Effects of US state paid family leave policies on perinatal and postpartum health: A quasi-experimental analysis

Abstract

The US is the only high-income country without a national paid family leave (PFL) policy, although several states have implemented policies recently. This study evaluated whether PFL policies in six states improved maternal and infant health. We used difference-in-differences, a quasi-experimental approach, to estimate the impact of state-level policy implementation. We leveraged recently developed methods designed to account for staggered policy implementation and treatment effect heterogeneity. Data were drawn from the Pregnancy Risk Assessment Monitoring System 2004–2021 waves. Primary outcomes included breastfeeding, maternal postpartum depressive symptoms, and attendance at a postpartum check-up; secondary outcomes included birth outcomes. Multivariable regressions were adjusted for possible confounders. PFL policies led to increased breastfeeding duration (0.53 weeks; 95% CI: 0.06 to 0.99) and decreased depressive symptoms (−0.93 percentage points; 95% CI: −1.84 to −0.01). Policies were also associated with worsening of some birth outcomes, possibly reflecting selection in utero, data limitations, or true negative effects. Estimates were largely robust to alternative specifications, with subgroup differences by race/ethnicity and income. This study adds important evidence on the health effects of state-level PFL policies at a critical point when many states are considering or enacting policies, and during ongoing conversations about national PFL policy implementation.

Introduction

Despite evidence on the importance of paid family leave (PFL) for parental and child health,^1,2 the US remains the only high-income country without a national PFL policy.³ The Family Medical Leave Act (FMLA) provides 12 weeks of job-protected unpaid leave, although nearly half of workers are ineligible, and there is no federal mandate for paid leave.⁴ While some employers offer PFL, many Americans lack access to paid leave after birth or adoption of a child, with racial/ethnic and social inequities in access to and use of PFL.⁵ Some states have addressed this gap by enacting their own policies. State-level variation in PFL adoption provides an opportunity to examine the policies’ effects.

PFL may influence parent and infant health through several pathways. PFL improves financial security and parent labor market outcomes,^6,7 which are important for promoting parent health since financial stress and job insecurity negatively impact health.^8,9 PFL also increases time available for obtaining mental health support, alleviating sleep disruption, and bonding with infants, which can impact parent mental health and postpartum depression.⁶ Postpartum depression impacts a substantial proportion of new parents and influences not only parents’ health but also child health and development.^10,11 Increased time and resources may also facilitate postpartum check-ups, critical for child vaccination and maintaining health postpartum and beyond, although many parents do not attend postpartum care with disparities in access.^12,13 PFL also influences ability to breastfeed,^14–16 which has been linked with improved child health and development.¹⁷ Finally, PFL could influence anticipatory financial and psychological stress during the prenatal period, which could influence birth outcomes, although this connection is less clear.¹⁸

California and New Jersey were the first states to implement PFL (2004 and 2009, respectively). Studies have found these policies increased breastfeeding^14–16 and improved parental psychological distress and health behaviors,^19–21 child health,^13,22–25 and post-neonatal mortality rates.^26,27 In recent years additional states have followed suit: as of early 2024, 13 states and Washington DC have implemented mandatory PFL policies.²⁸ Evidence examining effects of PFL implementation in recently adopting states is needed to enhance generalizability and inform policymaking.^29,30

Recent literature has highlighted methodological challenges in quasi-experimental evaluations of staggered policy adoption or heterogeneous treatment effects. Existing methods such as traditional difference-in-differences (DiD) may result in biased estimates,^31,32 which newer methods seek to address.^33,34 No studies to our knowledge have examined the effect of multiple state-level PFL policies using quasi-experimental approaches accounting for staggered policy implementation.

