Open Collections

UBC Graduate Research

Vulnerable Populations: A Spatial Assessment Of Social Vulnerability to Earthquakes In Vancouver, British… Fox, Jana Christine May 31, 2008

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

Item Metadata


310-SCARP_2008_gradproject_Fox.pdf [ 1.49MB ]
JSON: 310-1.0107184.json
JSON-LD: 310-1.0107184-ld.json
RDF/XML (Pretty): 310-1.0107184-rdf.xml
RDF/JSON: 310-1.0107184-rdf.json
Turtle: 310-1.0107184-turtle.txt
N-Triples: 310-1.0107184-rdf-ntriples.txt
Original Record: 310-1.0107184-source.json
Full Text

Full Text

     VULNERABLE POPULATIONS: A SPATIAL ASSESSMENT OF SOCIAL VULNERABILITY TO EARTHQUAKES IN VANCOUVER, BRITISH COLUMBIA  by  JANA CHRISTINE FOX  B.A., Willamette University, 2005  A PROJECT SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF ARTS (PLANNING)  in  THE COLLEGE FOR INTERDISCIPLINARY STUDIES  School of Community and Regional Planning  We accept this project as conforming to the required standards   ---------------------------------------------  ---------------------------------------------  ---------------------------------------------   THE UNIVERSITY OF BRITISH COLUMBIA May 2008 © Jana Christine Fox, 2008                                                                                                                                     May 2008   Executive Summary  Spatial assessment of social vulnerability is an expanding research area in the disaster and risk management field. Social vulnerability is an important factor in understanding and effectively planning for disaster events. Physical risk, another key component of vulnerability, has been extensively studied and in many cases is well understood by planners and emergency management personnel. While the physical components of disasters are fairly well understood, the knowledge of social vulnerability is lacking in many areas. This study works to understand and assess social vulnerability to earthquakes of Vancouver, British Columbia using two existing social vulnerability indexing models, the SoVI model developed by Susan Cutter (Cutter et al. 2003) and the Pareto Ranking model developed by Lisa Rygel (Rygel et al. 2006). Both methods will be used to analyze social vulnerability in Vancouver and the results compared and explained between the two models in order to create a more complete understanding of social vulnerability.  This study reviews the relevant literature on social vulnerability indexes, specifically to define social vulnerability, review models, and identify social vulnerability indicators. The two chosen social vulnerability index models are then fully explained and used to analyze a set of base indicator data derived from the 2001 Census of Canada. The social vulnerability scores for each model are mapped at the census tract level. The resulting maps are then analyzed for significant differences between census tract scores from each method. Significant differences are then explained by looking at the underlying factors of vulnerability for that census tract. Potential planning uses of the data and further research possibilities are also discussed.                         i                                                                                                                                May 2008   Table of Contents  Terms of Reference          1  1.0  Introduction          2  2.0  Literature Review         2 2.1 Defining Social Vulnerability        2 2.2 Social Indicators         3 2.3 Social Vulnerability Indexes        3 2.4 Scale            5 2.5 Spatial Assessment of Social Vulnerability       6  3.0  The Vancouver Context        6 3.1 Earthquake Risk         7 3.2 Demographics/Influences        7  4.0  Social Vulnerability Indicators & Proxy Variables    8 4.1 Socioeconomic Status         9 4.2 Gender          9 4.3 Race & Ethnicity         9 4.4 Age           10 4.5 Employment Loss         10 4.6 Residential Property         10 4.7 Renters          11 4.8 Occupation          11 4.9 Family Structure         12 4.10 Education          12 4.11 Population Growth         13 4.12 Social Dependence         13 4.13   Summary of included Indicators       13 4.14 Excluded Indicators          14 4.15 Correlation and Interaction Between Variables     15  5.0  Methods          16 5.1 Social Vulnerability Index (SOVI) Model       16 5.2 Pareto Ranking Model         17  6.0  Analysis          19 6.1 Data Collection         19 6.2 Normalization (z-score)        20 6.3 Principal Component Analysis        21 6.4 SoVI Social Vulnerability Scores & Mapping     24 6.5 Pareto Ranking Social Vulnerability Scores & Mapping    25   ii                                                                                                                                May 2008   7.0  Findings          26 7.1 Highly Vulnerable Area        27 7.2 High Social Vulnerability Scores: Differences Between Models   30  8.0  Next Steps/Planning Implications       31 8.1 Planning Implications         31 8.2 Planning Recommendations        32 8.3 Further Research         32  9.0  Conclusion          33   Bibliography           34  Appendix 1: Factor Maps         37  Appendix 2: SoVI & Pareto Ranking Scores by Census Tract    41   List of Tables Table 1: Social Vulnerability Indicators and Proxy Variables    8 Table 2: Descriptive Statistics for Social Vulnerability Variables    14 Table 3: Total Variance Explained by Factors      21 Table 4: Rotated Factor Matrix        22 Table 5: Factors and Variable Loading       23  List of Figures Figure 1: Pareto Vulnerability Space of Census Tract ‘A’     5 Figure 2: Pareto Ranking Example        19 Figure 3: Factor Scores by Highly Vulnerable Census Tract     29  List of Maps Map 1: SoVI Social Vulnerability Map by Census Tract, City of Vancouver  25 Map 2: Pareto Ranking Social Vulnerability Map by Census Tract, City of Vancouver 26 Map 3: High Social Vulnerability Census Tracts      28     iii                                                                                                                                May 2008   Terms of Reference This project uses established social vulnerability models to assess social vulnerability in Vancouver, British Columbia. The focus is on earthquakes and the factors that increase social vulnerability in the context of an earthquake event. Analysis is done at the census tract level for the City of Vancouver and the University Endowment Lands. The two methods that are used, SoVI and Pareto Ranking, are chosen as they place emphasis on different ways of measuring vulnerability. By using two methods and comparing them, it is expected that those census tracts that might not have been identified as vulnerable by one method may be by the other. A comparative approach is also expected to yield insight into the robustness of findings across methods and the strengths and limitations of each method. This project does not assess the social vulnerability of individuals or families but census tracts as a whole.                                   1                                                                                                                                May 2008   1.0 Introduction  Currently, natural hazard and emergency management planning often lacks a clear understanding of social vulnerability. Without a clear understanding of the causes of social vulnerability, it is impossible to adequately plan for disasters and work to truly reduce vulnerability. Most vulnerability studies focus on the physical hazards that threaten a community and mitigation measures aim to simply address the physical components of risk. While these mitigation measure are absolutely necessary, without addressing the social factors that cause individuals, households, and communities to be more vulnerable, planning and mitigation measures will never reach their full potential. In order to create truly resilient communities that can adapt and respond to risk, both physical and social vulnerabilities must be understood, planned, and mitigated for. This project explores social vulnerability in the context of a municipality and aims to answer the question: what areas of Vancouver, British Columbia, have a higher relative level of social vulnerability to earthquakes and what underlying factors cause those increased levels of social vulnerability.  The results of this study could be used in a number of different ways to decrease social vulnerability. The base data developed by this study can be used in conjunction with other existing data on service levels and critical facilities to identify potential gaps in critical services. The maps may also be used to help develop target programs to decrease social vulnerability in areas that need it. The range of information available from these studies can help to develop more effective programs by targeting the underlying causes of social vulnerability in an area.   2.0 Literature Review  2.1 Defining Social Vulnerability Vulnerability is an often yet inconsistently used term within the disaster literature. The term is also widely used in other social science disciplines, which can lead to confusion between as well as within disciplines (Alwang et al. 2001). The definition of vulnerability that is chosen for a study is integral to the indicators chosen to quantify vulnerability and therefore must be clearly expressed (Rygel et al. 2006).  Specific to the disaster literature there are three main models of assessing vulnerability used in research (Cutter 1996, 2001, 2003; Dow 1992). The first model is an exposure and location oriented model, focusing on the physical conditions that make people or places vulnerable (Anderson 2000; Burton et al. 1993). This model of vulnerability fails to adequately take into account the social conditions that may add to the vulnerability of people or places by assuming that vulnerability is a pre-existing condition (Rygel et al. 2006). The second model focuses on vulnerability as a social condition and assumes that people display different patters of loss and recovery from disasters (Rygel et al. 2006; Hewitt 1997; Blaikie et al. 1994). This social condition focused model does not adequately identify physical location as an aspect of vulnerability. The third model weighs both social and physical conditions that affect vulnerability and resilience to 2                                                                                                                                May 2008   hazards (Rygel et al. 2006; Cutter et al. 2000, 2003; Kasperson et al. 1995). Cutter et al’s definition of vulnerability that arises from this third model provides a more complete picture of vulnerability and one that is relevant to this study.  “Social vulnerability is partially the product of social inequalities-those social factors that influence or shape the susceptibility of various groups to harm and that also govern their ability to respond. However it also includes place inequalities-those characteristics of communities and the built environment, such as the level of urbanization, growth rates, and economic vitality, that contribute to the social vulnerability of place.” (Cutter et al. 2003)  A number of researchers (Kleinosky et al. 2007; Rygel et al. 2006; Wu et al. 2002; Cutter et al. 2000; Yarnal 1994) have conducted case studies using the vulnerability of place model described above. The vulnerability of place model has the ability to provide for a spatially oriented form of social vulnerability analysis that can be used in conjunction with existing hazard data to form a more complete picture of social vulnerability for disaster planning purposes.  2.2 Social Indicators Social indicators have been used since the 1960s for social science research in areas of policy, environmental indicators, and quality-of-life indexes (Cutter 1985, 2003). Much of the current social indicator research is focused around popular ratings on quality of life and quality-of-place; for example studies such as America’s Top-Rated Cities (Garoogian 2005), Green Index (Hall and Kerr 1991), Green Metro Index (World Resources Institute 1993) and The Social Health of the Nation: How America is Really Doing (Miringhoff 1999). The most recent shift in social indicator studies has been primarily to the sustainability field with studies such as the United Nations Development Program’s Human Development Index which measures well-being (UNDP 2000). Despite the growing use of social indicators, there is no clearly defined set of indicators that measures human well-being or social vulnerability, and each study must choose and justify its own indicators based upon study goals (Cutter et al. 2003).  