The economic characteristics of one's childhood neighborhood have been found to determine long-term well-being. Policies enacted during childhood may change neighborhood trajectories and thus impact long-term outcomes for children. We use individual-level data from the Wisconsin Longitudinal Study to examine the enduring consequences of childhood exposure to local-area New Deal emergency employment work-relief activity. Our outcomes include adolescent cognition, educational attainment, midlife income, health behaviors, late-life cognition, and mortality. We find that children (ages 0–3) living in neighborhoods with moderate work-relief activity in 1940 had higher adolescent IQ scores, had higher class rank, and were more likely to obtain at least a bachelor's degree. We find enduring benefits for midlife income and late-life cognition for males who grew up in areas with a moderate amount of work relief. We find mixed results for males who grew up in the most disadvantaged areas with the highest levels of work-relief activity. These children had similar educational outcomes as those in the most advantaged districts with the lowest work-relief activity but had higher adult smoking rates. Our findings provide some of the first evidence of the long-term consequences of New Deal policies on children's long-term life course outcomes.
Over the last several decades, the relationship between childhood circumstances and educational attainment, economic well-being, and health over the life course has been established across economics, demography, epidemiology (Almond et al. 2018; Campbell et al. 2014; Hayward and Gorman 2004), and other fields. Furthermore, studies have documented positive long-term consequences for children whose households benefited from social safety net programs, such as the Earned Income Tax Credit (EITC) (Chetty and Friedman 2011), supplemental nutrition programs (Bailey et al. 2020), and Medicaid (Boudreaux et al. 2016). Several experimental studies have further evaluated the relationship between childhood circumstances and long-term outcomes. For example, the Moving to Opportunity (MTO) experiment randomly assigned families to receive a housing voucher to move to a low-poverty neighborhood. The study found that younger children who moved were more likely to attend college, had higher adult incomes, and were less likely to become single parents (Chetty et al. 2016).
As one aspect of childhood circumstances, where children live has recently received attention in the economics literature (Chyn and Katz 2021). A growing consensus suggests that childhood residential neighborhood is an important determinant of adult outcomes, although the mechanisms remain unclear. Other recent work suggests that proximate childhood neighborhood characteristics, such as poverty rates within one mile of the childhood home, are most important in predicting adult outcomes (Chetty et al. 2018). Neighborhood poverty is thought to affect lifelong outcomes through its impact on childhood cognitive capacity and adolescent risk behaviors. Cumulative advantage theories predict that the sequence of advantages or disadvantages is an important consideration and that early advantages propagate into early adulthood, midlife, and later-life advantages (DiPrete and Eirich 2006). Yet, these cumulative advantages are not homogeneous. For example, studies of the MTO experiment have found that lower neighborhood poverty has varying benefits for exposed children on the basis of their gender, age, and exposure duration (Nguyen et al. 2016; Osypuk et al. 2019; Schmidt et al. 2018).
In addition, government policies may impact neighborhoods in complex ways. For example, local policies can alter local labor markets, educational opportunities, and health/nutrition access (Lukes and Cleveland 2021; Richardson et al. 2017). These policies may interact with social networks to shape socioeconomic opportunities for the entire community well into the future (Jæger and Blaabæk 2021). Government policies aimed at improving child well-being have usually focused on providing economic support to children in poor households (i.e., EITCs, supplemental nutrition assistance programs, Medicaid, housing vouchers) rather than investing in place-based policies that reshape neighborhood opportunities. Yet, at certain historical moments, massive public investments in infrastructure and simultaneous fiscal redistribution to households could have affected local neighborhoods across large regions. A particularly notable example is the New Deal Era, when the U.S. government responded to the Great Depression with investments that increased local incomes and transformed local areas through the construction of roads, schools, parks and improved access to nutrition and sanitation (U.S. Federal Works Agency 1947; Wright 1974).
The current study aims to build on the literature describing the connection between social policies that transform children's neighborhoods and their subsequent long-term economic and developmental outcomes, but with two new contributions. First, we focus on work-relief policies enacted during the Great Depression. Children born in the Great Depression were exposed to their parents' economic adversities at the time. Parents in this era had higher and more sustained unemployment rates than any cohort has experienced since, with parental unemployment rates of 10% to 25% for several years. It is plausible that the size and duration of the economic shock had substantial negative long-term impacts on child development. At the same time, the New Deal social policy response was itself of unprecedented scale. This combination of extreme deprivation and extreme relief provides a unique opportunity to evaluate the impact of large government-instituted employment and infrastructure relief efforts on neighborhoods and their subsequent effect on children's economic and health trajectories. Although relief programs have been shown to limit city-level contemporaneous infant mortality (Fishback et al. 2007), no study has examined the long-term effects of local-area New Deal era relief programs on longer term child outcomes.
Second, we examine long-term outcomes measured at different points in the life course from ages 15 to 80. By examining the impact of New Deal emergency employment work-relief programs on educational, economic, and health outcomes across multiple points in the life course, we aim to elucidate the life course relationship between childhood neighborhood, adolescent outcomes, adult socioeconomic position, adult health behaviors, late-life cognition, and mortality outcomes. If some consequences of these work-relief programs endured but others faded as children moved to adulthood, then this approach will provide a more comprehensive picture than studies with more limited duration or outcome domains. We further extend the literature and examine how outcomes vary by gender over the life course.
