Patterns of Earnings and Employment by Worker Sex, Race, and Ethnicity Using State Administrative Data: Results from a Sample of Workers Connected to Public Assistance Programs

Abstract

During the strong economic conditions that predated the COVID-19 pandemic, many US workers, especially females and individuals of color, suffered from economic vulnerability. Despite growing research attention, we lack an understanding of how the prevalence and patterns of earnings and job instability vary with worker characteristics, particularly at the intersections between sex and race/ethnicity. This study uses longitudinal administrative data from a large, diverse state from 2015 through 2018 to document changes in earnings and jobs. We then examine variation in the size, frequency, and direction of these changes by worker sex and race/ethnicity among a subsample of workers who are connected to the public welfare system. Results indicate that, as expected, workers who are connected to the public welfare system experienced higher levels of economic vulnerability, but with substantial racial/ethnic and sex differences. As a consequence, a large number of workers—disproportionately those of color—were experiencing high levels of economic instability during a period of strong economic growth. Our findings have implications for policy and practice strategies.

Introduction

The period of economic growth that ended in 2020 with the COVID-19 pandemic was the longest in the post-World War II era and was exceptional across a range of economic measures, including historic low levels unemployment and high levels of wage growth (Edwards & Smith, 2020). Despite the overall strength of the economy in the period leading up to the COVID-10 pandemic, many U.S. workers, particularly low-wage and low-skilled workers who are more likely to be nonwhite and female, lacked financial stability and predictability (Board of Governors of the Federal Reserve System, 2018), with important negative consequences. For example, employment and work schedule instability—which is more frequently experienced by low-wage workers—leads to unpredictable and unstable household income and schedules, contributing to family stress, inconsistent routines, which in turn, have cascading, negative effects on children and families (Hardy, 2014; Hill et al., 2013; Lambert et al., 2019; Sandstrom & Huerta, 2013; Wolf & Morrissey, 2017).

Despite growing research attention to economic vulnerability, however, we lack a clear understanding of the full dimensionality of economic vulnerability, such as the frequency, magnitude, direction of earnings and job instability. Importantly, the extant research fails to document how the prevalence and patterns of economic vulnerability vary at the intersections between sex, race, and ethnicity. Further, much of the work in this area uses surveys based on self-reports and recall that may not accurately reflect earnings or job changes (Bollinger et al., 2019). This study addresses these gaps in the literature by examining the prevalence and patterns of economic vulnerability using longitudinal administrative data from a large, diverse state from 2015 through 2018. Specifically, we examined: earnings levels; the size, frequency, and direction of earnings shocks; and job changes among the broader population of workers covered by Unemployment Insurance (UI). We then focused on a subsample of workers, which we refer to as our Public Welfare Connected sample, who participated in either the Temporary Assistance to Needy Families [TANF] or the Supplemental Nutrition Assistance Program [SNAP] at some point from 2015 to 2018, and explore observed differences in earnings and job changes experienced by worker sex, race, and ethnicity.

This descriptive project uses the most complete earnings data available, administrative data from UI wage records, to highlight the contours of economic vulnerability at the height of the strong pre-pandemic economy. Our innovative use of longitudinal state administrative data provides the universe of a specific population of interest—low-wage workers eligible for and connected to public welfare programs—as well as the sample size necessary to explore economic vulnerability at the intersection between sex and race/ethnicity going beyond the Black-White dichotomy. Specifically, we investigate three research questions and hypotheses:

• 1. What is the prevalence of earnings and job instability among workers in a large, diverse state between 2015 and 2018? We hypothesize relatively low rates of both earnings instability and job changes during this period of historic economic growth for the full universe of covered employment.

• 2. How do patterns of earnings and job instability vary for a public welfare connected sample? We expect that those connected to public assistance programs will experience lower levels of earnings and higher rates of, and greater variation in, both earnings instability and job changes relative to the full sample of covered workers.

• 3. Among this subsample of public welfare connected workers, how do patterns of earnings and job instability vary at the intersection of worker sex and race/ethnicity? We hypothesize that male and female workers of color will experience higher rates of both earnings instability and job changes relative to their White female and male counterparts.

We find that workers who are connected to public welfare programs are more economically vulnerability across all measures examined, including earnings levels, earnings stability, and job stability. Furthermore, among those connected to the public welfare system, sex, race and ethnic differences in labor market experiences combine to disadvantage Black female and male workers more than their White, Asian and Hispanic male and female counterparts. Together, findings enhance our understanding of the scope of and disparities in earnings and job instability in the pre-pandemic economy, with implications for the recovery following the COVID-19 pandemic and beyond.

Economic Instability

Instability in economic resources, including job loss, job changes, and large swings in income, is a common and growing phenomenon (Dynan et al., 2012; Board of Governors, 2020; Hardy & Ziliak, 2014; Morris et al., 2015; Wolf et al., 2014; Ziliak et al., 2011). Structural changes in the economy, including a shift in financial risk from corporations to families combined with rollbacks in government regulation, unionization, and worker protections, contributed to reductions in worker financial security and a widening gap between low- and high-earning workers over the last few decades (Hacker, 2019; Kalleberg, 2011). In 2019, three in ten American adults had incomes that fluctuated from month to month, and this lack of economic stability is associated with financial hardship; over one-third of those with variable incomes, compared to one in ten adults overall, reported struggling to pay their bills (Board of Governors, 2020). Earnings changes from labor account for the largest source of income instability (Farrell & Greig, 2016). In turn, job loss, a change in family structure (e.g., divorce), and changes in health status are the most common causes of economic instability (Acs & Nichols, 2010; Acs et al., 2009; Western et al., 2016), but variable work schedules—more common in certain industries and among low-educated workers—also contribute to economic instability (Board of Governors, 2020; Morduch & Siwicki, 2017). Low-income households are the most likely to experience these wide changes in income, and may also have the fewest resources to buffer the negative effects (Mckernan et al., 2009). One study using bank account data from 2012 through 2015—a period of relative economic growth—found that 74% of those in the bottom quintile of income experienced a month-to-month change in income of 30% or more (Farrell & Greig, 2016). As a consequence, we expect that even during the strong economic conditions of the period preceding the pandemic, economic vulnerability was widespread for a subset of the labor force.

