Marriage, Work, and Racial Inequalities in Poverty: Evidence from the U.S

Abstract

This paper explores recent racial and ethnic inequalities in poverty, estimating the share of racial poverty differentials that can be explained by variation in family structure and workforce participation. The authors use logistic regression to estimate the association between poverty and race, family structure, and workforce participation. They then decompose between-race differences in poverty risk to quantify how racial disparities in marriage and work explain observed inequalities in the log odds of poverty. They estimate that 47.7–48.9% of black-white differences in poverty risk can be explained by between-group variance in these two factors, while only 4.3–4.5% of the Hispanic-white differential in poverty risk can be explained by these variables. These findings underscore the continued association between racial disparities in poverty and those in labor and marriage markets. However, clear racial differences in the origin of poverty suggest that family- and worked-related policy interventions will not have uniformly effective or evenly distributed impacts on poverty reduction.

Introduction

In the years since the Great Recession of 2007–2009, there is renewed interest in questions about work, poverty, and inequality. Recent attention to these questions is well-founded. By the end of the downturn, in 2010, the national poverty rate had reached 15.1%, its highest level in nearly twenty years. In absolute terms, this figure translated to over 46 million Americans living below the poverty line, which is the largest poor population recorded in the more than five decades the federal government has tracked poverty statistics (DeNavas-Walt et al. 2011). Underlying these overall trends, the negative impacts of the recession were unevenly distributed across racial groups. For example, inflation-adjusted median wealth fell by 53% among non-Hispanic black (henceforth “black”) households and 66% among Hispanic households, compared to just 16% among non-Hispanic whites (henceforth “white”), pushing racial wealth gaps to their highest levels in a quarter-century (Kochhar et al. 2011). Disproportionately negative impacts from the economic crisis for racial minorities vis-à-vis whites were similarly observed with respect to earnings, employment, and related economic outcomes (Hoynes et al. 2012). By 2010, the poverty rate among the black (27.4%) and Hispanic (26.6%) populations was more than twice that among whites (13.0%) (DeNavas-Walt et al. 2011).

Recent patterns of racial inequality are well documented, but few studies have attempted to unpack the proximate determinates of these differences. We begin to address this issue in the current paper, with a particular emphasis on understanding the respective contributions of family structure and employment to racial disparities poverty. This analysis builds on a recent article in this journal by Baker (2015), which documented changes over time in the respective associations between marriage and work on the one hand, and poverty on the other. Baker’s analysis showed substantial temporal changes in these relationships, but did not interrogate the large racial and ethnic disparities that are our focus here, and are central to questions about the links between work, family structure, and poverty in the U.S.

Our effort to address this gap contributes to two bodies of research. First is the literature examining the links between work and poverty (Brady et al. 2010, Iceland and Kim 2001, Thiede et al. 2015), which has shown that the poverty-reducing effects of employment are not as strong as sometimes assumed, and vary systematically by racial group. Second, and perhaps most centrally, our research adds to the large literature on the links between family structure and poverty (Eggebeen and Lichter 1991, Iceland 2003, Lichter and Landale 1995, Lichter et al. 2005, Sawhill 1988). This research has shown that changes in family structure have had substantial proximate effects on the poverty rate (Iceland 2003, Sawhill 1988), and has documented salient racial inequalities in these dynamics (Eggebeen and Lichter 1991). As well, and importantly for our purposes, prior analyses have shown that family structure has historically accounted for large shares of inequality between racial groups (Lichter et al. 2005). However, previous research has also shown that the link between family structure and poverty has varied over time due to changes in the labor market (e.g., changes in female labor market outcomes) and the family (e.g., the increase of cohabitation) (Baker 2015, Iceland 2003, Thomas and Sawhill 2005). This evidence suggests the need for updated estimates, particularly in light of continued changes in family formation and the period of economic crisis and stagnation around the Great Recession. We provide such updated estimates here, while also expanding upon the methodology of prior work in important ways.

More specifically, we use recent data from the Current Population Survey (2012–2014) to document disparities in family structure, work, and poverty, and compare the strength of the associations between these variables across racial groups. We then decompose racial disparities in poverty into components attributable to differences in family structure, work, and other characteristics expected to be associated with poverty. Our results complement Baker’s (2015) recent analysis of historical trends in the relationship between marriage, work, and poverty (among children). We expand upon her approach by not only estimating differences in the returns to work and marriage, but also using regression-based decomposition techniques to account for the contribution of compositional differences to racial differentials in poverty rates. This substantive focus also represents an update and extension of work by Eggebeen and Lichter (1991) and Lichter and colleagues (2005). These studies motivate our analysis by demonstrating that the importance of work and family structure for poverty reduction (or generation)—the focus of Baker’s (2015) paper—may differ from their contribution to racial inequality, which we address here. Our multivariate approach also represents a methodological extension of prior work on this topic, which has generally utilized traditional standardization approaches. Finally, our analyses also expand upon much of the previous research on this topic by disentangling the potentially confounding effects of family structure and work (see Lichter and Landale 1995 for a notable exception) and accounting for racial differences in cohabitation. Our overall goals are largely descriptive: to provide a statistical portrait of how racial differences in the attainment of marriage and work contribute to disparities in poverty, and to document the effects of other factors that have often received less attention in debates about poverty and social policy. Our findings show that work and marriage are important, but also highlight variation in the sources of poverty and racial inequality. A key implication of our findings is that universal policy interventions are unlikely to have uniformly effective and evenly distributed impacts on poverty reduction across racial groups.

Background

The effects of work and family structure on poverty risk are in many respects straightforward. In fact, the association between these variables is due in part to how poverty is officially defined in the U.S. According to the official measure, a person is identified as poor if they live in a family with pre-tax income below a threshold defined according to family size, age of the householder, and number of related children (<18 years) in the family. In 2015, for example, the poverty threshold for a single working-age adult was $11,367, $19,078 for a family of three with a single working-age adult and two children, and $24,036 for a family of four with two working-age adults and two children. In each of these scenarios, the earnings required for a single worker to remain above the poverty line is a function of family structure, as is the likelihood that one or more co-resident workers will supplement that primary worker’s earnings. In a mechanical manner, family size and structure determine the income required to escape poverty, as well as the number of workers potentially available to generate income.

