown to have limited effects on children's cognitive and behavioural scores, it is clear that parenting skills and family functioning have large and significant effects on all of the scores. When I test for endogeneity of maternal labour supply, I find limited evidence that suggests it may be endogenous. This evidence also indicates that maternal labour supply may have a significant negative impact on PPVT scores. These findings would indicate an effective children's agenda would put a great deal of focus on helping families with small children with parenting skills and family functioning issues. In addition, as child-care quality is increasing in income,65 there should be some consideration given to 6 5 The CNCCS data shows that higher income families spend more on child-care than lower income families. 67 assisting lower, income families with their child-care cost. While many low-income single parent families now qualify for child-care subsidies, the same is not true for lower income two parent families. Results also indicate that further research in this area should place emphasis on testing for the endogeneity of maternal labour supply and/or estimating instrumental variables models. In the presence of contradictory findings on the potential endogeneity of maternal labour supply, policy implications are more difficult to arrive at. If maternal labour supply has little or no effect on child outcomes, policies that encourage women to spend more time at home with their children may be misguided if they are costly and also negatively affect mothers' future work opportunities. Funds may be better spent supporting families with their parenting responsibilities and their child-care costs, and on increasing the quality of the child-care experience for children. On the other hand, if maternal labour supply negatively affects children outcomes, sufficiently generous extended parental/maternity leave policies can benefit children. 68 CHAPTER III CHILD-CARE COSTS AND CANADIAN WOMEN'S MARKET WORK: HETEROGENEOUS PREFERENCES 3.1 INTRODUCTION In this chapter, I examine the impact of child-care costs on the continuity of Canadian women's market work. In order to account for heterogeneity of preferences and habit formation, I use prior/current occupation ranking and weeks worked in the last twelve months as additional control variables in two separate estimation procedures. There has been prior research on the impact of child-care costs on the market work of Canadian women with preschool children.66 However, prior research does not control for occupation and weeks worked and focuses on all women with preschool children. The question of interest here concerns women's decision to return to work soon after the birth of their last child. The relevant population is defined as married women with children less than three years old, who have exhausted federal maternity benefit provisions, and who have participated in market work in the previous five years. Why should we be concerned about potential inhibiting effects of child-care costs on married women's labour supply? The availability of quality, affordable child-care has been an issue of importance to women's groups for many years. One argument made for affordable child-care is that women need to show a strong attachment to the labour force in order to maintain economic self-sufficiency and independence, and child-care costs can weaken this attachment. Phipps, Burton, and Lethbridge (2001) provide evidence for the family (wage) gap and find that a portion of the gap is due to changing jobs after a job interruption as well as fatigue or scheduling difficulties. Thus, if a woman is to maintain her wage level after taking time off for childbirth, she will be in a better 6 6 For example, see Cleveland, Gunderson, and Hyatt (1996), and Powell (1997) who examine the impact of child-care costs on the work of married Canadian women with children under six years of age. 69 position to do so if she returns to her pre-childbirth job. In order to do this, the length of her absence from work will have to be relatively short. Married women with children can therefore experience a reduction in their wage relative to that of their husband if they drop out of the labour force for a significant period of time. Bargaining theory predicts that under such conditions, the woman's ability to derive favourable bargaining outcomes in a marital union will be impaired. Furthermore, should marriage end in divorce, a woman may experience a considerable drop in her standard of living, unless the divorce settlement specifically compensates her for her loss of human capital while she took responsibility for child-rearing. Child-care policy has a number of objectives revolving around the welfare of children and their parents. In this chapter, I address the following question: can child-care policy exert a significant impact on the labour supply of married mothers? Section 3.2 provides a literature survey, section 3.3 includes a description of the data and some preliminary statistics, section 3.4 presents the model and discusses econometric issues, section 3.5 covers the results, and section 3.6 concludes the chapter. 3.2 L I T E R A T U R E S U R V E Y A number of US studies have found that the labour supply of mothers is sensitive to child-care costs (Blau and Robins, 1988; Ribar, 1992 and 1995; Connelly, 1992; Barrow, 1999), although other studies find little or no effect (Blau and Robins, 1991, Ribar, 1995). Most of these studies also find that the decision to pay for care is sensitive to costs. Blau and Robins (1989) find that higher child-care costs result in lower fertility. They also find that higher costs result in more mothers quitting work and fewer mothers returning to work. In a study of child-care usage, Blau and Hagy (1998) find that a lower price for a particular type of child-care leads to substitution toward that type and to an increase in the use of paid care; they 70 also find that a decrease in the price of care causes an increase in the hours of care demanded but to a decrease in the demand for what they define as quality-related attributes. These attributes include group size, staff/child ratio and caregiver training. The study is not restricted to working mothers. Hofferth and Wissoker (1992) find that both direct child-care subsidies and tax credits for child-care expenses increase the probability that group care will be used. In a study of the importance of care characteristics, Johansen, Leibowitz, and Waite (1996) find that parents who value developmental characteristics of care choose group care, while those for whom hours, location, and cost are important choose care at home. In a Canadian study, Cleveland, Gunderson and Hyatt (1996) estimate the joint decision of married women with preschool children to work and to choose paid care. They find that increasing expected child-care costs reduces the probability of using paid care and of working. In another paper, Cleveland and Hyatt (1996) carry out the same analysis for single mothers; the findings are the same, except that they also find that expected social assistance income appears to have a large negative impact on the probability of employment. In a similar Canadian study, Powell (1997) corroborates the findings for married Canadian mothers of preschoolers and also analyzes the determinants of hours of work for these mothers. Powell finds that increasing expected child-care costs also reduces hours of work. In a subsequent paper, Powell (1998) finds that expected child-care costs have a much stronger impact on the probability that married mothers of preschoolers will work full-time than on the probability that she will work part-time. In another Canadian study, Cleveland and Krashinsky (1998) do a cost/benefit analysis of the universal provision of quality licensed child-care to all children aged two to five years with employed parents, as well as enriched nursery school for children cared for primarily by their parents. They conclude that for every dollar spent on such a program, approximately two dollars worth of benefits are generated for children and their parents aside from any other benefits provided to society at large. 71 A series of studies of British and European mothers (Joshi, 1990; Joshi and Davies, 1992; Joshi and Davies, 1993) estimate the probability of withdrawal from the labour force of mothers and the resultant human capital costs of childbearing; they find that for women who increase their labour supply as a result of the availability of childcare, the value of the additional output generated exceeds the resource costs of the childcare. In another British study, Duncan and Giles (1996) argue that preschool childcare should be subsidized on the grounds that this would encourage those with childcare responsibilities to return to work and would minimize the potentially damaging financial consequences of a spell out of the labour market. They examine various forms of subsidies and find that childcare reforms are unlikely to pay for themselves through increased tax revenue. They argue that if universal provision cannot be justified for all preschool children, it should be for near school-age children, and that targeting towards low-income families is a sensible policy option because it addresses important distributional concerns. A number of studies have focused on evaluating impacts of particular policies or programs. In a US study, Michalopoulos, Robins and Garfinkel (1992) study the impact a refundable child-care tax credit would have on labour supply and child-care demand. They find that such a program would distribute child-care benefits more equally across the population by increasing the shares of subsidies received by low-income families and would result in a considerable increase in expenditures on paid care. Hofferth (1996) looks at the effects of public and private policies on the employment patterns of American mothers after childbirth. She finds that mothers who have access to part-time work or workplace childcare from their employers return to part-time work sooner, and that mothers who have access to liberal unpaid leave return to full-time work sooner than mothers without access to these policies. In their study of a similar group of mothers, Klerman and Leibowitz (1990) find that the expected level of child-care tax credits is positively related to the likelihood of a mother returning to work within three months of her first child's birth. In a related study, Leibowitz, Klerman, and Waite 72 (1992) corroborate the previous finding; they find that wages relate positively to early return to work, while higher family income delays return to work. They also find that, contrary to their expectations, tax credits do not affect child-care choice. Hotz and Kilburn (1995) find that state regulations increase the cost of child-care and reduce the usage of paid care, especially among households with non-working mothers. In another regulation study, Lowenberg and Tinnin (1992) argue that whether consumers or producers gain most from licensure can be determined by examining the effect of licensure on aggregate consumption: if aggregate consumption increases, consumers benefit, otherwise it is producers that primarily benefit. They find that increased regulation results in reduced consumption and therefore conclude that it primarily benefits producers. Chipty (1995) finds that regulations are binding on providers, that they have spillover effects on competitors and that certain types of regulations are more effective at raising quality than others. In his study on the supply of child-care labour, Blau (1993) estimates the elasticity of supply of child-care labour. His findings reveal elasticities of 1.2-1.9 and he concludes that the majority of childcare subsidies accrue to consumers of child-care. Walker (1992) examines family providers and their fees and finds the following: unregulated providers care for fewer children per establishment and offer a more adult-intensive form of care; licensed family providers exhibit more commitment to the profession; family providers receive no returns to experience or education; family providers offer large discounts in fees covering more than one child. The next group of studies does not focus on child-care issues, but rather on the idea that the traditional models used to estimate the labour supply of women do not adequately deal with heterogeneity issues. In particular, cross-sectional estimates of labour supply are poor predictors of a woman's future labour supply. Heckman and Willis (1977) use panel data to estimate the labour supply of married women. The model used takes into consideration the observation that current labour supply is not independent 73 of prior labour supply. They find that when prior labour supply is statistically controlled for, predicted participation probabilities are either near zero or near one. In their study on the work behaviour of married mothers, Nakamura and Nakamura (1985a) investigate the related hypothesis that a woman's (unobserved) preferences for a certain lifestyle are correlated with observed child status, her educational level and her work behaviour, along with some characteristics of her husband. Because these preferences are unobserved and are possibly correlated with explanatory variables for labour supply, coefficient estimates that do not take preferences into consideration are biased. Hence the traditional labour supply model applied to cross-sectional data suffers from omitted variables bias.67 Nakamura and Nakamura propose a number of methods to account for the heterogeneity. These methods are based on the assumption that previous labour supply is a good predictor of future labour supply and is a reflection of unobserved preferences. Often cross-sectional data include labour supply variables for the previous year, and the unobserved heterogeneity problem can be remedied by incorporating this information in the estimation procedure. See the appendix for a simple derivation of the labour supply model with heterogeneous preferences. Nakamura and Nakamura find that a woman's child status variables have little impact on her current work behaviour after controlling for her work behaviour in the previous year. Similarly, Woittiez and Kapteyn (1998) study the impact of habit formation on female labour supply. They find that habit formation contributes significantly to the explanation of female labour supply. Using Canadian data, Boothby (1984) estimates the conditional probability that a married woman will participate in a period, given her observed characteristics and her participation status in the previous year. He finds that there is considerably more continuity in married women's labour force participation than cross-section models of participation would imply. 6 7 Heterogeneity issues can be adequately addressed using fixed effect models with panel data, but since much of our labour force data is cross-sectional, other methods need to be used to deal with this issue. 74 In a follow-up to their 1985 study, Nakamura and Nakamura (1994) use Canadian and US census data to classify women into groups according to the number of weeks they worked in the previous year and to examine their labour supply on that basis. The specified weeks of work categories for the previous year are 0, 1-26, 27-47, and 48+ weeks. They find that when the data is used in a purely cross-sectional manner, without controlling for previous labour supply, female labour supply is negatively related to the number of children a woman has had. However, they find evidence that when weeks of work in the previous year are conditioned for, female labour supply is positively related to the number of children a woman has had. 3.3 DATA AND PRELIMINARY STATISTICS Two survey databases are used to conduct this research. The first is the Labour Market Activity Survey (LMAS, 1988) public use micro-data file. The second is the Canadian National Child-care Survey (CNCCS, 1988) public use micro-data file. The CNCCS was conducted as an addition to the Labour Force Survey, between September 19 and October 29, 1988. It includes data on child-care costs, detailed family composition, education, reference week work behaviour for husband and wife, and 1987 total annual incomes for each husband and wife. Unfortunately, in 1988, the Labour Force Survey questionnaire did not request wage information which is crucial to the estimation of structural probability of working estimates. Since wage variables used as regressors in labour supply equations suffer from endogeneity problems, researchers commonly replace actual wages with predicted wages derived from a prior wage equation estimate. In this research, a wage equation is first estimated using the LMAS data, and the coefficients thereby obtained are used to form a predicted hourly wage for each observation used from the CNCCS. The analysis is restricted to married women, twenty years old or more, with spouse present, who have worked at some point in the last five years, and whose youngest child is between the ages of 5 and 35 months inclusive. A number of other observations were deleted from the CNCCS data: women with children with a long-term condition or health problem or with children whose health 75 limits work opportunities or child-care arrangements, women who are permanently unable to work, women with chronic health problems, part-time students, unpaid family workers and those who were not available for work because of school. Women in households where the male is the 'designated adult' are also excluded.68'69 The variables used in the estimation fall into several groups: categorical variables for age, education, other income, immigrant status, provinces and regions, rural area, presence of teenager(s), presence of other adult, and husband's flexibility of hours. There are also categorical and numerical variables to represent child status. The LMAS data includes a variable for total family income and also a variable for earnings of the mother; it is therefore possible to calculate other income by subtracting earnings of the mother from total family income. This information is not available in the CNCCS public use micro-data file and the husband's total 1987 income is used as a proxy for other 1988 income. The CNCCS has very detailed data on children's ages including the age of the youngest child in months.70 This variable is used in the structural work equation to derive estimates of the probability that a woman will be at work as her youngest child ages. Except for older women in their 50's and 60's, it is expected that women's labour supply will increase as they age: as younger women age they are more likely to have work experience and thus have higher wage rates. More educated women are also expected to command higher wages rates. Hence, in a reduced-form equation where the wage rate is not available but age and education are, the probability of working is expected to increase with age and education. Conversely, women who have had several children are more likely to have taken time out of the labour force for child rearing and would thus have lower wage rates and a lower probability of being at work. Immigrant status may adversely affect the probability that a woman works, particularly if the immigrant is not fluent in either one of the official languages. Two separate immigrant status 6 8 The designated adult is the person most responsible for making child-care arrangements. 76 variables have been defined, one for immigrants who speak at least one of the official languages, and one for those who do not. Labour supply theory predicts that an individual is less likely to work with increases in unearned income. Here it is assumed that a married woman decides whether to work after having observed her husband's labour supply and income. The husband's income is therefore treated as unearned income. It is expected that a woman with more or with younger children is less likely to work. Younger children require more attention, as do larger numbers of children. When child-care costs are not controlled for, coefficients on children variables also pick up the child-care costs effects on labour supply. Provincial dummy variables control for regional variation in job opportunities. Additional adults or teenagers may provide free child-care services, which should result in a higher probability of being at work. Similarly, if the husband has flexible work hours, he may also be able to provide child-care. Table 3.1 shows means of dependent and independent variables for the sample data used in the estimations. This information is presented for occupation ranking and work history categories. Occupation categories include (1) professionals, (2) technicians and semi-professionals, supervisors and middle managers (TSM), (3) skilled, and (4) semi-skilled and unskilled. Work history categories are based on how many weeks the woman worked from October 1, 1987 to September 30, 1988. Weeks worked refers to how many weeks the woman was actually at work, rather than the number of weeks she was employed. The questions are formulated in this manner because the study is about child-care arrangements while parents are at work or school. Women who worked up to half a year and women who worked between 27 and 42 weeks in the previous 12 months are each separated into 6 91 also estimated the same model, omitting women who had at least one of their children spend time at work with them during the reference week; some of these results are presented in Table 3.4. 