RACIAL INEQUALITY IN THE UNITED STATES: ANALYZING THE WEALTH GAP A Thesis Submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy By Shehnaz Jagpal, B.A. Washington, DC April 10, 2007 TABLE OF CONTENTS Introduction……………………………………………………………………………….1 Literature Review……………………………………………………………………...….3 Data analysis and research design………………………………………………..………12 Hypothesis, Data sources and methodology…………………………………..…………13 Limitations………………………………………………………………….……………16 Determinants of racial inequality……………………………………….………………..19 Income and age…………………………………………………………………..21 Educational achievement and net worth………………………...……………….22 Life opportunities: the example of the U.S criminal justice system……………..23 The concentration of wealth and intergenerational disadvantages…...………….25 Wealth, gender and marital status……………………..…………………………27 Education and income……………………………………………………………29 Occupational categories and race…………………………...……………………31 Analysis…………………………………………………………..………………………35 Policy Implications………………………………………………………………………43 Conclusion……………………………………………………………………………….47 Appendix A: Determining the Proper Model and Specification...………………...…….49 Appendix B: Regression Diagnostic Graphs.……………………………………………50 Appendix C: Multicollinearity and Correlation Coefficients……………………………60 References………………………………………………………………………………..63 ii RACIAL INEQUALITY IN THE UNITED STATES: ANALYZING THE WEALTH GAP Shehnaz Jagpal, B.A. Thesis Advisor: Robert W. Bednarzik, Ph.D. Abstract Recent analyses of economic well-being by race have shifted their focus from income inequality to the wealth gap. Dr. Melvin L. Oliver and Dr. Thomas M. Shapiro’s pioneering work in Black Wealth/White Wealth: New Perspective on Racial Inequality (1995) suggests that public policies designed to achieve racial equality by closing the income gap are insufficient to address the depth of inequality revealed by the wealth gap. This thesis builds on their work, using 2001 data from the Survey of Income and Program Participation (SIPP) and an adapted model to gauge changes in wealth holdings during the 1990’s. Key findings include the negative association between race and gender in acquiring wealth; African-Americans, especially single-parent females, are disproportionately marginalized in wealth accumulation. Policy recommendations contribute to emerging deliberations on asset-based welfare policy and revising affirmative action criteria to account for class status. Specific instruments to advance low-income homeownership and provide opportunities to increase asset ownership for poor women are suggested as means to achieve racial equality. iii Introduction Concentrated poverty in urban areas is one of the most often cited causes of the persistence of an African-American ‘underclass’ (Massey & Denton, 1993). Income inequality rose in the United States between 1970 and 2001. This is evidenced by the Gini coefficient, which measures income concentration, and it rose from 0.40 in 1970 to 0.47 in 2001 (Census Bureau, 2001). Yet, income inequalities alone do not account for the persistence of the racial divide in the United States. Keister and Moller (2000) demonstrate the limitations of analyzing disparities in income by highlighting starkly varying U.S. Gini coefficients for wealth (0.84) as opposed to income (0.52) in 1989. They and others (Conley, 1999a; Hao, 2004; Oliver & Shapiro, 1995) argue that analyses of wealth are thus fundamental to developing appropriate policy options to confront racial inequality. Numerous advantages are correlated with wealth acquisition, which cannot be realized by gains in income alone; for example, the ability of better-educated parents with significant liquid assets to provide funds for a college education or a down payment for a house. Acquiring wealth offers the opportunity to break the cycle of poverty. And, because sizeable wealth is directly transferable from one generation to the next, scholars conclude that substantial wealth assures that life opportunities and social position remain within a family over long periods (Conley, 1999a; Oliver & Shapiro, 1995, 2006). 1 Although income is a necessary condition for generating wealth, it is not a sufficient one. Indeed, the correlation between income and wealth holdings is surprisingly weak. Lerman and Mikesell (1988) found that when asset-income, i.e., income generated from wealth, was excluded, the correlation between income and wealth was only 0.26. This implies that income-based explanations are deficient in accounting for inequality of opportunity across races (Keister & Moller, 2000). It also implies that policies that seek to address racial inequalities by narrowing income disparities are insufficient. Oliver and Shapiro (1995) broke new ground in discussions of racial inequality in the United States by analyzing wealth inequality. Using data from the 1987 Survey of Income and Program Participation (SIPP), they conducted multiple regression analyses with income, net worth (NW) and net financial assets (NFA) as the dependent variables. They found that NFA is the best indicator of control over future financial resources while NW more accurately indicates the intergenerational transmission of wealth. They also conclude that blacks’ claim to middle-class status relies on income and not assets. For example, they found that the NW of African-Americans was far less than the NW of whites; and especially that middle-class blacks owned only 15 cents for each dollar of NW owned by whites in the same income class. This thesis builds on Oliver and Shapiro’s work using an adapted model and more recent SIPP data with the goals of: 1) seeing if patterns they identified continue to apply 2 and 2) recommending policy options to close the racial wealth divide. Given their findings about the black and white middle classes, this thesis will assess NW and NFA changes for these groups. If wealth disparities persist, it is important to develop policies that enhance current economic, social and human capital among disadvantaged groups. Literature Review This literature review demonstrates the rationale for studying wealth in addition to income with an eye toward generating policy solutions that address racial inequality more comprehensively. Home equity represents the majority of NW for the American middleclass; this is especially so for middle-class blacks (Oliver & Shapiro, 1995). The review thus looks first at studies of homeownership and the evolution of the African-American middle-class. Because of the inability of income analysis to explain differences in blackwhite life opportunities, it looks next at the cross-generation disadvantages associated with minority wealth accumulation. For example, Spriggs (2004) suggests that wealth inequalities reflect the cumulative income differences from one generation to the next. Finally, it reviews assessments of recent Federal housing policies which aim to decrease racial inequality and considers their effects on the black-white wealth gap. 3 Progress of wealth analyses: why study wealth versus income Wealth data to evaluate the economic conditions of African-Americans were unavailable until the 1967 Survey of Economic Opportunity (SEO). The first comparative analysis of wealth accumulation by race was done in 1971 by Henry Terrell. His findings indicated that ongoing discussions of income inequalities understated the level of economic inequality for blacks significantly. Regardless of income levels, black households bore the burden of historically lower average incomes compared with whites of similar backgrounds. Terrell’s work brought attention to the fact that the history of disparities in opportunity has major consequences for addressing income inequalities (Terrell, 1971). Oliver and Shapiro (1990) echoed this view, showing that at least one-third of U.S. households are asset poor. They suggested that social welfare and redistributive policies founded on income analyses radically underestimated racial inequalities. In addition, Charles and Durst’s (2002) analysis of wealth inequalities by race, as evidenced by the transition to homeownership, suggests disadvantages faced by blacks because of the lack of access to intergenerational transmission of wealth. Similarly, Conley (1999b) argues that resulting class differences from historically-rooted family wealth differences are difficult to address because they are the outcomes of past economic trends. Keister and Moller (2000) clarify some of the intergenerational processes and their impact on wealth accumulation. They identify cultural capital, transferred through education and 4 other life experiences, as an important asset in wealth accumulation. Based on limitations of each type of transfer in the black community, they agree with Oliver and Shapiro’s conclusion that wealth inequalities will persist for future generations. Blau and Graham (1990) established the parameters for wealth analysis theory. They analyzed wealth data from the late 1970’s of younger households in order to determine the composition of wealth for different groups. Controlling for racial differences in income and a range of demographic factors, their analysis found that over three-quarters of the wealth gap between blacks and whites remained unexplained. They hypothesized that policies developed to address income differences would not suffice to narrow or close the racial divide where wealth inequalities are larger than income differences. Oliver and Shapiro (1995) expand this theory by adding sociological factors. The authors contend that social context creates radically different investment opportunities for blacks and whites. They use a three-staged approach to answer the question of why wealth accumulation varies radically by race. Stage one assesses the explanatory power of human capital, labor market, and sociological factors; the objective is to analyze how education and income, for example, work together to explain racial wealth differences. Stage two probes the role of policy and institutional discrimination in wealth disparities. Following Cose (1993), they note how residential segregation limits minorities’ access to high-quality schools and jobs. The third stage examines the intergenerational 5 transmission of inequality, demonstrating how historic racial inequality is reproduced across generations. Wolff (1998) expands on the theory of wealth analysis, suggesting that wealth analysis provides a more comprehensive understanding of economic disparities because: (1) liquid assets which can be converted into money are a source of consumption funds independent of current income; (2) homeownership provides more fiscal options to the owner; (3) times of economic crisis can be managed more easily by households with access to liquid assets, and; (4) there is a relationship between the distribution of power and wealth. In contrast, Conley (1999a) acknowledges the persistence of racial inequality but suggests that “the consolidation of race and class is the fundamental problem” (p. 23). Conley’s analysis of net worth begins with a full model that includes race and class as explanatory variables. He finds that parental wealth is the strongest predictor of his “wealth-inclusive class model”1 and importantly, that the impact of race disappears when parental class is controlled for. In short, Conley argues that public policies to address the wealth gap should change their focus from race to class. 