2.3 Social Vulnerability Indexes The quantification of social vulnerability is a complex process as the many manifestations of vulnerability are difficult to assess and proxy variables are necessary to quantify these indicators. Standard indicators that are included in most indexes are: gender, ethnicity, age, and income. A number of methods for creating vulnerability indexes have been developed; each has its own set of advantages and disadvantages. Most indexes create one end value that is comparable across geographic locations (Cutter et al. 2000. 2003, Rygel et al. 2006, Adger et al. 2004, Alwang et al. 2001). The most common types of analyses use non-weighted composite indexes. However, a method for using Pareto ranking has recently been applied to social vulnerability studies by Rygel et al. (2006).  3                                                                                                                                May 2008   The distinction between weighted and non-weighted indicators is key to understanding vulnerability studies. Weighted indicators require a subjective decision by the researchers as to what variables have more affect on social vulnerability. Those decisions must be adequately justified in order to validate the study results. Weighting can be a difficult process and is one of the leading reasons why many studies use non-weighted averaged indicators (Dwyer et al. 2004; Cutter et al. 2003, 2000; Clark et al. 1998). Simple averaged indicator studies can obscure extreme high or low scores which may be more important indicators for hazard planning than the overall composite scores (Rygel et al. 2006). Unweighted indicator indexes also carry bias in their assumptions, as equalizing all scores assumes that all indicators contribute equally to vulnerability, an assumption that does not hold true in most cases.  A simple process for index development is laid out in Cutter et al.’s Handbook for Conducting a GIS-Based Hazards Assessment at the County Level (1997) which was developed for the South Carolina Emergency Preparedness Division. This tool was developed for practitioners and avoids the more complex statistical analysis used in other studies. Although the process may vary slightly from study to study the general principles remain the same for most social vulnerability analyses and this method serves as an illustrative tool for basic social vulnerability analysis. The process involves first identifying vulnerability indicators and selecting the appropriate proxy variables for the vulnerability indicators. The second step is the acquisition of data, generally through the Census or similar demographic databases. In this case a fairly straightforward process of index calculation is used on the collected data. The first step in index calculation is to normalize the data to a value ranging from zero to one so that each indicator has equal weight. Cutter does this by dividing the indicator score (Xn) a maximum score for that indicator (Xmax) so that all values range between 0 and 1, with 1 signifying the most vulnerable subarea for that indicator, as shown below.  Normalized Equation:  Xn                 Xmax  Xn       = value of X for subregion n Xmax = the maximum value for X in the dataset of all subregions  For the final step, the indicator values are then summed to give a final vulnerability score to each geographic sub-region (Cutter et al. 1997). This represents the most basic approach to social vulnerability indexes; although it provides a rough sketch that may be helpful to planners, it does not provide a highly accurate picture in most cases.  In contrast to the summing approach used by Cutter et al. (1997), Rygel et al. (2006) uses a Pareto ranking system to organize data groups (i.e., groups of a subregion) into ranks. The advantage of the Pareto ranking system is that high value indicators are not obscured by the composite or averaging process. The Pareto method involves grouping data into sets using the principal of non-domination. If a datapoint is not dominated by any other 4                                                                                                                                May 2008   datapoint then it is grouped into Set One and then removed from the data set. The process is then repeated using the remaining data until all data has been placed into sets. The Pareto method assumes that a high score on any individual indicator is an indication of greater overall vulnerability (Rygel et al.  2006). Subregions in Set One are considered the most vulnerable, those in Set Two are the next vulnerable and so on. Figure 1 below shows the vulnerability spaces of point ‘A’. Other points located within the more vulnerable region relative to point ‘A’ would be given a higher vulnerability rank. Those points within the non-comparable areas would be given the same vulnerability rank as point ‘A’. Those points in the less vulnerable area are dominated by point ‘A’ and would be given lower vulnerability ranks.  Figure 1: Pareto Vulnerability Space of Census Tract ‘A’    Not Comparable  A   More Vulnerable    Less Vulnerable   Not Comparable                                                Component 1 C om po ne nt  2                                           Adapted from Rygel et al 2006  2.4 Scale Spatial and human scales play a central role in designing and implementing an effective social vulnerability assessment. Clearly defining a spatial scale is key for a social vulnerability index. The factors that affect vulnerability in a comparative study of larger areas, such as cities or regions, are not always the same factors that can be used to compare social vulnerability at a smaller scale, such as at the census tract level within a city. For example a study comparing the relative social vulnerability of regions in British Columbia may take into account the diversity of the economic base.  However data on economic diversity is difficult to deal with at the census tract level, proxy indicators are often not available at such a small level of detail, and if they are available they may be misleading. One area within a region will have different types of industry and development but they generally do not have an affect on the social vulnerability of the people living in the census tract in which the industry is located. However such data on a broader city wide level can provide insight into the ability of a city to recover from a disaster, a key factor in social vulnerability.  The human scale of social vulnerability is equally as important to understand as the spatial scale. While factors for vulnerability may change over the spatial scale they also change depending on the human scale. Social vulnerability can be measured at a number 5                                                                                                                                May 2008   of different levels; the person, household, or aggregate. Person and household scaled vulnerability studies are valuable to look at for underlying causes of vulnerability. However, data for these types of studies must be collected at the individual level, which can be quite difficult and time consuming. Along with quantitative data, qualitative data may also be used to assess vulnerability at this small scale. This type of study is very useful in understanding underlying causes of vulnerability and contributes greatly to the understanding of social vulnerability. However, for many planning agencies undertaking this type of analysis is too time consuming and resource intensive.  Social vulnerability studies at the aggregate level provide a way to assess social vulnerability based on existing or readily accessible data. Census data can provide most, if not all, of the data needed for conducting an aggregate level social vulnerability study. While this type of analysis can be very useful, it is important to understand that within the aggregated level, the social vulnerability of the people and households living there will range significantly.  Aggregation is used in all social vulnerability indexes. However, when interpreting aggregated data, it is important to consider that social vulnerability may spatially vary within the aggregated area. These studies are also appropriate for and can be tailored to different types of questions or policies, for example neighborhood prioritization.  2.5 Spatial Assessment of Social Vulnerability The spatial orientation of social vulnerability within a geographical area provides an easy way for planners and decision makers to identify areas that may need additional planning in order to prepare for disasters. Social vulnerability assessments involving indexes are often mapped as a way to visually represent the resulting data (Rygel et al. 2006; Dwyer et al. 2004; Cutter et al. 2003, 2000; Clark et al. 1998). In these studies, the index calculations are completed, grouped into classifications, and then mapped, generally using Geographic Information Systems (GIS) software. A second method of spatial assessment involves creating GIS layers for vulnerability indicators and then looking for geographic areas with high concentrations of vulnerable populations through spatial overlay analysis (Morrow 1999). This method can also be used in developing countries and areas without GIS capabilities through hand drawn maps and overlays. Ii sum, some of the key decisions that need to be made when conduction social vulnerability analyses are: the spatial scale of the assessment, the specific uses the study will have, and the social vulnerability indicators that apply to the study.   3.0 The Vancouver Context  The major social vulnerability index models all use indicators developed for the United States. While there are overwhelming similarities between the United States and Canada, not all of the same indicators are appropriate in both locations. In order to have an accurate social vulnerability index the indicators and proxy variables must take into account the unique conditions of the study location. Vancouver has distinct demographic patterns and influences that require special attention when developing a social vulnerability index. 6                                                                                                                                May 2008   3.1 Earthquake Risk Although Vancouver has not had a significant damaging earthquake in recent memory, this does not mean that Vancouver is not at risk. Vancouver is at risk for three types of earthquakes; megathrust, crustal, and intraslab earthquakes. A megathrust earthquake in the Vancouver area would occur in the Cascadia Subduction Zone which runs off the Pacific Coast from British Columbia to Northern California. The last Cascadia Subduction Zone earthquake occurred in 1700. These megathrust earthquakes, like the one that caused the Indian Ocean Tsunami in 2004, are very large earthquake events with a long duration period of two to four minutes. Crustal earthquakes can also be severe and occur within the North America Plate at a shallower depth than intraslab earthquakes, and are therefore more harmful to human settlements. The Northridge, Loma Prieta, and Kobe earthquakes were all crustal earthquakes and caused major damage to highly urbanized areas. Intraslab earthquakes take place in the Juan de Fuca Plate and are at a greater depth, the Nisqually earthquake in the Seattle area in 2001 is an example of a intraslab earthquake that occurred recently in the region. The high level of seismic activity in British Columbia and Vancouver specifically means planners must be prepared for large earthquake events which require planning for physical as well as social vulnerabilities (British Columbia Provincial Emergency Program 1999; Natural Resources Canada).  3.2 Demographics/Influences Vancouver has a broad and diverse population base. A significant portion of the Vancouver Census Metropolitan Area (CMA) has a mother tongue other than English or French, upwards of 40% (Statistics Canada, population 2006). The Vancouver CMA is a geographical area defined by Statistics Canada for the census and includes the same municipalities that are part of Metro Vancouver. Cantonese, Mandarin, Chinese (non major dialects) and Punjabi are the most predominant languages spoken besides English in the Vancouver CMA. A fairly large percentage of residents of Vancouver CMA are international migrants, around 8% have immigrated to Vancouver between 2001 and 2006 (Statistics Canada, 2006). This is a large percentage of the population which may not yet be acclimatized to the area and in the case of an earthquake may have problems accessing services. While there have been large numbers of recent immigrants to Vancouver, there are also well established visible minority groups from a broad range of countries. These groups can provide assistance to new immigrants help them to acclimatize, and teach them how to access necessary services as well as simply provide social support. This structure allows for minority status to be less of a determinant of vulnerability than it may be in other areas with smaller minority populations. The City of Vancouver has also embraced the concept of diversity and offers many services in multiple languages in order to include a broad range of the population. This set of social factors is quite different from the factors influencing the areas that the original SoVI and Pareto Ranking studies took place such as Hampton Roads, Virginia and Revere, Massachusetts.     