Background on New Deal Work-Relief Programs
The New Deal consisted of government programs instituted by President Franklin Roosevelt to combat the economic decline of the Great Depression (National Resources Planning Board 1943). Between 1935 and 1941, the New Deal programs led to what was, at that time, the largest temporally discrete increase in government-led infrastructure, services, and employment policy in U.S. history. The largest of these programs, the Works Progress Administration (WPA), employed more than 8 million individuals and provided income support and stability to households, reaching more than 20 million children. Work-relief programs, such as the WPA, composed a large share of the late-1930s interventions (National Resources Planning Board 1943). Several other programs targeting child development and health were also implemented or strengthened during this period, including the Children's Bureau and Aid to Families with Dependent Children (Kotch 2013).
Here, we focus on the work-relief program of the New Deal in the late 1930s, which included the National Youth Administration (NYA), the Civilian Conservation Corps (CCC), and the WPA. The NYA and CCC aimed to employ young adults to ensure that their skills would not atrophy and had limited requirements regarding family need. In contrast, the WPA was open to all adults older than 18, and employment through the WPA was based on family need (Table A1, online appendix). The WPA was the primary organization in charge of work-relief activities and made up about 70% of emergency employment jobs by June 1940 (National Resources Planning Board 1943).
Our study setting is Wisconsin. Emergency employment work relief in Wisconsin, as in other states, was enacted as a partnership between the federal, state, and local governments. The federal government's financial contribution was significant: three quarters of the funding for relief projects in Wisconsin in 1935–1938 was federally provided (U.S. Federal Emergency Relief Administration 1938). State and local governments had significant control over the projects and, within federal guidelines, had the discretion to identify local needs to be addressed. American public memory of the New Deal infrastructure has become somewhat attached to public buildings, such as post offices and schools. However, the work of emergency relief workers was diverse and varied over time. In Wisconsin, as nationally, road projects were important. For example, in 1934–1935, one quarter of the 9,880 relief projects in Wisconsin were improvements to “farm-to-market and other secondary roads.” The next largest category was “educational buildings,” accounting for 1,000 projects. Within schools, WPA workers served 15.4 million school lunches (U.S. Federal Works Agency 1947:134–136). In June 1940, shortly after the census, 30% of Wisconsin relief workers worked on road projects, compared with 43% nationwide. Compared with the nation, Wisconsin relief workers were more likely to be building sewers (16% vs. 10%), recreational facilities (12% vs. 6%), and conservation projects (7% vs. 3%).
New Deal and Potential Mechanisms for Human Development
Prior studies found that relief programs had income multiplier effects for benefitting states (Fishback 2017; Fishback and Kachanovskaya 2015). These studies show that work-relief interventions within a locality helped maintain an area's income and had multiplier effects beyond the beneficiaries. Infant mortality fell more in areas with higher relief spending, supporting the hypothesis that children, in particular, may have benefited (Fishback et al. 2007). Thus, there is reason to believe that areas with more emergency employment activity would have had reduced poverty rates and increased consumption and incomes for many households in the area, regardless of whether they directly received benefits.
Poverty reductions in communities with emergency employment activity could lead to a variety of positive outcomes for child development, especially cognitive development. Growing evidence points to poverty costs regarding children's neuroendocrine function, early brain development, and cognitive ability. Results from a randomized control trial showed that regular cash transfers given to low-income families increased infant brain activity at age 1 (Troller-Renfree et al. 2022). Poverty has been found to be associated with smaller white and cortical gray matter and hippocampal and amygdala volumes in childhood (Luby et al. 2013). Studies using the British 1946 birth cohort found that exposure to poor material home conditions (e.g., dated structure, poor repair, uncleanliness, crowding) at age 4 was strongly associated with lower cognitive ability in childhood and adolescence; negative effects of poverty on measures of verbal ability, memory, and concentration persisted into midlife (Richards and Wadsworth 2004).
Work-relief programs likely affected child development and later-life adult outcomes through direct and indirect pathways. The direct pathway would have impacted households in which family members were employed through emergency employment programs. These households would have had more resources to support their children. Although this pathway is important to examine, work-relief jobs were offered only to the long-term unemployed on relief rolls for months (National Resources Planning Board 1943). Comparisons that do not account for the disadvantageous selection process into these programs will likely be biased. Indirect pathways include area-level income and consumption effects or multiplier effects. The increased circulating resources could have improved the childhood environment for households in the local area that did not receive work relief—households that are likely to be less selected on individual characteristics. Moreover, communal resources built or maintained through emergency employment programs could affect children's environment through improved sanitation, better school access, recreational opportunities, local services, and other resources that in turn could have affected outcomes for all children in the area.