77 two groups, one for women whose youngest child is between 5 and 15 months of age, and one for women whose youngest child is between 16 and 35 months of age. Women with very young children could have been on a short-term maternity leave in the previous year and are potentially different from those with an older youngest child. The 43 - 52 weeks category is meant to include women who worked continuously but who may have taken regular or accumulated vacation or a short-term maternity leave, as well as teachers who may have as much as 10 weeks off per year. Of the women in the sample, 55 percent were at work in the reference week. This percentage varies considerably within categories however, with 91 percent of those who worked 43-52 weeks in 87/88 and 68 percent of professionals at work, and only 23-28 percent of those who worked 0-26 weeks in 87/88 and 47 percent of the semi- and unskilled at work. Table 3.1 also shows that the probability of paying for care for working women varies between 55 percent and 84 percent. Professionals and women who worked more than half a year in the previous 12 months are the most likely to pay for child-care. Average hourly cost of care for those who pay for care is $ 1.80 with the highest costs for professionals. In the occupation categories, professionals are older, have the most education, and are more likely than women in the other groups to live in Ontario. In the weeks worked in 87/88 categories, those who worked the full year are older and those who worked 27-42 weeks and had a very young child have the most education. This level of detail about the children's age is not available in the L M A S data. 78 TABLE 3.1 Means of Dependent and Independent Variables Variables A l l Occupations Weeks W o r k e d Prof . Semi- Semi-sk. Prof. S k i l l e d unsk. 43-52 27-42 0-26 all baby no baby baby no baby N 4,901 504 1,039 1,315 2,043 1498 381 318 1287 1417 Mother at work 55% 68% 62% 55% 47% 91% 91% 75% 28% 23% Pays for care/at work 69% 84% 79% 67% 57% 73% 72% 71% 61% 55% Hourly cost of care/payers $ 1.80 2.46 2.00 1.61 1.38 1.74 1.93 1.76 1.91 1.80 Mother's age 20-24 15% 4% 10% 12% 23% 8% 16% 11% 23% 16% 25-34 71% 70% 74% 76% 67% 73% 69% 74% 69% 71% 35 + 13% 26% 16% 12% 10% 18% 15% 15% 8% 12% Mother's education Did not complete high school 3% 0% 1% 0% 6% 2% 2% 1% 3% 4% Completed high school 51% 14% 31% 58% 69% 48% 43% 54% 55% 54% Some post-secondary 10% 6% 9% 13% 10% 11% 9% 7% 10% 11% Post-secondary diploma 20% 17% 35% 21% 11% 23% 25% 17% 18% 17% University degree 16% 62% 24% 7% 4% 15% 20% 21% 15% 14% Children/wage costs Number of children less than 17 1.79 1.7 1.8 1.7 1.8 1.8 1.7 1.9 1.7 1.8 Age of youngest in months 18.32 17.8 18.2 18.1 18.7 22.2 9.6 24.5 9.6 24.1 More than one child < 6 43% 45% 43% 44% 43% 41% 39% 43% 47% 44% Number of children 6-9 0.21 0.18 0.22 0.18 0.24 0.24 0.17 0.29 0.14 0.25 Estimated wage costs - 0-2 $ 2.06 2.09 2.07 2.04 2.07 1.95 2.22 1.90 2.28 1.95 Estimated wage costs - 3-5 $ 0.42 0.42 0.42 0.42 0.41 0.44 0.27 0.46 0.35 0.49 Husband 1987 annual income $0 - 10,000 7% 7% 7% 6% 9% 7% 6% 4% 9% 7% $10,001 - $20,000 19% 8% 16% 20% 24% 21% 21% 18% 19% 17% $20,001 - $30,000 30% 28% 30% 29% 31% 32% 30% 30% 27% 29% $30,001 - $40,000 25% 27% 25% 27% 22% 21% 25% 30% 27% 25% $40,001 - $50,000 11% 14% 13% 11% 9% 10% 12% 12% 11% 12% More than $50,000 8% 16% 10% 8% 4% 8% 5% 6% 7% 10% Spouse hours not flexible 71% 69% 69% 71% 72% 73% 78% 74% 68% 68% Teenage present 4% 2% 3% 3% 5% 5% 4% 4% 2% 3% Adult present 6% 4% 5% 6% 8% 7% 6% 5% 4% 7% Immigrant - with Eng. or Fr. 6% 7% 8% 5% 5% 7% 7% 3% 5% 6% Immigrant - no Eng. or Fr. 8% 9% 6% 5% 10% 8% 6% 11% 7% 8% Farming residence 2% 2% 1% 3% 1% 3% 1% 2% 1% 1% Provinces British Columbia 11% 9% 11% 12% 12% 10% 8% 11% 13% 13% Alberta 11% 9% 13% 12% 10% 11% 10% 12% 12% 10% Saskatchewan 4% 3% 4% 4% 4% 4% 5% 5% 4% 4% Manitoba 4% 3% 4% 4% 4% 4% 3% 5% 4% 4% Ontario 38% 46% 40% 34% 36% 40% 45% 33% 34% 37% Quebec 24% 21% 20% 26% 25% 24% 24% 26% 24% 22% New-Brunswick 3% 2% 2% 3% 3% 2% 3% 3% 3% 3% Nova Scotia 3% 5% 4% 3% 3% 4% 2% 3% 4% 3% Prince Edward Island 1% 0% 0% 1% 1% 0% 0% 0% 1% 1% Newfoundland 2% 2% 2% 2% 3% 1% 1% 2% 2% 3% Predicted wage $ 11.51 14.54 12.44 11.16 10.35 11.81 11.93 11.87 11.17 11.26 Worked part-time in 87/88 21% 22% 26% 19% 20% 31% 24% 33% 15% 13% Worked mixed-time in 87/88 8% 5% 9% 10% 8% 11% 15% 14% 5% 5% Worked full-time in 87/88 38% 49% 38% 40% 34% 58% 61% 53% 27% 15% 79 Estimated wage costs is an hourly cost of care estimated for two children age groups and is calculated as follows: legislated provincial staff-child ratios for centre care for the 0-2 and 3-5 age categories are multiplied with average provincial wages for centre care. The result is multiplied by the number of children each woman has in the 0-2 and 3-5 age categories.71 These variables are used in the estimation of child-care costs. Staff/child ratios and wages which vary by province, and the number of children a woman has in each of these two age groups, are the major determinants of costs of care. Other variables such as family income and/or education of the parents also have an impact on costs of care as indicated by the actual cost of care professionals pay relative to other groups, but the analysis focuses on quality-adjusted cost of care. Hence income and education are not included as explanatory variables in the cost of care equation.72 7 1 Provincial centre wages were available from a survey of centre care conducted in 1991 and used as proxies for 1988 wages. See "Caring for a Living" (1991). 7 2 Chapter 2 presents evidence that higher income families use higher quality care in terms of children's cognitive achievements. 80 3.4 M O D E L A N D E S T I M A T I O N P R O C E D U R E Given their husbands labour supply and income, mothers are assumed to maximize utility by choosing whether (and how much) to work, and whether (and how much) to pay for child-care given if they work. Although child-care quality (and thus cost if they are related) can be considered a choice variable, this is not taken into consideration here since no information about quality is available in the data. It is assumed that families face a certain market average cost/quality of care, which will affect whether the mother works.73 More specifically, EP = <P(W,C,L,s), with 8 ~ NtO.cj2) EP is the probability that a mother is working, W is the wage, C is the (net) market cost of care and L is a vector of explanatory variables including mother's education, and experience (or age), child status variables, and other family income. EP can be estimated using a probit equation. In order to generate estimates that can be used for policy analysis, we want to estimate structural parameters for child-care costs. However, since cost and wage data are available only for those who work and/or pay, it is not possible to estimate unbiased structural probability of working parameters at the outset. The econometrics procedure used to address these issues is similar to the one used by Connelly (1992), Cleveland, Gunderson and Hyatt (1996), and Powell (1997), and other researchers in this area. In addition, because of the lack of wage information in the CNCCS, the LMAS was used in conjunction with the CNCCS to construct a predicted wage. The following steps describe the procedure used: Although some mothers may report no cost of care, it is likely that indirect costs of care are incurred for small children. 81 (1) Estimate a reduced-form probit on the probability of working,74 using LMAS data. Whether a mother works is dependent on a vector of variables X. Let I represent an index function, where mother works if I > 0, and mother does not work if I < 0. I = Xp + 8, where s ~ N(0,CT2) EP = <D(XpVo), where O represents the cumulative standard normal. (2) Retrieve a selection term (the inverse mills ratio) from (1); using data on workers only, add the term as a regressor in a log wage equation and estimate with OLS: Ln W = Zcp + o"A.o where: Xo = (j)(Xp/a)/0(XB/a) where § is the standard normal distribution. (3) Use the coefficients obtained in (2) to estimate a wage for CNCCS workers and non-workers alike, where the Z variables in the CNCCS have been chosen to (precisely or approximately) match the Z variables in the LMAS: W(hat) = exp(Zcp(hat)) (4) Estimate a bivariate probit on the probability of working and paying for care using CNCCS data; since paying for care is dependent on working, a model which includes a selection adjustment on the pay for care equation is used: Prob(I pays n I works) = Prob(I pays 11 works)*Prob(I works) The model is: Zi = Qn + 8 i , where Si ~ N(0,cri2), governs the pay for care equation; z 2 = V5 + 8 2 , where e2 ~ N(0,CT22), governs the probability of working equation; 7 4 Powell (1998) and Cleveland, Gunderson, and Hyatt (1996) look at the probability that a woman is at work (works some hours) in the reference week. I do the same here. The unemployed, the not-in-the labour force 82 Sj, s2 ~ F(0,0,l,l,p), where F is bivariate normal, and p = Gn^i<^2-An individual works if > 0, and an individual pays if ZQ and z-,\ are both > 0. The likelihood function for this model is: Sw, p lnF2[Q7r ,VS, p] + Sw, n p lnF2[-Q7c ,V8, -p] + S n w lnO[-V5] where "w" means works, "p" means pays, "nw" means does not work, and "np" means does not pay. (5) Retrieve two selection terms from (4); using those who work and pay only, add the terms as regressors in a cost-of-care equation and estimate with OLS: C = X a + 9)A,i + e2A,2 + £ 3 , where e3 ~ N(0,CT 3 2), A , = {<K-Qn)«>[(-V5) - P(K)]/(1 - p2)1 / 2}/ F(Q7r,VS,p), and ^ 2 = {(f)(-V5)0[(-Q7r) - p(S)]/(l - p 2) , / 2}/ F(Q7t,V5,p) 7 5 where X represents a vector of variables thought to affect C, the net hourly costs of care. The cost of care thus estimated is the predicted net hourly cost of care.76 This is the after tax hourly cost of care that would be paid if the family paid for care. (7) Use W(hat), C(hat) and estimate a structural probit for the probability of working: EP = 0(W,C,L,e), with s ~ N(0,a 2) and where W is the wage, C is the net cost of care and L is a vector of explanatory variables. (8) Repeat 4-7 for each additional specification.77 and women who had jobs but were not at work (i.e., women on leave of any kind) were treated as not at work. 7 5 See Greene (1995), sections 22.3 and 28.5. 7 6 Parents reported total weekly (gross) cost of care and whether specific costs incurred for each of the care arrangements for each of their children were to be claimed on a tax return. Net costs of care have to be estimated. To do this, I use simplifying assumptions. I first assume that if the costs of the primary supplemental care arrangement for the youngest child are claimed, then total reported costs are claimed. To calculate tax savings, I use the total provincial and federal marginal tax rates in effect in 1988, excluding surtaxes, and apply them to the lower of total costs and the child-care expense deduction ceiling for the family. If the mother usually works more than 30 hours per week, I assume that her marginal tax rate is in the middle tax bracket. If she works less than 30 hours per week, I assume that her marginal tax rate is in the lowest tax bracket. For Quebec residents I apply tax rate of 46.5 percent and 34.7 percent respectively for these two groups. Net hourly cost is a cost that takes into consideration the hypothesis that parents who do not claim child-care costs pay less for their care. Costs that are not claimed are usually not claimed because the care 83 An additional question that is addressed in this research is the issue of heterogeneous preferences or habit formation, as discussed previously. In particular, the question to be investigated is whether taking into account heterogeneous preferences affects the estimated impact of child-care costs on labour supply. The appendix describes a simple variant of the inertia model developed by Nakamura and Nakamura (1985a). I estimate a model that uses prior weeks worked as an explanatory variable in the labour supply equation to account for heterogeneous preferences. I also estimate a model that uses current/prior occupational status to account for heterogeneous preferences. The hypothesis here is that as a woman moves up the occupational hierarchy, her commitment to the labour force is stronger. I also estimate specifications that do not include either occupation or past work history as explanatory variables in the structural probit equation for comparison purposes. 3.5 ESTIMATION RESULTS AND DISCUSSION (a) Reduced-form work probit and predicted wage The LMAS data is used to estimate a log wage equation.78 The wage rate is the average wage a person earned in 1988 from all paid jobs.79 A person worked if hours of work during week 42 of 1988 were greater than zero. The parameters from the wage equation are used to get a predicted wage for each observation of interest in the CNCSS data set. The results of this estimation, which consist of steps 1 and 2 outlined in section 3.4 are presented in Table 3.2.80 The results consist of probit coefficients for a reduced-form probability of working equation and selection-corrected OLS coefficients for a log wage equation. The number of children aged 0-2, 3-5 and other 1988 income provider does not provide receipts. If the child-care market is competitive, then these parents must be compensated for their losses of tax savings through lower fees. 7 7 One selection-corrected regression is used to estimate wages for women with children less than three years of age. 7 8 This equation is estimated for married mothers with children under the ages of two. 7 9 Workers for whom an average wage rate is not available are not included. 8 0 Except for the structural labour supply equation, standard errors for selection corrected regression have been adjusted. 84 are included in the reduced-form probit but not in the wage equation. While these variables are expected to affect the reservation wage, they are not expected to affect the observed wage. Of the 2,561 women in this sample, 1,245 (48.6 percent) worked at some time in 1988. This probability of working is lower than the CNCCS rate of 55 percent.81 The predicted wage from Table 3.2 is highest for the age 35 and older category; it is higher for those with a post-secondary diploma and higher still for those with a university degree, and is lower for residents of prairie and Atlantic provinces. The number of children less than 16 years old is associated with a reduction in the probability of working and in the wage rate. The number of young-children coefficient is not significant indicating that the number of young children does not decrease the probability that a woman is at work in a particular week. While this finding may be counterintuitive, note that the effect of children is already captured by the children less than 16 variable and that all of the women in the sample have a child less than three years old. The coefficient on the selection term is negative but not significant. A positive coefficient on the selection term would indicate that working-women have a higher predicted wage than non-working women and vice-versa. The lack of significance on this coefficient indicates that selection is not an issue and that working and non-working women have similar predicted wages. 1 In the CNCCS sample, women with babies and women who did not work in the last five years were excluded. 85 TABLE 3.2 Reduced-form Work Probit and Log-Wage Equation Coefficients Work Log-Variables Probit t-stat Wage t-stat N 2,561 1,245 Constant -0.321 -2.27 2.109 10.63 Mother's age (20 - 24) 25 - 34 0.285 3.71 0.261 5.22 35 + 0.437 4.02 0.401 5.71 Mother's education (Did not complete high school) Completed high school 0.486 6.46 0.089 1.19 Some post-secondary 0.488 5.09 0.065 0.82 Post-secondary diploma 0.632 7.63 0.279 3.13 University degree 0.755 8.30 0.498 4.87 Number of children less than 16 -0.166 -3.83 -0.054 -1.86 Immigrant, speaks French or English -0.088 -0.81 -0.051 -1.03 Immigrant, other -0.108 -1.04 -0.017 -0.35 Provinces British Columbia -0.230 -2.70 0.039 0.81 Alberta -0.069 -0.78 -0.027 -0.69 Manitoba/ Saskatchewan 0.017 0.17 -0.088 -1.96 Quebec -0.194 -2.81 0.002 0.05 Atlantic -0.211 -2.12 -0.121 -2.39 No. of children aged 0-2 -0.025 -0.30 -No. of children aged 3-5 -0.054 -0.85 -Other 1988 annual income $0 - 10,000 $10,001 - $20,000 -0.007 -0.06 -$20,001 - $30,000 0.194 2.01 -$30,001 - $40,000 0.144 1.75 -$40,001 - $50,000 0.126 1.46 -More than $50,000 -0.066 -0.69 -Presence of teenager -0.202 -0.94 -Lambda - -0.103 -0.51 Rsquared - 0.2846 Log-likelihood -1,667.9 86 (b) Reduced-form bivariate employment, pay for care probit and cost equation The next steps in the procedure (steps 4 and 5) consists of estimating a reduced form bivariate probit for the probability of working and of paying for care, and a cost of care equation. The results for this are shown in Table 3.3. The specification labeled 'occupations' refers to the model that uses standard labour supply explanatory variables in the work equation augmented by current/prior occupational status.82 I divided occupation status into four groups: (1) professionals, (2) technicians and semi-professionals, supervisors and middle managers (TSM), (3) skilled, and (4) semi-skilled and unskilled. I also divided women into five groups according to weeks worked from October 1, 1987 to September 30, 1988.83 Women who worked up to half a year and women who worked between 27 and 42 weeks were each separated into two groups, one for women with a child between 5 and 15 months of age, and one for women whose youngest child is between 16 and 35 months of age. The last group includes women who worked between 43 and 52 weeks. The specification labeled "weeks worked" includes dummy variables for these groups as explanatory variables in the work equation. In the 'occupations' specification, the work probit coefficients have the expected signs, except for the coefficient on the 'spouse hours not flexible' variable. Here it was expected that a spouse with flexible hours would be available for child-care duties, which would facilitate mother's labour force participation. It is assumed that the husband's hours of work and timing for work are pre-determined before the wife decides whether and when to work. As with the LMAS findings, labour supply does not decrease with the number of young children, although it does decrease with the total number of children. In the 'weeks worked' specification, the work probit coefficients on weeks worked are highly significant and generally reduce the significance of other explanatory variables relative to the 87 'occupations' specification. For example, only a post-secondary diploma or university degree has a significant positive impact on labour supply relative to the 'did not complete high school' reference category, compared to the occupations specification where completing high school also has a significant positive effect. These results support the habit persistence or heterogeneous preferences hypotheses. There is one exception to this general finding of reduced significance for explanatory variables other than occupations and weeks worked: the effect of having more than one child less than six years of age is not significantly negative in the occupations specification, but it is in the weeks worked specification. Hence, the continuity of work habits is less strong when women have more young children. Rho is negative and significant in both specifications, which indicates that mothers who are less likely to pay for care if they work are also more likely to work. This finding is expected: if child-care costs can be thought of as reductions in wage rates, the absence of costs results in a higher wage, which in turn increases the probability that a woman will be receiving her reservation wage.84 The correlation is particularly high in the occupation specification, but not so high in the weeks worked specification. Again, this is also to be expected: a woman who has already demonstrated some attachment to the labour force as indicated by the weeks she worked in the last year has already adjusted to child-care costs if any. The pay for care probit coefficients indicate that families are more likely to pay for care if they have more of the older 'young children',85 and they are less likely to pay for care if they have more children between the ages of 6 and 9. For women who are not currently working, but who have worked in the last five years, prior occupational status is included. 8 3 The reference week is between September 11 and October 29, 1988. 8 4 The reservation wage for an individual is the minimum wage she would accept to work. 8 5 Although child-care staff wages and staff child ratios affect estimate wage costs, the most important determinant of these estimated costs is the number of children in the age group. 