7                                                                                                                                May 2008   4.0 Social Vulnerability Indicators & Proxy Variables  Despite the lack of agreed upon social vulnerability indicators a number of common indicators emerge in the literature. Cutter et al. (2003) provides a clearly outlined description of standard indicators of social vulnerability and justifies their inclusion. The following discussion identifies commonly used measures of social vulnerability and indicators that have been chosen to quantify the level of social vulnerability to earthquakes in this study. As social vulnerability is a complex issue and each person’s vulnerability is different it is nearly impossible to create a complete measure of social vulnerability for any given area due simply to lack of available data. It is equally important to understand that even though a person may fall into one or more of the categories that are used to identify social vulnerability it does not mean that he or she as an individual is necessarily more vulnerable, as vulnerability is a measure of many different factors that affect individuals (Morrow 1999). Vulnerability is complex and many different combinations of factors can increase a person’s social vulnerability. The factors can vary from person to person and are quite difficult to assess. Table 1 below summarizes the vulnerability indicators and proxy variables chosen for this study.  Table 1: Social Vulnerability Indicators and Proxy Variables  Indicator Variable Socioeconomic Status % low income population Gender % women making less than $20,000 per year Race & Ethnicity % Population without knowledge of English Age % Population less than 19 years old  % Population 65 years and over Employment Loss Unemployment Rate Residential Property % Apartment 5+ storey built before 1980  % dwellings needing major repairs Renters % dwellings rented Occupation % population not employed as professionals or management Family Structure Average Household Size  % Single Parent Families Education % Population (20years+) with High School or less education level Population Growth % Population Change from 1996-2001 Social Dependence  % total income from government transfer payments 8                                                                                                                                May 2008   The rationale for these indicators and variables is provided below. For all variables increased values indicate higher levels of social vulnerability.  4.1 Socioeconomic Status Socioeconomic factors have a large effect on social vulnerability (Rygel et al. 2006; Dwyer et al. 2004; Cutter et al. 2003, 2000, 1997; Morrow 1999; Mileti 1999; Blakie et al. 1994). Indicators such as education level, income level, job type, and political power all affect people’s ability to respond to and recover from a disaster. Income, education level and job type are the often used indicators of social vulnerability caused by socioeconomic factors.  In this study the percentage of low income population per census tract is used as the proxy variable. This information is available from the Census of Canada as a percentage of economic families and unattached individuals. A higher percentage of low income individuals and families indicates higher social vulnerability of the census tract.  4.2 Gender Women have been shown to often be in positions of increased vulnerability during a disaster (Rygel et al. 2006; Dwyer et al. 2004; Cutter et al. 2003, 2000, 1997; Morrow 1999; Mileti 1999; Morrow and Enarson 1996). The primary factors affecting women’s vulnerability are lower wages, informal economy jobs, and family care responsibilities. The standard variable used for gender is the percentage of women but I have chosen a more refined variable that is explained below.  The proxy variable chosen to represent gender is the percentage of women fifteen years and over who are making $20,000 or less. This was calculated by adding the number of women in the income brackets which were $20,000 or less per census tract and dividing that by the total number of women fifteen and over for that census tract. This proxy variable was chosen because it represents one of the major factors that make women more vulnerable, low income status. The inclusion of gender is not because women are inherently more vulnerable but that their roles in society and within their families often place them within a context of higher vulnerability. It is important to note that there are cases where women do not work because they do not need to which can affect the accuracy of this variable. Overall this variables is more accurate in gauging a cause of vulnerability than using the percentage of women as the proxy variable.  4.3 Race & Ethnicity Ethnicity and language barriers can increase social vulnerability (Rygel et al. 2006; Dwyer et al. 2004; Cutter et al. 2003, 2000, 1997; Morrow 1999; Mileti 1999; Peacock and Gerard 1997). Language barriers can play a role in both pre and post disaster periods. Those with limited English skills may not be able to understand or may not trust government warnings. Post-disaster, it may be difficult for some minority groups to access services due to lack of ability to communicate. The number of ethnic minorities and non-English speakers in a community are generally used as indicators for measuring social vulnerability.  9                                                                                                                                May 2008   The proxy variable for race and ethnicity in this study is the percentage of the population without knowledge of English. This proxy variable was created by dividing the number of people who can speak English (aggregate of population that has knowledge of English and population that has knowledge English and French) by the population of the census tract to convert to a percentage of the population that speaks English, that percentage was then subtracted from one hundred percent to produce the percentage of the population that does not speak English. The higher the non-English speaking population, the higher the social vulnerability level for that census tract.  4.4 Age Age is one of the most commonly used indicators of social vulnerability (Rygel et al. 2006; Dwyer et al. 2004; Cutter et al. 2003, 2000, 1997; Morrow 1999). Within the age category two distinct groups stand out as more vulnerable to disasters: children and the elderly. Children must depend on adults during times of disasters and often require special assistance or care during emergencies. Similarly the elderly are less likely to have the same mobility and resources available to the adult population which can make them more vulnerable (Ngo 2001, Mileti 1999).  Two proxy variables were used for age: the percentages of the population nineteen years of age and younger, and sixty-five years of age and older. The age of nineteen was used as it is the age of majority in British Columbia. Sixty-five was used as the older cutoff as it is a typical retirement age. Percentages were calculated by dividing the population nineteen and under by the total population. The same procedure was used for calculating the percentage of the population sixty-five and over. An increased percentage of elderly or youth within a census tract indicates higher social vulnerability.  4.5 Employment Loss Employment loss both pre and post disaster play an important role in determining the recovery period of an area (Cutter et al. 2003, Mileti 1999). Post disaster employment loss can be severe in hard hit places where major industry has been destroyed. Areas with previously high levels of employment loss may have even longer economic recovery times. High levels of unemployment put a strain on resources and assistance programs within a community. The unemployment rate prior to a disaster is a way of gauging the employment loss in an area prior to a disaster as an aspect of social vulnerability.  The proxy variable for employment loss in this study is the unemployment rate. Statistics Canada calculates the unemployment rate for each census tract. That number was used directly for this study. The higher the unemployment rate per census tract the higher the social vulnerability.  4.6 Residential Property A number of aspects of residential property affect social vulnerability. Value, quality of construction, and earthquake engineering are significant in determining vulnerability of housing stock during a disaster (Cutter et al. 2003, 2000; Heinz Center for Science, Economics and the Environment 2000; Bolin et al. 1991). Properties with low quality of construction or mobile homes are often severely damaged or destroyed in an earthquake, 10                                                                                                                                May 2008   leading to extreme losses for the people living in those types of housing. Areas with higher rise buildings that were built before seismic codes are also at higher risk during earthquake events.  The proxy variable for residential property that was chosen for this study was the percentage of apartment buildings five or more stories high built before 1980. This variable was chosen as buildings that are 4 stories or less are most often wood frame and often fare better during earthquakes than buildings not built to recent seismic code that are concrete or brick buildings. Seismic codes were originally introduced in BC in the early 1980’s and it can be assumed that higher rise dwelling built since that time meet at least minimum seismic standards and can perform reasonably well in an earthquake event (Finn, 2004). The information for this variable had to be created by multiplying the percentage of high rise apartments per census tract by the percentage of buildings built before 1980 in order to create a proportion of apartments built before 1980. There are some problems with this variable in that it assumes that the rate of construction for all forms of dwelling are the same, but as there are no census variables that directly correlate building date and dwelling type it is the best proxy measure available. The higher the percentage of high rise buildings built before 1980 the higher the vulnerability of the census tract.  The census variable dwellings requiring major repairs was used as the second proxy variable to measure residential property. This is a self-reported category where those living in a dwelling unit assess the level of repair necessary to that dwelling unit. Buildings already in disrepair may be further damaged by an earthquake beyond their better constructed counterparts. The higher the percentage of dwellings requiring major repairs the higher the social vulnerability of the census tract.  4.7 Renters The number of renters plays a role in indicating social vulnerability (Cutter et al., 2003; Heinz Center for Science, Economics, and the Environment, 2000; Morrow, 1999) It is assumed that people rent because they are unable to purchase a house or they are transient. Renters may also have less adequate insurance or be completely uninsured leading to larger losses. Renters may also be priced out of the market quickly if rents rise post-disaster due to lack of adequate housing to meet the populations needs.  The proxy variable for renters is straightforward. The Census provides information on the number of households renting per census tract which can easily be converted to a percentage of households renting per census tract which is comparable across areas. A higher percentage of renters indicated higher social vulnerability.  4.8 Occupation Occupation can play a key role in indicating social vulnerability. Post-disaster there is often a loss of jobs, particularly in the unskilled, informal and service sectors of the economy. As disposable incomes decrease after a disaster as people lose jobs or are using their disposable income for personal recovery, the demand for services that are not 11                                                                                                                                May 2008   absolutely necessary drops. Professionals and managerial positions tend to be more insulated against these economic shocks post disaster.  