Abstract

A large literature has documented links between harmful early-life exposures and later-life health and socioeconomic deficits. These studies, however, have typically been unable to examine the possibility that these shocks are transmitted to the next generation. Our study uses representative survey data from the United States to trace the impacts of in utero exposure to the 1918 influenza pandemic on the outcomes of the children and grandchildren of those affected. We find evidence of multigenerational effects on educational, economic, and health outcomes.

Introduction

Understanding persistent poverty and poor life outcomes poses significant challenges for social science theories as well as crafting policy responses. Of particular interest is the so called long arm of childhood circumstances, wherein events and conditions occurring early in life (even in utero) may set in motion an accumulation of disadvantage. The developmental origins of health and disease theory (often referred to as the Barker hypothesis) suggests that insults in utero program the fetus in ways that lead to maladaptive responses to the environment that persist throughout the life course and explain poor long-term outcomes (Barker 2007). However, given that the likelihood of experiencing an early insult is related to parental circumstances, causally examining the impacts of early insults poses an empirical challenge. This issue ties into a key question of whether intergenerational persistence in poor outcomes may stem in part from genetic mechanisms that may be more resistant to policy efforts. In order to separate potential channels, some research has focused on exogenous (i.e., unrelated to genetics) health shocks during early life to explore long-term outcomes.

Indeed, a strong causal link between harmful early-life or in utero shocks and lowered later-life health and economic outcomes has been found in a number of clearly identified empirical studies. In addition to separating genetic and nongenetic mechanisms, many of these investigations have also been able to uncover biological (i.e., fetal programming) explanations by studying specific insults, including the effects of prenatal nutritional status and exposure to infectious disease. Without the use of experimental (i.e., animal model) research designs, these studies have used quasi-experimental designs by leveraging striking demographic events, such as the Dutch famine and the 1918 flu pandemic (Almond 2006; Lindeboom et al. 2010). The results generally suggest large, lifelong impacts of in utero insults on economic and health outcomes.

With these results established, a next question is whether in utero effects might persist into the outcomes of future generations. We explore the question of intergenerational persistence of fetal insults, extending previous work that used the 1918 influenza pandemic as a natural experiment (Almond 2006) by adding multigenerational data. We document that in utero exposure to the pandemic can be seen in the educational attainments of multiple generations. More specifically, we find a reduction of approximately one-tenth of a year of schooling (statistically insignificant) for the first generation, one-fifth of a year of schooling for the second generation, and one-sixth of a year of schooling for the third generation.1 We also show large effects on economic as well as health outcomes for the second generation. Our findings suggest the importance of in utero health insults that persist across multiple generations and allow a shift in our analytical frame from the long arm of childhood circumstances to the long arm of previous generation’s circumstances—or, alternatively, the long reach of history (Kuzawa and Eisenberg 2014). Our results can also be interpreted in the context of assessing benefits of policy and environmental conditions that reduce the likelihood of in utero health insults, where the full benefits may unfold over multiple generations.

Background Literature

Our research intersects several related literatures. A large literature has documented associations between early-life insults and later-life outcomes, following the original Barker hypothesis. In addition, we draw on literature that has explored multigenerational linkages of environmental exposures on life course outcomes. These literatures have rarely been unified in a single analysis, largely because of data constraints. We overcome these limitations by leveraging a three-generational data set that coincides with a key demographic event from the early twentieth century: the 1918/1919 influenza epidemic.

Almond’s (2006) seminal study was among the first to exploit the 1918 flu pandemic as a natural experiment and to test whether in utero exposure to infectious disease influences later-life outcomes. Using U.S. census microdata to situate respondent births in both time (birth quarter) and place (exploiting geographic variation in the severity of the flu), Almond found evidence of lower educational attainment, higher rates of physical disability, and lower income and socioeconomic status (SES) for cohorts exposed to the 1918/1919 flu while in utero (estimated using birth quarter, age, and timing of the flu) compared with unexposed adjacent cohorts. In a related study, Almond and Mazumder (2005) used a similar approach with data from the Survey of Income and Program Participation and found negative health effects for adults who were in utero (based on birth quarter) during the outbreak. Subsequent work has built on these outcomes, finding excess cardiovascular disease (Mazumder et al. 2010); increased likelihood of kidney disease, diabetes, and circulatory and respiratory problems in old age (Lin and Liu 2014); and increased old-age mortality in non-cancer-related causes (Myrskylä et al. 2013).2 More broadly, the negative effect from the flu dovetails with other negative in utero shocks (see, e.g., Catalano et al. 2011; Roseboom et al. 2001; Schulz 2010; Torche 2011).