Racial and Ethnic Disparities in Economic Vulnerability

In the United States, income, wealth, and economic security are highly stratified by race and ethnicity (Gibson-Davis et al., 2020; Board of Governors, 2020; McKernan et al., 2014). For example, in 2019, 79% of White adults compared to 66% and 65% of Hispanic and Black adults, respectively, reported they were doing ok financially. Among adults not working full-time, Black and Hispanic workers across educational levels were disproportionately more likely to report wanting to work more (Board of Governors, 2020). Hispanic and Black individuals are less likely to have health insurance (Buchmueller & Levy, 2020), more likely to receive high-cost mortgages (Bayer et al., 2018), more likely to be unbanked, and are more likely to face challenges accessing credit (Board of Governors, 2020) than their White counterparts, all of which increase economic vulnerability. Further, individuals of color disproportionately work in industries like retail and food that are more likely to entail variable shifts (U.S. Bureau of Labor Statistics, 2021). Hispanic workers of both genders are more likely to begin and end work at various times due to their employers’ request (Mccrate, 2021). Indeed, research has found that workers who are Black, young, and female are most at risk of having jobs with precarious scheduling practices, including volatile hours and unpredictable scheduling (Lambert et al., 2019).

A small but growing body of evidence has investigated patterns of income instability by race and ethnicity, finding patterns that differ somewhat from those of average income. Since the 1990s, Black workers are more likely to have been displaced from their jobs than White workers (Wrigley-Field & Seltzer, 2020). During the Great Recession, losses among Black workers about 40% greater than other racial groups (although men experienced greater losses compared to women) (Rinz, 2019). Using a nationally representative sample of families with children from the 2004 and 2008 panels of the Survey of Income and Program Participation (SIPP), Gennetian and colleagues found that across race and ethnicity, children in lower-income families experienced greater income volatility than their higher-income peers, but at the lowest income levels, Hispanic children were slightly more likely to experience stable incomes than their peers (Gennetian et al., 2019). Black, American Indian, and Hispanic individuals have a higher probability of experiencing downward income mobility than their White or Asian American counterparts (Akee et al., 2019), suggesting that this higher income volatility is generally not contributing to upward mobility for workers of color. We expect that nonwhite workers will have higher levels of economic vulnerability relative to White workers.

Intersections of Race, Ethnicity, and Sex in Economic Vulnerability

Like race/ethnicity, there is considerable evidence for earnings variation by gender, with women earning less than their male peers in nearly all occupations (Altonji & Blank, 1999; Greenman & Xie, 2008; Jones, 2021; Kahn, 2013): the 2020 female-to-male earnings ratio for full-time, year-round workers was 83% (Shrider et al., 2021). The economy’s demand for long and inflexible work hours disadvantage women who disproportionately take on caregiving roles (Goldin, 2014), and experimental evidence indicates that mothers are discriminated in the job search (Correll et al., 2007). Recent research has not found overall gender differences in working variable schedules, but mothers are less likely than men to work variable schedules (Mccrate, 2021), likely a result of disproportionate caregiving roles.

The intersectional forces of sex, race, and ethnicity may particularly disadvantage women of color who earn less than their same race and ethnicity counterparts regardless of level of education (Altonji & Blank, 1999; U.S. Bureau of Labor Statistics, 2021). In addition, patterns of racial and gender discrimination may lead Black or Hispanic women of similar educational or income backgrounds as their White male peers to show greater levels of job and earnings instability than both White male and female peers. For example, Schneider and Harknett (2019) document that female workers of color in one study of workers at food and retail companies were more likely to have unstable and unpredictable schedules than their White coworkers (Schneider & Harknett, 2019). Prior research also documents earnings differentials by sex and race/ethnicity, but suggest that gender and racial penalties may not be additive, but that women from minority groups experience a smaller gender gap with same-race men (though show large racial gaps) (Greenman & Xie, 2008).

To date, however, less research has examined the intersection of sex and race/ethnicity with regard to instability in earnings and jobs, however, which may have separate implications for household stress and well-being. One exception is Hardy (2012), who finds that while overall earnings volatility among Black women fell from the 1970s to the early 2000s, levels were consistently higher than those of White women (Hardy, 2012). We expect to find that female workers face higher levels of economic vulnerability relative to male workers of the same race.

Economic Vulnerability Among Public Assistance Participants

As discussed above, low-income households are more at-risk of experiencing economic instability and are often eligible for means-tested public assistance programs designed to temporarily provide resources or smooth temporary periods of economic instability related to labor market participation or low earnings.¹ The Supplemental Nutrition Assistance Program (SNAP), which provides cash-like vouchers that can be used to purchase food prepared at home, is one of the few public programs that reaches households with and without children, and participation is associated with a range of benefits for food security, health, and other outcomes (CBPP, 2019). Temporary Assistance to Needy Families (TANF) provides cash assistance to poor families, although the program reaches many fewer poor households with children than it did decades ago—just 23% of families with children in poverty in 2017 compared to 68% in 1996—and now has a five-year maximum participation limit in most states (Floyd et al., 2018).

Both SNAP and the TANF program are means-tested, and fluctuations in earnings and jobs change income eligibility and benefit levels. As a result, changes in earnings and jobs likely trigger increased reporting requirements and the associated administrative burden of compliance. Despite these program similarities, there is relatively little overlap of SNAP caseloads with TANF; nationally, only 4% households participating in SNAP, and only 10% of SNAP households with children, also received TANF benefits (Cronquist, 2021).

Notably, there are racial, ethnic, and sex differences in participation in SNAP and TANF. Women, particularly single mothers, and Black and Hispanic households average lower incomes and are more likely to be eligible for both means-tested programs. SNAP participation is higher among women; 58% of SNAP participants were female in FY 2019, and 26% and 16% of participants were Black and Hispanic, respectively (Cronquist, 2021). In FY 2020, TANF adult recipients were even more disproportionately female (85%), but the racial composition was somewhat similar to SNAP with 29% of TANF participants identifying as Black but a much higher 36% as Hispanic (US Department of Health & Human Services, 2021). In Virginia, the state we examine in this study, the racial composition of programs is slightly different, with Hispanic participation considerably lower.² In FY 2019, 42% of SNAP participants were Black and just 1.3% were Hispanic (Cronquist, 2021); similarly, 58% of TANF participants were Black and just 4.5% were Hispanic in FY 2020 (US Department of Health & Human Services, 2021). While workers connected to SNAP or TANF in Virginia are likely to be disproportionately Black and non-Hispanic than the broader state population, this population represents those likely most at-risk for economic vulnerability.