This study sought to address existing gaps by examining the effects of PFL policies in multiple US states on perinatal and postpartum health, using a large population-based dataset. The analysis incorporates recently developed applications of DiD that account for staggered policies with heterogeneous treatment effects,³³ novel quasi-experimental methods that are increasingly a core tool for social epidemiology.^35–37 This study also provides timely insights to inform ongoing state and federal policymaking on PFL.^38–40

Methods

Study sample

We analyzed data from the Pregnancy Risk Assessment Monitoring System (PRAMS), an annual surveillance system conducted by the US Centers for Disease Control and Prevention (CDC) and state and local health departments. PRAMS is repeated cross-sectional, surveying parents 2–6 months post-birth and linking with individual-level birth certificate measures. Full methodology is described elsewhere,⁴¹ and details are in Appendix S1.

Our sample included individuals from 2004–2021 PRAMS waves (n=715,718). We restricted to singleton births gestational age 20–44 weeks (Appendix Figure S1). PRAMS captures household income brackets; we excluded the lowest bracket to exclude parents with zero income who were ineligible for PFL. We excluded those missing data on all outcomes. Our main analysis restricted to non-missing covariates. In Callaway-Sant’Anna analysis (below), we used an annual timescale; if PFL was activated mid-year we excluded the remainder of the year to avoid misclassification (final n=450,626).

The PRAMS survey is periodically revised with core survey changes. Our main analysis included Phases 5–8 (2004–2021). Questions on postpartum depressive symptoms and postpartum check-ups were excluded or optional during Phases 5–6. Therefore, we restricted these outcomes to Phases 7–8 (2012–2021), resulting in the exclusion of New Jersey, the only PRAMS state with PFL pre-2012 (final n for these outcomes=239,373). California was the earliest PFL adopter but does not contribute PRAMS data.

Measures

Exposure

The primary exposure was whether the birth occurred in a state with an active PFL policy at time of birth. Policies vary widely across states including in duration (ranging from 4–12 weeks at the end of 2021) and maximum weekly benefit (ranging from $667 to $1,206 at the end of 2021). Appendix Table S1 provides a summary of PFL policies and dates. Among states participating in PRAMS, six (New Jersey, Rhode Island, New York, Washington, Massachusetts, and Washington DC) had PFL policies at the end of the study period in 2021. While New Jersey, Rhode Island, and New York contribute several years of “exposed time” post-implementation, Washington, Massachusetts, and Washington DC contribute only one or two years of exposed time (Appendix Figure S2). All other PRAMS states (n=41) were included in the control group.

Outcomes

The primary outcomes (self-reported) were breastfeeding, maternal postpartum depressive symptoms, and postpartum check-up attendance. We included two breastfeeding measures: whether the infant was ever breastfed (including pumped breastmilk), and number of weeks breastfed (either exclusively or non-exclusively). For those still breastfeeding at the time of the survey, we used the survey date (generally 2–6 months after birth) to calculate a lower bound on breastfeeding duration. Postpartum check-up attendance was captured dichotomously. Depressive symptoms were measured using questions adapted from the two-item Patient Health Questionnaire, a standard validated screener for depressive symptoms.⁴² Details are in the Appendix S1.

As secondary outcomes, we included birth outcomes based on the hypothesis that knowledge of future access to state PFL benefits may reduce prenatal maternal stress, which could influence birth outcomes.⁴³ Birth outcomes, identified from linked birth certificate data, included six dichotomous variables: preterm birth (<37 weeks), low birthweight (LBW, <2500g), very LBW (VLBW, <1500g), small-for-gestational-age (SGA, <10^th percentile), large-for-gestational-age (LGA, >90^th percentile), and appropriate-for-gestational-age (AGA, 10^th–90^th percentile). PRAMS does not provide continuous measures for these outcomes.

Covariates

Covariates were selected a priori based on hypothesized relationships between the exposure and outcomes (Appendix Figure S3). Individual-level covariates (self-reported) included maternal age, education, race/ethnicity, marital status, household income, and family size in the year before the child’s birth. Race and ethnicity were grouped into non-Hispanic White, non-Hispanic Black, Hispanic/Latina, and other. The latter category is heterogeneous but could not be further disaggregated due to small cell sizes. Household income was inflation-adjusted, harmonized across survey years and state modules, and dichotomized (>$50,000 versus ≤$50,000; details in Appendix S1). We included indicator variables (i.e., fixed effects) for birth year to account for underlying temporal (i.e., secular) trends and state fixed effects to account for time-invariant state factors.

