Trends in Deep Poverty from 1968 to 2011: The Influence of Family Structure, Employment Patterns, and the Safety Net
This paper examines the changing face of deep poverty in the United States over the past fifty years and the role of family structure, employment patterns, and governmental taxes and transfers in explaining these trends. Using a newly developed historical measure of poverty based on the Census Bureau’s supplemental poverty measure, we find that deep poverty rates have been fairly constant over the past fifty years, both overall and for families with children. In view of changes in family structure and government policy over this period, the intransigence of deep poverty is surprising. However, this overall stability obscures changes in the demographics of individuals and families in deep poverty, as well as the role of government policy. Governmental transfers reduce the risk of deep poverty for all subgroups examined, but the significance and the role of these programs have changed over time.
historical Supplemental Poverty Measure (SPM), historical poverty trends, antipoverty programs
An important URL in the references of this article has been updated. The updated link can be found in the DeNavas-Walt and Proctor 2014 reference on page 33.
One of the primary goals of government is to provide an adequate safety net to ensure that vulnerable members of society are protected from the most severe forms of deprivation. Public policies designed to target and aid certain groups necessarily create winners and losers over time, with certain demographic groups benefiting more from government intervention than others. Accurately measuring the size and demographics of the poorest segment of the population provides important insights into the functioning of the safety net. This article uses a newly developed measure of poverty to more fully capture the experience of those at the bottom of the income distribution, focusing on those primarily subsisting on less than half the poverty threshold. This article expands our current knowledge about the role of the safety net over the past fifty years and explores how effective the [End Page 14] safety net has been at targeting vulnerable families.
Understanding historical trends in severe deprivation in America is a challenging endeavor, both conceptually and technically. Many types of measures exist, and many are defensible. Severe deprivation is most commonly measured using the “deep poverty” rate, which is generally defined as having resources that total less than half of a specified poverty threshold. Indeed, this rate is published every year by the U.S. Census Bureau in its annual publication on poverty and income in the United States (see, for example, DeNavas-Walt and Proctor 2014). If a poverty threshold is understood as the least amount of income necessary to maintain a basic minimal living standard, those with resources less than half of this standard are thought to be in the most severe state of disadvantage. While other articles in this volume concentrate on other, and sometimes more severe, definitions of disadvantage, we focus on deep poverty given our ability to examine long-term trends in deep poverty rates and composition, as well as the role of social policies in ameliorating deep poverty.
The primary challenge in understanding historical trends in severe deprivation lies in the fact that current estimates of deep poverty are typically based on a fundamentally flawed measure of official poverty. This measure fails to fully capture the role of governmental safety net programs because it excludes the value of in-kind benefits—such as the Supplemental Nutrition Assistance Program (SNAP, formerly the Food Stamp Program) and housing assistance—as well as the role of the tax system, including tax credits such as the Earned Income Tax Credit (EITC). Deep poverty rates based on official measures also rely on an outdated poverty threshold, which is based solely on the cost of food and how that figured into family budgets in the 1950s and 1960s. Rates defined with reference to such thresholds fail to take into account changing living standards: some necessities, like food, have become a smaller part of family budgets, while others, like housing, are consuming a relatively greater share.
In this article, we utilize a recently developed and more comprehensive approach to poverty measurement to reanalyze trends in deep poverty in America. Recent analyses using such a measure show that considerable progress has been made in reducing overall poverty in the past fifty years (Fox et al. 2015; Wimer et al. 2013). At the same time, despite decades of economic growth, very little has changed for the poorest segment—that is, the share of the population with income below 50 percent of the poverty line. Indeed, as we show later, the rate of deep poverty in the United States has remained relatively constant over the past fifty years, hovering around 5 percent of the population.
This article seeks to understand such trends, not only the remarkable stasis over time but also the extent to which changes in deep poverty rates among key subgroups over time and the role of the social safety net for these groups have jointly contributed to this stasis. That is, do the flat overall deep poverty rates mask changes in who is most likely to be in severe deprivation over time? Who have been the winners, and who the losers? Specifically, this article aims to investigate whether the composition of the population in deep poverty has changed and whether policy has assisted some groups more, leaving others at a higher risk of falling into deep poverty. These questions are important given the changes in family structure in recent decades and the expansions in policies aimed at reducing poverty among specific groups, including seniors (such as Social Security and Medicare), working parents (for example EITC), and children (such as the Child Tax Credit [CTC] and the School Lunch Program).
Income below a poverty line is thought to be a statistical representation of an individual or family lacking the material resources required to meet their basic necessities over the course of a year. Deep poverty, defined as having resources less than 50 percent of the poverty threshold, represents a common measure of severe deprivation—the inability to meet even half of one’s annual basic necessities.
To properly measure trends in deep poverty over time as a marker of severe deprivation, we must first have an accurate measure of poverty. The United States has published official poverty rates for its population going back to 1959. [End Page 15] The original official poverty thresholds were based on the cost of a minimally adequate diet in the 1950s and the proportion of families’ budgets devoted to food, which at that time was one-third of the total budget (Fisher 1992). Since then, these thresholds have mostly just been updated for inflation, although some other minor changes have been made along the way.