Importantly, however, much research on safety net program participants relies on small samples that prohibit an in-depth examination of racial, ethnic, and sex differences, or survey samples which may not accurately reflect program participation. For example, the reliability of self-reported household earnings data has increasingly come into question as estimates by Bollinger et al. (2019) suggest that non-random, non-response in the Current Population Survey (CPS) in the tails of the distribution may bias earning differentials by as much as 20% (Bollinger et al., 2019). Relatedly, it has been well-established that recipients of social programs, such as SNAP and TANF, often do not accurately respond to survey questions regarding program participation (Meyer et al., 2015). Recent estimates indicate that as many as one in two SNAP recipients inaccurately report not receiving program benefits in the previous year when responding to the CPS (Celhay et al., 2018; Meyer, Goerge, et al., 2018) with significant consequences for studies of poverty and social welfare receipt (Bollinger & David, 1997; Meyer & Mittag, 2019; Meyer, Mittag, et al., 2018). Further, even prior research using administrative data typically only includes information during periods of program participation. As a result, little is known about the income and employment swings leading up to program participation or following exit and thus little is known about patterns of economic instability among this economically vulnerable population or the racial and ethnic disparities in instability.

The Current Study

This descriptive study uses two administrative datasets—the first, representing the universe of workers covered by the UI system in a large, diverse state from 2015–2018, and the second, a sample of these covered workers who participated in SNAP or TANF in any month during the 2015–2018 period. This second sample, which we refer to as Public Welfare Connected workers, comprises 1 in 5 workers in the universe of UI-covered employed in Virginia during 2015–2018. In what we believe is a novel contribution of this study, we focus on an economically vulnerable sample that is defined according to their connection to income support programs, TANF and SNAP, rather than selecting our sample based on education level or income level. For this group, we can observe earnings and employment over the entire period, rather than only the period in which they are directly connected to these means-tested programs as does much research with administrative data. Our Public Welfare Connected sample represents 31% of the full sample earning below the 2015–2018 median, 36% of the full sample earning below the 30th percentile of earnings, and 63% of the full sample earning below the 10th percentile of earnings. (See appendix Table 2) It is also important to note that 83% of adults who receive SNAP or TANF are observed employed in the UI wage data system at some point in our observation period, indicating that most individuals connected to the public welfare system engage in UI-covered work. Thus, our study provides novel insight on the economic vulnerability of a group of workers that are frequently the target of federal and state policy. Finally, our intersectional analysis allows a fuller explanation of how the contours of economic vulnerability vary by the salient features of sex, race, and ethnicity.

Methods

Data

We use linked administrative data from the Virginia Department of Social Services (VDSS) and the Virginia Employment Commission (VEC) Unemployment Insurance Wage Data to document patterns in the levels and stability of earnings between 2015 and 2018. Virginia is the 12th largest state by population (8.5 million in 2019) and the 35th in geographic size, including urban, suburban, and rural areas. Virginia is comparably racially and ethnically diverse and slightly more economically advantaged relative to the broader United States, with per capita income of $39,278 (vs. $34,103) and 10% of the population in poverty (vs. 11%), and labor force participation rates among men and women comparable to the rest of the country (U.S. Census Bureau, 2021). Together, the state’s size, diversity, and economic context make it ideal for examining worker earning and job changes.

The VEC data contains the universe of quarterly earnings for all covered employment in the Commonwealth of Virginia at the individual level for each job worked within the quarter. For each calendar year, we link individual-level data for all jobs held in each of the four quarters (as well as the fourth quarter of the prior year) in order to calculate measures of quarterly earnings stability. All analyses presented using the full VEC data are at the person-quarter level or person-year level depending on the analysis; for the 2015–2018 observation period, we have data for a total of 4,286,608 person-year observations.

We then link the VEC quarterly wage record data with VDSS data that contains demographic information on sex and race/ethnicity for all participants in TANF and SNAP between 2015 and 2018.³ We call this our Public Welfare Connected sample. We limit the Public Welfare Connected sample to those age 18–64 to remove retired workers from the sample.⁴ Using demographic information contained in the DSS file, we stratify this sample by sex, race, and ethnicity to give a more detailed picture of economic vulnerability at the intersection of race and sex at the end of the last economic cycle.

In total, the Public Welfare Connected sample contains 1,392,901 unique individuals. Race and ethnicity are separate in-take questions in the DSS. We combine the two fields to create four mutually exclusive race/ethnicity categories (Non-Hispanic White, Non-Hispanic Black, Non-Hispanic Asian, and Hispanic). We omit from our analysis a racial category that is labeled in the administrative data as “Other” because of the heterogeneity in experience and the small sample size (n = 196,005). We also drop 220,251 individuals who were missing information either on sex, race, or ethnicity.⁵ Finally, we lose 166,923 individuals from the Public Welfare Connected sample who do not work in any quarter during the 2015–2018 period.

The remaining sample of 809,722 individuals in our Public Welfare Connected sample are distributed across sex, race/ethnic categories such that 14.6% are Non-Hispanic White males (hereafter White male), 19.7% are Non-Hispanic White females (hereafter White female), 13.4% are Non-Hispanic Black males (hereafter Black male), 20.9% are Non-Hispanic Black females (hereafter Black female), 3.9% are Hispanic males, 4.6% are Hispanic females, 13.1% are Non-Hispanic Asian males (Asian male), and 9.8% are Non-Hispanic Asian females (Asian female). It is important to note that the Public Welfare Connected sample includes members with zero earnings in some quarters in our analysis. We compare descriptive statistics for employed Public Welfare Connected (PWC) sample to that of the employed population for Virginia and the United States using 5-year estimates for 2014–2018 in Appendix Table 3. The Public Welfare Connected sample is comprised of fewer Non-Hispanic White male and female workers, similar Hispanic male and female workers, and a disproportionately higher level Non-Hispanic Black and Asian male and female workers when compared to all employed workers in Virginia or the United States.