Given that we anticipate strong selection of benefitting households, we focus our estimation approach on the effects of local-area-level emergency employment, conditional on observable individual-level characteristics that predict emergency employment receipt and area-level characteristics correlated with emergency employment activity. We examine local areas, defined as census enumeration districts (EDs), as the area-level unit. We examine the overall effects of area-level benefits of New Deal work-relief programs on outcomes across the life course of individuals who were children in 1940.
We analyze data from the Wisconsin Longitudinal Study (WLS), a cohort born between 1937 and 1940. The WLS includes a random one-third sample of all male and female high school seniors in Wisconsin in 1957 (Herd et al. 2014; Wisconsin Longitudinal Study n.d.). We use WLS data from six survey waves (1957, 1964, 1975, 1992, 2004, and 2011; see Figure A1, online appendix). The WLS collects information on participants' educational attainment and experiences, economic attributes, family composition and relationships, health, and mortality. The study has had a remarkable response rate over time, at 86% in 2004 (47 years after the initial survey). The WLS also includes administrative linkages that augment the survey data.
The data include WLS records that have been linked to the 1940 decennial U.S. census. The WLS–census linkage is exceptionally complete compared with other U.S. cohort studies linked to the 1940 census. The match rate between the 1940 census and the WLS is approximately 90% even when conservative criteria are used for adjudicating matches. By contrast, most census linkages, even when linking records of independently transcribed versions of the same census, can link only 43% to 67% of the observations (Abramitzky et al. 2021).1 Moreover, previous linkage efforts have focused on men because women often change surnames at marriage. The WLS–census linkage includes both genders because women were first sampled in high school before marriage and a potential surname change.2
The full WLS includes 10,317 men and women. Our sample is restricted to the 9,207 individuals whose records could be linked to the 1940 census (for a comparison of matched and unmatched WLS participants, see Table A2, online appendix). From this matched sample, we extract participants' county and ED information in 1940. We then link characteristics of the county or ED where participants and their families resided in 1940. Some counties and EDs could not be matched. We further exclude non-White individuals (n = 39), respondents born before 1937 (n = 452), and those living on farms (n = 2,173). WLS participants whose families resided on farms in 1940 could have received targeted benefits under the Agricultural Adjustment Act, which we cannot determine. Further, research suggests that WLS participants who resided on farms had distinctly disadvantageous educational and economic trajectories (Herd et al. 2019). These restrictions and survey nonresponse for some covariates and outcomes lead to an upper limit for the analytic sample size of 5,303. Because we use complete case analysis, our sample sizes vary by outcome.3
ED Emergency Employment
The main exposure variable in our analysis is derived from the complete count 1940 census (Ruggles et al. 2020). The 1940 census included the following questions asked of all individuals aged 14 or older:
“Was this person at work for pay or profit in private or nonemergency Govt. work during the week of March 24-30?”
“If not, was he at work on, or assigned to, public Emergency Work (WPA, NYA, CCC, etc.) during the week of March 24-30?”
“Was this person seeking work?”
“If not seeking work, did he have a job, business, etc.?”
We use responses to these questions to calculate the percentage of the labor force that was employed by an emergency employment program within each ED in March 1940. EDs are the smallest geographic area in the 1940 census and were originally used to create catchment areas for census enumerators. EDs often conformed to existing neighborhoods and thus represent a unit of analysis where individuals likely interacted. Wisconsin had more than 3,400 EDs (average population = 924) and 71 counties in 1940.
We use the complete-count 1940 census to create quintiles of emergency employment based on the distribution of emergency employment in EDs within Wisconsin. The quintile cutoffs are 0.937%, 2.461%, 4.47%, and 7.71%. Table 1 displays sociodemographic characteristics at the ED level across the five quintiles from the complete-count 1940 census. As expected, emergency employment followed need and was higher in areas with higher unemployment and lower adult education. The last row of the table compares the 1893 EDs in the WLS analytic sample, which excludes those living on farms, with the full census. Our analytic sample slightly overrepresents EDs in the third and fourth quintiles and underrepresents EDs in the first quintile.
County and ED Control Variables
One of the most important factors that influenced the location of work relief was the local level of economic distress (Wright 1974). To capture the severity of economic distress throughout the Great Depression, we use existing measures of change in retail sales between 1929 and 1933 (Fishback and Kantor 2017), thereby capturing the initial shock at the county level (Haines and Inter-university Consortium for Political and Social Research 2010). We also account for the average farm size in the county because it determined other New Deal programs that targeted farmers. These county-level variables are correlated with emergency employment both nationally and in Wisconsin.
We also control for other socioeconomic factors within EDs: percentage foreign-born, percentage aged 25 or older with at least a high school education, percentage unemployed, percentage owning their home, and average value of owned homes in the ED.4 These five variables were shown to differ systematically across the five quintiles of emergency employment (Table 1) and could independently affect child outcomes.
Household-Level Control Variables
To account for children's household characteristics, we estimate associations conditional on a series of household/family–level factors. The 1940 linked WLS data allow us to recover measures of household conditions among WLS sample members' families as recorded in the 1940 census. We account for maternal and paternal education, whether the participant's father was U.S.-born, family homeownership, family size, and urban location.