In order to construct a proxy variable for occupation the percentage of the population over fifteen (the census age measurement for employment) not employed in professional or managerial positions was used. This was done by subtracting the number of people in professional and managerial positions from the population fifteen and over for that census tract and then dividing it by the population fifteen and over. The higher the percentage not employed in professional or managerial positions, the higher the social vulnerability for the census tract.  4.9 Family Structure Family structure plays a significant role in social vulnerability (Rygel et al. 2006; Dwyer et al. 2004; Cutter et al. 2003, 2000, 1997; Morrow 1999; Mileti 1999, Blakie et al. 1994). Family structures are complex and many different types of structures can contribute to increased vulnerability. A large number of children, indicated by high birth rates, can stretch financial resources and make it more difficult for families to recover from disasters. Single parent families are also more vulnerable due to single income and child care responsibilities in addition to work.  The percentage of lone parent families is used as the proxy indicator for family structure. The number of lone parent families is divided by the total number of families for each census tract in order to calculate percentage of lone parent families. The higher the percentage of lone parent families, the higher the social vulnerability of that census tract.  Average family size is used as the second proxy variable for family structure. The average family size is calculated by the census and available at the census tract level. The higher the average family size, the higher the social vulnerability of that census tract.  4.10 Education Education is correlated closely with socioeconomic status (Cutter et al, 2000; Heinz Center for Science, Economics, and Environment, 2000). Those with higher levels of education have higher lifetime earning potential. Those with lower education levels may have lower incomes as well as difficulty understanding emergency information and accessing recovery services.  The proxy variable used for education is the percentage of the population nineteen year of age and over with a high school level education or less. Anyone who has attended any level of schooling past high school is not included in this category, even if they did not complete another level of education. This was calculated by combining the population nineteen and over at each level from completed high school level and below and dividing by the overall population nineteen and over for that census tract. The higher the percentage of the population with a high school education or less, the higher the social vulnerability score for that census tract.   12                                                                                                                                May 2008   4.11 Population Growth High levels of population growth may increase vulnerability if social services, infrastructure, and housing stock are not adequate to deal with the influx in population (Cutter et al. 2003, 2000; Morrow 1999). This population growth can be indicated in the social vulnerability index by using growth rates.  Population growth is calculated by the census in five year intervals for each census tract. This growth is represented as a percentage and that percentage has been used directly in this study. The assumption is that the greater the population growth the higher the social vulnerability for that census tract.  4.12 Social Dependence A high degree of social dependence is a factor in social vulnerability (Cutter et al. 2003; Morrow 1999; Heinz Center for Science, Economics, and the Environment 2000; Drabek, 1996; Hewitt 2000). People who are socially dependent rely on social services such as income transfers, social housing, and other government social programs. Those who are dependent on social services generally require additional support in a post-disaster period, further exacerbating recovery.  In order to gauge social dependence, the used percentage of income from government transfer payments for each census tract was used. The data are available directly from the census and must simply be turned into a percentage of income for each census tract. A higher percentage of income from government transfer payments indicates higher levels of social vulnerability.  4.13   Summary of included Indicators Table 2 below shows the basic descriptive statistics for each proxy variable used in this study. Of the fifteen indicators chosen for this study most displayed a wide range of variance between the minimum and maximum values across census tracts. The largest variance between census tracts existed in the percentage of dwellings rented indicator. Percentage of apartments five or more stories built before 1980 and the percentage of the population with a high school education or less also varied significantly. The average household size had the smallest standard deviation. The unemployment rate and percentage of dwellings in need of major repairs also had small standard deviations.             13                                                                                                                                May 2008   Table 2: Descriptive Statistics for Social Vulnerability Variables    N Minimum Maximum Mean Std. Deviation Incidence of low income in 2000 % 106 7.70 80.00 27.26 11.50607 Government transfer payments % - composition of total income in 2000 106 3.00 53.30 11.69 7.84793 % Women making less than $20,000 per year 106 29.83 87.45 54.48 12.55076 Unemployment rate, population 15 years and over 106 3.30 28.70 8.57 3.94578 % 19 & U 106 2.60 27.66 18.14 7.25784 % 65+ 106 5.05 34.51 13.17 5.20261 % apartment 5+ storey before 1980 106 .00 77.39 10.07 18.42572 % Dwellings rented 106 15.10 95.47 51.99 20.70222 % dwellings needing major repairs 106 1.50 22.17 9.42 4.07635 Average number of persons in private households 106 1.10 3.70 2.44 .69131 % lone parent families 106 6.47 38.10 16.72 5.47124 % Pop 20+ w/ HS or less education 106 4.88 72.15 30.84 15.14707 Population percentage change, 1996-2001 106 -8.00 59.90 5.20 10.27764 % Population without knowledge of English 106 .31 42.10 7.82 6.68246 % Population not employed as professionals or management 106 31.62 89.44 60.96 14.93396  4.14 Excluded Indicators There are a number of indicators that are used in many different studies that have been excluded from this study as they do not apply to the Vancouver context, the spatial level of analysis did not fit, or the data were not available. The level of commercial and industrial development, infrastructure lifelines, medical services, rural/urban development, and special needs populations are among those indicators not included in this study but are discussed below. These indicators can be included in the planning process in different ways but are difficult to quantify for an index at the census tract level.  Special needs populations, such as the infirm and mentally ill, present a unique challenge in response and recovery. These populations are often not adequately considered in planning processes and may be overlooked during a disaster. Special needs populations often require assistance to perform a number of tasks and would need additional assistance in a disaster situation. The population of special needs persons in an area can serve as an indicator of social vulnerability (Cutter et al. 2003; Morrow 1999).  The data for special needs population were not available at the census tract level and therefore not 14                                                                                                                                May 2008   included in this study. However, later in the planning process, the location of facilities that house special needs populations can be mapped as an overlay to a social vulnerability index.  Rural and urban environments can both contribute to increased social vulnerability in different ways (Cutter et al. 2003, 2000). Rural environments tend to have a higher concentration of poor, children, and elderly populations. These areas also often rely on resource based economies which are highly vulnerable to disasters. Highly dense urban environments can be more difficult to evacuate in an emergency which increases vulnerability. Indicators for rural environments are generally not direct and proxies such as age and income are used. However population density can be used to represent high density areas. Although population density can play a key role, the level of urbanization throughout Vancouver is high but the corresponding level of infrastructure development is, as well. Issues of evacuation planning can be dealt with separately and be informed by social vulnerability but are difficult to place within the context of an index.  The level of commercial and industrial areas is important for assessing social vulnerability over large spatial areas, such as cities or regions. However, at the census tract level, these levels will vary significantly while not having an effect on the social vulnerability of the population. The Vancouver population has a wide range of access to commercial and industrial properties as the region is well integrated. Areas without significant commercial or industrial bases, such as resource based economies, can be devastated in the aftermath of a disaster. Without a broad range of industry and services, it is hard to recover after a disaster.  Infrastructure lifelines represent important service functions on a day-to-day level as well as during a disaster. Without functioning infrastructure response and recovery processes can be severely hindered. While this is an important consideration, it is one better assessed on a larger scale than at the census tract level. Vancouver enjoys a high level of service in all major infrastructure sectors and service levels do not vary significantly throughout the city. In addition, information on vulnerable areas is difficult to obtain from most providers as they are concerned about threats to their infrastructure.  Access to medical services is a very important part of response and recovery effort, yet a difficult indicator to quantify. Without adequate medical services in a disaster a city would have difficulty recovering, as demonstrated in New Orleans with Hurricane Katrina. However the way that the system is set up with a number of larger facilities within Vancouver, as well as smaller local clinics there is a broad network of medical services available. Post disaster some areas located near larger medical centers may have better access to medical services but this relationship is tenuous and hard to quantify. For this study it is assumed that a fairly even level of access to medical services is available throughout Vancouver.  4.15 Correlation and Interaction Between Variables One of the important factors to consider when conducting a social vulnerability assessment is the correlation and interaction between variables. In studies with large 15                                                                                                                                May 2008   numbers of variables there are often variables that measure essentially the same or very similar things. If too many variables that measure the same thing are chosen to be used in the study there can be too much emphasis on one vulnerability indicator and it can overshadow other important indicators. While this study selects a small number of variables and attempts to reduce the correlation as much as possible through the statistical technique of Principal Component Analysis (see section 5 on methods below) there is still some crossover in what the variables are measuring. For example the ‘women making less than $20,000 a year’ variable and ‘people in low income’ variables both measure the amount of people who are making small amounts of money that would be difficult to live off of. However the aim of the two variables is quite different. The ‘women making less than $20,000 a year’ variable focuses on measure a way that women are more vulnerable because the percentage of women is not a reliable proxy variable.   5.0 Methods  Social vulnerability is a complex and evolving topic with few well established standard quantitative indicators. However a number of studies have been conducted on quantifying and mapping social vulnerability, which allows for a comparison between methods. The hazard-of-place model (Cutter et al. 2000) provides a framework for assessing social vulnerability in the context of place. This type of assessment with a refined set of indicators that takes into account the unique social, economic, and special characteristics of Vancouver, British Columbia will be used to answer the question: What geographical areas of Vancouver have higher levels of social vulnerability? In order to answer this question two models within the hazards of place framework will be used to analyze Vancouver’s social vulnerability to earthquakes and then compared: Cutter’s SoVI model and Rygel’s Pareto Ranking model.  