Comprehensive E-cigarette Flavor Bans and Tobacco Use among Youth and Adults

Abstract

The vast majority of youth e-cigarette users consume flavored e-cigarettes, raising concerns from public health advocates that flavors may drive youth initiation and continued use of e-cigarettes. Flavors drew further notice from the public health community following the sudden outbreak of lung injury among vapers in 2019, prompting several states to enact sweeping bans on flavored e-cigarettes. In this study, we examine the effects of these comprehensive bans on e-cigarette use and potential spillovers into other tobacco use by youth, young adults, and adults. We utilize both standard difference-in-differences (DID) and synthetic DID methods, in conjunction with four national data sets. We find evidence that young adults decrease their use of e-cigarettes by about two to three percentage points, while increasing cigarette use. For youth, there is some suggestive evidence of increasing cigarette use, though these results are undermined by pre-trend differences between treatment and control units. The bans have no effect on e-cigarette and smoking participation among adults 25 and over. Our findings suggest that statewide comprehensive flavor bans may have generated an unintended consequence by encouraging substitution towards traditional smoking in some populations.

1. Introduction

1.1. Background

The past decade has seen a major disruption to the tobacco/nicotine market with the advent of electronic cigarettes (e-cigarettes), or more broadly electronic nicotine delivery systems (ENDS). Entering the United States (U.S.) market in 2007, e-cigarettes have surged in popularity among youth, surpassing cigarettes in 2014 and becoming the most widely used form of nicotine among youth. After witnessing more than a 100% increase in the prevalence of e-cigarette use among high school students (from 11.7% to 27.5%) over 2017–2019), the U.S. Surgeon General declared youth vaping a national epidemic (U.S. Department of Health and Human Services – DHHS 2018).¹

No form of tobacco is deemed safe, especially for youth and young adults for whom nicotine exposure can present adverse developmental consequences. Adolescence, in particular, is a key period for brain development, as the prefrontal cortex, which is responsible for executive function, rational decision making, and higher order cognitive abilities continues to develop until about the age of 24 (López-Ojeda and Hurley 2024; Arain et al. 2013). E-cigarette use among young adults has also been linked to respiratory symptoms (Tackett et al. 2024). While not completely safe, e-cigarettes are considered to be a safer alternative to combustible cigarette use, though there exists a degree of uncertainty with respect to the relative risk of these tobacco products. The Office of Health Improvement and Disparities in the United Kingdom (McNeill et al. 2022) recently reiterated its prior conclusion that nicotine vaping poses only a small fraction of the risk relative to smoking (about 5%), whereas a recent survey of 137 tobacco control experts reported a 37% relative health risk, on average, for e-cigarette use compared to smoking (Allcott and Rafkin 2022).

The heavy toll of smoking, responsible for over 480,000 deaths annually (U.S. DHHS 2014), coupled with the significantly lower relative risk profile of e-cigarettes, have presented a key regulatory challenge. Policymakers at the federal, state, and local levels have grappled with how best to regulate access to e-cigarette products such that their harm reduction potential is maximized (i.e. for adults who want to use these smoking alternatives to quit the habit or reduce their combustible cigarette consumption) while constraining uptake and use among youth. This uncertainty surrounding the optimal regulatory approach is reflected in the variance in the policy landscape across the country. For instance, e-cigarette taxes – an increasingly popular policy lever deployed by states and localities to curb e-cigarette use – are currently levied in only 32 states and in District of Columbia, along with a handful of local jurisdictions. In contrast to cigarettes, however, there is no federal tax on e-cigarettes.² And, even among states and localities that have adopted these taxes, policies vary widely in their structure (i.e. ad valorem vs. excise tax vs. specific sales tax) and in the amount of the tax (Dave et al. 2022). Moreover, several studies have shown that while higher e-cigarette taxes are effective in reducing vaping, especially among youth and young adults, they generate an unintended consequence in the form of increasing cigarette sales and smoking participation and deterring smoking cessation (Abouk et al. 2023; Cotti et al. 2022; Saffer et al. 2020).

