The PSID in Research and Policy

First Important Contributions of the PSID to Policy Research: Event History Analyses

The first big policy and practice use of the PSID was the study of poverty and welfare spells, exits and entries. These studies found that a small number of families were both poor over the long term and habitual welfare users—this work was both supported by augmented literature on the intergenerational transmission of poverty and welfare, and the culture of poverty (Gottschalk and Moffitt 1994; Gottschalk et al. 1994). Ultimately, these studies helped lead to Personal Responsibility and Work Opportunity Reconciliation Act (PWORA) and the mid-1990s welfare reform legislation that did away with Aid to Families with Dependent Children (AFDC) and introduced Temporary Assistance for Needy Families (TANF) in its stead (Haskins 2001; more generally, Duncan and Chase-Lansdale 2001).

Early on, Coe (1978), Rainwater (1982), and others examined the proportion of persons who were poor by some definition over a fixed time frame, typically eight or ten years, tabulating how many people were poor for, say, ten out of ten years, or five out of ten, or one out of ten. The approach was very simple, and produced a set of facts much like other poverty studies. The more difficult steps are to determine the causes of the time people are in poverty and why they go off and on public assistance. Rodgers and Rodgers (1993) summarized the literature to date nicely by creating an additively decomposable index of aggregate poverty that showed that chronic poverty was a more serious problem than previously thought, as poverty not only increased, it became more chronic and less transitory in nature as we moved from the 1970s through the 1980s.

In one of the first analyses of poverty dynamics, Duncan (1984) found a high volatility into and out of poverty. Ellwood and Bane (1986) showed this to be an incomplete characterization: the long-term poor are a small proportion of all those who are ever poor, but a very high proportion of all who are poor at a given point in time. Literally thousands of students of public policy, poverty, and welfare have been taught these realities for the last several decades. The influence of these studies and the value of their reported results have been immense, and the “spell” approach initiated by Bane and Ellwood (1986) has spawned a large literature using hazard regressions to study poverty exits and durations. The innovative analytical design suggested by Bane and Ellwood (1986) led to their central insight that the majority of those poor at a given time are in long-term poverty, and the majority of the person-years of poverty are accounted for by the long-term poor. The determinants of long-term poverty could therefore be used by social policy-makers and practitioners to focus anti-poverty efforts on the longer term poor populations, such as undereducated and never married single parents. While the targeting was correct, so far the policy remedies have not been enough to reduce U.S. single parent poverty to reasonable levels (Maldonado 2017).

In time, this led to even more targeted PSID studies of welfare dependence, including Plotnick (1983), Ellwood’s classic book, Poor Support, in 1986, and ending with the 1996 Welfare Realities book by Bane and Ellwood, which appeared right after welfare reform was initiated in the United States, undoubtedly contributing to the research evidence that led to that dramatic policy change (Haskins 2001) .

Most recently, to answer questions about upward mobility for children from poverty for a Gates Foundation initiative, a new and important PSID-based study of child poverty transitions for all children born into the PSID from 1968 to 1985 concluded that following children from birth through age 17 reveals that 11.8 percent of children, nearly 9 million in total, are persistently poor, meaning they spend at least half their childhoods living below the poverty level. Children of color fare much worse than average, with just over 40 percent of black children being persistently poor, compared with less than 6 percent of white children. Among the persistently poor, only about 35 percent are white, the rest are black (56 percent) or another race or ethnicity (9 percent) (Kalish and Ratcliffe 2017). Given the vast demographic changes being experienced in this nation (Frey 2014), children of color will be the focus of child poverty alleviation efforts for decades to come.

Volatility, Instability, and Mobility

The study of income volatility, year to year, month to month, across and within generations is possible only with longitudinal panel data. And the PSID has made more seminal contributions to policy research in this area than any other dataset (Gottschalk, Moffitt, and Zhang 2018; Moffitt and Zhang, this volume). The long 50 year period renders the PSID advantageous for life cycle research on volatility and is very important for determining how “transitory” events like job loss affect future income; how they affect family labor supply and fertility, and if they are short-lived or long-lived. In particular, information on the presence or absence of a spouse or cohabiting partner permits the PSID to be used to study volatility among single mothers—a large and typically disadvantaged subgroup in the United States that is known to have high economic volatility. ³