1 Conley’s net worth model begins with simple comparisons of young black and white families. He controls for income, age, gender, and education and finds dramatic differences by race. Next, he controls for parental class status, and the impact of race becomes insignificant. Controlling for parental net worth in his full model, neither race nor class are significant, suggesting the importance of inheritances in the transmission of wealth within a household (Conley, 1999a: 47). 6 Homeownership and the evolution of the African-American middle class Massey and Denton (1993) made a significant contribution to our understanding of racialized economic disparities and persistent residential segregation by showing that the evolution of the ghetto is the result of deliberate public policies. However, later analyses of private wealth delve deeper. Oliver and Shapiro (1995) argue that differential mortgage rate approvals by race offer evidence of institutionalized racism affecting home ownership. Charles and Hurst (2002) offer supporting evidence in a recent analysis of the transition to homeownership. Following a sample of black and white renters, the authors studied racial differences in the likelihood of being approved for a mortgage. They found significant evidence signaling that African-Americans are twice as likely as comparable white households to have mortgage applications rejected. This conclusion controls for credit histories and other proxy indicators of wealth, suggesting racialized home-lending practices. In a similar vein, Jackman and Jackman (1980) used national survey data to study racial inequality in homeownership. Controlling for socioeconomic characteristics and family composition, their analysis indicated that the probability of homeownership is significantly lower among African-Americans and whites of comparable backgrounds. They argued that the impediments blacks confront in the housing market severely restrict the possibility of improving their living standards. Kenneth Jackson (1985) argued that the most salient outcome of the ghetto is the impact on the black middle class, which did not enter the housing market on the same 7 terms as middle class whites. Jackson attributes this difference partly to policies of the Federal Home Owner’s Loan Corporation (HOLC). Established in 1933, the HOLC granted long-term mortgage loans to prevent foreclosure of urban homes until 1936. Jackson contends that the system of standardized property appraisals used by the HOLC institutionalized racist lending practices. Individual and community loans were provided on the basis of qualitative assessments of properties which included the racial composition of neighborhoods. With the infamous practice of redlining, mixed race and predominantly black neighborhoods were characterized as undesirable and not fit for investment. Thus, blacks’ access to Federal mortgage monies for homes in the suburbs was severely limited. More recently, although Alba et al. (2002) demonstrated that while suburban African-Americans are more integrated than central city blacks, a comparison of middleclass blacks and whites found no point at which the two groups were on par in terms of location. Middle-class African-Americans tend to live in more ethnically diverse neighborhoods but their white neighbors are often poorer than they are. These findings are consistent with Massey and Denton’s (1993) research which concluded that residential segregation systematically undermines the socioeconomic health of AfricanAmericans. 8 Recent Federal housing policy: impact on the wealth gap Let us turn to recent changes in Federal housing policy to see whether policies that foster mixed-income communities are a viable option for addressing racial imbalances in the United States. Since 1992, HUD has awarded billions of Federal dollars through Housing Opportunities for People Everywhere (HOPE VI) grants. The objective of HOPE VI grants is to alleviate concentrated poverty by supporting mixedincome communities in formerly blighted neighborhoods. Evaluations of the program remain limited. Despite some successes, several policy-relevant issues remain unanswered. These include appropriate targeting of limited resources and limited opportunities and choices faced by public housing residents (Popkin et al., 2004). A related measure, the Quality Housing and Work Responsibility Act of 1998, includes provisions to reduce concentrated poverty, promote economic self-sufficiency, improve public housing management via deregulation, rehabilitate or replace existing public housing facilities and streamline Section 8 voucher programs. Yet, the impact of this Act is also questionable. For example, although data suggest that concentrated poverty is being alleviated because of the Act’s income-based targeting, it is unclear whether self-sufficiency can be improved by this policy. Solomon (2005) found that while the number of working residents in public housing neighborhoods increased, incomes did not increase significantly. As such, improvements in self-sufficiency may owe more to the improving economy than to changes resulting from the Act. 9 Other Federal initiatives to promote mixed-income communities include tax credits for low income housing and community development block grants. However, the feasibility of these policies is also unclear because research on preconditions of these programs, including the economic background of residents, remains limited (Schwartz & Tajbakhsh, 1997). In general, effects of current Federal housing policies are unclear. Nonetheless, Conley (1999b) suggests that public policy shift its current focus on affirmative action in education and employment to promoting asset equality. He also recommends including net worth in determining class status for class-based quotas. Recent developments in wealth analysis and gaps in the literature Oliver and Shapiro recently issued a tenth anniversary edition of their original work, but they did not update their regression analyses. The 2006 edition includes two additional chapters. The first gauges changes in wealth since 1995 using data from various sources, including the Urban Institute, the Black Investor Survey and the Joint Center for Housing Studies. These recent studies indicate that although blacks made notable gains in wealth accumulation during the 1990’s, the recession and collapse of the stock market in the early 2000’s effectively erased most of the progress. The data also suggest improvements in the homeownership gap: black homeownership increased by 17.3 percent from 1995 to 2004, reducing the homeownership gap from 28.5 percent to 10 26.2 percent (Oliver & Shapiro, 2006: 213). However, gains in homeownership and increased access to credit were mitigated by sharp increases in debt. Moreover, because blacks and Hispanics continued to face subprime lending, Oliver and Shapiro suggest that minorities appear to be converting current debt into mortgage debt (Oliver & Shapiro, 2006:216). In short, they argue that macroeconomic shocks negatively impacted gains in asset accumulation during the 1990’s, returning minority net worth to its status in the mid 1980’s. 2 Indeed, a larger portion of African-American families had zero or negative net worth in the early 2000’s compared to the late 1980’s. Median black net worth peaked at $8,774 in 1999 but declined to $5,988 by 2002, effectively reverting to the levels observed in 1987 (Oliver & Shapiro, 2006: 215). Given the persistence of racial wealth inequalities, it is timely to revisit and revise the original model developed by Oliver and Shapiro and use updated SIPP data from 2001 to conduct a similar analysis. The adapted model used here analyzes income, NFA and NW as the dependent variables with race, gender, income and age as the main explanatory variables. Multiple regression analysis suggests that the adapted model has higher explanatory power than Oliver and Shapiro’s model, especially for NFA. As such, this model could serve as a springboard for future analyses of the racial wealth gap. 2 The second chapter expands on the idea of asset-based policy, first articulated by Michael Sherraden (1991) and briefly discussed in the first edition of Oliver and Shapiro’s book. 11 Data analysis and research design As discussed in the literature review, the model developed by Oliver and Shapiro (1995) serves as the basis for the relationships examined in this thesis. Using 2001 data from the Survey of Income and Program Participation (SIPP), the present study adapts their model with the goal of assessing changes in the wealth gap during the 1990’s. Preliminary analysis of the 2001 SIPP data suggests the continuing importance of age and income in analyzing wealth. Race, gender and marital status are important factors in determining income and consumption, and thus wealth. These relationships are analyzed in the following section, based on the 2001 SIPP data. The relationships among education, occupation and wealth are also analyzed. The model expands on Oliver and Shapiro’s (1995) model by including the type of household as an explanatory variable. Policy recommendations build on emerging asset-based policy and suggest ways to accelerate implementation. The following section provides details about the data and models used to analyze wealth during the 1990’s using the third wave of the 2001 SIPP.3 3 The SIPP was developed in the late 1970’s as a necessary complement to the Current Population Survey (CPS). The CPS is a cross-sectional database, and respondents were asked to recall their income and participation in Federal programs each March. In response to data collection issues with the CPS, the Department of Health, Education and Welfare (HEW) began the Income Survey Development Program (ISDP), a longitudinal survey, in which respondents were contacted every three months to track labor force participation and also to capture assets and liabilities data. The ISDP evolved into the SIPP, which was revised significantly in 1996. SIPP respondents are contacted every three months, and each household member older than 15 years is interviewed. Each household is contacted four times for lengthy interviews, making the three-month reference period cumbersome. As such, as of this writing, the Census Bureau is considering revising or replacing the SIPP. 12 Hypothesis, Data Sources and Methodology This section presents the hypothesis plus the data and model used to test them. Three measures of wealth will be examined: 1. Income 2. Net worth (NW) 3. Net financial assets (NFA) An analysis of income offers a baseline to gauge net worth and net financial assets (Oliver & Shapiro, 1995). They hypothesize that NFA is the most accurate predictor of control over future financial resources and that NW is a better tool to measure the intergeneration transfer of wealth, i.e., inheritances. Using their framework, the following hypotheses about wealth in the 1990’s will be tested: H1 (a): Race is positively correlated with wealth inequalities in the United States H1 (b): Age is negatively correlated with NW and NFA, and positively correlated with income. H1(c): Female-headed, single-parent households are less likely to acquire private wealth Ordinary least squares (OLS) regression analysis will be used to test these hypotheses. The regression analysis is based on the model developed by Oliver and Shapiro (1995) and pursuant work about wealth inequalities and their impact on public policy (Conley, 1999a; Keister & Moller, 2000; Darity & Frank, 2003). By using data from the 2001 SIPP, this analysis will allow for comparisons to Oliver and Shapiro’s findings. Besides using more current data, this paper builds on Oliver and Shapiro’s work by adding a dummy variable for household type. A comparison of white and black 13 female-headed households will also add to the previous work by providing a fuller picture of the impact of gender and marital status on wealth accumulation and life opportunities. 