Working with state administrative quarterly wage record data is not without limitations. First, we are limited to covered employment, inclusive of approximately 92% of all employed workers⁶ (U.S. Bureau of Labor Statistics, 2019). Uncovered workers include self-employed workers, independent contractors, undocumented immigrants, state and federal workers, and those working in the informal sector or the gig economy. In most states, the largest group of uncovered employment comprised of self-employed workers, who Bollinger et al. (2019) suggest are more likely to come from the bottom of the income distribution. However, the omission of federal workers from this analysis likely distributes this bias more evenly throughout the income distribution. As a consequence, our results may present an under-estimate of the level of economic vulnerability present in the full population.

Second, earnings data are reported quarterly as a single number, obscuring the wage and salary earned as well as the hours worked that produced the quarterly total. Third, quarterly wage data is reported without any demographic information for the worker, such as the race, sex, and age, data that are critical to social science research. We address this limitation by linking the quarterly wage data with DSS data that includes a robust set of demographic information which allows us to construct measures of economic vulnerability by sex and race/ethnicity. However, given that not all groups participate in social programs at the same rate due to stigma, lack of knowledge, and legal constraints (specifically for non-US born populations), in addition to state policy variation, this constrains the population to which our results by sex and race/ethnicity can be generalized. Our employed PWC sample is comprised of more Non-Hispanic Black and Asian workers (both male and female) and fewer Non-Hispanic White and Hispanic workers (both male and female) than in the United States (See Appendix Table 3). Fourth, despite our ability to link quarterly wage data with other DSS files, we still lack critical information on the occupation and industry of employment, fringe benefits received or flexibility of the work arrangement. Nonetheless, the use of administrative wage record data with individual-level longitudinal measures of quarterly earnings and jobs with the universe of covered workers and the public welfare connected sample provide a substantial contribution to the literature.

Measures

We operationalize economic vulnerability based on observed quarterly changes in earnings and the number of jobs. Throughout our analysis, quarterly earnings are adjusted for inflation and presented in constant dollars. In our samples of covered employment and disadvantaged workers who receive TANF and SNAP, we calculate a series of measures that capture different dimensions of economic vulnerability.

We begin by documenting differences in annual earnings (the sum of the earnings from all jobs held in the four quarters of a calendar year). Examining annual earnings provides a baseline measure of economic wellbeing and uses units (annual dollars earned) that are intuitively understood. However, it does not help us identify workers with low wages separately from those who are working less than full-time—or those for whom earnings instability is likely to be most difficult to weather. Therefore, we supplement this annual earnings measure by creating a measure to identify workers with low earnings in a quarter. We define a low earnings worker as one earning less than $2,610 per quarter (what a worker would earn if they worked at least 30 h per week earning the federal minimum wage of $7.25/h⁷). Our measure of low earnings worker likely contains some workers earning more than the minimum wage but working fewer hours per week as well as workers who worked full-time for part of the quarter but were not employed for several weeks. Consequently, this single measure over-estimates the extent that quarterly economic instability is present. To address this limitation, we supplement this indicator with additional measures.

We create a number of measures of earnings instability as assessed via changes in total earnings across quarters. First, to document the magnitude of the average earnings changes occurring quarter to quarter, we construct a measure of the standard deviation of the arc percent change (SDAPC) in quarterly earnings, or the average difference in the income between two time points relative to the mean value across the two time points (Dahl et al., 2011; McKinney & Abowd, 2020; Ziliak et al., 2011).⁸ The SDAPC is preferred relative to the percent change in earnings in cases when zero earnings are an issue, which is common among those connected to the public welfare system. Higher values of the SDAPC are interpreted as indicating higher levels of earnings volatility.

Second, we examine the direction of the quarterly earnings change relative to the initial earnings to determine if the change is greater than or less than 33% of the base earnings.⁹ We create four categories for this measure for each individual for each quarter that identifies the quarterly earnings increased by more than 33%, decreased by more than 33%, changed by 33% or less, or if no earnings were observed in two consecutive quarters. This measure provides important information both on the extent that the earnings instability reflects an increase in quarterly earnings versus a decrease, an important substantive difference with wholly different implications for household well-being. In addition, setting a quarterly earnings threshold of 33% ensures that the change is meaningful in terms of the financial resources available for consumption.

Third, we calculate the average frequency that the change in quarterly earnings exceeds 33% with each time-period. While coping with one sudden change in quarterly earnings of 33% or more might be difficult for a household to recover over a several-year period, doing so multiple times within a short period would likely make it quite difficult to plan or to try to smooth consumption. Note that we are unable to distinguish earnings changes from promotions or raises, demotions, or voluntary or involuntary shift changes. Of course, a quarterly earnings decrease one quarter followed by a quarterly earnings increase the following quarter might be a sign of a quarter in which a job loss occurred which was followed by a job gain (and conversely, an increase followed by a decrease may represent a job loss). For this reason, when the average frequency is above two this indicates that more than the typical level of economic instability that one might expect with a single employment disruption is present.

Finally, quarterly earnings instability can be due to changes in hours worked on the same job or a disruption in employment itself. We identify the extent to which there is a change in the number of jobs worked in a quarter with our fourth measure. We categorize each worker based on their number of jobs held in each quarter relative to the previous quarters as experiencing a job loss, job gain, no change in the number of jobs held or no jobs held in two consecutive quarters. This measure of quarterly job stability allows us to gain an understanding of how much of the earning instability is related to job losses and job gains based on the number of jobs from quarter to quarter. For the sake of this study focused on economic instability, the circumstances producing the job disruption are less important than the fact that the job disruption occurred.

Taken together, our two measures of average earnings (average annual earnings and low earnings) and these four measures of quarterly earnings instability provide a holistic picture of the economic situation experienced among our different samples and over time.

Analyses

We conduct a series of descriptive analyses to examine our measures of economic instability (at the annual and quarterly levels) in the broader VEC sample and the Public Welfare Connected sample. Further, within the Public Welfare Connected sample, we compare differences across sex and different racial/ethnic groups. All analyses presented are for the 2015–2018 period, a time of strong economic growth that precedes the COVID-19 economic disruptions.

While previous studies of income volatility have tended to focus on changes in household income (Dahl et al., 2011), following Ziliak et al. (2011), we focus on individual earnings. One benefit of focusing on individual-level administrative data is our ability to link earnings and job changes to individual characteristics, including sex, race, and ethnicity, analyses vital to shedding light on patterns of inequality. Notably, though, our focus on the individual level may either underestimate or overestimate household economic instability. For example, in a household with multiple individuals employed, an earnings decrease for one individual may lead to another household member to increase their own number of work shifts and earnings, smoothing household level earnings. Alternatively, observing just one individual’s job and earnings changes misses other household members who may be experiencing job and earnings changes.

Results

Earnings Distribution

Figure 1 presents the distribution of annual earnings for the full sample of workers in covered employment in the VEC data and the Public Welfare Connected sample for the 2015–2018 time-period; horizontal dotted lines show the average for each population. As one might expect, the earnings distribution for the full VEC sample had a higher mean and right-skewed relative to the PWC sample.

Fig. 1

Total Annual Real Earnings among all covered workers and the Public Welfare Connected sample, 2015–2018. Authors’ calculations using the Virginal Employment Commission (VEC) administrative data at the individual-quarter level. N = 17,146,432 person-year observations for the full VEC sample and 3,691,011 person-year observations for the Public Welfare Connected sample. Only nonzero earnings are shown. The 2018 VEC sample is trimmed at the 99th percentile. Dashed vertical lines show the median of the distribution

Figure 2 shows average annual earnings for 2015–2018 among the public welfare connected (PWC) sample by sex and race and ethnicity. All groups experienced a slight increase in annual earnings from 2015 to 2017 and a sharp increase in real wages from 2017 to 2018. Relative rankings of earnings remained relatively stable, with Asian female workers consistently having the highest earnings in each year followed by Asian and Hispanic male workers. Next, Hispanic female workers fell into the middle of the pack and just above White male workers. Finally, White female, Black female, and Black male workers consistently earned the least, relative to the other groups examined. Note that the low levels of annual earnings in our PWC sample, even in 2018 when average real annual earnings were highest, suggest that the average White female and Black male worker in our PWC sample living in a single-person household were earning just above the poverty line of $9,800 (in 2006 dollars) and that Black female, Hispanic female, and Hispanic male workers were earning around 200% of the poverty line for a single-person household.

Fig. 2

Average real annual earnings by sex, race, and ethnicity, Public Welfare Connected sample, 2015–2018 (N = 3,691,011). Authors’ calculations using the Virginia Employment Commission administrative data at the individual-year level. All racial differences shown are statistically significant at the p < .0001 level

Low Earnings Workers

One way to consider the size of the earnings differentials is to examine the average share of the sample estimated to have low earnings each quarter, those whose total earnings were less than $2610 per quarter (indicating that they worked less than 30 h weekly at Virginia’s minimum wage). See Table 1. Among the full VEC sample, about 18% meet our definition of a low earnings worker, on average, over the 2015–2018 time period. For the PWC sample, the share was 21%, about 17% higher than in the full sample. Among the full sample, approximately 65% earned enough to escape this classification whereas among the PWC sample, only 44% earned enough to not be classified as having low earnings during the 2015–2018 period.¹⁰

Table 1 Quarterly earnings instability measures, 2015–2018 period

Figure 3 presents the low earnings worker measure stratified by sex and race/ethnicity. Black and White female workers in the PWC sample were much more likely to be low earnings workers (28% and 26%, respectively) than were Hispanic and Asian female workers (22% and 15%, respectively). Asian female workers stand out as being the most likely (62%) to have been employed and earning more than our cut-off—relative to Hispanic female (50%), Black female (40%) and White female workers (34%). On this measure, White female workers are the most disadvantaged group in that they are the group of female workers most likely to be not employed for a quarter and the group of women least likely to have earnings above our low earnings worker threshold.

Fig. 3

Rates of low earnings workers by race and ethnicity for females and males in the Public Welfare Connect Sample, 2015–2018. Authors’ calculations using the Virginia Employment Commission administrative data linked with Department of Social Service data at the individual-quarter level. A low earnings worker is defined as one earning less than $2610 per quarter (what a worker would earn if they worked at least 30 h per week earning the minimum wage of $7.25/h). All racial differences shown are statistically significant at the p < .0001 level. Chi-square statistics indicate that racial and ethnic differences shown for each sex are statistically significant at the p < .001 level

Many fewer White and Hispanic male workers met our definition of low earnings worker relative to female workers, which may be due to higher wages, greater work hours, or both. Asian and Black male workers, however, had quarterly earnings that resulted in us classifying them as a low earnings worker at levels that are similar to Asian and Black female workers, respectively. However, while Asian male workers were particularly advantaged on this measure, Black male workers appeared the most disadvantaged as they are the group of men least likely to have had earnings higher than our low earnings worker cut-off and also had particularly high levels of quarterly labor force nonparticipation.

The Magnitude of Earnings Instability

Table 1 shows the mean and standard deviation of the APC (SDAPC) in quarterly earnings in constant 2006 dollars for the 2015–2018 period, with higher levels indicating more dispersion. The mean difference in the APC is twice as high in the PWC sample as it is in the full VEC sample, reflecting more frequent changes from zero earnings quarters, while the standard deviation differences are much smaller. We focus here on the standard deviation (SDAPC) because, following previous literature (Gennetian et al., 2018; Hardy & Ziliak, 2014), the APC is a ratio of the difference between two points in time to the average of those two points, essentially a percent change bounded by −200 and 200. Importantly, the starting point does not change, such that the SDAPC does not confound differences in means with differences in variation. Across all covered employment in Virginia, our full sample, we find levels of quarterly earnings instability around 86 SDAPC, while in the PWC sample it was 92. But, when we stratify the PWC sample by sex and race/ethnicity and calculate the average SDAPC in quarterly earnings, important differences appear. See Fig. 4. During the 2015–2018 period, White and Black female workers experienced high levels of quarterly earnings SDAPC measures (97 and 95, respectively) while Hispanic and Asian female workers experienced lower levels (89 and 72, respectively). All groups of men experienced higher levels of earnings instability than did the women: Black men had the highest levels of earnings instability (109), followed by White men (100), with Hispanic men and Asian men having much lower levels of earnings instability (95 and 89, respectively).

Fig. 4

Standard deviation of the Arc Percent Change (SD APC) in quarterly earnings by race and ethnicity for females and males in the Public Welfare Connected Sample, 2015–2018. Authors’ calculations using the Virginia Employment Commission administrative data at the individual-quarter level of the standard deviation in the Arc Percent Change in earnings from one quarter to the prior one. Chi-square statistics indicate that racial and ethnic differences shown for each sex are statistically significant at the p < .001 level

High levels of earnings instability indicates that economic resources are fluctuating widely from quarter to quarter, but this could be overall positive if earnings gains dominate earnings losses. The SDAPC is neutral regarding the direction of the change in the earnings from quarter to quarter so one explanation for the high levels of earnings instability is that workers were experiencing earnings gains or increases in hours during this period, whereas low levels of earnings volatility may indicate stagnant wages. We examine these differences with additional measures below.

Job Stability

Table 1 presents changes in job stability for the full sample in covered employment and the PWC sample from 2015 to 2018. For the full sample, we find that 75% of workers were stably employed at the same number of jobs each quarter. The share of all workers with an observable job loss was 6%, the share of all workers with an observable job gain was 7%, and the share not at work in any job was 13%. The PWC sample was marked by less job stability relative to the full sample: the share stably employed in two quarters was 55%, fully 20 percentage points lower than the full sample. However, the share who experienced a job gain or job loss were comparable to the full sample (6% each) and the real difference in the two samples is that the public welfare connected sample had nearly three times as many sample members with no job in two consecutive quarters (34% relative to 13% for the full sample). Given that this sample is defined as being connected to the public welfare system at some point during the 2015–2018 period, and that program eligibility is a function of household income, this is not all that surprising. But put another way, the strong connection to the labor market among a PWC sample underscores the extent to which this sample is defined by economic vulnerability primarily and not a lack of attachment to the labor market; economic vulnerability is occurring despite their attachment to the labor force.

Figure 5 presents the proportion of the female and male PWC sample that experienced a job loss, job gain, or remained working at the same number of jobs in different quarters. Within each racial and ethnic group, the percentage of workers who experienced either a job loss or job gain from one quarter to another followed similar patterns for male and female workers. In contrast, job stability was much higher among female workers and particularly high among Asian and Hispanic females (69% and 62%, respectively) relative to Black females (57%) and White female workers (49%). Having no job was more common among Asian, Black, and Hispanic male workers relative to female workers of the same race and while persistent non employment is associated with a measure of economic stability, it is problematic for household income.

Fig. 5

Rates of quarterly job gains and losses by race and ethnicity for females and males in the Public Welfare Connected Sample, 2015–2018. Authors’ calculations using the Virginia Employment Commission administrative data at the individual-quarter level. Job losses are calculated based on a change in the number of jobs held from one quarter to the base quarter. Chi-square statistics indicate that racial and ethnic differences shown for each sex are statistically significant at the p < .001 level

The Direction of Earnings Instability

Table 1 shows the percentage of those in covered employment and the PWC sample by those experiencing a quarterly earnings loss or gain greater than 33%; those employed workers who did not experience a change in quarterly earnings that rises to that magnitude; and those with no earnings in both quarters. Among the full sample, about 14% experienced a quarterly earnings loss greater than 33%, about 18% experienced earnings gain greater than 33%, about 56% experienced earnings that remained relatively stable across quarters and 12% were not employed in two consecutive quarters. In terms of the frequency of quarterly earnings shocks greater than 33% (in either direction), we find that the average for the total sample was 2.3, consistent with the finding that job losses were likely to be followed by job gains (and presumably, job gains followed by losses).

While the share of workers with earnings gains and losses was comparable with that of the full sample, the share of workers with stable quarterly earnings was much lower–about 42%–and the share of not employed was much greater (28%) in the public welfare connected sample. In addition, the average number of quarterly earnings shocks (either increases or decreases) experienced over our observation period that were 33% or greater was almost twice as high among the PWC sample than the full sample (4.0 versus 2.3). We estimate that the average PWC sample person experienced 4 such shocks (losses or gains) to quarterly earnings greater than 33% during the 2015–2018 period.¹¹ Given that job gains and losses were not also twice as high among the PWC sample, the higher frequency of significant earnings shocks was likely to be the result of significant changes in wages or hours from the same job for a large portion of the PWC sample. Keeping in mind the lower earnings profile for the PWC connected sample and that family resource changes, either positive or negative, can be disruptive to household routines and finances, these significant changes in quarterly earnings were likely quite disruptive to family incomes.¹²

Figure 6 presents results on the direction of economic shocks stratified by sex, race, and ethnicity for the PWC sample, beginning with results for females. We find that the share of female workers who experienced a quarterly earnings loss greater than 33% was similar among White (13%), Black (14%), and Hispanic (13%) women but lower for Asian women (10%). Additionally, stable earnings were more common among Asian (56%) and Hispanic women (46%) than among White (36%) and Black (40%) women. In contrast, the share with a quarterly increase was highest among White, Hispanic, and Black women (16%, 19%, and 18, respectively) and lower for Asian women (14%).

Fig. 6

Earnings gains and losses by race and ethnicity for females and males in the Public Welfare Connected Sample, 2015–2018. Note: Authors’ calculations using the Virginia Employment Commission administrative data at the individual-quarter level. Chi-square statistics indicate that racial and ethnic differences shown for each sex are statistically significant at the p < .001 level

The bottom panel of Fig. 6 presents the same set of results for male workers. The patterns are generally consistent with those presented for females as for White males, but important differences exist for the other three groups. Generally, Asian men in the PWC sample had worse labor market outcomes than Asian women; Asian men were more likely than Asian women to have experienced a large drop in quarterly earnings, less likely to have experienced stable earnings, and more likely to have no earnings relative to Asian women. In contrast, the magnitude of quarterly earnings changes was similar across male and female White, Black, and Hispanic workers in the PWC sample, although it appears that Black female workers were more likely to have earnings stability than Black male workers.

Figure 7 presents the number of quarterly earnings shocks greater than 33% in order to capture the frequency of large changes, another dimension of economic instability. The mean number of large quarterly earnings shocks experienced over the 2015–2018 period was 5 for Black men and women, 4 for White and Hispanic men and women, and 3 for Asian women but 4 for Asian men. Standard deviations were somewhat higher among Black, White, and Asian men than their female or Hispanic counterparts, suggesting that there may have been slightly greater variability in these groups.¹³ The high number of earnings shocks among Black workers may be partially explained by their higher rates of job gains and losses, as shown in Fig. 5: 13% of Black females and 15% of Black males experienced a job loss or gain from quarter to quarter, on average, from 2015–2018, higher rates relative to their peers.