Early findings by Smith and Morgan (1970), Benus and Morgan (1972), and Duncan et al. (1974) revealed a startlingly high level of dynamism and mobility, but also instability and turbulence, in the lives of American families. Benus and Morgan (1972), using the first four waves of the PSID, were the first to decompose earnings of the family head into several components in a simple version of an error components model. Shortly after, Lillard and Willis (1978) introduced a more sophisticated econometric technique for decomposing the variance and the field built from there. By far the best known and most cited articles are several contributions by Peter Gottschalk and Robert Moffitt, who for many decades have led the charge in the PSID use for volatility research (see, for example, Gottschalk and Moffitt 1994, 2009; Moffitt and Gottschalk 2001, 2012).

Defining volatility generally as the degree of change in a variable from one time period to a later one, the PSID has permitted studies of a wide variety of forms of economic volatility, including studies of individual or family earnings, family income, job mobility and labor market turnover, and employment persistence or change. The term volatility is non-normative because these variables’ implications for well-being depend upon whether they are permanent or transitory as well as how well they can be smoothed by assets or borrowing (Dynarski and Gruber 1997). However, the literature on volatility has also been characterized by the way family and individual change affects income position within generations (intragenerational mobility) or across generations (intergenerational mobility).

The PSID intragenerational mobility literature has focused on longer term movements up and down the income scale from a starting position. For instance, several studies have chronicled the way that “middle-class incomes” changed, with more upward than downward mobility during the 1970s and 1980s through the mid-1990s, to the opposite in more recent times (Bayaz-Ozturk, Burkhauser, and Couch 2013; DeBaker et al. 2010; Duncan, Rodgers, and Smeeding 1993; Duncan et al. 1996; Dynan et al. 2012; Latner 2018; Shin and Solon 2011).

Permanent income changes were generally positive before 2000, but then became increasingly negative during and after the Great Recession. Moffitt and Zhang (this volume) interpret the results to suggest that neither the permanent nor the transitory variance departed substantially from a process with fluctuations around a stable trend until 2008, when increases in both truly started to emerge, as we entered the Great Recession. And now, as we exit this turbulent period, both variances also show signs of starting to decline from their recession peaks. Finally, new efforts are underway to link the concepts of instability and mobility in policy relevant ways (see Latner 2018).

Many of the same income dynamics are interpreted by others as signs of economic instability, especially where one finds downward mobility or increased variability at lower income levels. The evidence is that lower income families cannot in fact easily smooth consumption, and hence find negative effects on child well-being, family status, and residential change (Hannagan and Morduch 2015; Hill et al. 2013, 2018). Policy implications include mechanisms to shore up incomes and put floors under unexpected income losses (Halpern-Meekin et al. 2018)

International Comparisons and Policy Effects

Domestic panel data research is parochial in that it focuses only on the institutions, dynamics, and policy effects in one nation, ignoring the multiple ways that different nations adopt similar and different policies that achieve different important social outcomes. To achieve comparability, researchers first have to make the individual national datasets comparable by “harmonizing” existing datasets. Harmonization of similar household income datasets means redefining and reorganizing demographic, social, and economic variables across different nations to make data more comparable. ⁴ In the early 1990s, work was undertaken to harmonize the PSID and its German and British equivalents to create harmonized panel income datasets, which created the Cross-National Equivalence File (CNEF; see Burkhauser and Lillard 2007). ⁵ Since that time, PSID data have also played an increasingly important role in international comparative research, most of which has important policy implications. Several recent studies have used PSID and European panel data to conduct comparative analyses of economic indicators, such as poverty, income inequality, and the effects of public policies in the United States with other countries, such as Germany and Belgium (Van Kerm 2004), the Netherlands (Headey et al. 2004), and Sweden (Evertsson and Nermo 2004). Couch and Dunn (1997) used PSID and German panel data to calculate comparable measures of intergenerational correlations of earnings, hours, and education in Germany and the United States, and found that the greater overall U.S. wage dispersion primarily reflects the substantially greater compression at the bottom of the wage distribution compared with other countries. Duncan et al. (1994) examined comparative poverty and social-assistance dynamics in North American and European countries. Many findings are quite surprising; for instance Burkhauser et al. (1991) found that German women do worse than American women after marital dissolution. Bjorklund and Jantti (1997) used the PSID to study intergenerational income mobility in Sweden and United States—one of the first articles in a now burgeoning field (Jantti and Jenkins 2009). There is also a substantial amount of comparative household and individual income dynamics and intragenerational mobility (e.g., Zhang 2010), much of it nicely summarized in Burkhauser and Couch (2009).