We account for father's work status in March 1940, categorized as (1) employed at a regular job (not emergency employment), (2) employed by an emergency employment organization, (3) unemployed, or (4) not in the labor force.5 In sensitivity analyses, we include fathers' income in 1940 for the subset of fathers for which that information was available.
Parental Income in 1957–1960, Adolescent Cognition, and Education Outcomes
WLS obtained the standardized IQ scores for participants in the freshman year of high school, the junior year, or both. The test score is based on a Henmon-Nelson test, a multiple-choice assessment containing 90 verbal or quantitative items (Lamke and Nelson 1958). This test was administered in all Wisconsin high schools from the 1930s through the 1960s as part of a cooperative effort of high schools and colleges to identify youth who might succeed in college. The test was widely used throughout the United States after its development in the early 1930s. We use the composite junior and freshman year IQ score, as the WLS recommends. Another outcome measure is high school class rank in 1957, ascertained directly from high school records.
IQ and class rank are very different measures. IQ scores can be compared across schools, are based on a single test, and are thought to be a measure of cognitive ability. Class rank is based on grades across all four years of high school (Halpern-Manners et al. 2020), which are determined not only by cognitive ability but also by effort and other noncognitive skills (e.g., conscientiousness). Moreover, class rank is a comparison of grade point averages within schools. Students with the same IQ score could have different class rank measures because of varying rigor of classes or grading, varying levels of sustained effort, or different student compositions within schools.
Another educational outcome we explore is whether the participants ever attended college and whether they earned a bachelor's degree or higher. These outcomes were ascertained across multiple surveys. We use the 2004 measure, which is a retrospective summary measure of the information from previous survey years. Finally, in the 1964 follow-up, the WLS used parental Social Security numbers to obtain tax data on parental incomes for the WLS participants whose parents consented. WLS used these data to calculate average parental income for 1957–1960.
Midlife Work and Wage Outcomes
We examine WLS participants' labor force participation and income in 1974 (at ages 35–37), when participants' careers were fairly established but well before they were considering retirement. We first examine labor force participation, which we anticipate will differ by gender because of the norms of the time. Next, we examine the log of reported wage income in 1974 and total income in 1974. We include only those participants with nonzero income. We also conduct supplemental analyses with a two-part model that includes those with zero income. The two-part model separately estimates the probability of having any income and the positive income. The model then estimates the combined marginal effects of having any positive income and the total income for those with positive values of income.
Body Mass Index and Health Behaviors
Adverse childhood conditions may lead to poorer health behaviors, such as smoking, poor nutrition, low physical activity, and excess alcohol consumption (van den Broek 2021). We examine measures of high body mass index (BMI), smoking behavior, and problematic drinking behavior. The WLS included a BMI measure in the 1992 survey, when participants were aged 52–55. Because the relationship between BMI and mortality is nonlinear and J-shaped, there is disagreement about the appropriate cutoff value of BMI to indicate poor health. However, BMIs greater than 35 have been shown to be unambiguously related to metabolic syndrome and indicate poor health (Tomiyama et al. 2016). Thus, we dichotomize BMI to distinguish those with very high BMI (>35) from those with lower BMI.
To capture smoking and drinking behavior, we use retrospective questions from the 2004 survey when participants were aged 64–67. One question asked whether the participant ever smoked regularly. We use the dichotomized responses—ever smoked regularly or did not smoke regularly. The relationship between health and alcohol consumption is nonlinear and J-shaped and varies by gender. Therefore, instead of using the number and frequency of drinks, we use a validated self-assessment scale, CAGE, which poses questions asking participants about potential problematic alcohol use and its consequences (Dhalla and Kopec 2007). The WLS's version of the CAGE screening instrument asks whether participants' alcohol consumption had (1) caused them guilt, (2) been criticized by others, (3) caused work-related problems, (4) caused family-related problems, and (5) led them to seek help. We use the sum of this scale as a measure of the extent of problem drinking (i.e., range = 0–5).
Late-Life Cognition Measures
In 1992, 2004, and 2011, the WLS administered an abbreviated Wechsler Adult Intelligence Scale (WAIS). This scale is meant to measure abstract reasoning ability. A six-, eight-, or nine-question version of this scale was used for all WLS participants. We examine the WAIS standardized scores as an outcome over a consistent set of participants in the main analyses.
In supplemental analyses, we examine other composite cognitive measures. In 2004 and 2011, a subset of WLS participants was tested on six cognitive tasks. Following Moorman et al. (2019), we use factor analyses to create two summary measures from the six cognitive tasks: working memory and language/executive function (for details, see Table A3 in the online appendix).
In supplemental analyses, we examine mortality. The WLS ascertained participants' mortality through 2019 using National Death Index data. For individuals who died, the year of death is noted. Because we include only WLS participants who were born between 1937 and 1940, we observe mortality only up to ages 79–82. After 2019, WLS participants who were still alive were censored.
We estimate models with ED-level emergency employment work-relief activity as the primary exposure of interest. We control for individual-level gender, parental/household characteristics, ED characteristics, and county-level economic distress.