4 Exhibit 1: Variable Definitions and Sources VARIABLE THTOTINC Net worth NFA Race Region Education Age Age sq THTOTINC Gender RNFKIDS EMS RHTYPE TJBOCC1 DEFINITION Total household income5 Net Worth (Total household assets minus total household debt) Net Financial Assets (Total household wealth, minus home and vehicle equity) White or black FIPS State Code. SIPP state codes are broken down into the four standard U.S. regions used by the Census Bureau: NE, MW, South, West Less than HS, High School, College, College plus Age as of last birthday Age*Age Total monthly household income Male or female Total number of children under 18 in family Marital Status Household head Occupation SOURCE SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP SIPP Based on the literature review and theory, the hypothesized relationships between the dependent and independent variables are summarized in Exhibit 2: 4 Oliver and Shapiro’s model includes a variable which measures the marginal impact of being widowed on NW and NFA. The model in this thesis will look more specifically at the differences in asset accumulation by marital status of the head of household, i.e., being married or a single man or woman householder. While their work analyzed these differences in separate regressions, household type was not among their explanatory variables for measuring wealth. 5 This variable takes on negative values in some cases because it includes a measure of total household property income, which can be negative. The other two variables captured in total household income are earned income and total ‘other’ income, which include only positive values. 14 Exhibit 2: Variable Description, Predicted Relationship, Rationale VARIABLE DESCRIPTION PREDICTED RELATIONSHIP N/A Y1 THTOTINC Continuous Y2 Y3 X1 X2 Net worth NFA Race Region Continuous Continuous Dummy Dummy X3 X4 X5 X6 X7 X8 X9 X10 Education Age Agesq THTOTINC Male RNFKIDS EMS RHTYPE Dummy Continuous Continuous Continuous Dummy Continuous Dummy Dummy N/A N/A Negative Negative (for South) Positive Negative6 Positive7 Positive Negative Negative Negative Negative X11 TJBOCC1 Categorical Positive RATIONALE/ PREVIOUS STUDIES Income regression provides baseline to gauge trends in wealth, Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Oliver & Shapiro, 1995 Theory (Conley, 1999a; Hao, 2004) Oliver & Shapiro, 1995 The data for this model will come from the third wave of the 2001 SIPP. Wave 3 includes information about assets and liabilities and thus provides relevant information for wealth analysis. The two main dependent variables are defined as follows: 1. Net worth (NW) is the value of total assets, including home and vehicle equity, minus all debt 2. Net financial assets (NFA) are the total available household financial resources, i.e., total household wealth, excluding home and vehicle equity The main difference between NW and NFA is that NFA excludes vehicle and home equity. The independent variables are drawn from previous studies and theory (Blau & Graham, 1990; Oliver & Shapiro, 1995; Keister & Moller, 2000). Race is 6 7 Coefficient on Age is positive only for THTOTINC regression Coefficient on Agesq is negative only for THTOTINC regression 15 predicted to be negatively correlated with wealth. Income, age, gender and race are expected to be the most significant variables in determining wealth holdings. Because of the prevalence of female-headed households in the black community, data on black women’s wealth will be included in the analysis. Oliver and Shapiro (1995) compared single households by race and found that single white households had a median NW of $20,083 while single black households had NW of only $800. They also compared median NW by gender and marital status for whites and blacks. White female-headed households had a median NW of $23,530 compared to $500 for black female-headed households. Divorced white households had a median NW of $14,342 while divorced black households had only $1,373. These dramatic differences provide a useful benchmark to assess changes in the racial wealth gap. Limitations Although the model described was adapted from Oliver and Shapiro’s (1995) work, it has limitations. As several analysts of wealth note, it is difficult to quantify the historical role of the intergeneration transmission of poverty or wealth (Conley, 1999a; Hao, 2004; Oliver & Shapiro, 1995). However, because wealth data became more readily available with the initiation of the SIPP, wealth analysis helps to gauge the historical consequences of inequality more accurately than income data. The SIPP tends to over-sample the poor, which allows for analyses of wealth beyond the limited sample of the Survey of Consumer Finances (SCF) but this feature of the data is also 16 problematic. Many households in the sample have either zero or negative values for net worth and net financial assets, presenting a challenge for appropriate model specification. Determining the appropriate functional form to minimize loss of data is a challenge, but one that is addressable by following other analyses of wealth using SIPP data (Hao, 2004). This analysis seeks to gauge the impact of inheritances on wealth to better understand their role in the intergenerational transmission of wealth. However, data on inheritances are not readily available in the SIPP (Hao, 2004), so it is not feasible to account for the marginal impact of this important factor. Conley’s (1999a) critique of Oliver and Shapiro’s (1995) methodology applies to this variable and their model, including the adapted one tested here, more generally. He highlights the limitations of using cross-sectional data from the SIPP versus longitudinal data from the Panel Study of Income Dynamics (PSID). Since 1984, the PSID tracks families that were in the original 1968 sample, allowing assets and liabilities to be reviewed every five years. Conley (1999a) argues that this feature of the PSID allows researchers to study changes in wealth over time. “…by putting parental wealth into a model as a contributing factor, we can determine a ‘truer’ net effect of race in determining net worth or other socioeconomic outcomes” (Conley, 1999a:46). In short, the argument is that controlling for parental class factors effectively eliminates any significant impact of race on wealth. 17 Yet, if the model proposed here supports the hypotheses based on human capital theory, this analysis can contribute to policy solutions that seek to increase education and life opportunities for minority children. Moreover, because the SCF analyses discussed in the next section include inheritance data, they provide a good complement to the SIPP analysis. 18 Determinants of racial inequality: discussion of key variables • Life opportunities: human and cultural capital Good public infrastructure, including quality public schools and healthcare options, facilitates access to wealth. Because most public schools in ghettos lack the infrastructure needed to build the human capital of young African-Americans, residential and income-based segregation result. Cultural capital, transferred through better education, health status, and life experiences, is correlated with acquiring wealth. The inability to invest in children’s human capital continues to place blacks at a disadvantage. Conley (1999b) highlights this disparity in his analysis of the racial wealth gap. He argues that when ‘class-equalized’ blacks and whites are compared, blacks are as likely to complete college as whites. Better infrastructure and equal access to life opportunities ensure that wealth is maintained across generations within a household. • Historical economic trends Public policies built to enhance equal opportunity are insufficient if they fail to account for the burden of historically lower wages and asset accumulation opportunities among African-Americans. As Conley (1999b) argues, when parental assets are equalized to compare the wealth holdings of young adults, the racial wealth gap vanishes. This implies that racial inequality could disappear if public policies are explicitly designed to close the racial wealth gap. The complexity lies in developing practicable policies that can achieve this. Moreover, quantifying the historical impact 19 of generations of disadvantages is challenging (Conley, 1999a; Oliver & Shapiro, 1995). • Liquidity constraints Spriggs (in Smith, 2000) argues that African-Americans’ financial behavior and asset preferences must be situated in the context of liquidity constraints that frame their economic conditions. African-Americans remain disadvantaged in access to credit and loans. Liquidity constraints impact both short and long-term consumption as seen in asset allocation behavior. Parents with access to liquid assets can provide funds for a down payment on a house or to help with higher educational costs (Conley, 1999b; Oliver & Shapiro, 1995; Keister & Moller, 2000). • Public policy The historical impact of racially-biased Federal home-lending policy, coupled with lack of investment in the ghetto, continues to impact wealth portfolios of AfricanAmericans today (Oliver & Shapiro, 1995). Current Federal housing policy, with a focus on mixed-income communities, may be the first of many steps needed to assure ‘equality of achievement’ versus ‘equality of opportunity’ (Oliver & Shapiro, 1995). Glover-Blackwell (2006) suggests increasing affordable housing in opportunity-rich neighborhoods as a tool to confront persistent residential segregation and corresponding limited life opportunities. 