In designating youth e-cigarette use a public health epidemic, the U.S. Surgeon General highlighted the popularity of flavored e-cigarettes among youth and the importance of reducing access to flavored tobacco products for young people (U.S. DHHS 2018).³ Advocates contended that flavored e-cigarettes were very appealing to youth and that restrictions on flavors could decrease tobacco use by youth (Chen et al. 2017). Among high school students who currently use e-cigarettes, the vast majority (~85%) use flavored ones (Wang 2020). Flavors have been linked to youth initiation of e-cigarette use (Zare et al. 2018; Villanti et al. 2017) and drew further notice from the public health community following the sudden outbreak of lung injury and deaths among vapers in 2019. This e-cigarette, or vaping product use-Associated lung injury (EVALI) outbreak, was later linked to vaping devices used to consume cannabis e-liquids which had harmful additives.

The 2009 Family Smoking and Tobacco Prevention Act had banned the sale of flavored cigarettes, though menthol and tobacco flavors were exempted, and other flavored tobacco products – notably flavored e-cigarettes – remained on the market. This was partially remedied when the Food and Drug Administration (FDA) extended the ban to cover cartridge-based e-cigarettes in February of 2020. However, menthol and tobacco-flavored cartridge-based e-cigarettes were allowed to remain on the market, and the FDA ban also permitted all flavors to continue to be sold in disposable e-cigarettes and in tank-based vaping devices.⁴ Because of these exemptions and substitution possibilities, the federal ban could be easily circumvented rendering its potential impact on flavored e-cigarette use and overall e-cigarette use to be minimal (Romm et al., 2022).

Largely in response to the sudden outbreak of severe lung injury among vapers in 2019/2020 and in recognition of the federal exemptions, several states enacted more sweeping restrictions aimed at flavored e-cigarettes by banning all flavors and/or extending the federal ban to all e-cigarette devices. The key regulatory dilemma of balancing harm reduction while constraining youth access also applies to these more stringent statewide restrictions on the sale of flavored e-cigarettes. Even if these bans are effective in reducing flavored and overall e-cigarette use among youth and young adults – though it is not certain that they are since the restrictions could still be circumvented through cross-border purchases, online purchases, purchases at exempted retailers, or users adding their own flavors – they could generate unintended consequences in the form of substitution to cigarettes. These possibilities were reported by Romm et al. (2022) in a survey of young adult e-cigarette users just following the 2020 federal ban, who reported how they would respond to comprehensive flavor restrictions. Some participants reported they would quit vaping or have ways to circumvent the restrictions and not be impacted, others said they would substitute to cigarettes.

1.2. Contributions

This study directly examines the effects of statewide flavor bans on youth, young adults and adults, and presents comprehensive evidence on how these bans have impacted e-cigarette use and smoking in general. We rely on information from four national datasets: the combined high school, state Youth Risk Behavior Survey (YRBS), the Monitoring the Future (MTF) survey, the Behavioral Risk Factor Surveillance System (BRFSS), and the Population Assessment of Tobacco and Health (PATH) Study. The main analyses rely on a generalized difference-in-differences approach and the synthetic difference-in-differences approach (Arkhangelsky et al. 2021; Clarke et al. 2023). The main conclusions are based on the weight of the evidence across multiple data sets and on the validity of the counterfactual assumptions.

We document several key findings. First, for youth (ages 14–17), while there is some indication of a small increase in the use of cigarettes, we find little evidence to suggest that the statewide flavor bans reduce their overall e-cigarette participation. Models that support parallel trends also do not indicate any meaningful spillovers into smoking participation for youth. Second, for young adults (ages 18–24), we find evidence that the comprehensive restrictions on flavored e-cigarettes lowered the use of the banned flavored e-cigarettes and reduced their overall e-cigarette participation by about two to three percentage points. The bans appear to have generated an unintended consequence by raising the smoking participation in this population. Finally, for adults aged 25 and over, the statewide bans have no discernible impact on either e-cigarette or cigarette use.

The remainder of this paper proceeds as follows: Section 2 provides background on the statewide restrictions on e-cigarette flavors and Section 3 summarizes the relevant literature. The multiple data sets are outlined in Section 4, and Section 5 describes our methods. Our main results, robustness checks, and extensions are reported in Section 6. Finally, Section 7 concludes by offering further context for our findings with respect to limitations and policy implications.

1.3. Background on statewide e-cigarette flavor bans

Between October 2019 and July 2020, eight states implemented far more sweeping restrictions on flavored e-cigarettes in relation to the federal ban, which went into effect in February 2020. Table 1 presents a timeline of these state-level restrictions. The adoption of more comprehensive restrictions on e-cigarette flavors was largely driven by concerns regarding the health effects of vaping as they unfolded over 2019–2020 with the outbreak of lung injury among vapers. Most of the states that enacted permanent bans on flavors also enacted, or attempted to enact, emergency flavor bans as a result of this nation-wide outbreak of severe lung disease linked to e-cigarettes and other vaping devices in 2019.⁵

Table 1.