The rich PSID data collected on wealth was also used to document large differences in the household wealth distribution in the United States and the UK at the top fifth of the wealth distribution (Banks et al. 2003). Even after controlling for age and income differences between the two countries, the authors showed that the median U.S. household accumulates more financial wealth than its UK counterpart. The CNEF could contribute greatly to our understanding of wealth dynamics in comparative perspective were it to harmonize wealth concepts and the transfer of wealth across generations as well as it has done with incomes and income transfers.

In a study of changes in a woman’s economic well-being as a consequence of widowhood, Burkhauser et al. (2005) used data from four countries, including U.S. PSID data. This work documented substantial differences among the United States, UK, Germany, and Canada in sources of replacement income, yet it showed that the size and distribution of income replacement in these countries were actually quite similar across age and prior-widowhood income level. The study highlighted the importance of including private sources of income in analyses of income replacement, especially in countries such as the United States, and suggested that widows are fairly well protected against income loss in all these countries.

Health Mobility and Health Insurance Effects on Health Outcomes

Health status and health insurance coverage are also best addressed with panel data. From the Barker hypothesis about the long-term effects of birth and early life conditions on health, to end-of-life care issues, panel data allow one to monitor and observe health status as it changes for the same individuals (Almond and Currie 2012; Barker 1995; Barker et al. 2002; Wolf 2018). ⁶ More recently, the PSID data have been used to examine the transmission of health status across generations, beginning with Davis, McGonagle, and Schoeni (2005) and, more recently, Halliday, Mazumder and Wong (2018), whose work investigates intergenerational health mobility using self-reported health status, a health measure used widely in epidemiology research, which has been shown to be highly predictive of mortality and that the PSID has captured since 1984.

Given frustrations with our ability to make significant changes in economic mobility, the possibility that policy can affect health mobility is important. Indeed Halliday, Mazumder, and Wong (2018) found that there is substantially greater health mobility than income mobility in the United States. Evidence suggests that social institutions and policies may be more successful at breaking the intergenerational link in health than they are in income (Wang, Wang, and Halliday 2018). Further, evidence suggests that the racial mobility gap in health is both smaller than the analogous racial mobility gap in income and that the health gap is being reduced among younger generations of Americans (e.g., Currie and Schwandt 2016). Wang, Wang, and Halliday also find that early life circumstances affect health mobility. They conclude that close to 40 percent of intergenerational health persistence can be explained by early childhood circumstances, which vary with the socioeconomic background of the parents, in explaining the transmission of health status across generations.

One policy mechanism in particular, access to health insurance, might play a large role in reducing the intergenerational transmission of health status. To examine this question, Halliday, Mazumder, and Wong (2018) used PSID data on health insurance coverage for household members, controlling for family income and parent education. They found that there is greater persistence in health status among families where parents did not have health insurance, thereby providing suggestive evidence that health insurance might be part of breaking the link in health across generations.

Others have also studied the dynamics and consequences of health insurance coverage in the PSID, including Medicare (Cogan et al. 2010; Kemper et al. 2006; McClellan and Skinner 2006; U.S. Department of Treasury 2009). ⁷ Because PSID data cover several decades and the full life course, they offer a unique resource for studying such dynamics. A notable high-profile study by the Department of Treasury (2009) found that as many as 4 percent of nonelderly Americans and 5 percent of those under 21 years of age are uninsured at some point over a 10-year span—a substantially larger fraction than over a shorter period and much higher than the 11–12 percent typically cited as being uninsured at a point in time using cross sectional data (Barnett and Berchick 2017).