88 TABLE 3.3 Bivariate Work and Pay for Care Probit and Cost Equation Coefficients Bivariate Work and Pay for Care Probit and Cost Equation Coefficients Variables Occupations t-stat Weeks Worked t-stat N 4901 4901 Work Probit: Constant -0.745 (-5.49) -1.293 (-6.47) Mother's age 25 -34 0.191 • (3.90) -0.008 (-0.12) 35 + 0.492 (6.68) 0.227 (2.41) Mother's education Completed high school 0.375 (3.60) 0.147 (0.99) Some post-secondary 0.496 (4.31) 0.235 (1.45) Post-secondary diploma 0.556 (4.98) 0.296 (1.92) University degree 0.596 (5.05) 0.567 (3.62) Mother's current or last occupation Professional 0.472 (6.76) T S M 0.354 (7-15) Skilled 0.159 (3.71) Weeks mother worked in 87/88: 43-52 2.089 (33.45) 27-42, age of youngest 5-15 months 2.141 (19.70) 27-42, age of youngest 16-35 months 1.410 (15.45) 0-26, age of youngest 5-15 months 0.318 (4.21) Children Number of children less than 17 -0.047 (-0.87) 0.108 (1.55) Age of youngest child in months 0.011 (5.58) 0.008 (2.01) More than one child < 6 -0.078 (-1.17) -0.197 (-2.27) Number of children 6-9 0.074 (1.05) -0.059 (-0.64) Husband 1987 annual income: $0-$10,000 -0.196 (-2.47) -0.064 (-0.72) $20,001 - $30,000 -0.035 (-0.65) 0.089 (1.41) $30,001 - $40,000 -0.288 (-4.97) -0.105 (-1.53) $40,001 -$50,000 -0.283 (-3.89) -0.045 (-0.53) More than $50,000 -0.380 (-4.65) -0.112 (-1.18) Spouse hours not flexible 0.199 (4.83) 0.091 (1.93) Teenage present 0.323 (2.41) -0.054 (-0.31) Adult present 0.206 (2.70) 0.123 (1.38) Immigrant - with English or French 0.010 (0.13) -0.014 (-0.17) Immigrant - no English or French -0.237 (-3.31) -0.264 (-2.85) Farming residence 0.724 (5.18) 0.599 (3.20) Provinces: British Columbia -0.175 (-3.14) -0.108 (-1.55) Alberta -0.132 (-2.30) -0.136 (-1.79) Manitoba & Saskatchewan -0.037 (-0.55) 0.023 (0.27) Quebec -0.069 (-1.51) -0.102 (-1.69) Atlantic -0.130 (-1.99) 0.006 (0.07) Rho -0.935 (-16.72) -0.415 (-7.13) 89 T A B L E 3.3 (continued) Variables Occupations t-stat Weeks Worked t-stat N 4901 4901 Pay for Care Probit: Constant 0.860 7.52 0.842 5.20 Children/wage costs Number of children 6-9 -0.097 -2.10 -0.088 -1.63 Estimated wage costs - 0-2 -0.037 -1.17 -0.088 -2.10 Estimated wage costs - 3-5 0.090 2.46 0.067 1.50 Husband 1987 annual income: ($10,001 -$20,000) $0 - $10,000 -0.102 -1.13 -0.183 -1.64 $20,001 - $30,000 0.094 1.47 0.144 1.95 $30,001 - $40,000 0.128 1.90 0.082 1.04 $40,001 - $50,000 0.171 2.00 0.221 2.22 More than $50,000 0.190 1.99 0.198 1.77 Spouse hours not flexible -0.035 -0.70 0.031 0.52 Teenage present -0.238 -2.21 -0.202 -1.61 Adult present -0.472 -5.77 -0.544 -5.67 Immigrant - with English or French -0.011 -0.12 0.032 0.31 Immigrant - no English or French -0.240 -2.87 -0.412 -4.21 Farming residence -0.664 -5.22 -0.563 -3.88 Worked part-time in 87/88 -0.109 -1.30 -0.374 -3.11 Worked mixed-time in 87/88 0.110 1.16 -0.086 -0.65 Worked full-time in 87/88 0.406 4.76 0.248 2.05 Los Likelihood -4699 -3733 Cost of Care Equation: N 1798 1798 Constant 1.243 104.82 0.743 62.07 Children/wage costs Number of children 6-9 0.136 20.64 0.208 31.59 Estimated wage costs - 0-2 0.415 83.70 0.360 71.28 Estimated wage costs - 3-5 0.545 109.64 0.502 100.35 Adult present -0.484 -32.57 -0.427 -28.15 Immigrant - with English or French 0.224 19.29 0.274 23.25 Immigrant - no English or French -0.200 -15.21 -0.334 -24.70 Lambda - Pay for Care Equation 0.668 40.42 0.361 19.41 Lambda - Work Equation -0.961 -70.94 -0.136 -18.68 Adjusted R squared 0.098 0.071 90 Families are also more likely to pay as the husband's income increases, if there is no teenage child or other adult present in the home, and if the mother worked full-time in 87/88. Families are less likely to pay for care if the mother is an immigrant who cannot speak either English or French. These results hold for both specifications with some variation in significance. Cost of care is defined as hourly cost of care. Estimated wage costs are based on 1991 child-care worker wages times provincial staff-child ratios in effect for the two age groups in 1988.86 The presence of an adult in the home is expected to reduce child-care costs as this adult might be expected to provide child-care for the children for a lower fee than is available elsewhere. The coefficients on estimated wage costs can be interpreted as follows: a coefficient of 0.4 indicates that on average, net child-care costs approximate 40 percent of a child-care worker's wage cost. Parents do not pay the full cost because many parents use care arrangements that cost much less than centre care, because governments subsidize a portion of centre care costs, and because net costs take into consideration the tax savings for those who claim their child-care costs. Note, however, that the magnitude of the constant term is fairly substantial relative to these coefficients. Hence, parents may still pay costs that are close to or even more than estimated wage costs, but the relationship is not fully explained by these variables. The coefficients on the lambda terms indicate that those with higher predicted costs of care are more likely to pay for care and that those with lower predicted costs of care are more likely to work. These results are sensible: if the cost of child-care is high, it is unlikely that someone else will provide it for free and as argued previously, if it is low, the likelihood that the reservation wage will be met increases. Staff child ratios were obtained from Pence, A.R., et. al. (1997). 91 (c) Elasticity of work with respect to wages and predicted cost of care Table 3.4 reports child-care costs and wage elasticities evaluated at the means of the S"7 fift RQ explanatory variable for some of the estimated specifications. ' ' Specification I allows the cost of care impact to vary with the categories. Specification II is a restricted version of specification I, where cost of care coefficients are restricted to be the same for all categories of women. Finally, specification III does not include occupation or weeks worked categories.90 Elasticities show the percentage increase in the probability of working from a 1 percent increase in the explanatory variable. 8 8 Child-care costs elasticities for weeks worked specifications are either not significant or not in the expected direction. 8 9 Structural coefficients for the base sample are reported in tables 3.5A and 3.5B. 9 0 However, the cost of care for each of the occupation specifications was derived from the same reduced-form bivariate probit which includes occupation categories in the work equation. Similarly, the cost of care for each of the weeks worked specifications is derived from the same reduced-form bivariate probit which includes weeks worked categories in the work equation. 92 TABLE 3.4 Child Care Costs and Wage Elasticities Occupation Weeks Worked Models Models dPW/dX Xbar Elasticity dPW/dX Xbar Elasticity BASE SAMPLE (N=4901) Specification I Predicted costs Professional -0.064 2.32 -0.22 T S M -0.116 * -0.44 Skilled -0.071 * -0.30 L o w skilled -0.092 * -0.45 Predicted wage 0.023 * 11.51 0.48 0.021 * 11.51 0.44 Specification II Cost - A l l -0.091 * 2.32 -0.38 -0.037 1.70 -0.11 Predicted wage 0.023 * 11.51 0.48 0.021 * 11.51 0.44 Specification III Cost - A l l -0.094 * 2.32 -0.40 -0.099 * 1.70 -0.31 Predicted wage 0.033 * 11.51 0.68 0.033 * 11.51 0.68 CARE FOR A CHILD AT WORK EXCLUDED (N=4412) Specification I Predicted costs Professional -0.082 2.44 -0.30 T S M -0.153 * -0.63 Skilled -0.128 * -0.61 L o w skilled -0.184 * -1.12 Predicted wage 0.020 * 11.52 0.46 0.023 * 11.52 0.52 Specification II Cost - A l l -0.160 * 2.44 -0.77 -0.037 1.65 -0.12 Predicted wage 0.020 * 11.52 0.46 0.021 * 11.52 0.48 Specification III Cost - A l l -0.163 * 2.44 -0.79 -0.184 * 1.65 -0.60 Predicted wage 0.035 * 11.52 0.80 0.035 * 11.52 0.80 * Statistically significant at the 10% level using a one-tailed test. In this table, I have also included child-care costs and wage elasticities for the same equations estimated for a subset o f the main sample. In this sub-sample, I excluded women who took care o f children while at work during the reference week. I excluded these for comparability with other 93 Canadian research.91 I also found interestingly different results on the magnitude of the effect of child-care costs when this group is excluded. Overall child-care costs elasticities for the main sample are -0.38 in the occupations specification and -0.11 and are not statistically significant in the weeks worked specifications. Wage elasticities increase from 0.48 and 0.44 to 0.68 when occupation and weeks worked explanatory variables are not included in the structural probit. Note that wages are estimates only and that they are not constructed with occupation or weeks worked information, and that these variables are likely correlated with actual or potential wages. In the base sample, child-care cost elasticities are not statistically significant for professional women, but are statistically significant for all of the other groups. When women who care for a child at work are excluded, the difference between the elasticities of the various occupational groups of women is much larger: elasticities for professional women are slightly higher but still not significant, but elasticities for low skilled women are almost four times that of professional women compared to two times when the base sample is used. Elasticities for low-skilled women rise from -0.45 using the base sample to -1.12 when women who care for a child at work are excluded. In other words, taking care of children while at work somewhat facilitates the labour supply of low skilled women and reduces their labour supply responsiveness to child-care costs. The overall cost elasticity is -0.38 using the base sample, but increases to -0.77 when women who care for a child at work are excluded. In specification III, occupations or weeks worked are not included as control variables in the structural probit for comparison purposes. Child-care costs elasticities are -0.79 and -0.60 for the sub-sample that excludes women who care for a of child at work are quite a bit higher than those reported by Powell (1997) who reports child-care cost elasticity -0.38 and Cleveland and Hyatt (1996) who report child-care costs elasticities of-0.39. Powell, and Cleveland and Hyatt also use the CNCCS data. However, they include married mothers with children less than six years of age, 9 1 Cleveland, Gunderson and Hyatt (1996) and Powel l (1997) exclude this group o f mothers. 94 mothers with young babies,92 and most mothers who did not work for the last five years.Vi When these latter two groups, who are the most likely to stay home and to not be largely influenced by child-care costs, are excluded, elasticities should be higher. Barrow (1999) who uses a sample of first birth mothers from the US National Longitudinal Survey of Youth who were employed prior to the first birth finds child-care cost elasticities of-0.18. I have similar findings not reported here for mothers with one preschool child in the occupation models, where I find an elasticity of -0.22.94 Barrow does not control for weeks worked in the previous year. Overall child-care cost elasticities are not affected by the inclusion of occupational status variables in the structural probit, but they are substantially reduced when weeks worked in the previous year is included as an explanatory variable in the bivariate work and pay probit and also in the structural work probit. The inclusion of occupation or weeks worked controls substantially reduces wage elasticities. I find wage elasticities for the model without the inclusion of occupation or weeks worked controls of 0.80. These are quite close to what is reported by Powell (1997) and Cleveland and Hyatt (1996). These authors report wage elasticities of 0.85 and 0.81 respectively. However, when either occupation or weeks worked are included as controls, wage elasticities are substantially reduced to 0.46 and 0.48 respectively.95 Since a lot of these mothers will either be 'usual' non-workers or mothers on paid maternity leave, they will not likely be affected by child-care costs. 9 3 The labour supply of mothers who have little or no labour force attachment are also not likely to be affected by child-care costs. 9 4 In models where predicted child-care costs vary with the number of children in the family, limiting the analysis to families with only one child substantially reduces the variation available in the child-care costs variable. 9 5 Since both of these controls should be highly correlated with wages, this finding is consistent with expectations. 95 (d) Probability of working as a function of age in months of the youngest child and predicted hourly cost of care The following figures are based on the structural work probit results presented in Tables 3.5A and 3.5B for occupation and weeks worked models. The tables follow Figures 3.1-3.4. Figure 3.1 is based on specification I of the occupations specifications. It shows the impact of the age of the youngest child on the probability that a mother will have returned to work. The in-sample range is from five to thirty-five months. This probability is estimated to increase by 13-14 percentage points over the course of this 30-month span. Although the probability that a woman will be at work is sensitive to the child's age, the estimates indicate that for most women, the decision to return to work is implemented when the child is very young. Essentially, most of the women returning to work return after their maternity leave. FIGURE 3.1 1988 Employment Probability (Occupations -1): Married Mothers with Children < 3, by Age of Youngest Child in Months i.oo T 1 0.90 5 6 7 8 9 1 0 11 12 13 1 4 15 1 6 1 7 18 1 9 2 0 21 2 2 2 3 2 4 2 5 2 6 2 7 2 8 2 9 3 0 3 1 3 2 3 3 3 4 3 5 Aye of Youngest Child in Months 96 Continuing with occupations specification I, Figure 3.2 shows the effect that child-care costs have on the probability that a woman will be at work. This specification allows for differential child-care cost effects for women in the different occupational categories. Coefficients in Table 3.5 show that for professional women, the coefficient on the cost of care variable is not significant, but that it is significant for all of the other groups. The TSM women experience the steepest declines in probability of work as child-care costs increase followed by the low-skilled group. Recall from the previous section and table 3.4 that when women who care for a child at work are excluded, the low-skilled women's labour supply is by far the most responsive to child-care costs. FIGURE 3.2 1988 Probability of Being at Work (Occupations -1): Married Mothers with Children < 3, by Hourly Cost of Care 1.00 Hourly Cost of Care These findings are consistent with the hypothesis that different groups of women might have different preferences for work. Professional women are more career-oriented and few are deterred 97 from working by child-care costs. Additionally, child-care costs represent a small proportion of potential market earnings for women with professional skills. Figure 3.3, which is based on occupations specification II, shows the effect of child-care costs on the probability that a woman will be working. Here the effect of child-care costs is restricted to be the same for all occupational categories. The effect is negative and significant. This is to be expected since professionals consist of only 10.3 percent of the sample and since child-care costs are negative and significant for all the other groups of women. From the lowest to the highest cost of care there is a 25-27 point decline in the probability that a woman works. FIGURE 3.3 1988 Probability of Being at Work (Occupations - II): Married Mothers with Children < 3, by Hourly Cost of Care - Professional - T S M - Skilled - Semi-skilled or unskilled Hourly Cost of Care Figure 3.4, which is based on weeks worked specification II, shows the effect of child-care costs on the probability that a woman will be working. Here the effect of child-care costs is restricted 98 to be the same for all occupational categories. The effect is negative but not significant/6 The most important determinant of whether a woman is now working is the number of weeks she worked in the previous 12 months. The predicted probabilities are virtually identical for women who worked 43-52 weeks and for women who worked 27-42 weeks and had a very young child. Tables 3.5A and 3.5B show the structural probit coefficients for specifications that include occupation or weeks worked categories. The effect can be considered statistically significant if a one-tailed test is used. 99 FIGURE 3.4 1988 Probability of Being at Work (Weeks Worked - II): Married Mothers with Children < 3, by Hourly Cost of Care The occupations specifications indicate that cost of care has a significant negative impact on the probability that a woman is currently at work, except for professional women. The probability of working is sensitive to the age of the youngest child, but findings suggest that for women who work when their children are very young, their decision to work or not to work has been made by the time the youngest child is five months old. The husband's income has a depressing effect on the probability of working. Women whose husbands do not have flexible work hours are more likely to have returned to work. Except for women living in Manitoba or Saskatchewan, Ontario women are more likely to have returned to work. 100 TABLE 3.5A Structural Probability of Being at Work Probit Coefficients Occupation I II III Variables Coeff. t Coeff. t Coeff. t N 4190 4190 4190 Constant -0.344 -1.18 -0.344 -1.26 -0.445 -1.65 Mother's current or last occupation (Semi-skilled or unskilled) Professional 0.218 0.61 0.382 5.30 T S M 0.441 1.67 0.295 5.64 Skilled 0.067 0.27 0.189 4.11 Age of the youngest child in months 0.012 5.46 0.012 5.43 0.011 5.04 More than one child < 6 0.054 0.63 0.058 0.69 0.073 0.87 Number of children 6-9 0.114 2.68 0.117 2.74 0.114 2.69 Husband 1987 annual income: ($10,001 - $20,000) $0-$10,000 -0.165 -2.07 -0.163 -2.05 -0.165 -2.07 $20,001 - $30,000 -0.009 -0.17 -0.010 -0.18 0.003 0.07 $30,001 - $40,000 -0.253 -4.39 -0.252 -4.39 -0.239 -4.17 $40,001 - $50,000 -0.202 -2.81 -0.201 -2.81 -0.191 -2.68 More than $50,000 -0.318 -3.91 -0.318 -3.91 -0.282 -3.48 Spouse hours not flexible 0.192 4.71 0.191 4.69 0.184 4.53 Teenager present 0.327 3.18 0.327 3.18 0.301 2.95 Other adult present 0.083 0.86 0.081 0.83 0.061 0.63 Immigrant, speaks French or English 0.128 1.56 0.128 1.55 0.149 1.82 Immigrant, other -0.269 -3.60 -0.271 -3.63 -0.289 -3.88 Lives on a farm 0.656 4.44 0.659 4.47 0.674 4.59 Province British Columbia -0.227 -3.65 -0.227 -3.64 -0.248 -4.00 Alberta -0.195 -2.70 -0.197 -2.73 -0.186 -2.59 Manitoba/Saskatchewan 0.025 0.34 0.023 0.31 0.047 0.63 Quebec -0.188 -3.29 -0.188 -3.31 -0.198 -3.49 Atlantic -0.155 -1.83 -0.158 -1.87 -0.131 -1.55 Cost of care: -0.230 -2.08 -0.238 -2.16 Professional -0.162 -0.94 T S M -0.293 -2.16 Skilled -0.178 -1.33 Low skilled -0.231 -1.95 Hourly wage 0.058 6.66 0.057 6.63 0.082 10.80 Log Likelihood -2746 -2749 -2786 continued 101 TABLE 3.5B Structural Probability of Being at Work Probit Coefficients Weeks Worked I II m Var i ab l e s Coeff . t Coeff . t Coeff . t N 4190 4190 4190 Constant -0.824 -2.42 -0.960 -2.91 -0.334 -1.18 Mother's weeks worked in 87/88 (0-26) 43-52 1.358 5.08 2.148 34.86 27-42 - with baby 0.651 1.48 2.011 21.03 27-42 - without baby 0.375 0.91 1.349 13.87 Age of the youngest child in months -0.005 -1.68 -0.005 -1.76 0.011 4.68 More than one child < 6 0.031 0.26 0.099 0.84 0.162 1.61 Number of children 6-9 0.115 1.31 0.151 1.80 0.178 2.50 Husband 1987 annual income: ($10,001 - $20,000) $0 - $10,000 -0.069 -0.68 -0.072 -0.70 -0.155 -1.78 $20,001 - $30,000 0.051 0.73 0.044 0.63 0.027 0.46 $30,001 - $40,000 -0.130 -1.74 -0.140 -1.88 -0.236 -3.78 $40,001 - $50,000 -0.026 -0.29 -0.041 -0.44 -0.164 -2.13 More than $50,000 -0.066 -0.62 -0.061 -0.57 -0.255 -2.87 Spouse hours not flexible 0.021 0.40 0.032 0.60 0.143 3.26 Teenager present 0.085 0.60 0.114 0.80 0.292 2.55 Other adult present 0.067 0.50 -0.005 -0.04 -0.005 -0.05 Immigrant, speaks French or English 0.087 0.69 0.133 1.08 0.275 2.63 Immigrant, other -0.260 -2.33 -0.312 -2.84 -0.335 -3.66 Lives on a farm 0.219 1.01 0.244 . 1.13 0.144 0.77 Province British Columbia -0.031 -0.39 -0.037 -0.46 -0.185 -2.73 Alberta -0.123 -1.28 -0.159 -1.68 -0.227 -2.85 Manitoba/Saskatchewan 0.186 1.88 0.171 1.74 0.051 0.61 Quebec -0.024 -0.33 -0.052 -0.71 -0.111 -1.82 Atlantic 0.157 1.43 0.124 1.14 -0.073 -0.80 Cost of care: -0.263 -1.40 -0.470 -2.90 Worked 43-52 wks 87/88 0.133 0.54 Worked 27-42 wks 87/88 (baby) 0.489 1.54 Worked 27-42 wks 87/88 (no baby) 0.240 0.78 Worked 0-26 wks 87/88 -0.335 -1.71 Hourly wage 0.069 6.53 0.068 6.53 0.093 10.56 Log Likelihood -1804 -1815 -2786 In the weeks worked specifications, the most important determinant of whether a woman is currently working is how much she worked in the previous year. Coefficients on most of the other 102 explanatory variables are not statistically significant. If the weeks worked variable can be thought of as a proxy for preferences, then preferences have a very strong impact on a woman's decision to work. The findings also indicate that women's labour supply is highly predictable from past behaviour and that most women who return to work within 35 months of the birth of their children have done so by the time their maternity leave expires. Once past work behaviour has been taken into consideration, there are no child-care cost effects. These latter results should be interpreted with caution however, as the initial decision to return to work was likely affected by child-care costs, hence controlling for prior labour supply (which is in most cases after the birth of the youngest child) already controls for child-care costs effects. Information on prior work would be more useful and relevant in this type of analysis if it consisted of information about the mother's labour supply before the birth of the last child. These results coupled with the results presented in Table 3.4 suggest that while the probability that certain women with young children will be at work is not affected by child-care costs, it is strongly affected by these costs for certain other women and that in particular, the labour supply of low skilled women who cannot take care of their children while at work is highly sensitive to child-care costs. Looking back at Table 3.1 we can see that the low skilled represent 41.6 percent of women with children less than three. The above findings have clear policy implications. If married mothers' labour force participation is to be encouraged on equity grounds, subsidies to child-care will have a larger impact on labour supply if they are targeted to the lower income groups. Lower skilled women who do not have the opportunity to take care of their children while working would be the ones most likely to benefit from targeted subsidies. Lower skilled women who work and take care of their children while working could also choose to substitute paid care in lieu of their own (free) care while working. 