America's Working Poor

Abstract

This article addresses measurement challenges that have stymied contemporary research on the working poor. The authors review previously used measurement schemes and discuss conceptual assumptions that underlie each. Using 2013 March Current Population Survey data, the authors estimate national- and race-specific rates of working poverty using more than 125 measures. The authors then evaluate the association between each measure and a latent construct of working poverty using factor analysis and develop a working poverty index derived from these results. Finally, the authors estimate multivariate regression models to identify key social and demographic risk factors for poverty among workers. The authors’ national estimates of working poverty range from 2% to nearly 19% and are highly sensitive to alternative assumptions. The authors’ analyses find that the latent construct is most highly correlated with empirical measures of working poverty that include part-time or part-year employment and that use poverty income thresholds that include both the poor and near poor. Crude rates and conditional risks of poverty among workers vary considerably among racial groups. This article provides a conceptual and empirical baseline for decisions about how best to estimate the magnitude and composition of America's working poor population.

Keywords

working poor poverty labor markets job quality

The Great Recession has refocused the nation's attention on long-term unemployment and chronic poverty (Jenkins, Brandolini, Micklewright, & Nolan, 2013; Mishel, Bivens, Gould, & Shierholz, 2012). Recently released poverty estimates from the United States (U.S.) Census Bureau indicate that 46.5 million people—15% of America's population—were poor in 2012 (DeNavas-Walt, Proctor, & Smith, 2013). Continuing year-to-year declines in median household income have also coincided with the nation's economic downturn during the Great Recession and a slow recovery. In 2012, real median household income was 8.3% lower than in 2007. Income declines are due in part to high unemployment rates, but the downturn in earnings among full-time workers employed year-round—both men and women—has also played a role (DeNavas-Walt et al., 2013). Indeed, America's workers have suffered from declining earnings and income since the mid-2000s. Working poverty is on the upswing.

This article places the spotlight on America's working poor, which has replaced the welfare poor as a focal point for policy intervention following the implementation of work-based welfare reform in the late 1990s (Lambert, 2009; Lichter & Crowley, 2002). Ours is an important task. Poverty is often explained as a matter of insufficient work effort or a lack of employment opportunities. Longstanding narratives about the American Dream suggest that hard work is the primary determinant of upward social mobility. Public policies, such as postreform welfare (e.g., Temporary Assistance for Needy Families [TANF]), the expansion of the Earned Income Tax Credit (EITC), and food stamp work requirements, are often based on the assumption that poverty can be reduced by incentivizing individuals to work, or to work more. Most poverty research reinforces this notion by focusing on (un)employment and the jobless poor (Brady, Baker, & Finnigan, 2013). To be sure, America's jobless face a remarkably high risk of poverty, but the so-called working poor has historically constituted a larger subset of the U.S. population—one that has seemingly grown during the postwelfare reform era (Brady, Fullerton, & Cross, 2010; Newman, 1999; Slack, 2010). In fact, recent estimates suggest that a majority of poor Americans live in families or households with at least one worker (Blank, Danzinger, & Schoeni, 2006; Brady et al., 2010).

Yet, the peer-reviewed literature on working poverty has been small and often inchoate.¹ In this article, we highlight a number of serious conceptual and measurement challenges, which arguably account for why the existing social science literature on working poverty remains underdeveloped. Previous literature has utilized multiple schemes for classifying workers and for defining poverty. Such differences affect estimates of the size and composition of the working poor and make comparisons across studies problematic. Moreover, the multiplicity of current measurement schemes reflects a general underconceptualization of the working poor as a sociologically distinct population that merits attention. Current definitions often pivot on assumptions that are discussed in insufficient detail to fully appreciate the extent or etiology of the problem. In short, methodological shortcomings—not a lack of relevance to contemporary social problems and public policy—have plagued research on America's working poor.

This article addresses these limitations by advancing five specific objectives. First, we identify the main assumptions that underlie choices about how to conceptualize and measure the working poor. Second, we develop a set of measures that represents the range of possible definitions of working poverty and use these measures—over 125 in total—to produce and compare a set of different national- and race-specific estimates of working poverty. Third, using a factor analytic approach, we estimate the degree to which each measure is associated with a latent construct of working poverty. Fourth, we construct a working poverty index and estimate the association between this indicator and a set of conventional social and demographic predictors (e.g., race, education) using multivariate regression. As a fifth and final empirical analysis of alternative measurements, we compare these results with those obtained using the most and least restrictive definitions of the working poor, as well as two measures we argue are substantively most appropriate. Overall, this article provides a conceptual and empirical baseline for evaluating how best to estimate the magnitude and composition of America's working poor population.

Measuring Working Poverty: Criteria and Assumptions

Conceptually, the working poor include those whose earnings from formal employment are insufficient to avoid poverty (Bureau of Labor Statistics, 2013; Slack, 2010). The working poor presumably play by the rules, and the normative expectation is that work should—if only for moral reasons—be rewarded with an above-poverty standard of living. This sentiment clearly underlies the establishment of minimum wage legislation, the EITC, so-called living wage ordinances, and other work supports. Indeed, the working poor are sometimes included among the deserving poor—that is, they are deserving of government help. They are contrasted with the undeserving poor, who presumably are culpable for their own disadvantaged circumstances (Katz, 2011). Current political debates effectively highlight this distinction, including arguments concerning increases in the federal minimum wage, the extension of unemployment benefits for the long-time unemployed, and restrictions on food stamp receipt among able-bodied working-age individuals.

Working poverty is a seemingly straightforward concept, but its measurement and interpretation are complicated by a number of questions that do not lend themselves to straightforward answers (Crettaz, 2013; Joassart-Marcelli, 2005). Specifically, there is little agreement about four key choices involved in measuring working poverty: (a) defining the universe of persons from which workers are identified; (b) determining the number of hours or weeks of work that are required to be classified as a worker; (c) identifying which kinds of income or in-kind benefits should be used to determine poverty status and, relatedly, choosing a dollar threshold that separates the poor and nonpoor; and (d) selecting an accounting unit for measuring income and poverty (e.g., person, family, or household). To be sure, no single set of assumptions is ever fully justified or entirely foolproof. Instead, our goal is to outline these assumptions, discuss their measurement and interpretative implications, provide a clearly defined set of estimates of the working poor, and evaluate the association of each uniquely specified measure of working poverty with an unobserved latent construct.

Defining the Universe of Potential Workers

The first step of measuring working poverty is to define the universe of persons eligible to be classified as workers (potential workers). Any definition should presumably capture persons who are expected to work and support themselves. Some populations are clearly not expected to provide their own income, even if they meet other criteria that define a worker. For example, from a public policy standpoint, employed school age teenagers are generally not classified as workers or included among the working poor. Their economic welfare is unambiguously the responsibility of their parents or guardians. Whether other populations should be eligible to be counted among the working poor is less straightforward. For example, should poor workers at retirement age be considered among the working poor? For them, earnings from work often supplement pension income or social security; or work may be motivated less by the need for economic self-sufficiency and more by personal fulfillment or the desire to stay active. Additionally, college students sometimes work at low-wage jobs that supplement the financial support that they receive from their families. Few people would define working students among the working poor. Yet, what if college students work because they receive little or no support from parents? How they should be classified is unclear.

Previous studies of working poverty typically restrict the population of potential workers to working-age adults, but there are differences in the definition of working age and the type of adults considered. A number of studies consider all adults aged 18 to 62 or 64 (Jensen, Findeis, Hsu, & Schachter, 1999; Jensen & Slack, 2003; Slack & Jensen, 2002), while others eliminate college-age individuals to avoid classifying employed students as working poor (DeJong & Madamba, 2001; Slack, 2010). By excluding college-age persons, however, one also excludes many full-time workers with a high school education or less, who are often in low-wage occupations. No known studies include persons of retirement age.

Some studies also identify the population of potential workers on the basis of their relationship to other family or household members. For example, Lichter, Johnston, and McLaughlin (1994) and Slack (2010) consider household and family heads, respectively. Hauan, Landale, and Leicht (2000) restrict their universe of potential workers to working-age male heads of household or their spouses (i.e., one person per household) living in urban labor markets. All three of these studies measure poverty on the basis of total family or household income. In doing so, they essentially construct family- or household-level measures that indicate whether families or households with working heads are poor or nonpoor. These estimates do not include poor households with multiple part-time workers among the working poor, even though many family members participate in the formal labor market and make below-poverty wages.

In contrast, Iceland and Kim (2001) and DeFina (2007) consider the combined work and earnings of all family members—seemingly regardless of age or householder status. The emphasis here is on whether the total hours worked by all family members is enough for that family to be considered working and, if so, whether the combined earnings are sufficient to lift the family above the poverty line. Such family-based measures equate the work of youth and the elderly with that of working-age adults, and as such implicitly assign responsibility for economic self-sufficiency to all members. For example, this measurement approach treats a family with a single full-time, year-round adult worker the same as a family with multiple part-time workers, including school-age youth—two fundamentally different positions in the labor market. Treating all workers as equivalent is at odds with normative and legal expectations about work and economic responsibility, which emphasize the role of working-age adults.

