Impact of minimum wage increases on smoking before, during and after pregnancy in the USA: a difference-in-differences analysis of cross-sectional surveillance data

Abstract

Introduction

Smoking before and during pregnancy has long-lasting adverse effects for children’s health and development. Financial stress is an important risk factor for maternal smoking, but there is little evidence on how economic policies affect tobacco use. We examined whether increases in state minimum wages had an impact on maternal smoking and cessation.

Methods

This was a secondary analysis of cross-sectional surveillance data from 46 states between 2004 and 2019. Our analysis sample included women without college education. Outcomes were use of cigarettes in the 3 months before pregnancy, the last 3 months of pregnancy, and in the postpartum period, as well as cessation of smoking during and after pregnancy. For the main analysis, a two-way fixed effects difference-in-differences analysis was used to estimate the effects of state-level minimum wage policies on outcomes, adjusting for individual covariates (age, race and ethnicity, marital status, parity, education and month of birth), state-level economic and tobacco policies, and state and year fixed effects. We also explored robustness to de Chaisemartin & D’Haultfœuille (dCDH)’s heterogeneity-robust alternative estimator. Subgroup analysis examined heterogeneous impacts by race and ethnicity, age, marital status and pregnancy intention. We tested differences by subgroup using interaction models and testing the joint null hypothesis of coefficients. We used the Benjamini-Hochberg procedure to adjust p values for multiple comparisons.

Results

The study analysed 421 884 women of whom 28% smoked in the 3 months prior to pregnancy. A 10% increase in the minimum wage was associated with a 0.75 (95% CI −1.32 to −0.19) percentage point decrease in the probability of smoking before pregnancy. Associations were strongest for women who were American Indian/Alaska Native, non-Hispanic White, married and had an unintended pregnancy. Two-way fixed effect results were consistent when using the dCDH estimator. There were no significant associations between minimum wages and smoking during pregnancy or postpartum, or on smoking cessation. The conclusions of the analysis do not change when adjusted for multiple hypothesis testing.

Conclusions

Estimates suggest that an increase from the current federal minimum wage of US$7.25–US$14 would be associated with a 17% decrease in smoking prior to pregnancy in a demographic group that is disproportionately affected by poor health outcomes.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Previous studies have found mixed associations between minimum wages and smoking in the general population. No studies have examined how minimum wage changes affect maternal smoking before and after pregnancy, and no studies have included measures of electronic cigarette use.

WHAT THIS STUDY ADDS

• Using a quasi-experimental design with a difference-in-differences approach that accounts for staggered adoption of policies across states, we found that increases in state minimum wages were associated with reductions in smoking prior to pregnancy among women without a college degree but not with prenatal or postpartum smoking. The associations of minimum wages with smoking before pregnancy were concentrated among those with unintended pregnancy and were strongest among American Indian/Alaska Native and non-Hispanic White women.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Estimates suggest that changes to federal minimum wage policy from US$7.25 to US$14 would be associated with a 17% decrease in smoking prior to pregnancy.

Introduction

In the USA, approximately 12% of women smoked cigarettes in the months before pregnancy in 2021, 5% reported smoking during pregnancy, and 7% smoked during the postpartum period.¹ Women with unplanned pregnancies are 30% more likely to smoke prior to pregnancy and 68% more likely to smoke during pregnancy than those with planned pregnancies.² Smoking before and during pregnancy, even at low levels, increases risk for adverse birth outcomes including preterm birth, low birth weight and perinatal death.^{3 4} Smoking during and after pregnancy also has significant detrimental effects for children across the life course, including increased risk of asthma, respiratory tract infections, cognitive impairment and obesity.³ Since their introduction in 2006, use of electronic cigarettes (ECs) increased to about 4% among pregnant women and to 29% among women of reproductive age by 2017.^{5 6} The evidence on EC use during pregnancy suggests similar associations with birth outcomes as conventional cigarettes.⁷ Cessation of tobacco use prior to pregnancy has large benefits; smokers who quit before pregnancy have a risk of preterm birth similar to that of nonsmokers.⁴ As a result, any interventions that reduce the prevalence of maternal smoking and EC use before and during pregnancy would produce substantial improvements in birth outcomes.⁸