103 3.6 C O N C L U S I O N In this chapter, I estimate the impact of child-care costs on the probability that married women with children less than three years old are at work. Women with infants and women who have not worked in the past five years are excluded from the analysis. Variables are included in the structural probit estimation of the probability of work to allow for heterogeneity of preferences. Phipps, Burton and Lethbridge (2001) show that the Canadian family gap is partially explained by women changing jobs after work interruptions. The authors do not find evidence that women who retain the same jobs after work interruptions experience decreases in pay. This points to the importance that the job held prior to a birth be resumed in order for wage rates to be maintained. This is unlikely to occur for women who stay home for several years after the birth of a child. The results in this chapter indicate that except for professional women, the labour supply of women with non-infant but very young children is quite responsive to cost of care. In the occupations specifications, cost of care elasticities range from -0.22 to -1.12 for the low skilled who did not take care of their children while at work. The results for low-skilled women are quite sensitive to whether women who take care of their children while at work are included in the analysis. This indicates that many low-skilled women have fewer child-care opportunities. If the main policy objective of child-care subsidies is to encourage the continued labour force participation of mothers, the findings in this chapter tend to support income-tested subsidies rather than across the board subsidies to child-care. 104 APPENDIX 3.1 Heterogneity Model for Work Preferences A very simple variant of the labour supply model as developed by Heckman (1974b) and reformulated by Nakamura and Nakamura (1985b) is as follows. Let the market wage for an individual be linearly determined as follows: (1) lnW i t =p 0 + Z i tp, + v i t where Z is a vector of explanatory variables for the market wage including experience and education and is v;t a white noise random variable. Let the marginal rate of substitution of leisure and income be linearly determined as follows: (2) In MRS i t= a 0 + Q i ta, + Xhit + 4>hjt., + p i t where Q is a vector of explanatory variables for the marginal rate of substitution, including family composition variables, non-labour income and other demographic characteristics, h i s hours of work at time t and where pu is a white noise random variable. The hjt_i variable is included to reflect person-specific tastes and preferences for work along with any other unobserved person-specific persistent effects. If hjt_i is omitted, person-specific effects will form part of the error term. Assuming correlation between the variables contained in Q and hit.i, the omission of h jt.i in the estimation procedure results in omitted variables bias. A person works if (3) lnW i t>lnMRS i t(h=o) or if (4) Po + ZuPi + v i t> a 0 + Qi tai + Xhit + <phit-i + Pit or if 105 (5) v i t - nit > (a 0 - B 0) + Q i ta, - Z i tp, + A,hit + <j>hiM Equation (5) can be estimated with a probit or a logit depending on the distributional assumption for the error term. 106 C H A P T E R TV P O T E N T I A L I N C O M E A N D T H E E Q U I T Y O F T H E C H I L D - C A R E E X P E N S E D E D U C T I O N 9 7 4.1 INTRODUCTION The existence and extent of the child-care expense deduction (CCED) available to Canadian taxpayers have been a source of controversy for some time. During the 1999 federal budgetary debates, opposition parties attacked the deduction for its apparent violation of horizontal equity. They argued that two-parent families with a stay-at-home parent were not entitled to the deduction and thus faced discrimination. Child-care funding supporters have also viewed the CCED less favourably than other forms of child-care funding, because it results in higher tax savings per dollar spent on child-care for higher income earners than for lower income earners. That is, the CCED is seen to violate vertical equity. In contrast with other Canadian research on the CCED, this chapter uses an empirical framework to examine how the CCED fares in terms of horizontal and vertical equity in taxation. To assist in the evaluation, I use data from the 1988 Canadian National Child-care Survey and the 1988 Labour Market Activity Survey to simulate tax benefits for Canadian families with children under the age of 7, based on parental earnings and child-care costs estimates. Benefits are simulated for 1988 and 1999. Estimated parental earnings and childcare costs for 1988 are "aged"98 to simulate benefits for 1999. Equity is evaluated using (predicted) potential and actual parental earnings as measures of ability to pay. Potential earnings99 are defined as the greater of a parent's predicted actual earnings and her predicted earnings based on a 37.5-hour workweek. I argue that potential earnings are a Chapter IV, with minor differences, has been accepted for publication by the Canadian Tax Foundation , to be published in the Canadian Tax Journal volume 49, number 3, 2001. The title of the article is the same title as used for Chapter IV. 9 8 The data is aged by translating 1988 earnings and costs in 1999 dollars using the Consumer Price Index as the weight. The tax laws in effect in 1999 are then applied to this aged data. There is an implicit (simplifying) assumption that both earnings and costs experienced average nominal growth. 9 9 Note that only earnings are estimated and that other income is excluded from the empirical analysis. 107 superior measure of ability to pay because, unlike actual income, they better describe families' opportunities by recognizing the value of leisure and household production. This chapter is structured as follows. In section 4.2 I describe how CCED rules have evolved since the inception of the deduction in 1972. In section 4.3 I adopt working definitions of vertical and horizontal equity and examine related literature. In section 4.41 examine how an increase in the ceiling on eligible expenditures in 1988 affected different income groups and the proportion of income that claimants in different income groups spent on child-care in 1987 and 1988. In section 4.5 I describe the procedure used to estimate annual child-care costs, and predicted and potential earnings. In section 4.6 I analyze how the CCED fares in terms of vertical and horizontal equity. Concluding observations are offered in section 4.7. 4.2 THE CANADIAN CHILD-CARE EXPENSE DEDUCTION The CCED is a deduction in the calculation of net income for tax purposes. To qualify for the deduction, a taxpayer must incur child-care expenses to permit the taxpayer or a supporting person of the child to pursue employment, business, training, or research activities.100 In effect, the deduction recognizes the expense necessary to earn income or engage in human capital investments, and thus reduces taxable income. The CCED was introduced in the 1972 taxation year. Until 1983, the deduction was available to women with earned income.101 Men could also claim the deduction if they were widowers, divorced, or separated (and also, starting in 1974, never married), or if the children's mother was incapable of caring for the children for any period during the taxation year. The deduction could be claimed by individuals with earned income, whether the income was from employment or from 1 0 0 For an excellent description of the current provisions under section 63 of the Income Tax Act see Gordon (1999). 1 0 1 In 2000, earned income includes employment income, self-employment income, the taxable part of scholarships, bursaries, fellowships, and similar awards, net research grants, and certain earnings supplements and disability benefits. Earned income does not include property income (dividends or interest income) or regular unemployment insurance benefits. 108 training in respect of which a grant was received. It was available for child-care expenditures for children under 14 years of age. The maximum claim was limited to the lesser of an amount per child times the number of eligible children (up to four children per family) and two-thirds of earned income. Per child and/or total family limits were adjusted in 1976, 1983, 1988, 1993 and 1997.102 The two-thirds of earned income limit is still in effect, except for single-parent students after 1995.103 In 1983, the rules were changed so that the spouse with the lower income must now ordinarily claim the deduction. In addition, if one of the parents is in full-time attendance at a designated educational institution or is otherwise incapable of caring for the children, the higher-income spouse can claim the deduction. Higher-income spouses are subject to additional weekly limits on the amounts deductible, because the deduction is prorated for the portion of the year in which the other spouse is not available to provide care. In the earlier years of the deduction, these additional limits applied generally to men. In 1988, the amount available for children aged under 7 years of age was increased relative to the amount available for children between the ages of 8 and 13. The limit of four children per family was also removed. These changes have remained in effect since then. Since 1988, then, families with more than four children are not affected by an additional limit. 1 0 4 In 1996, the age limit for older children without disabilities was increased to 16.105 In 1998, the deduction was extended to part-time students.106 Appendix 4.1 shows the evolution of the CCED limits between 1972 and 1999, in nominal and constant 1999 dollars.107 In 1999 dollars, the maximum annual per child deduction for children 102 Canadian Master Tax Guide, 34 , h-51 s t ed. (1975-1996); The National Finances, 1970-71 and 1971-72 ed.; Forms T778 E(97-99), (Revenue Canada Taxation). 1 0 3 Department of Finance Canada (1999). 104 Canadian Master Tax Guide, 34 t h-51 s t ed. (1975-1996); The National Finances, 1970-71 and 1971-72 ed.; Forms T778 E(97-99), (Revenue Canada Taxation). 1 0 5 Form T778 E (96), (Revenue Canada Taxation). 1 0 6 Department of Finance Canada . 1999. 1 0 7 These limits are established by the federal government and apply also to all provincial governments and territories that harmonize their tax systems with the federal system. Quebec is the exception. In 1988 Quebec made an effort to harmonize its tax system with the federal system, so the same ceilings were in effect for that year, but in general the C C E D deduction rules for Quebec were different from the federal government rules for other years. 109 under 7 years of age increased from $2,119 in 1972 to $7,000 in 1999, and for children between 8 and 16, the maximum deduction increased from $2,119 in 1972 to $4,000 in 1999. Figure 4.1 shows the number of CCED claimants from 1972 to 1996.108 The number of taxable returns has increased five-fold over the 15-year period, from 142,454 in 1972 to 759,540 in 1996.109 FIGURE 4.1 Number of CCED Claimants 1,000,000 -i 1 900,000 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 Year Figure 4.2 shows the amount of child-care expenses reported and allowed by taxable and all claimants for the years 1972-1996, in 1999 dollars.110 As the number of claimants has increased, so has the number of claims. Taxable and allowable claims increased eightfold from $268 million in 1972 to $2.129 billion in 1996. As a result of the increase in real limits of claims, the increase in the value of allowable claims has been greater than the increase in the number of claims. 1 0 8 Revenue Canada Taxation. 1974-1998. 1 0 9 Taxable returns are those that result i n a tax liability, while a l l returns include returns that do not result in a tax liabil i ty. 110 Figure 4.3 shows the proportions of CCED claimants who were affected by limits from 1972 to 1996. Limits include the two-thirds of earned income limit, weekly limits on men in the early years of the deduction and on the higher-income supporting person in later years, and the family limit based on an amount per child. Taxable claimants affected by the latter limit account for about 94.0 percent of constrained claimants, compared with 91.6 percent for all claimants.111 The proportion of claimants affected by limits has decreased over the 15-year period as real limits on the maximum allowable deduction for families have increased. FIGURE 4.2 Reported and Allowed Claimant Child Care Costs, ('000 of 1999 $): 1972-1996 3,000,000 -| • • • • • — — 1 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 Year From 1972 to 1996, each time the family limits were increased, the percentage of claimants affected by limitations declined. The 1983 increase in the limit restored the proportion of constrained claimants to its 1976 level, but subsequent increases in the limits in 1988 and 1993 further reduced 1 1 0 Revenue Canada Taxation. 1974-1998; Statistics Canada, The Consumer Price Index, Statistics Canada 111 the proportion of claimants affected by limits to all-time lows. As the limits for expenditures increased in real terms, the CCED likely became less progressive. It is important to remember, however, that the CCED is designed to address horizontal rather than vertical equity issues, and that removing deductibility ceilings promotes this objective. FIGURE 4.3 Percentage of Constrained Returns 35.0% 30.0% 25.0% 20.0% 15.0% 10.0% 5.0% A - taxable returns - all returns 0.0% 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 Year As more mothers of young children entered the labour force in the last 30 years, the CCED became more popular. The CCED limits have been gradually increased in real terms to better reflect the child-care costs parents are facing. The CCED has helped counteract the disincentive to work that untaxed household production creates. As large numbers of mothers of young children continue to work, it will be important to ensure that horizontal equity is maintained and the CCED is designed to achieve this objective. Catalogue 62-001. 1 1 1 Revenue Canada Taxation. 1974-1998. 112 4.3 VERTICAL AND HORIZONTAL EQUITY CONCEPTS AND RELATED LITERATURE In this section I consider general definitions of vertical and horizontal equity. Vertical equity is concerned with the extent to which those with higher incomes (or ability to pay) bear a higher share of the total tax burden (or the burden of a particular tax); or alternatively, the extent to which average tax rates rise with income. The aim of a progressive taxation system is to achieve vertical equity. Horizontal equity is achieved when those with equal ability to pay, share an equal proportion of costs, regardless of other differences such as source of income. The question arises as to the appropriate measure of "ability to pay". I consider two possible measures: actual and potential earnings. Actual earnings are pre-tax actual earnings, while potential earnings are the earnings of an individual who works for pay full-time. Most of the literature discussed below uses either one or both of these definitions to evaluate equity. Conclusions regarding the appropriate treatment of child-care expenses in the tax system are related to whether ability to pay is measured by actual or potential earnings. The distinction between actual and potential income is an important one in choice theory. Individuals (or families) are assumed to make rational choices about market consumption, market labour supply, home production, and leisure on the basis of their preferences and their opportunity set. The opportunity set consists of all the possible consumption-labour supply-home production-leisure combinations available to individuals or families given their wage rates and non-labour income that exhaust their resources.112 Some people prefer more leisure with less market consumption, while others prefer to work more and consume more market goods. Furthermore, in a household with young children, there are considerable opportunities for home production to replace market production and vice versa (mother's care versus purchased child-care, home-cooked meals versus frozen dinners or restaurant meals, etc.). Conceptually, potential income can be seen as the 1 1 2 The theory summarized here is based on a one-period model that does not allow for savings. 113 market value of actual consumption plus the market value of actual leisure and home production. In contrast, actual income captures only the market value of consumption and ignores the market value of leisure and home production. An individual who works the maximum number of hours possible in the market-place will earn actual income that is equal to her potential income, and thus actual income will thus reflect her opportunity set. In contrast, an individual who works less than the maximum number of hours will earn actual income that is less than her potential income, and thus actual income will understate the value of her opportunity set. This distinction is crucial to the analysis of equity in societies where families make widely different choices about their degree of market consumption versus leisure and home production. Table 4.1 offers insights into using potential earnings instead of actual earnings as a measure of ability to pay via a numerical example that employs three scenarios of a basic story to gauge horizontal equity. The table depicts how the three different scenarios affect the after-tax position of two individuals. The calculations are based on a simplification of the current Canadian income tax system for expository purposes. I subsequently consider the effect on the relative after-tax position of the two individuals of changing current tax policy to (a) eliminate the CCED and (b) keep the CCED but make it also available to families with a stay-at-home parent. The example involves a widower who needs child-care for his two children. In scenario 1, the widower hires a nanny to care for his children. In scenario 2, the widower and the nanny marry but the nanny stays home to care for the children. In scenario 3, the widower and the nanny marry and the nanny goes to work outside the home, earning the same as before she was married, and hires another nanny to replace all of her own services at the same price. This scenario thus models the dual-earner couple. In all cases the couple's potential earnings are $94,000, which are also their actual earnings in scenarios 1 and 3, when the nanny's services are compensated through a taxed market exchange. In scenario 2, however, when the nanny's services are compensated through untaxed marital transfers of resources, their actual income is $70,000. 114 TABLE 4.1 The Widower and Nanny Get Married (1) The widower and nanny are not married. (2) The widower and nanny are married and nanny stays home. (3) The widower and nanny are married and nanny works outside and family hires another nanny for the same price. Tax system: - Child care expense deduction equals $14,000 - Personal tax credit equals $1,500 - Married exemption credit equals $1,500 - Equivalent to married exemption credit equals $1,500 - Tax on 1st $30,000 is 26% - Tax on balance is 42% (1) (2) (3) Not married M a r r i e d Both work Widower Nanny Total Total Widower Nanny Total Gross earnings 70,000 24,000 94,000 70,000 70,000 24,000 94,000 Child care expense deduction 14,000 14,000 14,000 14,000 Taxable earnings 56,000 24,000 80,000 70,000 70,000 10,000 80,000 Tax on 1st 30,000 7,800 6,240 14,040 7,800 7,800 2,600 10,400 Tax on balance 10,920 10,920 16,800 16,800 16,800 Taxes before credits 18,720 6,240 24,960 24,600 24,600 2,600 27,200 Personal credit 1,500 1,500 3,000 1,500 1,500 1,500 3,000 Married credit - 1,500 -Equivalent to married credit 1,500 1,500 -Tax payable 15,720 4,740 20,460 21,600 23,100 1,100 24,200 After tax income 54,280 19,260 73,540 48,400 46,900 22,900 69,800 Household services cost 24,000 ** 24,000 Discretionary income 49,540 48,400 45,800 ** Since household services income is not recognized, household services cost are also not deducted. Notes: If the CCED was eliminated, couple (1) would have discretionary income of $49,540 -.42($14,000) = $43,660 and couple (3) of $45,800 - .26(14,000) = $42,160. With the CCED, couple (3) still has $2,600 less in discretionary income than couple (2). Without it, the discrepancy is increased to $6,240. Although the widower and the nanny, when considered as a couple, have the same before-tax opportunity set regardless of their situation,113 their after-tax position deteriorates significantly when the (married) nanny works outside the home. The unmarried couple in scenario 1 has the best discretionary income.114 The dual-earner couple in scenario 3 has the worst discretionary income. Relative to couple 1, couple 3 does not have an equivalent-to-married deduction and the lower-income spouse claims the CCED; the result is additional tax of $1,500 + [(0.42 - 0.26) x $14,000] = Assuming that the nanny is a live-in nanny. 