Underemployment refers to when very few minorities are able to find a job.
|
High underemployment has been a chronic structural feature of the rural United States for decades. In this paper, we assess whether and how inequalities in underemployment between metropolitan (metro) and nonmetropolitan (nonmetro) areas have changed over the course of the last five decades. Drawing on data from the March Current Population Survey (CPS) from 1968 to 2017, we analyze inequality in the prevalence of underemployment between metro and nonmetro areas of the United States, paying special attention to differences between white, black, and Hispanic workers. Our results show that the underlying risk of underemployment has increased in both metro and nonmetro areas over the last fifty years. Nonmetro workers have consistently faced greater employment hardship compared to their metro counterparts, and these differences cannot be fully explained by differences in population characteristics. Nonmetro ethnoracial minorities have experienced particularly poor labor market outcomes. The disadvantage of ethnoracial minority status and rural residence is especially pronounced for nonmetro black workers, among whom underemployment has remained persistently high with only modest convergence with other workers. Hispanic workers also face elevated risk of underemployment, but we observe a unique convergence between metro and nonmetro workers within this population. Show
Fifty years after President Johnson declared a War on Poverty in the United States, renewed attention is being paid to the legacy of that initiative (e.g., ). Less well-known than the broader War on Poverty is a 1967 publication authored by the President’s National Advisory Commission on Rural Poverty, entitled The People Left Behind. This report documented the especially high and persistent poverty that afflicted much of rural America, a pressing problem that the Commission argued was overshadowed by the more visible economic troubles of the nation’s inner-cities. One dimension of the poverty problem noted in the The People Left Behind was that “unemployment and underemployment are major problems in rural America” (1967: x) and that rural racial and ethnic minorities “suffer even more than low income whites from unemployment and underemployment” (1967: 5). The implication was that rural labor markets systematically failed to provide the employment opportunities needed to escape economic hardship, and that ethnoracial minorities in rural areas faced particularly steep barriers to gainful employment., Although the decade after the publication of The People Left Behind was characterized as one in which “remarkably few studies of rural labor markets exist” (: 53; for an exception, see ), employment hardship in rural areas began to capture greater attention among scholars in subsequent years (e.g., ), leading to seminal work on rural underemployment in the 1980s (; ). This paper returns to these themes by examining trends in rural underemployment and ethnoracial inequality therein, in the context of a half-century of social, economic, and demographic change. In doing so, we contribute to the literature on rural underemployment that has emerged over the intervening fifty years (e.g., ; ; ; Lichter, Johnston, and McLaughlin 1994; ; ; ; Thiede, Lichter, and Slack 2018; ). Our analysis perhaps builds most directly on study of ethnoracial and residential disparities in underemployment from 1968 to 1998. We extend this earlier research in at least two primary ways. First, we expand the temporal scope of the study by a full two decades, which witnessed major economic and demographic changes, including the dotcom boom-and-bust of the 1990s, the emergence of new rural immigrant destinations, the Great Recession of 2007–2009, and continued deindustrialization. Second, we use an alternative set of methods to more precisely identify trends in the probability of underemployment across rural and urban areas and between ethnoracial groups, and to assess the relative contribution of regional and demographic changes to such patterns. This approach enables us to assess how historical shifts in the regional concentration of workers and the demographic composition of the labor force have contributed to changes in ethnoracial and residential disparities in underemployment over time. On one hand, real gains in rural standards of living and socioeconomic progress among ethnoracial minorities provide reasons to expect that the metro-nonmetro gap in underemployment may have narrowed over time. On the other hand, uneven impacts of demographic and economic change, and related processes of selection into and out of the places classified as nonmetro, provide reasons to believe that the workers who remain in nonmetro America may be being left even further behind. On balance, we expect that the disadvantages of rural residence in an urbanizing economy and the long and difficult history of ethnoracial minorities in the rural U.S. will persist as significant axes of inequality in underemployment, even after controlling for the changing demographic characteristics of the labor force. BackgroundEmployment Hardship in the Rural United StatesComparatively high underemployment has been a longstanding feature of the U.S. countryside, reflecting both the structure of rural economies and the characteristics of the rural workforce (; ; ; ). Nonmetro labor markets exhibit less diverse employment prospects compared to metro areas, owing in part to geographic isolation, lower population density, reliance on place-specific natural resources, and forces that discourage agglomeration economies (i.e., the benefits that accrue to firms by locating near to one another in cities or industrial clusters) (). According to the USDA’s county economic typology, less than 30 percent of nonmetro counties are classified as economically diverse compared to a majority of metro counties (). A lack of economic diversity in local labor markets creates vulnerability for workers, whose job prospects are tied to a single industry. In many cases, this vulnerability has been compounded by rural communities’ dependence on volatile sectors, such as agriculture, mining, and non-durable manufacturing. It is also the case that nonmetro areas tend to be characterized by both lower average educational levels among the population and lesser returns to human capital endowments for workers (). The relatively low levels of human capital in rural areas are in turn reinforced by the selective out-migration of more highly educated residents to other—often metro—labor markets where workers’ investments in education are better rewarded (; ). Many of the macroeconomic changes that have resulted in more precarious employment for U.S. workers over the last 50 years are also inherently spatial and have tended to disadvantage rural areas. Deindustrialization, policy regimes for collective bargaining, and the devolution of social welfare programs all operate in ways that are spatially-unequal in terms of process and impact (). Take for example a factory closure in a small “company town” versus a diversified metropolis; in the former the closure stands to decimate the local economy, while in the latter the economic impact is likely to be more readily absorbed as other firms continue to operate. Even in cases where the quantity of jobs does not change significantly, declines in the quality of jobs may disproportionately affect rural areas. Furthermore, nonstandard employment—the temporary, contract, and on-call jobs that characterize an era of precarious work—has been shown to be more common in nonmetro than metro contexts (). Indeed, the negative impacts of industrial change on the livelihoods of people in rural and small-town America has recently been highlighted in the mass media narrative regarding the 2016 election of President Trump. While research suggests that Trump’s rural advantage was not sufficient to swing the election, there is evidence that he performed better in rural areas characterized by greater economic distress (). In short, despite the sociological literature’s emphasis on the impacts of deindustrialization and economic restructuring in urban settings, research makes clear that rural America has also been fundamentally, and perhaps especially, affected by these changes (e.g., ). Race, Ethnicity, and Economic Status in the Rural United StatesWhile popular accounts often portray rural America as racially and culturally homogeneous, it is in fact characterized by considerable diversity (; ; ; ), and ethnoracial inequality in rural communities has long been a salient social issue generally and in terms of underemployment specifically (). In his presidential address to the Rural Sociological Society (RSS) nearly three decades ago (1990), Gene Summers challenged rural sociologists to make persistent rural poverty, and its particular burden among rural minorities, a disciplinary priority. He stated: “The challenge to rural sociologists is to identify the barriers that prevent rural minorities from claiming their rights of citizenship in this prosperous land. As citizens they have a right to adequate education, sufficient income, decent housing, adequate health care, and full employment” (: 179, emphasis added). Summers’ RSS presidency gave rise to a task force on rural poverty, which included a working group focused on the status of rural ethnoracial minorities. The working group not only found evidence that these populations faced a “double jeopardy” of being rural and nonwhite, but also concluded that the social science literature on rural poverty was fundamentally limited by a “failure to address the position of minority communities within the larger social and economic matrix of the dominant society” (: 184). The history of black people in what is now the United States predates the inception of the nation by over 150 years. Millions of West and Central African people were forcibly transported to the American colonies, and later the U.S., as slave labor to be utilized in a range of industries, most notably plantation agriculture in the rural U.S. South. The institution of slavery stood not only as a system of brutal servitude, but of racial domination undergirded by the ideology of white supremacy (). From the antebellum period through the mid-twentieth century blacks were subject to overt systemic racial oppression, moving beyond slavery to sharecropping, Jim Crow, and other forms of institutionalized discrimination and segregation. In the second half of the twentieth century, however, these traditional and explicit barriers to full participation for black people in the economy and society began to be dismantled due to the victories of the Civil Rights Movement (). Despite broad social gains over subsequent decades that allowed for the emergence of a substantial black middle-class, a subset of more disadvantaged blacks remained significantly socially and economically marginalized (, ). Major accounts of the reasons for contemporary black-white economic disparities include urban segregation, deindustrialization, and related spatial mismatches in skills and jobs that have resulted in pernicious economic impacts for a portion of the black population, less educated urban black men in particular (; , ). Moreover, research has shown the uniquely disadvantaged position of African Americans in the U.S. labor market is significantly understated, because conventional labor force estimates fail to account for mass incarceration, a trend that has removed an ever-growing number of economically marginal black men from the labor market (; ). Importantly, the vast majority of research on the economic status of black Americans is characterized by a decidedly urban focus, despite evidence that rural blacks have historically experienced even greater economic hardship than their urban counterparts (; ; ; ). Rural African Americans are substantially more likely than those in urban areas to be underemployed () and to experience high and concentrated poverty (; Thiede, Kim, and Valasik 2018). Moreover, there is an unambiguous regional consideration for rural blacks; dating from slavery to present, “the presence of African Americans in rural areas is an overwhelmingly southern phenomenon” (: 441; see also ). As of 2000, the black poverty rate in the South remained nearly 170 percent higher than among whites in the region, a marked level of contemporary inequality linked to the historical concentration of slavery (). The history of Hispanics in the United States also predates the birth of the country, in this case by over 250 years (). Occupying and settling in today’s Southeast and Southwest, the ancestors of many Hispanics became “American” through political annexation and conquest. Others arrived later through immigration. As such, “Hispanics are both an indigenous and an immigrant community” (: 20). Although bound by a common language retained by many, the diversity of experiences and countries of origin evident in this group has called into question the validity of the “Hispanic” label itself (). With immediate origins in some 20 Spanish-speaking countries (), the Hispanic population in the United States is diverse. For simplicity, historical treatments often focus on the experiences of Mexicans, Puerto Ricans, and Cubans (). The settlement patterns of these subgroups are regionally distinct. Those of Mexican origin historically concentrated in the Southwest, proximate to Mexico. Once the U.S. expanded to the Pacific, immigration from Mexico continued to provide labor for the growing economies of the Southwest. These flows were halting, however, with the doors open amidst labor shortages and closed when jobs were scarce (e.g., the Great Depression). Puerto Rican migration to the United States, eased by its status as a semi-autonomous commonwealth whose residents enjoy U.S. citizenship, increased substantially after World War II, with the majority moving to New York City and other metro areas in the Northeast (). Immigrants from Cuba cluster in Miami and elsewhere in Florida (). Although fewer in number, immigration from the Caribbean also adds to the Hispanic mosaic in the United States. For example, Dominicans comprise a sizable enclave in New York City (). With the exception of Mexicans in the Southwest, historically, none of the major Hispanic subgroups have been prominent in rural and small-town America. However, in recent decades through both the internal migration of Hispanics and the emergence of new gateway destinations for newcomers, the growth of the nonmetro Hispanic population has been sizable (; ; ; ). New Hispanic destinations have cropped up across the rural landscape in the Southeast, Midwest, and Northwest. The economic draws are, for example, in timber and poultry processing in the Southeast, amenity-based employment in the Mountain West, and meatpacking and ancillary business in the Plains and Midwest heartland (; ; ). In these places Hispanic arrivals have been greeted by a sometimes wary native population, and the socioeconomic implications of these changes remain somewhat ambiguous to date. Researchers have probed the residential segregation of Hispanics from white populations in these new destinations and found mixed evidence for spatial assimilation (; Wahl, Breckenridge, and Gunkel 2007). Some research suggests that Hispanics in new rural destinations are better off economically than those living in traditional rural places (). However, this is balanced against more recent work suggesting Latino economic circumstances in new rural destinations deteriorated in the first decade of the 2000s (). Despite the general urban bias in the ethnoracial inequality literature, it has been noted that “racial segregation is as much a reality in the American countryside as it is in the cities” (: 126). Whether black populations in the Delta or Black Belt, Latino colonias, or American Indian reservations, all are identifiable by high concentrations of minority residents. At the national level, patterns of ethnoracial residential segregation in small town America have been shown to be remarkably similar to those in larger metro areas, with nonmetro blacks facing particularly pronounced isolation (). Relatedly, a major characteristic of rural areas with concentrated minority populations is that they are also among the very poorest places in American society. Many of the U.S. regions characterized by the highest and most persistent poverty are also be home to high numbers of ethnoracial minorities (; ; ). Again, the Delta, Black Belt, Lower Rio Grande Valley (or Borderlands), and American Indian reservations stand as prime examples. Juxtaposed to the urban ethnoracial minority experience, : 127) argues:
The sobering implication is that the history of systemic and institutionalized ethnoracial oppression in rural America continues to reinforce contemporary patterns of inequality (e.g., ). The long and difficult history of ethnoracial minorities in the rural U.S. is reason enough to carry forward a concern for their economic welfare today. However, emerging demographic patterns also underscore the call for a contemporary assessment of underemployment among nonwhite populations in rural America. Just as nationally the U.S. population is undergoing a “diversity transition” () and is on pace to be “majority minority” by mid-century (), America’s rural population is also increasingly comprised of ethnoracial minorities (). In fact, research shows that that while the nonmetro nonwhite population grew by 20 percent during the first decade of the 2000s, the nonmetro white population barely grew at all (). Much of the recent and projected growth of the rural minority population is due to Hispanic population change. While in absolute numbers the size of the nonmetro Hispanic and black populations were quite similar as of 2010—3.8 million versus 4.2 million, respectively—between 2000 and 2010, Hispanics accounted for 56 percent of all nonmetro population growth (). The precipitous increase in the rural Hispanic population is driven by both the dispersion of Mexicans and other Hispanic groups away from traditional places of settlement in the Southwest toward smaller cities and towns throughout the country, as well as the higher fertility of Hispanics vis-à-vis native whites and others in new destination places (; ). It is in this context that we return to theme of rural underemployment raised by President’s National Advisory Commission on Rural Poverty in The People Left Behind (1967). Our aim is to carry