As decades of research and commentary have demonstrated, the official measure of poverty used in the United States is deeply flawed (Blank 2008; Citro and Michael 1995). First, the poverty thresholds are outdated, as food no longer comprises such a large share of families’ budgets and other expenses like shelter have grown in importance (Hutto et al. 2011). This concern has led some to argue for a so-called relative or quasi-relative poverty threshold—one that changes over time as consumer expenditure patterns and living standards change (for a discussion, see Iceland 2005). Second, the American family has gone through tremendous changes over the past fifty years, with rising shares not only of single-parent families but also of cohabiting couples and cohabiting-parent families (Cancian and Reed 2009). This is problematic from a poverty measurement perspective since the official measure considers only those related by blood, marriage, or adoption as the unit sharing resources—that is, as “family” (see Provencher 2011). Third, and most important from our perspective, the official measure fails to count many of the resources devoted to alleviating poverty in the United States; these include near-cash or in-kind benefits like SNAP benefits and housing assistance as well as benefits that reach families through the tax system, like the Earned Income Tax Credit and the Child Tax Credit.
To remedy these and other deficiencies with the official measure, the National Academies of Science convened a panel of experts in the mid-1990s to recommend changes to the nation’s poverty measurement system (Citro and Michael 1995). The panel’s landmark report made numerous recommendations for improving the measurement of poverty, including innovations designed to reduce or eliminate the deficiencies noted here. Over the subsequent fifteen years, researchers at the U.S. Census Bureau and the Bureau of Labor Statistics (BLS) and in academia and think tanks experimented with measures based on these recommendations. In 2010 the Interagency Technical Working Group (ITWG) formed from across a number of government agencies issued a report with formal recommendations for the creation of a new Supplemental Poverty Measure (SPM) that the Census Bureau would publish each year alongside the official measure, in collaboration with the BLS and other agencies (ITWG 2010). Starting with the calendar years 2009 and 2010 (Short 2011), the Census Bureau began formally releasing the SPM in 2011, with annual releases thereafter.
As of this writing, the SPM has been released by the Census Bureau only for the calendar years 2009 through 2013, for reasons that are primarily technical: all of the data required to compute the SPM exist only for 2009 onward. This makes the SPM, for all its methodological improvements, inadequate for assessing long-term historical trends in either poverty or deep poverty. To fill this gap, in past work we have constructed an alternative time series using a newly developed measure that we call the historical Supplemental Poverty Measure, for all years between 1967 and 2012. The historical SPM time series attempts to implement the SPM in a consistent way over time to the best of our abilities given available data. In two recent papers using our historical SPM (as well as an alternative version of the historical SPM that uses an absolute or “anchored” poverty threshold) (Fox et al. 2015; Wimer et al. 2013), we have found that long-term trends in poverty as measured using the historical SPM are more favorable than official statistics would suggest. We find that much of the progress made in reducing poverty over the past fifty years, especially in recent years, is a result of government policies and programs, and especially those very programs not counted in official poverty statistics (with the notable exception of Social Security, which has reduced elderly poverty substantially and is included in the official poverty measure). We have also found that, regardless of whether we use a relative or anchored poverty threshold, deep poverty rates under our historical SPM time series have [End Page 16] been fairly flat since the 1960s, again largely as a result of resources coming from government policies and programs.
In this article, we explore long-term trends in deep poverty in more detail, taking a particular look at changes in family structure and employment, as well as government policies and programs. Our central questions are: (1) How have deep poverty rates changed for different types of families, and in particular for families with children? (2) What would deep poverty rates among families with children look like over time absent changes in family structure and changes in employment patterns? (3) How would deep poverty rates for different family structure and employment subgroups look absent accounting for government policies and programs? (4) What do the trends imply for the changing composition of the deep poor over this period?
data and methods
The data come from the 1968–2013 Annual Social and Economic Supplement (ASEC) to the Current Population Survey (CPS), also known as the March Supplement. It is important to note, as discussed in more detail later in the article, that this is a household-based survey of the non-institutionalized population. As such, it does not enumerate or capture some of the most severely disadvantaged individuals in American society, such as the homeless, the incarcerated, and those living in group housing in its many forms. Each survey covers income and associated topics in the prior calendar year, so these analyses cover the years 1967 to 2012. All figures are created using centered three-year moving averages, so our analysis covers the calendar years 1968 to 2011. We augment the annual CPS files to create our historical SPM series using information from the 1960–1961, 1972–1973, and 1980–2012 Consumer Expenditure (CEX) survey—a national survey tracking Americans’ expenditures in a comprehensive variety of domains—as well as administrative data sources where necessary. Here we outline our approach to constructing the historical SPM time series, including the creation of poverty resource-sharing units, historical SPM poverty thresholds, and SPM resources. For a full accounting of all the methodological choices underlying our historical SPM series, see Fox et al. (2015) and its detailed technical appendix.