Timeline of statewide e-cigarette flavor bans

Date		10/19	11/19	12/19	1/20	2/20*	3/20	4/20	5/20	6/20	7/20
State	EVALI
Maryland	12.10				P	P	P	P	P	P	P
Massachusetts	17.86		T	T	T	T	T	T	T	P	P
New Jersey	13.59							P	P	P	P
New York**	8.75								P	P	P
Rhode Island	5.00	T	T	T	T	T	P	P	P	P	P
Utah**	39.06										P
Montana	5.00	T	T	T	T
Washington	3.25	T	T	T	T
United States	6.67

Note: EVALI represents the approximate number of E-cigarette or Vaping Use-Associated Lung Injury (EVALI) hospitalizations or deaths reported to the Centers for Disease Control and Prevention (CDC), as of February 2020, per million population. T and P represent temporary and permanent bans on flavored vapes, respectively. The statewide flavor bans are for e-cigarettes only. No state bans tobacco flavor. Only Massachusetts bans menthol in all tobacco products. Maryland only prohibits the sale of cartridge-based and disposable e-cigarettes with flavors. Montana and Washington are excluded from the analyses presented in this paper due to implementing only temporary bans. California is also excluded because it implemented a statewide flavor ban that went into effect after 2020, in December 2022. Ban data are obtained from Tobacco Free Kids. See https://www.tobaccofreekids.org/assets/factsheets/0398.pdf.

^*Federal ban on cartridge based flavored e-cigarettes goes into effect.

^**Temporary bans blocked by legal challenges.

As shown in Table 1, eight states issued emergency rules to temporarily ban the sale of flavored e-cigarettes. As a result of legal challenges, these orders were blocked in four states. Temporary bans adopted in Rhode Island (RI) and Massachusetts (MA) became permanent in March and June of 2020, respectively. New York (NY) and Utah (UT), where bans were initially blocked by legal challenges, were able to enact permanent bans. New Jersey (NJ) and Maryland (MD) also enacted permanent bans. Montana (MT) and Washington state (WA) implemented temporary restrictions on flavored e-cigarette sales in October 2019, which did not convert into a permanent ban and expired in January 2020. We exclude these two states from the analysis since the bans were very short-lived; we also exclude these states from the control group given they have been previously treated, albeit for a short period of time. California enacted a statewide flavor ban in December 2022 which is much later than the other states. There are often dynamic effects in the post period which impact estimation when groups with different length post periods are averaged. Thus, we exclude California from the study to avoid this type of distortion.

1.1. Use of flavored e-cigarettes

Table 2 presents descriptive statistics based on the PATH data, on the percent of nicotine vapers who use banned flavors across the three age groups. In all treatment states other than MA, all flavors were banned in all e-cigarette devices except for tobacco and menthol flavors. MA further banned menthol flavors in e-cigarette products as well. These descriptives underscore three key points.

Table 2.

Weighted descriptive statistics on banned e-cigarette flavor use

Treated States Pre-Period	Treated States Post-Period	Difference	Control States Pre-Period	Control States Post-Period	Difference	Difference-In-Difference
Panel A: Youth
66.06%	76.34%	10.28	59.27%	70.68%	11.41	−1.13
Panel B: Young Adults 18–24
44.74%	61.63%	16.89	42.92%	64.07%	21.15	−4.26
Panel C: Adults 25+
28.92%	44.69%	15.77	29.98%	50.17%	20.19	−4.42

Note: Authors' calculations using PATH data from 2015–2023, excluding 2020 due to the COVID-19 pandemic and excluding 2014 as e-cigarette flavor use questions were not introduced until 2015. Table presents the percent of current e-cigarette users who reported using banned flavors. Difference columns are calculated by subtracting the pre-period from the post-period and represent percentage point difference. Difference-in-difference column is calculated by subtracting the control state difference from the treated state difference and represents percentage point difference. Treatment states are Maryland, Massachusetts, New Jersey, New York, Rhode Island, and Utah. Banned flavors are candy, fruit, chocolate, clove/spice, alcoholic drink, non-alcoholic drink, and other flavors. Unbanned flavors are tobacco and menthol, except for Massachusetts where tobacco is the only unbanned flavor. Data were adjusted for sample weights.