5.1 Social Vulnerability Index (SoVI) Model The Social Vulnerability Index (SoVI) was developed at the University of South Carolina by Susan Cutter, Bryan Boruff, and W. Lynn Shirley. The SoVI model is a hazard-of- place model which utilizes census data to quantify social vulnerability to environmental hazards at the county level for the United States (Cutter et al. 2003). SoVI is a comparative model in which final social vulnerability scores are given relative to each other and mapped using standard deviations from the mean to classify data. The SoVI model utilizes a seven step process to acquire, analyze, and map the data, as described below (Cutter 2008).  The first step is to download the chosen variables for each county from the US Census. The SoVI model starts with over 250 variables from which a subset is chosen according to the study goals. The second step is to standardize the variables into density, per capita, or percentage functions. The third step involves verifying the data’s accuracy by using minimum and maximum values to check for consistency. Missing data must also be added during this period by using the mean value for the variable to substitute for missing data. The creation of z-scores is the fourth step, in which the standardized variables are normalized to z-score variables that have a mean of zero and a standard deviation of 1. Z- 16                                                                                                                                May 2008   scores allow for data to be comparable across variables by normalizing all data to the same scale (Cutter et al. 2003, 2008).  The fifth step is to perform a principal component analysis (PCA) on the data using a varimax rotation and Kaiser criterion for variable selection (eigenvalues greater than 1.0). Eigenvalues greater than 1.0 are considered significant in explaining data variation, eigenvalues less than 1.0 are not considered significant. PCA is a statistical technique used to decrease the number of variables used in analysis by identifying the variables that account for the majority of data variance. PCA produces a set of components (or factors) that contribute significantly to the data. Each component is an amalgamation of the variables but each variable weighs differently in each component. Components are generally loaded highly on a small number of variables while other variables contribute small amounts to that component. The varimax rotation is used to find the best possible fit for the data set (Cutter et al. 2003, 2008).  Once the PCA has been performed, the sixth step is to interpret the resulting components. This is a key step and involves the judgment of the researcher. This step involves determining what the components broadly represent and which variables are significant within that component. Components must be aligned at this step to represent either increases or decreases in social vulnerability. The seventh step is to place all the components, with any necessary adjustments in correlation, into an additive model and sum to generate the overall SoVI score for each county (Cutter et al. 2003, 2008).  The eighth and final step is to map the SoVI scores using classifications based on standard deviations from the mean of the SoVI scores for each county. Scores greater than or equal to 1 indicate higher levels of social vulnerability and scores of less than or equal to -1 indicate lower levels of social vulnerability. The resulting map can then be used for spatial comparison of social vulnerability across the given area; in the SoVI case, counties across the United States (Cutter et al. 2003, 2008).  5.2 Pareto Ranking Model The Pareto ranking model of constructing a social vulnerability index was developed by Lisa Rygel, David O’Sullivan, and Brent Yarnal (Rygel et al. 2006). Like the SoVI model, the Pareto ranking model also uses a hazard-of-place theoretical framework. In addition the initial steps through the PCA are the same as those used by the SoVI model. The differences between the two models emerge when calculating an overall vulnerability score for each spatial unit. This approach was used initially to calculate social vulnerability to hurricane storm surge, however the model can be adapted to other types of hazard events by the selection of indicators specific to the hazard (Rygel et al. 2006).  The Pareto method begins with the selection of broad indicators from the literature to characterize social vulnerability. In the case of the original study, 57 variables were chosen to represent five broad vulnerability indicators. This analysis was conduced at the census block level as opposed to the county level analysis in the SoVI model. The 57 variables were run through a PCA with a varimax orthogonal rotation and Kaiser 17                                                                                                                                May 2008   Normalization applied to the solution. The PCA produced 3 significant components those components accounted for 50.83% of the data variation in the Rygel et al. study (2006). Within each component the significant variable scores were added together to produce an overall component score. The three component scores were then used to conduct a Pareto ranking. The Pareto ranking provided a set of 19 rank classifications to the data set. The variables were then sorted into 4 equal interval classes and mapped (Rygel et al. 2006).  Pareto ranking is a method used to order cases in a multiple criteria data set. Pareto ranking is based upon the concept of non-dominance, which assumes that a higher score on any component indicates higher vulnerability. In the hypothetical case of areas X and Y for a three component case (component 1, 2, and 3), if X has a score higher than Y on each component then, X dominates Y and has a higher vulnerability. However if X has a higher score on component 1 and 2 but Y has a higher score on component 3 there is no way to tell which area has higher vulnerability and they are both given the same rank. Figure 2 below shows the ranking for a two component case (Rygel et al. 2006).  In Figure 2 each data point represents a subarea and with scores on components 1 and 2. To illustrate the differences between SoVI and Pareto Ranking results a test case of subareas ‘A’ and ‘B’ is used. Subarea A has a score of 9 on component 1 and 1 on component 2. Subarea B has a score of 5 on component 1 and 6 on component 2. For example in the two component case below subarea ‘A’ has a score of 10 using the SoVI method which adds component scores, while subarea ‘B’ has a score of 11. With the SoVI method subarea ‘B’ would be considered more vulnerable that subarea ‘A’. However with the Pareto ranking subarea ‘A’ is considered more vulnerable because of the extreme score on component 1. This case shows that the Pareto Ranking method is more sensitive to extreme values on any one component.                     18                                                                                                                                May 2008   Figure 2: Pareto Ranking Example  Pareto Rank 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 Component 1 C om po ne nt  2 A 10 Rank 1 Rank 2 Rank 3 Rank 4Rank 5Rank 6Rank 7Rank 8Rank 9 Rank 10 B   6.0 Analysis  For this study, social vulnerability to earthquakes in Vancouver, BC was assessed using two indexing methods, the SoVI and Pareto Ranking methods described previously in section 5.0. The two models use the same methods through the Principal Component Analysis (PCA) step; only in the assigning of index scores and mapping do the two models vary. The process for using both models will be explained as well as the data generated during the analysis process. All statistical analysis was done in SPSS with data imported from Microsoft Excel. In the following section, 7.0 Findings, the results of the two models with be discussed and compared.  6.1 Data Collection The first step in both the SoVI and Pareto Ranking methods is to decide which vulnerability indicators apply to the study and then to choose proxy variables for those indicators. The chosen indicators and proxy variables were discussed earlier in section 4.0 Social Vulnerability Indicators & Proxy Variables (see Table 1). There were fifteen variables chosen: percentage of low income population, percentage of women making less than $20,000 per year, percentage of the population without knowledge of English, percentage of the population less than 19 years old, percentage of the population 65 years or older, unemployment rate, percentage of apartments five or more stories built before 19                                                                                                                                May 2008   1980, percentage of dwellings needing major repairs, percentage of dwellings rented, percentage of the population not employed as professionals or management, average household size, percentage of single parent families, percentage of the population (20 years or older) with a high school education or less, percentage population change in the last 5 years, and percentage of total income from government transfer payments. These variables represent the broad indicators of social vulnerability to earthquakes outlined in the existing literature.  All data for the proxy variables were taken from the 2001 Census of Canada and accessed through the E-STAT and CANSIM databases. In most cases, the data did require manipulation to either create proxy variables that were more accurate measures of social vulnerability indicators or to convert the data into a percentage or comparable standard. All variables were adjusted before analysis so that a higher score on any variable would correlate with a higher level of overall social vulnerability. This saved time in re- orienting the variables later in the process. There are a total of 105 census tracts within the City of Vancouver, the University of British Columbia and the surrounding endowment lands were also included in the analysis for a total of 106 census tracts.  6.2 Normalization (z-score) Data normalization, using z-scores, is the fourth step in both methods. Z-scores allow for the comparison of data across variables that may have different scales. With z-score normalization all data is converted into standard deviations from the mean. A mean of 0 and a standard deviation of 1 is used in this normalization. Scores with negative numbers are lower than the mean, indicating lower social vulnerability, and positive numbers are higher than the mean, indicating higher social vulnerability. The equation for z-score normalization is below.  z-score= χ - μ   σ where:  χ is a score to be standardized   μ is the mean of the scores over Vancouver census tracts   σ is the standard deviation of scores over Vancouver census tracts              20                                                                                                                                May 2008   6.3 Principal Component Analysis SoVI and Pareto Ranking models both use Principal Component Analysis (PCA) to reduce the number of variables through the formation of components. The PCA is performed with a varimax rotation and Kaiser criterion for component selection. Four components (factors) with eigenvalues of greater than 1 were identified (see table 3 below). The four factors with eigenvalues of 1 or more explain 77.75% of the variance in the data. Factors 1 and 2 explain a larger percentage of the data, with factor 1 explaining 34.55% and factor 2 explaining 26.55%. In Table 3 below in the initial eigenvalues section the ‘total’ column indicates the amount of variance accounted for by each factor. The ‘percentage of variance’ column gives the percentage of variance accounted for by the factor relative to the total variance. The ‘extraction sums of squared loading’ section shows the amount of variance explained by the factors through the PCA extraction; these numbers are the same as the initial eigenvalues numbers. The ‘rotation sums of squared loadings’ section explains the variance after the rotation has been applied to the data so the amounts of variance explained by each factor shift slightly but the overall amount of variance explained is the same as the initial extraction.  