To construct a historical SPM time series, the first step is to create a historically consistent poverty unit, which is the unit within a household deemed to be sharing resources to meet routine needs and expenses. Under the official measure, the poverty unit is the family, or anyone in the household related by blood, marriage, or adoption. The SPM makes a number of departures from this definition of the unit, in particular by including cohabiting unmarried partners together in the same unit, as well as by attaching unrelated children and foster children under the age of twenty-two to the household reference person. (For a full discussion of these issues, see Provencher 2011).
Constructing these poverty units consistently back to 1967 is challenging, in that not all unmarried partners in the household were identified in the CPS until 2007, and no unmarried partners were identified in households before 1995. In addition, foster children were not identified in the CPS until 1988. While we make no attempts to find foster children prior to 1988, given their extremely small sample size in any given year, we do attempt to identify unmarried partners and their children. To do this we use the Census Bureau’s adjusted Persons of the Opposite Sex Sharing Living Quarters (POSSLQ) method. Lynne Casper, Philip Cohen, and Tavia Simmons (1999) define an adjusted POSSLQ household as one in which two unrelated adults (ages fifteen and older) of the opposite sex live together, with no other adults except relatives and foster children of the reference person or children of unrelated subfamilies. In our construction of poverty units and the poverty universe, we also exclude people living in group quarters (for example, college dormitories) in all years.
Under the SPM, the Bureau of Labor Statistics computes poverty thresholds on an annual basis using the most recent five years of CEX data (for details on the procedures for setting SPM [End Page 17] thresholds in the CEX, see Garner 2011). The BLS first selects all consumer units with exactly two children and then estimates their expenditures on a core set of goods and services that includes food, clothing, shelter, and utilities (FCSU). They then find the average of the thirtieth to the thirty-sixth percentiles of expenditures on this basket for three different groups, defined by their housing status: renters, owners holding a mortgage, and owners not holding a mortgage. These figures are then multiplied by 1.2 to account for other common necessities (such as toiletries).
To estimate these thresholds historically we use historical data from the CEX. Because the CEX became an annual survey in 1980, the first year we are able to estimate a historical poverty threshold similar to the BLS threshold is 1984, covering the years 1980 to 1984. For 1980–1983, we use sequentially fewer years of data in estimating thresholds, so our 1983 threshold is based on 1980 to 1983, 1982 on 1980 to 1982, and so on. Prior to 1980, there were only two CEX surveys, one in 1960–1961 and one in 1972–1973. We thus construct a threshold in each of those years and then interpolate thresholds in intervening years using the rate of change in inflation. We also deviate from the Census Bureau and BLS in not adjusting our historical poverty thresholds for geographic differences in the cost of housing prices, given the lack of consistent and comparable data on these costs back to 1967.
To give some context, in 2012 the deep poverty SPM threshold for a two-adult, two-child family was $12,529. This was based on a typical two-adult, two-child family in deep poverty spending an average of $418 per month on food, $50 per month on clothing, $180 per month on shelter, and $187 per month on utilities.
The SPM makes a number of changes to the definition of the resources available to meet the expenses deemed necessary in the poverty thresholds. First, it considers after-tax income rather than pretax income, both by subtracting federal and state income tax liabilities and payroll taxes and by adding any tax credits such as the Earned Income Tax Credit or Child Tax Credit. Second, it adds a variety of in-kind or near-cash benefits to the definition of resources: SNAP, the School Lunch Program, the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), the value of government housing assistance, and the Low-Income Home Energy Assistance Program (LIHEAP). Third, it subtracts some nondiscretionary expenses from resources, including medical out-of-pocket expenses and work and child care expenses. Following is a brief description of our approach to including these resources in the CPS.
Taxes: Census Bureau tax calculator estimates are available in the CPS back to 1980 (for the calendar year 1979). Prior to that, we rely on the TAXSIM program of the National Bureau of Economic Research (NBER) (Feenberg and Coutts 1993) to estimate taxes for 1967 to 1978. Details of the tax model can be found in Fox et al. (2015).
In-kind benefits: Of the five in-kind benefits we added to resources, only LIHEAP is measured in the CPS in all the years that the program existed. For certain years, then, we must impute benefits for the remaining four programs. For SNAP, data are not available prior to 1980 (for the calendar year 1979). We thus impute SNAP from the 1972–1973 CEX for all years between 1967 and 1978, constraining the imputation to specific percentages of households based on the percentage of households receiving SNAP in 1980 (1979) and changes in SNAP caseloads between 1967 and 1979. We then estimate values for imputed recipients using distributions of 1972–1973 values, adjusted for inflation in a given year. A similar approach is used in the imputation of the School Lunch Program (also prior to 1980 ), housing assistance (prior to 1976 ), and WIC (prior to 2001 ). Values for school lunch are estimated in a similar manner to values for SNAP. Values for housing assistance are based on estimated household rental payments and the difference between estimated rental payments and the shelter component of the poverty threshold. Values for WIC are [End Page 18] estimated based on annual administrative data.