20 Graph 1: Median monthly household income in the SIPP sample by age and race in the United States, 2001 5000 4500 White Black 4000 Income 3500 3000 2500 2000 1500 1000 500 0 Adult (25- Middle Age 40) (40-55) Old Age (55-65) Elderly (6585) Age Group Source: Survey of Income and Program Participation, 2001 Income and Age Graph 1 shows that white households had substantially higher median incomes than black households in 2001 in the United States. A partial explanation lies in the fact that African-American workers are overrepresented in lower-paying sectors of the economy, such as services. Moreover, the gap initially widens with age as white incomes increase more during middle age than black incomes before narrowing in old age. However, income declines for both whites and blacks after their peak earning years, which is why researchers use age, as well as income, to explain wealth disparities (Hao, 2004; Oliver & Shapiro, 1995). 21 Educational achievement and net worth Net worth is typically defined as the sum of all assets, including home and vehicle equity, minus all debts at a given point. Net worth is an important predictor of wealth transfers across generations (Oliver and Shapiro, 1995). While wealth data were not readily available until the 1970’s, the advent of the SIPP now allows analysts to measure trends in wealth in the United States. SIPP data show net worth increasing with age (See graph 2). Moreover, the positive relationship between net worth and age is magnified by education. Graph 2 shows that higher education and higher net worth go hand-in-hand, regardless of age. For example, the median net worth of a household head aged 55-59 years for a high school dropout was $47,000 and climbed to over $300,000 for a college graduate in the same age cohort. In short, age and education appear positively correlated with wealth. 22 Graph 2: Age net-worth profiles of households with 2 or more people by household head education, 2001 Source: Urban Institute tabulations from wave 3 of the 2001 SIPP panel Life opportunities: the example of the U.S. criminal justice system The prevalence of female-headed households in the black community suggests limited opportunities to invest in the human capital of young African-Americans. Concentrated poverty, poor pubic infrastructure, and insufficient policy support for lowincome women trying to make socio-economic gains all affect the life opportunities of future generations. The impact of broken families on early childhood development serves to perpetuate racial inequality. This inequality is perhaps clearest in the U.S. criminal justice system. African-American males represented 13 percent of all monthly 23 drug users yet accounted for 35 percent of drug possession arrests, 55 percent of drug convictions and 74 percent of prison sentences (The Sentencing Project, 2001). One in every three black men can expect to be imprisoned during their lifetime. Black women are six times more likely than white women to be incarcerated during their lifetime and one of every 14 African-American children has at least one parent in prison (Mauer & King, 2004). Graph 3 shows the tremendous increase in the absolute numbers of AfricanAmericans in prison from 1954 to 2002. These figures are also alarming in a relative sense because the U.S. black population had only increased from about 10 percent to 12 percent from 1950 to 2000 (McKinnon, 2001). Graph 3: Number of African-Americans in prison or jail, selected years – 1954-2002 Source: Mauer & King, 2004: The Sentencing Project 24 The concentration of wealth and intergenerational disadvantages Recent analyses of the Survey of Consumer Finances (SCF) provide contrasting views about the role of race in asset accumulation. Wolff’s (2000) study of trends in wealth ownership found that white households invested 23 percent of their total assets in stocks compared to only 11 percent among black households. In contrast, Straight’s 2002 study of the 1998 SCF data suggested that asset allocation was more equally distributed across races. Comparing data from the 1995 and 1998 SCF, Straight argued that the proportions of specific types of asset holdings display differences by race, but that they were not very extreme. He concluded that asset allocation was more balanced when families with similar social and demographic characteristics were compared. In 1994, African-Americans families had an average net worth equivalent to less than one-fifth that of white families (Gittleman & Wolff, 2000). Over 24 percent of white households received an inheritance with the average transfer amount equaling 115 thousand dollars. By comparison, only 11 percent of black households reported receiving inheritance monies and the mean value of the inheritance was 32 thousand dollars (Wolff, 2000). This analysis highlights the importance of inheritances and the disadvantaged position of blacks in the intergenerational transmission of wealth. The current situation must be contextualized in the history leading up to the current wealth gap. The ratio of mean black to white wealth by family remains consistently low, at approximately 0.19 (Wolff, 2000). That is, whites are roughly five times wealthier than blacks. 25 Wolff’s findings on inheritances in the SCF were consistent with Hao’s (2004) analysis of immigration and wealth using SIPP data. Acknowledging the difficulties associated with collecting early-age inheritance data, Hao (2004) conducted a separate analysis for the youngest age cohort in the SIPP as a way to distinguish youth inheritances from future wealth accumulation. Her analysis identified the largest earlyage inheritance deficit among African-Americans at the lower tails of wealth distribution. These findings support the importance of intergenerational transmission of wealth in later-stage wealth acquisition. As many wealth analysts note, simply stated, wealth begets wealth. Graph 4: Monthly median household income by household head and race, 2001 Married Couple HH Male HH Race Female HH Black White 0 1000 2000 3000 Income Source: SIPP, 2001 26 4000 5000 Wealth, Gender and Marital Status Gender, race and marital status of the head of household all appear to be correlated with wealth holdings. Households headed by women had lower median incomes, and the proportion of female-only householders was much higher for black households. Thirty-two percent of black households were headed by women compared to only ten percent for white households. Coupled with lower income among black than white households, this suggests that gender, race and marital status of the householder could play a role in wealth differences. Oliver and Shapiro (1995) noted the significance of gender in their analysis. In trying to explain it, they propose a ‘sociology of race and wealth,’ which combines economic human capital theory with different investment opportunities blacks and whites face. This thesis identifies related elements of human capital theory and sociological trends which contribute to the disproportionately higher share of female householders in black households. Elements of human capital theory relevant to wealth analysis include education, work experience and skills. Basically, blacks and whites face differential opportunities to build human capital, and this is important to take into consideration when analyzing wealth (Hao, 2004; Oliver & Shapiro, 1995). Data from the 2001 SIPP show that households headed by married couples had higher median incomes than households headed alone by either a male or female. Among single-parent household heads, men had higher median incomes than women, with a monthly median income of $3,240 compared to $2,149. As Table 1 and Graph 4 27 illustrate, the intra-gender differences are more pronounced when race is factored in. For example, households headed by single-parent white women had a median income of $2,381 compared to $1,679 for black female headed households. Other wealth measures indicate an even more dramatic difference by race. Households headed by single black women had net financial assets equal to zero compared to $900 for similar white households. This is consistent with Hao’s (2004) analyses of wealth data using the SIPP, which suggested that non-married households remained at the low-end of wealth distribution compared to married couple headed households with children. The latter show greater variation in wealth accumulation, depending on which stage of the life cycle the household is experiencing, i.e., married couple households’ wealth is highly variable relative to that of non-married households. Table 1: Total household wealth by race, gender and household type, 2001 Household type Married Couple households Male-headed households Female-headed households • White Female • Black Female Number of observations Median monthly income Net worth Net financial assets 17,961 13,410 $4,360 $96,303 $26,800 1,042 $3,240 $26,272 $2,000 3,509 $2,149 $5,354 $280 2,357 $2,381 $9,948 $900 1,152 $1,679 $1,500 $0 Source: SIPP, 2001 NOTE: households are the unit of observation 28 Education and Income Higher levels of education are correlated with higher incomes. However, returns to education suggest racial bias. Monthly median earned income for African-Americans with educational levels beyond a Bachelor’s degree was lower than that of similarly educated whites ($3,275 versus $4,584 in 2001). These findings are especially relevant for an analysis of wealth among the U.S. middle class. Middle class status is defined using Oliver and Shapiro’s (1995) parameters. These include: college-level education, total household income, and occupational category. Yet, even when comparing returns to education for college-educated blacks and whites, the difference was significant: blacks earned about $1,300 a month less than equally educated whites.8 Table 2 shows the overall trends in educational achievement by race and gender in the United States in 2000. By gender, the percentage of men with a college or advanced degree was higher than for women. By race, whites had higher educational attainment than blacks in all education categories. While these data from the Census Bureau provide a good benchmark to compare the SIPP findings to, it is important to note that these educational categories are not exclusive. For example, the figure of 80.1 percent of men with a high school diploma includes those with advanced degrees, as they would have completed high school before entering higher education. By contrast, SIPP household heads report their highest level of education in exclusive categories. For example, of the 22,533 white household heads, 2.7 percent held either a professional or 8 Data from SIPP, 2001 29 doctoral degree in 2001. Similarly, of the 3,606 black households, just under one percent held either a professional or doctoral degree. Table 2: Educational achievement by race and gender, 2000 Characteristic Population 25 and over Gender High school + Some College + College + Advanced Degree Male 87,077,686 80.1% 52.5% 26.1% 10% Female 95,133,953 80.7% 51.1% 22.8% 7.8% White 143,085,659 83.6% 54.1% 26.1% 9.5% Black 19,858,095 72.3% 42.5% 14.3% 4.8% Race Source: Bauman, K.J. & Graf, N.L, Educational