The mechanisms underlying these broad effects from in utero influenza exposure are not fully understood in humans. However, studies of monkeys (Kobasa et al. 2007) and mice (Kash et al. 2006) have shown that the reconstructed 1918 flu virus triggers an exceptionally intense and prolonged innate immune response. More specifically, gene-expression analysis has shown that proteins involved in the innate immune system have a higher and sustained expression when triggered by the 1918 reconstructed virus than when triggered by the contemporary flu of the same H1N1 type (Kobasa et al. 2007). Although the specific 1918 flu strain cannot be studied in humans through purposeful exposure for ethical reasons, both animal and human studies have linked the maternal immune response during severe flu infection in pregnancy to offspring brain development as well as impaired adult behavior and cognitive outcomes (Brown and Derkits 2010; Canetta and Brown 2012; Fatemi et al. 2002; Li et al. 2014; Miller et al. 2013). These results align with research in psychology that links inflammatory processes in mothers—including stress and infection, as well as stress-induced immunosuppression—to poor birth outcomes (for a review, see Schetter 2011). Recent evidence suggests that in utero exposure to the 1918 flu may affect not only adult outcomes but also the outcomes of subsequent generations. Richter and Robling (2013) were the first to identify an effect of first-generation prenatal exposure (using birth trimester) to the 1918 flu pandemic on the outcomes of the second generation (the children of those exposed to the 1918 flu in utero). The authors used historical influenza morbidity data matched to birth information to identify potential exposure to the 1918 flu and found that first-generation maternal in utero exposure in the second trimester lowers educational attainment for female children by 2 to 2.5 months—that is, by 1.5 % to 1.8 %—but found no such effect for male children. An analogous result was identified for first-generation paternal exposure and male outcomes. First-generation exposure in the second trimester lowers educational attainment for male children by 2.4 to 3 months—that is, by 1.8 % to 2.2 %—but paternal exposure showed no such effect for females. Taken together, first-generation exposure to the 1918 flu while in utero results in 2 to 3 fewer months of schooling for the second generation.

Several pathways exist for the intergenerational transmission of early-life health shocks. Through socioeconomic channels, intergenerational persistence in poor outcomes could occur when a fetally insulted parent, marred by poorer health and socioeconomic outcomes, raises a child in a low-resource environment. Biologically, phenotype-to-phenotype transmission and epigenetic inheritance are hypothesized to be key mechanisms for intergenerational transmission (Kuzawa and Eisenberg 2014). In cases of shocks in utero or very early in life, phenotype-to-phenotype transmission impacts the outcomes of the next generation through changes in parental biological systems that lead to altered gestational and/or lactation environments for offspring (e.g., pre-pregnancy hypertension is linked to low birth weight). Importantly, socioeconomic and biological channels are not mutually exclusive and can also interact with one another; that is, the effects of adult phenotype on offspring extend beyond physiology and metabolism to include parental behavior/environmental response as a potential source of phenotypic transmission and potentially even cumulative intergenerational phenotypic change (Benyshek 2013). For example, stress experienced by a mother prenatally may alter stress regulation in offspring, which may in turn increase risk for the same adult phenotype in the offspring as well as in subsequent generations. Epigenetic inheritance, in comparison, occurs when parental experiences alter gene expression that is subsequently transmitted to offspring and future generations through the germ line. Here, again, there is an opportunity for socioeconomic circumstances to interact with biology—for example, where poverty, through the experience of stress, alters gene expression for generations. Our analysis of second-generation outcomes explores some of these channels, but we do not attempt to prioritize one over another.

To the best of our knowledge, work extending effects of in utero exposure to the 1918 flu to the third generation (the grandchildren of those originally exposed in utero) in humans does not exist. The strongest evidence for a potential biological channel across multiple generations comes from studies of historical data from the Överkalix region in northern Sweden that exploit variance in first-generation grandparental food supply during childhood. Bygren et al. (2001), for example, found that an excess of food during the period just before adolescence—a time labeled the slow growth period (SGP)—shortens a grandson’s longevity. A later study using the Överkalix data replicated the results of Bygren et al. (2001) in a second cohort and further extended the results to include an association between first-generation paternal grandmother’s food supply and granddaughter’s mortality risk (Pembrey et al. 2006).3

Our work builds on this evidence to conduct novel examinations of multigenerational effects of in utero exposures in human populations. The current research leverages a unique survey to measure the multigenerational impacts of in utero exposure to the 1918 flu pandemic. Our hypothesis is that the previously documented direct effects of such shocks extend into the outcomes of the second generation. Going further, we also estimate whether these effects continue into the third generation. In other words, we ask whether the singular in utero shock has a multigenerational effect: that is, on the adult outcomes of those exposed in utero, on their children, and on their grandchildren.

Data and Empirical Methodology

Data

To examine multigenerational effects of an in utero exposure, we require multigenerational data. Very few data sets in the United States have a multigenerational component and fit the relevant period for our exposure (i.e., birth cohorts around 1918). Our data come from the Wisconsin Longitudinal Survey (WLS), which is a random one-third survey of graduating high school seniors in Wisconsin in 1957. The majority of these respondents were born in 1939 and form our second generation. Thus, the parents of the WLS graduates form our first generation, with birth years overlapping the 1918/1919 in utero exposure period. This allows for the creation of our primary measure for in utero exposure to the 1918 flu epidemic: an indicator for either first-generation parent being born during the 1918–1919 range.4,5 Additional data are collected on the later-life outcomes of the WLS graduates and a selected sibling, as well as a limited number of outcomes for the children (third generation) of the graduates/siblings, providing the structure for our multigenerational analysis.