Nondiscretionary expenses: Medical and child care expenses have been measured in the CPS only since 2010 (for calendar year 2009). Other work-related expenses (such as commuting costs or uniforms) are always estimated in the CPS and never directly measured, even in the Census Bureau’s current SPM estimates. Thus, we impute medical and child care estimates for the entire time series and similarly estimate other work-related expenses for the entire time series using Census Bureau methods. Taking work expenses first, we estimate these as 85 percent of the median weekly work expenses calculated in the Survey of Income and Program Participation (SIPP) and then multiply by the number of weeks worked for each worker in the CPS. (Census Bureau researchers provided us with a historical table of these values going back to 1997.) We then calculate these values back to 1967, using changes in inflation. For medical expenses, we impute values from the CEX, attempting to mimic the distributions of medical expenses for key groups defined by income, number of elderly members of the poverty unit, and number of people in the poverty unit. For child care expenses, we take a similar approach, but first impute the incidence of child care expenses for units with children. Following the SPM, work and child care expenses are summed and capped at the level of the lowest-earning spouse’s or partner’s earnings. Because of the length of time over which we must impute and the lack of good benchmarks against which to assess them, our imputations in particular should be interpreted with caution. It is worth noting, however, that our main results are the same with or without the exclusion of medical and work and child care expenses from resources, at least in terms of the trends if not the overall levels.
Family Type, Family Structure, and Employment Status
We examine all three of the key constructs that we use to explore deep poverty trends—family type, family structure, and employment status—at the SPM-unit level. For family type, we define three mutually exclusive groups. We first identify the presence of working-age family members (ages eighteen to sixty-four) in a unit and then divide those units into families with children and those without children. The third category includes those families with no eighteen- to sixty-four-year-olds—these are elderly- only families. (The small number of SPM units with all members under the age of eighteen are dropped from the analyses.) Within these three primary family types, we define family structure by whether the unit is married, cohabiting, or single. Thus, if anyone in the SPM unit is married, we code everyone in that unit as residing in a married family. If no one is married but a cohabiting couple is part of the unit, the unit is coded as cohabiting. The remaining families are coded as single-headed families.
For employment status, we consider four mutually exclusive groups, focusing on units with at least one eighteen- to sixty-four-year-old member: units where all working-age adults are working full-time, full-year (defined as thirty-five hours or more per week for at least fifty weeks a year);1 units where all working- age adults are working, but at least one is not working full-time, full-year; units where at least one working-age adult is working but at least one working-age adult is not working (a status that includes both unemployed workers and workers out of the labor force for any other reason); and units with eighteen- to sixty-four-year-olds present but none are working.
We first present three rates of deep poverty: (1) for the overall population, (2) by family type, and (3) among families with children for key family structure and employment subgroups. We then present a formal “decomposition” of the role of family structure and employment [End Page 19] status in explaining long-term trends in deep poverty among families with children. This decomposition is followed by an assessment of the role of policies and programs in reducing poverty rates across family types and across subgroups of families with children. We conclude by documenting the changing composition of the deep poor that results from the trends that we detail. We use centered three-year moving averages for all figures.
Trends in Incidence of Deep Poverty
We begin by showing trends in deep poverty for the total population, with and without the inclusion of resources stemming from government policies and programs. Overall, the post-tax-and-transfer deep poverty rate in the United States has been fairly constant over the past fifty years, remaining around 5 percent of the non-institutionalized, civilian population. Over this time the role of government taxes and transfers in alleviating deep poverty has grown (see figure 1). Without these programs, the rate of deep poverty would have increased from 12.8 percent to 18.7 percent from 1968 to 2011 and also would have been more volatile over the time period.
While the overall incidence of deep poverty has been relatively unchanged over the past fifty years, different groups have experienced differing trends. Table 1 shows deep poverty rates for a number of demographic groups. This basic demographic analysis shows that the risk of falling into deep poverty has changed considerably for various subgroups over time. In 1968 elderly units and single-parent families with children were most likely to fall into deep poverty, but by 2011 working-age families without an employed adult had substantially higher deep poverty rates than any other group.
Since 1968, deep poverty rates for working-age families with or without children have been relatively constant, while for elderly families with no working-age adults present, there was a sharp decline in the deep poverty rate up until about the mid-1980s, followed by a gradual rise. Looking at rates by race-ethnicity, we can see [End Page 20]
[End Page 21]
that while deep poverty rates for whites have been fairly constant, there has been a considerable decline for blacks.
Focusing on families of working-age adults with children, we next examine trends in deep poverty by family structure. We find that single parents with children have experienced large declines in the likelihood of deep poverty; their deep poverty rates decreased from 19.0 percent in 1968 to 12.0 percent in 2011 (see figure 2). However, much of this decline occurred prior to 1977; deep poverty rates for this group have been relatively flat since then. Deep poverty rates for cohabiting and married families with children exhibit much less change over the period. While rates for cohabiting families with children have fluctuated a bit more than for married families, it is worth noting that this is a rather small group in the CPS, especially in the early portion of the time series.