First, banned flavors were most popular among youth, with the majority of youth who currently use e-cigarettes reporting use of banned flavored e-cigarettes, both before and after the bans. There is a steep age gradient in the use of the banned flavors among current e-cigarette users, with the popularity of these flavors declining substantially from youth to young adults and further decreasing among adults 25 and over. While this gradient partially narrows in the post-period between youth and young adults, the overall age pattern remains pronounced.

Second, a notable pattern emerges post-treatment: the use of banned flavors increases across all age groups in both treatment and control states, conditional on e-cigarette use. This suggests that the flavor bans were not very effective in their implementation. The relatively high use of flavors in the treatment states, even after the bans go into effect, may be in part due to exemptions for certain store types. For instance, MA exempts stores that primarily sell tobacco, e-cigarette establishments, tobacco/smoking bars, adult-only retailers, and liquor stores from all flavor bans. UT also exempts tobacco retail specialty businesses from flavor bans (Public Health Law Center, 2023). Users are also able to add their own flavors to the e-liquid mix by opening the e-cigarette cartridge or tank device. Because it is not difficult to make these modifications, a flavor ban could also result in a black market for flavored e-cigarettes. This is essentially what happened during the 2019 outbreak of lung injuries, which were linked to vape devices that had been modified and sold by black market operators. Hence, users may still be able to obtain flavored e-cigarettes through online purchases or illegally on the black market or through establishments due to lack of enforcement.⁶

Third, the unconditional difference-in-differences estimates in the last column reveal that the treatment states experienced smaller increases relative to control states. We observe a clear age gradient in this effect as well, with adults 25 and over showing the largest relative decline (4.42 percentage points), followed by young adults (4.26 percentage points), and youth showing the smallest decline (1.13 percentage points). Despite the challenges in implementation, these differences suggest that flavor bans may have still had some impact on flavor consumption patterns. The smaller effect among youth may reflect their stronger preference for flavored products and potentially greater motivation to seek alternative sources, as evidenced by their consistently higher usage rates across both pre- and post-treatment periods.

2. Prior studies

Flavor restrictions on e-cigarettes have been primarily motivated by concerns over their popularity among youth users, with policymakers aiming to prevent both initiation and continued use in this age group of particular concern. The literature examining the effectiveness of these restrictions is still developing, paralleling the evolving policy landscape itself. Earlier studies predominantly used sales data for their analyses, while more recent research has shifted toward self-reported usage data, enabling researchers to estimate effects across different age groups. This age-specific analysis is particularly crucial given the well-documented variations in nicotine’s neurological and developmental impacts across different life stages, as well as different levels and trends in tobacco use.

Ali et al. (2022) study the effects of flavor restrictions on e-cigarette sales in three states with a permanent ban (MA, NY, RI) and one state with a short-lived ban (WA) using early data through 2020, and thus identifying very short-term effects for up to a year post-treatment. They find substantial reductions in sales (on the order of 25–31%), largely driven by a reduction in the sale of non-tobacco flavored e-cigarettes. Xu et al. (2022), using a similar post-ban window extending through early 2020, widen the lens to study effects on cigarette sales. They focus on bans in three states (MA, RI, and WA) and find significant increases in cigarette sales in the short term on the order of 5–8%. Expanding on the number of treated localities (to include seven statewide bans as well as various sub-state local bans) and extending the post-treatment window through early 2023, Friedman et al. (2023) also find a significant reduction in ENDS sales, driven by a decrease in the sale of flavored products, and a substitution into cigarette sales, both overall and for brands disproportionately preferred by youth.

All of these studies rely on commercial sales data from Circana (formally known as Information Resources, Inc.), which cover sales from national chain convenience stores, large food stores, drug stores, mass merchandiser outlets, and military sales. This work identifies compelling effects on e-cigarette sales and potential substitution into cigarettes, but the use of these commercial sales introduces three main limitations to any analysis. First, sales from online retailers, independent convenience stores, independent food stores, other independent stores (excluding drug stores), and tobacco specialty stores such as vape shops are excluded. These exclusions omit a large share of tobacco sales. For instance, Selya et al. (2023) conclude that about 50% of the e-cigarette market is not recorded by Circana. Using the PATH data, we find that about 60% of youth purchases of vaping products occur through tobacco specialty stores; the corresponding shares for young adults and adults 25+ are 70% and 67%, respectively.