To estimate the effect of the ED emergency employment activity on economic and health outcomes, we use linear, logistic, Poisson, or time-to-event regression models, depending on the distributional properties of the outcome examined. Our stylized estimation equation is as follows:
Oidc = F(α+ β(%EEd) + γ(Genderi) + δ(Parental Zi) + χ(ED-chard) + φ(Change in Retail Salesc)+ λ(Avg.Farm Sizec) + εidc),
where Oidc is the outcome of interest for a child i in enumeration district d and county c. Our exposure of interest is the percentage of the labor force that was employed by emergency employment programs in the ED (%EEd) split into quintiles, as described earlier.
We include child gender as control for each individual, Genderi. We capture childhood household environment through a series of variables (Parental Zi1940), including paternal employment status, paternal and maternal education, an indicator of whether the participant's father was U.S.-born, family homeownership, and family size. We also account for whether the household lived in an urban area in 1940. We account for ED-level characteristics (ED-chard), including percentage foreign-born, percentage aged 25 or older with at least a high school diploma, percentage unemployed, percentage who own their home, and average home values. Finally, we include changes in retail sales between 1929 and 1933 and average county farm size in 1930 at the county level.
For all outcomes, we include an additional model in which we examine the relationship between ED-level emergency employment activity and the WLS participant's gender, operationalized as an interaction term between %EEd and the WLS participants' gender, Genderi. Because ED-level measures are not independent, all standard errors are clustered at the ED level. For continuous outcomes, we present clustered standard errors for regression estimates in brackets. For binary, count, and mortality outcomes, we present 95% confidence interval bounds using clustered standard errors in parentheses.
Match and Representativeness
We first describe the 1940 census-matched and unmatched WLS samples (Table A2, online appendix). The approximately 1,100 unmatched WLS participants were systematically younger, had more-educated parents, and were more likely to have lived in a rural area.
A key issue in interpreting the results is how representative the WLS participants and their families are relative to external populations. To assess the degree of representativeness, we compare the matched WLS sample's parents' education to parents' education in Wisconsin, the Midwest, and the United States in 1940.6 The WLS sample is overrepresentative of higher educated White parents compared with the United States and the Midwest in 1940, but it is representative of the racial composition and educational attainment of parents in Wisconsin in 1940 (results available on request).
To further assess WLS participants' representativeness, we examine several characteristics from the 1940 census across the WLS matched sample, our analytic sample, and a comparison group from the 1940 census. We limit the census comparison group to households in Wisconsin with a child under age 10. Table 2 presents the comparisons. Although the comparisons are imperfect, they provide evidence that WLS participants' family members were less likely to own their home in 1940 and were more likely to have a U.S.-born father and to have a father with regular employment in 1940. Because we limit the sample of WLS participants to nonfarm dwellers in 1940, the analytic sample is less likely to have lived in a rural area.
Next, we describe the main exposure variables capturing emergency employment activity at the ED level. Figure A2 (online appendix) presents the centiles of emergency employment activity in Wisconsin in 1940. Using this distribution, we create quintiles of emergency employment within EDs as our primary exposure variable. The median value of ED-level emergency employment as a percentage of the labor force is 3.3%. Some EDs had very high emergency employment activity relative to the labor force, but 90% of EDs had less than 10% of the labor force in emergency employment jobs. Figure A3 (online appendix) shows the spatial distribution of county-level emergency employment. Northern counties had the highest concentration of such employment in 1940.
Table 3 provides the summary statistics for the outcome variables examined in this study. Note that the number of observations varies across outcomes because these data were collected in different study waves and had different response rates.
WLS participants' average parental income was approximately $7,000 in 1957–1960 and varied widely. We standardized IQ scores within the entire WLS cohort so that the analytic sample's IQ has a mean of 0.08 and a standard deviation of 0.98. The average class rank is 51. Approximately 51% of the WLS cohort in our analytic sample ever attended college, and 29% earned a bachelor's degree or higher. Approximately 78% of the WLS cohort was employed in 1974. The average yearly wage or salary for WLS participants in 1974 was about $8,500 ($51,000 in 2022 dollars). For BMI and health behaviors, 34% of WLS participants had a BMI above 35, and 59% had ever smoked regularly. All late-life cognitive scores are standardized to have a mean of 0 and a standard deviation of 1 for the entire WLS sample. Our analytic sample has slightly higher late-life cognitive scores, with a mean of approximately 0.05. By 2019, 41% of the men and 29% of the women in the analytic sample had died.
The regression tables are organized by type and timing of outcomes. For all the regression tables, the first column for each outcome presents estimates that are not allowed to vary by gender. The second column for each outcome adds an extra interaction term between area-level emergency employment activity and the WLS participant's gender to allow for gender differences in the estimated relationships.
Table 4 presents the association between ED emergency employment activity in 1940 and WLS participants' adolescent IQ, class rank, college attendance, and parental income in 1957–1960. Columns 1–2 of Table 4 examine standardized IQ test scores. Relative to WLS participants in EDs with the lowest emergency employment activity (Q1) in 1940, those in EDs with moderate emergency employment activity (Q2 and Q3) had a standardized IQ score that was 0.09–0.12 standard deviations higher. The results do not differ by gender. Relative to WLS participants with unemployed fathers in 1940, those whose fathers held emergency employment jobs had an IQ score that was 0.15 standard deviations lower.
Columns 3–4 of Table 4 present the association between high school class rank in 1957 and ED-level emergency employment activity. WLS participants who lived in EDs in the second quartile of emergency employment activity had a high school class rank that was about three units higher in 1957. Although female participants have a higher class rank in general, the relationship between ED emergency employment activity and class rank does not differ by gender. Moreover, WLS participants whose fathers held emergency employment jobs in 1940 had a class rank that was four units lower than those whose fathers were unemployed; this result, however, was marginally statistically significant.
Columns 5–6 present the association between earning a bachelor's degree or higher and ED-level emergency employment activity. WLS participants living in EDs in the second quartile of emergency employment activity had a higher likelihood of completing a bachelor's degree (odds ratio [OR] = 1.35, p < .05). However, when examining the association of emergency employment activity while allowing for gender differences, we find that male WLS participants living in areas with moderate or high emergency employment were more likely to earn a bachelor's degree or higher than those who lived in areas with low emergency employment activity. The magnitude of this effect suggests that male WLS participants who lived in areas with moderate emergency employment (Q2 and Q3) were 53% more likely to earn a bachelor's degree (OR = 1.53, p < .05). The magnitude is similar in EDs with the highest emergency employment activity (OR = 1.58, p < .05). The opposite relationship holds for women: female WLS participants who lived in areas with moderate or high emergency employment were less likely to earn a bachelor's degree than those who lived in areas with low emergency employment activity. The magnitude of this effect is only marginally significant for female WLS participants in the third quintile of emergency employment activity (OR = 0.64; p < .10).
Columns 7–8 of Table 4 examine the average parental income of WLS participants based on tax records from 1957–1960. Parental income was 3% to 5% higher in areas with higher emergency employment activity (Q2–Q4) than in areas with low emergency employment activity, but the estimated coefficients are not statistically significant. In sensitivity analyses, we examine parental income patterns with nonparametric models in detail.
Table 5 presents the association between ED-level emergency employment activity and labor force participation and log income in 1974. Columns 1 and 2 show that there is no relationship between emergency employment activity in 1940 and labor force participation in 1974. Columns 3–6 display income outcomes for the sample limited to those with positive wages in 1974. Column 3 shows that wages were approximately 12% higher in 1974 for WLS participants living EDs with moderate emergency employment activity (Q2) than for those living in EDs with low emergency employment activity (p < .01). Likewise, column 5 shows that WLS participants who lived in EDs with moderate emergency employment activity (Q2) had approximately 14% higher total earnings in 1974 (p < .01) than those who lived in EDs with low emergency employment activity. Note that results in columns 3–6 were for those with positive income.
In Table A4 of the online appendix, we present the combined marginal effects for labor force participation and total income using a two-part model to account for zero wages and gender differences in labor force participation. The results show that WLS participants who lived in EDs with moderate emergency employment activity (Q2) had approximately $880 higher total income in 1974 than those who lived in EDs with low emergency employment activity (roughly equivalent to $5,000 in 2022; p < .05). Examining differences by gender, we find that male WLS participants living in EDs with moderate emergency employment activity (Q2) had higher incomes than male participants who were in EDs with low emergency employment activity, but that women living in EDs with moderate emergency employment activity (Q2) had statistically significantly lower own incomes than women who lived in EDs with low emergency employment activity.
Table 6 presents the association between ED-level emergency employment activity and each of three health-related behaviors: BMI, having ever smoked regularly, and drinking problems. ED-level emergency employment activity in 1940 was not associated with BMI. ED-level emergency employment activity is not associated with ever smoking regularly or problematic alcohol consumption when we consider both genders together (columns 3 and 5). This result obscures differences by gender. Male WLS participants living in EDs with high emergency employment activity (Q4 and Q5) had a higher likelihood of ever smoking regularly (Q4 OR = 1.77, p < .01; Q5 OR = 1.61, p < .05) than male participants who were in EDs with low emergency employment activity. Men who lived in EDs with moderate emergency employment activity also had elevated smoking rates. Female WLS participants living in EDs with high emergency employment activity (Q4 and Q5) were approximately 50% less likely to have ever smoked regularly. Male WLS participants living in EDs with moderate emergency employment activity (Q2) reported more drinking problems, whereas female WLS participants living in EDs with moderate emergency employment activity reported fewer drinking problems.
Table 7 presents the relationship between ED-level emergency employment activity and the WAIS score for three periods over a consistent subset of participants. WLS participants living in EDs with moderate activity (Q3) had consistently higher WAIS scores across three waves than those who lived in EDs with low emergency employment activity in 1940. The results are stronger for male WLS participants. We present results for the relationship between ED-level emergency employment activity and late-life cognitive measures of executive function and memory in the online appendix (Table A5). Results suggest higher language and executive function scores for men who lived in EDs with moderate activity (Q3) relative to men who lived in areas with low emergency employment activity. There was no differences in working memory scores.
Finally, we examine survival rates and emergency employment activity in Table A6 (online appendix). Our follow-up captures deaths up to approximately ages 79–82. ED-level emergency employment activity was not associated with survival rates for either gender.
We used the complete-count 1940 census to determine the emergency employment activity quintile cut points. To validate these cut points, we show the results of nonparametric regressions with the same control variables for the relationship between emergency employment activity and two outcomes: average parental income in 1957–1960 and IQ score. As shown in Figure 1, the relationship for both variables is nonlinear. As emergency employment activity increases from 0% to 2%, the mean values of both outcomes increase. After 2%, the relationship slowly decreases. At low and high levels of emergency employment activity, the mean values are imprecisely estimated, with large standard errors. Although each outcome has a slightly different shape, these analyses suggest that our quintile split captures different sections of the implied relationship.
Another concern is that controls for parental emergency employment do not sufficiently control for household-level benefits. In Table A7 (online appendix), we exclude WLS participants whose father was employed at an emergency employment job in 1940. For brevity, we examine three outcomes that were collected nearest in time to 1940 (IQ score, high school class rank, and parental income in 1957–1960) and find that the coefficients are unchanged when we exclude this group. This finding suggests that nonbeneficiary households did benefit.
The previously discussed models did not include fathers' income because that could (and we hypothesize should) increase in areas with higher emergency employment activity. Since father's income itself could directly affect a variety of life course outcomes for WLS participants, we excluded it as a control variable in the main models. In Table A8 (online appendix), we examine additional models that include the impact of father's employment and income on the regression estimates. We present three models. The first does not include the father's employment status or income in 1940. The second includes the father's employment status (as in the previously presented models). Finally, the third model includes the father's employment status in 1940, as well as two 1940 income variables: the father's wage income in 1940 and whether the father had a nonwage income greater than $50 in 1940. We examine the three outcomes that were collected nearest in time to 1940. The addition of these variables does not change the relationship between area-level emergency employment activity and parental income in 1957–1960, IQ score, or high school class rank.
This study examined the association between childhood exposure to area-level New Deal emergency employment activity and life course outcomes. Specifically, we examined a small-area measure of emergency employment activity in 1940 and the sequential life course outcomes, conditional on whether the household head/father was engaged in emergency employment. We found that children living in areas with moderate emergency employment activity had higher scores on adolescent IQ exams and higher class rank than children living in EDs with low emergency employment activity. The results for IQ scores and the interpretation that children living in EDs with moderate emergency employment had better IQ outcomes than those living in EDs with low emergency employment were supported with nonparametric models. Our results further suggest that children who grew up in the most disadvantaged districts with the highest levels of work-relief activity (i.e., Q4 and Q5) had similar adolescent IQ scores and class rank to children who grew up in the most advantaged districts (Q1).
In interpreting these findings, it is important to acknowledge that work-relief activity was intended to counter high rates of unemployment in local areas but that areas with high unemployment had other preexisting disadvantages. Indeed, areas with the highest work-relief activity had the highest unemployment rates and the lowest adult education levels. In the most disadvantaged areas, work-relief programs may have neutralized some preexisting disadvantages even if they did not completely overcome them. In contrast, the more beneficial life course impacts observed in moderately disadvantaged areas (Q2 and Q3) could be attributable to the success of work-relief jobs, income, and investments in overcoming the smaller disadvantages in these areas, thereby leading to better outcomes for children in those areas than for children even in the least disadvantaged areas. The incomplete countervailing impact of work-relief activity likely explains the nonmonotonic relationship between the amount of work-relief activity and several socioeconomic outcomes.
Our results did not indicate gender differences in adolescent IQ scores and class rank. However, once we examined outcomes for WLS participants at older ages when they would have started their households, gendered patterns emerged. Importantly, we found that higher rates of bachelor's degree completion were concentrated among male WLS participants who lived in EDs with moderate or high emergency employment. Female WLS participants were consistently less likely to complete a bachelor's degree, although this result was not always statistically significant. Examining economic outcomes in early adulthood, we found no labor force participation differences for WLS participants who were exposed to emergency employment activity. WLS graduates living in EDs with moderate emergency employment (Q2) had 12% higher incomes in midlife than those living in EDs with low emergency employment activity. In models accounting for both labor force participation and wages, gender differences emerged: males who lived in EDs with moderate emergency employment (Q2) had higher adult incomes, whereas females who lived in these areas had lower adult incomes compared with their matched sex counterparts who lived in EDs with low emergency activity. Gender differences also emerge in late-life cognitive scores: male WLS participants who lived in EDs with moderate emergency employment activity consistently exhibited higher scores than males who lived in EDs with low emergency employment activity, whereas female WLS participants who lived in EDs with higher emergency employment activity had either lower late-life cognitive scores or the same scores as females who lived in EDs with low emergency employment activity.
Late-life health behaviors exhibit both lasting markers of the early-life disadvantage and gender norms. We found higher rates of regular smoking in adulthood for males who lived in EDs with high emergency employment activity than for males who lived in EDs with low emergency employment activity.