Fig. 7

Mean number of shocks in quarterly earnings by race and ethnicity for females and males in the Public Welfare Connected Sample, 2015–2018. Authors’ calculations using the Virginia Employment Commission administrative data at the individual-quarter level. Quarterly earnings changes must be at least 33% greater than the base quarter in either direction. Chi-square statistics indicate that racial and ethnic differences shown for each sex are statistically significant at the p < .001 level

Discussion

We use recent state administrative data from Virginia to describe overall patterns of workers’ experiences in instability in earnings and jobs, and then to examine a sample of workers connected to the public welfare system to examine how these patterns vary by socioeconomic status, sex, and race/ethnicity. Our findings suggest that economic instability was quite widespread among workers, even during a recent period of economic growth. Among our sample of public welfare connected workers, nearly half experienced a quarter-to-quarter change in income of more than 33%. As expected, frequent, large earnings shocks and job changes were more common among individuals who are connected to public assistance programs than the full sample of workers, confirming our first and second research hypotheses. Further, we find divergent economic patterns that suggest the salience of sex and race/ethnicity for economic outcomes in the United States, confirming our third research hypothesis. Specifically, we find that sex, race, and ethnicity interact to produced very different labor market outcomes even during period of strong economic growth. Large and recurrent fluctuations in earnings, particularly the high rate experienced by Black men and women, suggest the very different labor market experiences across individuals of different racial and ethnic backgrounds.

Findings indicate that, consistent with other research (Morris et al., 2015; Wolf & Morrissey, 2017), economically vulnerable workers showed higher levels of earnings and job instability than the broader working population. Overall, across all covered employment in Virginia, we find levels of quarterly earnings instability around 86%, and as expected, in the sample participating in public assistance programs, this figure was about 7% higher. Notably, our unique data provide for a full picture of earnings instability among this sample, not limited to the periods during which they are participating in public assistance programs. To put these findings in context, recent evidence by McKinney and Abowd (2020) finds that the SDAPC for 2016 were around 64% for all males aged 25–59,¹⁴ and that while earnings volatility has been declining or is stable, excluding recessions, significant heterogeneity exists in the experience of earnings instability. They estimate that in 2016, as many as 67% of workers experienced very high earnings stability and regular earnings increases and that most of the earnings volatility was attributed to a smaller set of workers who experienced large levels of earnings instability characterized by large decreases in earnings (McKinney & Abowd, 2020).

Our estimates indicate that on average, our PWC sample experienced 4.0 shocks or quarterly changes in earnings of 33% or more (losses or gains) from 2015 to 2018, nearly twice the mean for the entire covered worker sample, which was still sizable at 2.3 shocks. That is, workers connected to public welfare programs were experiencing high levels of economic instability during a period of strong economic growth—well before entering 2020, a year of unprecedented job loss and economic turmoil. Notably, too, because our sample includes only those workers covered by the state’s unemployment insurance system, it excludes self-employed workers and those in the informal market, both of which are likely to have higher rates of unstable earnings. Thus, despite the high levels we identify, it is likely that our findings underestimate the prevalence of economic instability.

Importantly, as the COVID-19 pandemic and its economic downturn have disproportionately harmed the economic outcomes of female workers and workers of color (Hooper et al., 2020; Karageorge, 2020), our findings suggest that these groups faced the most precarious economic situation prior to the COVID-19-induced economic crisis. Building on previous research identifying racial and ethnic—but not gender—differences in work schedule variability (Mccrate, 2021), we find that Black males and females had the highest frequency of quarterly earnings instability and job changes across all racial and ethnic groups we examined. We also find, consistent with previous research documenting race gaps (Greenman & Xie, 2008), that Black females and males earn lower average wages over time relative to their peers. Falling or unstable earnings is more problematic for groups that already were starting from a place of low earnings relative to other groups, such as Black males and females. Indeed, whereas White women had lower average earnings, Black men experienced the greatest level of earnings instability. In contrast, White women showed the greatest level of stability, but with greater levels of labor force nonparticipation and lower average earnings. Our data cannot tease apart whether instability in earnings is attributed to fluctuations in wages, fluctuations in hours, or a combination, from quarter to quarter.

Further, the well-documented racial and ethnic differences in wealth, income, and access to credit (Bandelj & Grigoryeva, 2021; Baradaran, 2017; Gibson-Davis et al., 2020; McIntosh et al., 2020; Perry, 2020) suggest that racial and ethnic minority families have fewer buffers for adjusting to swings in household resources. For example, homeownership represents a major asset that may buffer against swings in income or employment, but whereas more than 70% of White Americans owned homes in 2019, fewer than half of Black or Hispanic Americans did (U.S. Census Bureau, 2020b), and Black-owned homes tend to average lower values (Perry, 2020; Sedo & Kossoudji, 2004). Further, individuals of color average lower educational attainment (22% of Black and 17% of Hispanic individuals had a BA degree or higher in 2018, compared to 36% of White individuals (U.S. Census Bureau, 2020a), hold higher average student loan debt (Scott-Clayton & Li, 2016), and face discrimination in the job market and elsewhere (Bertrand & Mullainathan, 2004; NPR, 2017), compounding their ability to attain financial security. Together, our results suggest that Black individuals participating in public assistance show both low earnings and high levels of earning instability relative to peers, creating a double layer of economic disadvantage.

More work is needed to shed light on the economic instability patterns among Latinx or Hispanic individuals and those of Asian descent, and that explicitly examines how public benefits smooth or exacerbate household resource instability. Previous research using nationally representative survey data finds that Hispanic children are more likely to live in lower-income but more economically stable households, and that social program participation does little to improve economic stability (Gennetian et al., 2019). In our public welfare connected sample, Hispanic and Asian male workers had similar levels of earnings and measures of instability, and Hispanic women had higher average earnings and lower measures of instability. This is different from previous research finding Hispanic individuals experience more employer-driven variable work schedules than Black or White workers (Mccrate, 2021). However, it is possible that our data undercount Hispanic workers because Hispanic individuals are disproportionately low-income and to be recent immigrants (Budiman et al., 2020; Creamer, 2020), and thus often ineligible for social programs. This is one of the first studies to distinguish workers with Asian backgrounds, and we find relative levels of stability in earnings and employment and relatively higher average levels of both earnings and employment, compared to their peers also participating in public assistance. Future research should explore the implications of this relative stability and the role that racial, ethnic, and immigrant selection into public assistance programs plays in these figures.