Summary statistics are presented in Table 1. Approximately 10 % of WLS parents were born in either 1918 or 1919. On average, fathers were born in 1907, and mothers were born in 1911. Consequently, a birth year of 1918 or 1919 is closer to the right tail of the distribution of births, and the rate of 10 % is driven primarily by relatively young mother exposures.6 Because WLS graduates’ fathers tend to be, on average, four years older than mothers, this right-tail problem is larger for Generation 1 males, as shown in Fig. 1. Although not the primary focus of the WLS data collection, several parental (first-generation) outcomes are available, including years of schooling, occupational prestige, and family SES in 1957.7

The primary focus of the WLS data collection is Wisconsin high school seniors in 1957, the second generation of our study. Given this focus on graduates (and their siblings), a large number of economic and health variables are available in each wave of the WLS (irregular intervals roughly 10–15 years apart: 1975/1977, 1992/1993/1994, 2003–2005/2004–2007, and in 2011). In addition to examining years of schooling, we include several additional dependent variables for the second generation to capture broad differences in economic and health well-being.8 These include income during the peak earning years (i.e., family income collected when graduates were age 53, on average); net worth at initial retirement age (i.e., net worth collected when graduates were age 65, on average); and general indicators of health, measured by an index of self-reported health, height, and body mass index (BMI). Finally, the WLS data contain information collected from the second generation about the third generation; we focus on years of schooling as the main outcome of interest.

Empirical Methodology

Following Almond (2006), we examine harmful effects of being exposed in utero on later-life outcomes. We then extend this analysis by estimating multigenerational impacts on the second and third generations. The primary estimating equation is given by the following form:
yif,g=β0+β1YOB=1918/1919f,1+γTf,1+δXif,g+εif,g,

Our primary focus is on the coefficient β1, which measures the effect of having a parent/grandparent born in 1918–1919 on a number of outcomes for i individuals in f families for generation g. Parent year-of-birth time trends and their square are denoted by γTf, 1; δXif, g represents generation-specific controls; and εif, g is representative of a family-clustered error term.

For the first generation, we control for birth year and its square, capturing age-specific trends that are tied to our first-generation outcomes of interest (e.g., years of schooling). For the analysis of second and third generations, birth years and squares for both first-generation parents are included as controls along with generation-specific controls for sex, age, and birth order.

Our estimation strategy follows an intent-to-treat design. The actual incidence of flu likely differed by SES and social standing (Mamelund 2006; Sydenstricker 1931). Our use of year of birth, however, avoids potential confounding by SES because all—both rich and poor—are defined as treated if they were born in one of the two years. Although avoiding confounders associated with actual disease incidence, this estimation strategy will underestimate the true effect of flu exposure, lowering the magnitude of our coefficient of interest.

Results

First Generation

Our initial analysis explores the direct effects of being born during the 1918 influenza epidemic. Although health data are sparse for the WLS graduates’ parents (first generation), a number of economic outcomes are available, especially during the initial sample year of 1957. Table 2 explores the relationship between these economic variables and an indicator for birth during 1918/1919. Columns 1–3 of Table 2 show the relationship between a WLS graduate’s father being born during 1918/1919 and the father’s years of schooling, the father’s occupational prestige, and the family’s index of SES in 1957 (i.e., when the first-generation members were approximately age 40). Although statistically insignificant, a negative association is observed between years of schooling and being born in 1918 or 1919.9 This is carried over into father’s job prestige in column 2, which shows that the indicator of in utero exposure to the 1918 flu is associated with an approximate 0.1 standard deviation decline in the index of occupational prestige. These effects culminate in column 3, which shows a statistically significant negative effect of in utero exposure to the flu and later-life economic well-being: a 1.3 decline in the SES index, corresponding to a decline of roughly 10 % of a standard deviation. The findings of Table 2 corroborate past research showing that in utero exposure to the 1918 flu led to poorer economic outcomes later in life (Almond 2006).10

Mirroring the results for males in the first generation, a negative but statistically insignificant association is observed between an indicator of birth in 1918/1919 and schooling for first-generation females. This association, however, becomes statistically significant at the 1 % level for the index of family SES, the coefficient being nearly identical to that of a male’s in utero exposure to the flu, although it is unclear whether these effects flow from labor market and/or marriage market sources. The findings of Table 2 are extended in Table 3, which explores potential marriage market effects of early-life exposure to the flu epidemic.11

Column 1 of Table 3 regresses the indicator for male in utero exposure on an identical measure for spouse’s exposure. Females who were born in 1918/1919 were 5 percentage points more likely to marry men who were also born during the same period. Furthermore, as shown in columns 2 and 3, these women were more likely to marry men with fewer years of school and lower job prestige. These effects are significant at the 1 % level. Similar effects are shown in columns 4–6 for males born in 1918/1919, who are 14 percentage points more likely to have married flu-exposed females (p = .00) and married females with 0.26 fewer years of schooling (p = .101). Given the findings of Table 3, however, we cannot rule out that this is a marriage market effect.12

Second Generation

Our hypothesis is that the direct effects observed in Tables 2 and 3 extend into future generations. To address this hypothesis, we regress a number of economic and health outcomes of the WLS graduates and siblings, the offspring of the first generation examined in Tables 2 and 3. These estimations are performed in Tables 4 and 5. In both tables, column 1 focuses on an indicator for either parent being born in 1918/1919; the primary regressor of column 2 is an indicator for fathers’ (Generation 1 males’) in utero exposure to the 1918 flu; column 3 considers mothers’ (Generation 1 females’) exposure; and column 4 includes separate indicators for both fathers’ and mothers’ exposure. Controls included in all columns include father’s year of birth and its square, mother’s year of birth and its square, an indicator for sex in the second generation, a measure of birth order in the second generation, and second-generation year of birth and its square.