Socioeconomic disadvantage is a strong correlate of maternal smoking.⁹ In 2021, nearly 20% of women with high school education or less reported smoking in the 3 months before pregnancy, compared with only 8% of women with more than a high school education.¹ Material hardship and financial stress (such as low income, poor housing or difficulty making ends meet) are strongly associated with smoking and barriers to quitting.^{10 11} Socioeconomic disadvantage is also associated with increased risk of smoking during pregnancy, with even small differences in resources (1 additional year of education or US$15 000 in annual income) distinguishing smokers from quitters and nonsmokers.¹²

From a public health perspective, policies that increase income and decrease financial stress may reduce smoking among women experiencing socioeconomic disadvantage.¹³ Raising the minimum wage has increasingly gained political interest as a means of increasing wages of lower-income earners.¹⁴ Raising the minimum wage disproportionately benefits workers without college education, particularly women and those from racial and ethnic minority groups.^{14 15} In 2024, 24 states increased their minimum wages, with 15 states setting minimum wages that reached or exceeded US$14 per hour, while 20 states continued to use the federal minimum wage of $7.25, creating large geographic disparities in the income of minimum wage earners.¹⁴

Economists have long studied minimum wages with mixed findings on employment and poverty; however, the most recent evidence finds that raising the minimum wage results in meaningful reductions in poverty and food insufficiency.15,18 Evidence also indicates minimum wage policies are associated with reductions in experiences of stressful life events among pregnant women and with lower rates of mental distress among adults.19,21 Minimum wage increases may therefore affect individual smoking behaviour through reductions in financial stress and improved mental health. Alternatively, higher wages may make cigarettes more affordable, increasing their use.¹⁸

A growing literature has found that minimum wages are associated with improved population health, including improved maternal, infant and child health.1519 22,24 However, previous studies have found mixed associations between minimum wages and smoking in the general population.1825,28 Importantly, behavioural responses to minimum wage increases among women in the perinatal period may be different from those for women in the general population, as there may be more motivation to quit smoking before and during pregnancy.²⁹ Studies have found that most pregnant smokers are aware of the harmful effects of smoking on infant health and desire to stop smoking.³⁰ However, scant research has examined the effects of minimum wages on maternal smoking.²² No studies, to our knowledge, have examined impacts of minimum wage increases on maternal EC use, which is vital given the growing rates of EC use (and possible substitution from cigarettes to ECs as a ‘safer alternative’), particularly among younger individuals.³⁰

In this study, we used quasi-experimental methods to investigate the effects of minimum wages on maternal smoking using US population-based surveillance data merged with US birth records from 2004 to 2019. In addition to two-way fixed effects models that have been traditionally used to implement difference-in-differences analysis, our approach uses recent methodological developments that account for variation in the timing of policy adoption across states. We examined impacts on both smoking prevalence and cessation behaviour in the periods before, during and after pregnancy and included information on EC use in our measures. We also explored heterogeneous impacts by demographic characteristics to better understand who benefits most from minimum wage changes.

Materials and methods

Data

We used data from the Centers for Disease Control and Prevention (CDC)’s Pregnancy Risk Assessment Monitoring System (PRAMS), a national surveillance system that represents about 80% of all US births, and which has been used to study socioeconomic determinants of maternal health.31,34 For each state and year in the PRAMS data, birth records are used to select a representative sample of women who delivered a live-born infant. A mail survey is initially sent to this sample; if there is no response, a follow-up telephone interview is conducted. We use the term women to maintain consistency with PRAMS, which does not ask questions on gender identity. This study followed Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines for cross-sectional studies. The datasets used in this study are publicly available.³⁵

States opt in to participation in PRAMS, and in addition must meet the response rate threshold for data to be available for researchers; this results in a changing composition of states across years. Consistent with prior research,^{19 36} in our main analysis, our eligibility criteria were all states with at least 2 years of data over the study period from 2004 to 2019 (46 states including DC, n=614 736 participants) and within those states, individuals with some college education or less (n=421 844), as this group is most likely to be affected by minimum wage changes.¹⁵ In robustness checks, we alter our selection of states, restricting the sample to states with fewer than 3 years of missing data over the study period (N=245 430 in 17 states).