McClellan and Skinner (2006) used the PSID to examine distributional transfers associated with Medicare in the context of a lifecycle model. The authors found that the distributional consequences are much larger than a simple (annual) accounting exercise would suggest, with a net benefit flowing to lower-income households. The PSID has also been used to study the interplay of Medicare coverage for individuals with disabilities with the private insurance market. Cogan et al. (2010) found that removing high-cost individuals from private insurance markets leads to greater coverage of individuals who are similar but not as high cost. And the PSID has also allowed for the exploration of long-term care expenses over the life course. Kemper et al. (2006) have found that people turning 65 will need long-term care for three years on average, but with a wide difference between men’s and women’s use. The authors document that an important share of needed care will be covered by public programs and some private insurance, but much of the care will be an uninsured private responsibility of individuals and their families.

Researchers are also beginning to use the health data in conjunction with the socioeconomic data to examine the extent to which transmission of health across generations within the same family accounts for the transmission of socioeconomic status, and vice versa (Johnson and Schoeni 2009). To look within generations of the same family, a large number of adult siblings are interviewed independently in each wave. Several recent articles have used this unique feature to control for unobserved family effects in models examining the link between childhood family structure and adult outcomes (Boisvert and Wright 2008; Conley et al. 2007; Conley 2008; Cunha et al. 2005; Duncan et al. 2010; Metzger and McDade 2009; Vartanian and Buck 2005).

Child Development: Life Events, Family Change, and Public Policy

The effects of family change, economic change, and instability on children are a major policy concern. There are several key findings in the policy literature that could not have been obtained without following parents and children over a long period of time in the PSID. These include classic findings that have influenced social policy: that parental divorce is harmful for kids (e.g., McLanahan and Sandefur 1994); that teenage childbearing is likely to be harmful for the youngest mothers (e.g., Hoffman et al. 1993), that neighborhoods have effects on children’s development independent of family characteristics (Brooks-Gunn et al. 1993); and that family incomes matter for child development (Duncan, Kalil, and Ziol-Guest 2010, 2018; Duncan et al. 1998).

Several recent articles illustrate the unique value of PSID data for understanding the long-term consequences of policy changes on youth. I first focus on seminal work by economists and then the sociologists and developmental psychologists who began the CDS.

The first key exemplar is the work of Jackson et al. (2017) and Jackson and Johnson (2018), who use PSID geocode files to link children to school districts, to compare those who attended elementary and high school before and after court-mandated school finance reforms that began in 1971. Unlike most evaluations of school finance reform that focus on test scores and lose track of the students after graduation, Jackson et al. followed them into adulthood. They found that court-mandated funding reforms, such as state school revenue equalization formulas, which increased resources and school spending in poor school districts, had substantial effects on increasing completed education rates and adult earnings for low-income children in those formerly underfunded districts, especially for black children.

Equally important, Hoynes et al. (2016) employed a similar strategy to evaluate long-term impacts of food stamps. Based on the gradual expansion of the program across counties between 1961 and 1975, they found that access to food stamps in early childhood leads to a significant reduction in an index of poor health conditions (obesity, high blood pressure, and diabetes) when those children reach adulthood. For women, there was also an increase in a separate “self-sufficiency” index. Both articles exploit information on family economic conditions at the time of the policy “treatment” to check that effects are concentrated among those from the poorest families.

Starting in 1997, additional data were collected about children aged 0–12 and their parents, with follow-ups of these families in 2002/3 and 2007/8 in the Child Development Supplement (CDS) to the PSID. ⁸ With the addition of the CDS, the PSID has begun to make substantial contributions to the child development literature and how programs and policies affect incomes, living arrangements, and child outcomes. Recent policy relevant work includes the examination of children’s time use (e.g., Sandberg and Hofferth 2001), fathers’ involvement with children (Marsiglio et al. 2000), the connection between TV, media, and obesity (Vandewater et al. 2004), the effects of welfare reform on children (e.g., Hofferth et al. 2000), and determinants of child achievement (e.g., Yeung and Conley 2008; Hofferth and Reid 2002; Loveless 2003).

An important focus of families, and those who study families, is the set of investments that parents make in the development of their children. Del Boca et al. (2014) estimate a structural model of investments in children and labor supply decisions of parents, combining data from the CDS and the main PSID interview. They found that mother’s and father’s time spent with children is important to their cognitive development, and that financial resources are moderately important for older children. Kalil et al. (2014) assessed the importance of family structure as a determinant of child investments, highlighting the roles of noncustodial parents, step-parents, and grandparents compared with biological parents who are married or cohabiting. Focusing on how parents’ influence their children’s time use, Weininger et al. (2015) found that mother’s education has a large influence on children’s time in organized extracurricular activities, which have been shown in other work (Kaushal et al, 2011) to have positive impacts on adult outcomes, whereas income, wealth, and mother’s work hours have more modest effects.

Finally, Newman and Chaloupka (2013) found higher cognitive development in children whose families are in public housing, owing to the fact that rent is limited to 30 percent of income. While the exact mechanisms that connect reduced rent and school performance are not identified, the study breaks new ground on the effects of housing policy on child outcomes.

Intergenerational Social and Economic Dynamics

One of the fundamental questions facing the United States today is whether there is a sufficient degree of equality of opportunity. Do all children, irrespective of their parents’ socioeconomic status, have the same chances of achieving lifetime economic success? Or are there fundamental disparities that persist from generation to generation, creating dynastic patterns of separation in society that will ultimately lead to an even more unequal society? Researchers who would have asked that question 50 years ago would have very little to say, as there was no nationally representative intergenerational sample that would allow for a fully satisfactory answer to that question. However, with the introduction of the PSID in 1968, along with other data sources, we have actually learned a great deal about intergenerational persistence and intergenerational mobility in the last few decades (Mazumder, this volume).

The ability to track individuals as they form new households and the broad coverage of many dimensions of socioeconomic status over large portions of the lifecycle make the PSID especially ideal for studying intergenerational mobility. The economics and policy literatures have used these key aspects of the data to substantially contribute to the modern view of the United States as one of the least economically mobile countries among advanced economies. ⁹ A fairly broad consensus has emerged that intergenerational income mobility is actually relatively low in the United States, especially when compared with other advanced economies (e.g., Corak 2013). This suggests that the view of the United States as a highly mobile society and a “land of opportunity” may be unwarranted.

The fact that both parents and their adult children are interviewed in the PSID has supported extensive analyses of the intergenerational transmission of economic status as measured by earnings and income and the mechanisms behind intergenerational persistence and how persistence has changed over time. ¹⁰ The PSID has also been instrumental in contributing to our understanding of many other dimensions of intergenerational mobility including occupation; wealth; education; consumption; health; and group differences in mobility by gender, race, and region. Literally hundreds of publications have used the PSID to examine intergenerational effects on socioeconomic status (Eberharter 2008; Gouskova et al. 2010a; Lee and Solon, 2009; Mayer and Lopoo 2005; Vartanian, Buck, and Gleason 2007), and philanthropic behavior (Wilhelm 2008). Policy researchers using the PSID to study intergenerational mobility have been able to exploit individual-level data on educational attainment to highlight the importance of human capital investment as an important transmission channel of mobility and an area on which policy-makers should focus attention (Altonji and Dunn 1996; Reeves and Venator 2014) .

Among all of these, wealth is rising to the top as a dynastic force that limits intergenerational mobility and preserves economic status. Recent research by Pfeffer and Killewald (2018); Pfeffer (2018); and Fisher et al. (2017) builds on the extensive PSID work on intergenerational income mobility that began in the early 1990s (e.g., Corcoran et al. 1992; Solon 1992). Wolff and Gittleman (2011) found that inheritances are an important component of wealth for the 30 percent of households that receive them, but found little evidence of an inheritance “boom,” largely because inherited wealth comes much later in the life course. This finding and others led to the new intergenerational transfer module for the PSID to capture inter vivos transfers across generations (Schoeni et al. 2015). The patterns of intergenerational transfers at key times in children’s and young adult’s lives may go a long way toward understanding patterns of integrational persistence at the top and bottom of the distribution (Gornick and Smeeding 2018; Pfeffer 2018; Smeeding 2016a, 2016b).

Other articles have expanded on the intergenerational transmission theme in recent years to explore the intergenerational persistence and mobility of lifetime earnings (Moffitt and Gottschalk 2002, 2012). For instance, Hendricks (2007) found lifetime earnings are substantially more persistent across generations than suggested by previous estimates based on short panel data. Focusing on differences in intergenerational wage and earnings elasticities of adopted and biological children, Liu and Zeng (2009) have found that the earnings correlation across generations would be 50 percent smaller if inheritable ability were removed. They concluded that inheritable ability plays a very important role in the intergenerational transmission of earnings.