Table 4 focuses on economic outcomes of the WLS graduates. To reiterate, our hypothesis is that in utero exposure to the 1918 influenza pandemic has effects that persist for multiple generations. Table 2 shows the direct first-generation effects; Table 4 begins to show the indirect effects that are transmitted to offspring. Panel A regresses years of schooling in the second generation on indicators of first-generation exposure to the 1918 flu.13 As shown in column 1, either parent being born in 1918/1919 is associated with a statistically significant decline of 0.21 years of schooling in the second generation. This estimate is likely understated because the sampling design of the WLS focuses on high school seniors; thus, individuals with fewer than 12 years of schooling are underrepresented in the data.14 Columns 2–4 disambiguate this effect into the maternal and paternal lines and show that the effect of column 1 seems to be driven by mother’s in utero exposure to the flu.15

The findings of panel A are extended into panel B, which replaces years of schooling as the dependent variable with the natural log of family income for the 1992 wave, a time when second-generation respondents were, on average, age 53—a time of peak earning in the life course. As with years of schooling, a negative association is seen throughout the specifications of panel B. Having either parent born in 1918/1919 is associated with a 22 % decline in family income. Once again, this effect seems to be driven by the mother’s (not the father’s) exposure. Panel C replaces income with a measure of net worth for the 2004 wave. This measure of net worth is when the second-generation respondents were, on average, 65 years of age and is representative of earnings throughout the life course. The dual indicator of column 1 is negative but statistically insignificant at conventional levels (p = .23). As shown in column 3, however, the effect of mother’s exposure becomes statistically significant at the 10 % level (p = .069), implying that those with mothers born during 1918/1919 had approximately $36,000 less in net worth by 2004.

Following the broader health focus of the Barker hypothesis, we replace the economic effects of Table 4 with health measures in Table 5. We consider three broad measures of general health and well-being later in life: self-reported health in later life (~53 years old), height, and BMI (again, at ~53 years old). The use of self-reported health is intended to capture general well-being later in life; this is tested in panel A of Table 5. Mazumder et al. (2010) provided evidence that in utero exposure to the 1918 flu is associated with a reduction in height and increased cardiovascular disease, so these negative health outcomes could possibly be transmitted to offspring; this is tested in panel B of Table 5.16 And given prior work linking early-life exposure to the Dutch famine to BMI (Stein et al. 2007) and metabolic function (Lumey et al. 2009) in adulthood, as well as the fact that both the Dutch famine and the 1918 flu impacted maternal stress and nutritional status,17 we examine BMI as another broad indicator of health for the second generation; this is tested in panel C of Table 5. A consistent pattern emerges across all panels. Like SES outcomes in Table 4, Generation 1 female exposure to the flu is shown to have negative and statistically significant association with all health outcomes in the second generation.

From panel A, parental (Generation 1) flu exposure reduces self-reported health by roughly one-half point (p < .05) on a Likert scale ranging from 1 to 5, with 1 being very poor health and 5 being excellent health. As shown, and like prior estimates, this effect is driven by Generation 1 female exposure. For adult height, Generation 1 female exposure is associated with a reduction of roughly 0.2 inches (p < .10), which is larger than the direct effect of 0.05 inches that Mazumder et al. (2010) documented; however, Mazumder et al.’s estimated effect falls within the 95 % confidence interval of our estimated coefficient. Panel C considers the effect on second-generation BMI. From column 1, either parent having been born in 1918/1919 is associated with a statistically significant increase in the offspring’s BMI of 0.41 points. As with the economic effects of Table 4, this increase in BMI appears to be driven by mother’s exposure. This is shown in columns 2–4, which estimate a statistically significant positive coefficient for mother’s exposure but a coefficient that is statistically indistinguishable from 0 for father’s exposure.

A persistent effect of in utero exposure to the 1918 flu is shown in Tables 4 and 5. First-generation exposure consistently has a statistically significant and economically meaningful effect on second-generation health and economic outcomes. This effect appears to be driven solely by mother’s exposure; because of marriage market associations, though, we cannot determine definitively that mother’s exposure produces this multigenerational effect through biological mechanisms.

Tables A10A11 in the online appendix reestimate Tables 4 and 5, splitting the sample by Generation 2 sex. In short, the negative effects of first-generation exposure appear to be more prominent in second-generation males, such that the harmful multigenerational effects are most prominent in second-generation men whose mothers were exposed to the 1918 flu.18 Potential reasons for this sex disparity could be tied to general trends in the differences between sons and daughters in intergenerational mobility (Chadwick and Solon 2002). In other words, sons are more likely to inherit their family’s SES standing than daughters. This implies that the negative Generation 1 shock from in utero exposure is more likely to affect sons through the proposed SES channel than daughters, resulting in a larger estimated effect for Generation 2 males. Furthermore, there appears to be a growing gap in the intergenerational elastiticy between sons and daughters during the period in question (Olivetti and Paserman 2015). Another potential reason for the sex difference could be the larger standard deviation in Generation 2 male outcomes, as shown in Table A9 (online appendix). Finally, the findings shown in Tables A10A11 (online appendix) differ from those of Richter and Robling (2013), who showed homogeneous sex effects from in utero flu exposure for education: Generation 1 female exposure led to a reduction in Generation 2 female education, and Generation 1 male exposure is associated with a reduction in Generation 2 male education. Given this disparity between the two studies, our underrepresentation for Generation 1 male exposure, and the mixed findings in Tables A1011 (online appendix), we view our general finding of Generation 1 female exposure working through Generation 2 male outcomes as preliminary and suggestive. Indeed, as shown in Table A12, many of the sex differences in Generation 2 are not statistically significant.