the statistical record forward over the last half century—one characterized by major social, economic, and demographic change—to evaluate national progress on residential and ethnoracial inequality in underemployment duly noted by the Commission 50 years ago. MethodsDataTo meet our objectives, we draw on data from the March Current Population Surveys (CPS) from 1968 to 2017. We extract harmonized CPS data from the Integrated Public Use Microdata Series (IPUMS) database produced by the University of Minnesota (). The CPS, conducted by the U.S. Census Bureau for the U.S. Bureau of Labor Statistics, is a monthly survey that serves as the primary source of labor force statistics for the U.S. population. The March survey, known formerly as the Annual Demographic File and currently as the Annual Social and Economic Supplement (ASEC), provides detailed information about the socioeconomic characteristics of households, families, and individuals as of the interview date and for the prior calendar year (). Although the March CPS instrument has remained largely consistent through the years, modifications to survey items and answer categories have been made. The harmonized variables available through IPUMS allow users to efficiently recode variables and maximize consistency in measurement over time. Using these 50 years of data, we examine employment circumstances among all working-age individuals, defined as those ages 18 to 64 years at the time of the survey. We apply the CPS person-weights designed specifically for the individual-level ASEC data to produce representative estimates. We also note that the CPS uses a rotating sampling design, whereby sampled households are surveyed once per month for four months, drop out for four months, and are then interviewed again for four consecutive months before dropping out of the sample altogether. We therefore limit our analytic sample to individuals in households that are in their fifth through eighth month in the survey to avoid analyzing data for the same individuals in successive years. After imposing these restrictions, our dataset includes a total of 1,909,439 observations (unweighted). MeasuresWe measure underemployment, our dependent variable in this analysis, according to the Labor Utilization Framework (LUF) developed by and advanced by Clogg and Sullivan (; ; ). This measure, designed specifically for use with the CPS, defines four operational states of underemployment as:
Workers who do not fall into one of the four states of underemployment above are defined as adequately employed, while individuals who are not employed and do not indicate a desire to be are defined as not in the labor force. The latter are excluded from our analysis, and therefore the trends we document should be viewed as being conditional upon realized or desired participation in the labor force. Many of our analyses focus on a binary measure of underemployment which distinguishes between the adequately-employed and underemployed, irrespective of the specific type of underemployment. Our main explanatory variables of interest are metropolitan status of residence, ethnoracial group, and time period. We define residence using the U.S. Office of Management and Budget (OMB) metropolitan classification system, which we operationalize as a three-category variable distinguishing between residents of metro areas, nonmetro areas, and a small subset of respondents for whom metro status is not identified. Generally, metro areas consist of counties with a city of at least 50,000 residents (or a total urbanized area of 100,000 or more), plus surrounding counties that are significantly bound economically to such core counties via commuting patterns. Counties that do not meet the metro criteria are defined as nonmetro. Over the course of our study period, some counties’ metro status designation was reclassified due to changes population size, as well as changing levels of integration with urbanized cores. Here we use a floating definition of metro status that allows for such changes, in part because it is not possible to operationalize a fixed metro-nonmetro definition with public-use CPS data. County identifiers are suppressed for a substantial share of the public-use sample. The new metropolitan classifications issued by OMB on a decennial basis are adopted by the March CPS in years ending in five. The result is that the universe of metro and nonmetro areas from which workers are sampled in our analysis is not constant over time. Such reclassification is a selective process: many of the nonmetro areas with growing populations and robust economies during the first part of our study period transitioned to metro status in subsequent decades. However, use of a floating universe does capture the metro and nonmetro character of places at any given time point, and allows us to accurately track the welfare of workers who remained peripheral to the urban economy as it expanded geographically over the fifty years we study. We also examine ethnoracial differences in underemployment, with a focus on four groups defined as: non-Hispanic white (white), non-Hispanic black (black), Hispanic of any race (Hispanic), and non-Hispanic other race (other). The CPS did not begin collecting data on Hispanic ethnicity until 1971, therefore we cannot distinguish this population from other groups in the first three samples we include in our dataset (1968–1970). Since most Hispanics defined themselves as racially white when asked this question in 1971, we expect that this reclassification primarily affects estimates of white underemployment in the first 5-year period in our analysis, upwardly-biasing that figure. Our final variable of interest is time period, which we measure using 5-year intervals, beginning with 1968–1972 and ending with 2013–2017. We use the 1993–1997 period as the reference group in all models, since the abovementioned limitation to Hispanic classification make the first period a poor reference category. One advantage to pooling surveys in 5-year groups is the ability to obtain a greater number of cases in each period to facilitate ethnoracial group-by-residence comparisons. In interpreting the results, however, it is also important to note that pooling annual surveys serves to smooth year-to-year variation in underemployment. We also include a number of control variables in our models. First, we control for residence in the nine regional census divisions to account for geographic labor market variations and shifts in the CPS sample. Second, we control for a range of demographic factors to account for the changing composition of the U.S. labor force. These include age measured in years and as a quadratic function, and a series of indicator variables for sex (male = reference), marital status (married = reference), educational attainment (less than high school = reference), and industry of employment (agriculture, forestry, fishing, and mining = reference). Given significant changes in industry classification over time, we use an IPUMS-constructed variable that is harmonized to the Census Bureau’s 1950 industrial classification system (). Our choice of controls is motivated by prior research showing that underemployment is generally higher among the youngest and oldest workers, females, the non-married, the less educated, and those in extractive and service industries (; ). Further, regional controls are merited given the shifting geographic distribution of counties classified as nonmetro, and the uneven distribution of ethnoracial groups across the country (; ; ; ). Our approach enables us to assess how historical shifts in the regional concentration of workers (e.g., rural blacks in the South, and rural Hispanics traditionally in the Southwest but increasingly in new destinations in other regions of the country) and the demographic composition of the labor force (e.g., increasing educational attainment and female labor force participation) have contributed to the risk of underemployment over time. Analytic StrategyOur analysis begins with a series of descriptive statistics. We first describe overall trends in underemployment for the 50 years spanning 1968 to 2017, estimating the overall underemployment rate and the prevalence of each of the four types of underemployment. We then describe trends in underemployment by residence and ethnoracial group, with the goal of understanding whether and how underemployment has changed over time for different groups of workers. Next, we assess whether and how the residence- and ethnoracial group-specific differentials in underemployment, and changes therein, can be explained by changes in the demographic composition and geographic distribution of the labor force over the last five decades. To do so, we estimate a series of linear probability models predicting the likelihood that a member of the labor force is underemployed as a function of our independent variables and controls. Our modeling strategy is as follows. We begin by analyzing underemployment across the full national sample. Our goals are, first, to assess whether the changes in the composition of the labor force in metro and nonmetro areas explains trends in the residential gap in the probability of underemployment over the past fifty years; and second (and relatedly), to estimate