115 $3,740. Relative to the married couple with a stay-at-home parent in scenario 2, couple 3 must report $24,000 of additional income. Since only $14,000 of this income is deductible, the result is additional tax of $10,000 x 0.26 = $2,600. Looking at the alternative policy options described above, I calculate that if the CCED were eliminated, couple 3 would pay $3,640 ($14,000 x 0.26) more in tax and couple 1 would pay $5,880 ($14,000 x 0.42) more. If the CCED were extended to married couples with a stay-at4iome parent, couple 2, with a deduction of $14,000, would save an additional $5,880. Either one of these measures (eliminating or extending the CCED) would increase the horizontal inequity relative to couple 3. It is clear that the married nanny would work outside the home only if she could deduct the full cost of her replacement, or if she could earn an income that exceeds the cost of her replacement by an amount sufficient to offset the tax liability that arises from the inability to deduct that cost. Hence, if a secondary wage earner is to work outside the home, either the value of her market production must be greater than the amount that the family would be willing to pay for child-care and other household services, or the value of her market production must be significantly greater than the value of her household production. In this example, when potential income is used to evaluate horizontal equity, income taxes are divided by $94,000 in scenarios 2 and 3 to measure the tax rate the couple faces in each of the two situations. The benefits of household production are thus recognized in determining the measure of ability to pay. The tax rates are 23.0 percent for the couple with a stay-at-home parent and 25.7 percent for the couple with two parents who engage in market work.115 1 1 4 The after-tax income and household services situation in scenario 1 improve when the nanny's earnings are lower for a given widower's wage, since the nanny pays tax on her wages. 1 1 5 How could the tax system be neutral? Taxing the stay-at-home parent for the imputed value of her home production and giving the working parent an equivalent deduction while also allowing the higher-income earner in the dual-earner family a similar deduction to defray home production replacement costs would result in neutrality. Alternatively, simply taxing the imputed value of home production and providing no deductions to either type of family would be horizontally neutral in the labour supply dimension, but not in the childless versus with-children dimension. 116 A number of authors either implicitly or explicitly use potential income as the appropriate measure of ability to pay. Gentry and Hagy (1996) examine horizontal and vertical equity concepts in relation to the tax treatment of child-care expenses in the United States. They point out that the Haig-Simons definition of income, supports the deductibility of child-care expenses in families where these expenses are incurred to earn income 1 1 6 (as opposed to treating child-care expenses as a tax credit). The Haig-Simons definition of income says that income is the monetary value of increases in potential consumption net of the costs incurred to earn that income. If one can argue that child-care expenses are incurred to earn income, then employment income can be defined as employment earnings less child-care expenses. While a child-care tax credit is tied to child-care expenses, it makes no adjustment to income to reflect that this expense is incurred to earn income. However, one of the purposes of a tax deduction is to adjust income to reflect expenses incurred to earn the income. In Canada, similar status is accorded moving expenses when a person is moving for employment reasons, and annual union, professional or like dues. The central point of Gentry and Hagy's analysis is that since the actual income of the family is dependent on labour supply choices, actual income is not (necessarily) the appropriate measure of ability to pay. Using a sample of families with children 1 1 7 and a measure of actual income, Gentry and Hagy find that the US child-care tax credit is regressive in the lowest quintile of the income distribution and progressive for the higher quintiles. When potential income 1 1 8 is used as the measure of ability to pay, the benefits are progressive throughout the income distribution. Gentry and Hagy evaluate horizontal equity by examining the distribution of benefits across family characteristics but they fail to analyze this issue across families with and without a stay-at-home parent. Since the gap between potential earnings and actual earnings is large for families with a stay-1 1 6 Gentry and Hagy also point out that the argument for any tax consideration of child-care expenses is weakened when a longer time horizon is used-for example, when the family-planning stage is included. In this case, child-care expenses can be considered as expenses related to the decision to have children, rather than as expenses related to the decision to work. 1 1 7 Families with children only avoid the life-cycle earnings issue. 117 at-home parent but is small or absent for a family without one, an analysis that promotes the concept of potential income as the appropriate measure of ability to pay should look at the effect on horizontal equity of redefining income from actual income to potential income. Feldstein and Feenberg (1996) also implicitly favour a potential income measure in their analysis of the taxation of dual-earner families. They argue that the US system of taxing families instead of individuals is unfair because it imposes the same burden of taxation on a married couple with one earner as it does on a dual-earner couple with the same income. The unfairness arises because the dual-earner couple usually works more hours in the market place and has less untaxed home services. In the Canadian tax system, this inequity is somewhat mitigated by individual filing combined with a progressive taxation system. Similarly, when Krashinsky and Cleveland (1999) examine the issue of tax fairness for single- and dual-earner families, the idea that single-earner households (with two parents) benefit from untaxed household production is central to their analysis. The authors examine the role of the CCED in taxation and also conclude that it partially corrects for the non-taxation of household production and thus contributes to the reduction of horizontal inequity between single- and dual-earner households. The idea that potential income should be used as a measure of ability to pay can be reduced to the alternative that child-care expenses (and the cost of other household production substitutes) incurred to earn income should be tax deductible. Vincent and Woolley (2000) point to this relationship: they note that the CCED is needed to achieve horizontal equity since it merely recognizes the costs of working, or alternatively, recognizes that (the benefit of) parent-provided child-care is untaxed. They conclude that child-care costs incurred to earn income should be untaxed. Gentry and Hagy define potential labour income as the sum of each parent's wage rate (imputed i f not available) times 50 hours per week for 52 weeks per year. 118 The strand of the literature that implicitly or explicitly focuses on actual income as the appropriate measure of ability to pay reaches opposite conclusions with respect to horizontal equity. In addition, when the underlying economic rationale for the CCED-the promotion of horizontal equity-is ignored, issues of vertical equity are often brought to the fore. Boessenkool (1999) examines the Canadian tax treatment of the family. He argues that the CCED contributes to horizontal inequity because the choice of parents who do not raise their children at home is "subsidized", while the choice of parents who do is not. In an earlier analysis, Boessenkool and Davies (1998) state: "Critics of the CCED ignore the fact that parents who provide child-care themselves, instead of purchasing it in the market, in effect already enjoy a child-care deduction. The value of their child-care services, which is a form of in-kind income, is not taxed. Allowing for a deduction [the CCED for dual-earner families] when these are paid out of earned income is thus a simple requirement of horizontal equity."119 [Emphasis added.] This statement contradicts Boessenkool's later claim that the CCED violates horizontal equity. The interpretations differ because in one case horizontal equity is evaluated on the basis of actual income, while in the other the value of household production is also recognized. Potential income implicitly puts a market value on household production. Gordon (2000) examines the adequacy of section 63 of the federal Income Tax Act (the CCED) as the principal mechanism for state funding of child-care. Gordon points to Hogg and Magee's (1997) argument that "[t]he failure to tax imputed income is one factor tending to discourage women from seeking work outside the home.... [T]he partial deductibility of child-care expenses helps reduce the barrier against work outside the home."120 Gordon dismisses this fundamental point, however, by stating that many families are denied the option of staying at home (by tax policy for example) and that others may simply be unemployed. 119 1 2 1 It is neither clear how the tax system denies a mother the option to stay home nor why unemployment is a relevant aspect of the equity of the CCED. The fact that a person is unemployed should not impede in their ability to enjoy leisure or engage in household production although they may also be looking for work; however, a relatively small percentage of individuals may be unable to work because of illness or other handicaps, which should not detract from the overall analysis. Gordon views the CCED as a subsidy for child-care expenses rather than as a tool of horizontal equity, and argues that a program of grants and subsidies to approved child-care facilities would be preferable to the CCED. She further suggests that the implicit child-care costs of stay-at-home parents should be subsidized.122 Hung (1998) argues that the CCED encourages employment in the outside labour force by limiting the deduction to actively earned income and that it discourages the recognition of the domestic labour force. She argues that the CCED violates horizontal equity by not providing the same tax benefits and assistance to those in the same economic position. Hung uses the concept of actual income to determine economic position. Given two households with the same actual income, but with differing labour supplies, one household may qualify for the CCED while the other (the houshold with a stay-at-home parent) would not. This interpretation of horizontal (in)equity is widespread and implicitly assumes that actual income is the appropriate measure of ability to pay. Hung commits the same error she accuses the CCED of: she ignores the value of household production. The CCED does not recognize the value of household production as an expense eligible to be treated as a deduction from income because the value of household production is not treated as income. Young (1994) examines the Symes v. Canada case in terms of its vertical equity implications. Symes, a self-employed professional, argued that any child-care expenses not deductible under 1 2 1 The 1988 CNCCS indicates that 83 percent of stay-at-home designated adults were out of the labour force rather than unemployed. 1 2 2 This argument ignores the fact that home child-care 'costs' are already being subsidized by not being taxed. 120 section 63 should be deductible as business expenses. The Supreme Court of Canada disagreed and ruled that child-care expenses are not deductible as business expenses. Young concludes that although a positive result for Symes might have improved the situation for women, already privileged women would have benefited most, and that since the pool of child-care funding is limited, already disadvantaged women would have been worse off. In effect, a positive Symes result would have reduced vertical equity. The arguments common in papers that focus on actual income as the appropriate measure of ability to pay are that the CCED, or any form of state funding to child-care for working parents, violates horizontal equity between two-parent single- and dual-earner families with identical actual income, and/or that the CCED violates vertical equity. Untaxed household production or costs of working are given little or no consideration. The debate in Canada over these issues has been limited to a theoretical one. The dividing line between the positions taken is based on whether authors recognize the value of household production (and implicitly the concept of potential income as the appropriate measure of ability to pay). This study uses an empirical framework to evaluate the horizontal and vertical equity of the CCED and to demonstrate how the use of actual versus potential income can lead to markedly different conclusions. An additional concern that arises in measuring ability to pay is whether annual measures of income are more appropriate than lifecycle income measures. Altshuler and Schwartz (1996) use these two different measures to evaluate the progressivity of the US treatment of child-care expenses and find the system to be progressive. When all families are included in the evaluation of a child-care expense related deduction or credit, the benefit may appear more progressive because individuals eligible for the deduction or credit are in young families who tend to earn less than individuals in older families. Young families eventually become old families and it is thus not appropriate to treat them as "different" families or families with different ability to pay. Using the lifecycle approach to 121 defining income compensates for this. Gentry and Hagy (1996) avoid this issue by restricting their analysis to families with young children. That approach is followed here. 4.4 CCED CEILINGS AND CHILD-CARE PAYMENTS BY INCOME GROUPS, 1987/1988 In this section I examine how claimants were affected by the increase in eligible child-care expenditures for children under 7 years of age from $2,000 in 1987 to $4,000 in 1988. I also examine the proportion of income that claimants in different income groups spend on child-care. Note that in this section when vertical equity is discussed, it is evaluated with the traditional concept of actual rather than potential income as the measure of ability to pay. Figure 4.4 shows the percentage of taxable claimants affected by the family limit for 1987 and 1988 by income groups. Revenue Canada publishes annual information on the number of tax returns with allowed child-care expenses.123 This information is organized by income group and includes the total number of claimants, the number of children claimed, total payments for care, allowable deductions, and the number of claimants that are constrained by the various limits. Higher-income claimants, who tend to spend more per child on child-care, are more likely to be affected by the family limit. Hence, although higher-income claimants may receive a higher tax rebate on the last dollar they spend on child-care, this occurs only if their expenditure is below the limit. Once the limit is reached, there is no additional tax deduction. Because the limits are more likely to affect higher income individuals, they increase the vertical equity of the CCED. The years 1987 and 1988 are included to show the effect of a large increase in the limit on the distribution by income group of individuals who are affected by the limit. As can be seen from the change in the percentage of claimants affected by the limit when it is lifted, individuals in the upper-middle income groups are the ones most likely to benefit from the change. Revenue Canada Taxation. 1974-1998. 122 FIGURE 4.4 Percentage of Taxable Claimants Affected by Family Limits: 1987 & 1988 by Income Groups Figure 4.5 shows reported child-care expenses as a percentage of claimants' income, by income group, for all taxpayers with CCED claims in 1987 and 1988. Note that reported child-care expenses are total 'receipted'124 child-care expenses and are greater or equal to the allowed deduction, and that high-income earners are more likely to be affected by ceilings. To calculate the percentages, the denominator is set equal to the midpoint of the income cell, except for the highest income group, where $100,000 is used to represent income. Although the average income for this category is certainly higher than $100,000, the selection of any particular amount is arbitrary. As income rises, child-care expenses decline as a proportion of income. Receipted child-care expenses are those the taxpayer can claim. Taxpayers report their receipted child-care expenses and then calculate whether they w i l l be eligible to cla im the full amount or w i l l be affected by one o f the ceilings. 123 FIGURE 4.5 Reported Child Care Payments as Percentage of Income: 1987 & 1988 by Income Groups Figures 4.4 and 4.5 show that high-income earners are more likely to be affected by CCED ceilings and that they spend a smaller proportion of their income on child-care. These effects work favour of vertical equity of the CCED. Although these statistics provide insight into vertical equity, horizontal equity has to be analyzed using data and procedures that consider the heterogeneity of taxpayer choices. These are discussed in the following sections. 4.5 DATA AND ESTIMATION PROCEDURE The 1988 Canadian National Child-care Survey (CNCCS) provides data on family characteristics and child-care expenditures for a sample of 24,155 families with children under 13 years of age. A total of 42,131 children are represented in the survey. Data include the different types of supplementary care used for each of four children in the family in the reference week, the 124 individual cost by care arrangement and child, the total cost of care for the family, and whether the parents planned to claim each particular cost associated with each child and each of the child's care arrangements on their tax return. I estimate an annual cost of care and related CCED benefit for each family with child-care costs. To do this, I first estimate an hourly cost of care equation for families who plan to claim the primary cost of care for their youngest child. 1 2 5 Hourly cost of care for non-claimants is lower than hourly cost of care for claimants. Non-claimants usually do not claim their child-care expenses because the care provider does not supply receipts. I assume that in a competitive child-care market 1 2 6 the child-care purchaser will require either receipts from the provider or a lower fee as compensation for the loss of the CCED benefit.127 I predict a "full cost" of care for each family with child-care expenses and estimate the CCED benefit on the basis of this full cost for each family. The full-cost is the predicted cost of care using coefficient estimates for claimed care. The annual predicted cost equals the weekly cost times 52 for families who claim the primary care cost for their youngest child.1 2 8 For other families with child-care costs, the predicted cost of care equals the predicted hourly cost of claimed care for this type of family times the designated adult's usual weekly hours of work times 52. I estimate predicted annual earnings for the designated adult and the spouse. In order to do this, I estimate a selection-corrected wage equation using the 1988 Labour Market Activity Survey and use the coefficients 1 2 9 from this equation with parallel characteristics in the CNCCS to estimate wage rates for the CNCCS designated adults and spouses. I restrict observations to individuals with See Appendix 4.2 for the equation and its coefficients. 1 2 6 For a discussion of the supply of child-care see, David M . Blau (1993). 1 2 7 The average hourly cost of care for families who pay for care but do not categorically plan to claim is $ 1.56, while it is $2.49 for those who pay and plan to claim. The 'grossed up' average for the first group is $2.27 per hour based on the equation in Appendix 4.1. 1 2 8 Note that in most child-care centers and family day homes, fees are still paid during summer vacation if one wishes to retain the child's space. Less formal arrangements (babysitters) likely result in a break in fees, but it is not possible to estimate the length of this break. 1 2 9 The coefficients are reported in Appendix 3.3. 125 children less than 17 years of age and divide them into four groups: single and married males and females. The wage equation can be written as follows: LnW = yX + o-wh/rjhA, where A, = <t>(-pZ/o-h)/l -0(-pZ/ah) X = a vector of explanatory variables for wage rates Z = a vector of explanatory variables for the probability of working during 1988 The wage equation is estimated using workers only. Lambda is obtained from a probit equation on workers and non-workers that estimates the probability an individual worked during 1988. The matrix Z is identical to the matrix X , except that it also includes the number of children age zero to two, the number of children aged three to five, and whether a teenager is present. For married females, husband income is also included in Z. The wage rates obtained from the above procedure are multiplied by usual hours of work to arrive at predicted annual earnings. To estimate full-time annual earnings, the predicted wage rates are multiplied by 1,950 hours (37.5 hours per week). Potential earnings are the greater of actual and full-time earnings. For example, using the coefficients from appendix 4.3, the 1988 predicted log-wage rate for a Canadian-born 30-year-old married woman who has a university degree, lives in Ontario (the reference group) and has one child under the age of 16 is: 1 3 0 LnW = 1.834 + 0.208 + 0.699-0.018 =2.723 Lambda is not used to calculate the predicted wage; the above calculation yields an unbiased estimate of the wage for workers and non-workers; unbiased wage estimates of workers only can also be obtained by including lambda in the calculation. Lambda is included as a regressor in the wage equation because the wage equation is estimated using data for workers only, but is used to predict wages for both workers and non-workers. See, Killingsworth (1983), p. 143. 