The growth of and disparities in the experience of economic instability are concerning given the research demonstrates the importance of economic resources (Brooks-Gunn & Duncan, 2000; Duncan et al., 2011; McLoyd, 1998) and systemic racism and discrimination (Braveman et al., 2010, 2011; Karlsen & Nazroo, 2002; Williams & Jackson, 2005) for health, educational, and economic outcomes, and the experience of economic instability may help explain inequalities beyond levels alone. Given the growing racial and ethnic diversity of Americans (U.S. Census Bureau, 2019), and of the higher rates of single-mother households, particularly among racial and ethnic minority families (Livingston, 2018), investigating the prevalence of economic instability by sex and race/ethnicity is particularly important in understanding the implications of today’s economic context.

Policy Implications and Limitations

Results have implications for the types of and targeting of economic relief and investments to promote economic stability and equity following the pandemic-induced economic crisis; notably, that the promotion of economic stability may require different policies or interventions for different groups. For example, increasing the level of earnings may well be accomplished by an increase in the minimum wage for some workers or some geographic areas. Indeed, In May 2021, the minimum wage in Virginia increased to $9.50, and is set to increase incrementally to $15.00 per hour in January 2026.¹⁵ Public assistance programs, typically intended to temporarily smooth household resources, may introduce additional instability (Hardy, 2017; Morrissey et al., 2020), and the replacement rate is likely to be low in states such as Virginia given the low SNAP and TANF benefit amounts and the limited eligibility for both programs for men who are able-bodied and without dependent children. Program designs that reduce administrative burden and smooth benefit “cliffs” may help stabilize individual and household resources (Davis et al., 2017; Herd & Moynihan, 2018) and improve family well-being, although more research is needed. Likewise, state unemployment insurance provisions that require high levels of quarterly earnings or longer durations of employment to qualify for Uninsurance Benefits may disqualify a disproportionate number of Black men and women. Indeed, given the recent large drop in female labor force participation (Albanesi & Kim, 2021; U.S. Bureau of Labor Statistics, 2021), the limits of an employment-based safety net are going to fall particularly heavily upon women of color.

Results also have implications for economically vulnerable families with children. Finding and affording child care is difficult, but regulated child care programs that offer care during non-standard hours or that accommodate shifting schedules are few and far between (Chaudry et al., 2021). Given that stability in care environments and in household resources and routines is important to children’s development (Duncan et al., 2011; Sandstrom & Huerta, 2013), this economic vulnerability and instability—and the disparities in these experiences—have implications for families struggling to meet their employment and child caregiving demands as well as for the design of public child care subsidies to accommodate these frequent changes. In particular, subsidy systems that require specific documentation of minimum weekly hours worked or specific incomes to certify or recertify eligibility may reduce participation by those with high levels of job or earnings instability, who are more likely to be Black.

While our data span a period of economic growth prior to the COVID-19 pandemic, results are relevant to the period of the pandemic and its recovery. Notably, we find that the groups in the most difficult economic situation—Black female and male workers—entered the pandemic-induced economic crisis with, on average, lower and more volatile earnings, and presumably fewer assets to weather the downturn. Moreover, emerging research indicates that these groups were then also disproportionately harmed by the pandemic and its fallout. Economic vulnerability, particularly among lower-income families and individuals of color, increased quickly and dramatically. Of workers who had been employed in October 2019, 22% and 23% of Black and Hispanic workers were laid off as of July 2020 compared to 18% of White workers (Board of Governors, 2020). The higher job loss in the service sector, school closures, and the child care crisis disproportionately affected women of color, and it is expected that this time out of the labor force will have cascading and long-term effects on their and their families’ financial security (Karageorge, 2020; Malik & Morrissey, 2020). Our findings suggest that research and policies attend to racial, ethnic, and sex differences in the recovery from the pandemic, and even in strong economic times.

Several limitations of this analysis should be noted. First, while quarterly wage record data cover the vast majority of employed individuals, the self-employed, independent contractors, state or federal workers and those working “off the books” or illegally are not included. Second, as noted above, we focus on individual-level economic instability, which allows us to investigate the intersection of sex and race/ethnicity at the individual level, but we lack a full picture of economic resources or changes at the household level. Third, we only have demographic information on the sample of individuals that participated in SNAP or TANF between 2015 and 2018. While SNAP and TANF participants are low-income by definition, there are many low-income workers who do not receive these programs for a variety of reasons including social stigma, lack of eligibility due to immigration status (particularly among Hispanics and Asians), and lack of knowledge of eligibility. Nonetheless, our PWC sample includes approximately one of every five workers in Virginia, highlighting the reach of and need for the SNAP and TANF programs among the working population. Further, this population of low-income individuals intermittently eligible for and participating in public welfare programs, while also connected to employment, is arguably the population most of interest for those designing or administering public programs or considering low-wage worker conditions and protections. Second, while our administrative dataset is rich in terms of the coverage of workers, it lacks critical information such as the number of hours worked, the occupation and industry of employment, fringe benefits received, or flexibility of the work arrangement.

Conclusion

This study contributes to our understanding of earnings and job instability among workers by public welfare connection history, sex, and race/ethnicity using a longitudinal, administrative database representing the universe of workers covered by unemployment insurance in a large, diverse state. Prior research suggests that both levels in earnings and instability in levels may cumulatively interact to affect measures of economic well-being, (Brown et al., 2021; Miller & Votruba-Drzal, 2017). A next step for future research is to examine the implications of racial, ethnic, and sex disparities in economic instability for long-term health, educational, and other outcomes. More research on how the frequency, magnitude, direction, and predictability of changes in economic resources affect family functioning and children’s outcomes can shed light on targets for policy interventions that smooth and bolster household resources and enhance equity in economic opportunity.

Appendix 1

See Tables

Table 2 Share of covered employment at different points in the distribution composed of public welfare connected sample

Table 3 Comparison of race, ethnicity and sex distribution of employed sample

Table 4 Sensitivity analysis using different earnings loss cut points, overall

Table 5 Sensitivity analysis: direction of quarterly earnings

Table 6 Sensitivity analysis: average number of earnings shocks

2, 3, 4, 5, 6