Looking at deep poverty rates by family employment status among families with children, we find that families without an employed adult are almost twice as likely to fall into deep poverty as they were fifty years ago: the deep poverty rate increased from 21.6 percent in 1968 to 39.6 percent in 2011 (see figure 3). Meanwhile, families with all adults employed full-time, full-year, have had consistently low rates of deep poverty—between 1 and 2 percent. For families with all adults working and at least one working part-year or part-time, deep poverty rates fell over the period, from about 5 percent in 1968 to 2.3 percent in 2011. And deep poverty rates for families with some but not all members working have been essentially flat over the time period. In the next section, we examine how changes in family structure and employment have interacted to produce long-term trends in deep poverty, focusing specifically on families with children.
Decomposing Deep Poverty Trends for Families with Children
By estimating the share of individuals in families with children who would fall into deep poverty by alternately holding family structure and employment constant, we can estimate the role of each in accounting for the total change in deep poverty rates from 1968 to 2011. To estimate the rate of deep poverty [End Page 22] in 2011 holding family structure constant, we use the family structure distribution (share of families with children living in single, cohabiting, or married families) from 1968 and apply this to the rates of deep poverty for each subgroup in 2011 to arrive at an estimate of what the overall deep poverty rate would have been in 2011 had the family structure of families with children stayed the same as it was in 1968 but subgroup poverty rates changed. We generate similar predicted values by estimating the impact of changing employment patterns as well as by examining corresponding predicted values in 1968 using distributions from 2011 and rates from 1968. Consider the following equation:
where Yj is the overall deep poverty rate for working-age families with children in year j, for j equal to either 1968 or 2011; β is the share of the population in a given demographic group d (either by family type [single, cohabiting, or married] or by employment status [all employed full-time, full-year; all employed at least part-time; at least one not employed; all not employed]), and γ is the deep poverty rate for a given demographic group d.
The share of individuals who would have fallen into deep poverty in 2011 if the distribution of either family structure or employment status had remained at 1968 values can then be expressed as
and the share in deep poverty if deep poverty rates had remained constant within demographic groups would be
These counterfactuals can be compared with actual 1968 and 2011 values to indicate the role of changes in family structure or employment patterns in accounting for the total change in the rate of deep poverty from 1968 to 2011. [End Page 23]
Table 2 details this decomposition and shows that changes in family structure, absent changes in family work patterns, would have led to increasing rates of deep poverty for families with children. If the share of individuals in single, cohabiting, or married-couple families had remained constant from 1968 to 2011, the rate of deep poverty for families with children would have declined 1.5 percentage points, from 4.8 to 3.4 percent. If, however, the employment status of families had remained constant while family structures changed, the rate of deep poverty would have been unchanged. Specifically, despite the increase in single-parent families (from 8.4 percent to 20.9 percent of the population), the overall rate of deep poverty among families with children remained constant as the share of the population in families with all adults employed full-time, full-year, increased as well (from 10.9 to 25.4 percent). A key factor here is the increased full-time, full-year employment of single parents (from 23.9 to 32.4 percent) (see appendix figure A1). We find similar patterns if we reverse the decomposition and impose 2011 demographics on 1968 poverty rates, although in this case we find that changing family structure absent changes in employment would have led to a higher deep poverty rate (6.8 percent).
The lower panel of table 2 examines the same relationships, but instead decomposes the pretax and pretransfer rate of deep poverty. Absent government taxes and transfers, the rate of deep poverty would have increased from 8.8 to 14.3 percent. We find that changes in family structure account for most of this increase (4.2 out of 5.0 percentage points). Taken together, the results in table 2 suggest that changes in family structure alone would have increased deep poverty among families with children, but that these trends were offset by both changes in the antipoverty effects of government policies and, to a much lesser degree, changes in family employment patterns. [End Page 24]
The Role of Government
The decomposition results in table 2 suggest that government policies and programs have largely offset a rise in deep poverty that would have occurred given changes in family structure over recent decades. The next set of analyses therefore focus specifically on the role of these policies and programs in reducing estimated poverty rates. Table 3 shows trends in deep poverty over time for key family types and subgroups with and without the inclusion of resources from government programs. The role of government over time operates in different ways for different groups. Overall, as we saw in figure 1, deep poverty has been flat over the period, but absent government transfers it would have actually risen by nearly six percentage points.
Looking at family type, we see that for working-age families with children, deep poverty absent government transfers would have risen by five percentage points over the period, but after including transfers the deep poverty rate in 2011 was almost the same as our estimate for 1968. A similar story is evident for working-age families without children, where we see what was about a one-percentage-point rise in deep poverty over the period but would have been a rise of about five percentage points absent government transfers. For the elderly, deep poverty rates fell both with and without government transfers, but we note that without government transfers (Social Security), deep poverty rates would have been extremely high for this group in all years.