As WLS participants moved into their own households, other factors may have gained importance. Specifically, the strong gender norms of the time might have led to more divergent outcomes for men and women. The median age at first marriage was approximately 20 for women in the WLS birth cohorts (Rele 1965), suggesting that these women were likely to marry and change households shortly after graduating high school or college. Norms around working for women, particularly mothers, may have affected female WLS participants' occupational choices and salaries (Yellen 2020). Although we observed increased income for male participants and decreased income for female participants living in EDs with moderate emergency employment activity compared with their matched sex counterparts living in low emergency activity areas, these results should be interpreted with caution and with consideration of the social context of the time. Future assessments of women's midlife outcomes may need to include their husbands' attributes and/or household income because these measures may be important markers of women's socioeconomic status in this era. Further, social norms regarding smoking and drinking were highly gendered, and retrospective survey responses may encapsulate real behavior differences and mirror expected adherence to social norms.
This study contributes to an understanding of early-life environment by using a longitudinal cohort to examine a significant, specific government intervention that impacted neighborhoods in the historical context of the Great Depression. The longitudinal nature of the WLS allowed us to study both medium-term and longer term outcomes. Notably, we found mixed results for male participants: living in areas with moderate levels of emergency employment activity in 1940 was related to higher education, higher adult income, and higher cognitive scores compared with living in areas with low emergency employment activity, but living in areas with the highest emergency employment activity was related to higher adult smoking rates compared with living in areas with low emergency employment activity. These results may explain why we did not find relationships with mortality. Resources from emergency employment may have had some positive effects on long-term outcomes (cognition, education, and income) for children in moderately disadvantaged areas, even if they could not undo other negative exposures in the most distressed areas (smoking behavior).
Our study has some notable limitations. First, the WLS participants included in our analysis—White individuals living in a progressive state—were relatively advantaged. Wisconsin enacted a state unemployment insurance plan in 1932 and was the only state to offer unemployment insurance benefits before 1935 (Nelson 1967). Moreover, because the WLS sample did not include sufficient numbers of African American or Native American participants, we excluded these groups from our analysis. Thus, our results generalize to White high school graduates, who nevertheless represent two thirds of the current U.S. population in these birth cohorts (Herd et al. 2014). Moreover, because of our sample selection, our findings relate to nonfarm residents only. Second, our measure of emergency employment is a point-in-time measure in March 1940. This measure fails to capture the emergency employment activity during the entire duration of the Great Depression, which would have been a better indicator of ongoing neighborhood investments. Many households that did not have emergency employment jobs in 1940 may have been employed through these programs in earlier years of the Depression. At the community level, some areas that are categorized as having low emergency employment rates in 1940 could have had high or moderate emergency employment activity in prior years. Both of these limitations—a relatively advantaged sample and potential misclassification of the exposure—would lead to an underestimate of the effects of emergency employment on the outcomes examined. Third, because we cannot account for other, simultaneously occurring relief programs or for the type of local investment (schools vs. roads), the estimated effects could be biased by these unobserved variables. Finally, for more urban areas, students from multiple EDs attended the same high school. However, in more rural areas, EDs may coincide with school district boundaries, and we therefore cannot disentangle benefits resulting from particular high schools or school districts.
Future studies should examine whether benefits accrue differently by child's age at exposure, gender, and race. This study, one of the first to examine long-term outcomes for children resulting from New Deal work-relief programs, explored outcomes by gender. Research should examine these associations among a more geographically, socioeconomically, and racially diverse population of Americans.7 Furthermore, as new data emerge on the actual outcomes of the New Deal investments, studies should examine possible mechanisms related to increased sanitation, better nutrition, and higher area-level consumption supported by New Deal expenditures.
This project was generously supported by the Sloan Foundation (G-2018-11079) and the National Institute on Aging (R01AG059791 and R01AG050300). We use data from the Wisconsin Longitudinal Study, which has been funded by the National Institute on Aging (R01AG009775 and R01AG033285). Support also came from the Minnesota Population Center, which receives core funding (P2CHD041023) from the Eunice Kennedy Shriver National Institute for Child Health and Human Development. The funders had no role in study design, data collection and analysis, decision to publish, or manuscript preparation.
Common names (e.g., John Smith) and transcription errors from the original file and the digitization process account for the high nonmatch rates.
Although the minimum marriage age was 16 years, women rarely married before completing schooling. Given that the WLS is a sample of high school graduates, most females in the WLS were unlikely to be married at the start of the survey.
Although the WLS eventually added the initial participants’ siblings to the sample, our analysis examines outcomes only for the initial participants.
The average is calculated for all owned homes with nonzero and nonmissing values.
Some WLS fathers classified in regular work, unemployed, or not in the labor force may have held an emergency employment job before or after March 1940.
The analysis of the complete-count census is limited to households with a person under age 20.
Future analyses that include Americans of different races and ethnicities may also have to contend with the neighborhood effects of redlining. Although five areas in Wisconsin (Milwaukee, Madison, Racine, Kenosha, and Oshkosh) were redlined, we examined effects among White Americans who were likely less affected by redlining, and all our models controlled for local house prices.