A few published studies using the PSID have also attempted to address the question of trends in mobility, including Mayer and Lopoo (2005, 2008), Hertz (2008), and Lee and Solon (2009). The latter two studies, which used a wide range of PSID cohorts, concluded that there is no evidence of a change in trend. In contrast, a few recent studies using PSID data (Hartley, Lamarche, and Ziliak 2017; Justman and Krush 2013; Justman et al. 2017) show evidence of a striking increase in intergenerational income persistence.

Research Frontiers: Multigenerational Processes and Links with Administrative Data

There is also growing recognition in the literature that there may be important mobility processes that are ignored by focusing only on transmission across one generation. The empirical interest in three generations was sparked by Mare in his PAA presidential address (Mare 2011). Several studies have exploited the fact that the PSID now extends beyond two generations to examine multigenerational processes with respect to education (Pfeffer 2014a, 2014b; Song 2016; Song and Mare 2017), occupation (Hertel and GrohSamberg 2014), income (Wightman and Danziger 2014), and wealth (Pfeffer and Killewald 2018; Pferrer 2018: Killewald, Pfeffer, and Schachner 2017). Cross-national studies across multiple generations are already appearing based on the CNEF (Neidhöfer and Stockhausen 2018). As a whole, these studies demonstrate that we may gain a deeper and more accurate understanding of the degree of socioeconomic mobility and important differences by race and SES by looking beyond two generations. This is a clear research direction for the PSID as it matures into its sixth decade.

An extremely valuable resource for researchers would be greater linkages between the PSID and other data sources under secure protocols that continue to protect respondent confidentiality. For example, linkages to birth records, health records, school records, government programs (e.g., Medicaid, food stamps), credit bureau data, banking records, and income tax records would dramatically improve the value of the data. Such linkages would facilitate much more research on exposures that may influence intergenerational processes, and it would allow researchers to study new outcomes. In addition, with such linkages in hand, researchers could better understand the potential problems with both survey and administrative data, and consider ways to combine them to produce the most robust results concerning intergenerational mobility.

Recently, researchers with access to administrative data files have provided new estimates of intergeneration mobility (Chetty et al. 2014; Mazumder 2005a, 2005b; Mitnik et al. 2015). But already it appears that, once one standardizes design issues such as the ages at which parents’ and children’s income is measured, PSID estimates line up well with those from administrative data (Mazumder 2015). Moreover, the PSID provides additional leverage for understanding intergenerational outcomes that administrative data do not—correlations in consumption and wealth, as well as earnings (Charles et al. 2014; Killewald, Pfeffer, and Schachner 2017; Pfeffer and Killewald 2015, 2018). It also provides a much richer menu of explanatory variables to begin the process of understanding the forces that produce these correlations.

There are many ways in which PSID data linked to administrative data could improve inferences from both data sources. Links to administrative records would allow researchers to better understand the importance of attrition in the PSID and to improve methods to correct for bias. On the other hand, self-reported data from the PSID on intergenerational family linkages can be compared with administrative data linkages using protected identity keys (PIKs) to compare the efficiency of PIKs to actual records based on following the same families for decades (Johnson et al. 2015). Further, administrative data on the income of nontax-filers may better allow us to understand what happens when tax data are missing. In addition, tax records are not ideal for identifying family relationships since they rely on whether children are claimed as dependents for tax purposes at a point in time. With the increasing complexity of family arrangements in the United States, the ability to track all the parental figures over the life course may loom as a major challenge in intergenerational mobility research going forward, and one in which the PSID can play an important role.

The PSID has remained a valuable vehicle for evidence-based policy research for decades and should remain so for many more. Even simple studies of mobility into and out of poverty that were at the forefront of social policy analysis in the era of welfare reform can be brought up to date and serve as prima facie evidence for debating patterns of mobility out of poverty in the twenty-first century (Ratcliffe and Kalish 2017). In our current era of “alternative facts” (Conway 2017) and “forensic social science” where the answers are dictated and the evidence is chosen to fit the results (Rivlin 1973), evidence-based policy is under attack. But time and again, better and more consistent evidence will rise to the top and good science will lead to better policy.