126 This yields a wage of $15.23 and, assuming a 37.5-hour workweek, full-time annual earnings of $29,690. For 1999, all earnings and child-care costs estimates are multiplied by 1.303, which is based on changes in the overall Canadian consumer price index from 1988 to 1999. Hence, the 1999 full-time earnings for the same woman are predicted as $38,687. I estimate the 1988 tax expense with and without the CCED for each (potential) claimant and for the family as a whole. I estimate actual and potential earnings for each parent using usual and potential hours of work and predicted wage rates. I then estimate taxable earnings with and without a CCED in place. The CCED is allocated to the lower income spouse and is calculated based on the predicted annual cost of care, but is subject to the two-thirds earned income and the per child ceilings. I then apply marginal tax rates to each income slice. Tax calculations are based on province of residence and include federal surtaxes, family allowances in 1988 1 3 ' , the federal child tax credit, the federal sales tax credit, and provincial surtaxes and credits, but do not include provincial child benefit payments.132'133'134 Canada Pension Plan and employment insurance payments are also not included in the calculation: it is assumed that the net costs of these equal their net benefits. Although Quebec allowed the CCED to be deducted by higher-income earners in 1988 and had converted the deduction to a refundable credit by 1999, for the purpose of comparison, benefits for families living in Quebec are simulated as though Quebec used the same rules as the rest of Canada. Once taxes have been calculated for the family, the benefit from the CCED is calculated as the difference between tax in the presence of and tax in the absence of the CCED. This procedure is 1 3 1 Family allowances have been allocated to the higher income earner. 1 3 2 Sources for tax information include: Treff and Perry (1999), Canadian Master Tax Guide, 44th and 45th ed. Thorne Ernst & Whinney (1989). 1 3 3 Although the base amount for the Canada child tax benefit for Alberta differs by age group, the Canadian benefit has been used. 1 3 4 The data does not identify children that are 17 years old - these children are omitted in calculations where they might otherwise be included. 127 repeated for actual and potential earnings. I calculate (predicted) average family tax rates, and CCED benefits as a proportion of actual family earnings and potential family earnings. 128 4.6 ANALYSIS a) Horizontal Equity Horizontal equity is evaluated by comparing the tax rates faced by two-parent families with children under 6 years of age.135 The analysis in this section is limited to two-parent families because single parent families are likely to collect social assistance when the parent is not employed, and are also likely to receive provincial child-care subsidies when the parent is employed. Families are sorted into two groups: stay-at-home (at least one parent who does not usually work), and "both at work". FIGURE 4.6 1988 Average Family Tax Rates as a Proportion of Predicted Family Earnings: Two Parent Families with Children < 6 0.30 < 20,001 20,001 - 30,001 - 40,001 - 50,001 - 60,001 - 70,001 + 30,000 40,000 50,000 60,000 70,000 Predicted Family Earnings 1 3 5 Families with pre-school and kindergarten children are the ones most likely to have significant child-care costs when parents work outside the home. 129 FIGURE 4.7 1999 Average Family Tax Rates as a Proportion of Predicted Family Earnings: Two Parent Families with Children < 6 0.30 < 20,001 20,001 - 30,000 30,001 - 40,000 40,001 - 50,000 50,001 - 60,000 60,001 - 70,000 70,001 + Predicted Family Earnings Figures 4.6 and 4.7 show 1988 and 1999 average family tax rates as a proportion of predicted actual family earnings for two parent families with children under the age of 6. At income levels of $30,000 or more in 1988, and $40,000 or more in 1999, families with a stay-at-home parent face higher tax rates than families without, regardless of the CCED. The higher tax rates for families with a stay-at-home parent are due to of individual progressive taxation; the CCED only exacerbates the discrepancy, which adds fuel to the argument that it violates horizontal equity. Figures 4.8 and 4.9 show 1988 and 1999 average family tax rates as a proportion of potential family earnings for two-parent families with children under the age of 6. In both years at all earnings levels, families with a stay-at-home parent face a much lower tax than families in which both parents work. The CCED reduces the horizontal inequity that results from untaxed household production, although most of the inequity remains.136 Note that 1999 tax rates tend to be lower than 1988 rates. I calculated Plotnick indices for these series, which are discussed in Appendix 4.4. See Plotnick (1981, pp. 283-288. 130 The child tax credit is treated as a negative tax in both years. This benefit is much larger in 1999 because it replaced family allowances of earlier years.137 FIGURE 4.8 1988 Average Family Tax Rates as a Proportion of Potential Family Earnings: Two Parent Familes with children < 6 0.30 0.25 0.20 0.15 - Both at work - bef. CCED - Stay-at-home - Both at work - aft. CCED X 0.10 0.05 0.00 4 -0.05 < 30,001 30,001 - 40,000 40,001 - 50,000 50,001 - 60,000 60,001 - 70,000 70,001 -Potential Family Earnings It is clear that when people work more they earn more taxable income, so the fact that families who work more hours are more likely to have a higher tax rate in a progressive income tax system is not an unusual discovery. Family allowances for 1988 are included in the calculation of taxes but are neither treated as income nor as a benefit (or tax reduction) when tax rates are calculated. Family allowances were not affected by the CCED, but the child tax credit is. 131 FIGURE 4.9 1999 Average Family Tax Rates as a Proportion of Potential Family Earnings: Two Parent Familes with children < 6 0.25 30,001 -40,000 40,001 -50,000 50,001 -60,000 60,001 -70,000 70,001 + Potential Family Earnings Figures 4.6 to 4.9 highlight that using actual earnings as a measure of ability to pay results in a view that it at odds with reality. Whether families choose to have a stay-at4iome parent who specializes in household production or not has an important impact on the value of the actual income a family earns. The actual earnings of a family with a stay-at-home parent do not need to be used for child-care. In addition to providing child-care, the stay-at-home parent can prepare less costly home cooked meals, do gardening and house cleaning and can spend more time shopping for bargains. Families with different labour supplies are not comparable in terms of the benefits they derive from a given level of actual earnings. Potential earnings are a more comprehensive measure of family well being and with this measure, the CCED promotes horizontal equity. 132 b) Vertical Equity Since the main purpose of the CCED is to rectify horizontal inequity,1381 will focus my analysis of vertical equity on potential claimants-that is, on parents who work and incur child-care costs. There are five effects at work in terms of vertical equity of the CCED. These can be examined with the following simplified equation for CCED benefits as a proportion of income 1 3 9 , 1 4 ° : percent CCED benefit = CCED benefit/income (CC*MTR)/income if CC < limit, and (Limit*MTR)/income if C O limit where: CC = child-care expenses MTR = marginal tax rate The five effects are as follows: (1) As income rises, child-care expenses per child also rise, but not as fast as income. The ratio of benefits to income declines as income rises because child-care expenses decline as a proportion of income. This effect works in favour of vertical equity. (2) As income rises, child-care expenses per child also rise. The annual deduction limit is approached and, eventually exceeded. Above the limit, there are no tax benefits to additional Krashinski and Cleveland (1999) make this point. I am mankful to Gordon Cleveland and Jon Kesselman for observing that the attainment of vertical equity is not the goal of the CCED. Other child-care programs such as child-care subsidies, programs such as social assistance, as well as the progressive tax rate schedule, are specifically designed to address vertical equity issues. 1 3 9 Here income refers to total income before deductions; the empirical work in this study focuses solely on predicted employment and self-employment income and excludes property income. 4 0 This equation works provided the CCED does not result in the taxpayer moving into a different tax bracket; otherwise, the benefit will be a weighted average of two different marginal tax rates. 133 expenditures on child-care. Thus the ratio declines as income rises. This effect works in favour of vertical equity. (3) As income rises the marginal tax rate rises at discrete intervals, which causes the ratio to rise. This effect works against vertical equity. (4) (5) When income is very low, the two-thirds of earned income limit is more likely to bind, and low-income taxpayers will receive little or no benefit. This effect works against vertical equity. When income is very low, or when a lower-income taxpayer has a lot of personal tax credits, the benefit from the CCED cannot be enjoyed because there is little or no tax liability and the benefit is not refundable. This effect works against vertical equity. FIGURE 4.10 Potential CCED Benefits: Two Parent Family 8000 # # # .# a# ^ ^ ^ v# a# # V *)' <V <V <V' rj»< r{v rgl' nj>' ^ fr fr fr fr fr fr fr <<V Earnings of Claimant 134 Much of the debate about the vertical equity of the CCED focuses on the absolute amount of benefits received by claimants based on their income. Figures 4.10 and 4.11 show the 1988 and 1999 nominal-dollar potential benefits when claimants' expenditures on child-care are equal to or greater than the CCED ceiling.141 The calculations are based on British Columbia tax rates. Figure 4.10 shows benefits for two parent families, while Figure 4.11 shows benefits for single parent families. Cit is clear that potential benefits from the CCED increase with income. There are no benefits for low-income earners, and potential benefits jump with increases in marginal tax rates.142 FIGURE 4.11 Potential CCED Benefits: Single Parent FamUy 8000 7000 6000 „ 5000 - O - One child < 7, 1988 - O - Two children < 7, 1988 - A - One child < 7,1999 - X - T w o children < 7, 1999 # # # # # # # # # # # # # # # # # # # # V V < V < V < ^ r j v ' c{r rg>< n j > ' ^ fr #1- fr fr fr fr fr fr / f \ < Earnings of Claimant A single parent begins to enjoy the benefit only when she reaches an income level that is almost twice that of the married parent. This occurs because the single parent is entitled to an The analysis can be extended to all users of paid child-care if we assume that benefits are either derived from the tax system or from the lower fees of care providers who do not issue receipts. 1 4 2 It is important to note however that the reason low-income potential claimants do not receive any benefit is either because they have no tax liability, or more rarely because their child-care costs are unusually large in relation to their earnings. 135 equivalent-to-married deduction and pays no taxes until her income exceeds that of the married parent. Critics of the CCED who are concerned about vertical equity point to this apparent regressivity. Low-income individuals do not benefit from the CCED, the amount of the benefit increases with income, and the benefit limit is reached only when income exceeds $60,000. The regressivity of a tax measure is evaluated, however, by the impact of the measure on tax rates as income rises. Hence, it is the benefit rate rather than the absolute benefit that should be considered. This is consistent with the definition of vertical equity, which states that those with higher incomes should pay a greater proportion of their income in taxes. FIGURE 4.12 Potential CCED Benefit as a Percentage of Earnings: Two Children < 7 Figure 4.12 shows the 1988 and 1999 nominal-dollar potential CCED benefits as a percentage of earnings of single- and two-parent families with two children under 7 years of age. The 136 benefit is very regressive at lower income levels, becomes progressive for low-middle income levels, is regressive again when the second marginal tax rate is reached, and becomes progressive again for the balance of higher income levels. It should be remembered, however, that low-income earners in single-parent families can usually qualify for full or partial child-care subsidies and thus may not face any child-care costs. Nevertheless, these subsidies are often subject to ceilings and availability constraints, and the CCED would not likely provide any additional relief for the remainder of these costs for a low-income single parent. However, child-care subsidies are specifically intended to address vertical equity concerns, and if they are inadequate, the CCED itself should not be maligned. At this point, it is useful to consider an example of how the definition of actual versus potential earnings can affect vertical equity. The tax calculation in table 4.2 is similar to that of Table 4.1. Here, the three situations are as follows: 1) both the husband and the wife work, together earning actual and potential income of $70,000, and they are entitled to a CCED of $14,000, which covers all of their child-care costs; 2) both the husband and the wife work, together earning actual and potential income of $70,000, but the CCED does not exist; and 3) the husband works, earning actual income of $70,000, and the wife stays at home, but could earn $40,000 if she worked full-time, so the couple's potential income is $110,000. Hence the third couple is not comparable to couples 1 and 2 from a potential earnings perspective. Critics of the CCED, however, routinely engage in this type of comparison by focusing on actual income. Tax is calculated for all three situations, and at the bottom of the table tax liability is expressed as a percentage of pre-tax actual, discretionary, and potential earnings. When tax is expressed as a percentage of actual earnings, the system appears to suffer from horizontal inequity because couple 3 has a much higher rate of taxation than couples 1 and 2, even though they all have the same actual income. The CCED only exacerbates this problem. When tax is expressed as a percentage of discretionary income, which is defined as pre-tax income less child-care expenses, the system no longer violates horizontal equity and is progressive because the couple with the higher 137 discretionary income faces a higher tax rate. The CCED increases the progressivity of the system. When tax is expressed as a percentage of potential income, the tax system is regressive without the CCED, and moderately progressive with it. T A B L E 4 .2 H o r i z o n t a l a n d V e r t i c a l E q u i t y a n d the C C E D (1) Husband and wife both work full-time and earn $70,000 altogether. They incur child care costs of $14,000 per year. These costs are fully deductible. (2) Husband and wife both work full-time and earn $70,000 altogether. They incur child care costs of $14,000 per year. These costs are not deductible. (3) Husband works full-time and earns $70,000. Wife does not work, but could earn $40,000 if she worked full-time. Tax system: - Personal tax credit equals $1,500 - Married exemption credit equals $1,500 - Equivalent to married exemption credit equals $1,500 - Tax on 1st $30,000 is 26% - Tax on balance is 42% (1) With CCED (2) Without CCED (3) 1 works Husband Wife Both Husband Wife Both Worker Gross earnings 35,000 35,000 70,000 35,000 35,000 70,000 70,000 Child care expense deduction 14,000 14,000 -Taxable earnings 35,000 21,000 56,000 35,000 35,000 70,000 70,000 Tax on 1st 30,000 7,800 5,460 13,260 7,800 9,100 16,900 7,800 Tax on balance 2,100 2,100 2,100 2,100 16,800 Taxes before credits 9,900 5,460 15,360 9,900 9,100 19,000 24,600 Personal credit 1,500 1,500 3,000 1,500 1,500 3,000 1,500 Married credit - - 1,500 Tax payable 8,400 3,960 12,360 8,400 7,600 16,000 21,600 After tax income 26,600 31,040 57,640 26,600 27,400 54,000 48,400 Pre-tax discretionary (actual) income 56,000 56,000 70,000 Potential income 70,000 70,000 110,000 Tax payable/actual income (gross earnings) 0.18 0.23 0.31 Tax payable/pre-tax discretionary income 0.22 0.29 0.31 Tax payable/potential income 0.18 0.23 0.20 If actual income is the measure of ability to pay, there appears to be a problem with horizontal equity. If potential income is the appropriate measure of ability to pay, vertical equity is violated when no CCED is allowed, because unequals are treated as equals. The progressiveness of the tax system and individual filing mitigate this problem. Note that when income is spread over two taxpayers as in cases 1 and 2 compared with one taxpayer, as in case 3, individual filing and a progressive tax rate structure tend to promote vertical equity when ability to pay is measured in terms of potential income. These features compensate to 138 some degree for untaxed household production. The same features appear to violate horizontal equity when ability to pay is measured in terms of actual income. Figures 4.13 to 4.16 show predicted CCED benefits as a proportion of predicted and potential family earnings for two parent dual-earner families with one child under the age of 7 and child-care costs, for 1988 and 1999. In these graphs, each dot represents one household. Figures 4.13 and 4.14 show the predicted CCED benefits as a proportion of predicted and potential family earnings 1 4 3 , for 1988. Figures 4.15 and 4.16 show the same information for 1999. The data are plotted for families with one child less than seven to limit the comparison to similar families with different earnings 1 4 4 , and because the ceiling on the deduction changes at age 7. Note that since predicted and potential earnings are based on a predicted wage rate, the earnings range is narrower than it would have been if actual wage rates had been used. 1 4 4 The results are similar for families with two children under 7 years of age. 139 FIGURE 4.13 Predicted Benefit as a Proportion of Predicted Family Earnings: Two Parent Families with One Child < 7,1988 Dual Earner Families with Child Care Costs 0.06 0.05 TJ • • 0 20,000 40,000 60,000 80,000 100,000 120,000 Predicted Family Earnings It should be noted first that the CCED benefit as a proportion of predicted family earnings is quite low, ranging from 0 to 5 percent in 1988 to up to 6 percent in 1999. Many of the observations show zero benefits at lower earnings levels. Since the observations are restricted to dual-earner families with child-care costs, families with no benefit are those where the income of the lower-income spouse is so low that there is no tax liability, even without the CCED. As the income measure is changed from predicted to potential family earnings, the bottom of the distribution moves to the right. This occurs because families with very low or zero benefits are more likely to have a parent who is not working full-time, and hence potential earnings tends to be higher than predicted earnings for the majority of these families. With zero-benefit observations less concentrated at the bottom of the earnings distribution, we would expect the vertical equity of the CCED to improve when ability to pay is measured in terms of potential rather than actual earnings. 140 FIGURE 4.14 Predicted Benefit as a Proportion of Potential Family Earnings: Two Parent Families with One Child < 7,1988 Dual Earner Families with Child Care Costs 0.05 • • QJ 20,000 40,000 60,000 80,000 100,000 120,000 Potential Family Earnings 141 FIGURE 4.15 Predicted Benefit as a Proportion of Predicted Family Earnings: Two Parent Families with One Child < 7, Dual Earner Families with Child Care Costs - 1999 Taxes It is difficult to assess the degree of vertical equity from figures 4.13 to 4.16. Figures 4.17 and 4.18 assist in this evaluation by showing the predicted CCED benefit as a percentage of earnings for deciles of predicted claimant earnings, predicted family earnings and potential family earnings. A downward-sloping benefit curve is an indication of progressivity or vertical equity in the system, while an upward-sloping benefit curve is an indication of regressivity or vertical inequity. When predicted claimant earnings are used in the denominator, the 1988 benefit curve first slopes upward, indicating regressivity at lower earnings levels and then slopes downward, indicating progressivity. When predicted family earnings are used in the denominator for 1988, a similar, but less sharp pattern emerges, although the curve slopes upward again at higher earnings levels. When potential family earnings are used in the denominator, the 1988 benefit curve consistently slopes downward, indicating consistent but very moderate progressivity. 