Yet many other variables intersect with family size and structure, and have the effect of producing systematic differences in economic outcomes between social groups. Race and ethnicity are prominent and longstanding axes of inequality in the U.S. labor market, along which marked differences in the likelihood of employment, occupational attainment, and earnings exist. For instance, the current unemployment rate among black adults is more than twice the white unemployment rate, and earnings among black workers are only 60% of what their white counterparts earn (Bureau of Labor Statistics 2016). This type of inequality has clear implications for the strength of association between work and poverty across racial groups. Indeed, a recent study estimated that 6.0–7.2% of white workers were poor in 2012, less than half the rate observed among black (16.2–19.7%) and Hispanic workers (19.9–22.1%) (Thiede et al. 2015). Given evidence that such disparities are driven in part by differential labor market experiences—including discrimination—and accumulation of human capital between majority and minority groups, we expect both levels of work and the returns to work to vary significantly across racial groups (del Rio and Alonso-Villar 2015, Gradín 2013, Grodsky and Pager 2001, Massey 2007).

Regarding family structure, recent evidence shows a strong relationship between adults’ marital status and total family income. For example, in 2014 the median income for married-couple households was $81,025. This figure is more than twice the median income for households headed by single females ($36,151), and nearly $30,000 more than that of single-male headed households ($53,684) (DeNavas-Walt and Proctor 2015). These differences are in large part a function of labor supply within the family. All else equal, a family with two working-age adults in the household is more likely to have at least one worker, and therefore generate more income, than families with only a single adult. Moreover, marriage can provide additional benefits through two widely-recognized mechanisms (Amato and Maynard 2007, Baker 2015, Becker 1981, Thomas and Sawhill 2005, Waite 1995): economies of scale (i.e., sharing expenses) and specialization within the household, which increases efficiency and wages.

Unsurprisingly, in a context of stagnating median wages and declines in the real minimum wage, dual- and multi-worker households have become increasingly common (Waite and Nelson 2001). For a growing share of the population today, multiple workers are needed for families to avoid poverty. This imperative has led to growing disadvantages for those with just one worker, which by definition is an unavoidable circumstance for single adults and the families they head (Baker 2015). An important caveat here is the rise of families headed by cohabiting adults (Seltzer 2000, Smock 2000). Because, like married couples, cohabiting adults have the potential advantage of dual-earner strategies—as well as economics of scale and household specialization in the division of labor—their families tend to attain higher economic status than those headed by single adults (Manning and Brown 2006, Thomas and Sawhill 2005). It is therefore important to conceptualize cohabiting adults as a distinct family type.

The link between family structure and poverty may have important implications for racial inequality given well-documented racial disparities in marriage and non-martial fertility. Black (55%) and Hispanic children (31%) are far more likely to live with only a single parent compared to white children (21%) (Vespa et al. 2013). The material implications of these differences are clear in recent estimates that show a poverty rate of 6.2% (3.7 million families) among married couples, 15.7% (1.0 million families) among single-male heads, and 30.6% (4.8 million families) among single-female heads (DeNavas-Walt and Proctor 2015). Other research shows that cohabitating couples occupy a middle ground between families headed by married couples and single adults in terms of the likelihood of being poor (Manning and Brown 2006). Gender inequality in the labor market may also be a key factor undergirding poverty differences according to family structure. Gender inequality compounds the disadvantages of single women relative to single men, and in turn intersects with racial inequality (Browne and Misra 2003). That is, in addition to economic differentials attributable to race and family structure, the disproportionate representation of women among families headed by a single parent adds yet another layer of disadvantage for those in such living arrangements. Such multidimensional disadvantage is expected to be particularly strong among families headed by black and Hispanic women, who face constraints associated with both race and gender.

Another important finding in the literature on family structure and well-being is that differentials by family structure often reflect selection effects. People who marry and remain in stable marriages are increasingly selective with respect to education and income (Brown 2010, Gibson-Davis 2009, Kuo and Raley 2014, Lichter et al. 2006, Musick et al. 2012). As a consequence, one might expect marriage to reduce poverty odds above and beyond the simple effect of joining incomes and providing more degrees of freedom in terms of labor supply. The extent to which selection processes are present and vary across groups will further shape racial differences in work-poverty and family structure-poverty linkages. And again, prior research suggests this is the case. For example, Wilson (1987, 1996) argued that the less attractive employment situations of black men create disincentives for the formation of marriages, a situation compounded by disproportionately high black male incarceration rates (Lopoo and Western 2005, Western and Wildeman 2009). This example points to a broader point: structural constraints distribute the likelihood and benefits of marriage unequally across groups (Huffman and Cohen 2004, Lichter et al. 1992, McLoyd et al. 2000). This notion is supported by research showing that marriage provides greater economic advantages to white children compared to their black and Hispanic counterparts (Manning and Brown 2006).

Finally, we note that family age composition is associated with poverty odds. For one, the poverty threshold is set according to whether the householder is below age 65 (all families in our study), and the number of family members below the age of 18. Children are assumed to use fewer resources than adults, so each additional child in a family increases the poverty threshold less than an additional adult. However, the practical implications of such age composition effects are more ambiguous given that children also lack income-generating potential. As well, fertility rates have often been higher among low-income women than among those in the middle- and upper-range of earners, for which relatively low fertility has been more common (Eggebeen and Lichter 1991). This observation has important overlap with racial inequality in light of evidence that a large and growing share of births is occurring to mothers from historically disadvantaged racial and ethnic groups (Johnson and Lichter 2008). Such racial and ethnic variation in fertility corresponds to a large share of children, particular from racial minority groups, being born into poverty (Lichter et al. 2015).