Statistical analysis

Primary Analysis

We summarized sample characteristics separately for individuals in treated states (with PFL policies by the end of the study period) and control states (without PFL policies during the study period).

To estimate policy effects, we used DiD analysis, a commonly used approach to evaluate policies while accounting for secular trends.^44,45 Specifically, DiD compares pre-post changes in treated states with pre-post changes in control states. Our primary analysis leveraged the Callaway-Sant’Anna (CS) DiD approach, designed to address bias in generalized DiD study designs with staggered entry of units into treatment (in this case, state PFL policy adoption).^33,34 Recent literature has highlighted pitfalls of using generalized DiD two-way fixed-effects estimates for staggered policy adoption, because already-treated states act as a contaminated control group.^31,46 The CS DiD estimator addresses this bias by separately estimating treatment effects for each policy activation date, ensuring already-treated groups are never used as a control group. Moreover, we use a doubly-robust CS DiD estimation approach that includes inverse probability weighting of covariates as well as covariates in the outcome regression.⁴⁷ To explore changes in policy effects over time, we used CS event-study estimation to assess outcome trends relative to year of policy activation. The number of states present in each event-study period varies due to staggered adoption (Appendix Figure S2); our main event-study plots display post-periods where at least two treatment states are present. We used the csdid command in Stata.⁴⁸ Details are in Appendix S1.

As is standard for DiD analysis, and given differences in interpretations of interaction terms in non-linear models, we used multivariable linear regression for both continuous and binary outcomes.^49,50 For binary outcomes, coefficients from linear probability models represent the percentage-point change in risk. We adjusted for covariates listed above and clustered standard errors by state to account for correlation of observations within state.⁵¹ We did not include survey weights in our main analysis since weighting is less appropriate for our goal of estimating causal effects rather than population-level statistics, and since we adjusted for variables related to the sampling approach.^52,53

Analysis was performed in Stata 17, and code review was performed by a second analyst.⁵⁴ The analysis plan and statistical code were drafted prior to initiating the analysis. Ethical approval was provided by the Institutional Review Board at the first author’s institution (protocol #18-26719).

DiD assumptions

A key assumption of DiD is that the trends in outcomes among the treated and control groups would have remained parallel if the intervention had not happened. While this counterfactual scenario cannot be directly tested, we followed the typical approach for exploring this assumption using event-study plots to visually inspect parallel trends during the pre-treatment period. DiD also assumes there are no differential compositional changes between the treated and control groups over time. We examined this assumption by assessing whether pre-post differences in observed covariates were similar in treated and control groups. Details are in Appendix S1.

Subgroup analyses

In addition to the aggregated DiD estimate, we examined cohort-specific estimates of PFL policy effects, where a cohort corresponds to all states that implemented PFL within the same year.

PFL policies may disproportionately benefit advantaged parents; the leave often covers only a portion of salaries, and lower-income or unmarried parents with a single income may not be able to afford time off without full pay.¹⁴ Black and Hispanic parents have less access to PFL compared to White and Asian parents—in part due to structural racism and discrimination in employment—and may therefore benefit less.^14,55 While some prior research found married, White, and high-income parents benefited more from California’s PFL policy,¹⁴ other research found historically disadvantaged women benefited more from California’s PFL policy.¹⁶ Therefore, we examined heterogeneity in PFL effects by conducting subgroup analyses by race/ethnicity, household income, and marital status. We conducted stratified analyses using the CS estimator; this estimator does not allow interaction terms. Given potential concerns for overadjustment bias due to adjusting for covariates correlated with the stratifying factor, and the potential for adjusting for mediators between the stratifying factor and outcome, we also included a minimally adjusted stratified analysis adjusting for maternal age and fixed effects for year and state.

Sensitivity analyses

The COVID-19 pandemic and related impacts on work arrangements may have modified the effect of PFL policies on our outcomes. For example, shelter-in-place orders and workplace closures in early 2020 were associated with increased breastfeeding duration.⁵⁶ To explore whether results were influenced by pandemic-era changes, we performed a sensitivity analysis excluding 2020–2021, when Washington DC and Massachusetts implemented PFL.