142 FIGURE 4.16 Predicted Benefit as a Proportion of Potential Family Earnings: Two Parent Families with One Child < 7, Dual Earner Families with Child Care Costs - 1999 Taxes Results for 1999 are similar to those for 1988, although all three benefit slopes turn upward at the eighth decile. While in 1988 federal surtaxes were levied on all income earners, in 1999 federal surtaxes were limited to high-income earners. Thus, an increase in the progressivity of the overall tax system has resulted in a decrease in the progressivity of the CCED. Another contributing factor could be the increase in the CCED ceiling, which primarily benefits higher income earners because they tend to spend more on child-care. Finally, since the CCED reduces net income, this increase in the CCED ceiling may also result in an increase in the child tax credit when incomes are not too high. Note that the simulations for 1999 incorporate no behavioural responses in terms of child-care expenditures that might have occurred as a result of changes in the tax system, changes to the CCED, or changes in relative prices. The new tax structure is merely applied to CPI adjusted earnings and 143 child-care costs. These costs are therefore assumed to maintain the same relationship to earnings that they had in 1988. FIGURE 4.17 Predicted CCED Benefit as Percentage of Earnings by Earnings Deciles: Two Parent Families with One Child < 7, 1988 Dual Earner Families with Child Care Costs Income Decile The Kakwani index of progressivity is included in the legends for each of the benefit curves in figures 4.17 and 4.18. This index can vary from -1 to 1. A negative value for this index is an indication of a regressive system; an increase in the index is indicates an increase in progressivity. Both a weighted and an unweighted index were calculated 1 4 7 , with the weighted index shown to the left. Note that the index confirms that the progressivity of the CCED declined from 1988 to 1999. When the analysis is restricted to working families who use paid child-care, the 1999 index indicates For a discussion o f this index and its antecedents see, Gelardi (1999. See Appendix 4.5 for information on how this index was derived. 144 that the CCED is now a neutral benefit in terms of vertical equity, while it was a slightly progressive benefit in 1988. The unweighted index also indicates that progressivity increases when ability to pay is measured in terms of potential rather than actual income. Again, it should be noted that the CCED is designed to address horizontal equity issues rather than vertical equity. FIGURE 4.18 Predicted CCED Benefit as Percentage of Earnings by Earnings Deciles: Two Parent Families with One Child < 7, Dual Earner Families with Child Care Costs - 1999 Taxes Income Decile The C N C C S data are weighted data. Using weights in the calculation o f the index results i n the repetition o f each observation n times, where n is the observation's weight. This tends to reduce the variance of the income and benefit distribution. 145 c) Equity Within Households and Women's Equality The previous analysis addresses the issue of vertical equity across households, but it does not address the issue of equity within households. Fuchs (1989) argues that women still are not faring very well in terms of economic equality because they take relatively greater interest than men in the welfare of children. He uses bargaining theory to back up the claim that programs targeted toward the welfare of children will also increase the welfare of mothers. Phipps and Burton (1996)148 reach the same conclusions. The CCED facilitates the use of paid child-care, which is likely to benefit children if the only reason parents do not use child-care is that the secondary earner cannot afford it. The CCED is designed to take into consideration an expense that arises from the decision of the secondary earner to work for pay. In the absence of fully state-funded child-care for working parents, the CCED results in higher after-tax earnings in the pockets of the low-income parent in a family. 4.7 CONCLUSION The tax treatment of child-care expenses is a complex and controversial issue. My analysis in this chapter of how the CCED fares in terms of horizontal and vertical equity shows that the CCED promotes horizontal equity and is relatively neutral in terms of vertical equity when the focus is on dual-earner families with child-care costs. Since the rationale behind the CCED is to promote horizontal equity, the deduction should not be evaluated on the basis of its vertical equity properties. Any concerns about the vertical equity of the personal income tax can be addressed directly by changes to the rate schedule. The CCED promotes horizontal equity because it recognizes that parents, engaging in market work, substitute taxed earnings for valuable but untaxed household production and/or leisure, and that some of the household production must be replaced with costly services. The CCED also recognizes They provide an excellent overview o f the literature on decision-making within households. 146 that potential income is a better measure of the household's opportunity set and ability to pay than actual income. Although the CCED promotes horizontal equity between families with children who have different labour supply choices, it does not address the issue of horizontal equity between families with children and those without. This issue can only be addressed through tax or benefit measures that favour families with children over families or individuals without children. The CCED fares reasonably well in terms of vertical equity. The analysis here of this issue is restricted to potential beneficiaries of the CCED (workers who use paid child-care) because it is recognized that vertical equity is not the objective of the CCED. The CCED also fares well in terms of vertical equity by providing a higher discretionary income for the lower-income earner within a household, which may result in higher expenditures on child-care. This is an important result if discretionary income has anything to do with distribution within the household and if increases in parental expenditures on child-care lead to an increase in the quality of child-care and the welfare of children. I did not attempt in this chapter to evaluate how the CCED fares relative to alternatives such as child-care subsidies because the policy objectives of child-care subsidies generally differ from those of the CCED. 1 4 9 Nevertheless, it is clear that if child-care costs for working parents were fully and directly funded by governments, the CCED would no longer be needed. For a discussion of child-care subsidies and their goals, see Cleveland and Hyatt (1997). 147 APPRENDIX 4.1 T A B L E 4A1.1 CCED Limits 1972-1999 I I Nominal dollars Constant 1999 dollars Max. annual deduction Weekly limit Max. annual deduction Weekly limit Year Child age < 7 8-13 Max. Family Child age < 7 8-13 Child age < 7 8-13 Max. Family Child age < 7 8-13 1972 500 500 2,000 15 15 2,119 2,119 8,475 64 64 1973 500 500 2,000 15 15 1,970 1,970 7,881 59 59 1974 500 ' 500 2,000 15 15 1,776 1,776 7,105 53 53 1975 500 500 2,000 15 15 1,603 1,603 6,413 48 48 1976 1,000 1,000 4,000 30 30 2,982 2,982 11,930 89 89 1977 1,000 1,000 4,000 30 30 2,762 2,762 11,047 83 ;83 1978 1,000 1,000 4,000 30 30 2,535 2,535 10,139 76 76 1979 1,000 1,000 4,000 30 30 2,323 2,323 9,291 70 70 1980 1,000 1,000 4,000 30 30 2,109 2,109 8,435 63 63 1981 1,000 1,000 4,000 30 30' 1,876 1,876 7,504 56 56 1982 1,000 1,000 4,000 30 30 1,692 1,692 6,769 51 51 1983 2,000 2,000 8,000 60 60 3,198 3,198 12,793 96 96 1984 2,000 2,000 8,000 60 60 3,065 .3,065 12,261 92 92 1985 2,000 2,000 8,000 60 60 2,947 2,947 11,787 88 88 1986 2,000 2,000 8,000 '60 60 2,830 . 2,830 11,319 85 85 1987 2,000 2,000 8,000 60 60 2,712 2,712 10,847 81 81 1988 4,000 2,000 n/a 120 60 5,212 2,606 n/a 156 78 1989 4,000 2,000 n/a 120 60 4,966 2,483 n/a 149 74 1990 4,000 2,000. n/a 120 60 4,737 2,369 n/a 142 .71 1991 4,000 2,000 n/a 120 60 4,487 2,244 n/a 135 67 1992 4,000 2,000 n/a 120 60 4,420 2,210 n/a 133 66 1993 5,000 3,000 n/a 150 90 5,427 3,256 n/a 163 98 1994 5,000 3,000 n/a 150 90. 5,417 3,250 n/a 163 98 1995 5,000 3,000 n/a 150 90 5,302 3,181 n/a 159 95 1996 5,000 3,000 n/a 150 90 5,217 3,130 n/a 157 94 1997 5,000 3,000 n/a 150 90 5,135 3,081 n/a 154 92 1998 7,000 4,000 n/a 175 100 7,122 4,070 n/a 178 102 1999 7,000 4,000 n/a 175 100 7,000 4,000 n/a 175 100 Note: the limits apply federally and to all provinces and territories except for Quebec. Sources: Canadian Master Tax Guide, 1975-96, T778 E (1997-1999), The National Finances, 1970-71, 1971-72. _ j : L _ 148 A P P E N D I X 4.2 The hourly cost of care equation for child-care expense claimants (t statistic in brackets) is as follows: Hourly cost = 0.793 + 0.044*dawage + 0.352*wnc02 + 0.152*wnc35 + 0.043*wnc612 (6.39) (3.03) (12.85) (6.25) (1.50) -0.013*whrssm3 + 0.000*whrss3sq - 0.003*whrs35 + 0.000*whrs35sq (-6.91) (4.93) (-1.74) (.70) -0.002*whrsbg6 + 0.000*whrsbg6q (-1.02) (1.22) where: dawage = designated adult predicted wage; wnc02 = dawage*number of children less than three years old; wnc35 = dawage*number of children three to five years old; wnc612 = dawage*number of children six to twelve years old; whrssm3 = dawage*designated adult usual hours*number of children less than three years old; whrss3sq = dawage*(designated adult usual hours)2*number of children less than three years old; whrs35 = dawage*designated adult usual hours*number of children three to five years old; whrs35sq = dawage*(designated adult usual hours)2*number of children three to five years old; whrsbg6 = dawage*designated adult usual hours*number of children six to twelve years old; whrsbg6q = dawage*(designated adult usual hours)2*number of children six to twelve years old. It is assumed that hourly cost depends on the designated adult's wage, hours of works and number of children in various age groups. Adjusted R 2 for the equation is .30. 149 APPENDIX 4.3 T A B L E 4A3.1 Log-Wage Equation Coefficients Married Single Married Single Variables Female t-stat Female t-stat Male t-stat Male t-stat Constant 1.834 60 30 1.701 49 32 2 194 27 09 1.709 21 56 Age of mother 25-34 0.208 9 48 0.383 17 39 0 234 9 42 0.416 9 24 Age of mother 35-44 0.293 13 16 0.481 22 41 0 346 7 95 0.795 16 73 Age of mother 45+ 0.233 8 20 0.423 8 99 0 325 4 42 0.420 4 45 Completed high school 0.174 11 96 0.128 4 52 0 132 5 61 0.187 4 18 Some post-secondary education 0.260 13 19 0.177 6 34 0 228 6 03 0.234 4 74 Post-secondary diploma 0.386 21 24 0.364 10 75 0 277 7 50 0.351 6 03 University degree 0.699 34 31 0.595 13 12 0 402 18 91 0.568 8 88 Number of children less than 16* -0.018 -2 14 -0.039 -3 57 0 003 0 42 -0.008 -0 60 Immigrant, speaks French or English -0.042 -2 03 -0.024 -0 80 -0 047 -2 56 -0.018 -0 32 Immigrant, other -0.098 -5 98 -0.212 -6 86 -0 115 -4 36 -0.189 -3 09 British Columbia -0.015 -0 86 -0.045 -1 61 0 004 0 20 -0.016 -0 32 Alberta -0.036 -2 02 0.007 0 26 -0 046 -1 78 0.047 1 24 Manitoba/Saskatchewan -0.110 -5 61 -0.048 -1 54 -0 128 -4 34 -0.106 -2 04 Quebec 0.009 0 63 -0.029 -1 18 -0 048 -3 94 -0.022 -0 58 Atlantic -0.021 -11 03 -0.224 -6 94 -0 218 -12 26 -0.158 -3 12 Lambda** 0.032 0 76 0.043 0 66 0 165 0 54 0.283 1 25 7177 2267 8550 1665 * While the L M A S provides information on the number of children less than 16, the corresponding variable in the CNCCS is the number of children less than 17. **Lambda is the inverse mills ratio from a probit on the probability of having worked for pay. Lambda is used to account for potential self selection bias, although the insignificant t statistic on the coefficients for lambda indicates that self selection is not an issue. ***The regression is done for the subset of the 1988 L M A S that includes individuals with children less than 16. It is restricted to individuals who have annual earnings and hours worked information. 150 APPENDIX 4.4 Table 3 shows Plotnick's Preordered Inequity Index (PII) 1 5 0 using unweighted data. 1 5 1 The index is calculated as follows. The area below the standard Lorenz curve 1 5 2 for post-tax income is deducted from the area below the preordered Lorenz curve. This difference is normalized. The preordered Lorenz curve consists of a post-tax income Lorenz curve where families are ranked according to their initial pre-tax income ranking rather than according to their post-tax income ranking. The normalization factor is twice the difference between 0.5 and the area below the standard Lorenz curve, times 100, which yields an index that ranges from zero to 100; an increase in the index indicates an increase in inequity. Plotnick calculates the PII for certain income and tax and transfer measures and finds very low values ranging from 0.2 to 1.8. In this analysis, the only transfer payments that are included are the child tax benefit and other refundable tax credits. With the majority of transfer payments excluded, the calculated indices are extremely low. Furthermore, the analysis is restricted to two-parent families with at least one child under 6 years of age; this is done to increase homogeneity and to coincide with the results shown in figures 3-6. Indices are calculated for 1988 and for 1999. There is an index for actual and for potential income, with and without the CCED 1 5 3 . The results indicate that the CCED increases horizontal equity when potential income is the measure of ability to pay, but reduces horizontal equity when actual income is the measure of ability to pay. This result is consistent with the results found using other analytical tools. TABLE 4A4.1 Preordered Inequity Index 1988 1999 Actual Potential Actual Potential Earnings Earnings Earnings Earnings Without the C C E D 0.0027 0.0124 0.0056 0.0169 With the C C E D 0.0032 0.0098 0.0060 0.0141 % Change 17% -21% 8% -17% 1 5 0 Supra footnote 48. 1 5 1 Weighted data would not allow identification of rerankings. 1 5 2 Lorenz curves are illustrated in Appendix 4.5. 1 5 3 Although changing the tax system (e.g., eliminating the CCED) would likely generate behavioural responses that would impact actual family income, there is no attempt to model these responses in this study. 151 APPENDIX 4.5 The Kakwani154 index is based on the Lorenz curve. Two such curves are illustrated in Figure 19. The index is calculated as twice the area below the benefit curve less the area below the income curve.155 In this particular case, since the two curves almost coincide, the index is nearly zero. This implies that the benefit is neutral or proportional to income. A benefit curve lying above the income curve generates a positive index, indicating progressivity of the benefit, while a benefit curve lying below the income curve generates a negative index indicating regressivity of the benefit. The index is calculated as follows: K = 2j B(T) dT - 2lY(T)dT where the integral is evaluated from 0 to 1, and T = cumulative percentage of taxpayers B(T) = Lorenz curve for CCED benefits Y(P) = Lorenz curve for pre-tax income Supra footnote 57. 5 5 This modified version of the Kakwani index is used for tax benefits rather than tax costs. For tax costs, the signs of the areas are inverted. 152 FIGURE 4A5.1 Kakwani Index - 1999 Potential Earnings 153 CHAPTER V CONCLUDING COMMENTS 5.1 CONCLUDING COMMENTS In this thesis, I examine the effects of parental labour supply and child-care on children, the impacts that child-care costs have on the labour supply of mothers, and the fairness of the tax system with respect to child-care costs. In Chapter II, I find that maternal labour supply does not have a large impact on children outcomes. I also find that the use of substitute child-care is associated with some poorer behavioural results for children aged two to five, but that it is also associated with improved prosocial behaviour for these children. In addition, I find evidence that the quality of substitute care in terms of its impact on children's cognitive outcomes is increasing in household income. I also find that children's behavioural outcomes are largely explained by the parents' own behaviour and emotional health, and by the quality of family functioning. Chapter II also provides limited evidence for the potential endogeneity of maternal labour supply in the PPVT equation and its negative impact on PPVT scores. In Chapter III, I find that labour supply of mothers is largely habit persistent. I also find that low-skilled mothers' labour supply is significantly affected by child-care costs and that low-skilled mothers appear to have limited child-care choices. In Chapter IV, I argue that in order to evaluate the CCED in terms of equity, it is important to draw the distinction between potential and actual income. This is equivalent to recognizing the value of untaxed household production. Alternatively, one can adopt the Haig-Simons definition of income. With these in mind, I find empirical evidence that the CCED supports principles of horizontal equity in the tax system. With 75 percent of married women between the ages of 25-44 working in 2000, it is clear that most women have a strong commitment to the labour force. Nevertheless, women with young children are less likely to participate, and women who leave a job to raise children experience 154 subsequent drops in earnings. Furthermore, low-skilled women are the least likely to return to work, have limited child-care choices, and are thus are most likely to eventually re-enter at even lower skill levels and rates of pay. Finally, with indications that family income levels contribute to the quality of child-care, it becomes clear that child-care policy with an emphasis on the working poor would be beneficial. However, as is pointed out in Chapter IV, the source of funding for targeted child-care policy should not be the CCED. The CCED is an important tool for the promotion of horizontal equity in the tax system. While targeted child-care policy may improve the welfare of poor women and children, it is also evident that programs that assist all families with relationship and parenting skills could go a long way in helping children's cognitive and behavioural outcomes. Furthermore, such programs would likely reduce the likelihood that children are thrown into poverty because of marital breakdowns. 155 B I B L I O G R A P H Y Altshuler, Rosanne and Amy E. Schwartz. 1996. "On the progressivity of the child-care tax credit: snapshot versus time-exposure incidence." National Tax Journal 49( 1 ):55-71. Amemyia, Takeshi. 1979. "The estimation of a simultaneous equation tobit model." International Economic Review 20( 1): 169-81. Apps, Patricia F., and Ray Rees. 1988. "Taxation and the household." Journal of Public Economics 35:355-369. . 1999a. "Individual versus joint taxation in models with household production." Journal of Political Economy 107(2):393-403. . 1999b. "On the taxation of trade within and between households." Journal of Public Economics 73:241-263. Barrow, Lisa. 1999. "An analysis of women's return-to-work decisions following first birth." Economic Inquiry 37(3):432-51. Baydar, Nazli and Jeanne Brooks-Gunn. 1991. "Effects of maternal employment and child-care arrangements on preschoolers cognitive and behavioural outcomes: evidence from the children of the National Longitudinal Survey of Youth." Development Psychology 27(6):932-945. Beaujot, Roderic. 1997. "Parental preferences for work and childcare." Canadian Public Policy, 23(3):275-88. Benjamin, Dwayne, Gunderson, Morley, and W.Craig Riddell. 1998. Labour Market Economics: Theory, Evidence, and Policy in Canada, fourth edition. Toronto: McGraw-Hill Ryerson. Berger, Marc C , and Dan A. Black. 1992. "Child-care subsidies, quality of care, and the labor supply of low-income, single mothers." Review of Economics and Statistics 74(4):635-42. Blau, David M . 1991. "The quality of child-care: an economic perspective." In The Economics of Child-care, ed. by David M . Blau, 145-73. New York: Russell Sage Foundation. . 1993. "The supply of child-care labor." Journal of Labor Economics 11(2):324-47. . 1997. "The production of quality in child-care centers." Journal of Human Resources 32:354-87 . 2000. "Child-care subsidy programs." NBER Working Paper no. 7806. Cambridge, Mass.: National Bureau of Economic Research. Blau, David M. , and H. Naci Mocan. 1999. "The supply of quality in child-care centers." NBER Working Paper no. 7225. Cambridge, Mass.: National Bureau of Economic Research. Blau, David M. , and Philip K. Robins. 1988. "Child-care costs and family labor supply." Review of Economics and Statistics, 70:374-81. . 1989. "Fertility, employment, and child-care costs." Demography 26(2):287-99. . 1991. "Child-care demand and labour supply of young mothers over time." Demography 28(3):333-51. . 1998. "A dynamic analysis of turnover in employment and child-care." Demography 35(l):83-96. Blau, David M. , and Alison P. Hagy. 1998. "The demand for quality in child-care." Journal of Political Economy 106(1): 104-46. Blau, David M. , and V. Joseph Hotz. 1992. "Special issue on child-care: introduction." Journal of Human Resources 27(1): 1-39. Blau, Francine D., and Adam J. Grossberg. 1992. "Maternal labor supply and children's cognitive development" Review of Economics and Statistics 74(3):474-81. Boessenkool Kenneth J., and James B. Davies. 1998. "Giving mom and dada a break: returning fairness to families in Canada's tax and transfer system." CD. Howe Institute Commentary, No. 117. Toronto: C D . Howe Institute. 156 Boessenkool, Kenneth J. 1999. "Putting tax policy in its place: how social policy took over the tax treatment of the family." In It Takes Two: the family in law andfinance, eds. Allen D.W., and John. Richards. Toronto: C D . Howe Institute. Boothby, Daniel. 1984. "The continuity of married women's labour force participation in Canada." Canadian Journal of Economics 17(3):471-80. Bover, Olympia. 1991. "Relaxing intertemporal separability: a rational habits model of labor supply estimated from panel data." Journal of Labor Economics 9(l):85-99. Brandon, Peter D. 1999. "Income-pooling arrangements, economic constraints, and married mothers' child-care choices." Journal of Family Issues 20(3):350-370. Brandon, Peter D. 2000. "An analysis of kin-provided child-care in the context of intrafamily exchanges: linking components of family support for parents raising young children." American Journal of Economics and Sociology 59(2)190-216. Brayfield, April. 