In sum, broad associations between marriage, work, and poverty are well documented, but these relationships are also characterized by considerable complexity. The fact that poverty is defined as a function of income, family size, and age composition would suggest straightforward associations between poverty on the one hand, and the number of workers and family structure on the other. However, group differences in the likelihood of employment and wages—driven by racial and gender discrimination, and an array of structural factors affecting workers’ skills—influence whether adult family members are employed, and whether their earnings are sufficient to avoid poverty. Other characteristics associated with marital status and fertility may also shape estimated associations between these outcomes and poverty across various groups. As a consequence, the seemingly mechanical relationship between work, family structure, and poverty is in fact moderated by structural factors that operate above and beyond individual decisions about marriage and work. Capturing these nuances can provide new insights for social science research on poverty, as well as for social policy aimed at ameliorating economic hardship. These insights are much needed in a context of rapidly increasing racial and ethnic diversity, changing family structures, and an economy still marked by the devastating impacts of the Great Recession (Lichter 2013). We unpack these considerations in the analysis that follows.

Data

Our analyses draw on micro-data from the March Supplement of the Current Population Survey (CPS) (King et al. 2010). The CPS is based on a nationally representative sample of approximately 60,000 households, and is the primary source of labor force statistics for the U.S. The March Supplement includes detailed information on prior-year income and employment, and is a frequently used source of data for research on work and poverty. Our main analyses pool data from the 2012–2014 CPS, which corresponds to the 2011 to 2013 period for prior calendar year income used to determine poverty status.

We restrict our sample in three ways. First, we drop members of the fifth through eighth rotating groups in the 2013 and 2014 data to account for repeated observations attributable to the CPS sampling structure. Second, we consider only household reference persons (i.e., householders) aged 18–64 at the time of the survey: families with working-age reference persons are our units of analysis. The householder is defined by the Census Bureau as the person to whom the place of residence is owned or rented, and can be either spouse when a married couple jointly owns or rents a dwelling. Since poverty is a family-based measure (see below), this restriction has the implication of limiting our analysis to members of primary families (i.e., excluding members of unrelated secondary families within households), except when the reference person has a cohabitating partner. In those cases, we include that partner and their family as a part of the primary family unit used in the analysis. Third, we consider only householders self-identified as non-Hispanic white, non-Hispanic black, or Hispanic (any race). For brevity, we refer to the former two groups as white and black throughout the text. We do not disaggregate the Hispanic population by ethnicity or country of origin, but note that 89.3% of the Hispanic population in our sample self-identified as white and 5.2% as black. The three most common countries or regions of origin are Mexico (66.3% of our sample), Puerto Rico (10.8%), and Central and South America (10.8%). The small number of observations from other racial and ethnic groups—including the population identifying as multi-racial—precludes reliable estimates of group-specific multivariate models. Note that our main analytic sample includes family and single-person households, which reduces the risk of biases associated with selection into family households (e.g., via fertility). For comparison with other studies, however, we replicated our analysis with only family households and found the substantive conclusions from our main analysis to be robust.

Measures

Our dependent variable is family poverty status as defined by the official, family size-adjusted U.S. government thresholds for the year of observation. For families with cohabiting heads, we define poverty thresholds using the adjusted family unit described above. While acknowledging serious limitations to the official poverty measure (Brady 2003), we employ it in this analysis for the purposes of comparability with other relevant studies and the commonly-published headline estimates of poverty that continue to drive most social policy decisions. That said, we note that the official poverty measure does not account for the effects of the Earned Income Tax Credit (EITC), which benefits many low-income workers. Holding work-related costs and other factors constant, omitting the effects of the EITC will downwardly bias estimates of the poverty-reducing effects of work.

Our analyses focus on work and family structure, which we operationalize as two categorical variables in our main analysis. The first variable distinguishes between families with 0 (reference), 1 or 2+ workers within each marital status group. We define workers as any person in the primary family unit between the ages of 18 and 64 who worked at least one week during the preceding calendar year. This definition has the benefit of not applying an arbitrary threshold of hours or weeks to classify workers, but does allow considerable heterogeneity among workers in terms of hours and weeks employed. We discuss this issue further below. We restrict our definition of workers to working-age adults given that the presence of teen and/or 65+ year-old workers may be a function of poverty. This restriction is also consistent with normative claims that youth and older adults should not be forced to work by economic circumstances.

The second main variable of interest accounts for marital status and, among families headed by single adults, the gender of the family head. That is, we disaggregate families according to whether the head was married (reference), cohabitating, an unmarried male, or an unmarried female. This approach allows us to account for how gender inequalities in the labor market translate into unique economic disadvantage among families headed by single women, but treats married-couple households as homogeneous since identification of the reference person among spouses can be arbitrary. We do not distinguish cohabitating families by the reference person’s gender due to similar questions about self-reporting.

There are two potential limitations to how we have operationalized these variables. First, the number of workers and family structure may be highly correlated such that certain work-family structure combinations are very rare; and second, our measure of work may be too coarse and not capture differences in the number of hours and weeks that each worker was actually employed. To address the first concern, we combine all work-family structure combinations (so defined) into a single twelve-category variable—included in the descriptive statistics (Table 1) for reference—and replicate the regression and decomposition analyses described below. For each between-race comparison, the share of the poverty gap explained by this twelve-category variable is very similar to the total share explained by the two separate work and family structure variables. We present results using the latter pair of variables since it provides detail about the respective contributions of each factor to racial inequalities in poverty. To address the second concern, we perform additional supplementary analyses, this time defining work according to the total annual hours worked by all working-age adults in the family (i.e., the product of total weeks worked and usual hours worked per year). Here, we distinguish between families with combined work of 0 hours, 1–1,749 hours, and 1,750+ hours (note that 1,750 = 1 full-time equivalent). The substantive conclusions that we draw below are robust to this alternative measure.

Table 1.

Descriptive Statistics, by Race

Variable	Non-Hispanic white	Non-Hispanic black	Hispanic
	Proportion/Mean (SD)	Proportion/Mean (SD)	Proportion/Mean (SD)
Poverty rate	0.099	0.255	0.228
Number of adult workers
No worker	0.107	0.197	0.102
One worker	0.427	0.504	0.452
Two+ workers	0.467	0.299	0.445
Family structure (ref. = married)
Married	0.555	0.302	0.514
Cohabitating	0.072	0.060	0.090
Single male-headed	0.176	0.206	0.159
Single female-headed	0.197	0.431	0.237
Combination of family structure and the number of adult workers
Married, no worker	0.032	0.027	0.025
Married, one worker	0.146	0.086	0.185
Married, two+ workers	0.378	0.190	0.304
Cohabitating, no worker	0.004	0.006	0.005
Cohabitating, one worker	0.015	0.015	0.026
Cohabitating, two+ workers	0.053	0.039	0.060
Single male, no worker	0.032	0.059	0.024
Single male, one worker	0.130	0.132	0.101
Single male, two+ workers	0.013	0.016	0.034
Single female, no worker	0.039	0.106	0.049
Single female, one worker	0.135	0.271	0.141
Single female, two+ workers	0.023	0.055	0.048
Educational attainment
Less than high school	0.050	0.112	0.294
High school diploma	0.452	0.558	0.464
Some college/Associate	0.117	0.108	0.087
College degree or more	0.380	0.222	0.156
Age of family head	44.770	42.994	40.347
	(12.224)	(12.350)	(11.917)
Adults aged 65+
No adult aged 65+	0.958	0.959	0.950
One+ adult aged 65+	0.042	0.041	0.050
Number of children
No child	0.633	0.590	0.451
One child	0.160	0.187	0.203
Two children	0.138	0.131	0.197
Three+ children	0.069	0.091	0.150
Number of working-age adults
One working-age adult	0.340	0.505	0.288
Two working-age adults	0.530	0.363	0.490
Three working-age adults	0.095	0.098	0.142
Four+ working-age adults	0.035	0.034	0.081
Nativity and citizenship
Native	0.955	0.885	0.530
Immigrant citizen	0.026	0.066	0.165
Immigrant non-citizen	0.019	0.049	0.305
Region
Northeast	0.186	0.159	0.134
Midwest	0.266	0.192	0.088
South	0.349	0.565	0.380
West	0.199	0.085	0.398
N (unweighted)	55,838	11,299	14,082

Note: Descriptive statistics calculated using weighted data. Ref. = reference. T-tests and χ² tests show that differences in means and distributions (reference = white) are statistically significant at the 0.01 level for all variables except for black-white differences in the presence of adults aged 65+.

We also include several factors that affect the likelihood of poverty and vary across racial groups: education, age, and family composition. Education is measured with a set of three binary variables indicating whether a respondent attained a high school diploma, some college or an associate degree, or a college degree or higher (reference group = less than high school diploma). Age is measured in years and modeled as a quadratic function to account for a commonly-observed non-linear relationship between age and poverty odds. To capture differences in family age composition, we consider three categorical variables that respectively account for the number of working-age adults (reference = 1), the number of children (reference = 0), and the presence of an adult aged 65+ years in the family (reference = no). We also account for immigration and naturalization status given racial differences in this factor and its strong correlation with the variables of interest. We distinguish between the native-born population (reference), naturalized citizens, and non-naturalized immigrants. Finally, we include a series of controls to account for survey year (reference =2011 [i.e., the 2012 CPS]) and residence in the four primary census regions (reference= Northeast).

Methods

Our analyses proceed as follows. First, we summarize differences in poverty, workforce participation, and family structure across the racial and ethnic groups of interest. We then estimate a set of logistic regression models to identify the association between family structure and work and the likelihood of poverty, conditional on the factors described above. Here, we proceed by first pooling data from all racial and ethnic groups of interest and estimating a model that allows the coefficients on all explanatory variables to vary by race. We then assess between-group differences using a Wald test of the joint significance of all coefficient estimates on the race variable and race-by-explanatory variable interaction terms (i.e., a Chow test). We find statistically significant differences, and therefore proceed by estimating models stratified by race.

We then extend our regression analyses by performing a decomposition of between-race differences in poverty risk, following an approach used in prior research (Phillips and Sweeney 2006, Sweeney and Philips 2004, Tzeng and Mare 1995). For our purposes, we estimate how differences in family structure and workforce participation across racial and ethnic groups contribute to the overall gap in the log odds of poverty between groups. Using this method, the difference in the expected log odds of poverty between any two groups is separated into a component due to differences in observed group characteristics and an unexplained component, which reflects the contribution of unobserved variables in addition to rate or coefficient effects. We only draw substantive conclusions about the estimated composition effects since the latter lacks a straightforward interpretation. The choice of a standard population is not determined a priori, so we decompose each pairwise difference in poverty odds twice, using each group as the standard (Phillips and Sweeney 2006). We present both results for each between-group comparison and interpret them as a range of estimated composition effects. As an additional robustness check, we perform comparable regression and decomposition analyses using a linear probability framework, and find that our conclusions do not change, qualitatively, with this alternative approach. Finally, we underline that our analyses focus squarely on the role of compositional differences explaining between-group inequalities in poverty: they do not speak to the potential poverty and inequality reducing effects of changes in composition beyond the level of the reference group in each comparison (e.g., marriage among white families).

Results

Descriptive Statistics

We begin by describing trends in work, family structure, poverty, and other key demographic characteristics across our sample (Table 1). Our estimates show considerable differences in the poverty rate and population composition across racial groups. The respective poverty rates among black and Hispanic householders were more than double than that of whites. Approximately one-tenth (9.9%) of white families were poor, while 25.5% of black and 22.8% of Hispanic families lived below the poverty line. Our main explanatory variables of interest are marriage and the number of workers in the family. More than half of white (55.5%) and Hispanic (51.4%) families were headed by married couples, compared with less than a third (30.2%) of black families. Cohabitation was most common among Hispanic (9.0%) and white (7.2%) families, and least common among black families (6.0%). A full 43.1% of the latter were headed by single women, nearly 20 percentage points more than the share of Hispanic families (23.7%) and more than twice the share of white families (19.7%). With respect to work, nearly 90% of white (89.3%) and Hispanic (89.8%) families had one or two earners, versus 80.3% among black families. And, among these working families, one-worker families were disproportionately represented among blacks (62.7%) relative to whites (47.8%) and Hispanics (50.4%).

We also show the distribution of observations across 12 combined work-family structure categories, which, as noted above, we used in a robustness check. The modal category among whites (37.8%) and Hispanics (30.4%) was a married-couple family with two earners, while single-earner, female-headed families were most common among the black population (27.1%). Notably, white families headed by married or cohabitating adults were more likely to have two or more earners (68.6%) than their black (63.1%) and Hispanic (60.2%) counterparts. White and Hispanic families headed by single adults were near-equally likely to have at least one earner (80.9% and 81.7% respectively), and more so than black families (74.2%). Finally, we note that a number of work-family structure combinations were, as expected, quite rare (e.g., less than 2% of the sample). However, the absolute number of observations in these cells remain sufficient for estimating the regression models in our robustness checks.

In addition to marriage and family structure, we also describe other relevant group characteristics included in our regression analyses. These variables include additional indicators of family structure and composition, such as the family heads’ age (modeled as a quadratic function), number of children, number of working-age adults, and a binary indicator of the presence of adults aged 65 years and older. We also control for region of residence and year. Our analyses show that the Hispanic population has distinctively younger householders, more children, more working-age adults, and more adults aged 65+ than the other groups. With respect to other controls, we underline differences in educational attainment across groups—with clear white advantage and Hispanic disadvantage: 29.4% of heads lacked a high school diploma, versus 11.2% of black and 5.0% of white householders. Hispanic families were also much less likely to be headed by a native-born citizen (53.0%) relative to black (88.5%) and white (95.5%) families. Finally, we note racial differences in geography, with black families overrepresented in the South, and the Hispanic population underrepresented in the Midwest.

It is notable that black and Hispanic families have considerable differences with respect to these compositional variables, but comparably high poverty rates. In contrast, our estimates show remarkably similar patterns of work and marriage among white and Hispanic families. Such results suggest that these factors are unlikely to explain much of the large Hispanic-white poverty gap. In contrast, large black-white differences in work and family structure are consistent with the much higher poverty rate among black families. There are two plausible explanations for these patterns. On the one hand, racial differences in other variables, such as educational attainment, may offset the similarities in family structure and work. On the other, the returns to these variables (i.e., the poverty-reducing effect of work) may vary by group. We explore these alternative possibilities in a decomposition analysis, which allows us to estimate the extent to which these individual factors explain group differences in poverty odds.

Social and Demographic Risk Factors

To quantify whether and how racial differences in the factors we observe are linked to racial inequalities in poverty, we conduct a series of regression analyses and regression-based decompositions. We first estimate a model using the pooled sample of all white, black, and Hispanic observations within our analytic sample. We estimate this initial model with a set of interaction terms between race and the explanatory variables in our model. Here, our main goal is to determine if there is sufficient overall variation in the associations between these explanatory variables and poverty odds to require the use of race-stratified models in our subsequent analyses. A secondary goal is to estimate the conditional associations between work and marriage and the likelihood of poverty, and assess the extent of racial differences in these patterns. Regarding the first objective, results of the Chow test indicated statistically significant overall racial differences in coefficient estimates (χ²=3156.7, p<.001). For the decomposition analysis below, we therefore proceed by estimating logistic regression models of the likelihood of poverty using race-stratified models.

To address our second goal, we first estimate the associations between different work and marriage arrangements and the log odds of poverty (i.e., the net coefficient estimates) for each racial group (Table 2). The respective reference groups for these two sets of net coefficients are same-race householders in families with no workers and same-race married householders. Overall, the results indicate a strong negative association between the number of workers in a family and poverty, as well as elevated odds of poverty among families headed by single and cohabitating adults vis-à-vis married-couple families. Using these estimates, we then test for between-race differences within each category of the work and marriage variables. In the two right-most columns of Table 2, we present the raw differences in net coefficient estimates among black and Hispanic families relative to whites, as well as tests of their statistical significance (Table 2). For example, the disadvantage in terms of the log odds of poverty among families headed by single-women versus married-couple families (reference group) was an estimated 0.193 larger among black than white families, but this difference was not statistically significant. Surprisingly, between-race differences in these coefficient estimates were generally not statistically different from zero, with the one exception being the difference in the coefficient for single male-headed families between the white and Hispanic populations. This gap is seemingly driven by the similarity in poverty odds being married-couple and single-male headed Hispanic families, which in turn reflects unique labor market and family circumstances among this group. This case aside, the similarity in coefficient estimates indicates that the results of the Chow test described above are driven by differences in other coefficients. Of course, the justification for estimating race-stratified models remains.

Table 2.

Net Coefficient Estimates for Work and Marriage, by Race

Variable	Net coefficienta		Between-group differenceb
	B	SE	B	SE
Number of adult workers + number of adult workers X race
One worker, Non-Hispanic White	−2.560***	0.059	–	–
One worker, Non-Hispanic Black	−2.596***	0.094	0.036	0.111
One worker, Hispanic	−2.699***	0.116	0.139	0.130
Two+ worker, Non-Hispanic White	−4.393***	0.094	–	–
Two+ worker, Non-Hispanic Black	−4.507***	0.157	0.114	0.183
Two+ worker, Hispanic	−4.496***	0.142	0.103	0.170
Family structure + family structure X race
Cohabitation, Non-Hispanic, White	0.522***	0.099	–	–
Cohabitation, Non-Hispanic, Black	0.508*	0.210	0.014	0.232
Cohabitation, Hispanic	0.388**	0.139	0.134	0.170
Single male-headed, Non-Hispanic White	0.530***	0.103	–	–
Single male-headed, Non-Hispanic Black	0.723***	0.134	−0.193	0.169
Single male-headed, Hispanic	0.194	0.124	0.336*	0.161
Single female-headed, Non-Hispanic White	0.977***	0.096	–	–
Single female-headed, Non-Hispanic Black	0.803***	0.115	0.175	0.150
Single female-headed, Hispanic	0.853***	0.108	0.124	0.144

^aThe reference groups for these variables are same-race householders with zero workers and same-race married householders, respectively.

^bDifference in work or family coefficient estimates for each racial group relative to non-Hispanic whites.

^*p<.05.

^**p<.01.

^***p<.001.

Before proceeding to the decomposition analysis, we also present predicted probabilities of poverty for each race-work and race-family combination, holding all other variables at their means (Figures 1 and 2, respectively). These figures help illustrate the combined differences in poverty across race, according to work and family structure, and race-work and race-marriage interactions. Disadvantages among families without workers (Figure 1), and those headed by single adults (Figure 2)—particularly women—are pronounced. For example, the expected probability of poverty among black householders in worker-less families was 0.72, but 0.15 among comparable householders in one-worker families and 0.02 for those in two-worker families. Likewise, the expected probability of poverty among married black householders was 0.04, compared with 0.22 among non-cohabitating single males and 0.27 among non-cohabitating single females. For black cohabitating householders, the expected probability of poverty was 0.09. Notably, the relative disadvantages among single female householders is largest among Hispanics, with the expected probability of poverty among single Hispanic women (0.32) more than twice that of single men (0.13). Single-female to single-male disparities among white and black householders were smaller—approximately 52% and 26%, respectively.

Figure 1.

Predicted Probabilities of Poverty with 95% CIs, by Race and Work

Figure 2.

Predicted Probabilities of Poverty with 95% CIs, by Race and Family Structure

While the pattern of variation in poverty odds along work and family structure lines is consistent across racial groups, racial disparities in poverty within each category are also apparent (e.g., between white and non-white families without workers), as are differences in the magnitude of the marginal benefits of work and marriage. For instance, the expected probability of poverty among white householders in one-worker families was 0.06, just 14.5% of the expected probability faced by white householders in families with no workers. The expected probability of poverty for black and Hispanic householders in one-worker families was 0.15 and 0.21, respectively. Both of these statistics were still more than 20% of the expected probability among comparable householders in worker-less families (21.4% among blacks and 26.8% among Hispanics). The reduction in the probability of poverty between non-married and married householders was similarly smaller for black and Hispanic householders than whites. These figures illustrate that non-white minorities not only face higher odds of poverty within each work and family structure category, but also benefit less from work and marriage.

Explaining Between-Group Differences in Poverty Risk

Building on our initial regression estimates, we next estimate race-stratified regression models (Table 3) that parallel the fully-interacted model discussed above. Using those results, we decompose racial differentials in the expected log odds of poverty into percentages explained by key compositional factors (Table 4). Here, recall that a positive percentage indicates that between-group differences in a given variable contributed to the observed differential in the log odds of poverty between groups. In the absence of those compositional differences, the poverty gap would have been smaller. In contrast, a negative percentage indicates that compositional differences between groups offset, or suppressed, the observed differences in expected odds of poverty. That is, the differential between groups would have been larger had there been no racial differences in that factor.

Table 3.

Logistic Regression Models Predicting Poverty, by Race

Variable	Non-Hispanic white		Non-Hispanic black		Hispanic
	B	SE	B	SE	B	SE
Number of adult workers (ref. = no)
One worker	−2.560***	0.059	−2.596***	0.094	−2.699***	0.116
Two+ workers	−4.393***	0.094	−4.507***	0.157	−4.496***	0.142
Family structure (ref. = married)
Cohabitating	0.522***	0.099	0.508*	0.210	0.388**	0.139
Single male-headed	0.530***	0.103	0.723***	0.134	0.194	0.124
Single female-headed	0.977***	0.096	0.803***	0.115	0.853***	0.108
Educational attainment (ref. = less than HS)
High school diploma	−0.833***	0.077	−0.686***	0.103	−0.666***	0.084
Some college/Associate	−1.298***	0.098	−1.229***	0.150	−1.264***	0.144
College degree or more	−1.763***	0.087	−1.981***	0.145	−1.540***	0.140
Age of family head	−0.094***	0.013	−0.053*	0.021	−0.064**	0.023
Age of family head2	0.001***	0.000	0.000	0.000	0.001	0.000
Adults aged 65+ (ref. = no)	−1.202***	0.150	−1.026***	0.221	−0.738**	0.236
Number of children (ref. = no child)
One child	0.414***	0.068	0.599***	0.106	0.434***	0.114
Two children	0.749***	0.072	1.164***	0.118	1.040***	0.109
Three+ children	1.456***	0.082	1.959***	0.120	1.617***	0.112
Number of working-age adults (ref. = one)
Two working-age adults	0.017	0.096	0.020	0.108	0.155	0.117
Three working-age adults	0.378**	0.125	0.687***	0.163	0.691***	0.153
Four+ working-age adults	0.931***	0.181	1.100***	0.247	0.296	0.174
Nativity and citizenship (ref. = native)
Immigrant citizen	0.571***	0.137	0.205	0.168	−0.012	0.120
Immigrant non-citizen	0.777***	0.149	0.558**	0.169	0.613***	0.086
Region (ref. = northeast)
Midwest	0.149*	0.071	0.411**	0.131	0.012	0.163
South	0.195**	0.068	0.218	0.114	0.179	0.110
West	0.155*	0.074	0.257	0.157	0.116	0.108
Year (ref. = 2011)
2012	0.024	0.050	−0.077	0.077	0.104	0.070
2013	0.047	0.052	−0.227**	0.082	0.115	0.081
Wald χ2	4426.72***		1713.91***		1671.00***
Pseudo R2	0.335		0.357		0.321
N (unweighted)	55,838		11,299		14,082

Note: Regression estimates are based on weighted data. Ref. = reference. HS = high school.

^*p<.05.

^**p<.01.

^***p<.001.

Table 4.

Decomposition of Change in the Log Odds of Poverty Explained by Racial Differences in Composition

Variable	White-Black differential (%)		White-Hispanic differential (%)		Black-Hispanic differential (%)
	1	2	3	4	5	6
	White as standard	Black as standard	White as standard	Hispanic as standard	Black as standard	Hispanic as standard
Number of adult workers	33.85	34.88	1.83	1.74	511.51	504.88
Family structure	15.03	12.84	2.67	2.55	170.21	158.75
Educational attainment	12.71	15.82	28.61	25.31	−216.37	−186.76
Age of family head	3.97	2.89	11.10	5.53	−70.12	−50.35
Adults aged 65+	0.09	0.08	−0.68	−0.42	9.58	6.90
Number of children	2.37	3.20	11.98	14.05	−194.34	−164.45
Number of working-age adults	−0.18	−0.18	4.00	2.66	−81.96	−62.07
Nativity and citizenship	2.93	1.58	20.29	11.68	−158.41	−151.41
Region	0.84	−0.80	0.70	1.79	2.41	−2.10
Year	0.04	−0.19	0.00	0.04	−3.39	0.53
Total % explained by differences in composition	71.64	70.11	80.49	64.93	−30.88	53.93

Black-White Poverty Differential

Columns 1 and 2 in Table 4 show the share of the black-white gap in the expected log odds of poverty explained by racial differences in the compositional factors listed in the table. We find that regardless of whether the white or black populations are used as the standard, the poverty gap would have decreased substantially in the absence of the observed compositional differences in work, family structure, family composition, and householder education. Specifically, if whites had the same compositional characteristics as blacks with respect to the factors in our analysis, the black-white poverty differential would be reduced by 71.6%. Conversely, if blacks had the same characteristics as whites, the poverty gap would be reduced by 70.1%. We take these two estimates as the range of total compositional effects. Black-white differences in the number of adult workers in the family explain the largest share of the gap in poverty risk between these groups. Regardless of which population was used as the standard, just over one-third (33.9-34.9%) of the difference in poverty between these groups can be accounted for by disparities in the distribution of families by number of workers. Differences in family structure (12.8–15.0%) and educational attainment (12.7–15.8%) explained substantial, and the second-largest, shares of the black-white poverty gap. In contrast, family composition—age of head and the numbers of children, working age adults, and adults aged 65+—explained a relatively small share of this gap (6.0–6.3% combined).

Hispanic-White Poverty Differential

Hispanic-white differences in compositional factors accounted for between 64.9% (Hispanic as the standard) and 80.5% (white as standard) of the poverty gap (Columns 3 and 4, Table 4). The magnitude of these compositional effects are comparable to those explaining the black-white poverty gap: the midpoint of the range in composition effects (72.7%) was only negligibly different than black-white composition effects. That said, we find that the particular compositional variables driving the differences in poverty between Hispanics and whites are substantially different from those that explain the black-white gap. Whereas disparities in work and marriage, combined, explained a majority of the white-black poverty differential, compositional differences in these two factors combined explained only 4.3–4.5% of the Hispanic-white gap in poverty log odds. This is a strikingly low figure given widespread and longstanding assumptions about the drivers of poverty in the U.S. In contrast, disparities in family heads’ educational attainment explained the largest share of the Hispanic-white poverty gap, estimated at 25.3–28.6%. Differences in nativity and citizenship status also accounted for a large share of Hispanic-white differences (11.7–20.3%). Of course, this variable may be partially capturing correlated factors such as English language proficiency, occupational attainment, and wages (e.g., suppressed wages among undocumented workers). Finally, we underline that differences in two dimensions of family composition—age of head and the average number of children per family—are also important, having explained a combined 19.6–23.1% of the Hispanic-white differences in the log odds of poverty.

Black-Hispanic Poverty Differential

Finally, we analyze how compositional differences between the Hispanic and black populations contribute to disparities in poverty among these two groups (Columns 5 and 6, Table 4). Here, we note that the difference in the log odds of poverty between black and Hispanic families (0.103 higher among blacks than Hispanics) was small in comparison to that between whites and blacks (1.592 higher among blacks) and whites and Hispanics (1.489 higher among Hispanics). The small Hispanic-black gap in poverty odds has the effect of inflating the estimated composition effects since they are reported as percentages of the observed gap, so readers should be careful when drawing comparisons between these results and the other decompositions. Nonetheless, our findings still reveal important differences in the relative contribution of different factors to the black-Hispanic gap, and suggest two key points. First, black families are considerably disadvantaged in terms of marriage and the number of workers vis-à-vis Hispanics, just as they were relative to white families. Holding all other factors constant, the Hispanic advantage in poverty log odds would have been between more than six and a half (663.6–681.7%) times greater than observed if combined differences in work and family structure remained in place. Such a scenario would translate to a gap in the log odds of poverty of approximately 0.692, which was more than a third of the very large gap observed between black and white families. The second key point that emerges from the black-Hispanic comparison is that low educational attainment, high fertility, and low rates of nativity and citizenship among Hispanics are key sources of disadvantage. This pattern holds true in reference to both the black and white populations.

Discussion and Conclusion

Work and marriage are commonly portrayed as panaceas to poverty, at least among some policymakers. Indeed, these were the primary ideas underlying the 1996 welfare reform legislation and continue to undergird the targeting of public benefits to workers and certain demographic groups (e.g., mothers) (Lichter and Jayakody 2002). However, if marriage and entry into the labor market are indeed universal pathways out of poverty in the U.S., then one would expect disparities in (a) the number of workers per family and (b) family heads’ marital status to translate into large shares of poverty gaps between racial groups. In contrast, if work and marriage have weak or less universal effects on poverty, then any compositional differences in these factors are likely to explain much less of observed racial inequalities. We addressed this issue here, building on a recent study by Baker (2015) and contributing to poverty research on the respective effects of marriage and work (e.g., Brady et al. 2010, Eggebeen and Lichter 1991, Iceland 2003, Iceland and Kim 2001, Lichter and Landale 1995, Lichter et al. 2005, Sawhill 1998, Thiede et al. 2015, Thomas and Sawhill 2005).

Our results suggest that racial disparities in work and marriage play an important role, but one limited to explaining gaps in poverty between black families and their white and Hispanic counterparts. Combined, lower rates of work and marriage—particularly disproportionate shares of families headed by single women, who also face gender-related disadvantages—among black families explain approximately half of that population’s disadvantage in poverty vis-à-vis whites. Moreover, the black-Hispanic gap in the log odds of poverty would have been more than six times smaller if rates of work and marriage among black families converged to levels observed among Hispanics. Such patterns of disadvantage among black families today are similar to those documented in previous decades (Eggebeen and Lichter 1991, Lichter et al. 2005), and consistent with assumptions about the proximate effects of work and marriage on poverty. Notably, in both black-white and black-Hispanic comparisons, differences in work were seemingly more important than in marriage. Racial disparities in the number of workers (33.9–34.9%) explained more than twice the share of the white-black poverty gap than did family structure (12.8–15.0%), with a threefold difference with respect to the black-Hispanic gap (504.9–511.5% versus 158.8–170.2%). The substantial contribution of differences in work to racial poverty disparities diverges somewhat with prior findings, which emphasized the role of family structure (Eggebeen and Lichter 1991, Lichter et al. 2005). The differences in results may be partially attributable to prior studies’ focus on child poverty in particular, but it may also reflect changes in the labor market over recent decades—including the residual effects of the Great Recession.

Work and marriage explain relatively little of the white-Hispanic gap in poverty, which reflects compositional similarities with respect to these two factors rather than their lack of importance as determinants of poverty (DeNavas-Walt and Proctor 2015, Oropesa et al. 1994). The structure and number of workers per Hispanic and white family are, on average, relatively similar. The substantial difference in the poverty rate between these two groups (12.9 percentage points) is therefore less about work or marriage per se than it is about human capital disparities and structural factors that determine why otherwise-comparable Hispanic and white families face very different poverty odds. This finding is consistent with prior work that suggests poverty among Hispanics is more affected by macroeconomic conditions than family structure (Iceland 2003). It also demonstrates that Hispanic economic disadvantage vis-à-vis whites does not fit the dominant narrative regarding poverty in the U.S. Hispanics families are largely as ‘intact’, and include as many workers as white families, but remain systematically more likely to face impoverished circumstances. The idea that working and marrying—‘playing by the rules’—will be rewarded with at least an above-poverty standard of living is proving false for a disproportionate share of Hispanics in the U.S. (Thiede et al. 2015). The substantial increase in the Hispanic share of the population over recent decades, and likely into the future (Johnson and Lichter 2008, 2010), makes this finding particularly relevant for policymakers.

Our results further indicate that a Hispanic disadvantage in educational attainment plays an important role, explaining between 25.3–28.6% of the observed poverty gap vis-à-vis whites. As well, disadvantages in terms of nativity and citizenship (11.7 –20.3%) and relatively high fertility, as indicated by the number of children per family (12.0–14.1%), also make substantial contributions to the white-Hispanic gap. The composition of the Hispanic population in terms of these factors is also a major source of disadvantage vis-à-vis blacks, and in fact offsets a substantial share of the Hispanic advantage in terms of family structure and number of adult workers. Notably, educational attainment, nativity and citizenship, and fertility are also associated with the multi-generational process of immigrant integration in the U.S. (Fry 2003, Waldinger and Perlmann 1998). To the extent that compositional differences between the Hispanic population and both whites and blacks diminish over immigrant generations, our results suggest declining Hispanic-white gaps and growing Hispanic advantage vis-à-vis poverty among the black population. Of course, this still raises questions about immigrants (and their children). Here, our findings have straightforward implication for policy: implement programs and policies that increase education and reduce the economic ‘penalty’ experienced by non-native workers. Our results suggest that such interventions have the potential to reduce Hispanic disadvantages vis-à-vis whites, even among the large share of Hispanics who are first or second generation immigrants.

It is important to note that our estimates only reflect the effects of compositional differences. Racial disparities in wages, hours worked, and other factors that determine the economic returns to work, marriage, and education are partitioned into the unexplained portion of our decomposition. As such, our estimates likely understate the significance of the labor market and human capital in explaining racial disparities. These findings underscore the potentially large impact of interventions that improve education and skills, which are arguably more likely to succeed than marriage promotion policies. Importantly, though, policies that increase skills and education are also likely to have the second-order effect of increasing employment and job quality, and marriage and family stability—with clear positive impacts in terms of families’ economic status (Lichter et al. 1992, Musick et al. 2012).

Of course our analyses do not capture these dynamic processes and interactions: the decomposition approach used here show only the static changes that would occur were compositional differences to decline. Another limitation of our approach is the focus on only a single point in time. While beyond the scope of this article, future research could expand upon both Baker (2015) and our findings by examining temporal changes in the compositional determinants of racial disparities in poverty. We suggest future research also build upon our findings regarding the sources of Hispanic disadvantage, which imply that compositional changes across immigrant generations may have important consequences for poverty dynamics (see also Turner and Thiede 2016). Relatedly, future research should also explore differences within the Hispanic population according to ethnicity and country of origin, since in some cases patterns of marriage, work, and related factors vary between sub-groups. In our sample of Hispanic householders, rates of marriage were notably lower, and rates of worker-less householders higher, among those of Puerto Rican origin relative to the other largest sub-groups. Heterogeneity within the Hispanic population merits future attention.

Despite these limitations, our analyses yield robust evidence that racial disparities in poverty have complex foundations that are not amenable to universal policy solutions. Reducing inequality in poverty rates—and reducing poverty more generally—will require attention to racial differences in poverty generating processes, which include but are not limited to questions of marriage and work. In an era of economic stagnation, growing inequality, and increasing racial diversity, attention to these complexities is increasingly needed to shape effective social policy regimes.