Defining Work

Once the universe of potential workers is defined, the next step is to distinguish between those who do and do not work. This distinction has clear implications for estimating the size of the working poor population: Working persons constitute the base population from which the working poor are identified.² One's definition of work reflects assumptions about the amount of work that should be required to avoid poverty, as well as meet other normative expectations about work effort (Correll, Kelly, O'Connor, & Williams, 2014; Kmec, O'Connor, & Schieman, 2014). Most studies of the working poor separate workers from nonworkers using thresholds defined by the number of hours or weeks worked during the previous year. Not surprisingly, previous research has used very different thresholds. Several articles, for example, have defined workers as individuals working 27 or more weeks during the past year (e.g., Caputo, 2007; Hauan et al., 2000; Lichter et al., 1994; Slack, 2010). Alternatively, DeJong and Madamba (2001) define workers as those who usually worked 35 or more hours per week during the previous year. Conceptually, the drawback to measures that consider only hours or weeks worked is that they may capture persons who work a significant number of weeks (hours) but only a few hours per week (weeks per year).

Iceland and Kim (2001) and DeFina (2007) account for the number of both hours and weeks worked during the previous year. They define working families as those whose members worked a combined 1,750 hours or more the previous year. This is equivalent to a single individual working full-time (35+ hours) and year-round (50+ weeks) during the previous year.³ In contrast, other studies set neither hour nor week thresholds, considering all employed persons as workers regardless of the number of hours or weeks worked (e.g., Brady et al., 2010; Hauan et al., 2000) or, alternatively, as those with any labor market earnings (e.g., Slack & Jensen, 2008). This is a low and debatable threshold for defining workers. Yet, research on working poverty from the Bureau of Labor Statistics (BLS) utilizes an even more inclusive definition of workers, including all individuals in the labor force whether they are actually working or not (Bureau of Labor Statistics, 2013). In this case, simply looking for work is sufficient to be defined as a worker. Although the efforts of active job seekers arguably distinguish them from discouraged workers, the usual conceptual distinctions between the working poor and the jobless poor are largely lost or rendered ambiguous by this definition.

Measuring Income and Poverty

Identifying poor workers clearly involves making decisions about how to measure income and poverty. The measurement of poverty has a longstanding and contentious history—one that has produced few straightforward answers or an easy consensus (Citro & Michael, 1995). Indeed, our task raises the usual questions about the most appropriate method for separating the poor from the nonpoor. For example, the assumptions used to set official U.S. poverty income thresholds—those based on absolute family income—are often difficult to justify theoretically or empirically (e.g., the need to account for taxes, transfers, geographic differentials in the cost of living; Iceland, 2005; Lichter, 2005). We will not attempt to adjudicate between alternative measures here, but we do examine whether estimates of the working poor are sensitive to the selection of different poverty thresholds.

Additionally, if part-time or part-year workers are included among the working population, then one must decide whether to adjust observed earnings to a full-time and full-year equivalent. This choice reflects assumptions about whether the distinctive feature of working poverty is subpoverty wages or subpoverty levels of annual income. The decision may also rest on assumptions about the amount of work required to be considered a full-time worker. For example, if one considers an individual working 9 months of the year to be fully employed—and therefore presumably deserving of a standard of living above the poverty line—then poverty status should arguably be based on reported annual earnings rather than on a full-time, full-year equivalent. Poverty-level earnings from a full-time job, so defined, represent a labor market failure worthy of policy attention. The decision to adjust earnings may also depend on assumptions about the usual reasons for part-time or part-year work. On the one hand, adjustments may artificially inflate the earnings of workers without access to full-time or year-round employment (i.e., involuntary part-time employed). On the other, annualizing earnings might appropriately correct for the below-maximum earnings of voluntary part-time workers, some of whom could escape poverty by choosing to work full-time.

Previous literature provides few straightforward guidelines for addressing these conceptual issues. For example, a number of studies adjust individual-level earnings to their full-time year-round equivalent, and in some cases identify the working poor using a threshold of 125% of the official U.S. poverty line for individuals (e.g., Jensen et al., 1999; Jensen & Slack, 2003; Slack & Jensen, 2002). This approach identifies workers with earnings near or below the poverty line rather than workers living in poor families. DeJong and Madamba (2001) also use 125% of the official U.S. poverty line as the poverty threshold but consider individuals' average weekly earnings. The use of earnings rather than income places the emphasis squarely on the quality of the job rather than on whether persons have access to sufficient income or income support (e.g., government cash assistance, the incomes of other family members).

Other studies measure income and poverty using the official poverty definition, which does not adjust earnings (Lichter et al., 1994; Slack, 2010). Workers are considered poor if they live in poor families, as officially defined by the U.S. Census Bureau. Alternatively, Iceland and Kim (2001) and DeFina (2007) define poverty using experimental income poverty thresholds based on recommendations by the National Academy of Sciences Panel on Poverty and Family Assistance (Citro & Michael, 1995). These thresholds are based on both money and in-kind income and account for other factors including government transfers, access to Medicare or Medicaid, and empirically based economies of scale in different sized households.

Other recent studies of the working poor, such as Brady et al. (2010, 2013), use a relative poverty line benchmarked at one-half of the U.S. median household income. Relative definitions of poverty are commonly used in the European Union (EU) and are consonant with theoretical concerns about social exclusion (Lohmann, 2009). Here, workers are considered poor if they cannot fully participate in society: Their incomes are insufficient to enjoy the material advantages or living standards that are commonplace in society. Such an approach is rarely used in U.S. poverty research but may have increasing relevance as income poverty thresholds have dropped well below median family income over recent decades.

Unit of Analysis

Finally, one must choose an appropriate unit of analysis. Decisions about the unit for which work, income, and poverty are measured reflect assumptions about whether income generation, consumption, and poverty are best viewed as individual- or family-level phenomena. For example, strictly individual-level measures may not account for the number of dependents or for resource pooling among family members. Choices about the unit of analysis may also reflect an assumption about which household members are responsible for the economic well-being of the family. Such choices determine, for example, whether families should be considered poor if the earnings of teenagers or retirement-age family members are all that keep them out of poverty.

Previous studies that considered all working-age adults (i.e., not limited to the household or family head) typically use individual-level income and poverty measures (e.g., DeJong & Madamba, 2001). However, a number of these studies utilize family- or household-level income and poverty measures (e.g., Brady et al., 2010; Caputo, 2007). In these cases, work status is defined at the individual level, but poverty is a function of total family (household) income and family (household) size. Studies that consider only family or household heads as potential workers also utilize family- or household-level income and poverty measures (Lichter et al., 1994; Slack, 2010). These studies also link individual-level work to family-level income and poverty but focus on the work of the family head, who is often the primary earner. To our knowledge, no study that considers the work of family heads has measured income and poverty at the individual level. Finally, as previously mentioned, Iceland and Kim (2001) and DeFina (2007) take a relatively unique family-level approach that aggregates the work and earnings of all family members.

Current Study

Research on the working poor requires a number of necessary but contested measurement decisions that defy easy consensus. Previous studies have employed many different approaches but often lack serious discussion of the underlying conceptual assumptions that may affect estimates of the size and composition of the working poor and interpretations of those estimates. We provide a conceptual and empirical baseline for measurement decisions in research on the working poor, and therefore contribute to a more coherent literature on this topic. We examine how a comprehensive set of alternative definitions of working poverty affects estimates of the share of U.S. workers experiencing poverty, as well as how these estimates are revealed unevenly across racial and ethnic groups. We then use factor analysis to assess the association (i.e., factor loadings) between alternative measures and an unobserved latent construct of working poverty. We use these factor loadings to construct a composite working poverty index that serves as an outcome variable in multivariate models that identify social and demographic risk factors of poverty among U.S. workers. As a final assessment of how alternative measurements affect conclusions about working poverty, we compare these results with others obtained using the most and least restrictive definitions of the working poor, as well as two measures we argue are substantively most appropriate. Our analyses demonstrate that estimates of working poverty are sensitive to underlying assumptions and provide important insights about the implications of alternative measurement frameworks.

Data, Measurement, and Methods

Current Population Survey

Our analysis draws upon the public use microdata sample from the March Supplement of the 2013 Current Population Survey (CPS; King et al., 2010). The March CPS provides perhaps the most detailed information available on employment, earnings and income, and poverty in the past year (2012). The CPS is a nationally representative household survey of approximately 60,000 households and, importantly, identifies the 1-year work history and sources of income (including earnings) of all household members. It is the most common source of data used in previous studies of working poverty in the U.S.⁴

Measurement

Below, we examine the six key parameters that underlie study-to-study variation in national estimates of the working poor:

Potential workers. Persons included in the universe of potential workers include (a) all individuals aged 18 to 64, (b) heads of primary families,⁵ and (c) families (sum of all working family members).