Outcomes

Our primary outcomes of interest were maternal smoking in the 3 months before pregnancy, in the last 3 months of the pregnancy and postpartum (assessed at the time of questionnaire completion, with 90% of participants responding between 2.5 and 6.5 months after delivery). The three questions used to estimate these measures were as follows: (1) ‘In the 3 months before you got pregnant, how many cigarettes did you smoke on an average day?’; (2) ‘In the last 3 months of your pregnancy, how many cigarettes did you smoke on an average day?’ and (3) ‘How many cigarettes do you smoke on an average day now?’ If a respondent reported any cigarette use during the specified period, they were categorised as smoking during that period. For the years 2016–2019, additional questions were asked about how often respondents used EC products in the 3 months before and the last 3 months of pregnancy, with responses ranging from 1 day a week or less to more than once a day. We created two additional outcome variables indicating a respondent either smoked cigarettes or used ECs in those time periods. Across our primary outcomes, <2% of individuals had missing data; therefore, we used complete case analysis.

As secondary outcomes, we examined smoking cessation, including quitting smoking during pregnancy (defined as cigarette use in the 3 months before pregnancy, but not during the last 3 months of pregnancy), quitting between prepregnancy and postpartum (defined as use in the 3 months before pregnancy but not postpartum), and quitting between pregnancy and postpartum (defined as cigarette use in the last 3 months of pregnancy but not postpartum). Finally, we examined relapse in the postpartum period, defined as having quit smoking during pregnancy but reported smoking postpartum. Because EC questions were not asked about for the postpartum period, we limited cessation and relapse variables to tobacco products only.

Exposure

Yearly data on state minimum wages were obtained from the University of Kentucky Center for Poverty Research (UKCPR), which sources these data from the US Bureau of Labor Statistics.³⁷ We used the logged real state minimum wage on 1 January of each year, inflated to 2020 dollars using the annual average Consumer Price Index.^{15 20} In states without minimum wage laws (five states) or states with rates below the federal minimum wage (two states), the minimum wage was set to the federal rate, which legally applies. Consistent with previous literature on minimum wages and smoking,^{25 27} we measured the exposure in the 2 years prior to the measurement of outcome. For outcomes measured at 3 months before pregnancy (1 year before delivery), the exposure was defined as the average of the 1-year and 2-year lagged minimum wage prior to the month and year of delivery, while for those measured during and after pregnancy, the exposure was defined as the average of the current and 1-year lagged minimum wage prior to delivery.

Covariates

Covariates were chosen based on prior literature on minimum wages and population health, as well as literature on predictors of maternal smoking.^{1 9 20 25 27} We included individual characteristics predictive of smoking behaviour as well as state-level tobacco and Medicaid policies and economic conditions that could confound the relationship between minimum wage policy changes and smoking behaviour. Individual characteristics included demographic and socioeconomic variables (mother’s age, race and ethnicity, marital status, parity, education), and the child’s month of birth (due to seasonality of smoking behaviour³⁸). To classify race and ethnicity, we used variables provided in PRAMS that are derived from the birth certificate and are self-reported. The final categories included non-Hispanic American Indian/Alaska Native (AI/AN), non-Hispanic Asian/Pacific Islander, Hispanic, non-Hispanic Black, non-Hispanic White and other (which included subcategories of non-Hispanic other race, mixed race and missing race, categorised together due to small sample sizes). We included race and ethnicity variables because individuals from non-White groups are disproportionately likely to be earning minimum wages and because of the large and systemic racial disparities in maternal health outcomes.³⁹ Individuals with missing covariate information were coded using a separate missing category indicator or as part of a larger category (see online supplemental table 1).

State-level time-varying characteristics were lagged 2 years from the year of delivery. Based on prior literature,^{20 25 27} these included policies that could affect smoking rates including state cigarette excise tax rates and three variables on smoking bans in bars, restaurants and private workplaces, obtained from the CDC.^{40 41} We also included Medicaid eligibility levels for pregnant women and parents, obtained from the Kaiser Family Foundation.⁴² In addition, we controlled for state-level economic conditions that may be correlated with smoking behaviour, obtained from UKCPR: including the unemployment rate, gross state product, per cent of children uninsured, poverty rate, state Earned Income Tax Credit (EITC) rate (a tax refund for low-income to moderate-income working families) and combined monthly maximum for Temporary Assistance for Needy Families (TANF) and Supplemental Nutrition Assistance Program (SNAP) benefits for a two-person family (public programmes that provide cash and food assistance to lower-income families). In robustness checks, we removed controls for state-level unemployment and poverty rates. State-level variables had no missing data.