In the third panel of table 3, we focus on working-age families with children, comparing the role of government taxes and transfers for single, cohabiting, and married-parent families. Among single-parent families, pre-tax-and-transfer deep poverty rates fell fairly consistently over time, by about 11 percentage points. After including government transfers, deep poverty rates for this group fell between 1968 and 1978, but then stayed fairly flat at between 11 and 12 percent. Thus, government programs are reducing single-parent family deep poverty less in absolute terms over time, though before government programs are taken into account, single-parent families are less likely to be falling into deep poverty today than in the past.2 For both cohabiting and married-parent families, deep poverty would have risen more absent government taxes and transfers than we see after accounting for these, though the differences are not as substantial as they are for single-parent families.
In the last panel of table 3, we examine trends by employment status among working-age families with children. With or without government taxes and transfers, fully employed families exhibited declines in deep poverty rates over the period, though for families with not all working-age adults employed full-time, full-year, deep poverty rates would have slightly increased over the period absent accounting for transfers. More interestingly, we see the importance of the safety net in blunting the rise in deep poverty that might have occurred for families with at least one adult not employed and for families with no employed adults. For those with at least one adult not employed, deep poverty rates would have risen by about eight percentage points absent government transfers, but remained essentially flat after including those transfers. For families with no adults employed, deep poverty rates would have risen from about 60 percent to nearly 90 percent over the period, whereas after including government transfers, deep poverty rates rose by about 18 percentage points, from 21.6 percent to 39.6 percent. While this is still a large increase in deep poverty rates over time (for a group shrinking in size in relative terms), the figures in table 3 show the growing importance of government programs in ameliorating their deep poverty rates.
Figure 4 presents counterfactual estimates for the rate of deep poverty for single-parent families with children in the absence of specific programs. We focus on three major sets of antipoverty programs—cash welfare, the EITC, and nutrition programs—and examine trends in the importance of these programs in reducing deep poverty for single-parent families with children. The antipoverty role of tax credits increased [End Page 25]
substantially after the expansions of the EITC in the early 1990s (Grogger 2004; Hoynes 2014). Around the same time, cash welfare decreased in importance following the federal welfare reform of 1996, which time-limited the program and added work requirements. Subsequently, caseloads dropped precipitously (Blank 2002). Nutrition assistance programs like SNAP have expanded dramatically in recent years following the decline in importance of cash welfare assistance (Ganong and Liebman 2013). We focus on single-parent families with children, a particularly vulnerable group and a key group affected by these policy changes.
Figure 4 shows that, overall, taxes and transfers have a considerably smaller effect on the deep poverty rate today than they did in earlier years. In 1968 taxes and transfers decreased the deep poverty rate from 46.7 to 19.0 percent, while in 2011 the reduction was from 35.4 to 12.0 percent. Part of the reduction in the role of government can be accounted for by a decline in the rate of pre-tax-and-transfer deep poverty among single-parent families. Up until the early 1990s, cash welfare played a major role in reducing the incidence of deep poverty among single-parent families, cutting deep poverty rates by more than half in many years. However, [End Page 27] even prior to welfare reform, the importance of cash welfare transfers had been steadily declining since the early 1970s. At the same time, the role of nutrition programs grew, peaking in importance in 1992, when it reduced deep poverty by more than ten percentage points. Since the mid-1990s, the EITC has consistently reduced deep poverty rates among individuals living in single-parent families by two to three percentage points. Overall, then, the declines in the importance of cash welfare since the 1970s have been offset by the EITC and nutrition programs, alongside greater pre-tax-and-transfer resources among single-parent families that have probably been driven by the increases in employment generated by the combination of the expanded EITC in the 1990s, the 1996 welfare reform that transformed AFDC into TANF, and the growing economy of the late 1990s. It is important to note, however, that the employed have benefited more than the non-employed (Moffitt 2014).
The Composition of Deep Poverty
In the final set of analyses, we examine how all of the trends detailed here have combined to transform the composition of the deep poor over time between the 1960s and today. Although the overall incidence of deep poverty has not changed much over the past fifty years, the demographic composition of the deep poor has changed. While the share of the deep poor living in elderly households has hovered around 10 percent for most of the period, the share of the deep poor in working-age households with or without children has varied over time, showing opposite trends: an increasing share of the deep poor are those without children, whereas the share of the deep poor with children has been declining (see appendix table A1). In 1968, 60 percent of the deep poor were in working-age families with children, 23 percent were in working-age families with no children, and about 15 percent were in elderly families. By 2011, the portrait had changed: 52 percent were in families with children, 36 percent were in working-age families with no children, and just over 10 percent were in elderly families.