levels and trends in the magnitude of metro-nonmetro differences in underemployment risk that cannot be explained by the control variables included in our model. To meet these goals, we begin by estimating a base model that predicts underemployment as a function of metropolitan status, time period, and a metropolitan status-by-time period interaction, which allows the association between metropolitan status and the probability of underemployment to vary over time. We then estimate a second model that adds region fixed-effects to this base model, thereby holding the regional distribution of the rural and urban labor forces constant across time. In the third model, we drop the region fixed-effects and introduce the full suite of demographic variables to control for changes in the demographic profile of the labor force over time. And finally, in the fourth, full model, we add both region fixed-effects and demographic controls to the base model. To assess the contribution of shifts in the demographic and geographic composition of the metro and nonmetro labor forces to changes in the probability of underemployment, we derive predicted probabilities of underemployment by residence and time period from each of the four models, and then compare changes in the magnitude of these probabilities, and metro-nonmetro differences therein, across the different models. To the extent that the predicted probability of underemployment increases (or decreases) over the base model after introducing a set of controls, then changes in the labor force with respect to those changes have, together, contributed to a reduction (or increase) in the probability of underemployment. Likewise, if introducing blocks of control variables increases (or decreases) the metro-nonmetro differential in the probability of underemployment, one can conclude that changes in those population characteristics have contributed to metro-nonmetro convergence (or divergence) in underemployment. We then stratify the sample according to metropolitan status and estimate a series of nonmetro- and metro-specific models. Here, our goal is to assess whether ethnoracial disparities in underemployment within metro and nonmetro areas have changed over time, and whether these changes can be explained by changes in the demographic and regional composition of each group. In this sequence of models, our base model predicts the probability of underemployment as a function of ethnoracial group membership, time period, and interaction terms between these vectors of variables. Similar to our first sequence of models, we then estimate a second model that introduces region fixed-effects to the base model and a third model that includes the base model plus all demographic controls. The fourth, full model, adds both region fixed-effects and demographic controls to the base model. We then estimate predicted probabilities of underemployment for each model, and again conduct cross-model comparisons to assess the contribution of demographic and geographic compositional shifts to ethnoracial inequalities in underemployment within nonmetro and metro areas. ResultsTrends in Underemployment, 1968–2017We begin by describing the share of the U.S. labor force that is underemployed, both overall and by category of underemployment, in 5-year periods from 1968 to 2017 (Table 1). These time trends at the national level are illustrated in Figure 1, and generally show a lagging countercyclical relationship between underemployment and national macroeconomic conditions. Over this 50-year span, underemployment averaged 17.5 percent of the labor force and ranged from a low of 14.4 percent in 1998–2002, reflecting the economic expansion related to the dotcom boom of the late 1990s, to a high of 21.1 percent in 1983–1987, following the deep recession at the beginning of that decade. We find a similarly-high level of underemployment (20.5%) in 2008–2012, in the aftermath of the Great Recession.  Open in a separate window Figure 1. Percent underemployed by residence, 1968–2017 Source: March Current Population Surveys. Notes: Changes in metro status classifications issued by OMB on a decennial basis were adopted by the CPS in the years of 1975, 1985, 1995, 2005, and 2015. During the time observed, the U.S. economy experienced recessions during the following periods: 1969–1970, 1973–1975, 1980–1980, 1981–1982, 1990–1991, 2001, and 2007–2009, with the longest occurring between 2007–2009 (18 months), 1981–1982 (16 months), and 1973–1975 (16 months). Table 1.Percent underemployed overall and by type, 1968–2017 YearUnderemployedLow incomeLow hoursUnemployedDiscouraged1968–197214.95.52.64.62.21973–197717.04.93.46.72.01978–198217.75.73.97.01.11983–198721.16.94.98.01.41988–199218.36.74.46.30.91993–199718.17.13.96.01.11998–200214.46.22.74.90.72003–200715.36.03.15.40.82008–201220.55.45.38.61.32013–201717.66.14.46.01.2 Open in a separate window Source: March Current Population Surveys. Notes: Changes in metro status classifications issued by OMB on a decennial basis were adopted by the CPS in the years of 1975, 1985, 1995, 2005, and 2015. During the time observed, the U.S. economy experienced recessions during the following periods: 1969–1970, 1973–1975, 1980–1980, 1981–1982, 1990–1991, 2001, and 2007–2009, with the longest occurring between 2007–2009 (18 months), 1981–1982 (16 months), and 1973–1975 (16 months). Table 1 also shows trends from 1968 to 2017 disaggregated into the prevalence of underemployment by type. The two most common forms of underemployment during this time span are unemployment and underemployment by low income, followed by underemployment by low hours and discouragement, respectively. Over the 50 years examined, the average unemployed share of the labor force was 6.4 percent, while underemployment by low income averaged 6.1 percent. In contrast, underemployment by low hours and discouragement averaged 3.9 percent and 1.3 percent of the labor force, respectively. The distribution of the underemployed across these four categories has remained relatively consistent over time, except for a decline in the proportion of discouraged workers over the first three periods in our study. The stability of underemployment over the last five decades is remarkable given major concurrent economic and demographic changes in U.S. society. In Figure 1, we examine variation in long-term underemployment trends between metro and nonmetro areas. At least two important aspects of metro-nonmetro differences stand out. First, underemployment among the nonmetro labor force has exceeded that of metro areas at every time point over the last 50 years. Overall, nonmetro underemployment has averaged 20.4 percent—or one-in-five nonmetro workers—compared to a lower rate of 16.6 percent in metro areas. However, we also find that the nonmetro-metro gap has been shrinking over the last two decades, falling from a peak of over six percentage points in the mid-1980s to below two percentage points in the years after 2008. This trend is particularly notable since many of the nonmetro areas with the fastest population growth during the early periods of our analysis were reclassified as metro in subsequent years. The implication is that the areas remaining nonmetro toward the end of our analysis are in a least some respects negatively selected, a process that would be expected to place upward pressure on nonmetro-metro inequalities. The results also suggest a third finding: despite major changes in the demographic composition of the labor force and structure of the economy over the last 50 years, things in 2013–2017 are much as they were in 1968–1972 in terms of the prevalence of nonmetro underemployment (a difference of just 1.1 percentage points). We next turn to the second focus of this paper: differences by race and ethnicity. We analyze levels of underemployment, and changes therein, among white, black, and Hispanic workers in nonmetro and metro areas, respectively (Figure 2). A comparison of the group-specific figures illustrates how trends in employment experiences have differed by residence and ethnoracial group over time. Underemployment among white workers largely parallels the trends and differences outlined above, reflecting their relatively larger share of the population. There is evidence of metro-nonmetro convergence in underemployment among white workers over recent decades. Nonetheless, nonmetro white underemployment has consistently exceeded that of metro white workers across the last half-century, averaging 18.5 and 13.7 percent, respectively, over the study period—an average residential gap of 4.8 percentage points. Open in a separate window Open in a separate window Open in a separate window Figure 2. Percent underemployed by residence and race/ethnicity, 1968–2017 Source: March Current Population Surveys. Notes: Changes in metro status classifications issued by OMB on a decennial basis were adopted by the CPS in the years of 1975, 1985, 1995, 2005, and 2015. During the time observed, the U.S. economy experienced recessions during the following periods: 1969–1970, 1973–1975, 1980–1980, 1981–1982, 1990–1991, 2001, and 2007–2009, with the longest occurring between 2007–2009 (18 months), 1981–1982 (16 months), and 1973–1975 (16 months). Among the black population, the results suggest