Statistical analysis

In a generalised difference-in-differences (DID) framework, we used two-way fixed effect (TWFE) models to estimate the associations between state-level minimum wage (treatment) and outcomes. This TWFE DID approach compares groups experiencing different evolutions of their exposure to treatment over time.⁴³ We estimated the associations between treatment and outcomes, adjusting for individual covariates, state-specific time-varying factors, and state and time fixed effects. We did not include state-specific linear time trends in our main analysis as recent work has cautioned that the use of unit-specific time trends may not always be appropriate²⁰; however, for completeness, we show results including these in robustness checks. To account for within-state correlation, standard errors (SEs) were clustered at the state level.⁴⁴ We then used results from these models to estimate linear predicted values of smoking prevalence at the value of the current federal minimum wage (US$7.25) and the maximum state minimum wage observed in our sample (US$14), and tested the difference using contrasts of marginal linear predictions. As our continuous models were based on the log of the minimum wage, our hypothesis tests were also based on this log transformation. Next, we stratified models by race and ethnicity (as defined above), age (<20, 20–29, ≥30), marital status (married vs not married), parity (nulliparous vs parous) and pregnancy intention (wanted pregnancy then or sooner vs wanted pregnancy later, not at all, or was not sure) to examine if minimum wage laws have differential impacts across demographic groups. We tested differences by subgroup with models interacting minimum wage with each subgroup and used a Wald test to jointly test the interaction coefficients.

While DID models have been widely implemented using a TWFE approach, recent research has shown that they can be biased in contexts in which the timing of the treatment varies,⁴³ as applies with minimum wage policies. To address this concern, we augmented our analysis with the de Chaisemartin & D’Haultfœuille (dCDH) heterogeneity-robust DID estimator.^{43 45} Following previous literature, we operationalised minimum wage increases as absorbing binary events defined by large minimum wage changes. We considered event sizes of US$1 or more year-to-year increases in the state nominal minimum wage as well as US$0.75 or more increases. Because our event study analysis of dCDH estimates shows the parallel trends assumption is more likely to hold when defining events by US$0.75 or more, these results are presented in the main text. However, results for US$1 or more year-to-year increases are similar and are shown in the online supplemental technical appendix. We focused on the outcome of smoking before pregnancy as our TWFE models did not provide consistent evidence that other outcomes were impacted. Complete specifications are provided in the online supplemental section 1.

The pattern of missingness for all individual covariates and outcomes is shown in online supplemental table 1; less than 2% of data were missing across all variables. All analyses were weighted using survey weights provided by the CDC to account for the complex design. All statistical tests were two-sided with a significance level of 0.05. Analyses were conducted using Stata V.19.

Sensitivity analyses

We conducted several sensitivity analyses of the TWFE model. We used the Benjamini-Hochberg procedure to adjust p values among each set of primary outcomes, secondary outcomes, and outcomes by subgroup.^{46 47} This procedure, which is designed to correct for elevated type I errors associated with conducting multiple hypothesis tests, produces ‘sharpened’ false discovery rate-adjusted q-values, with q-values less than 0.05 interpreted as statistically significant. To examine the robustness of model specification, we used logistic regression models, we used an alternative measure of exposure (two calendar years before delivery), we included state time trends as controls, we removed unemployment rate and poverty rate as controls, and we adjusted for within-state correlation using the Wild Cluster Bootstrap. To address the changing composition of states over time, we restricted the sample to states with fewer than 3 years of missing data over the study period (N=245 430 in 17 states). As a placebo test, we estimated models using a sample of women with at least college education, a group unlikely to be substantially affected by minimum wage changes.²² We also conducted a placebo test using the state minimum wage 2 years after delivery as the exposure, which should not affect outcomes.

Patient and public involvement

As this was a secondary data analysis of previously collected data by the CDC, participant involvement in the design, conduct, reporting or dissemination of this research was not possible.

Results

The main analysis sample size was 421 884 women with some college education or less and who gave birth in 2004–2019 in 46 states (online supplemental figure 1). In this population, 1.2% were AI/AN, 2.7% were Asian/Pacific Islander, 21.6% were Hispanic, 17.8% were non-Hispanic Black, 53.9% were non-Hispanic White, and 2.8% were non-Hispanic other race (table 1). More than one-fifth (22.5%) did not complete high school, while 38.8% had a high school diploma, and 38.7% had some college or associate’s degree. Just over half were unmarried at the time of the birth (52.3%). Over a quarter (28%) reported smoking in the 3 months before pregnancy, 15% reported smoking in the last 3 months of pregnancy, and 21% reported smoking postpartum, with higher rates of smoking in all time periods among participants who were younger than 30 years, AI/AN, non-Hispanic White, unmarried, and who did not want the pregnancy then. In the sample, 6% reported using ECs before pregnancy and 2% during pregnancy, with higher rates among younger and non-Hispanic White participants.