Again focusing on working-age families with children, we next examine differences by family structure subgroup (single family, cohabiting family, and married family). As shown in figure 5, married families with children comprise a declining share of the deep poor families with children, whereas single-parent and cohabiting families have each been increasing in share. In 1968, the majority of deep poor families with children (about two-thirds) were married-couple families. By 2011, the largest group was single-parent families, who now make up fully half of the deep poor families with children. The increasing representation of single-parent families in deep poverty is primarily due to the increase in the prevalence of single-parent families as the risk of deep poverty among these families has declined over time (as shown in figure 2). The increasing share of cohabiting families in deep poverty, from close to zero in 1968 to about 10 percent in 2011, is due to increases in the prevalence of this family structure subgroup as well as increases in the risk of deep poverty for this group.
We next examine the distribution of the deep poor among families with children by family employment status, using the employment groups defined earlier. Not surprisingly, figure 6 indicates that having a job is an increasingly important factor in the composition of deep poverty. Working-age families without an employed adult represent a dramatically increasing share of the deep poor families with children since 1990, with the proportion at 22 percent in 1968, 34 percent in 1990, and 51 percent in 2011 (consistent with Moffitt 2014). For the group with all adults ages eighteen to sixty-four employed, the share in deep poverty steadily declined from 42 percent in 1968 to 35 percent in 1999, then rapidly declined to 17 percent by 2011.3
We further explore the relationship between employment and deep poverty by examining the prevalence of disability and low-wage work among the deep poor over time. First, we [End Page 28]
[End Page 29]
look at the share of the deep poor who lived in a family with at least one member self-reporting that he or she was not employed the previous year owing to illness or disability. In 1967, 10 percent of those in deep poverty lived in such a family, and by 2012 the likelihood of living in such a family had nearly doubled, to 18 percent. During this period, the overall rate of non-employment due to illness or disability increased from 4 to 10 percent. In other words, in 1967 one in ten of the deep poor lived with someone not employed owing to illness or disability. By 2012, that share had increased to one in five of the deep poor, mimicking general trends among the non–deep poor.
Second, we looked at low-wage work by first converting annual earnings into an hourly wage rate, then identifying workers whose estimated wage rates were $1 or less above the federal minimum wage. Looking at the share of the deep poor who had an employed adult earning $1 or less above the federal minimum wage, we found that this share has actually decreased slightly over time, from 36 to 31 percent of the deep poor population from 1967 to 2012. In 2012, deep poor families were much less likely to have an employed adult than they would have been in 1967, but if they had a worker they were much more likely to have a low-wage worker (71 percent in 2012 versus 43 percent in 1967). So overall, as a growing share of deep poor families become disconnected from employment, fewer of these families rely on a low-wage worker.
Taken together, the demographic characteristics of the deep poor have changed considerably over the past fifty years. In 1967, the typical person in deep poverty was living in a married family with children and at least one worker. By 2012, the typical person in deep poverty lived in a household without an employed adult and often with at least one adult reporting not working owing to illness or disability. The workers in the families of those living with employed adults most likely earned near-minimum wage.
Although our analyses provide a picture of deep poverty trends using a consistently and commonly measured indicator of severe deprivation in the United States, they do have a number of important limitations. First and foremost, an analysis of deep poverty using survey data necessarily misses a sizable portion of the deep poor population. The Current Population Survey is nationally representative, but it covers only the non-institutionalized population; thus, individuals living in prisons, mental institutions, and other institutional living quarters are excluded. Additionally, those living in group quarters are excluded from the poverty universe. Finally, homelessness or transitory living conditions make a sizable share of the truly disadvantaged inaccessible to a household survey. Other contributions to this volume shed considerable light on these important subpopulations.
In addition, the results presented here do not adjust for the underreporting of income. It is well known that the underreporting of benefits such as SNAP or food stamps and WIC has been growing over time (Meyer, Mok, and Sullivan 2009). Addressing this underreporting is an area of future research. In addition, income questions in the CPS focus on regular income from a variety of sources and are thus likely to miss some informal sources of support that may be critical to the severely deprived (see, for example, Edin and Lein 1997).
Following previous research on the Supplemental Poverty Measure, we value several forms of in-kind benefits (such as SNAP and the EITC) at their face value; however, doing so may overvalue some benefits for some families. One could argue that, since inefficiency is introduced with in-kind benefits and, as a result, recipients value them less than their face value, they should be discounted in a poverty resource measure. Similarly, one could argue that an annual lump sum payment like the EITC should be discounted owing to debt that might need to be serviced to smooth consumption of this income across a full year (not to mention the especially high interest rates that recipients may face for tax refund anticipation loans). Although we acknowledge that these are areas of future research, we follow current National Academy of Sciences (NAS) guidelines for the inclusion of these resources at full value in our measure. [End Page 30]
Finally, our analyses rely heavily on imputed values, which introduce an element of uncertainty to our results. Because many components of the SPM are not measured historically in the Current Population Survey, these imputations are essential to estimating historically consistent poverty rates. Although they add some uncertainty, many of the imputed programs were quite small during the periods of imputation, and those that are more substantial do not alter our understanding of trends over time. Owing to the complexity of these, we have not been able to estimate standard errors for our poverty results, and this is an important area for future research.