a number of key points. First, underemployment among black workers in both nonmetro and metro contexts is consistently much higher than among their white counterparts, demonstrating the lasting power of ethnoracial stratification processes in the U.S. Over the last 50 years, the black-white difference in underemployment has averaged 16.9 percentage points in nonmetro areas and 12.6 percentage points in metro areas. Second, nonmetro black workers are plainly disadvantaged relative to their metro black counterparts. Between 1968 and 2017, nonmetro black underemployment averaged 35.3 percent—more than one-in-three black workers in nonmetro areas—and stood as high as 43.5 percent in the mid-1980s. The gap between black workers in nonmetro and metro areas has averaged nine percentage points over the last fifty years. The comparative advantage of black workers in metro areas relative to their nonmetro peers should not distract from the fact that underemployment among this population is nonetheless high, averaging 26.3 percent between 1968 and 2017. Taken together, this is a truly extraordinary level of employment hardship. Finally, our results demonstrate a consistent disadvantage in underemployment among Hispanic workers relative to white workers in both metro and nonmetro settings, again illustrating the persistence of ethnoracial disparities in the U.S. Over the last 50 years, the Hispanic-white difference in underemployment has averaged 10.5 percentage points in nonmetro areas and 11.1 percentage points in metro areas. Although, here too we see evidence of convergence in the Hispanic-white gap in underemployment in more recent decades, especially since the peak of Hispanic underemployment in the 1980s. For example, the nonmetro Hispanic-white gap was 12.9 percentage points in 1983–1987 but declined to 5.9 percentage points in 2013–2017. Greater employment hardship among nonmetro Hispanics compared to their metro counterparts is also apparent over the period as a whole. During the last half-century, nonmetro Hispanic underemployment averaged 29.0 percent compared to 24.8 percent among Hispanic workers in metro areas. Of special note, however, is that Figure 2 also illustrates a unique feature of Hispanic employment patterns: a decline in the metro-nonmetro gap since the 1990s culminated in a reversal of residential disadvantage following the Great Recession of the late 2000s. Since 2008, metro Hispanics have experienced higher rates of underemployment than Hispanic workers in nonmetro areas. This trend parallels other evidence of post-recession improvements in the relative standards of rural Hispanics vis-à-vis other ethnoracial groups (). We speculate that this pattern could reflect the uneven regional impacts of the economic crisis, including disproportionate declines in housing markets and related contractions in construction employment in some areas of the country but not others. The longer-term pattern of nonmetro-metro convergence in Hispanic underemployment also corresponds to the emergence of new Hispanic destinations from the mid-1990s forward (e.g., ). Overall, it noteworthy that the ‘nonmetropolitanization’ of the Hispanic population has been associated with an improvement in economic status among this population, at least as measured by underemployment. Multivariate Models of UnderemploymentWe next estimate a series of regression models, with the goal of understanding whether and how changes in the demographic profile and regional distribution of workers explains temporal trends in underemployment and both nonmetro-metro and ethnoracial disparities therein. As elaborated above, we begin with a base model, followed by the addition of region fixed-effects (i.e., we hold the regional distribution of the labor force constant across time), then drop the region fixed-effects and introduce the full suite of demographic variables (i.e., we control for changes in the demographic profile of the labor force over time), and finally estimate a full model in which region fixed effects and demographic controls are both added to the base model. To facilitate a more parsimonious presentation, we display the results as predicted probabilities and include only findings from the base and full models in our figures. Our first series of models focuses on nonmetro-metro disparities in underemployment over time. It predicts underemployment as a function of residence, time, a time-by-residence interaction, and in some models a series of demographic and regional covariates. Figure 3 shows the predicted probability of a worker being underemployed by residence, 1968–2017. The results from the base model illustrate that the probability of underemployment has been higher in nonmetro than metro areas in every period over the last 50 years. The probability of a nonmetro worker being underemployed averaged 20.4 percent over the entire time period, ranging from a low of 17.2 percent in 1998–2002 to a high of 26.1 percent in 1983–1987. In metro areas the corresponding average over the entire period was 16.6 percent, with a low of 13.3 percent in 1968–1972 and a high of 20.3 percent in 2008–2012. However, while the probability of underemployment has been consistently higher for nonmetro workers, the gap with their metro counterparts has been variable: the difference was largest in the mid-1980s (6.8 percentage points) and smallest in the wake of the Great Recession (1.4 percentage points). Open in a separate window Open in a separate window Figure 3. Predicted probability of being underemployed by residence, 1968–2017 Source: March Current Population Surveys. Notes: “Base” model includes main predictors and interaction terms. “Full” model adds region fixed-effects and demographic controls. Changes in metro status classifications issued by OMB on a decennial basis were adopted by the CPS in the years of 1975, 1985, 1995, 2005, and 2015. During the time observed, the U.S. economy experienced recessions during the following periods: 1969–1970, 1973–1975, 1980–1980, 1981–1982, 1990–1991, 2001, and 2007–2009, with the longest occurring between 2007–2009 (18 months), 1981–1982 (16 months), and 1973–1975 (16 months). The intervening models adding region fixed-effects and demographic controls indicate that both considerations hold explanatory power across residential settings, but that changes in the demographic composition of the labor force have had greater impacts on the probability of underemployment than have regional effects. In other words, the relative contribution to these trends from demographic change is greater than what can be explained by conditions specific to particular regions of the country. A comparison of the predicted probabilities from the base and full models (Figure 3) reveals that after introducing the demographic and regional control variables, the predicted probabilities of underemployment shift downward, and disproportionately so over the first half of the time series. That is, instead of the relatively flat trend observed in the base model we find a trend toward an increasing probability of underemployment in both nonmetro and metro areas over the last five decades. This is important in its own right, and suggests that the demographic and regional characteristics of workers represented a source of greater disadvantage in earlier periods compared to later periods in our analysis. Nonetheless, the likelihood of underemployment among nonmetro workers continues to exceed that of their metro counterparts in every period observed. The probability of a nonmetro worker being underemployed averaged 17.3 percent over the entire time period, ranging from a low of 12.0 percent in 1968–1972 to a high of 21.1 percent in 2008–2012. In metro areas the corresponding average over five decades was 13.5 percent, with a low of 8.1 percent in 1968–1972 and a high of 19.1 percent in 2008–2012. In the presence of the full range of controls, the gap in the probability of underemployment between nonmetro and metro workers was greatest in the mid-1980s (6.1 percentage points) and smallest in the wake of the Great Recession (2.0 percentage points). We next estimate a series of residence-specific regression models, with the goal of describing ethnoracial disparities by metro status over time and evaluating the contribution of demographic and regional shifts to these trends. Figure 4 shows the predicted probability of a worker being underemployed by residence and ethnoracial group, 1968–2017. The results from the base models provide a number of important insights into ethnoracial disparities in underemployment across the rural-urban divide. On average over the 50 years observed, the probability of underemployment is greater for workers in nonmetro compared to metro areas among every ethnoracial group: 18.5 percent versus 13.7 percent among whites, 35.3 percent versus 26.3 percent among blacks, and 29.0 percent versus 24.8 percent among Hispanics. Moreover, ethnoracial minorities are consistently disadvantaged relative to their white counterparts within each residential context. The average gap in the probability of underemployment among black and white workers has been 16.9 percentage points and 12.6 percentage points in nonmetro and metro areas, respectively; between Hispanics and whites the comparable