1995. "A bargain at any price? Child-care costs and women's employment." Social Science Research 24:188-214. Brayfield, April, and Sandra L. Hofferth. 1995. "Balancing the family budget: differences in child-care expenditures by race/ethnicity, economic status, and family structure." Social Science Quarterly 76(1): 158-177. Burchinal, Margaret R. 1999. "Child-care experiences and developmental outcomes." Annals of the Academy of Political and Social Science 563:73-97. Burchinal, Margaret R., Joanne E. Roberts, Laura A. Nabors, and Donna M . Bryant. 1996. "Quality of center child-care and infant cognitive and language development." Child Development 67:606-620. Burchinal, Margaret R., Joanne E. Roberts, Rhodus Riggins, Jr., Susan A. Zeisel, Eloise Neebe, and Donna Bryant. 2000. "Relating quality of center-based child-care to early cognitive and language development longitudinally." Child Development 71(2):339-357. Canadian Child Day Care Federation and Canadian Day Care Advocacy Association. 1991. Caring for a Living. Ottawa: CCDCF. Casper, Lynne M. , and Martin O'Connell. 1998. "Work, income, the economy, and married fathers as child-care providers." 1998. Demography 35(2):243-250. CCH Canadian Limited. 1972-1989. Canadian Master Tax Guide: A Guide to Canadian Income Tax (1972-1989 editions), Don Mills: CCH Canadian Limited. Child-care Resource and Research Unit. 1994. Child-care in Canada: Provinces and Territories, 1993. Toronto: Childcare Resource and Research Unit, University of Toronto. Child-care Resource and Research Unit. 1997. Child-care in Canada: Provinces and Territories, 1995. Toronto: Childcare Resource and Research Unit, University of Toronto. Childcare Resource and Research Unit. 2000. Early Childhood Care and Education in Canada: Province and Territories, 1998. Toronto: Childcare Resource and Research Unit: University of Toronto. Chipty, Tasneem. 1995. "Economic effects of quality regulations in the day-care industry." American Economic Review 85(2):419-24. Cleveland, Gordon H., Morley Gunderson, and Douglas E. Hyatt. 1996. "Child-care costs and the employment decision of women: Canadian evidence." Canadian Journal of Economics 29(1): 132-51. Cleveland, Gordon H. and Douglas E. Hyatt. 1996. "La garde des enfants, Paide sociale et le travail: les meres seules ayant des enfants d'age prescolaire." HRDC Working Paper No. W-96-2F. Ottawa: Human Resources Development Canada. . 1997a "Subsidies to consumers or subsidies to providers: how should governments provide child-care assistance." HRDC Working Paper No. R-97-7F. Ottawa: Human Resources Development Canada. . 1997b. "Using the NLSCY to study the effects of child-care on child development." HRDC Technical Paper No. T-97-6E, Ottawa: Human Resources Development Canada 157 Cleveland, Gordon H. and Micheal Krashinsky. 1998. "The benefits and costs of good child-care: the economic rationale for public investment in young children." CRRU Monograph no. 1. Toronto: Childcare Resource and Research Unit, University of Toronto. . Connelly, Rachel. 1992. "The effect of child-care costs on married women's labor force participation." Review of Economics and Statistics 74:83-91. Connor, Sarah and Satya Brink. 1999. "The impacts of non-parental care on child development." HRDC Working Paper No. W-00-2E. Ottawa: Human Resources Development Canada Crockenberg, Susan, and Cindy Litman. 1991. "Effects of maternal employment on maternal and two-year-old child behaviour. Child Development 62:930-953. Cryer, Debby. 1999. "Defining and assessing early childhood program quality." Annals of the Academy of Political and Social Science 563:39-55. Currie, Janet, and Duncan Thomas. 1995 "Does Head Start make a difference?" American Economic Review 85:341-64. Currie, Janet. 1997a. "The intergenerational transmission of "intelligence": down the slippery slopes of "the bell curve"." CIAR Working Paper No. 106. Toronto: Canadian Institute for Advanced Research. . 1997b. "The effects of welfare on child outcomes: what we know and what we need to know." CIAR Working Paper No. 108. Toronto: Canadian Institute for Advanced Research . 1997c "Are public housing projects good for kids?" CIAR Working Paper No. 109. Toronto: Canadian Institute for Advanced Research. . 1997d " School quality and the longer-term effects of Head-Start." CIAR Working Paper No. 110. Toronto: Canadian Institute for Advanced Research. Culkin Mary, John R. Morris, and Suzanne W. Helbum. 1991. "Quality and the true cost of child-care." Journal of Social Issues 47(2):71-86. Department of Finance, Canada. 1999. Government of Canada Tax Expenditures 1999. Ottawa: Minister of Supply and Services. Dooley, Martin D., Curtis, Lori, Lipman, Ellen L., and David H. Feeny. 2000. "Child psychiatric disorders, poor school performance and social problems: the roles of family structure and low-income." In Labour Markets, Social Institutions, and the Future of Canada's Children, Miles Corak, ed. Statistics Canada, Catalogue No. 89-553-XIE. Ottawa: Statistics Canada. Duclos, Jean-Yves. 1997. "Measuring progressivity and inequality." In Research on Economic Inequality, Sourushe Zandvakili, ed. Greenwich, Connecticut: Jai Press Inc. Duclos, Jean-Yves, and Peter J. Lambert. 2000. "A normative and statistical approach to measuring classical horizontal inequity." Canadian Journal of Economics 33(1):87-113. Duncan, Alan, and Christopher Giles. 1996. "Should we subsidise pre-school childcare, and if so, how?" Fiscal Studies 17(3):39-61. Fast, Janet, and Moreno Da Pont. 1997. "Changes in women's work continuity", Canadian Social Trends, Catalogue 11-008-XPE, No. 46, Statistics Canada. Ottawa: Government of Canada. Feldstein, Martin, and Daniel R. Feenberg. 1996. "The taxation of two-earner families." In Empirical Foundations of Household Taxation, eds. Feldstein, Martin, and James M . Poterba. Chicago: The University of Chicago Press. Fox Folk, Karen, and Andrea H. Beller. 1993. "Part-time work and child-care choices for mothers of preschool children." Journal of Marriage and the Family 55:146-157. Fuchs, Victor. 1989. "Women's quest for economic equality." Journal of Economic Perspectives, 3:25-41. Garfinkel Irwin, Daniel Meyer and Patrick Wong. 1990. "The potential of child-care tax credits to reduce poverty and welfare recipiency." Population Research and Policy Review 9(l):45-63. Garfinkel, Irwin., Philip K. Robins, Patrick Wong, and Daniel Meyer. 1990. "The Wisconsin Child Support Assurance System." Journal of Human Resources 25(1): 1-31. Gelardi, Aleander M.G. 1999. "The progressivity of helping families with children: the Canadian experience, 1985-1994." Advances in Taxation, 11:61-88. 158 Gelardi, Alexander M.G. 1998. "The effect of flattening tax brackets and converting deductions to credits on the horizontal equity and progressivity: The case of the 1987 Canadian federal income tax reform." Frances in Taxation, 10:91-127. Gentry, William M . , and Alison P. Hagy. 1996. "The distributional effects of the tax treatment of child-care expenses." In Empirical Foundations of Household Taxation, eds. Feldstein, Martin, and James M . Poterba. Chicago: The University of Chicago Press. Goleman, Daniel. 1998. Working With Emotional Intelligence. New York: Bantham Books. Goodman, M . and D. Ort. 1989.1988 Tax Developments. Toronto: Canadian Tax Foundation. Gordon, Debra. 1999. "Child-care expense deduction." Canadian Tax Journal 47:1588-1626. Gornick, Janet C , Marcia K. Meyers, and Katherin E. Ross. 1998. "Public policies and the employment of mothers: a cross-national study." Social Science Quarterly 79(l):35-54. Greene, William. H. 1998. Limdep User's Manual Version 7.0, revised edition. New York: Econometric Software, Inc. Greene, William H. 1993. Econometric Analysis, 2 n d ed. Macmillan Publishing Company: New York. Greenstein, Theodore H. 1993. "Maternal employment and child behavioural outcomes." Journal of Family Issues 14:323-54. . 1995. "Are the "most advantaged" children truly disadvantaged by early maternal employment? Effects on child cognitive outcomes." Journal of Family Issues 16:149-69. Gustafsson, Siv, and Frank Stafford. 1992. "Child-care subsidies and labor supply in Sweden." Journal of Human Resources, 27(l):204-30. Hagy, Alison P. 1998. "The demand for child-care quality." Journal of Human Resources 33(3):683-710. Hanushek, Eric A. 1992. "The trade-off between child quantity and quality." Journal of Political Economy 100:84-117. Hao, Lingxin, and Arleen Leibowitz. 1994. "Public policies, private support and single mothers' schooling, work and child-care." RAND Working Paper Series 94-24. Santa Monica: RAND. Hausman, Jerry A. 1978. "Specification tests in econometrics." Econometrics 46(6): 1251-71. Head, John G., Lars Osberg, L. Green, et al., eds. 1993. Fairness in Taxation: Exploring the Principles. Toronto: University of Toronto Press. Heckman, James J. 1974a. "Effects of child-care programs on women's work effort." Journal of Political Economy. 82(2):S136-S163. . 1974b. "Shadow prices, market wages, and labor supply." Econometrica. 42(4): 679-94. . 1976. "The common structure of statistical models of truncation , sample selection and limited dependent variables and a simple estimator for such models." Annals of Economic and Social Measurement 5(4):475-492. . 1995 "Lessons from the bell curve." Journal of Political Economy 103:1091-120. Heckman, James J., and Robert J. Willis. 1977. "A beta-logistic model for the analysis of sequential labor force participation by married women." Journal of Political Economy 85(1): 27-58. Hill, M . Anne, and Jane O'Neill. 1994. "Family endowments and the achievement of young children with special reference to the underclass." Journal of Human Resources 29:1064-1099. Hofferth, Sandra L., and Douglas A. Wissoker. 1992. "Price, quality and income in child-care choice." Journal of Human Resources 27(1):70-111. Hofferth, Sandra L. 1996. "Effects of public and private policies on working after childbirth." Work and Occupations 23(4):378-404. Hogg, Peter.W., and John.E. Magee. 1997. Principles of Canadian Income Tax Law, 2 n d ed. Scarborough: Carswell. Hotz, V. Joseph, and M . Rebecca Kilburn. 1995. "Regulating child-care: the effects of state regulations on child-care demand and its cost." RAND Working Paper Series 95-03. Santa Monica: RAND. James, Simon, and Christopher Nobes. 1996. The Economics of Taxation: Principles, Policy and Practice, Hemel Hempstead: Prentice Hall Europe. 159 Jenkins, Stephen P. 1988. "Empirical measurement of horizontal inequity." Journal of Public Economics 37:305-329. Johansen, Anne S., Leibowitz, Arleen and Linda J. Waite. 1994. "Parents' demand for child-care", RAND Working Paper Series 94-13. Santa Monica: RAND. . 1996. "The importance of child-care characteristics to choice of care." Journal of Marriage and the Family 58(3):759-72. Johnson, William .R., and Jonathan Skinner. 1996. "Labor supply and marital separation." American Economic Review 76(3):455-69. Joshi, Heather. 1990. "The cash opportunity costs of childbearing: an approach to estimation using British data." Population Studies 44:41-60. Joshi, Heather, and Hugh Davies. 1992. "Day care in Europe and mothers' forgone earnings." International Labour Review 132(6):561-79. . 1993. "Mother's human capital and childcare in Britain.", National Institute Economic Review 146:50-63. Kennedy, Peter. 1992. A Guide to Econometrics, third edition. Cambridge, Mass.: MIT Press. Killingsworth, Mark. R. 1983. Labor Supply. New York: Cambridge University Press. Kimmel, Jean. 1995. "The effectiveness of child-care subsidies in encouraging the welfare-to-work transition of low-income single mothers." American Economic Review 85(2):271-75. Klerman, Jacob A., and Arleen Leibowitz. 1990. "Child-care and women's return to work after childbirth." American Economic Review, 80(2):284-88. Kohen, Dafha E., Clyde Hertzman, and Michele Wiens. 1998. "Environmental changes and children's competencies." HRDC Working Paper No. W-98-25E. Ottawa: Human Resources Development Canada Kohen, Dafha E., Clyde Hertzman, and Jeanne Brooks-Gunn. 1998. "Neighbourhood influences on children's school readiness." HRDC Working Paper No. W-98-15E. Ottawa: Human Resources Development Canada Krashinsky, Michael, and Gordon Cleveland. 1999. "Tax fairness for one-earner and two-earner families: an examination of the issues." CRPN Discussion Paper No. F|07. Ottawa: Canadian Policy Research Network. Kuhlthau, Karen, and Karen Oppenheim Mason. 1996. "Market child-care versus care by relatives." Journal of Family Issues 17(4): 561-578. Lambert, Peter J. 1995. "Equity, equality and welfare." European Economic Review. 39:674-682. Lambert, Peter J., and J. Richard Aronson. 1993. "Inequality decomposition analysis and the gini coefficient revisited." The Economic Journal 103:1221-1227. Lambert, Peter J., and Xavier Ramos. 1997. "Horizontal inequity and reranking: a review and simulation study." In Research on Economic Inequality, Sourushe Zandvakili, ed. Greenwich, Connecticut: Jai Press Inc. Lee, Lung-Fei. 1978. "Unionism and wage rates: a simultaneous equations model with qualitative and limited dependent variables." International Economic Review 19(2):415-433. Lee, Lung-Fei, G.S. Maddala, and R.P. Trost. 1980. "Asymptotic covariance matrices of two-stage probit and two-stage tobit methods for simultaneous equation models with selectivity." Econometrica 48(2):491-503. Lefebvre, Pierre, and Philip Merrigan. 1998. "Family background, family income, maternal work and child development." HRDC Working Paper No. W-98-12E. Ottawa: Human Resources Development Canada . 2000. "The effects of child-care arrangements on development outcomes of young children." Paper prepared for the Conference on Families, Labour Markets, and Weil-Being of Children, University of British Columbia, Vancouver, Canada, June 1-2, 2000. Leibowitz, Arleen., Jacob A. Klerman, and Linda J. Waite. 1992. "Employment of new mothers and child-care choice: difference by children's age." Journal of Human Resources 27(1): 112-33. 160 Leibowitz, Arleen., Linda J. Waite, and Christina Witsberger. 1988. "Child-care for preschoolers: differences by child's age." Demography 25(2):205-19. Lipman Ellen L., Michale H. Boyle, Martin D. Dooley, and David R. Offord . 1998. "Children in lone-mother families: an investigation of factors influencing child well-being." HRDC Working Paper No. W-98-1 IE. Ottawa: Human Resources Development Canada Lipps, G., and J. Yiptong-Avila. 1999. "From home to school - how Canadian children cope." Culture, Tourism and the Centre for Education Statistics, Statistics Canada, Catalogue No. 89F0117XIE. Ottawa: Statistics Canada. Lowenberg, Anton D., and Thomas D. Tinnin. 1992. "Professional versus consumer interests in regulation: the case of the US child-care industry." Applied Economics 24:571-80. Maddala, G.S. 1983. Limited-Dependent and Qualitative Variables in Econometrics. New York: Cambridge University Press. Mason, Karen O., and Karen Kuhlthau. 1992. "The perceived impact of child-care costs on women's labor supply and fertility." Demography 29(4):523-43. Maume, David J., and Karen R. Mullin. 1993. "Men's participation in child-care and women's work attachment." Social Problems 40(4):533-546. Meyers, Marcia K. 1995. "Child-care, parental choice, and consumer education in JOBS welfare-to-work programs." Social Service Review 69:679-702. Meyers, Marie K., and Theresa Heintze. 1999. "The performance of the child-care subsidy system. Social Service Review 73:37-64. Michalopoulos, Charles., Philip K. Robins, and Irwin. Garfinkel. 1992. "A structural model of labor supply and child-care demand." Journal of Human Resources 27(1): 167-203. Mocan, H. Naci. 1995. "Quality-adjusted cost functions for child-care centers." American Economic Review 85(2):409-413. More, Dennis R. 1996. "Substitute child-care at different ages: relationship to social-emotional functioning in Preschool." American Journal of Orthopsychiatry 66(2):305-308. Mott, Frank L. 1991. "Developmental effects of infant care: the mediating role of gender and health." Journal of Social Issues 47(2): 139-158. Nakamura, Alice, and Masao Nakamura. 1985a. The second paycheck: a socioeconomic analysis of earnings. Orlando, Florida: Academic Press, Inc. . 1985b. "A survey of research on the work behaviour of Canadian women." In Work and Pay: The Canadian Labour Market, ed. W. Craig Riddell. Toronto: University of Toronto Press. . 1994. "Predicting female labor supply: effects of children and recent work experience." Journal of Human Resources 29(2):304-327. Nelson, Forrest and Lawrence Olson. 1978. "Specification and estimation of a simultaneous-equation model with limited dependent variables." International Economic Review 19(3):695-709. NICHD Early Child-care Research Network. 1997. "The effects of infant child-care on infant-mother attachment security: results of the NICHD study of early child-care." Child Development 68(5):860-879. O'Brien Caughy, Margaret, Janet A. DiPietro, and Donna M . Strobino. 1994. "Day-care participation as a protective factor in the cognitive development of low-income children " Child Development 65:457-471. Pence, Alan R., Sandra Griffin, Linda McDonell, Hillel Goelman, Donna S. Lero, and Lois M . Brockman. 1997. Shared Diversity: An Interprovincial Report on Child-care in Canada. Canadian National Child-care Study. Ottawa, Ont: Statistics Canada and Human Resources Development Canada. Peters, H.E. .1996. "Marriage and divorce: informational constraints and private contracting." American Economic Review 76(3):437-54. Phillips, Meredith, and Jeanne Brooks-Gunn et al. 1998. "Family background, parenting practices and the Black-White test score gap." In The Black-White Test Score Gap, ed. C. Jencks and M . Phillips. Washington, DC: Brookings Institution Press. 161 Phipps, Shelley A. and Pierre S. Burton. 1996. "Collective models of family behaviour: implications for economic policy." Canadian Public Policy 22:129-143. Phipps, Shelly., Pierre Burton and Lynn Lethbridge. 2001. "In and Out of the Labour Market: Long-term income consequences of child-related interruptions to women's paid work." Canadian Journal of Economics 34:411-429. Plotnick, Robert. 1981. "A Measure of Horizontal Inequality." Review of Economics and Statistics 63:283-288. Posner, Jill K., and Deborah L. Vandell. "Low-income children's after-school care: are there beneficial effects of after-school programs?" Child Development 65:440-456. Powell, Lisa. 1997. "The impact of child-care costs on the labour supply of married mothers: evidence from Canada." Canadian Journal of Economics, 30(3):577-94. . 1998. "Part-time versus full-time work and child-care costs: evidence for married mothers." Applied Economics 30:503-511. Revenue Canada Taxation. 1974-1998. Taxation Statistics. Ottawa: Minister of Supply and Services. Ribar, David C. 1992. "Child-care and the labor supply of married women." Journal of Human Resources 27(1): 134-65. . "A structural model of child-care and the labor supply of married women." Journal of Labor Economics 13(3):558-97. Robins, Philip. K. 1990. "Federal financing of child-care: alternative approaches and economic implications" Population Research and Policy Review. 9:65-90. Ross, David P., Paul A. Roberts, and Katherine Scott. 1998. "Variations in child development outcomes among children living in lone-parent families." HRDC Working Paper No. W-98-7E. Ottawa: Human Resources Development Canada Ryan, Bruce A., and Gerald R. Adams . 1998. "Family relationships and children's school achievement: data from the national longitudinal survey of children and youth." HRDC Working Paper No. W-98-13E. Ottawa: Human Resources Development Canada. Ruhm, Christopher J. 2000. "Parental employment and child cognitive development." NBER Working Paper No. 7666. Cambridge, Mass: National Bureau of Economic Research Stata User's Guide Release 7 (2001). Stata Press: College Station, Texas. Statistics Canada. 1998. The Daily, Catalogue no. 11-00IE, June 9, 1998. Ottawa: Minister of Supply and Services. Statistics Canada. 1992. National Child-care Survey: Microdata Users' Guide. Ottawa: Minister of Supply and Services. Sui-Chu, Ester Ho, and J. Douglas Willms. 1996. "Effects of parental involvement on eight-grade achievement." Sociology of Education 69:126-41. Thorne Ernst & Whinney. 1989. Tax Faxts: 1988-1989. Toronto: De Boo. Treff, Karin, and David B. Perry. 1999. Finances of The Nation. Toronto: Canadian Tax Foundation. Waite Linda J., Arleen Leibowitz, and Christina Witsberger. 1991. "What parents pay for: child-care characteristics, quality, and costs." Journal of Social Issues 47(2):33-48. Walker, James R. 1992. "New evidence on the supply of child-care: a statistical portrait of family providers and an analysis of their fees." Journal of Human Resources 27(l):40-69. Woittiez, Isolde and Arie Kapteyn. 1998. "Social interactions and habit formation in a model of model of female labour supply." Journal of Public Economics 70:185-205. Wooley, Frances. 1998. "For a radical redesign of our tax treatment of the family." Policy Options September 1998:7-10. Young, Claire. 1994. "Child-care and the charter: privileging the privileged." Review of Constitutional Studies 2:20-38. Zaslow, Martha J. 1991. "Variation in child-care quality and its implications for children." Journal of Social Issues 47(2): 125-138. Zukewich Galam, Nancy. 1997. "Attitudes toward women, work and family." Canadian Social Trends, Catalogue No. 11-008-XPE, No. 46, Statistics Canada . Ottawa: Government of Canada. 162 

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:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0090537/manifest

Comment

Related Items