In addition to capturing only a limited fraction of tobacco sales, estimates using the Circana data may further present a distorted picture of the impact of bans since many of the retailers not represented in the data (i.e. vape dispensaries, specialty tobacco retailers) were also exempted by the flavor bans in certain states. If the bans shifted sales away from traditional retailers to these specialty retailers, either because they were exempted or less vigorously enforced, then the identified treatment effects in studies using the Circana data may be overstated.

Second, sales do not equate to use. A reduction in sales in the banned states could be offset by an increase in cross-border sales or through illicit purchases. Indeed, in recent work using the Circana data Chen et al. 2023 uncovered strong evidence of spatial spillovers, even over a short post-treatment window (through February 2020). They find that bans implemented in four states (MA, WA, RI, and MT) resulted in significant increases in ENDS sales in neighboring counties. One other concern is that these aggregate sales data are not age specific and thus cannot be used to investigate the heterogeneous effects on use across youth vs. adults or across other sub-populations of interest.

There are only a few quasi-experimental studies of comprehensive flavor bans that have gone beyond the effects on sales, and they have largely focused on a single state or locality prior to the 2020 federal ban. Several studies have explored the effects of restrictions on flavored tobacco that were adopted in the San Francisco Bay Area over 2018–2019. In their analysis of the impact of these bans among high school students, using the California Healthy Kids Survey, Dove et al. (2023) find no effects on current or ever use of e-cigarettes over a post-policy window of one year.⁷ They attribute this finding to potential substitution from the banned to the non-banned flavors and/or cross-border purchases. Friedman (2021), utilizing data on high-school students from the district YRBS, finds robust evidence that San Francisco’s ban also resulted in youth substituting into cigarette use, even over the study’s short post-policy window.⁸ Hawkins et al. (2022) study how local restrictions on flavored tobacco products in MA counties, which predated the federal flavor ban, affect youth use of e-cigarettes and cigarettes. Using biennial data from the 2011–2017 Massachusetts Health Surveys, they find significant reductions in both e-cigarette and cigarette use among youth.

Based on an online survey of 1,624 adult e-cigarette users, a recent study (Yang et al. 2023a) explores pre-post changes in e-cigarette use and flavored e-cigarette use associated with flavor restrictions in three states (WA, NJ, and NY). Following the ban, 8.1% of e-cigarette users stopped using e-cigarettes, and overall, the use of non-flavored e-cigarettes increased from 5.4% to 25.4%. Descriptive evidence indicates that e-cigarette users were able to obtain the banned flavors after the restrictions through various means: in-state retailers, cross-state purchases, online purchases, black market, mixing the flavors themselves, and stocking up on e-cigarettes prior to the ban. Their finding that 45% of e-cigarette users continued to be able to purchase the banned flavors from in-state retailers suggests that compliance and enforcement were not high.

More recently, Cotti et al. (2024) and Friedman et al. (2024) examined the effects of flavor bans on tobacco outcomes using a subset of the data sets employed in this study. The Cotti et al. paper appeared in the same month as the prior working paper version of our study (Saffer et al. 2025), while the Friedman et al. paper appeared a few months later. It is reassuring that all these studies find similar effects, though our study differs in several ways. Most importantly, we provide evidence using two additional datasets, the MTF survey (for youth), and the longitudinal PATH (for youth and adults). As shown in Table 2, a key strength of the PATH is that it contains information on the specific e-cigarette products that were used, including the banned flavors. We also provide estimates for adults 25+ using the BRFSS, while Friedman et al. (2024) only consider young adults in this data set. Similar to our study, Cotti et al. (2024) also obtain estimates for all adults using the BRFSS. For youth, we provide estimates using the MTF in addition to the YRBS. Additionally, our analyses leverage a synthetic difference-in-differences (SDID) estimation strategy. This estimator has several advantages compared to the standard two-way fixed effects and synthetic control methods, which are discussed in Section 5.

3. Methods

3.1. Data

To obtain a comprehensive view of the effects of flavor ban policies on smoking and e-cigarette use, we capitalize on information from four national data sets listed above, each offering complementary strengths. Figure 1 illustrates trends in e-cigarette and cigarette use in our study period, separately for youth (ages 14–17), young adults (ages 18–24), and adults 25 and over, across the four data sets.⁹ Appendix Figure A1 presents trends further broken down by treatment status. The outcome variables we use in this figure and in all our analyses are dichotomous past 30-day use of cigarettes or e-cigarettes.¹⁰ A summary of all data sets with descriptive statistics on outcome variables is also presented in Appendix Tables A1 and A2.

Figure 1. