Economic Downturns and Inequities in Birth Outcomes: Evidence From 149 Million US Births

Abstract

Using birth certificate data for nearly all registered US births from 1976 to 2016 and monthly data on state unemployment rates, we reexamined the link between macroeconomic variation and birth outcomes. We hypothesized that economic downturns reduce exposure to work-related stressors and pollution while increasing exposure to socioeconomic stressors like job loss. Because of preexisting inequalities in health and other resources, we expected that less-educated mothers and black mothers would be more exposed to macroeconomic variation. Using fixed-effect regression models, we found that a 1-percentage-point increase in state unemployment during the first trimester of pregnancy increased the probability of preterm birth by 0.1 percentage points, while increases in the state unemployment rate during the second/third trimester reduced the probability of preterm birth by 0.06 percentage points. During the period encompassing the Great Recession, the magnitude of these associations doubled in size. We found substantial variation in the impact of economic conditions across different groups, with highly educated white women least affected and less-educated black women most affected. The results highlight the increased relevance of economic conditions for birth outcomes and population health as well as continuing, large inequities in the exposure and impact of macroeconomic fluctuations on birth outcomes.

Numerous studies have examined the effect of macroeconomic conditions on population health (1–9). Four studies on newborn health using US data have come to differing conclusions: Three earlier studies found newborn health to improve during economic downturns (10–12), while a recent study found increased risks of preterm birth (PTB) during downturns experienced in the first trimester and decreased risk of PTB for downturns experienced in the second trimester of pregnancy (13). The earlier studies found substantial variation in the impact of economic conditions by maternal race (10, 11), while the latter found none (13). In this study, we attempted to reconcile these findings and shed new light on the unequal impact of economic downturn in recent data covering the Great Recession.

Protective effects likely reflect a reduction in economic activity, work hours, and pollution during economic downturns (14–17). Ambient air pollution, which increases the risk of adverse birth outcomes (18–21), declines during downturns (16, 17, 22). Both physical work and work-related psychosocial stress have been linked to adverse birth outcomes (23–25), and by reducing work hours, pregnant women’s exposure to work-related stressors might decline during recessions.

Because racial/ethnic minority and low socioeconomic status are associated with increased exposure to environmental hazards and air pollution (26–30), we expected economic downturns to have stronger effects on black women and less-educated women. Similarly, low education and racial/ethnic minority status increase the risk of working in unsafe/hazardous working conditions, including intense physical work, exposure to toxic substances, discrimination, and harassment (24, 31–33). These elevated toxic exposures would likely magnify the protective impact of economic downturns among less-educated women and black women, as found in previous studies (10, 11).

Other research suggests that recessions have adverse effects on newborn health (34, 35). One study found that downturns experienced during the first trimester increase the risk of PTB, while second-trimester downturns decrease the risk of PTB (13). Hazardous first-trimester effects could be due to increased exposure to (or perceived risk of) adverse financial and economic events that occur with greater frequency during recessions: loss of employment, income, assets, and social status (14, 15, 36). Less-educated women and black women would likely be more affected. Both are at greater risk of job loss (37, 38) and, because of lower wealth, more preexisting health conditions, and less access to high-quality health care (39–42), more likely to experience health declines in response to economic shocks.

Our study period encompassed the labor market downturn and recovery related to the Great Recession, during which we expected the effect of economic conditions to be amplified. Unlike other downturns in our 40-year study period or since the Great Depression, the labor market downturn during the Great Recession was exceptionally deep, long-lasting, and flanked by a historic decline in household assets (14, 15, 42). These conditions likely combined to magnify the impact of economic conditions on birth outcomes, especially for groups that were more vulnerable economically.

METHODS

Data

Data on birth outcomes came from the National Center for Health Statistics (NCHS) and are based on birth certificates for all registered births occurring in the US from 1977 to 2016. Using information on gestational age, we estimated the month of conception and included all births that resulted from conceptions between March 1976 and March 2016 for mothers residing in one of the 50 US states or the District of Columbia—150,745,268 births in total. We excluded births with missing information on race, birth weight, gestational age, and state economic conditions. The resulting data set included 148,973,111 births.

Gestational age is reported in weeks, based on either the last menstrual period or an obstetrical/clinical estimate. Only year and month of birth are available in the National Center for Health Statistics data. To estimate calendar month of conception, we converted gestational age from weeks to months by dividing by 4.3 and rounding to the nearest integer. We then subtracted monthly gestational age from calendar month of birth to obtain calendar month of conception. A simulation (see Web Appendix 1, available at https://academic.oup.com/aje) shows that this method correctly identifies calendar month of conception in 79% of cases and is never off by more than 1 month.

This study was approved by the Brandeis University Human Research Protection Program for the protection of human subjects (institutional review board protocol 18008).

Outcomes

We defined PTB as births occurring before the 37th week of gestation (vs. after) and low birth weight (LBW) as being born weighing less than 2,500 g (vs. higher). A newborn is classified as small for gestational age (SGA) if weighing less than the 10th percentile of their sex- and plurality-specific (single vs. multiple) birth weight distribution for a given gestational age in weeks (vs. greater) (43).

Economic conditions

We used seasonally adjusted, monthly state unemployment rates as the primary measure of macroeconomic conditions. Data are from the Bureau of Labor Statistics, “Employment status of the civilian noninstitutional population, seasonally adjusted” series (44). Previous research suggests that state-level measures of economic conditions have larger effects on birth outcomes than do county-level measures (12). For sensitivity analysis, we used monthly employment-population ratios (100 times the number of employed divided by the population), which produced qualitatively similar results but smaller coefficient estimates.

We measured unemployment prior to conception as the 3-month average unemployment rate during the 3 months before the month of conception. During pregnancy, we measured first-trimester unemployment as the 3-month average unemployment rate from the month of conception to the end of the first trimester. We measured unemployment during the second and third trimesters as the 6-month average unemployment rate from the beginning of the second to the end of the third trimester. To reconcile our findings with previous studies that have used annual measures of state economic conditions (10–12), we also measured economic conditions as the 12-month average unemployment rate from 3 months prior to conception through the end of the third trimester.

Modifiers

Maternal race was categorized as black, other, or white. Hispanic origin cannot be consistently distinguished over the observation period. Maternal education was categorized as less than high school (0–11 years of education), high school (12 years), some college (13–15 years), and college or higher (≥16 years).

We conducted separate analyses for the most recent 10 years covering the Great Recession (i.e., all conceptions between March 2006 and March 2016). While the National Bureau of Economic Research’s Business Cycle Committee dates the Great Recession from December 2007 to June 2009, the labor market downturn persisted well beyond the official end date.

Other covariates

Other covariates included: maternal age in years (15–46 years), nativity (US- or foreign-born), plurality (singleton vs. twins or more), live birth order (first, second, third, or fourth or later), and infant’s sex (male, female).

Estimation

Estimation was performed using grouped data, where each cell is defined by all existing combinations of values of the variables analyzed, using the number of births in each cell as frequency weights. We relied on linear probability models estimated via ordinary least squares, which is computationally efficient, given the large number of observations and parameters we fitted, and yields estimates that have an intuitive interpretation, revealing the magnitude of effects in absolute terms. Analysis was performed in Stata, version 14 (MP) (StataCorp LLC, College Station, Texas), using the reghdfe ado (45). Results using logistic regression are reported in Web Appendix 2 and Web Table 1. Our baseline model for individual birth outcomes takes the following form:

$$\begin{array}{c}P(Y=1{)}_{\mathit{ijt}}={\mathrm{\beta }}_1\mathit{Pr}{e}_{\mathit{jt}}+{\mathrm{\beta }}_2\mathit{Firs}{t}_{\mathit{jt}}+{\mathrm{\beta }}_{3}S\mathit{econdT}h\mathit{ir}{d}_{\mathit{jt}}\hfill \\ +{\mathrm{\alpha }}_j+{\mathrm{\theta }}_t+{\mathrm{\delta }}_{j}T+\mathbf{X}\mathbf{\gamma }+{\epsilon }_{\mathit{ijt}},\hfill \end{array}$$ (1)

where i is individuals, j indexes 51 states of maternal residence, and t indexes calendar months (month of conception) from March 1976 to March 2016. $P(Y=1)$ is the probability of an adverse birth outcome. ${\mathrm{\alpha }}_j$ are 51 state fixed effects, ${\mathrm{\theta }}_t$ are 470 monthly calendar-date fixed effects, and ${\mathrm{\delta }}_{j}T$ are 51 state-specific linear trends (i.e., interactions between state fixed effects and a linear trend variable with $T=1,\dots ,471$). $\mathbf{X}\mathbf{\gamma }$ is a matrix of covariates and its coefficient vector, included only in some sensitivity analyses.

We included the average unemployment rate from 3 months prior to the month prior to conception, $Pr{e}_{jt}$, to control for selection effects. Our main focus is on the average unemployment rate from the month of conception to the end of the first trimester, $Firs{t}_{jt}$, and the average unemployment rate for the second and third trimester, $SecondThir{d}_{jt}$. Not all pregnancies are carried to term, so the second/third trimester economic measures partially capture economic conditions after birth, introducing measurement error for some births. We repeated the analysis using average conditions in the second trimester only, which did not alter results.

Equation 1 estimates the effect of economic conditions on birth outcomes from within-state, month-to-month deviations in economic conditions relative to the national mean, exploiting variation in economic conditions across 24,021 state × calendar month contexts. State fixed effects, ${\mathrm{\alpha }}_j$, control for unobserved confounders that are state-specific and time constant. ${\mathrm{\theta }}_t$ calendar month of conception fixed effects control for time-varying confounders that are shared across states. State-specific linear trends, ${\mathrm{\delta }}_{j}T$, control for time-varying state-level variables that confound the association between economic conditions and birth outcomes.

Equation 1 differs from the empirical models in previous work (13). Instead of calendar month of conception, ${\mathrm{\theta }}_t$, Margerison-Zilko et al. (13) controlled for year of conception and month of birth. Month of conception controls for the effect of unobserved maternal characteristics that predict newborn health (46) and are confounded with economic conditions because of selective conceptions/terminations. In Web Appendix 2 and Web Figure 1, we illustrate that omitting month of conception likely introduces omitted variable bias in trimester-specific unemployment rate estimates, substantially inflating the size of estimated associations.

To reconcile our findings with previous work using annual measures of economic conditions (10–12), we replaced the 3 economic measures in equation 1 with a 12-month average unemployment rate from 3 months prior to conception to the end of the third trimester.

Robustness checks

We replaced state-specific linear trends with the full set of interactions between state and year-of-conception fixed effects (2,040 parameters), controlling nonparametrically for year-to-year change in unobserved confounders at the state level, such as state-specific year-to-year changes in health of mothers, socioeconomic conditions, and policies. To adjust for selection effects either prior to or during pregnancy, we adjusted for individual covariates that are fixed at conception. These covariates, $\mathbf{X}\mathbf{\gamma }$, enter either linearly or fully interacted (i.e., all possible interactions between all covariate levels (1,426 parameters)). While this might partially adjust for selection, the covariate distribution at birth is determined by partly unobserved selection processes, and adjusting for selected covariate values can induce rather than eliminate bias. These results should therefore be interpreted with some caution.

Statistical inference

We report 95% confidence intervals that are corrected for clustering of observations at the state level (47). Because our data includes nearly all registered births over a period of 40 years, we focus interpretation on the magnitude rather than statistical significance of estimated associations.

RESULTS

Table 1 presents summary statistics for all variables used in the analyses. Table 2 presents descriptive statistics for monthly state unemployment rates.

Table 1.

Descriptive Statistics for Births Used in Multivariate Analysis of Registered Births According to Conception Date, National Center for Health Statistics Natality Files, United States, 1976–2016

Variable	Conception
	March 1976 to March 2016		March 2006 to March 2016
	Births	%	Births	%
Sample size	148,973,111	100.00	40,704,848	100.00
Outcomes
Preterm birth	16,504,810	11.08	4,804,732	11.80
Low birth weight	11,070,374	7.43	3,303,381	8.12
Small for gestational age	15,367,463	10.32	4,217,684	10.36
Maternal race
Black	23,394,952	15.70	6,515,142	16.01
Other	8,311,507	5.58	3,146,833	7.73
White	117,266,648	78.72	31,042,874	76.26
Maternal education
Less than high school (0–11 years)	28,240,622	18.96	6,451,742	15.85
High school (12 years)	45,171,128	30.32	9,458,615	23.24
Some college (13–15 years)	30,990,860	20.80	10,108,445	24.83
College or more (≥16 years)	30,965,604	20.79	10,406,195	25.57
Missing	13,604,898	9.13	4,279,852	10.51
Nativity
US-born	120,781,672	81.08	31,060,762	76.31
Foreign-born	27,687,344	18.59	9,382,729	23.05
Missing	504,092	0.34	261,358	0.64
Maternal age, years
15–17	5,940,951	3.99	999,447	2.46
18–24	49,918,324	33.51	11,762,213	28.90
25–35	80,387,688	53.96	23,246,306	57.11
≥36	12,726,144	8.54	4,696,883	11.54
Live birth order
First	60,426,532	40.56	16,075,463	39.49
Second	47,873,480	32.14	12,853,571	31.58
Third	24,180,304	16.23	6,771,178	16.63
Fourth or later	16,195,820	10.87	5,004,637	12.29
Missing	296,977	0.20	0	0
Infant sex
Female	72,711,368	48.81	19,876,668	48.83
Male	76,261,744	51.19	20,828,180	51.17
Plurality
Singleton	144,770,160	97.18	39,302,584	96.56
Twins or more	4,202,950	2.82	1,402,265	3.44

Table 2.

Descriptive Statistics for State Unemployment Rate Variables Used in Multivariate Analysis, National Center for Health Statistics Natality Files, United States, 1976–2016^a

Average State Unemployment Rate	Minimum	25th Percentile	50th Percentile	75th Percentile	Maximum
Three months prior to conception to end of third trimester	−4.60	−1.19	−0.28	1.02	9.66
One to 3 months prior to conception	−4.98	−1.23	−0.26	1.05	11.60
Conception to end of first trimester	−4.95	−1.24	−0.27	1.04	11.57
Second to end of third trimester	−4.79	−1.24	−0.30	1.00	10.73

^a Unemployment were rates de-meaned using their state-specific means. Descriptive statistics are weighted by the number of births in each state and calendar month.

Table 3 shows estimates of the association of average state unemployment rates and the probability of PTB. The first data row reports estimates for the 12-month average unemployment rate covering the period from 3 months prior to conception until the end of the third trimester. Model 1 was our baseline model (equation 1), minus state-specific linear trends and covariates $\mathbf{X}\mathbf{\gamma }$. We observe that the negative estimate obtained using model 1 is not robust to adjustment for linear state-specific trends (model 2) or state × year-of-conception fixed effects (model 3). While the 12-month average unemployment around conception and during pregnancy is not robustly associated with PTB risk, we found, using more fine-grained measures of economic conditions, that the impact of economic conditions varied by stage of pregnancy.

Table 3.

Estimates of the Association of State Unemployment Around the Time of Conception and During Pregnancy With the Probability of Preterm Birth (n = 148,973,111), National Center for Health Statistics Natality Files, United States, 1976–2016^a

Average State Unemployment Rate	Model 1b		Model 2c		Model 3d
	B	95% CI	B	95% CI	B	95% CI
Estimates Using a Single Measure of State Unemploymente
Three months prior to conception to end of third trimester	−0.105	−0.172, −0.037	−0.018	−0.048, 0.011	0.021	−0.040, 0.082
Estimates Using 3 Separate Measures of State Unemploymentf
One to 3 months prior to conception	−0.104	−0.170, −0.038	−0.080	−0.122, −0.038	−0.066	−0.095, −0.036
Conception to end of first trimester	0.117	0.079, 0.156	0.122	0.081, 0.162	0.124	0.082, 0.167
Second to end of third trimester	−0.121	−0.159, −0.082	−0.063	−0.102, −0.025	−0.060	−0.121, 0.001

Abbreviations: B, ordinary-least-squares regression coefficient; CI, confidence interval.

^a All registered births to US residents conceived in the period from March 1976 to March 2016. Ordinary-least-squares linear probability estimates, multiplied by 100, and robust 95% confidence intervals.

^b All models controlled for state and calendar month of conception fixed effects.

^c Model 1 plus state-specific linear trends.

^d Model 2 plus full set of interactions between state and year-of-conception fixed effects.

^e Here we report results from a model that includes a single measure of state unemployment around the time of conception and during pregnancy (i.e., the average state unemployment rate over the period from 3 months prior to conception to the end of the third trimester).

^f Here we report results from a model that includes 3 separate measures of state unemployment within the above-noted period around the time of conception and during pregnancy (see equation 1, Methods).

We observed a negative association between state unemployment rates prior to pregnancy and PTB risk (Table 3), suggesting that during downturns fewer babies are conceived that would be born preterm. Second, we observed a negative association between unemployment rates during the second/third trimester and PTB risk, suggesting that declines in economic conditions experienced after the first trimester have small protective effects. For unemployment rates both before conception and at the third trimester, we observed that estimates shrank considerably after controlling for state-specific time-varying confounders. In contrast, we obtained robust evidence that declines in economic conditions during the first trimester were associated with increased PTB risk.

Model 2 is our preferred specification, yielding results similar to model 3, using many fewer parameters. The respective estimate for first-trimester unemployment indicates that with a 1-percentage-point increase in state unemployment rates during the first trimester, the probability of PTB rose by 0.12 percentage points. In 2016, 388,669 of 3,945,875 births were preterm, corresponding to a PTB rate of 9.85% (48). A hypothetical increase in this rate by 0.12 percentage points would translate into approximately 4,700 more babies born preterm, or 1.2% of all babies born preterm in that year.

Web Tables 2 and 3 contain result from identical analyses using LBW and SGA as outcomes. Results using the 12-month unemployment measure mirror those for PTB: For before-conception, first-trimester, and second/third-trimester unemployment, estimates were smaller and not statistically significant in most cases.

Table 4 reports covariate-adjusted estimates of the association between before-conception and trimester-specific unemployment rates and PTB. For model 1, we refitted the baseline model without covariates but omitting all observations with missing covariate information. Model 2 added the main effects of covariates, and model 3 adjusted for covariate main effects and all possible interactions.

Table 4.

Covariate-Adjusted Estimates of the Association of State Unemployment Around the Time of Conception and During Pregnancy With the Probability of Preterm Birth (n = 134,756,919), National Center for Health Statistics Natality Files, United States, 1976–2016^a

Average State Unemployment Rate	Model 1b		Model 2c		Model 3d
	B	95% CI	B	95% CI	B	95% CI
One to 3 months prior to conception	−0.080	−0.131, −0.029	−0.103	−0.152, −0.054	−0.104	−0.154, −0.055
Conception to end of first trimester	0.140	0.088, 0.191	0.168	0.111, 0.225	0.168	0.112, 0.225
Second to end of third trimester	−0.079	−0.123, −0.034	−0.086	−0.130, −0.041	−0.086	−0.130, −0.042

Abbreviations: B, ordinary-least-squares regression coefficient; CI, confidence interval.

^a Excluding births with missing covariate data. Ordinary-least-squares linear probability estimates, multiplied by 100, and robust 95% confidence intervals.

^b All models controlled for state and calendar month of conception fixed effects and state-specific linear trends. Observation with missing data on any covariate were excluded.

^c Model 1 plus the following covariates, main effects only: race, education, nativity, maternal age, live birth order, infant sex, and plurality.

^d Model 1 plus the following covariates, main effects and all possible interactions: race, education, nativity, maternal age, live birth order, infant sex, and plurality.

Omitting observations with missing covariate information increased all coefficient estimates (Table 3, model 2 vs. Table 4, model 1). Adjusting for covariates again increased coefficient estimates somewhat. Because of the large number of missing observations, primarily for maternal education, and because covariates are measured at birth and not at conception, these results must be interpreted with care. Taken at face value, the increase in coefficient estimates is not consistent with selection processes inducing substantial bias in coefficient estimates.

Web Tables 4 and 5 report results from identical analyses using LBW and SGA as outcomes. Compared with PTB, before-conception and first-trimester unemployment effects were smaller and not statistically significant. Second/third-trimester effects indicate small protective effects on LBW and SGA.

Table 5 reports results for race × education subgroups. All estimates are based on the baseline model specification (equation 1), excluding controls for maternal/infant characteristics. Results for births to all mothers are displayed, followed by results for births to highly educated (some college and college or higher education) white mothers, less-educated (at most a high-school diploma) white mothers, highly educated black mothers, and less-educated black mothers.

Table 5.

Group-Specific Estimates of the Association of State Unemployment Around the Time of Conception and During Pregnancy With the Probability of Preterm Birth, National Center for Health Statistics Natality Files, United States, 1976–2016^a

Average State Unemployment Rate	Full Sample (n = 148,973,111)b		White, High Education (n = 50,279,618)b		White, Low Education (n = 56,177,695)b		Black, High Education (n = 7,377,873)b		Black, Low Education (n = 14,210,821)b
	B	95% CI	B	95% CI	B	95% CI	B	95% CI	B	95% CI
One to 3 months prior to conception	−0.080	−0.122, −0.038	−0.036	−0.076, 0.003	−0.085	−0.150, −0.021	−0.032	−0.153, 0.088	−0.243	−0.348, −0.139
Conception to first trimester	0.122	0.081, 0.162	0.096	0.037, 0.155	0.133	0.065, 0.201	0.141	0.038, 0.244	0.331	0.201, 0.462
Second to end of third trimester	−0.063	−0.102, −0.025	−0.070	−0.118, −0.022	−0.062	−0.116, −0.008	−0.147	−0.252, −0.041	−0.169	−0.306, −0.031

Abbreviations: B, ordinary-least-squares regression coefficient; CI, confidence interval.

^a All registered births to US residents conceived in the period from March 1976 to March 2016. Ordinary-least-squares linear probability estimates, multiplied by 100, and robust 95% confidence intervals. All models controlled for state and calendar month of conception fixed effects and state-specific linear trends.

^b The number of births in the group-specific analyses do not add up to the number of births in the full sample because the full sample includes births to mothers of all races, including mothers classified as “other” in the National Center for Health Statistics Natality Files, and births to mothers for whom education data was missing.

Consistent with results reported in Tables 3 and 4, we observed for all groups a negative association between before-conception unemployment and PTB risk, a positive association of first-trimester unemployment and PTB risk, and a negative association between unemployment and PTB risk in the second/third trimester. However, the magnitude of estimates differs. Highly educated white mothers are least affected, while less-educated black mothers are most affected. For example, for highly educated white mothers, a 1-percentage-point increase in state unemployment rates was associated with a 0.1-percentage-point increase in the probability of PTB. The corresponding estimate for less educated black mothers is 0.3 percentage points, more than 3 times the size. We observed similar inequalities for before-conception and second/third-trimester estimates.

Web Tables 6 and 7 report results from identical analyses for LBW and SGA. The sign of most estimates was similar to PTB results, although coefficient estimates were smaller, and we did not observe a gradient across groups.

Table 6 reports results by period for the full sample, for conceptions between March 1976 and February 2006 and those from March 2006 to March 2016. All measures of economic conditions were more strongly associated with PTB in the most recent period compared with the preceding period. Economic measures during pregnancy approximately doubled in size.

Table 6.

Period-Specific Estimates of the Association of State Unemployment Around the Time of Conception and During Pregnancy With the Probability of Preterm Birth, National Center for Health Statistics Natality Files, United States, 1976–2016^a

Average State Unemployment Rate	March 1976 to March 2016 (n = 148,973,111)		March 1976 to February 2006 (n = 108,268,262)		March 2006 to March 2016 (n = 40,704,849)
	B	95% CI	B	95% CI	B	95% CI
One to 3 months prior to conception	−0.080	−0.122, −0.038	−0.069	−0.109, −0.030	−0.091	−0.142, −0.039
Conception to end of first trimester	0.122	0.081, 0.162	0.097	0.054, 0.139	0.211	0.116, 0.305
Second to end of third trimester	−0.063	−0.102, −0.025	−0.055	−0.099, −0.011	−0.106	−0.188, −0.024

Abbreviations: B, ordinary-least-squares regression coefficient; CI, confidence interval.

^a All registered births to US residents conceived in the period from March 1976 to March 2016. Ordinary-least-squares linear probability estimates, multiplied by 100, and robust 95% confidence intervals. All models controlled for state and calendar month of conception fixed effects and state-specific linear trends.

Web Tables 8 and 9 report results for LBW and SGA. In the most recent period (March 2006 to March 2016), associations for LBW mirror those for PTB, as do results for SGA, although the latter were generally small. These findings suggest that economic conditions primarily affected the duration of pregnancy, although during the Great Recession we also observed effects on LBW.

DISCUSSION

Using data on nearly all registered births over the period from 1977 to 2016, we found that a 1-percentage-point increase in state unemployment during the first trimester of pregnancy was associated with a rise in the probability of PTB by 0.1 percentage points, while increases in the state unemployment rate during the second and third trimester reduced the probability of PTB by 0.06 percentage points. During the period encompassing the Great Recession, the estimated magnitude of these associations doubled in size.

How do these results compare with previous studies? Using data for 1976–1999 and annual measures of state unemployment, Dehejia and Lleras-Muney (10) found protective effects of economic downturns, which were larger among babies born to black mothers. We replicated this protective effect for 2 birth outcomes (Web Tables 10 and 11), but we also showed that it disappears if we extend the analysis period through 2016. More importantly, our analyses illustrate that the 12-month average or annual measures of unemployment used in previous studies (10–12) fail to capture substantial variation in the impact of state unemployment according to stage of pregnancy. Margerison-Zilko et al. (13) have reported a similar result before, but our results differed from theirs in 2 key respects: After controlling for month of conception (rather than month of birth), we found smaller effects on average, and we found substantial variation in the impact of economic conditions across different groups, with highly educated white women least affected and less-educated black women most affected. This socioeconomic gradient is consistent with well-known inequalities in exposure to pollution, toxic work stress, and job loss. Our results therefore highlight social inequities that leave less-educated and black women and babies more vulnerable to socioeconomic forces beyond their control.

What role does selection play in generating these effects? Our results suggested that before-conception increases in unemployment reduced the prevalence of PTB, indicating positive selection into pregnancy during recessions. Furthermore, exposure to socioeconomic stress during the first trimester could lead to spontaneous abortion of frail fetuses, resulting in positive selection among fetuses surviving the first trimester. The protective effects of second/third-semester downturns could therefore be due to a selection effect (13). However, second/third-trimester effects remained essentially unchanged after conditioning on maternal/infant characteristics, which does not support the selection explanation. Furthermore, consistent with positive selection into pregnancy during downturns, we found that first-trimester effects on PTB rose after controlling for maternal/infant characteristics. In other words, failing to adjust for positive selection, we might underestimate the hazardous impact of economic downturns during the first trimester.

Rather than selection, we interpret these results as the joint outcome of 3 intertwined processes. First, recessions increase mothers’ exposure to toxic socioeconomic stressors (job loss, loss of social status). Second, recessions diminish mother’s exposure to toxic work-related and environmental stressors. While these exposures would likely be highly correlated over the course of pregnancy, vulnerability to socioeconomic stressors might diminish over the course of pregnancy (49, 50), while vulnerability to work-related and environmental stressors does not change (51–53). Previous research has suggested that women’s emotional responses to life events and the physiological stress response to psychosocial stressors become attenuated over the course of pregnancy (49, 50). Furthermore, some households consolidate or improve their socioeconomic standing over the course of pregnancy. For example, fathers experience an increase in wages and increase in work hours around childbirth (54, 55). Both processes would leave women less vulnerable to socioeconomic stressors triggered by downturns in the later stages of pregnancy. At the same time, they would likely benefit from reduced exposure to work-related stressors and pollution during downturns throughout pregnancy. At the aggregate level, the hazardous effects might dominate the protective effects in the early stages of pregnancy, while the protective effects might dominate in the later stages of pregnancy, generating the distinct trimester-specific pattern we observe.

A limitation of our study is that our estimated dates of conception and length of gestation are affected by measurement error. While the gestational age variable we used was, until 2013, largely based on an estimate of the last menstrual period, clinical estimates of gestation were available for all states starting in 2007. As a robustness check, we repeated the analysis using only clinical estimates of gestation, and this did not alter the results.

Another limitation of our study is that we do not explicitly model the multiple pathways through which recessions affect pregnant women. Our trimester-specific estimates are average effect estimates across all pregnant women, regardless of whether and how they were affected by business cycle–related exposures. Failing to distinguish women who experience downturns as protective (vs. hazardous), we obtain an average effect estimate that likely masks heterogeneity in individual-level effects. More detailed socioeconomic information would allow future research to clarify the underlying mechanisms and reasons that vulnerability to certain stressors might change over the course of pregnancy.

Abbreviations

LBW

low birth weight

PTB

preterm birth

SGA

small for gestational age