Table 3: Total Variance Explained by Factors  Factor Initial Eigenvalues Extraction Sums of Squared Loadings Rotation Sums of Squared Loadings   Total % of Variance Cumulative % Total % of Variance Cumulati ve % Total % of Variance Cumulative % 1 5.1 34.5 34.5 5.1 34.5 34.5 4.8 32.0 32.0 2 3.9 26.5 61.1 3.9 26.5 61.1 3.9 26.4 58.5 3 1.3 8.8 69.9 1.3 8.8 69.9 1.6 11.1 69.6 4 1.1 7.8 77.7 1.1 7.8 77.7 1.2 8.0 77.7 5 .7 5.0 82.8 6 .6 4.6 87.4 7 .5 3.6 91.1 8 .4 2.8 93.9 9 .2 1.9 95.8 10 .2 1.5 97.4 11 .1 .8 98.3 12 .0 .5 98.8 13 .0 .4 99.3 14 .0 .3 99.6 15 .0 .3 100.0   Table 4 below shows the component matrix after the varimax rotation has been applied. The values in the table represent the correlation between the z-score variable and the rotated factor, for example incidences of low income and factor 1. Correlations can range from -1 (reflecting a negative linear correlation) to +1 (reflecting a positive linear correlation) with 0 indicating no correlation. The higher the absolute value of the  21                                                                                                                                May 2008   correlation the larger the effect on the factor. Results are interpreted by looking for common themes among the variables that load heavily within a factor.  Table 4: Rotated Factor Matrix  Factors  1 2 3 4 Incidence of low income in 2000 %  .703 .525 .276 .154 Government transfer payments % - composition of total income in 2000 .878 .278 -.002 -.224 % Women making less than $20,000 per year .896 -.191 -.041 .038 Unemployment rate, population 15 years and over .728 .330 .230 .129 % 19 & U .071 -.929 -.092 .083 % 65+ .186 .500 -.473 -.570  % apartment 5+ storey before 1980  -.001 .780 -.049 .032 % Dwellings rented .308 .762 .440 .080 % dwellings needing major repairs -.023 .300 .736 -.200 Average number of persons in private households .163 -.882 -.250 .128 % lone parent families .491 .071 .620 -.059 % Pop 20+ w/ HS or less education .831 -.355 .018 -.158 % Population change, 1996-2001 .045 .034 -.229 .821 % Population without knowledge of English .789 .041 -.017 .122 % Population not employed as professionals or management .707 -.414 .252 -.069   22                                                                                                                                May 2008   Table 5 is a summary of the variables that load heavily on each factor. In the case of this analysis all 15 variables were significant within a factor. As Table 5 below shows the four factors identified in the PCA were named in order to better describe their component variables; economic (factor 1), household (factor 2), families and dwellings (factor 3), and population change (factor 4). Variables were assigned to factors based upon the highest correlation value for each variable. The sign adjustment shows whether the factor increases or decreases social vulnerability. Because all variables were oriented as to increase vulnerability with higher scores at the data collection stage all factors in this case increase social vulnerability.  Table 5: Factors and Variable Loading Sign Adjustment Factor Name Variables Correlation  % low income population .703 % total income from government transfer payments .878 % women making less than $20,000 per year .896 Unemployment Rate .728 % Population with HS education of less (20+) .831 % Population without knowledge of English .789 + 1 Economic % Population not employed as professionals or management .707 % Population 19 years old or less .-929 % Apartment 5+ storey built before 1980 .780 % Dwellings Rented .762 + 2 Household Average Household Size -.882 % dwellings needing major repairs .736+ 3 Families & Dwellings % Single Parent Families .620 % Population Change from 1996-2001 .821+ 4 Population Change % Population 65 years & older -.570 23                                                                                                                                May 2008   The economic factor contains seven variables, which generally relate to economic factors that affect social vulnerability. The variables range from those that are directly economic such as the percentage of the population in low income and the percentage of income from government transfer payment to those factors that affect peoples ability to earn higher wages, such as their job type and knowledge of English. The second factor is named household and contains four variables. The household variables pertain to the personal characteristics of those living in a household, such as age as well as the type of structure and whether or not it is rented. The families and dwellings factor contains two variables: dwellings in need of major repairs the single parent families. The population change factor also contains two variables: percentage population change from 1996-2001 and the percentage of the population 65 years and older.  Scores for each of the four factors were created by adding the z-score of each of the variables within that factor by census tract. From these factor scores the SoVI and Pareto Ranking use different methods to assign a final social vulnerability score to each census tract and map the resulting vulnerability scores. The methods for each model are described below, starting with the SoVI model.  6.4 SoVI Social Vulnerability Scores & Mapping Vulnerability scores for the SoVI model are simply the summed scores from the four factors. The mapping process for the SoVI scores involves mapping the vulnerability scores by standard deviations from the mean. A base map of Vancouver census tracts is used and the SoVI vulnerability scores are merged with census tract data and then the layer is mapped based on the SoVI scores using standard deviations from the mean. Map 1 below shows the outcome of the SoVI social vulnerability assessment by census tracts. Factor maps can also be made using the scores for individual factors. Appendix 1 contains maps for each of the four factors from the PCA. These maps can be useful in identifying more specific aspects of social vulnerability. In Map 1 below the light yellow and yellow shaded census tracts represent the least vulnerable census tracts. Census tracts in light blue indicate a level of social vulnerability near the mean for Vancouver. Green represents social vulnerability levels higher than the mean for Vancouver while the dark blue census tracts represent the most socially vulnerable census tracts. Social vulnerability ratings from the Pareto Ranking model will be compared to the SoVI ratings and those census tracts with significant differences in vulnerability ratings will be discussed in detail later on.            24                                                                                                                                May 2008    Map 1: SoVI Social Vulnerability by Census Tracts, City of Vancouver     6.5 Pareto Ranking Social Vulnerability Scores & Mapping Pareto Ranking social vulnerability scores are calculated using the Pareto ranking method described in section 5.2 where non-dominance is used to create data ranks. The four factor scores created in the final steps of the PCA process are used as the four variables upon which the Pareto ranking is based. The data set created 8 vulnerability ranks. A vulnerability score of 1 represents the most socially vulnerable census tracts while a score of 8 indicates the lowest levels of social vulnerability for a census tract. The Pareto Ranks are mapped in four equal intervals. The gradient for the map is done in blue with the darkest blue representing the highest level of social vulnerability, a score of 1-2, and the lightest blue representing the lowest levels of social vulnerability, a score of 6-8. Map 2 below shows similar patterns of spatial distribution of social vulnerability as the SoVI model however there are some census tracts that do show significant differences. Those differences will be discussed in the following section.   25                                                                                                                                May 2008   Map 2: Pareto Ranking Social Vulnerability by Census Tracts, City of Vancouver     7.0 Findings  A number of spatial patterns appear in many of the proxy variables: these patterns influence the overall pattern of social vulnerability in Vancouver and lead to expectations about where the pockets of high vulnerability may be located.  Those indicators related primarily to economic factors, such as income, government transfer payments, English comprehension, professional or managerial position, high school education and the unemployment rate all have strong east-west tendencies. Those census tracts in the west tend to exhibit higher incomes, higher education levels, and larger numbers of professionals while the east tends to have higher unemployment rates, lower incomes, greater non-English speaking populations, and higher government transfer payment levels. A number of other factors also show a strong spatial correlation with the downtown core and periphery. The average household size is significantly smaller in the downtown area, housing type plays a large factor in who chooses to live in downtown areas as it can be difficult for large families to have enough space in condominium units. The population change is also greatest in the downtown, primarily in the False Creek North and Yaletown areas which are rapidly being developed with high rise apartment 26                                                                                                                                May 2008   and condominium developments. The large concentrations of dwellings needing major repairs are located in the Strathcona neighborhood and False Creek South. The Downtown East Side provides an interesting case as it is at the convergence of the downtown periphery and the beginning of the east side of Vancouver. While it did not rate extremely high on a number of individual factors it was generally in the higher rankings and did not rank low on any variable except youth. While there may not have been a large number of extremely high rankings on individual variables the many social problems of the Downtown East Side are well known and the area is generally considered to be extremely vulnerable.  The SoVI and Pareto Ranking social vulnerability maps show similar overall trends. Both models show the western half of Vancouver as less vulnerable while the eastern half is generally more socially vulnerable. This fits with the general expectations and knowledge of Vancouver. The western half of Vancouver generally contains more expensive real estate and the people living there accordingly have a higher average income than those in the eastern half of the city.  While the overall trends in vulnerability are the same between the SoVI and Pareto Ranking methods there are significant discrepancies between the two models in some census tracts. When trying to interpret these two models it is important to look at where they produce different results and to examine why the scores may be different. In some cases the reason one area scores high in one study and not another may be significant. On the other hand it may be a less significant indicator that is causing the increased vulnerability score where it may not necessarily be warranted. The following sections examine the areas that were rated as highly vulnerable in each model, as well as those that scored highly in both models. These individual factors and indicators are looked at for these highly vulnerable census tracts in order to gain a better understanding of what factors are affecting vulnerability in those areas.  7.1 Highly Vulnerable Areas The SoVI identified five census tracts with the highest vulnerability rating while the Pareto Ranking model identified thirteen census tracts with a vulnerability score of one. Three of those census tracts overlapped and were rated as highly vulnerable by both models. The three highly vulnerable census tracts, 9330056.01 (56.01), 9330057.01 (57.01), and 9330057.02 (57.02) are all located in the Downtown East Side and Strathcona areas, as would generally be expected. The map below shows the highly vulnerable areas identified by both studies. Areas in pink are the highest rated census tracts for SoVI, blue are the highest rated census tracts for Pareto Ranking and purple are the census tracts identified by both models.        27                                                                                                                                May 2008   Map 3: High Social Vulnerability Census Tracts    In order to understand the factors influencing these highly vulnerable census tracts it is important to understand the factors that are heavily influencing their social vulnerability score. The three census tracts that were identified as highly vulnerable by both studies will be discussed in this section and those census tracts that were identified by either study as highly vulnerable will be discussed in the following section. Figure 3 below shows the factor scores for each census tract rated highly vulnerable by SoVI or Pareto Ranking methods in order to allow for quick comparison between factor scores.         