Table 1. Descriptive characteristics of analysis sample, 2004–2019.

	N (%)*	Fraction who smoked cigarettes			Fraction who used ECs
		3 months before pregnancy	Last 3 months of pregnancy	Postpartum	3 months before pregnancy	Last 3 months of pregnancy
Age at delivery (years)
<20	49 608 (11.0%)	0.30	0.14	0.23	0.10	0.03
20–29	246 181 (59.5%)	0.31	0.16	0.23	0.06	0.02
30–39	116 502 (27.4%)	0.22	0.12	0.16	0.04	0.02
40+	9593 (2.1%)	0.18	0.11	0.13	0.03	0.01
Race and ethnicity
AI/AN¶	20 513 (1.2%)	0.44	0.23	0.33	0.05	0.02
Asian/Pacific Islander	20 553 (2.7%)	0.13	0.05	0.08	0.02	0.00
Hispanic	78 985 (21.6%)	0.11	0.04	0.07	0.02	0.00
Non-Hispanic Black	85 517 (17.8%)	0.20	0.11	0.17	0.02	0.01
Non-Hispanic White	198 305 (53.9%)	0.38	0.21	0.28	0.09	0.03
Other†	18 011 (2.8%)	0.31	0.16	0.23	0.06	0.02
Education
Less than HS	94 830 (22.5%)	0.29	0.19	0.24	0.05	0.02
HS diploma	164 491 (38.8%)	0.31	0.17	0.24	0.06	0.02
Some college	162 563 (38.7%)	0.24	0.11	0.16	0.05	0.01
Marital status
Not married‡	224 066 (52.3%)	0.36	0.20	0.27	0.07	0.02
Married	197 818 (47.7%)	0.20	0.10	0.13	0.04	0.01
Nulliparous
No	257 133 (61.7%)	0.27	0.16	0.21	0.04	0.02
Yes	162 901 (37.9%)	0.30	0.13	0.20	0.08	0.02
Missing	1850 (0.4%)	0.25	0.12	0.16	0.07	0.03
Pregnancy intention§
Did not want pregnancy then/not sure	213 527 (51.3%)	0.34	0.19	0.25	0.07	0.02
Wanted pregnancy then	200 116 (48.7%)	0.23	0.11	0.16	0.04	0.01
Total	421 884	0.28	0.15	0.21	0.06	0.02

Notes: Sample is individuals who gave birth between 2004 and 2019 with some college education or less.

^*Statistics shown are unweighted sample sizes and survey-weighted percentages.

^†Other race and ethnicity category includes the subcategories of ‘other race’, ‘mixed race’ and missing race; see online supplemental table 1 for details.

^‡Missing marital status (0.1%) included in not married category.

^§Defined as wanted later, wanted not at all or not sure, compared with wanted then or sooner. Postpartum smoking was measured at time of survey completion (about 2.5–6.5 months after delivery). Use of EC was measured only between 2016 and 2019. Some college includes Associate’s degree.

^¶ AI/AN and Asian/Pacific Islander are non-Hispanic.

AI/AN, American Indian/Alaska Native; EC, electronic cigarette; HS, high school.

Figure 1 shows trends in outcomes (subgraph A) and minimum wages (subgraph B) across the study period. Smoking rates in the preconception (3 months before pregnancy), prenatal (last 3 months of pregnancy), and postpartum periods generally decreased over time. State minimum wages (in real 2020$) increased from a mean of US$7.68 in 2004 to US$9.37 in 2019, with minimum and maximum rates varying widely within and across years.

Figure 1. (A) Trends in prevalence of maternal smoking and (B) real minimum wage rates (2020 $) in the study sample, 2004–2019. Notes: Study samples include individuals with some college education or less who gave birth between 2004 and 2019 in 46 states including DC. Smoking postpartum is defined as smoking at time of survey completion (about 2.5–6.5 months after delivery). Smoking is defined as at least one cigarette smoked in the relevant time period. Smoking prevalence estimates are weighted with PRAMS survey weights. DC, District of Columbia; PRAMS, Pregnancy Risk Assessment Monitoring System.