We additionally performed analyses including PRAMS sample weights to compare weighted and unweighted estimates. Additional robustness checks are detailed in Appendix S1, including comparison to generalized DiD two-way fixed effects, multiple imputation to address missingness, control for state-level covariates, alternative control group specifications, inclusion of the lowest income group, restriction to earlier adopting states, and investigation of non-response and time between birth and PRAMS survey.

Results

Sample Characteristics

The final sample included 83,955 people in states with PFL policies during the study period and 366,671 people in states without PFL policies during the study period (Table 1). Compared to people in non-PFL states, people in PFL states were older, more likely to have finished college, more likely to be Black, Hispanic/Latina, or other race/ethnicity, more likely to have a family income over $50,000, and more likely to have a smaller family size. People in PFL states also reported breastfeeding longer, higher likelihood of ever breastfeeding, and lower likelihood of postpartum depressive symptoms, and had lower likelihood of LBW, very LBW, preterm birth, and SGA births, and higher likelihood of AGA births.

Table 1.

Sample characteristics by state paid family leave policy status

	States with PFLa (n=83,955)	States with no PFLb (n=366,671)
	% or Mean (SD)	% or Mean (SD)
A. Sample characteristics
Baby born under PFL policy (%)	25.9	0.0
Respondent’s age (years)
<25	14.1	21.7
25–34	60.9	60.1
35+	25.0	18.1
Respondent’s education
Less than high school	7.6	8.6
High school	18.5	22.8
Some college	25.4	30.0
College +	48.5	38.6
Respondent’s race/ethnicity
White	45.6	60.5
Black	14.3	12.6
Hispanic/Latina	18.3	12.8
Other	21.7	14.0
Married (%)	72.5	72.3
Prenatal household income >$50,000 (%)	65.9	59.9
Family size
1	6.9	5.7
2	37.5	34.2
3	31.5	30.9
4+	24.1	29.3
B. Primary outcomes
Weeks breastfed	11.5 (7.6)	10.8 (7.6)
Ever breastfed (%)	88.8	87.0
Postpartum check-up (%)	92.8	92.2
Postpartum depressive symptomsc (%)	11.7	12.3
C. Secondary outcomes
Very low birthweight (%)	2.9	4.4
Low birthweight (%)	16.6	21.5
Preterm birth (%)	14.0	17.9
Small for gestational age (%)	13.1	14.7
Appropriate for gestational age (%)	77.1	75.3
Large for gestational age (%)	9.8	10.0

^aStates with PFL policy benefits active during study period (New Jersey, Rhode Island, New York, Washington, Washington DC, Massachusetts)

^bStates without PFL policy benefits active during study period (n=41 states)

^cRespondents answered two questions derived from the Patient Health Questionnaire to assess depressive symptoms. Similar to prior literature, we used a scoring system where a response of “always” or “often” to either question was classified as at-risk for postpartum depression.

Notes: Data drawn from the Pregnancy Risk Assessment Monitoring System years 2004–2021. N=450,626. Abbreviations: Paid family leave (PFL).

Model Assumptions

Examination of the parallel trends assumption was generally reassuring although more pronounced deviations from zero for postpartum check-up attendance in the pre-period highlight need for caution interpreting estimates for this outcome (Figure 1, Appendix Figure S4). There were also minimal differential compositional changes in the treated versus control states (Appendix Table S2). Details are in Appendix S1.

Figure 1.

Event study plot of the time-varying effects of paid family leave policies on breastfeeding and maternal outcomes

Notes: Data drawn from the Pregnancy Risk Assessment Monitoring System years 2004–2021. N=450,626. The reference period is T-1 (omitted from figure). Change in outcome represents percentage point change for binary outcomes. Callaway-Sant’Anna difference-in-differences models adjusted for maternal age, education, race/ethnicity, marital status, household income, family size, and fixed effects for infant’s year of birth and state. Models include robust standard errors clustered by state. Abbreviations: Percentage point change (pp). Changes in the number of treatment states present at each time point are due to differences in timing of policy implementation. These result in fluctuations in the estimates, since each estimate is the aggregation of treatment effects for treatment states present in that time period. Plots display post-periods that contain at least two treatment states; plots with full post-periods are shown in Appendix Figure S4. The presence of treatment states by years to treatment is shown in Appendix Figure S2.

Effect of PFL Policies on Perinatal and Postpartum Health

In CS DiD analyses, state-level PFL policies were associated with increased breastfeeding duration (0.53 weeks; 95% CI: 0.06 to 0.99) and decreased depressive symptoms (−0.93 percentage points [pp]; −1.84 to −0.01) (Figure 2). We observed increases in the percent ever breastfed (2.31pp; −0.65 to 5.27) and no evidence of impact on postpartum check-up attendance (0.03pp; −0.77 to 0.82). CS DiD event-study analyses allowed for more granular temporal analyses, supporting these findings (Figure 1).

Figure 2.

Effect of paid family leave policies on breastfeeding, maternal outcomes, and birth outcomes

Notes: Data drawn from the Pregnancy Risk Assessment Monitoring System years 2004–2021. N=450,626. Change in outcome represents percentage point change for binary outcomes. Callaway-Sant’Anna difference-in-differences models adjusted for maternal age, education, race/ethnicity, marital status, household income, family size, and fixed effects for infant’s year of birth and state. Models include robust standard errors clustered by state. Abbreviations: Percentage point change (pp).

For secondary outcomes, state-level PFL policies were associated with increased preterm birth (1.61pp; 0.51 to 2.71), LBW (3.46pp; 1.22 to 5.71) and SGA births (2.02pp; 0.67 to 3.38) and decreased AGA births (−2.25pp; −4.02 to −0.48). Other secondary outcomes are shown in Figure 2. Appendix Figure S4 shows the event-study plots for secondary outcomes.

Subgroup Analyses

Cohort-specific estimates showed that breastfeeding duration increased in New Jersey, Rhode Island, and New York, and decreased in Washington DC and Massachusetts. They also showed that ever breastfeeding increased in New Jersey, postpartum check-up attendance increased in Washington and decreased in Rhode Island, and postpartum depressive symptoms decreased in New York and Washington (Appendix Table S3). Estimates were somewhat noisy due to smaller sample sizes, and effect estimates for multiple outcomes in the most recently adopting states were consistent with no overall effect.

Stratified subgroup analyses suggested some effect heterogeneity for the primary outcomes (Appendix Figures S5–S6). For breastfeeding duration, PFL led to an increase in all groups except other race/ethnicity parents (−0.36 weeks; −0.95 to 0.23). For ever breastfeeding, PFL led to an increase in all groups except Hispanic/Latina parents (−1.77pp; −2.77 to −0.77). PFL led to increased postpartum check-up attendance among other race/ethnicity parents (1.64pp; 0.86 to 2.42) and those with lower household income (1.56pp; 0.33 to 2.78). For postpartum depressive symptoms, estimates were similar for most subgroups, although the decrease was more pronounced among Black parents (−1.88pp; −3.98 to 0.22) and those with higher income (−1.26pp; −1.71 to −0.80). Minimally adjusted analyses yielded similar results to the fully adjusted analysis (Appendix Figures S5–S6). Given formal statistical tests of interaction are not possible using the CS estimator, subgroup differences should be treated qualitatively.

Sensitivity Analyses

Results restricted to the pre-pandemic period were largely consistent with the main analysis, although a smaller decrease in postpartum depressive symptoms (Appendix Figure S7). Estimates from weighted analysis were broadly consistent with main results, and we also observed increased postpartum check-up attendance (0.56pp; −0.04 to 1.15) (Appendix Figure S8). Results of additional robustness checks are shown in Appendix S1 and Appendix Figures S9–S15.

Discussion

This study found that recent state-level PFL policies increased the duration of breastfeeding by 0.53 weeks (4.7% increase from baseline) and decreased postpartum depressive symptoms by 0.93pp (8.2% decrease from baseline). We also found PFL was associated with increased preterm birth, LBW, and SGA births and decreased AGA births. Our findings are in line with prior research on the first states adopting PFL policies—California and New Jersey—which found increased breastfeeding duration and decreased parental psychological distress.^{14–16,19–21} The findings on birth outcomes contribute to currently limited literature on these outcomes. Our research provides evidence on additional, more recent state policies, and also contributes insights using recent methodological developments to address potential bias in generalized DiD approaches. This study adds to a growing body of research in social epidemiology examining the health effects of state policies that address income and other social determinants of health.^57–60

Previous research examining PFL implementation across multiple states has not yet utilized recent approaches accounting for staggered policy implementation; prior research was therefore subject to potential bias highlighted in recent methodological literature.³³ The analytic approach leveraged in the current study via the Callaway-Sant’Anna estimator addresses this concern by separately estimating valid comparisons using doubly robust methods and performing weighted aggregation of unit-level effect estimates.³³ The current research adds rigor to existing estimates in the field. This approach also allowed examination of effect heterogeneity by state cohort. The largest effect estimates were mostly observed for New Jersey and Rhode Island, likely because they were earlier PFL adopters. Although cohort differences were limited due to smaller sample sizes and very recently adopting states contributed less substantially, overall estimates can be thought of as an aggregation or meta-analysis of multiple cohort-specific estimates. To achieve the most complete estimates in settings with staggered policy adoption, it is important to aggregate results from all available states rather than including only earlier-adopting states or states with significant effects, as well as to revisit analysis as more years of data become available.

Notably, three PFL states in our sample (New Jersey, New York, and Rhode Island) offer Temporary Disability Insurance (TDI) that can be used immediately before and/or after birth—sometimes supplementing or replacing PFL;⁶¹ this may influence use of PFL or modify its effect. However, given these states have offered TDI for pregnancy since the 1978 Pregnancy Discrimination Act, our study design should isolate the effect of changes in PFL independent of ongoing presence of TDI as the inclusion of state fixed effects account for state-specific time-invariant confounders.⁶²

Our finding that PFL increased breastfeeding duration is consistent with prior studies of early-adopting states.^14–16 PFL policies may enable parents to continue breastfeeding longer instead of returning more immediately to work. Recent research on workplace closures during the COVID-19 pandemic similarly found increased breastfeeding duration—but not likelihood of initiating breastfeeding.⁵⁶ Breastfeeding is widely regarded to have benefits for both parents and children, and the American Academy of Pediatrics recommends exclusive breastfeeding for six months.⁶³ Therefore, PFL policies represent an important strategy for improving parent and child health by enabling increased breastfeeding. Future research should examine other interventions to increase breastfeeding initiation.

Our finding that PFL decreased depressive symptoms in the postpartum period is consistent with prior research finding PFL policies in early-adopting states to be associated with improved parental mental health more generally.^19–21 Our finding is also consistent with recent investigation in the two PFL states (New Jersey and New York) where PRAMS captures self-reported paid or unpaid leave usage; this study found lower postpartum depression among those who took paid compared to unpaid leave, although the analysis was not quasi-experimental.⁶ Potential mechanisms include decreased parental stress, reduced financial stress and time pressure, increased access to healthcare, and increased time for bonding and self-care. Postpartum depression has an estimated prevalence of 13–19%, and also impacts child health and development;^10,11 therefore, PFL policies improving postpartum depressive symptoms may represent an important strategy for improving parent and child health at a population level.

We did not see evidence of a substantial impact on postpartum check-up attendance overall; however, we did find an increase in postpartum check-up attendance among lower-income people and other race/ethnicity parents. There is inconsistent evidence from other studies on PFL impacts on postpartum healthcare utilization. One study found no effect on children’s visits to a health professional in the past year.²² Another found a decrease in late child vaccinations,¹³ and a third found no overall effect on the likelihood of children being up-to-date on vaccinations but a potential improvement among more disadvantaged families.¹⁶ Another correlational study found that state PFL generosity was associated with fewer depressive symptoms and greater postpartum visit attendance.⁶⁴ Additional time and resources provided by PFL may be more salient for enabling families with lower income to access postpartum care.