Third Generation

As mentioned previously, the focus of the WLS data collection is high school graduates in 1957, but additional data have been collected for a number of variables for samples of the children of these graduates. Table 6 explores the effects of the WLS parents on the years of schooling for WLS children, estimating the effect of flu exposure across three generations.19

Consistent with the second-generation estimations, a persistent effect of the in utero flu exposure is observed, and this effect seems to be driven by grandmother exposure (Generation 1 female). The joint indicator for either parent is negative and close to statistical significance at the 10 % level (p = .106), suggesting that exposure in the first generation is associated with 0.12 fewer years of schooling in the third generation. The effect of grandfather exposure is insignificantly different from 0, but the coefficient of grandmother exposure is similar in magnitude to the joint estimate of column 1 and is statistically significant at the 10 % level (p = .066). Furthermore, the decline in schooling seen in the third generation is similar but slightly smaller in magnitude compared with the second-generation effects shown in panel A of Table 4.20 This again indicates a persistent effect that may be attenuating. Importantly, although the estimate for grandfather exposure is insignificant, it is consistent with intergenerational inheritance (Kuzawa and Eisenberg 2014). To say the same for grandmother exposure would require effects to persist into the fourth generation (given the differences in gametic development), and these data are not available in the WLS.

Conclusion

This study presents novel evidence of multigenerational effects of in utero health insults. We use the sudden and unexpected influenza pandemic in 1918/1919 to trace out the effects of in utero exposure to infectious disease on own outcomes for the first generation, children’s outcomes, and grandchildren’s outcomes. We find that this exposure reduces educational attainment and related economic outcomes across three generations. For example, individuals in our second generation—the WLS graduates and their siblings who have mothers who were exposed to the 1918 flu in utero—completed one-fifth year (2.4 months) less schooling. The size of our estimated effect aligns well with those of Richter and Robling (2013), who found that first-generation in utero exposure to the 1918 flu results in one-sixth fewer years of schooling (2 to 2.5 fewer months of schooling) for the maternal line. We then extend results to the third generation for the first time in the literature and find that individuals in the third generation, who have grandmothers who were exposed to the pandemic in utero, complete one-seventh fewer years of schooling (1.7 fewer months of schooling) than individuals without affected grandmothers.

Several potential mechanisms could explain the persistence of poor outcomes across three generations we find in our data. As previously outlined, the intergenerational persistence in poor outcomes due to early-life health shocks could occur through socioeconomic channels, whereby a fetally insulted person with resulting low educational attainment raises a child in a low-resource environment that reduces opportunities for high educational attainment of the child. For multigenerational effects to be revealed, the reduced opportunities for high educational attainment of the child must then reduce opportunities for their own children. Alternatively, intergenerational transmission may occur through epigenetic inheritance through the germ line, a distinct possibility based on findings of the Överkalix studies, or phenotype-to-phenotype transmission and cumulative intergenerational phenotypic change.

Our analysis is unable to fully distinguish between these proposed intergenerational channels—the socioeconomic and biological—but it is important to note that the channels are by no means mutually exclusive. Further analysis (Tables A3A5, online appendix) that is intended to partially account for the socioeconomic mechanism leads to mixed results in which the coefficient of first-generation flu exposure is attenuated to insignificance for some outcomes but not others. This provides evidence for a socioeconomic channel but does not eliminate the possibility of epigenetic mechanisms or other biological channel through which SES “gets under the skin” to influence outcomes of subsequent generations (e.g., the socioeconomic circumstances of one generation may be passed via biology and/or socioeconomics to other generations). Likewise, the estimates of grandfather effects on grandchildren are imprecise but consistent with transgenerational inheritance (Kuzawa and Eisenberg 2014). Additionally, although the WLS provides a unique framework to analyze multiple generations, its measurement of flu exposure is somewhat crude (from self-reported parents’ year of birth) and is in the right tail of the distribution of parent birth years, suggesting the possibility of confounding from resource-poor younger mothers.21 That is, exposure to influenza in the first generation is mechanically tied to mother’s ages of 20 and 21 because the WLS sample is drawn based on having a child who was born in 1939 (and thus who graduated from high school in 1957). Although we cannot definitively separate these two effects, our ability to compare outcomes of slightly older mothers (age 22 or 23, born in 1917 or 1916) provides evidence more consistent with in utero exposure to influenza than impacts of having a mother who is 20 or 21 years old.

From a policy perspective, our evidence may suggest a novel source of multigenerational persistence in poverty through biosocial factors and may point to a need to consider evidence of transgenerational social and/or biological mechanisms. We document the extent to which harmful early-life environments cascade through generations, promoting a disadvantaged start for those whose grandparents were exposed to a hazardous early-life environment.

Acknowledgments

This research uses data from the Wisconsin Longitudinal Study (WLS) of the University of Wisconsin–Madison. Since 1991, the WLS has been supported principally by the National Institute on Aging (AG-9775, AG-21079, AG-033285, and AG-041868), with additional support from the Vilas Estate Trust, the National Science Foundation, the Spencer Foundation, and the Graduate School of the University of Wisconsin–Madison. Since 1992, data have been collected by the University of Wisconsin Survey Center. A public-use file of data from the Wisconsin Longitudinal Study is available from the Wisconsin Longitudinal Study, University of Wisconsin–Madison, 1180 Observatory Drive, Madison, WI 53706; and at http://www.ssc.wisc.edu/wlsresearch/data/. The opinions expressed herein are those of the authors. Forgues acknowledges support from the National Institute on Aging Training Grant T32 AG00129 at the University of Wisconsin–Madison. Authorship for this article is alphabetical.

Notes

1

As mentioned in greater detail in the Data section, the first generation is composed of the parents from the Wisconsin Longitudinal Study (WLS). This is the generation that is treated, or exposed in utero to the 1918 flu epidemic. The second generation is composed of the WLS graduates and siblings, who are the focus of the WLS, and the third generation is composed of the offspring of the WLS graduates and siblings.

2

Cohen et al. (2010) and Fletcher (2018) found no effects on overall mortality.

3

Van den Berg and Pinger (2014) externally validated the potential for transmission across three generations by analyzing the impact of the German famine of 1916–1918 on the mental health outcomes (an index from survey questions accounting for general mental health, emotional functioning, social functioning, and vitality) of the children and grandchildren of those exposed to the famine during their SGP. The authors found that paternal (maternal) grandfathers’ (grandmothers’) exposure during their SGP is associated with better mental health in grandsons (granddaughters).

4

Parent year of birth is recorded from the WLS graduates. We use this self-reported measure to create the indicator of flu exposure: those parents born in 1918 or 1919. To maximize sample size, we use differing waves of the WLS and reported birth years from biological siblings to supplement the reported year of birth from WLS graduates. Parent birth years are first collected for the 1992/1993 graduate wave. Missing observations are then sequentially added from graduate reports in later waves—2003 and 2011. After using all graduate-reported year of parent birth variables, we then fill in observations from identically reported measures from a selected biological sibling. Specifically, for mothers (Generation 1 females), 7,976 observations come from graduates in the 1992/1993 wave, 366 additional observations come from the 2003/2005 wave, 84 observations come from 2011 wave, and 68 observations come from the selected sibling. This leaves a base sample size of 8,494, ~94 % of which is from the graduate reported 1992/1993 wave. For fathers (Generation 1 males), 7,929 observations come from graduates in the 1992/1993 wave, 361 additional observations come from the 2003/2005 wave, 77 come from the 2011 wave, and 91 observations come from siblings, leaving a base sample size of 8,458 (again, ~94 % of which is from the graduate-reported 1992/1993 wave). From this base, 7 observations are dropped for Generation 1 female’s (WLS mother’s) year of birth because reported Generation 2’s year of birth was 10 or fewer years after Generation 1’s year of birth; 45 additional observations are lost if this threshold is increased to 15 years. For Generation 1 male’s year of birth, 2 observations are removed for identical reasons; this increases to 25 additional observations for the 15-year threshold. We attribute this reduction in the sample mostly to measurement error.

5

The 1918 flu epidemic in Wisconsin was from September 1918 through December 1918, but it did not reach the severity experienced in many other states. According to historical records, Wisconsin had the fourth-lowest number of deaths of 25 reporting states (Burg 2000; Shors and McFadden 2009).

6

Tables A7 and A8 in the online appendix provide evidence that our results are not driven solely by confounding between exposure to the influenza pandemic and being young mothers.

7

The index of SES is a factor-weighted score combining data on father’s and mother’s years of schooling, father’s occupational prestige, and average parental income. Replacing this measure with average parental income (see Table A1, online appendix) does not change the effect of the flu indicator. Job prestige measures for both mother and father are based on Duncan’s socioeconomic index, which is a measure of job prestige based on income, education, and surveyed perceptions of general social standing for certain occupations (Duncan 1961).

8

Our three generations of individuals are drawn from three nonoverlapping sets of birth years; the mean birth year for Generation 1 is ~1910, Generation 2 is ~1940, and Generation 3 is ~1965. Thus, because we are conduct the analysis separately by generation, we control for much of the time variation in the meaning of education. We also conduct the analysis stratified by sex in Tables A10A12 (online appendix) so that we can control for the differential meaning of education in each generation.

9

First-generation years of schooling are reported by the second-generation WLS graduates. Measurement error is likely, which may result in the insignificant coefficients of Table 2.

10

Richter and Robling (2013: table 12) found a similar effect for in utero female exposure to flu in the first trimester.

11

Mothers and fathers of WLS graduates are assumed to be married.

12

Table A2 in the online appendix repeats the estimation of Table 3 while also controlling for an indicator of own flu exposure and own year-of-birth measures.

13

Table A6 in the online appendix includes a number of alternative measures for years of schooling in place of the simple count used in Table 4.

14

Table 3 includes selected siblings of the WLS graduates. These siblings do not have to be high school graduates.

15

This finding differs from that of Richter and Robling (2013), who found that maternal exposure was tied to daughters’ outcomes and paternal exposure tied to sons’ outcomes.

16

Table A14 in the online appendix tests indicators for a number self-reported health conditions. A weak positive association is shown between Generation 1 female flu exposure and an indicator of cardiovascular disease during the 1992/1993 wave of the WLS (Table A15, panel A, online appendix). This association, however, is reduced in magnitude and statistical significance for later-life waves of the WLS (Table A15, online appendix, panels B and C).

17

The Dutch famine influenced maternal nutritional status directly through caloric restriction; the 1918 flu influenced similar nutrition through symptoms, such as appetite loss, vomiting, and/or diarrhea.

18

When we omit those with no family income during the 1992 wave of the WLS, negative effects from Generation 1 female flu exposure are primarily seen in second-generation males. This indicates, however, a greater frequency of no income among second-generation females from Generation 1 flu exposure. Additionally, when looking at net worth in the 2011 wave of the WLS, we see no difference by Generation 2’s sex.

19

Analysis of the WLS graduate/sibling children (i.e., the third generation) restricts the sample to those children who are biological children and who were age 35 or older by the 2003/2004 wave of the WLS.

20

For Generation 1 female exposure in column 4, the p value for a difference in coefficients between panel A of Table 4 and Table 6 is 0.411, implying that the effect in the third generation is statistically indistinguishable from that in the second generation.

21

We specifically explore the confounding effects of younger mothers in Tables A7 and A8 (online appendix). These tables show that younger mothers are indeed initially disadvantaged (i.e., fewer years of schooling); however, this young-mother disadvantage does not persist in subsequent generations. Rather, only those mothers born in the 1918–1919 range have significant negative effects on later generations.

References

Almond, D. (
2006
).
Is the 1918 Influenza pandemic over? Long-term effects of in utero influenza exposure in the post-1940 U.S. population
.
Journal of Political Economy
,
114
,
672
712
. 10.1086/507154
Almond, D., & Mazumder, B. (
2005
).
The 1918 influenza pandemic and subsequent health outcomes: An analysis of SIPP data
.
American Economic Review: Papers & Proceedings
,
95
,
258
262
. 10.1257/000282805774669943
Barker, D. J. (
2007
).
The origins of the developmental origins theory
.
Journal of Internal Medicine
,
261
,
412
417
. 10.1111/j.1365-2796.2007.01809.x
Benyshek, D. C. (
2013
).
The “early life” origins of obesity-related health disorders: New discoveries regarding the intergenerational transmission of developmentally programmed traits in the global cardiometabolic health crisis
.
American Journal of Physical Anthropology
,
152
(
S57
),
79
93
. 10.1002/ajpa.22393
Brown, A. S., & Derkits, E. J. (
2010
).
Prenatal infection and schizophrenia: A review of epidemiologic and translational studies
.
American Journal of Psychiatry
,
167
,
261
280
. 10.1176/appi.ajp.2009.09030361
Burg, S. (
2000
).
Wisconsin and the great Spanish Flu epidemic of 1918
.
Wisconsin Magazine of History
,
84
(
1
),
36
56
.
Bygren, L. O., Kaati, G., & Edvinsson, S. (
2001
).
Longevity determined by paternal ancestors’ nutrition during their slow growth period
.
Acta Biotheoretica
,
49
,
53
59
. 10.1023/A:1010241825519
Canetta, S. E., & Brown, A. S. (
2012
).
Prenatal infection, maternal immune activation, and risk for schizophrenia
.
Translational Neuroscience
,
3
,
320
327
. 10.2478/s13380-012-0045-6
Catalano, R., Goldman-Mellor, S., Saxton, K., Margerison-Zilko, C., Subbaraman, M., LeWinn, K., & Anderson, E. (
2011
).
The health effects of economic decline
.
Annual Review of Public Health
,
32
,
431
450
. 10.1146/annurev-publhealth-031210-101146
Chadwick, L., & Solon, G. (
2002
).
Intergenerational income mobility among daughters
.
American Economic Review
,
92
,
335
344
. 10.1257/000282802760015766
Cohen, A. A., Tillinghast, J., & Canudas-Romo, V. (
2010
).
No consistent effects of prenatal or neonatal exposure to Spanish flu on late-life mortality in 24 developed countries
.
Demographic Research
,
22
,
579
634
. 10.4054/DemRes.2010.22.20
Duncan, O. (
1961
).
A socioeconomic index for all occupations
. In Reiss, A. J.Jr (Ed.),
Occupations and social status
(pp.
109
138
).
New York, NY
:
Free Press
.
Fatemi, S. H., Earle, J., Kanodia, R., Kist, D., Emamian, E. S., Patterson, P. H., … Sidwell, R. (
2002
).
Prenatal viral infection leads to pyramidal cell atrophy and macrocephaly in adulthood: Implications for genesis of autism and schizophrenia
.
Cellular and Molecular Neurobiology
,
22
,
25
33
.
Fletcher, J. M. (
2018
).
Examining the long-term mortality effects of early health shocks
.
Applied Economics Letters
,
26
,
902
908
. 10.1080/13504851.2018.1520960
Kash, J. C., Tumpey, T. M., Proll, S. C., Carter, V., Perwitasari, O., Thomas, M. J., … Katze, M. G. (
2006
).
Genomic analysis of increased host immune and cell death responses induced by 1918 influenza virus
.
Nature
,
443
,
578
581
.
Kobasa, D., Jones, S. M., Shinya, K., Kash, J. C., Copps, J., Ebihara, H., … Kawaoka, Y. (
2007
).
Aberrant innate immune response in lethal infection of macaques with the 1918 influenza virus
.
Nature
,
445
,
319
323
.
Kuzawa, C. W., & Eisenberg, D. T. (
2014
).
The long reach of history: Intergenerational and transgenerational pathways to plasticity in human longevity
. In Weinstein, M., & Lane, M. A. (Eds.),
Sociality, hierarchy, health: Comparative biodemography: A collection of papers
(pp.
65
94
).
Washington, DC
:
National Academies Press
.
Li, W-Y, Chang, Y-C, Lee, L-H, & Lee, L-J (
2014
).
Prenatal infection affects the neuronal architecture and cognitive function in adult mice
.
Developmental Neuroscience
,
36
,
359
370
. 10.1159/000362383
Lin, M-J, & Liu, E. M. (
2014
).
Does in utero exposure to illness matter? The 1918 influenza epidemic in Taiwan as a natural experiment
.
Journal of Health Economics
,
37
,
152
163
. 10.1016/j.jhealeco.2014.05.004
Lindeboom, M., Portrait, F., & Van den Berg, G. J. (
2010
).
Long-run effects on longevity of a nutritional shock early in life: The Dutch Potato Famine of 1846–1847
.
Journal of Health Economics
,
29
,
617
629
. 10.1016/j.jhealeco.2010.06.001
Lumey, L., Stein, A. D., Kahn, H. S., & Romijn, J. (
2009
).
Lipid profiles in middle-aged men and women after famine exposure during gestation: The Dutch Hunger Winter Families Study
.
American Journal of Clinical Nutrition
,
89
,
1737
1743
. 10.3945/ajcn.2008.27038
Mamelund, S. E. (
2006
).
A socially neutral disease? Individual social class, household wealth and mortality from Spanish influenza in two socially contrasting parishes in Kristiania 1918–19
.
Social Science & Medicine
,
62
,
923
940
. 10.1016/j.socscimed.2005.06.051
Mazumder, B., Almond, D., Park, K., Crimmins, E. M., & Finch, C. E. (
2010
).
Lingering prenatal effects of the 1918 influenza pandemic on cardiovascular disease
.
Journal of Developmental Origins of Health and Disease
,
1
,
26
34
. 10.1017/S2040174409990031
Miller, V. M., Zhu, Y., Bucher, C., McGinnis, W., Ryan, L. K., Siegel, A., & Zalcman, S. (
2013
).
Gestational flu exposure induces changes in neurochemicals, affiliative hormones and brainstem inflammation, in addition to autism-like behaviors in mice
.
Brain, Behavior, and Immunity
,
33
,
153
163
.
Myrskylä, M., Mehta, N. K., & Chang, N. K. (
2013
).
Early life exposure to the 1918 influenza pandemic and old-age mortality by cause of death
.
American Journal of Public Health
,
103
(
7
),
e83
e90
. 10.2105/AJPH.2012.301060
Olivetti, C., & Paserman, M. D. (
2015
).
In the name of the son (and the daughter): Intergenerational mobility in the United States, 1850–1940
.
American Economic Review
,
105
,
2695
2724
. 10.1257/aer.20130821
Pembrey, M. E., Bygren, L. O., Kaati, G., Edvinsson, S., Northstone, K., Sjostrom, M., & Golding, J. (
2006
).
Sex-specific, male-line transgenerational responses in humans
.
European Journal of Human Genetics
,
14
,
159
166
. 10.1038/sj.ejhg.5201538
Richter, A., & Robling, P. O. (
2013
).
Multigenerational effects of the 1918–19 influenza pandemic in Sweden
(Working Paper Series 5/2013).
Stockholm, Sweden
:
Stockholm University, Swedish Institute for Social Research
.
Roseboom, T. J., van der Meulen, J. H., Ravelli, A. C., Osmond, C., Barker, D. J., & Bleker, O. P. (
2001
).
Effects of prenatal exposure to the Dutch famine on adult disease in later life: An overview
.
Molecular and Cellular Endocrinology
,
185
,
93
98
. 10.1016/S0303-7207(01)00721-3
Schetter, D. (
2011
).
Psychological science on pregnancy: Stress processes, biopsychosocial models, and emerging research issues
.
Annual Review of Pyschology
,
62
,
531
558
. 10.1146/annurev.psych.031809.130727
Schulz, L. C. (
2010
).
The Dutch Hunger Winter and the developmental origins of health and disease
.
Proceedings of the National Academy of Sciences
,
107
,
16757
16758
. 10.1073/pnas.1012911107
Shors, T., & McFadden, S. H. (
2009
).
1918 influenza: A Winnebago County, Wisconsin perspective
.
Clinical Medicine & Research
,
7
,
147
156
. 10.3121/cmr.2009.863
Stein, A. D., Kahn, H. S., Rundle, A., Zybert, P. A., van der Pal-de Bruin, K., & Lumey, L. H. (
2007
).
Anthropometric measures in middle age after exposure to famine during gestation: Evidence from the Dutch famine
.
American Journal of Clinical Nutrition
,
85
,
869
876
. 10.1093/ajcn/85.3.869
Sydenstricker, E. (
1931
).
The incidence of influenza among persons of different economic status during the epidemic of 1918
.
Public Health Reports
,
46
,
154
170
. 10.2307/4579923
Torche, F. (
2011
).
The effect of maternal stress on birth outcomes: Exploiting a natural experiment
.
Demography
,
48
,
1473
1491
. 10.1007/s13524-011-0054-z
Van den Berg, G. J., & Pinger, P. R. (
2014
).
A validation study of transgenerational effects of childhood conditions on the third generation offspring's economic and health outcomes potentially driven by epigenetic imprinting
(IZA Working Paper No. 7999).
Bonn, Germany
:
Institute for the Study of Labor
.

Supplementary data