Weeks worked. We consider three alternative classifications of workers on the basis of weeks worked in 2012: those working (a) year-round (50 + weeks per year), (b) at least half of the year (27 + weeks per year), and (c) at least 1 week per year. This threshold is applied at the individual and family levels.

Hours worked. We consider the number of hours usually worked per week during 2012. We distinguish between persons and family members who worked (a) full-time (35 + hours per week), (b) part-time or more (17 + hours per week), and (c) any hours (1 + hours per week).

Income. We measure the working poor based on (a) unadjusted annual income and (b) income adjusted to a full-time, full-year equivalent.⁶ Following previous research, all family-level poverty measures are based upon unadjusted income.

Poverty threshold. We alternately distinguish the poor from the nonpoor based on (a) the official poverty thresholds, (b) 125% of the official poverty thresholds, (c) 50% of the median U.S. family or individual income,⁷ and (d) the Census Bureau's supplemental poverty measure (SPM; Short, 2013).

Poverty unit. We consider income and poverty at both the individual and family levels. The former evaluates whether work provides sufficient income from earnings to lift a single person out of poverty. The latter measures when work lifts the family out of poverty. We use the supplementary resource unit (SR) when using the SPM, which broadens the definition of the family vis-à-vis the official measure.

Based on these six criteria, we calculate 126 different estimates of the percentage of workers who are poor. These analyses are extensive, but not exhaustive. By necessity, much of the following discussion is limited to comparisons with significant measurement or substantive implications. A complete set of descriptive results is included in the Appendix.

Analytic Strategy

Our analyses proceed as follows. First, we use a set of alternative measurement criteria to estimate the percentage of U.S. workers or working families that experienced poverty in 2012.⁸ We consider the impact of alternative criteria on estimates of poverty among all workers, as well their implications for rates of poverty among workers in specific racial and ethnic groups. We then use factor analysis to estimate the association between alternative measures of working poverty and an unobserved latent construct.⁹ Factor analysis is useful for the purpose of data reduction, especially for summarizing substantively similar variables that are likely to be correlated with an unobserved or poorly measured variable (Fryer, Heaton, Levitt, & Murphy, 2013; Loehlin, 1987). In our analysis, factor loadings identify the measures of working poverty that are most highly correlated with the unobserved latent construct, and as such are most likely to correctly identify whether a person or family is in this latent state of working poverty.

Formally, we estimate the following equation:

Y_{ik} = β_{k} wpoo r_{i} + ɛ_{ik} .

where Y represents observed working poverty status, wpoor represents the (unobserved) true working poverty status, subscript i denotes a particular unit of observation (e.g., working-age individual, family), and subscript k represents particular measures of working poverty. We assume that a given unit i's observed working poverty status (Y) according to poverty measure k is a function of that unit's true working poverty status (wpoor). We further assume that this function takes a linear form with a slope of β_k, the factor loading with respect to the first factor. We use a simple one-factor model because this first factor explains a substantial proportion of observed variance, and additional factors provide relatively little marginal explanatory power.¹⁰ We estimate separate equations for individuals, family heads, and families because these represent fundamentally different analytic samples that are not comparable.

Using these estimates, we then construct a working poverty index. The index score (wpind) for each i unit is the weighted average of the observed working poverty status Y for all k measures of working poverty, where the weight (w) is given by the standardized coefficient for each k observed measure of working poverty (Fryer et al., 2013; StataCorp, 2013). This composite measure takes the form:

wpin d_{i} = \sum_{k = 1}^{p} w_{k} Y_{ik} .

Finally, we estimate a series of regression models of working poverty to compare results based upon multiple outcomes, including the index score (wpind), the most and least restrictive measures (as defined by national estimates of working poverty), and two measures we argue are substantively most appropriate. These models predict the poverty status of individual workers, working family heads, and working families (Y)—as indicated by each of the outcomes—using a set of social and demographic factors that previous research has shown to be associated with working poverty. When using the index (wpind) as an outcome variable, the linear equation we estimate takes the form of:

Y = β_{0} + β_{1} x_{1} + \dots β_{k} x_{k} + ɛ .

Substantively, these models estimate observations' proximity to the latent state of working poverty. Positive coefficient estimates indicate a positive association between a given variable and the likelihood an observation falls into the latent state of working poverty.¹¹ The other outcomes we model are binary, indicating the poverty status among workers, so defined. For these outcomes, we estimate logistic regression models, which take the form:

\ln (\frac{P (Y = 1)}{1 - P (Y = 1)}) = β_{0} + β_{1} x_{1} + \dots β_{k} x_{k} + ɛ .

These models predict the odds that a worker is poor, according to each unique definition of workers and poverty.¹² In both forms, explanatory variables (x) in the models include workers' race/ethnicity,¹³ sex, age in years, age squared,¹⁴ education,¹⁵ marital status,¹⁶ and the metropolitan status¹⁷ and region¹⁸ of residence. We apply the characteristics of the family head to family-level measures and estimate robust standard errors that account for the CPS stratified sampling strategy.

Results

Estimating Working Poverty Rates Under Alternative Assumptions

Defining workers: Week and hour thresholds

In this section, we demonstrate how changes in the parameters used to identify the working poor affect national- and race-specific estimates of the rate of poverty among workers in 2012. We begin by comparing results using different approaches to distinguish workers from nonworkers. The estimates in Table 1 demonstrate the implications of using different week-based thresholds to identify workers.

Table 1.

Estimated Poverty Rate (per 100) Among Workers Using Alternative Weeks-Worked Thresholds.

Worker unit	Individual			Individual			Individual			Family head			Family
Hours	N/A			N/A			N/A			N/A			N/A
Poverty unit	Individual			Individual			Family			Family			Family
Income	Unadjusted			Adjusted			Unadjusted			Unadjusted			Unadjusted
Poverty threshold	Official			Official			Official			Official			Official
Weeks	50+	27+	1+	50+	27+	1+	50+	27+	1+	50+	27+	1+	50+	27+	1+
All	6.2	8.4	14.0	6.0	6.6	7.4	4.3	5.3	7.3	4.8	6.1	8.4	5.1	6.3	8.4
Non-Hispanic White	5.3	7.3	12.4	4.7	5.3	6.1	2.6	3.4	4.9	2.9	3.7	5.4	2.9	3.7	5.3
Non-Hispanic Black	6.9	9.7	17.2	7.4	8.1	9.2	6.9	8.8	12.8	8.1	10.8	15.2	8.1	10.7	14.7
Hispanic	9.5	12.3	18.6	10.7	11.2	11.7	10.0	11.5	13.9	12.0	13.7	16.4	13.3	15.3	17.6
Asian and Pacific Islander	5.7	7.5	12.2	5.6	6.0	6.4	3.4	4.1	5.3	3.6	4.7	6.1	4.6	5.4	7.0

Note. The data are from the March 2013 Current Population Survey.

As expected, the working poverty rate increases when the work threshold is lowered from a full- to half-year equivalent and further still when the threshold is dropped to a single week of work. Lowering the weeks-worked threshold expands the population of workers, a disproportionate share of who are poor or make subpoverty wages. For example, when considering all individuals aged 18 to 64 and measuring poverty using individual earnings, the working poverty rate changes from 6.2% to 8.4% when the week-based threshold is lowered from a full- to half-year equivalent. The working poverty rate is 14.0% when a single week threshold is used. Lowering the weeks-worked threshold increases estimated rates of working poverty using other measures, including those measuring income and poverty at the family level. Significantly, lowering the weeks-worked threshold also results in higher estimates of working poverty when earnings are adjusted to their full-time, full-year equivalent. Using adjusted income, the share of workers who are poor increases from 6.0% to 6.6% and 7.4% when lowering the weeks-worked threshold from 50 weeks to 27 weeks and 1 week, respectively. This suggests that low earners are overrepresented among persons who worked only part of the year.

We also consider race-specific rates. Compared with non-Hispanic White and Asian workers, estimates of poverty among Black and Hispanic workers are generally more sensitive to changes in the number of weeks worked used to distinguish workers from nonworkers. For example, when this threshold is lowered from 50 weeks to 1 week, the rates of poverty among non-Hispanic White and Asian workers aged 18 to 64 (all individuals, unadjusted income) living in poor families increase by 2.3 and 1.9 percentage points, respectively; however, they increase by 5.7 and 3.9 percentage points among non-Hispanic Blacks and Hispanics, respectively. This suggests that relatively high proportions of workers in the latter two groups are both poor and employed less than year-round.

Across all 126 measures of working poverty that we consider, we find that, net of variation associated with the other criteria for measuring working poverty, estimated rates of poverty among all workers were approximately 1.5 and 4.1 percentage points higher when the weeks-worked threshold was dropped from 50 weeks to 27 weeks and 1 week, respectively (analysis not shown¹⁹). Similar patterns hold true for the race-specific estimates, although estimated rates of poverty among non-Hispanic Black and Hispanic workers are systematically more sensitive to changes in the weeks-worked threshold than other groups.

We next turn to estimates based upon measures that define workers according to the number of hours usually worked per week in the previous year (Table 2). Lowering the hours-worked threshold to be considered a worker, not surprisingly, has the effect of increasing estimates of working poverty. Here, the rate of poverty (using an individual-level measure) among individual workers is 6.3%, 11.3%, and 14.0% when using 35-, 17-, and 1-hour thresholds, respectively. As before, estimates are much lower if we adjust income to an annual equivalent, use the family as the poverty unit, or measure hours-worked from a family rather than individual perspective. In fact, adjusting the hours-worked threshold causes virtually no change in the estimated rate of working poverty when work, income, and poverty are measured at the family level. Across all 126 measures of working poverty that we consider, we find that, net of variation associated with the other criteria for measuring working poverty, estimated rates of poverty increase an average of 2.3 and 3.2 percentage points when lowering the hours-worked threshold from 35 hours to 17 hours and 1 hour, respectively (analysis not shown).

Table 2.

Estimated Poverty Rate (per 100) Among Workers Using Alternative Hours-Worked Thresholds.

Worker unit	Individual			Individual			Individual			Family head			Family
Weeks	N/A			N/A			N/A			N/A			N/A
Poverty unit	Individual			Individual			Family			Family			Family
Income	Unadjusted			Adjusted			Unadjusted			Unadjusted			Unadjusted
Poverty threshold	Official			Official			Official			Official			Official
Hours	35+	17+	1+	35+	17+	1+	35+	17+	1+	35+	17+	1+	35+	17+	1+
All	6.3	11.3	14.0	5.9	7.2	7.4	4.8	6.7	7.3	5.3	7.6	8.4	8.3	8.3	8.4
Non-Hispanic White	5.1	9.6	12.4	4.6	5.8	6.1	2.9	4.4	4.9	3.1	4.7	5.4	5.2	5.2	5.3
Non-Hispanic Black	8.9	15.1	17.2	7.8	9.0	9.2	8.4	11.8	12.8	9.9	14.2	15.2	14.6	14.6	14.7
Hispanic	9.9	16.0	18.6	10.3	11.5	11.7	10.6	13.1	13.9	12.2	15.5	16.4	17.5	17.6	17.6
Asian and Pacific Islander	5.2	9.7	12.2	5.2	6.2	6.4	3.3	4.6	5.3	3.6	5.1	6.1	6.9	6.9	7.0

Note. The data are from the March 2013 Current Population Survey.

With respect to between-race differences, estimates of poverty among non-Hispanic White and Asian workers are generally less sensitive to changes in the hours-worked threshold. For example, the share of non-Hispanic White and Asian workers (all working-age persons) living in poor families increases by 2.0 percentage points when dropping this threshold from 35 hours to 1 hour. For Hispanic and non-Hispanic Black workers, however, dropping the hours-worked threshold increases the estimated rate of family poverty by 3.3 and 4.4 percentage points, respectively. A relatively high percentage of workers in these two groups are both poor and employed less than year-round.

With few exceptions, estimated rates of working poverty are higher when using hour-based thresholds than week-based thresholds. For example, the share of working family heads living in a family below the poverty line is 4.8% when full-time work is defined in terms of weeks worked (i.e., 50 + weeks); but it is 5.3% when full-time work is defined in terms of hours usually worked per week (35 + hours). This may reflect a high wage penalty for part-time work compared with part-year work, or alternatively, part-time workers may also be more likely to work only part of the year. To examine this further, we compare similar hour- and week-based thresholds using adjusted individual income. This shows poverty (defined at the individual level) among all full-time workers is slightly lower using hour-based thresholds (5.9%) than week-based thresholds (6.0%) when income is adjusted to a full-time, year-round equivalent. This result suggests that the proportion of part-time workers who also work only part of the year is comparable with the proportion of part-year workers who usually work part-time hours.

As a coda to these analyses, we also calculated the share of individuals, family heads, and families that were classified as nonworkers according to each of these definitions (results not shown). Under the least restrictive work thresholds—1 hour or 1 week of work—24.7% of working-age individuals were classified as nonworkers, as were 20.8% of family heads, and 24.4% of families. In contrast, under the most restrictive work threshold—both 50 weeks and 35 hours—49.0% of working-age individuals, 42.7% of family heads, and 34.4% of families were classified as nonworkers. Differences in the universe of workers from which the working poor are identified vary greatly across measures and can influence estimates of the size of the working poor population.

Universe of potential workers

We also consider how restrictions on the universe of potential workers affect estimates of working poverty. Table 3 shows two series of estimates, each of which considers three different universes of potential workers: (a) all individuals aged 18 to 64, (b) all heads of primary families, and (c) all families (i.e., combined work of families).

Table 3.

Estimated Poverty Rate (per 100) Among Workers Using Alternative Units to Measure Work.

Weeks	50+			27+
Hours	N/A			N/A
Poverty unit	Family			Family
Income	Unadjusted			Unadjusted
Poverty threshold	Official			Official
Worker unit	Ind.	F. Head	Fam.	Ind.	F. Head	Fam.
All	4.3	4.8	5.1	5.3	6.1	6.3
Non-Hispanic White	2.6	2.9	2.9	3.4	3.7	3.7
Non-Hispanic Black	6.9	8.1	8.1	8.8	10.8	10.7
Hispanic	10.0	12.0	13.3	11.5	13.7	15.3
Asian and Pacific Islander	3.4	3.6	4.6	4.1	4.7	5.4

Note. The data are from the March 2013 Current Population Survey. Ind. = individual; F. Head = family head; Fam. = family.

The percentage of all workers living in poor families is lower than the percentage of working family heads living in poor families. For instance, when using a 50-week threshold to distinguish workers from nonworkers, 4.3% of working individuals lived in poor families, 0.5 percentage points less than the share of working family heads living in poor families (4.8%). If family heads are the primary earners, then many nonfamily head working-age individuals are likely in a position of supplementing family income. Under this assumption, nonfamily heads who meet the criteria for work are seemingly more likely to live in families with multiple workers (e.g., in which both they and the family head are working), and therefore face lower risk of poverty. In contrast, the earnings of working family heads appear less likely to be supplemented by other family members, thus increasing their risk of poverty relative to the population of all other workers.

Family-level measures of work often produce higher estimates of working poverty than individual- and family head-based measures. Again using a 50-week threshold to identify workers, we estimate that 5.1% of working families were poor, 0.8 and 0.3 percentage points more than the share of working individuals and family heads, respectively. Because the work of multiple family members may be captured within this family-level work measure, a given week or hour threshold is effectively less stringent than for individual- and family head-based measures of work. Some or all of the persons in working families, as defined by family-based measures, may fail to meet the criteria for work when defined at the individual level (e.g., work 17 + hours per week). In contrast, many of the persons who meet the criteria for work using individual-level measures are embedded in families that include other income-generating members. Across all of the measures we consider in this analysis, estimated national rates of working poverty were approximately 0.5 percentage points higher when considering family heads than all working individuals, and 1.5 percentage points higher when using family-level measures of work (analysis not shown).

Measuring income and poverty: Individual- versus family-level

Next, we consider the difference between estimates based on individual- and family-level measures of income and poverty (Table 4). Estimates using family-level measures are markedly lower than those based on individual earnings, which again reflects the poverty-reducing effect of having multiple workers within some families. Income pooling matters. Net of variation in estimates associated with other criteria for identifying the working poor, estimates that measured income and poverty at the family level were an average of 1.6 percentage points lower than individual-level measures (analysis not shown).

Table 4.

Estimated Poverty Rate (per 100) Among Workers Using Alternative Units to Measure Poverty.

Worker unit	Individual		Individual
Weeks	50+		27+
Hours	N/A		N/A
Income	Unadjusted		Unadjusted
Poverty threshold	Official		Official
Poverty unit	Ind.	Fam.	Ind.	Fam.
All	6.2	4.3	8.4	5.3
Non-Hispanic White	5.3	2.6	7.3	3.4
Non-Hispanic Black	6.9	6.9	9.7	8.8
Hispanic	9.5	10.0	12.3	11.5
Asian and Pacific Islander	5.7	3.4	7.5	4.1

Note. The data are from the March 2013 Current Population Survey. Ind. = individual; Fam. = family.

Reductions in poverty were especially evident when individuals working only part of the year were classified as workers. Using a 27-week threshold to distinguish workers from nonworkers, 8.4% of all workers aged 18 to 64 were poor when considering individual income and using the individual poverty threshold. This estimate is 3.1 percentage points higher than the rate when using family-level income and poverty measures. However, this pattern of results was less apparent among Black and Hispanic workers than White and Asian workers. Using the same 27-week threshold, the shift from individual- to family-level income and poverty measures decreased the estimated rate of poverty among non-Hispanic White and Asian workers by 3.9 and 3.4 percentage points, respectively. Among non-Hispanic Black and Hispanic workers, this change was associated with drops of only 0.9 and 0.8 percentage points, respectively. These results suggest that workers from these groups are underrepresented in households with multiple earners, or that the poverty-reducing impact of having multiple workers within families is lower among these groups.

Measuring income and poverty: Poverty threshold

Poverty researchers routinely debate the merits of the official U.S. poverty line and its many possible alternatives (Iceland, 2005; Lichter, 2005). Here, we demonstrate how the choice of poverty thresholds affects estimates of working poverty using a series of individual-level measures (Table 5). The official poverty threshold produces the most conservative estimates of national- and race-specific rates of poverty among workers. For example, we estimate that 4.3% of working-age persons who worked 50 or more weeks in 2012 lived in families below the official poverty line, at least 2.5 percentage points less than estimates using any other poverty threshold. Two common alternatives to the official poverty line—one set at 125% of the official poverty line and the other set at 50% of median U.S. individual (or household) income—result in markedly higher estimates of working poverty: 7.7% and 11.7%, respectively. We also consider the new SPM advocated by a National Academy of Sciences Panel on Poverty and Assistance's report (Citro & Michael, 1995; Short, 2013). Here, we find that 6.8% of workers (using a 50-week threshold) lived in poor families, as defined by the SPM. National estimates using the SPM typically fall between the official poverty line and 125% of that line. Thresholds based on 50% of the median income usually yield the highest estimates of family poverty.

Table 5.

Estimated Poverty Rate (per 100) Among Workers Using Alternative Poverty Thresholds.

Worker unit	Individual			Individual				Family head				Family
Weeks	50+			50+				50+				50+
Hours	N/A			N/A				N/A				N/A
Poverty unit	Individual			Family				Family				Family
Income	Unadjusted			Unadjusted				Unadjusted				Unadjusted
Poverty threshold	Off.	125% O.	50% M.	Off.	125% O.	50% M.	SPM	Off.	125% O.	50% M.	SPM	Off.	125% O.	50% M.	SPM
All	6.2	9.7	8.5	4.3	7.7	11.7	6.8	4.8	7.7	12.5	6.8	5.1	8.1	11.9	7.9
Non-Hispanic White	5.3	8.0	7.1	2.6	4.8	8.5	4.0	2.9	4.7	8.9	4.2	2.9	4.9	8.4	4.8
Non-Hispanic Black	6.9	11.3	9.7	6.9	12.0	19.5	10.3	8.1	13.2	23.0	10.8	8.1	13.1	21.2	11.9
Hispanic	9.5	16.1	14.0	10.0	17.4	21.0	16.1	12.0	17.9	22.1	16.0	13.3	20.0	22.2	18.9
Asian and Pacific Islander	5.7	8.4	7.3	3.4	6.5	8.4	7.8	3.6	5.8	8.7	7.7	4.6	7.0	9.4	9.2

Note. The data are from the March 2013 Current Population Survey. Off. = official poverty line; 125% O. = 125% of the official poverty line; 50% M. = 50% of median income; SPM = supplemental poverty measure.

The choice of poverty threshold has heterogeneous effects across major racial and ethnic groups. Estimates of working poverty among historically disadvantaged groups—such as Blacks and Hispanics—are especially sensitive to alternative poverty thresholds. Relatively large proportions of Black and Hispanic workers seemingly are bunched between 100% and 125% of the official poverty line, and between the official poverty line and 50% of median income. For instance, the share of all non-Hispanic White workers aged 18 to 64 (using a 50-week threshold) living in a poor family was only 2.2 percentage points higher using 125% of the official poverty lines relative to the official measure itself. The difference was a full 5.1 percentage points among Black workers, and an even greater 7.4 percentage points among Hispanic workers.

Estimates of working poverty among Black families are less sensitive, however, when the official poverty line is replaced with the SPM; Black estimates are, in fact, the least sensitive of all non-White groups. Estimates of working poverty are most sensitive to redefinitions of poverty among Asian workers. Working poverty rates among Asians are in some cases higher using the SPM than using 125% of the official poverty line; they are the only racial and ethnic group for which we observe this. Given the differences between the official poverty line and SPM, both cases likely reflect disproportionate transfers or living costs among working Black and Asian families (Short, 2011).

Measuring income and poverty: Adjusted/unadjusted income

The final comparison (Table 6) considers the effect of adjusting individual annual earnings to their full-year, full-time equivalent (FTE). Among all workers, the estimated rate of poverty is generally lower when using adjusted income to define poverty. By definition, the effect of adjusting income is greater when individuals employed less than full-time or year-round are considered workers. For example, the difference between the working poverty rates is only 0.2 percentage points greater using unadjusted income than adjusted income when using a 50-week threshold to define workers. However, when workers are defined using a 1-week threshold, the rate of poverty among workers is 6.6 percentage points higher when using unadjusted income than adjusted income. Notably, when workers are defined using both the 50-week and 35-hour per week thresholds, the rate of working poverty is higher after adjusting income to a full-time year-round equivalent. A substantial share of workers seemingly escapes poverty only because they work in excess of 1,750 hours annually. Similar trends are observed among Hispanic and non-Hispanic Black workers when work is defined using a 50-week threshold, and among Hispanics when using a 35-hour threshold. A disproportionate share of the non-Hispanic Black and Hispanic persons who work more than full-time (so defined) is actually working for subpoverty wages.

Table 6.

Estimated Poverty Rate (per 100) Among Workers Using Adjusted and Unadjusted Individual Income.

Worker unit	Individual		Individual		Individual		Individual		Individual		Individual		Individual
Weeks	50+		27+		1+		N/A		N/A		N/A		50+
Hours	N/A		N/A		N/A		35+		17+		1+		35+
Poverty unit	Individual		Individual		Individual		Individual		Individual		Individual		Individual
Poverty threshold	Official		Official		Official		Official		Official		Official		Official
Income	Unadj.	Adj.	Unadj.	Adj.	Unadj.	Adj.	Unadj.	Adj.	Unadj.	Adj.	Unadj.	Adj.	Unadj.	Adj.
All	6.2	6.0	8.4	6.6	14.0	7.4	6.3	5.9	11.3	7.2	14.0	7.4	2.5	5.0
Non-Hispanic White	5.3	4.7	7.3	5.3	12.4	6.1	5.1	4.6	9.6	5.8	12.4	6.1	2.1	3.8
Non-Hispanic Black	6.9	7.4	9.7	8.1	17.2	9.2	8.9	7.8	15.1	9.0	17.2	9.2	2.8	6.1
Hispanic	9.5	10.7	12.3	11.2	18.6	11.7	9.9	10.3	16.0	11.5	18.6	11.7	4.6	9.4
Asian and Pacific Islander	5.7	5.6	7.5	6.0	12.2	6.4	5.2	5.2	9.7	6.2	12.2	6.4	2.3	4.5

Note. The data are from the March 2013 Current Population Survey. Unadj. = unadjusted income; Adj. = income adjusted to full-time year-round equivalent.

Summarizing and Evaluating Alternative Estimates

Factor analysis

The parameters used to estimate the size of the working poor population reflect many assumptions. Our descriptive analyses clearly demonstrate that these choices often have significant consequences for estimates of the size and the racial and ethnic composition of the working poor population. To be sure, decisions about how to measure working poverty should be driven by sound theory, but they should also be judged by the measure's empirical properties, which we provide in this section. For each subpopulation of comparable units (i.e., individuals, family heads, and families), we use the factor analytic methods described above to estimate the association between the working poverty status—coded 1 = working poor and 0 = working nonpoor or nonworking—as assigned according to each of the 126 measures we consider, and the latent working poverty variable.

The factor loadings are shown in Tables 7 through 9. All of our measures of working poverty are at least moderately correlated with the latent construct of working poverty (factor loadings > 0.60). But there also is considerable variation in the loadings—particularly among individual-level measures—which suggests that some measures of working poverty are better at identifying individuals (and other units) that fall in the latent working poverty category than others.²⁰

Table 7.

Factor Loadings, All Specifications of Working Poverty for All Individuals, Aged 18–64.

Poverty unit	Income	Poverty threshold	Weeks	Hours	Factor loading
Ind.	Adj.	125% O.		17+	0.9391
Ind.	Adj.	125% O.	27+		0.9362
Ind.	Adj.	50% M.		17+	0.9345
Ind.	Adj.	50% M.	27+		0.9307
Ind.	Adj.	125% O.	1+		0.9293
Ind.	Unadj.	125% O.	27+		0.9236
Ind.	Unadj.	50% M.	27+		0.9215
Ind.	Adj.	50% M.	1+		0.9207
Ind.	Adj.	50% M.		1+	0.9207
Ind.	Adj.	50% M.	50+		0.9150
Ind.	Unadj.	50% M.	50+		0.9146
Ind.	Adj.	125% O.	50+		0.9142
Ind.	Unadj.	125% O.	50+		0.9133
Ind.	Adj.	50% M.		35+	0.9132
Ind.	Adj.	50% M.	50+	35+	0.9132
Ind.	Unadj.	125% O.	1+		0.9128
Ind.	Unadj.	Off.	50+	35+	0.9111
Ind.	Unadj.	125% O.		17+	0.9103
Ind.	Adj.	125% O.		35+	0.9102
Ind.	Adj.	Off.	50+	35+	0.9101
Ind.	Unadj.	50% M.	50+	35+	0.9062
Ind.	Unadj.	50% M.		17+	0.9029
Ind.	Adj.	Off.		17+	0.8973
Ind.	Adj.	Off.		35+	0.8927
Ind.	Unadj.	125% O.	50+	35+	0.8917
Ind.	Unadj.	50% M.	1+		0.8911
Ind.	Adj.	125% O.	50+	35+	0.8902
Ind.	Adj.	Off.	27+		0.8893
Ind.	Adj.	Off.	50+		0.8832
Ind.	Unadj.	Off.	50+		0.8829
Ind.	Unadj.	Off.	27+		0.8808
Ind.	Adj.	Off.	1+		0.8682
Fam.	Unadj.	Off.	27+		0.8563
Ind.	Unadj.	50% M.		35+	0.8525
Ind.	Unadj.	125% O.		35+	0.8497
Ind.	Unadj.	Off.		17+	0.8463
Fam.	Unadj.	125% O.	27+		0.8436
Fam.	Unadj.	Off.	50+		0.8421
Fam.	Unadj.	Off.		17+	0.8367
Fam.	Unadj.	125% O.		17+	0.8346
Fam.	Unadj.	125% O.	1+		0.8332
Fam.	Unadj.	Off.	1+		0.8273
Ind.	Unadj.	Off.	1+		0.8201
Fam.	Unadj.	125% O.	50+		0.8171
Ind.	Unadj.	Off.		35+	0.8111
SR	Unadj.	SPM	27+		0.7955
SR	Unadj.	SPM		17+	0.7887
SR	Unadj.	SPM	1+		0.7853
SR	Unadj.	SPM		1+	0.7842
Fam.	Unadj.	Off.		35+	0.7814
Fam.	Unadj.	Off.	50+	35+	0.7783
Fam.	Unadj.	50% M.	1+		0.7782
Fam.	Unadj.	50% M.		1+	0.7782
SR	Unadj.	SPM	50+		0.7691
Fam.	Unadj.	50% M.		17+	0.7655
Fam.	Unadj.	125% O.		35+	0.7576
Fam.	Unadj.	50% M.	27+		0.7574
Fam.	Unadj.	125% O.	50+	35+	0.7328
SR	Unadj.	SPM		35+	0.7204
Fam.	Unadj.	50% M.	50+		0.7129
SR	Unadj.	SPM	50+	35+	0.6808
Fam.	Unadj.	50% M.		35+	0.6572
Fam.	Unadj.	50% M.	50+	35+	0.6074

Note. The data are from the March 2013 Current Population Survey. Ind. = individual; Fam. = family; SR = supplementary resource unit; Adj. = income adjusted to full-time year-round equivalent; Unadj. = unadjusted income; Off. = official poverty line; 125% O. = 125% of official poverty line; SPM = supplemental poverty measure; 50% M. = 50% of median income.

Table 8.

Factor Loadings, All Specifications of Working Poverty for Family Heads.

Poverty unit	Income	Poverty threshold	Weeks	Hours	Factor loading
Fam.	Unadj.	125% O.	1+		0.9557
Fam.	Unadj.	125% O.		1+	0.9557
Fam.	Unadj.	125% O.		17+	0.9520
Fam.	Unadj.	125% O.	27+		0.9402
Fam.	Unadj.	50% M.	1+		0.9276
Fam.	Unadj.	50% M.		1+	0.9276
Fam.	Unadj.	Off.	27+		0.9251
Fam.	Unadj.	Off.		17+	0.9223
Fam.	Unadj.	50% M.		17+	0.9179
Fam.	Unadj.	Off.	1+		0.9155
Fam.	Unadj.	Off.		1+	0.9155
Fam.	Unadj.	Off.	50+		0.9133
Fam.	Unadj.	125% O.	50+		0.9109
Fam.	Unadj.	Off.	50+	35+	0.9098
Fam.	Unadj.	125% O.		35+	0.9048
Fam.	Unadj.	Off.		35+	0.8993
Fam.	Unadj.	50% M.	27+		0.8878
SR	Unadj.	SPM		17+	0.8795
SR	Unadj.	SPM	1+		0.8781
SR	Unadj.	SPM		1+	0.8781
Fam.	Unadj.	125% O.	50+	35+	0.8705
SR	Unadj.	SPM	27+		0.8651
Fam.	Unadj.	50% M.	50+		0.8379
Fam.	Unadj.	50% M.		35+	0.8362
SR	Unadj.	SPM	50+		0.8351
SR	Unadj.	SPM		35+	0.8332
SR	Unadj.	SPM	50+	35+	0.7829
Fam.	Unadj.	50% M.	50+	35+	0.7693

Note. The data are from the March 2013 Current Population Survey. Fam. = family; SR = supplementary resource unit; Unadj. = unadjusted income; Off. = official poverty line; 125% O. = 125% of official poverty line; SPM = supplemental poverty measure; 50% M. = 50% of median income.

Table 9.

Factor Loadings, All Specifications of Working Poverty for Families.

Poverty unit	Income	Poverty threshold	Weeks	Hours	Factor loading
Fam.	Unadj.	125% O.		35+	0.9703
Fam.	Unadj.	125% O.		17+	0.9703
Fam.	Unadj.	125% O.		1+	0.9698
Fam.	Unadj.	125% O.	1+		0.9698
Fam.	Unadj.	125% O.	27+		0.9569
Fam.	Unadj.	Off.	1+		0.9562
Fam.	Unadj.	Off.	27+		0.9556
Fam.	Unadj.	Off.	50+		0.9478
Fam.	Unadj.	Off.	50+	35+	0.9478
Fam.	Unadj.	Off.		35+	0.9440
Fam.	Unadj.	Off.		17+	0.9433
Fam.	Unadj.	Off.		1+	0.9412
Fam.	Unadj.	50% M.		35+	0.9408
Fam.	Unadj.	50% M.		17+	0.9408
Fam.	Unadj.	50% M.	1+		0.9400
Fam.	Unadj.	50% M.		1+	0.9400
Fam.	Unadj.	125% O.	50+		0.9277
Fam.	Unadj.	125% O.	50+	35+	0.9277
SR	Unadj.	SPM		35+	0.9061
Fam.	Unadj.	50% M.	27+		0.9061
SR	Unadj.	SPM		17+	0.9059
SR	Unadj.	SPM		1+	0.9049
SR	Unadj.	SPM	1+		0.9049
SR	Unadj.	SPM	27+		0.8898
Fam.	Unadj.	50% M.	50+		0.8661
Fam.	Unadj.	50% M.	50+	35+	0.8661
SR	Unadj.	SPM	50+	35+	0.8587
SR	Unadj.	SPM	50+		0.8577

Among individuals, for example, the measure using a 17 + hour a week threshold to distinguish workers from nonworkers, adjusting income, and using 125% of the official poverty threshold for single persons has the highest factor loading (0.9391). From a purely statistical standpoint, this measure of working poverty is best relative to others we consider and suggests that 12.4% of American workers are poor (see Appendix). In contrast, the measure that defines workers as persons working 50 + weeks per year and 35 + hours per week, does not adjust income, and identifies the poor using thresholds based on 50% of median family income has the lowest factor score of individual-level measures (0.6074). Using this definition, an estimated 9.2% of workers are poor. Factor loadings among family head- and family-based measures also vary from 0.7693 to 0.9557 and 0.8577 to 0.9703, respectively. Although these measures are not strictly comparable, it is worth noting that within all three analytic populations, the measure with the lowest factor loading had an extremely high work threshold (e.g., must have worked both 50 + weeks or 35 + hours per week) and used either the SPM or a poverty threshold based upon 50% of median income; those with the highest tended to use a work threshold that included part-time or part-year employees as workers, and 125% of the official poverty line to define poverty.

Considered together, the factor scores among individual-level measures indicate that measures based on a highly individualized conceptualization of the working poor have the strongest correlation with the latent working poverty category. These measures tend to define poverty status according to individual income only (i.e., excluding income from other family members), and to use poverty thresholds that capture the near poor or the relative poor, as well as part-time workers whose wages would leave them below the poverty line regardless of whether or not they worked full-time, year-round (i.e., adjusted income). In short, these results support a conceptualization of working poverty that focuses largely on job quality (Kalleberg, 2011).

Next, we consider measures of working poverty among family heads (Table 8) and family-based measures (Table 9). By definition, these sets of measures shift the emphasis from individual earnings to family income. Yet, in many respects, these results mirror the analysis of measures of working poverty among all working-age individuals shown in Table 7. The factor loadings are highest among measures classifying family heads as workers even if they are employed only part-time or part-year—or in the case of family measures, if family members' combined work efforts amount to only a part-time or part-year equivalent. Factor loadings are also highest among measures using poverty thresholds above those officially set by the government.