For secondary outcomes, this study found PFL policies were associated with increased preterm birth, LBW, and SGA births and decreased AGA births. These counterintuitive findings are potentially consistent with the “selection in utero” hypothesis, such that improved socioeconomic conditions lead to fewer miscarriages and fetal losses, and consequently delivery of at-risk infants who have worse outcomes at birth.⁶⁵ In this case, the apparent adverse birth outcomes would not necessarily be a “negative” impact of the policy as the hypothesis would suggest the policy may have reduced miscarriages. Alternatively, these findings could result from residual confounding, e.g., if states with PFL policies experienced greater increases in maternal age compared to non-PFL states over time. Increased maternal age is linked with higher preterm birth,^66–68 and we observed slight compositional changes by maternal age in Appendix Table S2. Our measure of maternal age was restricted to the three categories available in PRAMS, which could fail to fully adjust for the potential confounding impacts of recent changes in older age fertility that are linked to birth outcomes. Conversely, it may be that the design of PFL policies—in that most provide less-than-full replacement of wages—may paradoxically increase financial stress, adversely impacting downstream birth outcomes. If this were the case, the negative impact on birth outcomes would likely outweigh the positive impacts of PFL on breastfeeding and postpartum depressive symptoms. More research is needed to investigate whether PFL impacts pregnancy trajectories and the proportion of pregnancies that end in miscarriage, and to adjust more granularly for maternal age. Linkages with administrative data on leave-taking could also help reduce measurement error in the exposure. In contrast to our findings, one evaluation of PFL implementation in California²⁶ and an evaluation of San Francisco’s paid parental leave ordinance found no effects on birth outcomes.⁶⁹ Other research found decreased LBW following implementation of TDI, perhaps because TDI enables leave immediately prior to birth, which could more directly impact birth outcomes.⁶² Research on FMLA found no effect on birth outcomes overall but among college-educated and married mothers found increased birth weight and decreased premature birth.⁴³ Two recent studies found reduced post-neonatal infant mortality following PFL implementation in California.^26,27 Finally, while parallel trends appeared reassuring for these outcomes, other co-occurring policies may have confounded the relationship between the exposure and these outcomes; this is a limitation of all DiD analyses.

This study has several strengths. Use of a large national dataset specifically designed to study the perinatal and postpartum period and linkage to birth outcomes enabled us to examine multiple perinatal and postpartum outcomes in a relatively large sample. We used a rigorous quasi-experimental approach designed to account for staggered policy intervention. We also assessed potential policy heterogeneity by key subgroups to provide insight on how the policy may differentially impact at-risk populations.

This study also has limitations. Most states in PRAMS do not collect self-reported use of PFL. Our analysis therefore identifies the intent-to-treat effects of the policy on population outcomes rather than the impact of individual receipt and likely underestimates the effect among those who were able to take advantage of it. Given small sample sizes, we were not able to examine the impact of differing aspects of PFL generosity offered by states (e.g., duration, wage replacement rates, job protection, eligibility restrictions). While we have highlighted potential for a specific form of residual confounding related to maternal age in our secondary outcomes, other forms of residual confounding could also influence our primary outcomes. We are limited by having few states with long post-period data available; future work revisiting this analysis when more time has passed will be valuable. Finally, our main outcomes and covariates were self-reported and subject to standard reporting biases.

In conclusion, this study contributes evidence on the health impacts of PFL by evaluating more recent PFL policy changes using a robust quasi-experimental approach. The findings from this research advance substantive and methodological knowledge in social epidemiology and are critical to ongoing policymaking regarding PFL policies in the US.

Data Availability Statement:

The data described in the manuscript and a codebook are available upon request from the Pregnancy Risk Assessment Monitoring System (https://www.cdc.gov/prams/php/data-research/index.html). The analytic code is available upon request from the authors.