In addition to these limitations, we have been unable to address several important questions. While informative about snapshots of deep poverty, this article has not examined poverty duration. We have also not examined the severity of deep poverty among those experiencing it—that is, whether their incomes are just below 50 percent of the poverty threshold or considerably lower. Such analysis is potentially feasible using the income data from the CPS, although measurement may be particularly noisy for those with very low incomes (and may be confounded by the underreporting of benefits among this population).
Our analysis has four main findings. One, we find that while trends in overall deep poverty have been relatively flat since 1968, this constancy belies considerable change in the predictors and correlates of deep poverty. Families without an employed adult were much more likely to fall into deep poverty in 2011 than in 1968, whereas single-parent families are less likely to fall into deep poverty today than in the past. Two, despite the decline in the risk of deep poverty in single-parent families, their share in the deep poverty population has steadily increased on account of the rising proportion of single-parent households in the United States. Three, results from a simple “decomposition” analysis suggest that changes in family structure since 1968 would have increased deep poverty among families with children, but that these trends were offset by both changes in the antipoverty effects of government programs and, to a lesser degree, employment patterns. Four, governmental taxes and transfers reduce the risk of deep poverty for all subgroups examined, but the significance and role of these programs has changed over time. Specifically, our analysis points to a declining role for cash welfare and a growing role for nutrition and tax programs. We also find that for families without an employed adult, the antipoverty role of taxes and transfers steadily increased from 1968 through 1988, but has been declining ever since.
Although an increase in employment has made single-parent families with children less likely to fall into deep poverty today than fifty years ago, it is not clear that these families are unambiguously better off. This analysis of poverty provides insight into one dimension of family financial well-being. Previous research has found that increases in the female labor supply have had heterogeneous effects on total hours spent with children: single mothers may have reduced the time they spend with their children, but married mothers may have been able to preserve more of this time (Fox et al. 2013). In addition, stress may reduce the quality of the time that mothers spend with their children. Additionally, depending on their family’s access to quality, affordable child care, the children of working parents could experience either improvements or declines in well-being. To understand the full effect of changes in employment patterns, we would need to investigate a number of measures of child and family well-being. Finally, the shift from cash assistance to in-kind nutrition assistance and onetime tax refunds is also likely to have increased stress among mothers.
Taken together, our results suggest some fundamental shifts in the nature of deep poverty. Today fewer of the deep poor are elderly or families with children, but a growing share—now nearly 40 percent—are working-age adults without children, a group for whom the safety net is the thinnest. The makeup of deep poor families with children has also undergone striking changes. In 1968 the typical deep poor family with children was headed by a [End Page 31] married couple and had at least one adult employed—albeit possibly part-time, part-year—but in 2011 the typical deep poor family with children was headed by a single parent or had no adult employed, reflecting the fact that these latter two subgroups are those for whom deep poverty rates are particularly high (at 12 and nearly 40 percent, respectively). The net result of these changes is one of relative stability, with deep poverty rates rarely fluctuating much below or above 5 percent. This stability is largely thanks to the role of government programs and, to a lesser extent, employment, which have held deep poverty rates at bay.
It is debatable, of course, whether the overall long-term trend of stability in deep poverty rates is good news or bad news. Given macroeconomic growth, one could argue that deep poverty should have declined instead of remaining stagnant. Conversely, it is good news that even in steep economic downturns like the double-dip recessions of the early 1980s and the recent Great Recession, the safety net has held the overall deep poverty rate down. However, as we have seen, there has also been considerable change in the composition of the deep poor underlying this stasis, and the groups that are in deep poverty today, being more isolated from society, may be more difficult to target with policy interventions than the deep poor of fifty years ago. As the deep poor become increasingly isolated from employment, child care, and school systems and experience more and more limitations due to health or disability, designing popular programs to target this population will be a challenge.
Liana E. Fox is postdoctoral researcher at the Swedish Institute for Social Research at Stockholm University.
Christopher Wimer is research scientist and co-director of the Center on Poverty and Social Policy at the Columbia School of Social Work.
Irwin Garfinkel is Mitchell I. Ginsberg Professor of Contemporary Urban Problems at the Columbia School of Social Work.
Neeraj Kaushal is professor at the Columbia School of Social Work.
JaeHyun Nam is a graduate student at the Columbia School of Social Work.
Jane Waldfogel is Compton Foundation Centennial Professor for the Prevention of Children’s and Youth Problems at the Columbia School of Social Work.
1. Weekly hours are based on responses to the question of how many hours were usually worked per week in the preceding year. While prior to 1975 respondents were asked only about actual hours worked the previous week, not usual hours worked the previous year, they were also separately asked whether they worked full-time, part-time, or not at all in the previous year. We use the latter variable for classification for 1967–1974.
2. The decline in pre-tax-and-transfer single-parent poverty is largely due, of course, to increases in employment among this group, which is in turn affected by government policy.