Modeling the working poor

We have sought to identify specific measures ofworking poverty that are most correlated with the latent construct of working poverty. We have revealed a range of empirical estimates of working poverty and identified specific measures with the largest factor loadings, which indicate the extent to which each measure is correlated with the latent construct of working poverty among our observations. An alternative but complementary approach is to develop a single summary index based on the factor loadings, which gives different weights to each distinctive empirical indicator of working poverty. Using these weights and observed empirical indicators, we construct a composite index of working poverty. We then model this mathematically derived summary measure as the dependent variable in a linear regression analysis that includes basic social and demographic characteristics. We also estimate regression models using the most and least restrictive measures of working poverty—as indicated by the rate of poverty among workers estimated with each measure—and two measures we argue are substantively most appropriate. These analyses provide evidence of concurrent validity if the models yield results that are consistent with each other, as well as with our understanding of existing group-to-group differences in poverty among workers based on previous studies.

Table 10 shows the results from these models. In the interest of parsimony, we limit our results and discussion to basic demographic predictors of race, sex, and education. Moving left to right, the first model utilizes the most conservative measure of working poverty, with which only 2.5% of workers were estimated to be poor. The second model utilizes the least restrictive measure,²¹ which yields an estimated rate of poverty of 18.5%. The third and fourth models use similar definitions of working poverty that we feel are substantively most appropriate (see Discussion section below). Using these measures, estimated rates of poverty among workers are 9.3% and 11.0%, respectively. The final three columns present results using the working poverty index for individual-, family head-, and family-based measures as the outcome variable.

Table 10.

Select Coefficient Estimates From Regressions Predicting Poverty Among Workers, by Measure of Working Poverty.

Worker unit	Ind.	Ind.	Fam. H.	Fam. H.	Index, individual	Index, family head	Index, family
Worker weeks	50+	–	27+	–
Worker hours	35+	1+	–	17+
Poverty unit	Ind.	Ind.	Fam.	Fam.
Income	Unadj.	Unadj.	Unadj.	Unadj.
Poverty threshold	Off.	125% O.	125% O.	125% O.
Race
Non-Hispanic White	(ref)	(ref)	(ref)	(ref)	(ref)	(ref)	(ref)
Non-Hispanic Black	0.097	0.193***	0.672***	0.705***	0.031***	0.054***	0.059***
Hispanic	0.288***	0.147***	0.736***	0.698***	0.043***	0.059***	0.106***
Asian/Pacific Islander	0.191	0.119**	0.422***	0.354***	0.015***	0.017***	0.042***
Sex
Male
Female	0.467***	0.728***	0.582***	0.562***	0.023***	0.015***	0.023***
Education
Less than high school
High school degree	−0.742***	−0.819***	−0.969***	−0.939***	−0.035***	−0.048***	−0.042***
Some college	−1.239***	−1.021***	−1.449***	−1.385***	−0.053***	−0.063***	−0.059***
Bachelors+	−1.841***	−1.921***	−2.639***	−2.592***	−0.088***	−0.111***	−0.102***

Note. The data are from the March 2013 Current Population Survey. Ind. = individual; Fam. H. = family head; F = family; Unadj. = unadjusted income; Off. = official poverty line; 125% O. = 125% of the official poverty line.

p < .05. ***

p < .01.

Estimates from these regression models indicate that regardless of the measure of working poverty used, White workers are nearly always less likely than racial and ethnic minorities to be poor and thus included among the working poor. Only when using the first, most restrictive measure of working poverty are the differences between Whites and some minority groups—Blacks and Asians—nonsignificant. Female workers are also at greater risk than men of being in working poverty across all models.²² Finally, we find statistically significant differences in the risk of working poverty across levels of educational attainment, regardless of how working poverty is defined.

Discussion

The Great Recession and its aftermath have raised new public policy concerns about putative increases in the share of American workers who are poor (Brady et al., 2013; Bureau of Labor Statistics, 2013). However, any broad consensus on the magnitude of the problem—and its distribution across major racial and ethnic groups—is undermined by serious conceptual and measurement problems. Our article critiques existing approaches by estimating a range of alternative measures of working poverty using data from the 2013 March CPS and identifying key social and demographic factors associated with a range of indicators of working poverty.

At a minimum, the results of our empirical analyses demonstrate that estimates of poverty among workers depend heavily on the assumptions used. In fact, the range of overall rates was 16 percentage points, from a low of 2.5% to a high of 18.5%. Our best judgment—based on our review, conceptual discussion, and the factor analyses—is that between 9.3% (27 + week work threshold; 125% official poverty line) and 11.0% (17+hour/week threshold; 125% of poverty line) of working family heads are poor. This means that the promise that work provides a pathway to a decent, above-poverty standard of living is broken for roughly one in ten working family heads and their dependents. Using these two preferred measures—based on well-justified conceptual and empirical criteria—we estimate that between approximately 6.4 and 8.0 million working family heads were poor in 2012, and between 20 and 24 million persons lived in poor families with a working head.²³

More broadly, we find substantial variation in the correlation between each of the 126 measures we consider and a latent construct of working poverty. This suggests that some measures are more likely to accurately indicate whether an observation falls into the latent working poverty category than others. In general, those measures with the highest correlations include part-time or part-year employment, use poverty income thresholds that include the near poor (i.e., less than 125% of the poverty level), and, when considering individual workers, adjust income to a full-time, year-round equivalent. Yet, despite these differences, the conditional risk profiles we estimate using regression techniques reveal remarkably similar results. Ethnic and racial minorities, women, and individuals with low levels of education have a higher risk of working poverty. These conclusions are broadly supported by each of the regression models we estimate, including those using the indices derived from factor analysis. These results suggest that while conclusions about the share of workers in poverty are sensitive to measurement choices, those regarding the risk factors for poverty among workers generally are not. This may be reassuring to researchers using alternative definitions of the working poor.

Our measurement approach and analysis suggest several key takeaway points and lessons for future research. First, and most generally, we show that technical choices about the criteria used to define workers and poverty have significant implications for how the working poor are conceptualized as a socially and analytically distinct population. Such choices are inextricably linked to normative assumptions about the amount of work effort required to escape poverty, the age at which financial independence is expected, and how the burden of work and its benefits should be distributed among family members. If nothing else, our article makes clear the unrecognized normative choices that underlie conventional measures of the working poor.

Second, we also demonstrate that measurement choices have clear implications for national- and race-specific estimates of the working poor, and therefore also for the perceptions of the problem. Our estimates, based on different indicators, reveal substantial variation (range = 16.0 percentage points). Given contemporary debates about unemployment benefits, food stamps, and other public sector support for economically disadvantaged working-age adults, such estimates can undoubtedly serve as a catalyst for public policy interventions (e.g., raising the minimum wage or providing other work supports) or as a justification for inaction. Our analyses make clear the technical and conceptual basis on which recent policy judgments rest.

Third, our results also provide several important lessons pertaining to the best set of criteria for measuring working poverty. While each measurement scheme rests on debatable assumptions or other measurement limitations, our conceptual critique and empirical analyses suggest that some measures of working poverty are statistically better than others. Our factor analytic approach identifies those measures of working poverty that are most strongly correlated with the latent construct, and arguably most valid. From a statistical perspective, the best measures of working poverty include individuals working part-time and part-year. Workers should be defined to include individuals and families living both below and slightly above the official poverty income thresholds (i.e., 125% of poverty). That is, the best measures include those who may not be officially poor but nonetheless lack the resources to fully participate in mainstream American society or accumulate savings and wealth. Among individual-level measures, those that emphasize job quality are strongest. Finally, use of the family as the unit of analysis for measuring work effort, and employing poverty thresholds based on the SPM generally, receives less support than other measurement approaches.

A fourth important lesson for future research pertains to the universe of potential workers. Given the noncomparability of the three distinct populations we considered, analytic decisions must be based upon theory and the demands of a particular research question. In general, however, we recommend identifying the poverty status of individual workers rather than aggregated families or households. Family-level measures consider the work of all members regardless of age, even family members not usually considered responsible for their own economic support (e.g., youth). Moreover, it is difficult to set a week or hour threshold for aggregated family-level measures: Using the equivalent of a single person working full-time or year-round is a rather low threshold for the combined work of all family members, but the choice of and justification for a higher alternative is not necessarily clear.

Fifth, our results also support restricting the sample to family heads rather than considering all working-age individuals, a population more likely to include voluntary part-time workers (e.g., secondary workers) and working college students. If these subpopulations are systematically more likely to be nonpoor or live in nonpoor families, their inclusion among the population of workers will lead to underestimates of the size of the working poor population.