28                                                                                                                                May 2008   Figure 3: Factor Scores by Highly Vulnerable Census Tract  -20.00 -10.00 0.00 10.00 20.00 30.00 40.00 3.0 1 3.0 2 49 .02 50 .02 55 .01 56 .01 57 .01 57 .02 58 .00 59 .03 59 .06 60 .01 61 .00 62 .00 67 .00 Census Tract Fa ct or  S co re Economic Factor Household Factor Single Parent & Dwelling Standard Factor Population Factor   Census tract 56.01, in the Grandview-Woodlands neighborhood, has high scores across all four vulnerability factors; however its score is significantly higher on the single parent and dwelling standard factor, with a score of almost twice the other two highly vulnerable census tracts. Within that factor the two components, percentage of single parent families and percentage of dwellings in need of major repairs both score very highly. The percentage of single parent families is over 4 standard deviations above the mean, the highest score for any census tract. This extreme score on the single parent and dwelling standard factor allows the census tract to be identified during the Pareto ranking as well as the SoVI averaging process.  Census tract 57.01, in the Strathcona neighborhood, has the highest overall social vulnerability score in the SoVI model. The economic factor score for 57.01 is almost twice as high as any other census tract. This census tract also has the highest score on the household factor, but not by such a large margin. The scores on the other two factors are very high as well. Within the economic factor 57.01 scored highly on four of the seven indicators; low income, transfer payments, unemployment rate, and non-English 29                                                                                                                                May 2008   speaking. The high score in the household factor is influenced heavily by the high percentage of rentals in the census tract.  Census tract 57.02, in the Strathcona neighborhood, scores highly on all four factors but particularly heavy on the economic as well as the single parent and dwelling standard factors. Within the economic factor the scores are high across all factors, but slightly higher in the percentage of transfer payments and the percentage of the population who does not speak English. Both indicator scores, single parent families and dwelling standards, score highly as well leading to an overall high score in the single parent dwelling standard factor.  7.2 High Social Vulnerability Scores: Differences Between Models Two census tracts, in the Downtown East Side neighborhood, were identified as being highly vulnerable in the SoVI model but not in the Pareto Ranking model, census tracts 9330058 (58.00) and 9330059.06 (59.06). Census tract 58.00 scores very high on the economic factor but much lower on the other factors. The economic factor score was raised by the high percentage of transfer payments as a portion of income in census tract 58.00. The high score on the economic factor is enough to raise the average of the relatively low scores on the other factors to put this census tract in the highest vulnerability group. Like census tract 58.00, census tract 59.06 scores very highly on the economic factor but relatively low on other factors; however the unemployment rate variable is the most influential variable in the economic factor score of 59.06.  Ten census tracts were identified as being highly socially vulnerable in the Pareto Ranking model that were not identified in the SoVI model; census tracts 9330003.01 (3.01), 9330003.02 (3.02), 9330049.02 (49.02), 9330050.02 (50.02), 9330055.01 (55.01), 9330059.03 (59.03), 9330060.01 (60.01), 9330061 (61.00), 9330062 (62.00), and 9330067 (67.00). Census tracts 3.01 and 3.02, in the Sunset neighborhood, both have high scores on the economic factor as well as the household factor. Within the economic factor 3.01 weighs heavily upon the non-English speaking variable while 3.02 weighs higher on the transfer payments and women in low income variables. Within the household factor both 3.01 and 3.02 weigh heavily upon the average household size variable. Census tract 49.02, in the Fairview neighborhood, scored highly on the single parent and dwelling standard factor: primarily on the dwellings in need of major repairs variable. Census tracts 50.02 and 55.01, in the Grandview-Woodlands neighborhood, scored highly on the economic factor and single parent and dwelling standard factor. The low income variable was the dominant variable in the economic factor for census tract 50.02 while 50.01 was split between a number of variables. Census tracts 59.03, in the Downtown neighborhood, received a Pareto rank of 1 due to its high score in population growth. This high score in the population change factor is due to the fact that this area is predominantly made up of the False Creek North development and during the 1996-2001 time period the population grew rapidly as the area developed. The scores for all other factors for 59.03 were quite low. While this is an extreme level of population growth, the low scores in all other factors indicate that this area may not actually be highly socially vulnerable. Census tracts 60.01, 61.00, 62.00 and 67.00, in the West End neighborhood, all score highly on the household factor, more specifically they score highly on the 30                                                                                                                                May 2008   percentage of apartment buildings built before 1980 that are five or more stories tall and the percentage of rentals. This census tract is in the West End and has a long history of dense development predating the 1980s. The high vulnerability score would be justified if the buildings within these census tracts have not been retrofitted or built to a level that would withstand major damage from an earthquake and therefore protecting the occupant of the building. In order to truly assess the social vulnerability of this census tract information on the specific buildings would need to be reviewed.  While both SoVI and Pareto Ranking identify different census tracts as being highly vulnerable neither model is completely wrong or completely right. To truly assess the social vulnerability score given to a census tract the underlying factors of that score must be understood. Effective planning measure will come from understanding the causes of social vulnerability in a given area and targeting programs or plans to meet those vulnerabilities.   8.0 Next Steps/Planning Implications  In order to fully utilize this study within a planning context it is important to find ways to apply the knowledge gained that will support planning processes as well as aid and protect socially vulnerable citizens. The data provided by this study can be applied in a number of planning context and at all four disaster preparedness phases; mitigation, preparedness, response, and recovery. While this study provides useful data on social vulnerability in Vancouver the methods can be refined to create more accurate models and applied to other locations and hazard contexts.  8.1 Planning Implications There are a number of ways that a social vulnerability assessment can be applied to the planning process. Understanding how a study will be applied can aid in the design process in order to have an outcome that gives the best possible information in terms of its application. Studies can also be designed in a way to provide useful information to a broad range of planning contexts; this study was designed in that way. The final products of this study, overall social vulnerability maps and vulnerability component maps, can provide a large amount of information on their own and more so when combined with additional municipal data that helps to fill in the broader picture of vulnerability and development.  One of the primary uses of social vulnerability assessments is to identify areas with high levels of social vulnerability in the event of an earthquake. This study may reveal pockets of social vulnerability that planners had not taken into consideration previously, such as the two high vulnerability census tracts in the Sunset neighborhood near the Oak Street Bridge. Understanding the aspects of social vulnerability and the factors that affect each area planners can make better decisions about how to allocate resources or develop new planning strategies to aid those areas with high social vulnerability. When social vulnerability information is combined with other information, such as infrastructure 31                                                                                                                                May 2008   service levels or resource access information, social and physical service levels can be assessed and gaps in services can be targeted to areas that are in the greatest need.  Social vulnerability assessments can also be used in conjunction with physical hazard information to prepare for disaster response. Areas with highly vulnerable populations may need more assistance in the case of evacuation or accessing services before, during, or after a disaster. The combination of social and physical vulnerability information can provide a picture of which specific areas may be at greater risk and require greater assistance. The applications of social vulnerability studies are almost endless: the more creative municipalities and governing bodies are in using and applying the data, the more useful the studies become. An example of a specific program tailored to an area of high vulnerability would be to provide earthquake information or preparedness classes in multiple languages to target the large non-English speaking population in census tract 3.01 as identified by this study. These targeted programs could provide valuable information that could lessen vulnerability in the case of an earthquake, in a relatively inexpensive way by providing information that would have otherwise been inaccessible.  8.2 Planning Recommendations There are a few basic planning recommendations that come out of this research. Many other planning possibilities can come out of the data produced by this project with further analysis and information. However the two basic planning recommendations are; conduct an assessment of high rise apartment buildings built before current seismic safety codes and develop strategies for providing additional assistance to areas with high social vulnerability. The assessment of high rise apartments would allow for the identification of buildings that could sustain major damage in an earthquake event. Once the buildings were identified retrofits could begin to bring them up to current code. Until the retrofits are complete residents should be told of the risk and taught what to do in the event of an earthquake to protect themselves, including things like bracing shelves and where and how to properly take shelter during an event. The second recommendation, providing additional assistance is somewhat more nebulous of a recommendation: however it is very important. The potential needs of the highly vulnerable area will need to be assessed and programs tailored to specifically meet those needs. An additional component of this recommendation could be to begin an information campaign in these areas, and throughout the city, on earthquake risk and what to do in the case of an earthquake event.  8.3 Further research While both the SoVI and Pareto Ranking models of assessing social vulnerability provide useful results there are ways that the models can be refined or new models created to more accurately assess social vulnerability. One of the major ways the models can be improved is to develop a method for weighting social vulnerability indicators. Neither method allows for the weighting of vulnerability indicators: While the weighting process is quite complicated, in the end it could produce much more accurate results. The SoVI model simply deals with the problem of weighting by averaging the indicators and therefore assuming they have equal weight. The Pareto Ranking model assumes the problem of weighting is taken care of by the ranking process which gives high social vulnerability scores to census tracts which have an extreme value on any indicator, 32                                                                                                                                May 2008   assuming that one extreme indicator is an indicator of higher overall vulnerability. While both methods have their advantages, neither are as informative as a well done weighted study could be. In order to develop an accurate weighted social vulnerability model, a large amount of research as well as interviews and discussions with experts, decision makers, and other stakeholders would have to take place. While this may be a large amount of work, it would be a large step forward for this type of analysis.  