numbers are 10.5 percentage points and 11.1 percentage points. Open in a separate window Open in a separate window Open in a separate window Open in a separate window Figure 4. Predicted probability of being underemployed by residence and race/ethnicity, 1968–2017 Source: March Current Population Surveys. Notes: “Base” model includes main predictors and interaction terms. “Full” model adds region fixed-effects and demographic controls. Changes in metro status classifications issued by OMB on a decennial basis were adopted by the CPS in the years of 1975, 1985, 1995, 2005, and 2015. During the time observed, the U.S. economy experienced recessions during the following periods: 1969–1970, 1973–1975, 1980–1980, 1981–1982, 1990–1991, 2001, and 2007–2009, with the longest occurring between 2007–2009 (18 months), 1981–1982 (16 months), and 1973–1975 (16 months). Again, the intervening models adding region fixed-effects and demographic controls indicate that changes in both sets of characteristics contribute to the trends observed, but that changes in the demographic composition of the labor force have had greater impacts on the probability of underemployment by residence and ethnoracial group than have regional effects. The trends for the full model displayed in Figure 4 show that introducing all of the controls considerably diminishes ethnoracial disparities in both metro and nonmetro areas. In fact, in the most recent period of our analysis the introduction of controls eliminates nonmetro Hispanic-white disparities in underemployment completely; and the residual differences in many other periods are more than halved. In contrast, introducing controls appears to account for much less of the observed nonmetro black-white disparities. The average gap in the probability of underemployment between black and white workers in the full model is 12.4 percentage points and 8.9 percentage points in nonmetro and metro areas, respectively; between Hispanics and whites the comparable numbers are 5.5 percentage points and 6.1 percentage points. The introduction of control variables does not clearly shift the temporal trend in the probability of group-specific underemployment in nonmetro areas, but it does lead to a more pronounced upward trend in underemployment risk in metro areas. Even so, on average, the probability of underemployment remains greater in nonmetro areas across every ethnoracial group: 16.6 percent versus 11.4 percent among whites, 29.0 percent versus 20.3 percent among blacks, and 22.1 percent versus 17.4 percent among Hispanics. The 1980s and the aftermath of the Great Recession represent especially difficult times for nonmetro workers across ethnoracial groups, while in metro areas the aftermath of the Great Recession continues to stand out. Finally, our multivariate results are consistent with the initial descriptive evidence of a residential reversal in terms of disadvantage for Hispanic workers in the final period analyzed: in 2013–2017 the probability of underemployment among nonmetro white and Hispanic workers reached parity for the first time at 17.6 percent, while the probability of underemployment stood at 19.7 percent among metro Hispanics. Overall, that after introducing the full range of controls we do not observe a downward trend across residential settings and ethnoracial groups (and in some cases the opposite) points to the increasing importance of other structural forces in the economy and American society in shaping the risk of underemployment. DiscussionThe People Left Behind (1967) documented the high and persistent poverty that afflicted much of rural America in the 1960s, noting the disproportionate prevalence of underemployment in rural areas, and ethnoracial inequality therein, as a critical dimension of economic deprivation in the U.S. countryside. Our analysis suggests the Commission’s (1967) basic observations on rural underemployment have persisted, and perhaps grown even more salient, over the 50 years subsequent to the publication of its report. From 1968 to 2017, rural workers have faced more prevalent underemployment compared to their urban counterparts, with especially pronounced disadvantages for rural ethnoracial minorities. Rural blacks, in particular, have faced a disproportionate risk of underemployment compared to other workers in rural and urban areas, while rural Hispanics, also historically subject to a disadvantage compared to those in urban areas, have witnessed a residential reversal of employment fortunes in more recent years. Finally, rural white workers, while relatively advantaged compared to ethnoracial minority workers, have nonetheless experienced greater underemployment than among whites in urban areas in every period over the last 50 years. Overall, after introducing the full range of controls, we observe either relatively flat or upward trends in the risk of underemployment by residence and ethnoracial group. The implication is that structural forces beyond demographic and regional considerations increasingly place workers at risk of underemployment. Our analysis tells a story of both persistence and change. By providing an extensive historical overview of empirical trends in underemployment and ethnoracial disparities therein, we build on the work of the rural poverty commission (1967, 1968) and contribute to the extant rural sociological literature on underemployment (e.g., ; ; ; ; ; ; ; ; ; ). There should be little doubt that since The People Left Behind (1967) was published the material conditions and infrastructure of rural America have improved in many regards (e.g., ). And we do find evidence that the rural-urban gap in underemployment has been narrowing over time (albeit with the underlying risk of underemployment increasing in both settings). Nonetheless, we also show underemployment has stood as a chronic disadvantage for most rural workers over the last five decades. It is worth reiterating that many nonmetro areas that were growing in early periods of our analysis were reclassified as metro in subsequent years, so that nonmetro areas in latter periods of our analysis are negatively selected on key characteristics. From this perspective, that we do not show nonmetro workers to be increasingly left behind by those in metro areas is good news. A straightforward implication of this research is that employment alone is not a sufficient condition for escaping economic hardship, a fact with special significance in rural America. Too often the media and policymakers focus on the unemployment rate as the sole arbiter of how well the economy is doing for working people. The reality is that unemployment as a labor force measure only tells us how many people are out of work and actively looking for a new job. It does not tell us anything about job quality among the employed, including if workers are able to secure jobs that pay the wages and/or provide the hours needed to escape economic hardship. It also does not tell us anything about those who have dropped out of the labor force due to discouragement with their job prospects. In other words, unemployment captures only a slice of the broader range of labor market hardships that people face. Conceptualizing employment hardship as underemployment using the LUF, as we have done here, helps to expand the net to capture such considerations. But even using the LUF, much of what constitutes common perceptions of good and bad jobs is left out (e.g., whether jobs come with fringe benefits like health insurance or a retirement plan with an employer match). And all of the findings presented here must be viewed as conditional upon realized or desired participation in the labor force; the LUF does not capture the critical issue of labor force drop-out beyond discouragement. As such it excludes drop-out due to causes including significant care-giving responsibilities, disability, poor health, substance abuse, or the stain of a criminal record. As with other measures of economic well-being, these dimensions of hardship are likely to be stratified by residence and ethnoracial group membership. The impacts of mass incarceration, which eliminates an ever-growing number of economically marginal black men from the U.S. labor market, stands as a stark example (; ) Our findings hold a range of implications for public policy. To the extent that the underemployed fit societal conceptions of the “deserving poor,” there may be real opportunities for improving economic well-being among many working people if combined with sufficient political will. Two straightforward examples include increasing the federal minimum wage and expanding the Earned Income Tax Credit (EITC). The value of the federal minimum wage is established by Congress and is not regularly adjusted to address increases in the cost of living. The last increase in the federal minimum wage was to $7.25/hour in 2009. Since that time it has lost nearly 10 percent of its purchasing power due to inflation; its peak value in terms of purchasing power was 50 years ago in 1968 (), coincidently the very first year we analyzed in this study. While many states and localities have established their own higher minimums (), few of these places are in the South and other areas where rural minorities are most highly represented. The EITC, a refundable tax credit for low-income working people, is a popular and successful program. Expanding and increasing the credit—at the federal level and in states where it is offered—would serve to increase the incomes of those who work at low wages. In addition, other programs aimed at offsetting the opportunity costs associated with work, like those related to child care and transportation, stand to be helpful to the underemployed. These benefits are most likely to be realized when implemented in a way that accounts for spatial differences in these opportunity costs, including the near-absence of transportation and licensed child care providers in many rural areas. It is also notable that many of the states where rural ethnoracial minorities are most concentrated have the most meager social safety nets and regressive systems of taxation (). Last, it would seem that migration to urban economies, once a tried and true strategy for rural workers seeking economic opportunity, may increasingly offer a less promising path of escape from employment hardship. Indeed, our results suggest that among Hispanics the opposite has become true, with those in nonmetro areas now facing less risk of underemployment than their metro counterparts. The understood the centrality of quality employment to the rural poverty problem 50 years ago. It boldly called for a federally enforced policy of full employment and a national manpower program to help rural workers navigate the vicissitudes of the labor market, the latter aimed at issues such as job training and providing people with access to accurate information to help facilitate migration decisions. The Commission’s leading recommendation for eliminating rural poverty was straightforward: a national policy of full employment at above-poverty-level wages, with a public sector employment guarantee when the market falls short of producing this goal. While the macroeconomic implications of such a policy would be hotly debated today, there is less doubt that the chronic underutilization of so many workers evidenced here is objectionable to most Americans, particularly the millions who are trying to “play by the rules” but are stuck in the structural trap of underemployment. To conclude, this study makes clear that the intersection of race, ethnicity, residence, and underemployment continues to matter 50 years after the publication of The People Left Behind (1967). There is no question that the stated goal of the Commission a half-century ago—to “achieve the elimination of underemployment and low income of rural people” (1967: vi)—remains very much an unmet aspiration. Underemployment continues to stand as a critical obstacle to economic well-being in contemporary rural America. AcknowledgmentsEarlier versions of this paper were presented in 2018 at the “Rural Poverty: Fifty Years After The People Left Behind” conference in Washington, DC, organized by the Rural Policy Research Institute (RUPRI) with support from the U.S. Department of Agriculture, and the annual meetings of the Southern Demographic Association in Durham, NC. The authors acknowledge the assistance provided by the Population Research Institute at The Pennsylvania State University, which has core support from the Eunice Kennedy Shriver National Institute on Child Health and Human Development (P2CHD041025). This work was also supported by the USDA National Institute of Food and Agriculture and Multistate Research Project #PEN04623 (Accession #1013257) titled, “Social, Economic and Environmental Causes and Consequences of Demographic Change in Rural America.” AppendixTable A1.Percent underemployed by residence and race/ethnicity, 1968–2017 Nonmetro Metro YearTotalWhiteBlackHispanicTotalWhiteBlackHispanic1968–197218.0%16.1%37.7%23.4%13.3%11.8%23.8%19.4%1973–197719.117.334.528.616.014.325.821.21978–198220.518.637.427.516.414.028.923.11983–198726.123.943.536.819.416.432.828.21988–199223.421.440.036.317.014.028.226.81993–199721.319.433.733.717.314.027.329.51998–200217.215.329.725.113.810.921.223.02003–200717.816.028.326.414.811.921.922.12008–201221.719.735.629.320.316.428.130.12013–201719.117.033.022.917.413.725.424.6 Open in a separate window Source: March Current Population Surveys. Table A2.Linear probability models predicting underemployment (yes=1), 1968–2017 Base Region F.E. Demographic Full BS.E.BS.E.BS.E.BS.E.Residence Metro (ref). Nonmetro.040***.003.041***.003.041***.003.04***.003 Not identified−.001.012.004.012.010.011.016.011Ethnorace White (ref.) Black.090***.001.093***.001 Hispanic.065***.001.062***.001 Other.041******.036***.002Period 1968–1972−.040***.002−.039***.002−.081***.002−.082***.002 1973–1977−.013***.002−.012***.002−.057***.002−.058***.002 1978–1982−.009***.002−.009***.002−.053***.002−.054***.002 1983–1987.021***.002.021***.002−.020***.002−.021***.002 1988–1992−.003*.002−.003.002−.032***.002.032***.002 1993–1997 (ref.) 1998–2002−.035***.002−.035***.002−.034***.001−.04***.001 2003–2007.026***.002−.026***.002−.027***.001.028***.001 2008–2012.030***.002−.030***.002.027**.002−.003.002 2013–2017.001.002.000.002−.003.002.002Residence*Period Nonmetro*68–72.007*.004.006.004−.004.004−.003.003 Nonmetro*73–77−.009*.004−.009*.004−.019***.004−.018***.004 Nonmetro*78–82.001.004.002.004−.007**.004−.005.004 Nonmetro*83–87.028***.004.028***.004.017***.004.018***.000 Nonmetro*88–92.025***.004.025***.004.018***.004.018***.004 Nonmetro*92–97 (ref.) Nonmetro*98–02−.006.004−.005.004−.004.004−.004.004 Nonmetro*03–07−.009*.004−.008*.004−.005.004−.005.004 Nonmetro*08–12−.025***.004−.024***.004−.022***.004−.023***.004 Nonmetro*13–17−.022***.004−.021***.004−.020***.004−.021***.004 Not ident.*68–72(empty)(empty)(empty)(empty) Not ident.*73–77.033*.014.032*.014.023.013.018 Not ident.*78–82.033**.012.032**.012.027*.012.021 Not ident.*83–87.035**.015.033**.013.030*.012.024* Not ident.*88–92.031**.015.029*.015.023*.014.021 Not ident.*93–97 (ref.) Not ident.*98–02.005.020.014.020.005.019.009 Not ident.*03–07.008.016.033.016−.011.015−.016 Not ident.*08–12−.013.016−.020.016−.026.015−.031 Not ident.*13–17.013.016.005.016−.004.015−.008Region New England (ref.) Middle Atlantic.018***.001.003*.001 East North Central.025***.001.132***.001 West North Central.004**.002−.002.001 South Atlantic.023***.001−.008***.001 East South Central.041***.002.014***.001 West South Central.037***.002.002.001 Mountain.025***.002.006***.001 Pacific.046***.001.022***.001Age−.017***.000−.017***.000Age 2.000***.000.000***.000Female.038***.001.038***.001Marital status Married (ref.) Never married.074***.001.074***.001 Sep., div., wid..055***.001.055***.001Education < high school (ref.) H.S. - < 4 yrs. coll.−.090***.001−.090***.001 4+ yrs. coll.−.150***.001−.150***.001Industry Extractive (ref.) Construction.008***.002.009***.002 Durable manuf.−.100***.002−.101***.002 Non- durable manuf.−.081***.002−.080***.002 Trans. and comm.−.103.002−1.03***.002 Trade−.038***.002−.038***.002 FIRE−.112***.002−.112***.002 Services−.065***.000−.064***.002 Other.042***.002.043.002R-squared.005.006.080.090n (unweighted)1,909,4431,909,4431,909,4391,909,439 Open in a separate window Source: March Current Population Surveys. Notes: F.E.=fixed effects. Ref.=reference group. B=unstandardized coefficient. S.E.=robust standard error. *p<.05 **p<.01 ***p<.001. Table A3.Linear probability models predicting nonmetro underemployment (yes=1), 1968–2017 Base Region F.E. Demographic Full BS.E.BS.E.BS.E.BS.E.Ethnorace White (ref.) Black.143***.013.145***0.013.095***.012.099***.012 Hispanic.143***.013.140***0.013.086***.013.083***.013 Other.135***.017.129***0.017.106***.016.100***.017Period 1968–1972−.032***.003−.032***0.003−.083***.003−.083***.003 1973–1977−.020***.003−.019***0.003−.073***.003−.072***.003 1978–1982−.008**.003−.0060.003−.058***.003−.057***.003 1983–1987.045***.004.046***0.004−.006.004−.005.004 1988–1992.020***.003.020***0.004−.016***.004−.016***.004 1993–1997 (ref.) 1998–2002−.041***.004−.041***0.004−.035***.003−.036***.003 2003–2007−.034***.004−.034***0.004−.029***.003−.029***.003 2008–2012.002.004.0030.0040.005.004.005.004 2013–2017−.023***.004−.023***0.004−.022***.004−.022***.004Ethnorace*Period Black*68–72.073***.015.072***0.015.075***.015.074***.015 Black*73–77.028**.016.028**0.016.030*.015.030.015 Black*78–82.044**.016.043**0.016.045**.015.043**.015 Black*83–87.053**.016.053**0.016.051***.015.051***.015 Black*88–92.044**.017.044*0.017.039**.016.039*.016 Black*93–97 (ref.) Black*98–02.000.017.0000.017−.002.016−.003.016 Black*03–07−.020.016−.0210.016−.018.016−.019.016 Black*08–12.017.017.0160.017.017.016.016.016 Black*13–17.017.018.0150.018.019.017.018.017 Hispanic*68–72−.070**.026−.075**0.026−.047*.025−.051*.025 Hispanic*73–77−.031.020−.034*0.020−.013.019−.016.019 Hispanic*78–82−.054**.178−.058***0.018−.036*.017−.040*.017 Hispanic*83–87−.014.019−.0160.019.005.018.004.018 Hispanic*88–92.006.020.0070.020.019.019.019.019 Hispanic*93–97 (ref.) Hispanic*98–02−.06**.017−.044**0.017−.043**.017−.042*.017 Hispanic*03–07−.039**.016−.037**0.016−.034*.016−.032*.016 Hispanic*08–12−.047**.017−.046**0.017−.035*.017−.034*.017 Hispanic*13–17−.085***.017−.085***0.017−.084***.017−.083***.017 Other*68–72−.025.033−.0310.033−.022.032−.028.032 Other*73–77.020.039.0210.039.039.038.039.038 Other*78–82−.025.027−.0230.027−.015.027−.012.027 Other*83–87.015.026.0140.026.025.025.024.025 Other*88–92−.030.024−.0310.024−.016.023−.017.023 Other*93–97 (ref.) Other*98–02−.031.024−.0290.022−.036.021−.035.021 Other*03–07−.042.020−.0410.020−.051**.019−.050.019 Other*08–12−.056.21−.0530.021−.060**.020−.057.020 Other*13–17−.017.22−.0150.022−.025.021−.023.021Region New England (ref.) Middle Atlantic.017***0.003.013***.003 East North Central.026***0.003.020***.003 West North Central.015***0.003.008***.003 South Atlantic.010***0.003.000**.003 East South Central.037***0.003.029.003 West South Central.024***0.003.015***.003 Mountain.023***0.003.017***.003 Pacific.049***0.004.042.004Age−.016.000−.016***.000Age 2.000.000.000***.000Female.072.002.072***.002Marital status Married (ref.) Never married.101***.002.101***.002 Sep., div., wid..074***.002.074***.002Education < high school (ref.) H.S. - < 4 yrs. coll.