Online supplemental table 2 further shows the variation in minimum wage changes by state over time. The minimum wage increased from a minimum of US$5.15 in 2004 (20 states) to a maximum of US$14 in 2019 (in Washington, DC). On average over the study period, the real minimum wage increased from US$6.33 to US$9.04. Minimum wage events of US$0.75 or greater occurred across the study period, ranging from 2005 to 2019, with 15 states experiencing their first event before 2010 and 11 states after 2010.

Table 2 shows the results of the TWFE analysis, with models fully adjusted for individual-level and state-level factors. A 10% increase in the minimum wage was associated with a 0.75 percentage point (95% CI –1.32 to −0.19 percentage points) decrease in the probability of smoking in the last 3 months before pregnancy (table 2). The full set of regression results is shown in online supplemental table 3. These estimates suggest that an increase from the current federal minimum wage of US$7.25 to US$14, the maximum state wage observed in our sample, would be associated with a reduction in prevalence of smoking in the 3 months before pregnancy among women with some college education or less from 29.0% to 24.1%, a 17% decrease (online supplemental table 4). Results were similar when including use of ECs. For both outcomes, q-values adjusted for multiple comparisons were less than 0.05. There were no significant associations between minimum wages and smoking in the last 3 months of the pregnancy or in the postpartum period. There were also no significant associations between minimum wage increases and cessation of smoking or relapse among smokers.

Table 2. Estimates of a 10% increase in the minimum wage on maternal smoking.

Outcomes	Estimate	95% CI	P value	Sharpened q-value	N	Sample mean
(A) Primary outcomes
Smoked 3 months before pregnancy*	−0.75†	(−1.32 to −0.19)	0.011	0.048	414 262	0.282
Smoked/used EC 3 months before pregnancy*	−0.71†	(−1.29 to −0.13)	0.018	0.048	414 314	0.286
Smoked last 3 months of pregnancy‡	−0.08	(−0.37 to 0.21)	0.592	0.799	414 516	0.150
Smoked/used EC last 3 months of pregnancy‡	−0.04	(−0.34 to 0.25)	0.779	0.879	414 544	0.152
Smoked postpartum‡	−0.19	(−0.57 to 0.19)	0.331	0.495	414 465	0.207
(B) Secondary outcomes
Smoked before pregnancy, quit during pregnancy*	−0.39	(−0.98 to 0.21)	0.218	0.541	125 870	0.471
Smoked before pregnancy, quit postpartum*	0.40	(−0.29 to 1.10)	0.262	0.541	125 664	0.281
Smoked during pregnancy, quit postpartum‡	−0.24	(−0.65 to 0.18)	0.263	0.541	71 704	0.0651
Quit during pregnancy but relapsed postpartum‡	−0.32	(−1.60 to 0.95)	0.655	0.541	54 325	0.473

All coefficients have been multiplied by 10 to facilitate interpretation, and reflect a 10% increase in the state minimum wage. Postpartum smoking was measured at time of survey completion (about 2.5–6.5 months after delivery). Use of ECs was measured only between 2016 and 2019. All models adjusted for state and year fixed effects, maternal age, race and ethnicity, marital status, parity, education and month of child’s birth, and the following state time-varying variables lagged by two years: state cigarette excise tax rates, smoking bans in bars, restaurants and private workplaces, unemployment rate, gross state product, per cent of children uninsured, poverty rate, state Earned Income Tax Credit rate as a percentage of the federal rate, combined monthly maximum for Temporary Assistance for Needy Families and Supplemental Nutrition Assistance Program benefits for a two-person family, and Medicaid eligibility levels as per cent of the federal poverty limit for pregnant women and for parents. 95% CIs reflect clustered SEs. Sharpened q-values obtained using the Benjamini-Hochberg procedure for multiple hypothesis testing.

^*Exposure is log of average minimum wage in the 1–2 years prior to delivery.

^†p<0.05.

^‡Exposure is log of average minimum wage in the 0–1 years prior to delivery.

EC, electronic cigarette.