9.0 Conclusion  This study has a distinct advantage for Vancouver over other social vulnerability studies that have been done in other places in that two different models of analysis are applied to the same location and data. Having two different models allows for the benefits of both models to come through and provide a more complete picture of social vulnerability. The Pareto Ranking method has the advantage of catching census tracts which have high scores on any one factor, assuming that the one high score may be indicative of social vulnerability which may not show up if the scores are averaged. However, the SoVI method identifies those census tracts that score moderately high across a number of factors as highly vulnerable, something the Pareto Ranking model does not do. The SoVI method gives a good overall picture of vulnerability while the Pareto Ranking method focuses the attention on more specific aspects of vulnerability. The differences between subareas in both models is useful in considering the social vulnerability of an area.  The focus on earthquakes in this study does not preclude further similar studies focusing on other hazard events. Having one type of hazard as the focus of the study allows for more accurate vulnerability indicator and proxy variable selection. Multi-hazard social vulnerability indexes can lose a lot of the precision by not choosing a specific hazard to focus on and may miss important pockets of vulnerability to certain types of hazards. In terms of this study the most earthquake specific variables are those related to dwelling standards. Dwelling standards play a key role in earthquake events and unlike floods or other limited location hazards they apply across the entire region. The selection of the proxy variable ‘apartments five or more stories high built before 1980’ is specific to Vancouver and when seismic standards in the building code were adopted and the types of buildings built in Vancouver.  A different proxy variable may be necessary in a different study location.            33                                                                                                                                May 2008   Bibliography Adger, W. N., Brooks, N., Bentham, G., Agnew, M., & Eriksen, S. (2004). New indicators of vulnerability and adaptive capacity (No. 7). Norwich, UK: Tyndall Centre for Climate Change Research. Alwang, J., Siegel, P. B., & Jorgensen, S. L. (2001). Vulnerability: A view from different disciplines (No. 0115). Washington, DC: Social Protection Unit, Human Development Network, The World Bank. from Discussion-papers/Social-Risk-Management-DP/0115.pdf Anderson, M. B. (2000). Vulnerability to disaster and sustainable development: A general framework for assessing vulnerability. In J. R. Pielke, & S. R. Pielke (Eds.). London: Routledge. Blaikie, P., Cannon, T., Davis, I., & Wisner, B. (1994). At risk: Natural hazards, people's vulnerability, and disasters. London: Routledge. British Columbia Provincial Emergency Program. (1999)  British Columbia Earthquake Response Plan. Burton, I., Kates, R. W., & White, G. F. (1993). The environment as hazard (2nd ed.). New York: Guildford. Clark, . (1998). Assessing the vulnerability of coastal communities to extreme storms: The case of revere, MA., USA. Mitigation and adaptation strategies for global change, 3(1), 59 Cutter, S. L. (2000). Revealing the vulnerability of people and places: A case study of georgetown county, south carolina. Annals of the Association of American Geographers, 90(4), 713 Cutter, S. L. (1985). Rating places: a view on quality of life. Washington D.C.: Association of American Geographers. Cutter, S. L. (1996). Vulnerability to environmental hazards. Progress in Human Geography, 20(4), 529. Cutter, S. L. (2001). American hazardscapes: The regionalization of hazards and disasters. Washington D.C.: Joseph Henry Press. Cutter, S. L., Boruff, B. J., & Shirley, W. L. (2003). Social vulnerability to environmental hazards. Social Science Quarterly, 84(3), 242. 34                                                                                                                                May 2008   Cutter, S. L., Mitchell, J. T., & Scott, M. S. (1997). Handbook for conducting a GIS-based hazards assessment at the county level, South Carolina Emergency Preparedness Division, Office of the Adjutant General. Cutter, S.L. (2008). The SoVI Recipe. Unpublished manuscript from author. Dow, K. (1992). Exploring differences in our common future(s): The meaning of vulnerability to global environmental change. Geoforum, 23(3), 417. Dwyer, A., Zoppou, C., Nielsen, O., Day, S., & Roberts, S. (2004). Quantifying social vulnerability: A methodology for identifying those at risk to natural hazards, Commonwealth of Australia. Finn, W. D. Liam (2004). Earthquake Engineering in Canada: A Selective Overview, Garoogian, D. (2005). America's top-rated cities: A statistical handbook. Millerton, NY.: Grey House Publishing. Hewitt, K. (1997). Regions of risk: A geographical introduction to disasters. Essex, U.K.: Longman. Kasperson, J. X., Kasperson, R. E., & Turner, B. L. (Eds.). (1995). Regions as risk: Comparisons of threatened environments. Tokyo: United Nations University Press. Kleinosky. (2007). Vulnerability of hampton roads, virginia to storm-surge flooding and sea-level rise. Natural hazards, 40(1) Mileti, D. S. (Ed.). (1999). Disasters by design: A reassessment of natural hazards in the United States. Washington D.C.: Joseph Henry Press. Miringhoff, M. L. (1999). The social health of the nation: How America is really doing. New York: Oxford University Press. Morrow, B. H. (1999). Identifying and mapping community vulnerability. Disasters, 23(1), 1. Morrow, B. H., & Ennerson, E. (1996). Hurricane andrew through women's eyes: Issues and recommendations. Interantional Journal of Mass Emergencies and Disasters, 14(1), 5. Natural Resources Canada. accessed 2/2008 35                                                                                                                                May 2008   36 Ngo, . (2001). When disasters and age collide: Reviewing vulnerability of the elderly. Natural hazards review, 2(2), 80 Peacock, W. G., & Girard, C. (1997). Ethnic and racial inequalities in hurricane damage and insurance settlements. In W. G. Peacock, B. H. Morrow & C. Girard (Eds.), Hurricane Andrew: Ethnicity, gender and the sociology of disasters (pp. 171). London: Routledge. Rygel, L., O'Sullivan, D., & Yarnal, B. (2006). A method for constructing a social vulnerability index: an application to hurricane storm surge in a developed country. Mitigation and Adaptation Strategies for Global Change (2006). 11:741-764. United Nations Development Program (UNDP). (2000). Human development report 2000. New York: Oxford University Press. World Resources Institute. (1993). The 1993 information please environmental almanac. Boston, Mass.: Houghton Mifflin. Wu, S. Y., Yarnal, B., & Fisher, A. (2002). Vulnerability of coastal communities to sea-level rise: A case study of cape may county, new jersey, USA. Climate Research, 22, 255. Yarnal, B. (1994). Agricultural decollectivization and vulnerability to environmental change: A Bulgarian case study. Global Environmental Change, 22, 255.                                                                                                                                               May 2008   Appendix 1: Factor Maps   37                                                                                                                                May 2008    38                                                                                                                                May 2008    39                                                                                                                                May 2008   40    May 2008   41 Appendix 2: SoVI & Pareto Ranking Scores by Census Tract  Census Tract CT Short Hand Economic Score Household Score SP & DS Score Population Score SoVI Score Pareto Rank 9330001.01 1.01 0.30 0.74 1.00 -0.39 1.65 3 9330001.02 1.02 1.98 0.44 1.98 -0.11 4.29 3 9330002.01 2.01 -0.54 -0.15 1.82 0.42 1.55 2 9330002.02 2.02 2.72 0.07 1.98 -0.04 4.73 3 9330003.01 3.01 9.63 2.04 0.07 -0.63 11.11 1 9330003.02 3.02 9.52 2.29 -0.18 0.06 11.68 1 9330004.01 4.01 4.85 -0.61 -1.18 0.67 3.73 3 9330004.02 4.02 8.96 1.70 0.56 0.30 11.52 2 9330005.00 5.00 -0.02 -0.74 0.18 -1.25 -1.83 6 9330006.01 6.01 -0.54 -0.33 0.31 -0.65 -1.21 6 9330006.02 6.02 4.82 0.49 0.68 0.27 6.26 4 9330007.01 7.01 -2.90 -0.46 0.19 -0.69 -3.86 7 9330007.02 7.02 -3.54 -0.20 -0.02 0.16 -3.60 6 9330008.00 8.00 -8.35 -0.34 -0.10 -0.41 -9.21 7 9330009.00 9.00 -3.88 0.01 -0.33 0.77 -3.43 4 9330010.01 10.01 0.80 0.42 -0.10 1.22 2.33 3 9330010.02 10.02 0.48 1.05 -1.47 0.70 0.76 3 9330011.00 11.00 6.07 0.63 -0.68 0.42 6.43 3 9330012.00 12.00 2.66 0.06 -0.36 0.08 2.45 6 9330013.01 13.01 8.97 0.77 -0.02 0.39 10.11 2 9330013.02 13.02 9.24 1.14 -0.20 0.31 10.48 2 9330014.00 14.00 9.28 0.66 0.76 0.55 11.25 2 9330015.01 15.01 4.90 0.04 -1.99 -0.14 2.81 6 9330015.02 15.02 6.03 -0.10 -0.84 0.75 5.85 2 9330016.01 16.01 10.12 1.76 0.17 0.05 12.10 2 9330016.03 16.03 6.03 1.59 0.58 1.35 9.55 2 9330016.04 16.04 6.41 0.56 1.24 -0.95 7.26 3 9330017.01 17.01 6.59 0.79 0.08 0.01 7.46 3 9330017.02 17.02 10.86 0.70 2.25 0.41 14.23 2 9330018.01 18.01 7.93 0.96 0.02 -0.17 8.74 3 9330018.02 18.02 4.49 0.07 1.19 -0.48 5.27 3 9330019.00 19.00 1.83 -0.31 2.33 0.52 4.36 2 9330020.00 20.00 -3.16 -1.88 -1.46 2.97 -3.54 2 9330021.00 21.00 -9.55 0.41 -1.68 -0.75 -11.57 5 9330022.00 22.00 -6.58 1.68 -0.79 2.83 -2.87 2 9330023.00 23.00 -6.28 0.01 -0.94 -0.18 -7.40 7 9330024.00 24.00 -7.84 0.05 -1.25 -0.91 -9.95 7 9330025.00 25.00 -6.67 -0.78 -0.94 -0.48 -8.88 7 9330026.00 26.00 -4.54 -0.24 -1.16 0.01 -5.94 7 9330027.00 27.00 -3.44 -0.72 -0.58 1.02 -3.72 4 9330028.00 28.00 -6.35 -0.61 -0.74 -0.51 -8.21 7   May 2008   42 9330029.00 29.00 -6.57 -0.91 1.72 -0.20 -5.96 4 9330030.00 30.00 1.86 -0.14 1.59 -0.77 2.54 4 9330031.01 31.01 -5.36 -0.96 0.78 -1.05 -6.59 5 9330031.02 31.02 1.39 0.00 3.73 -0.66 4.47 2 9330032.00 32.00 5.95 0.73 0.98 -0.38 7.28 3 9330033.00 33.00 6.85 0.66 1.16 -0.06 8.61 3 9330034.01 34.01 9.98 0.62 0.89 0.17 11.66 4 9330034.02 34.02 8.18 1.04 2.05 0.02 11.28 2 9330035.01 35.01 7.39 0.34 -0.16 -0.20 7.36 5 9330035.02 35.02 6.87 0.49 0.24 0.70 8.30 2 9330036.01 36.01 6.81 0.21 0.69 0.16 7.88 5 9330036.02 36.02 5.36 -0.07 0.03 -0.71 4.61 6 9330037.01 37.01 5.91 0.34 3.04 -1.29 8.00 2 9330037.02 37.02 6.26 -0.28 1.43 -0.75 6.65 3 9330038.00 38.00 3.41 -0.79 3.00 -0.66 4.96 2 9330039.01 39.01 -6.04 -1.66 1.09 -1.22 -7.83 5 9330039.02 39.02 -4.41 -2.07 2.55 0.12 -3.82 2 9330040.01 40.01 -7.90 -2.30 0.69 -1.57 -11.08 6 9330040.02 40.02 -8.26 -1.50 0.35 -0.48 -9.90 6 9330041.01 41.01 -7.97 -1.60 -1.64 0.92 -10.29 5 9330041.02 41.02 -8.98 -2.21 -0.25 1.34 -10.10 3 9330042.00 42.00 -6.00 -1.37 0.65 -0.77 -7.49 6 9330043.01 43.01 -6.94 0.27 -0.15 -0.93 -7.74 5 9330043.02 43.02 -9.26 -0.94 -0.35 -0.63 -11.19 8 9330044.00 44.00 -9.14 -1.16 -1.02 0.14 -11.19 7 9330045.01 45.01 -10.11 -2.28 0.37 -0.22 -12.23 6 9330045.02 45.02 -7.43 -1.93 -0.04 -1.32 -10.73 8 9330046.00 46.00 -6.50 -2.64 1.90 -0.87 -8.12 4 9330047.01 47.01 -8.59 -2.26 -0.23 0.33 -10.75 5 9330047.02 47.02 -9.21 -2.95 0.41 -1.95 -13.70 7 9330048.00 48.00 -7.07 -2.40 2.46 -0.61 -7.62 3 9330049.01 49.01 -9.59 -2.99 2.88 -0.39 -10.10 2 9330049.02 49.02 -9.29 -1.41 4.44 0.79 -5.47 1 9330050.02 50.02 6.92 -0.29 5.13 -0.90 10.86 1 9330050.03 50.03 2.27 -1.49 3.00 -0.61 3.17 2 9330050.04 50.04 8.90 -0.20 6.08 -1.66 13.13 2 9330051.00 51.00 6.98 0.04 1.02 0.11 8.15 3 9330052.01 52.01 10.66 0.68 1.02 0.37 12.73 3 9330052.02 52.02 7.23 0.59 1.37 0.62 9.81 2 9330053.01 53.01 7.17 0.01 0.99 0.23 8.40 4 9330053.02 53.02 2.16 -0.86 1.45 -0.63 2.12 4 9330054.01 54.01 3.09 -0.79 1.45 -0.69 3.06 3 9330054.02 54.02 5.30 -0.57 -0.05 0.35 5.03 4 9330055.01 55.01 10.17 -0.35 4.20 -0.88 13.15 1 9330055.02 55.02 4.85 -1.33 2.60 -0.75 5.38 2 9330056.01 56.01 10.92 0.52 7.08 -1.29 17.23 1   May 2008   43 9330056.02 56.02 0.03 -0.69 4.12 -1.21 2.24 2 9330057.01 57.01 36.72 1.84 2.34 3.79 44.69 1 9330057.02 57.02 19.85 1.00 3.76 0.65 25.26 1 9330058.00 58.00 26.32 -0.63 -0.81 1.38 26.26 2 9330059.03 59.03 -8.00 -2.34 -2.19 4.10 -8.43 1 9330059.04 59.04 -2.04 -1.86 -0.37 -0.01 -4.29 7 9330059.05 59.05 -1.60 -1.89 -1.53 2.40 -2.62 2 9330059.06 59.06 18.73 0.21 0.65 0.66 20.24 2 9330060.01 60.01 -3.80 3.31 -0.61 -1.94 -3.03 1 9330060.02 60.02 -2.64 1.76 0.53 -0.30 -0.65 2 9330061.00 61.00 -4.18 3.67 -0.89 -0.78 -2.18 1 9330062.00 62.00 -6.47 2.21 -0.81 1.58 -3.49 1 9330063.00 63.00 -4.30 1.01 1.37 -1.47 -3.39 3 9330064.00 64.00 -3.74 1.65 -0.26 -1.65 -3.99 3 9330065.00 65.00 -1.00 1.72 -0.79 -0.66 -0.73 3 9330066.00 66.00 -5.63 -2.45 -1.40 3.58 -5.91 2 9330067.00 67.00 0.28 2.17 0.30 0.91 3.65 1 9330068.00 68.00 -5.69 2.20 -1.76 -0.52 -5.77 2 9330069.00 69.00 -4.27 0.74 0.22 0.16 -3.15 3 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items