−.089***.002−.089***.002 4+ yrs. coll.−.163***.002−.162***.002Industry Extractive (ref.) Construction.028***.004.029***.004 Durable manuf.−.096***.003−.097***.003 Non-durable manuf.−.076***.003−.074***.003 Trans. and comm.−.095***.004−.094***.004 Trade−.027***.003−.026***.003 FIRE−.113***.004−.112***.004 Services−.054***.003−.054***.003 Other.059***.004.059***.004R-squared.021.022.092.093n (unweighted)448,008448,008448,008448,008 Open in a separate window Source: March Current Population Surveys. Notes: F.E.=fixed effects. Ref.=reference group. B=unstandardized coefficient. S.E.=robust standard error. *p<.05 **p<.01 ***p<.001. Table A4.Linear probability models predicting metro underemployment (yes=1), 1968–2017 Base Region F.E. Demographic Full BS.E.BS.E.BS.E.BS.E.Ethnorace White (ref.) Black.133***.005.135***.005.092***.004.095***.004 Hispanic.155***.004.152***.004.094***.004.091***.004 Other.031***.006.027***.006.034***.006.029***.006Period 1968–1972−.022***.002−.023***.002−.077***.002−.078***.002 1973–1977.003.002.002.002−.052***.002−.053***.002 1978–1982.000.002−.001.002−.052***.002−.053***.002 1983–1987.023***.002.023***.002−.022***.002−.023***.002 1988–1992.000.002.000.002−.032***.002−.032***.002 1993–1997 (ref.) 1998–2002−.032***.002−.032***.002−.027***.002−.027***.002 2003–2007−.021***.002−.021***.002−.017***.002−.016***.002 2008–2012.024***.002.024***.002.029***.002.029***.002 2013–2017−.003.002−.003.002.003.002.004*.002Ethnorace*Period Black*68–72−.013*.007−.014*.007−.012.006−.013*.006 Black*73–77−.018**.007−.018**.007−.015*.006−.015*.006 Black*78–82.016*.007.015*.007.015*.006.015*.006 Black*83–87.031***.007.031***.007.027***.006.027***.006 Black*88–92.010.007.009.007.008.006.007.006 Black*93–97 (ref.) Black*98–02−.029***.006−.029**.006−.025***.006−.025***.006 Black*03–07−.032***.006−.032***.006−.027***.006−.027***.006 Black*08–12−.016**.006−.015*.006−.011.006−.011.006 Black*13–17−.016**.006−.016*.006−.011*.006−.011.006 Hispanic*68–72−.079***.011−.076**.011−.046.010−.025***.006 Hispanic*73–77−.085***.007−.084***.007−.054***.007−.043***.010 Hispanic*78–82−.064***.006−.063***.006−.039***.006−.053***.007 Hispanic*83–87−.036***.006−.035***.006−.017***.006−.037**.006 Hispanic*88–92−.026***.006−.026***.006−.014**.006−.016*.006 Hispanic*93–97 (ref.) Hispanic*98–02−.033***.005−.032***.005−.035***.005−.034***.005 Hispanic*03–07−.052***.005−.051***.005−.053***.005−.052***.005 Hispanic*08–12−.018***.005−.017**.005−.015**.005−.014**.005 Hispanic*13–17−.045***.005−.044***.005−.038**.005−.037***.005 Other*68–72−.041.013−.043**.013−.043***.012−.046***.012 Other*73–77−.017.012−.018.012−.011.012−.013.012 Other*78–82−.008.010−.008.010.008.010.008.010 Other*83–87.003.009.002.009.023*.009.022.009 Other*88–92.000.009.000.009.014.008.014.008 Other*93–97 (ref.) Other*98–02.007.007.007.007.008.007.008.007 Other*03–07−.008.007−.007.007−.007.007−.006.007 Other*08–12−.002.007−.001.007.001.007.003.007 Other*13–17−.005.007−.004.007−.004.007−.003.007Region New England (ref.) Middle Atlantic.002.002 East North Central.017***.002.000.001 West North Central.002.002.012***.001 South Atlantic−.007***.002−.003.002 East South Central.017***.002−.011***.002 West South Central.004*.002.008.002 Mountain.010***.002−.001*.002 Pacific.021***.002.004***.002Age−.017***.000−.017***.000Age 2.000***.000.000***.000Female.029***.001.029***.001Marital status Married (ref.) Never married.069***.001.069***.001 Sep., div., wid..050***.001.049***.001Education < high school (ref.) H.S. - < 4 yrs. coll.−.089***.001−.089***.001 4+ yrs. coll.−.148***.001−.148***.001Industry Extractive (ref.)−.010**.003 Construction−.012***.003−.117***.003 Durable manuf.−.116***.003−.097***.003 Non-durable manuf.−.098***.003−.119***.003 Trans. and comm.−.120***.003−.055***.003 Trade−.056***.003−.127***.003 FIRE−.129***.003−.082***.003 Services−.082***.003.024***.003 Other.023***.003−.010***.003R-squared.022.023.089.088n (unweighted)1,400,5781,400,4781,400,5741,400,574 Open in a separate window Source: March Current Population Surveys. Notes: F.E.=fixed effects. Ref.=reference group. B=unstandardized coefficient. S.E.=robust standard error. *p<.05 **p<.01 ***p<.001. Footnotes1In what follows, we use the term “ethnoracial” (e.g., ) to capture both ethnic and racial group identities. 2For ease of exposition, we use the terms urban-rural, metropolitan-nonmetropolitan, and metro-nonmetro interchangeably in this paper. However, note that our analysis employs the metro-nonmetro delineations produced by the U.S. Office of Management and Budget. 3An anonymous reviewer correctly noted that The People Left Behind (1967) and its companion volume Rural Poverty in the United States (1968) did not engage the concept of underemployment empirically. That research came later (e.g., ; ; ; ; ). We invoke the report because it underscores the importance of underemployment for rural poverty, and is a key historical statement on rural economic well-being. 4In the Economic Research Service county typology, each county in the U.S. is assigned a mutually exclusive economic type: farming-dependent, mining-dependent, manufacturing-dependent, federal/state government-dependent, recreation-dependent, or nonspecialized. 5Of course, some rural regions with disproportionately white populations, such as Central Appalachia and the Ozarks, are also characterized by high and persistent poverty. 6The U.S. labor force is defined as the sum of those employed and unemployed. People with formal jobs are employed. People who are jobless, looking for a job, and available for work are unemployed. People who are neither employed nor unemployed are not in the labor force (). 7We use the U.S. government’s official poverty measure. To define underemployment by low income, we take the average weighted poverty threshold for one person, divide by 48 weeks, and then figure 125 percent of that weekly poverty threshold. To determine the threshold across the entire range of years in our data, we begin by calculating the weekly poverty threshold for 2012, which is the most recent year for which we could find the individual weighted poverty threshold from the U.S. Census Bureau. We then apply the CPI inflator/deflator provided by IPUMS to adjust this threshold across the other years in our data. 8In our data, these are the CPS years of 1975, 1985, 1995, 2005, and 2015. 9Sensitivity analysis using 4-year and 6-year intervals resulted in similar substantive findings (available from the authors upon request). 10The regional census divisions include New England (reference), Middle Atlantic, East North Central, West North Central, South Atlantic, East South Central, West South Central, Mountain, and Pacific. 11We estimate linear probability models since they permit the straightforward estimation and interpretation of the nested model sequence conducted here. A similar approach is not possible using logistic regression. However, we note that the substantive conclusions from LPM are robust to re-estimating our models as logits (available from the authors upon request). 12We calculate linear predictions of the probability of underemployment holding the independent and control variables included in a given model at their means. For indicator variables (e.g., region), these means represent the average proportion of the sample in a given category. 13During the time examined here, the U.S. economy experienced contractions (recessions) during the following periods: December 1969-November 1970, November 1973-March 1975, January 1980-July 1980, July 1981-November 1982, July 1990-March 1991, March 2001-November 2001, December 2007-June 2009. In terms of duration, the longest contractions occurred between 2007 and 2009 (18 months), 1981 and 1982 (16 months), and 1973 and 1975 (16 months) (). 14Descriptive statistics used to generate Figures 1 and and22 can be found in the . 15The collapse of the housing market, a signature feature of the Great Recession, had distinct geographic patterns. Five states saw median home prices fall by more than 30%—Nevada (−49%), Florida (−38%), Arizona (−38%), California (37%), and Michigan (−34%)—while in nearly half of the states in the U.S. (N = 22) housing prices held steady or even increased during the crisis (). 16The results of the full series of regression models can be found in the . Contributor InformationTim Slack, Louisiana State University. Brian C. Thiede, The Pennsylvania State University. Leif Jensen, The Pennsylvania State University. References
|