Attainment: 2000, Census 2000 Brief 30 Table 3: Wealth by Race and Different Conceptions of the Middle Class, 2001 Monthly Net Net Financial Income Worth Assets White College degree $5,217 $161,548 $59,000 White collar $5,471 $93,454 $30,645 College degree $4,369 $55,280 $4,000 White collar $4,546 $21,214 $0 Black Source: SIPP, 2001 NOTE: Table 3 is an adaptation of Oliver and Shapiro’s definition of the middle class. All figures are median values. White collar is defined as earned monthly income between $3,248 and $6,496 in the Executive, Administrative, Managerial or Professional Specialty occupational category. Occupation categories and race Occupational categories provide an important tool to measure middle-class status which could offer wealth gains that are transferable from the current generation to the next. A larger percentage of African-Americans were in service sector occupations in 2001 (14.2 percent versus 6.9 percent for whites). Conversely, a much smaller percentage of blacks were in the executive managerial and professional sectors. These disparities imply that despite gains in employment opportunities, other factors play a role in keeping racial divisions in the workplace. Service sector jobs are associated with lower education, training and wages compared to managerial and professional jobs. As 31 such, blacks working in this sector are less able to provide enhanced life opportunities to their children. Lack of access to high-quality education and training in residentially segregated neighborhoods leaves blacks marginalized in the two occupational categories most closely associated with middle-class status, managerial and professional specialty. These differences result in a lack of intergenerational transfer of wealth. This perpetuates the black-white wealth gap as reflected in limited life opportunities, which lead to disparate life achievements. Lastly, an analysis of income and wealth by occupation supports the hypothesis that wealth measures reveal deeper racial inequality than income analyses. As demonstrated in Table 3, middle-class African-Americans earned $4,546 compared to $5,471 earned by middle class whites. The differences in net worth and net financial assets (NFA) were much more dramatic: middle-class blacks owned $21,214 in net worth compared to middle-class whites who owned $93,454. The NFA figures were even more striking, with middle-class blacks having $0 compared to $30,645 for whites. An interesting trend emerges when comparing Asian households to either white or black households; Asian households appear to outperform all other races in terms of income. An analysis of median incomes in the professional sector provides evidence supporting this trend: whites had a median monthly income of $5,585, blacks had $4,421 and Asians had $6,485 in 2001.9 Yet, wealth measures demonstrated marginalization by race: median wealth for Asian households in the 2001 SIPP was $58,285, substantially 9 These figures do not account for full or part time work status, which would attenuate the differences here. For example, women tend to work more part-time jobs than men, and African-American women are more likely to work part-time or be unemployed. Moreover, single African-American mothers were more likely than single white mothers to be in the labor force, suggesting sharper economic burdens on single black women (Bureau of Labor Statistics, 2006) 32 lower than white households. The figure for median NFA among middle-class Asian households was also significantly lower at $18,200. These findings are consistent with Hao’s (2004) comparison of wealth concentration between white and Asian households. She concluded that although Asian households have higher wealth holdings than other minorities, wealth is distributed more unequally for this group than white households. 33 Table 4 Regression of Income, Net Worth and Net Financial Assets Dependent variable Independent variable lnIncome lnNW lnNFA Household Income Black South Some College Age Age Squared Male Children Widow Lone parent Singles KBOccs Intercept R-squared N F-stat -0.271 (-9.59 )*** -0.154 (-8.88)*** 0.378 (20.76)*** 0.051 (14.56)*** -0.0005 (-15.48)*** 0.157 (8.98)*** -0.038 (-3.90)*** 0.262 (8.78)*** -0.529 (-20.58)*** -0.976 (-39.58)*** 0.611 (33.47)*** 0.0002 (21.49)*** -2.082 (-15.59)*** -0.053 (-0.63) 1.04 (11.81)*** 0.331 (21.38)*** -0.002 (-13.85)*** 0.499 (5.75)*** -0.015 (-0.32) 0.801 (5.52)*** -1.675 (-12.12)*** -1.473 (-12.89)** 0.29 (2.51)** 0.0002 (22.34)*** -1.972 (-24.80)*** -0.739 (-13.26)*** 2.011 (33.77)*** 0.101 (10.76)*** -0.0005 (-4.95)*** 0.3044 (5.33)*** -0.3541 (-11.94)*** 0.295 (2.73)*** -1.715 (-20.35)*** -1.588 (-20.22)*** 0.955 (14.11)*** 6.734 (71.52)*** -3.965 (-9.36)*** 2.808 (11.87)*** 0.19 26,139 720.4 0.18 26,139 503.07 0.27 26,139 907.42 Heteroskedasticity robust t-statistics in parentheses * significant at 10% level; ** significant at 5% level; *** significant at 1% level 34 Analysis Table 4 presents the results of the regression analyses for the three different dependent variables: income, net worth (NW) and net financial assets (NFA).10 The model builds on the work of Oliver and Shapiro (1995). However, because of the noted collinearity between income and total number of household workers, the model here used different independent variables including categorical variables for the head of household.11 The income regression provides a baseline to measure the NW and NFA findings against. Ordinary least squares (OLS) regression was used to tests three hypotheses: 1. Race is negatively associated with acquiring wealth for African-Americans 2. Age is positively correlated with income and negatively associated with NW and NFA 3. Female-headed, single-parent households are disadvantaged in acquiring wealth. These three hypotheses are a subset of the overarching hypothesis that wealth analysis reveals more deeply entrenched racial inequality than does income analysis. 10 The original model included the total number of household workers and work experience as independent variables. These were not included in the models here for several reasons including the potential collinearity between income and the number of workers, and the statistical insignificance of the work experience variable in all three models; dropping these variables did not significantly affect the results. 11 The independent variables used were tested for multicollinearity by running correlations between the different explanatory variables. The results suggest no collinearity in the explanatory variables. See Appendix C for correlation coefficient matrices for each regression. 35 Because the SIPP tends to over-sample the poor, many observations in the dataset had zero or negative values for NW and NFA. Indeed, it is precisely this feature of the SIPP that makes it such a powerful database for analyzing economic well-being among lower and middle-income U.S. households. Moreover, as Hao (2004) notes, mean and median values of NW demonstrate the positive skew of wealth measures. In the 2001 SIPP, mean NW was $166,365 while median NW was $53,085. Oliver and Shapiro (1995) substituted values of 1 for households with 0 values for NW, arguing that this allowed them to address the highly negative skew of wealth and include households with 0 wealth in their regressions. It is unclear how they dealt with negative values of NW. They then ran two versions of their regressions: the first used logged values of NW and NFA while the second did not. Because their findings were basically identical, they did not use logs in their final models, citing ease of interpretation and the ability to present unit changes in dollars (Oliver & Shapiro, 1995: 220). The study here uses logs to provide an equation that better fits the data. The initial OLS regressions on NW and NFA without logs had weak predictive power. However, logging the NW and NFA dependent variables truncated the data significantly, resulting in the loss of 4,500 to over 9,000 observations, depending on which variable was logged. 36 A better solution was found in Lingxin Hao’s (2004) work, which used SIPP data to analyze the impact of immigration on the distribution of U.S. wealth. To include zero and negative values of NW, Hao created transformed logs of NW as follows: 1. If NW if positive, the log of NW is simply the log of the value 2. If NW is zero, the log of NW is set equal to zero 3. If NW is negative, the absolute value of NW is logged and assigned a negative sign.12 Following Hao, NW and NFA were transformed into logs as described. For consistency, household income was also logged for the baseline regression. The Rsquared and F-values of these models suggests much better specification: for NW, the Rsquared is 0.18, close to the 0.20 value reported by Oliver and Shapiro and for NFA, it is 0.27, significantly higher than the 0.10 value reported in Oliver and Shapiro’s model.13 Yet, because NFA are considered a stronger predictor of control over future financial resources, i.e., inheritances, the model seems promising for future analyses of wealth using SIPP data and the adapted NFA log-transformed data. This is especially 12 Hao, 2004. Immigration and wealth inequality in the U.S. These findings suggest that the adapted model, explains more of the variation in wealth inequality than the Oliver and Shapiro’s model. This is likely the result of using the transformed log variables for NW and NFA, and retention of more of the observations for analysis. The regression diagnostics included in Appendix B show that the models appear properly specified and free of multicollinearity; however, there is the possibility of heteroskedasticity in the NW and NFA models. Despite the potential of heteroskedasticity, the models are still robust with the Huber-White transformed standard errors and tstatistics shown in parentheses in Table 4. This correction assumes heteroskedasticity is present and makes changes to the standard errors as described in the notes for Table 4. See Appendix B for the original regression results from Black Wealth/White Wealth. 13 37 relevant because of the policy implications for addressing the racial wealth gap. All else equal, blacks accumulate about 2 percent less wealth measured by either NW or NFA, relative to white households. This important finding supports the hypothesis that wealth analysis shows deeper inequality than does income analysis. That is, income differences attenuate the depth of racial inequality. Race has a negative impact in all three models, with the largest impact seen in the NW model. That is, African-Americans have lower measures of wealth relative to whites. The findings on race parallel those of previous studies on the wealth gap. Oliver and Shapiro (1995) found a highly negative and statistically significant impact of race on income, NW and NFA. As Hao’s (2004) quantile regression analyses suggest, the disproportionate representation of African-Americans in the lower distribution of wealth reflects important trends: discrimination in the housing market, an inheritance deficit, and low returns to lower levels of human capital acquired during childhood. Similarly, Conley (1999a) finds that distribution of wealth in the black community is more uneven than in the white community. Conversely, Conley (1999a) found that class impacts NW more than race, though he notes potential reverse causality between lower class status and race. Regressions were run using three different measure of wealth as the dependent variable – income, net worth (NW) and net financial assets (NFA). The results of these regressions follow. Except for the dummy variables for southern region residence and 38 children, all the other independent variables were statistically significant across all three models. The region and children variables were insignificant in the NW specification, which uses the sum of all assets minus liabilities, as the dependent variable. Oliver and Shapiro (1995) found a negative and statistically significant relationship between southern-region residence in their income and NW. The models here demonstrate that regional location was significant and negative for the income and NFA models but not for the NW model. This discrepancy may be the result of using different years of data, plus controlling for the type of household where NW can vary widely. Children were only statistically significant and negative in Oliver and Shapiro’s (1995) NFA model. This is only partially consistent with the findings here. The negative impact children can have on NFA was confirmed, but they also appeared to have a negative effect on income. In contrast to Oliver and Shapiro (1995), the impact of children on NW was highly variable. This could stem from the inclusion of the household type variable, as different households make different decisions about how much to invest in children, depending on which stage of the life-cycle households are in. For example, older households may experience a higher negative effect from having children as parents transfer lifetime savings to their children (Hao, 2004). The significance of income and age in determining NW and NFA provided empirical evidence supporting human capital theory. Age plays a positive and significant role in determining household income, NW and NFA. This is contrary to the hypothesis based on Oliver and Shapiro’s (1995) model, in which age was positive for the income 39 regression only, while negative for NW and NFA. The results in Table 4 demonstrate age having a positive impact in all three regressions. The relationship was strongest for NW: a one-year increase in age increases predicted NW by 39.23 percent, income by 5.23 percent and NFA by 10.63 percent.14 Similarly, the high t-statistics on income in the NW and NFA models suggest a robust positive relationship between income and these variables. In short, households acquire more wealth as they age, but at a diminishing rate, and higher incomes are associated with increases in wealth holdings. The sign on the age-squared variable was negative across all three models. As regression diagnostics [graphs (f) and (i)] in Appendix B of predicted values for NW and NFA against this variable indicate, diminishing returns seem to set in around the age of 64, consistent with Hao’s (2004) findings. In contrast, Oliver and Shapiro (1995) found a positive relationship between this variable and the two wealth measures. However, the finding here is consistent with trends depicted in Graph 1 and makes intuitive sense to the extent that older households transfer money for future generations. Hao’s (2004) work on immigration and wealth also suggests that households experience diminishing returns to age after the age of 64 as parents invest in the human capital of their children or provide other forms of intergenerational wealth transfers. 14 Applying simple algebraic properties of the logarithmic and exponential functions provides the exact percentage change in the predicted dependent variables: %∆y = 100 [exp(β)-1]. This is not an unbiased estimator because exp (.) is a nonlinear function. But it is a consistent estimator of 100[exp(β)-1] because the probability limit passes through the continuous functions while the expected value operator does not (Wooldridge, J.M., 2006: 198) 40 Another important difference from Oliver and Shapiro’s (1995) findings is the sign on the dummy variable for gender in all three models. They found a statistically significant positive relationship between males and income and negative relationships for men and NW and NFA.15 The regression results from the 2001 SIPP here demonstrate that being male appears to have a significant impact on acquiring income, NW and NFA. Coupled with the strong negative impact of lone-parent head of household status, the models here support the hypothesis that female-headed, single-parent households are less likely to acquire income and wealth. In the NW model, for example, a black femaleheaded household was predicted to have a decrease in NW based on race, household type, and being female. In other words, black female-headed households face a cumulative deficit in NW compared to white, male and married couple-headed households. These findings are consistent with previous studies on household type and wealth acquisition; both Conley (1999a) and Hao (2004) found that female-headed black households were disadvantaged in accumulating wealth. Conley’s (1999a) analysis of wealth among young households used the Panel Study on Income Dynamics and found that femaleheaded households had lower NW even after accounting for their lower incomes. The findings for the head of household independent variable were statistically significant and in the hypothesized direction. Compared to married couple-headed households, single-parent and never-married individual headed households were 15 Oliver and Shapiro (1995) note that the negative sign on the male variable in their NW and NFA specifications is a “statistical anomaly . . . created by the interaction between maleness and the number of workers in a household. In fact, the zero-order correlation between wealth and male householders is almost zero.” 41 marginalized in acquiring income and wealth. These findings are rooted in human capital theory and are further supported by the positive and statistically significant signs on the knowledge-based occupations variable in all three models. That is, knowledge-based occupations generate more income, NW and NFA than non-knowledge-based jobs such as services. The positive and significant coefficients for income and education in all three specifications suggest the need to complement current public policy based on human capital theory with additional policy measures to accelerate closing the racial wealth gap. For example, all else equal, having some college-level education results in a 37.8 percent increase in income, a 104 percent increase in NW and an over 200 percent increase in NFA. Moreover, income is positively and significantly related to both measures of wealth: each additional dollar of income generates an increase of 0.02 percent in NW and NFA or wealth more generally. The magnitude of the coefficient on income is supported in Lerner and Mikesell’s (1988) findings of the weak correlation between income and wealth, suggesting that while income is a necessary condition for wealth acquisition, it is not a sufficient one. Importantly, Hao (2004) argues that investments in human capital may be ineffective in closing the wealth gap if inheritances are the primary determinant of positive wealth during childhood. Following the classic life-cycle theory, Hao (2004) suggests that early-age inheritance is the most important factor in determining future 42 asset accumulation. Data on youth inheritances are difficult to obtain because they are hard to collect (Hao, 2004); future wealth research would benefit from enhanced data collection tools for younger age cohorts.16 Policy Implications Since the Civil Rights Act, many contend that African-Americans are offered equal legal protections and thus, equal opportunities. Yet, the prevalence of concentrated poverty and blacks’ disproportionate representation in lower-income jobs and impoverished neighborhoods imply that no policies to date have effectively bridged the racial divide. Blacks are disproportionately represented in service-sector jobs and earned less income for professional and managerial jobs compared to whites in 2001. As wealth analysts argue, policymakers’ exclusive focus on income as a measure of inequality obscures more complex, historically-based racial inequality. Cumulative disadvantages across generations contribute to the persistence of a black underclass, despite gains in education and access to better jobs. Nowhere is the role of intergenerational transmission of poverty more evident than in the black middle class and specifically, in regard to home equity. Although blacks have made significant strides in education, employment and politics, they continue to bear the brunt of the 16 Moreover, the 2001 SIPP has evolved since the time that the original models were done and macroeconomic shocks in the early 2000’s were shown to have erased blacks’ wealth gains during the 1990’s (Oliver & Shapiro, 2006). 43 cumulative disadvantages in acquiring wealth. Yet, the data analysis suggests that having some college-level education is positively associated with acquiring wealth, with the highest increase seen in NFA. Thus, expanding opportunities for higher education based on wealth criteria is crucial to closing the racial wealth gap. As the models tested here suggest, race is significantly and negatively correlated with wealth for African-Americans. Moreover, black women are disproportionately disadvantaged in wealth acquisition supporting the work of Conley (1999a); Hao (2004); and Oliver and Shapiro (1995 & 2006). It follows that policies designed to increase income equality are deficient in addressing racial inequality. A new set of policies to combat this situation began to develop in the early 1990’s. They included: (1) implementation of Individual Development Accounts (IDAs) which demonstrate the capabilities of lower-income people to save; (2) the 1998 passage of the Assets for Independence Act, which provides Federal funds for IDAs; and (3) revisions to the Temporary Assistance to Needy Families (TANF) eligibility criteria, allowing recipients to accumulate assets (Oliver & Shapiro, 2006: 245). Michael Sherraden (1991) proposed individual development accounts (IDAs) as a mechanism for asset-based welfare policy. IDAs are universal, matched savings accounts. Implementation of these individual-level accounts is in its nascent stages, but recent data suggest that IDAs are having a positive impact on homeownership (Mills et 44 al., 2004). In addition, preliminary impact evaluations of these programs demonstrate the positive “multiplier” effects of IDAs: grant recipients purchase homes, accumulate assets and display enhanced capacity to manage household finances (Mills et al., 2004). Similarly, Sherraden and Zhan (2003) find a positive association between asset accumulation and children’s educational achievement among single mothers. This is especially important given the findings that being African-American, female, and especially being a single female head of household are negatively and significantly correlated with accumulating wealth. More than 34 state-level TANF plans reference IDAs and 13 states are using their TANF funds for IDA programs (Parrish et al., 2006). Affirmative action programs need to be reevaluated. They are facing increased judicial resistance, suggesting new approaches should be explored. The findings here indicate that African-Americans are underrepresented in middle and upper-class jobs, and marginalized in higher education. Conley (1999a), for example, recommends a shift toward class-based affirmative action policies. Individual net worth would be accounted for in determining eligibility, thus providing African-Americans with most of the benefits, while simultaneously extracting race from the affirmative action debate. Moreover, current affirmative action policies do not do nearly enough within their existing framework to confront gender discrimination. More vigorous enforcement is desirable. Maximizing the net benefit from gender-based affirmative action under the current rules could address the disproportionate marginalization of low-income women in 45 wealth accumulation seen in the SIPP 2001 analysis. This would be especially important for single black parents, as suggested by the cumulative negative impact of race, gender and marital status of the household head described in the data analysis. Holzer and Neumark (2006) conclude that persistent education and employment gaps for American minorities warrant policies that complement affirmative action. The positive impact of knowledge-based occupations on wealth suggests the importance of education and training in determining economic well-being. Thus, expanding affirmative action in employment should be coupled with job training to ensure that individuals have the needed skills when considered for a job. International development policy may suggest a useful tool in addressing the wealth gap. The Grameen Bank’s system of microfinance has made remarkable achievements in reducing poverty in Bangladesh.17 For example, the Grameen Bank provides cell phones to women who live in rural areas with limited communications infrastructure. The recipient becomes the communications point-person for her village, and in turn, she begins to accumulate income and assets from her work (Stiglitz, 2006). The Grameen Bank’s micro-credit strategies could serve as a blueprint for empowering low-income American women, which is in keeping with the cumulative disadvantages faced by lone-parent female-headed households in asset accumulation shown here. 17 Indeed, Bangladesh appears to be the only developing country that can realistically reach the UN’s Millennium Development Goal of reducing poverty by half (M. Yunus, interview with Charlie Rose, 2006). 46 This policy tool is justifiable on two grounds: first, documented international success in poverty reduction by using this instrument; and, second, the ongoing statelevel encouragement of microenterprise work. Moreover, because the 1996 welfare revisions require TANF recipients to work, microenterprise offers the dual advantage of allowing recipients to continue receiving benefits and providing innovative opportunities to break the cycle of poverty. The latter is particularly important because of the demonstrated weakness of income alone to break families out of poverty. Finally, these recommendations are in line with the regional equity and asset building strategies proposed by scholars working in the new framework of asset-based policy (Oliver & Shapiro, 2006; powell, 2006; Rank, 2005; Shapiro, 2004). Building net worth and net financial assets are the sufficient conditions to complement the necessity of first increasing income. Conclusion Policymakers’ exclusive focus on income as a measure of inequality obscures more complex and historically-based asset inequality. Oliver and Shapiro’s (1995) analysis demonstrated that comparing blacks and whites of similar backgrounds still left about three-fourths of the wealth gap unexplained. The analysis of 2001 SIPP data supports the hypothesis that wealth differences suggest more racial inequality than do income differences. It follows that policies designed to increase income equality are deficient in addressing racial inequality. Cumulative disadvantages across generations 47 contribute to racial division in the United States. Based on the 2001 SIPP data, this thesis offers Civil Rights and low-income advocacy organizations a range of actionable policies to consider in closing the racial wealth gap. 48 APPENDIX A: Determining the Proper Model and Specification A. Addressing potential heteroskedasticity The ‘hettest’ function in STATA performs three versions of the Breusch-Pagan and Cook-Weisberg test for heteroskedasticity. All three versions of this test present evidence of heteroskedasticity.18 The steps for the Breusch-Pagan test are as follows: 1. Estimate OLS model and obtain the squared OLS residuals for each observation 2. Using them as the dependent variable, run the regression of u2=σ0+σ1x1+σ2x2 + . . . . . +σkxk +error. Save the R2 value. 3. Form the F-statistic using the R2 value from the preceding equation using the Fk, n-k-1 distribution; if the p-value is smaller than the chosen significance level, reject the null hypothesis of homoskedasticity. Because heteroskedasticity is observed in all three regressions, the STATA ‘_robust’ command is used to produce standard errors. This procedure computes a robust variance estimator based on a list of independent variables of equation-level scores and a covariance matrix. The heteroskedasticity robust t-statistics are generated using the White test, which adds the squares and cross products of all independent variables. Heteroskedasticity robust Huber-White t-statistics are presented in parentheses in Table 4.19 For large samples, robust standard errors are valid whether heteroskedasticity exists or not. 18 Adapted from STATA help menu Adapted from Introductory Econometrics: A Modern Approach (2006) by Jeffrey L. Wooldridge (3rd Edition). 19 49 APPENDIX B: REGRESSION DIAGNOSTIC GRAPHS To test for heteroskedasticity, residuals were plotted against the dependent variable. The initial results suggested that assumptions of OLS were violated, and that the error was correlated with the dependent outcome. Yet, because the ‘_robust’ command in STATA assumes violation of the homoskedasticity assumption, heteroskedasticity is accounted for as described in Appendix A. Graph a shows the pattern of the residuals and the dependent variable, the log of income; graph b shows predicted values of the log of income based on the heteroskedasticity robust model run with STATA, and it illustrates a higher degree of randomness. -20 -15 Residuals -10 -5 0 5 a) Heteroskedasticity test: residuals and lnHHINC -10 -5 0 lhhinc 50 5 10 -20 -15 Residuals -10 -5 0 5 b) Residuals and predicted values for lnHHINC (model specification scatterplot) 6 7 Fitted values 8 9 To assess non-linear returns to age, the predicted values for the log of income were plotted against the Age-squared variable. As discussed in the analysis section, diminishing returns to age appear to set in around 64 for all three measures of wealth. Graph c shows the fitted values of the log of income plotted against age-squared. 51 6.5 7 Fitted values 7.5 8 c) Fitted values of HHINC and Age squared variable: 0 2000 4000 Agesq 6000 8000 To visually assess the presence of heteroskedasticity, residuals of the log of net worth were plotted against the dependent variable, the log of NW. Graph d shows that the pattern of the residuals is non-random, i.e., there is the potential of heteroskedasticity. Still, using the Huber-White transformed t-statistics provides robust standard errors, and the coefficients themselves remain unbiased. 52 -30 -20 Residuals -10 0 10 20 d) Heteroskedasticity test: residuals and lnTNW: -20 -10 0 lNW 10 20 To test whether the model for the net worth regression was correctly specified, residuals were plotted against the dependent variable, the log of NW. Although the patterns of the residuals in Graph e are non-random, the ‘_robust’ command in STATA assumes violation of the homoskedasticity assumption and generates heteroskedasticity robust Huber-White standard errors; thus, model specification is not a problem. 53 -30 -20 Residuals -10 0 10 20 e) Residuals and predicted values for lnTNW: 0 10 20 Fitted values 30 40 To test the returns to age, the age-squared variable was plotted against the predicted values for the log of NW. As Graph f indicates, the data suggest diminishing returns to age set in around the age of 64, consistent with other analyses of wealth and age. 54 2 4 Fitted values 6 8 10 12 f) Fitted values of log Net Worth and Age squared variable: 0 2000 4000 Agesq 6000 8000 A plot of residuals against observed net financial assets (NFA) was used to test for the presence of heteroskedasticity. As Graph g indicates, the errors are not normally and identically distributed, so the STATA ‘_robust’ command was used to allow for violation of the homoskedasticity assumption to make the model robust despite this problem. 55 -30 -20 Residuals -10 0 10 20 g) Heteroskedasticity test: residuals and NFA -20 -10 0 lNFA 10 20 As with the income and net worth regressions, residuals were plotted against the predicted values of NFA to check for model specification. Because the heteroskedasticity is accounted for by the Huber-White correction, the model is robust despite the non-random pattern of residuals seen in Graph h, which shows a larger degree of randomness. 56 -30 -20 Residuals -10 0 10 20 h) Residuals and predicted NFA (model specification scatterplot): -10 0 10 Fitted values 20 30 Plotting the predicted NFA values based on the model used in this analysis against the age-squared variable indicates similar results to those seen in the income and NW models, i.e., that diminishing returns to age set in around the age of 64 as seen in Graph i. 57 5 6 Fitted values 7 8 9 i) Fitted log NFA values and Age squared variable: 0 2000 4000 Agesq 58 6000 8000 Table B-1: Regression results from Black Wealth/White Wealth, Oliver & Shapiro (1995: 130) Regression of Income, Net Worth, and Net Financial Assets Intercept Race South Highest Grade Completed Age Age Squared Work Experience Upper-White Collar No. of Workers Household Income Male Children Widow 2 R N *p<0.05 Income Net Worth Net Financial Assets -33,035.00*** -5,176.76*** -3,072.80*** 831.37*** 1,390.49*** -12.91*** -40.42 32,243.00 -27,075.00*** -9,352.00*** 666.18** -3,620.57*** 72.03*** -104.33 44,888.00*** -14,354.00*** -1,958.15 396.29 -3,442.74*** 54.31*** -6.29 3,705.02*** 11,401.00*** 28,635.00*** -10,715.00*** 23,841.00*** -9,259.61*** 28,11.54*** 300.23 -6,173.47*** 0.386 7,625 2.28*** -9,389.11 -778.70 -6,898.71 0.203 7,625 1.31*** -6,703.24*** -2,783.85** -8,581.18 0.107 7,625 **p<0.01 ***p<0.001 59 60 lnHHINC Black South SomeColl Age Agesq Male Children Widow Lonepar Singles KBOccs lnHHINC 1 -0.1268 -0.0699 0.2112 -0.0487 -0.0791 0.1316 0.07 -0.1437 -0.0965 -0.2633 0.2319 1 0.1562 -0.0841 -0.0316 -0.0331 -0.116 0.0697 0.0195 0.2098 -0.0008 -0.0581 black 1 -0.0529 0.0089 0.0096 -0.0182 -0.0031 0.0172 0.0212 -0.021 -0.0304 South 1 -0.1488 -0.1643 0.044 -0.0096 -0.1601 -0.0982 -0.003 0.3288 SomeColl 1 0.9857 -0.0264 -0.3815 0.4851 -0.1394 0.1575 -0.1858 Age 1 -0.042 -0.386 0.5237 -0.133 0.1859 -0.204 Agesq 1 -0.033 -0.244 -0.265 -0.095 0.0047 male 1 -0.1749 0.2121 -0.4227 0.0295 Children 1 0.0742 0.3393 -0.14 Widow 1 -0.3098 -0.0595 Lonepar 1 0.0174 Singles 1 KBOccs To test for multicollinearity, correlations were run between the explanatory variables for each of the three models. The only high correlation (over 0.7) is seen between the Age and Age-squared variables; these figures are highlighted in the matrices below. This is to be expected given that one is a transformation of the other. Appendix C: Multicollinearity and Correlation Coefficients 61 lnTNW HHINC Black South SomeColl Age Agesq Male Children Widow Lonepar Singles KBOccs lnTNW 1 0.195 -0.1565 -0.0311 0.0918 0.2965 0.2694 0.0938 -0.0943 0.0737 -0.1589 -0.06 0.0407 1 -0.1121 -0.0668 0.2532 -0.1113 -0.1454 0.1345 0.0953 -0.1796 -0.117 -0.2363 0.2705 HHINC 1 0.1562 -0.0841 -0.0316 -0.0331 -0.116 0.0697 0.0195 0.2098 -0.0008 -0.0581 black 1 -0.0529 0.0089 0.0096 -0.0182 -0.0031 0.0172 0.0212 -0.021 -0.0304 South 1 -0.1488 -0.1643 0.044 -0.0096 -0.1601 -0.0982 -0.003 0.3288 SomeColl 1 0.9857 -0.0264 -0.3815 0.4851 -0.1394 0.1575 -0.1858 Age CORRELATION COEFFICIENT MATRIX: NET WORTH REGRESSION 1 -0.042 -0.386 0.5237 -0.133 0.1859 -0.204 Agesq 1 -0.033 -0.244 -0.265 -0.095 0.0047 male 1 -0.1749 0.2121 -0.4227 0.0295 Children 1 0.0742 0.3393 -0.14 Widow 1 -0.3098 -0.0595 Lonepar 1 -0.0174 Singles 1 KB Occs 62 lnNFA HHINC Black South SomeColl Age Agesq Male Children Widow Lonepar Singles KBOccs 1 0.3359 -0.2361 -0.1194 0.2853 0.1543 0.1307 0.1309 -0.1056 -0.0382 -0.2155 -0.0983 0.1856 1 -0.1121 -0.0668 0.2532 -0.1113 -0.1454 0.1345 0.0953 -0.1796 -0.117 -0.2363 0.2705 1 0.1562 -0.0841 -0.0316 -0.0331 -0.116 0.0697 0.0195 0.2098 -0.0008 -0.0581 1 -0.0529 0.0089 0.0096 -0.0182 -0.0031 0.0172 0.0212 -0.021 -0.0304 1 -0.1488 -0.1643 0.044 -0.0096 -0.1601 -0.0982 -0.003 0.3288 1 0.9857 -0.0264 -0.3815 0.4851 -0.1394 0.1575 -0.1858 1 -0.042 -0.386 0.5237 -0.133 0.1859 -0.204 1 -0.033 -0.244 -0.265 -0.095 0.0047 CORRELATION COEFFICIENT MATRIX: NET FINANCIAL ASSETS REGRESSION lnNFA HHINC black South SomeColl Age Agesq male 1 -0.1749 0.2121 -0.4227 0.0295 Children 1 0.0742 0.3393 -0.14 Singles 1 -0.3098 1 -0.0595 -0.0174 Widow Lonepar 1 KB Occs References Alba, R.D., Logan, J.R. & Stults, B.J. (2000). How segregated are middle-class African Americans? Social problems, 47, 4, 543-558. Atlanta Journal & Atlanta Constitution (1989). The color of money: Home mortgage lending practices discriminate against blacks. Atlanta, GA, The Atlanta Journal and The Atlanta Constitution. Avery, R.B., Canner, G.B. & Cook, R.E. (2005). New information reported under HMDA and its application in fair lending enforcement. Federal reserve bulletin, Summer 2005, 344-394. Bauman, K.J. & Graf, N.L. (2003). Educational attainment: 2000. Census 2000 Brief, Washington, D.C., U.S. Census Bureau. Bureau of Labor Statistics (2006). Charting the U.S. labor market in 2005. Washington, D.C., U.S. Department of Labor. Blau, F.D. & Graham, J.W. (1990). Black-white differences in wealth and asset composition. The quarterly journal of economics, 105, 2, 321-339. Browne, R.S. (1972). The economic case for reparations to black America. American economic review, 62, 39-46. Charles, K.K. & Hurst, E. (2002). The transition to homeownership and the black-white wealth gap. The review of economics and statistics, 84, 281-297. Conley, D. (1999a). Being black, living in the red: Race, wealth and social policy in America. Berkeley, CA, University of California Press. Conley, D. (1999b). Getting into the black: Race, wealth and public policy. Political science quarterly, 114, 4, 595-612. Consumer Federation of America (2005). Consumer federation of America analysis of the Federal reserve bulletin’s release of the 2004 Home Mortgage Disclosure Act (HMDA) data. Washington, D.C., Consumer Federation of America. Retrieved on November 24, 2006 from www.consumerfed.org. Cose, E. (1993). The rage of a privileged class. New York, NY, Harper & Collins. Darity, jr. W. & Frank, D. (2003). The economics of reparations. American economic review, 93, 2, 326-329. 63 Gittleman, M. & Wolff, E.N. (2000). Racial wealth disparities: Is the gap closing? Jerome Levy economics institute, Working paper No. 311, New York, Bard College. Glover-Blackwell, A. (2006). Ensuring broad access to affordable neighborhoods that connect to opportunity. In The covenant with black America (pp. 97-121). Chicago, Third World Press. Hao, L. (2004). Immigration and wealth inequality in the U.S. Paper presented at: The Population Association of America 2004 Annual Meeting, Boston, MA. Holzer, H.J. & Neumark, D. (2006). Affirmative Action: What do we know? Journal of policy analysis and management, 25, 2, 463-490. Jackman, M.R. & Jackman, R.W. (1980). Racial inequalities in homeownership. Social forces, 58, 4, 1221-1234. Jackson, K.T. (1985). Crabgrass frontier: The suburbanization of the United States. New York, NY. Oxford University Press. Keister, L.A. & Moller, S. (2000). Wealth inequality in the United States. American review of sociology, 26, 63-81. Lerman, D.L. & Mikesell, J.J. (1988). Rural and urban poverty: An income/net-worth analysis. Policy Studies Review, 7, 765-81. Lerman, R.I. (2005). Are low-income households accumulating assets and avoiding unhealthy debt? A review of recent literature. Opportunity and ownership project, 1. Washington, D.C., The Urban Institute. Massey, D.S. & Denton, N. (1993). American Apartheid: Segregation and the making of the underclass. Cambridge, MA, Harvard University Press. Massey, D.S. & Denton, N. (1988). Suburbanization and segregation in US metropolitan areas. American journal of sociology, 94, 592-626. Mauer, M. & King, R.S. (2004). Schools and prisons: 50 years after Brown v. board of education. Washington, D.C., The Sentencing Project. McKinnon, J. (2001). The black population: 2000. Census 2000 Brief, Washington, D.C., U.S. Census Bureau. Mills, G., Patterson, R., Orr, L., & Demarco, D. (2004). Evaluation of the American dream demonstration, final report. Cambridge, MA, Abt Associates. 64 Munnell, A.H., Tootell, G.M.B., Browne, L.E & McEneaney (1996). Mortgage lending in Boston: Interpreting HMDA data. American economic review, 86, 1, 25-53. Oliver, M.L. & Shapiro, T.M. (1990). Wealth of a nation: a reassessment of asset inequality in America shows at least one-third of families are asset poor. American journal of economics and sociology, 49, 2, 129-151. Oliver, M.L. & Shapiro, T.M. (1995). Black wealth/white wealth: New perspective on racial inequality. New York, NY, Routledge. Oliver, M.L. & Shapiro, T.M. (2006). Black wealth/white wealth: New perspective on racial inequality (10th anniversary ed.). New York, NY, Routledge. Popkin, S.J., Katz, B, Cunningham, M.K., Brown, K.D., Gustafson, J. & Turner, M.A. (2004). A decade of HOPE VI: Research findings and policy challenges. Washington, D.C., The Urban Institute & The Brookings Institution. Parrish, L., McCulloch, H., Edwards, K. & Gunn, G (2006). State policy options for building assets. Washington, D.C., St. Louis, MO, New America Foundation & Center for Social Development. powell, j. (2002). Racism and metropolitan dynamics: The Civil Rights challenge of the 21st century. Briefing paper prepared for the Ford Foundation. Minneapolis, MN, The Institute on Race & Poverty. Rank, M.R. (2004). One nation underprivileged: Why American poverty affects us all. New York, NY, Oxford University Press. Rose, Charlie. (September 1, 2006). The Charlie Rose show. New York, NY: WNET. Schwartz, A. & Tajbakhsh, K. (1997). Mixed income housing: Unanswered questions. Presented at an international conference on housing and the built environment organized by the International Sociological Association, Lexington, VA. Shapiro, T.M. (2004). The hidden costs of being African-American: How wealth perpetuates racial inequality. New York, NY, Oxford University Press. Sherraden, M. (1991). Assets and the poor: A new American welfare policy. New York. Sharpe. Smith, E.L. (2000). Prescription for wealth: Declaration of financial empowerment for African Americans. Black Enterprise, January, New York, NY. 65 Solomon, R. (2005). Public housing reform and voucher success: Progress and challenges. Discussion paper prepared for the Brookings Institution Metropolitan Policy Program. Washington, DC, Brookings Institution. Spriggs, W. (2004). Event transcript: Wealth inequality panel. Washington D.C., Center for American Progress. Straight, R.L. (2002). Wealth: Asset-accumulation differences by race: SCF data, 1995 and 1998. American economic review, 92, 2, 330-334. The sentencing project (2001). Report summary - Young black Americans and the criminal justice system: Five years later. Retrieved on November 26, 2006 from http://sentencingproject.org/pdfs/9070smy.pdf (last revised April, 2001). Stiglitz, J. E. (2006). Making globalization work. New York, NY, London, U.K., W.W. Norton & Co. Survey of Income and Program Participation (2001). Survey of income and program participation: User’s guide (3rd ed.). Washington, DC. Census Bureau. Prepared by Washington DC, Mathematica Policy Research, Inc. & Rockville, MD, Westat. Terrell, H.S. (1971). Wealth accumulation of black and white families: the empirical evidence. The journal of finance, 26, 2, 363-377. U.S. Census Bureau (2001). Current Population Survey, Table IE-6 Measures of Household income inequality 1967-2001. Retrieved on October 12, 2006 from http://www.census.gov/hhes/income/histinc/ie6.html (last revised May 14, 2004) United Nations (1969). International convention on the elimination of all forms of racial discrimination. Retrieved from http://www.ohchr.org/english/law/pdf/cerd.pdf on December 1, 2006. Wolff, E.N. (1998). Recent trends in the size distribution of household wealth. The journal of economic perspectives, 12, 3, 131-150. Wolff, E.N. (2000). Recent trends in wealth ownership, 1983-1998. Jerome Levy economics institute, Working paper No. 300, New York, Bard College. Wolff, E.N. (2002). Inheritances and wealth inequality, 1989-1998. American economic review, 92, 2, 260-264. Wooldridge, J.M. (2006). Introductory econometrics. Mason, OH, Thomson SouthWestern. 66 Zhan, M., & Sherraden, M. (2003). Assets, expectations, and children’s educational achievement in single-parent households. Social service review, 77, 2. 191-211. 67