Figure 2 shows associations between increases in the minimum wage and the likelihood of smoking before pregnancy across subgroups. There were significant differences in associations by race and ethnicity (p=0.01), marital status (p=0.05), age (p<0.001) and pregnancy intention (p=0.01). The largest associations were found for individuals who were AI/AN (−1.38 percentage points, 95% CI −2.77 to 0.02 percentage points), non-Hispanic White (−1.24 percentage points, 95% CI −2.05 to −0.44 percentage points), married (−0.96 percentage points, 95% CI −1.62 to −0.29 percentage points) and had an unintended pregnancy (−0.94 percentage points, 95% CI −1.74 to −0.14 percentage points) (online supplemental table 5). When adjusting for multiple hypothesis testing, significant associations (q-values <0.05) were maintained for non-Hispanic White and married subgroups.

Figure 2. Association of minimum wage increases with smoking before pregnancy by demographic subgroup. Notes: Separate models stratified by each characteristic. Coefficients from regression models have been multiplied by 10 to facilitate interpretation and represent the impact of a 10% increase in the state minimum wage rate. The p value is from the joint test of the null hypothesis of each subgroup. All models adjusted for state and year fixed effects, maternal age, race and ethnicity, marital status, parity, education and month of child’s birth, and the following state time-varying variables lagged by 2 years: state cigarette excise tax rates, smoking bans in bars, restaurants and private workplaces, unemployment rate, gross state product, per cent of children uninsured, poverty rate, state Earned Income Tax Credit rate as a percentage of the federal rate, combined monthly maximum for Temporary Assistance for Needy Families and Supplemental Nutrition Assistance Program benefits for a two-person family, and Medicaid eligibility levels as per cent of the federal poverty limit for pregnant women and for parents. 95% CIs reflect clustered SEs. MW=minimum wage.

Results were robust to the full range of robustness checks we describe in the sensitivity analysis section (online supplemental table 6).

Figure 3 shows the dCDH event-study plot for a US$0.75 or more minimum wage event for the outcome of smoking before pregnancy. The preperiod shows no parallel trends (joint null test of placebos p=0.68), while estimates are negative and significant in the postperiod. The largest impact was in the first year after the treatment, with a reduction of 1.8 percentage points in the likelihood of smoking before pregnancy. The average effect over 3 years was −0.015 (95%CI −0.026 to −0.004). Average effects for all outcomes are shown in online supplemental table 7. The effect estimates for outcomes of smoking during pregnancy and smoking postpartum were negative but not significant (event plots shown in online supplemental figure 2). Overall, we find that results from TWFE and dCDH are consistent, with both indicating statistically significant reductions in smoking before pregnancy associated with minimum wage increases.

Figure 3. Event-study plot for a US$0.75 or more year-to-year minimum wage increase on outcome of smoking before pregnancy. Notes: Results obtained from de Chaisemartin & D'Haultfœuille’s DID estimator using did_multiplegt_dyn function (see online supplemental section 1). DID, difference-in-difference.

Discussion

Increases in state minimum wages were associated with reductions in smoking prior to pregnancy among women without a college degree, but not with prenatal or postpartum smoking. Estimates suggest that an increase from the current federal minimum wage of US$7.25 to US$14 would be associated with a 17% decrease in smoking prior to pregnancy in a demographic group that is disproportionately burdened by poor health outcomes. Difference-in-differences results were consistent using TWFE and the more recent heterogeneity robust dCDH approach that accounts for staggered adoption of minimum wage policies.

We found that the estimates of associations of minimum wages with smoking before pregnancy were concentrated among those with an unintended pregnancy. This is consistent with a model of the minimum wage improving well-being through mechanisms such as decreased financial stress and increased access to healthcare and preventive services, which can in turn lead to reductions in smoking.^{10 13 19} Conversely, among those with intended pregnancy, changes in smoking behaviour may already be motivated by the pregnancy itself, regardless of income. Previous studies of the general population have found mixed associations between minimum wages and smoking, possibly reflecting differences in study population, measurement of smoking or analytical methodology.1825,28 One study of the impact of minimum wages on infant health using U.S. birth records found no evidence for reductions in maternal prenatal smoking as a potential pathway, which is consistent with our results.²² Our study is the first to investigate impacts of minimum wage increases on preconception smoking.

Public health and policy implications

Smoking before pregnancy significantly increases risk of preterm birth.⁴ Moreover, most women who smoke before pregnancy continue to smoke during pregnancy, which increases the risk of maternal and perinatal morbidity and mortality and has lifelong detrimental consequences for children.³ Use of ECs has become particularly popular among younger pregnant women, and initial studies indicate that ECs are also strongly associated with neonatal complications.³⁰ Consistent with previous literature on the impact of the minimum wage on infant health,²² our results suggest raising the minimum wage has the potential to result in significant societal cost savings, through, for example, reducing rates of preterm birth, low birth weight or other adverse outcomes. Future research quantifying the potential cost savings attributable to minimum wage increases is necessary for a comprehensive analysis of costs and benefits.

Our findings also have implications for addressing health disparities. We found that the strongest associations of minimum wages with smoking before pregnancy were among those who were non-Hispanic AI/AN and White. While the estimate for the subgroup of AI/AN women did not reach the threshold for statistical significance at conventional levels (p-value was 0.053), it is a noteworthy result given the small size of this subgroup (1.2% of analysis sample). Recent evidence has found that increasing minimum wages reduces racial inequities in earnings.⁴⁸ In our sample of non-college educated women, AI/AN and White women had the highest rates of smoking prior to pregnancy, which is consistent with national statistics.^{1 4} AI/AN women also experience disproportionately high rates of hypertension, gestational diabetes and preterm birth.⁴⁹ Our results suggest that raising minimum wages may be associated with a reduction in health disparities among AI/AN women.

More broadly, our findings contribute much needed evidence on the impact of social and economic policies on tobacco use and smoking behaviour. Social and structural determinants are significant contributors to maternal and infant health, and financial stress in particular is an important predictor of maternal smoking behaviour.^{12 49} A small but growing body of research has found positive impacts of income support and social safety net programmes such as EITC, Medicaid, SNAP and TANF on maternal and perinatal health.⁵⁰ Several studies found that higher state EITC is associated with reduced smoking rates, particularly for non-Hispanic White mothers with lower levels of education.51,53 Research on the effectiveness of social and economic policies, in isolation and in tandem, on maternal smoking and other maternal behaviours such as drinking alcohol and smoking marijuana is needed. Additionally, more research is needed on effective policies for reducing smoking during pregnancy and postpartum.

Strengths and limitations

This study has several strengths. First, PRAMS follows a standardised data collection methodology that results in consistent measures of smoking before, during and after pregnancy for 44 states and over 16 years. Second, we included a measure of EC use in our outcomes in later years. Third, the study period spanned large changes in minimum wages at both federal and state levels, allowing us to use a quasi-experimental approach that adjusts for unobserved confounding. In addition to the two-way fixed effects approach, we validated our main findings by implementing more recent alternative estimators that account for treatment effect heterogeneity. Fourth, we were able to conduct a number of sensitivity and placebo checks to confirm results were consistent across various specifications and alternative approaches to measurement of key variables.

This study also has limitations. First, treatment effect heterogeneity or non-linearities associated with pretreatment minimum wage levels are potential sources of effect variation that could bias estimates of the average treatment effect in TWFE models.⁵⁴ While we assess this possibility using an alternative estimator, future research should continue to explore heterogeneity in minimum wage impacts. Second, measures of smoking are self-reported and may be under-reported due to social stigma or recall bias. Regarding recall bias, one previous study found that retrospective recall of prenatal smoking status more than 14 years later, compared with prospective cotinine positivity, had high sensitivity and specificity (96% and 80%, respectively).⁵⁵ In addition, the PRAMS questions on smoking ask respondents about cigarettes specifically and do not ask about cigars or cigarillos, which individuals may believe are safer alternatives.⁵⁶ Questions on ECs were only asked beginning in 2016. We examined binary measures of smoking, which do not capture any potential effects of minimum wages on reductions in the number of cigarettes. Second, our analysis captured the ‘intention-to-treat’ effect on the group of individuals most likely to be affected by the minimum wage, but did not consider their individual wages or employment characteristics. Fourth, PRAMS data are not nationally representative because of some states’ decisions not to participate and due to low response rates which prevent some state-years from being publicly released. Data are representative of each state-year and cover about 80% of US births. However, results may not generalise to the USA as a whole.

Conclusions

Smoking before and during pregnancy is a significant public health problem that has detrimental effects on the health of women and children across the life course.³ Our findings provide evidence that higher minimum wages are associated with lower rates of smoking prior to pregnancy. Our study contributes to a growing body of research on the public health impacts of economic policies using quasi-experimental methods, which researchers and policy makers should consider in cost-effectiveness analysis of social programmes.

The funders/sponsors had no role in the design or conduct of the study.

Data availability statement

Data are available in a public, open access repository.