Conclusion

To conclude, our fundamental goal was to provide a conceptual and empirical baseline for judging alternative measurement criteria. Our baseline estimates highlight the size and composition of America's working poor population at a time of growing income inequality, continuing high rates of poverty in the aftermath of the Great Recession, and declining real earnings, especially among America's historically disadvantaged minority populations. It is time to seriously reconsider a number of assumptions about poverty and employment, including whether or what kind of job growth is an effective poverty reduction strategy; and relatedly, what type of other policies can best improve economic outcomes among workers in the U.S. (and elsewhere; Andress & Lohmann, 2008; Bernhardt, 2012; Lambert, 2009; Leicht, 2010).

Given the evidence presented here and in other recent policy reports, renewed debate about increasing the federal minimal wage should be seen as a positive development (Cooper, 2013). Such an increase has the potential to improve the welfare of many hard-working adults—not just teenagers, as critics suggest—in America. A substantial proportion of working adults take home below-poverty earnings even when they were able and willing to work full-time and year-round. Although debates over the minimum wage often take on a partisan tone, the evidence that many Americans are playing by the rules but unable to escape poverty should provide a compelling reason for bipartisan action.

Finally, our consideration of the working poor and its many conceptual and measurement issues is not simply an academic exercise. Alternative assumptions about the extent and etiology of poverty, and the amount of work and income that is required to escape poverty, reflect important value judgments. They also guide policy decisions that have real-world social, psychological, and material implications for working Americans at the bottom of the income distribution.

Footnotes

Author Note

An earlier version of this article was presented at the 2013 annual meetings of the American Sociological Association.

Acknowledgments

We would like to thank John Hoffman for his methodological advice as well as Jahaan Chandler and Danielle Thomas for their comments on previous iterations of this article. We would also like to thank editor Daniel Cornfield and three anonymous reviewers for their constructive feedback during the review process. All remaining errors of any kind are our responsibility.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Cornell Population Center (R24 HD058488) through infrastructure funding by the National Institute of Child Health and Human Development.

Notes

Author Biographies

Brian C. Thiede is an assistant professor of sociology at Louisiana State University. His current research includes projects examining the relationship between demographic change, labor markets, and household economic status in the U.S., as well as demographic change in low-income contexts.

Daniel T. Lichter is the Ferris family professor in the department of policy analysis and management, professor of sociology, and Director of the Cornell Population Center. His recent work has focused on changing ethno-racial boundaries, as measured by changing patterns of interracial marriage, residential segregation, and concentrated poverty in the U.S.

Scott R. Sanders is an assistant professor of sociology at Brigham Young University. His research examines variations in poverty and health care access across different demographic groups and geographic locations. His research also examines the link between poverty reduction and globalization in Southeast Asia.

References

Andress

H. J.

Lohmann

(2008) The working poor in Europe: Employment, poverty, and globalization, Cheltenham, England: Edward Elgar.

Bernhardt

(2012) The role of labor market regulation in rebuilding economic opportunity in the United States. Work and Occupations 39: 354–375.

Blank

R. M.

Danzinger

S. H.

Schoeni

R. F.

(2006) Working and poor, New York, NY: Russell Sage Foundation.

Brady

Baker

R. S.

Finnigan

(2013) When unionization disappears: State-level unionization and working poverty in the United States. American Sociological Review 78: 872–896.

Brady

Fullerton

A. S.

Cross

J. M.

(2010) More than just nickels and dimes: A cross- national analysis of working poverty in affluent democracies. Social Problems 57: 559–585.

Bureau of Labor Statistics (2013) A profile of the working poor, 2011 (BLS Report # 1041), Washington, DC: U.S. Bureau of Labor Statistics.

Caputo

R. K.

(2007) Working and poor: A panel study of maturing adults in the U.S. Families in Society 88: 351–359.

Citro

C. F.

Michael

R. T.

(1995) Measuring poverty: A new approach, Washington, DC: National Academy Press.

Cooper

(2013) Raising the federal minimal wage to $10.10 would lift wages for millions and provide a modest economic boost, (Economic Policy Institute Briefing Paper #371) Washington, DC: Economic Policy Institute.

10.

Correll

S. J.

Kelley

E. L.

O'Connor

L. T.

Williams

J. C.

(2014) Redesigning, redefining work. Work and Occupations 41: 3–17.

11.

Crettaz

(2013) A state-of-the-art review of working poverty in advanced economies: Theoretical models, measurement issues, and risk groups. Journal of European Social Policy 23: 347–362.

12.

DeFina

R. H.

(2007) A comparison of poverty trends and policy impacts for working families using different poverty indexes. Journal of Economic and Social Measurement 32: 129–147.

13.

DeJong

G. E.

Madamba

A. B.

(2001) A double disadvantage? Minority group, immigrant status, and underemployment in the United States. Social Science Quarterly 82: 117–130.

14.

DeNavas-Walt

Proctor

B. D.

Smith

J. C.

(2013) Income, poverty, and health insurance coverage in the United States: 2012. Current Population Reports, Washington, DC: U.S. Census Bureau.

15.

Fryer

R. G.

Heaton

P. S.

Levitt

S. D.

Murphy

K. M.

(2013) Measuring crack cocaine and its impact. Economic Inquiry 51: 1651–1681.

16.

Hauan

S. M.

Landale

N. S.

Leicht

K. T.

(2000) Poverty and work effort among urban Latino men. Work and Occupations 27: 188–222.

17.

Iceland

(2005) Measuring poverty: Theoretical and empirical considerations. Measurement: Interdisciplinary Research and Perspectives 3: 199–235.

18.

Iceland

Kim

(2001) Poverty among working families: New insights from an improved poverty measure. Social Science Quarterly 82: 253–266.

19.

Jenkins

S. P.

Brandolini

Micklewright

Nolan

(2013) The Great Recession and the distribution of household income, New York, NY: Oxford University Press.

20.

Jensen

Findeis

J. L.

Hsu

Schachter

J. P.

(1999) Slipping into and out of underemployment: Another disadvantage for nonmetropolitan workers? Rural Sociology 64: 417–438.

21.

Jensen

Slack

(2003) Underemployment in America: Measurement and evidence. American Journal of Community Psychology 32: 21–31.

22.

Joassart-Marcelli

(2005) Working poverty in southern California: Towards an operational measure. Social Science Research 34: 20–43.

23.

Kalleberg

A. L.

(2011) Good jobs, bad jobs: The rise of polarized and precarious employment systems in the United States, 1970s–2000s, New York, NY: Russell Sage.

24.

Katz

(2011) The undeserving poor, New York, NY: Pantheon Books.

25.

King

Ruggles

Alexander

J. T.

Flood

Genadek

Schroeder

M. B.

Vick

(2010) Integrated public use microdata series, Current Population Survey: Version 3.0, [Machine-readable database] Minneapolis: University of Minnesota.

26.

Kmec

J. A.

O'Connor

L. T.

Schieman

(2014) Not ideal: The association between working anything but full time and perceived unfair treatment. Work and Occupations 41: 63–85.

27.

Lambert

S. J.

(2009) Lessons from the policy world: How the economy, work supports, and education matter for low-income workers. Work and Occupations 36(1): 56–65.

28.

Leicht

K. T.

(2010) Nickels and dimes won't fix this: The future of work and pay in America. Work and Occupations 37(2): 225–233.

29.

Lichter

D. T.

(2005) In search of the “best” poverty measure. Measurement: Interdisciplinary Research and Perspectives 3: 253–258.

30.

Lichter

D. T.

Crowley

M. L.

(2002) Poverty in America: Beyond welfare reform. Population Bulletin Vol 57, Washington, DC: Population Reference Bureau.

31.

Lichter

D. T.

Johnston

G. M.

McLaughlin

D. K.

(1994) Changing linkages between work and poverty in rural America. Rural Sociology 59: 395–415.

32.

Loehlin

J. C.

(1987) Latent variable models: An introduction to factor, path, and structural analysis, Hillsdale, NJ: Erlbaum.

33.

Lohmann

(2009) Welfare states, labour market institutions, and the working poor: A comparative analysis of 20 European countries. European Sociological Review 25: 489–504.

34.

Mishel

Bivens

Gould

Shierholz

(2012) The state of working America, Ithaca, NY: Cornell University Press.

35.

Newman

K. S.

(1999) No shame in my game: The working poor in the inner city, New York, NY: Vintage Books.

36.

Short

(2011) The research supplementary poverty measure: 2010. Current Population Reports, Washington, DC: U.S. Census Bureau.

37.

Short

(2013) The research supplementary poverty measure: 2012. Current Population Reports, Washington, DC: U.S. Census Bureau.

38.

Slack

(2010) Working poverty across the metro-nonmetro divide: A quarter century in perspective, 1979–2003. Rural Sociology 75: 363–387.

39.

Slack

Jensen

(2002) Race, ethnicity, and underemployment in nonmetropolitan America: A 30-year profile. Rural Sociology 67: 208–233.

40.

Slack

Jensen

(2008) Birth and fortune revisited: A cohort analysis of underemployment, 1974–2004. Population Research and Policy Review 27: 729–749.

41.

StataCorp (2013) Stata's user guide: Release 13, College Station, TX: StataCorp LP.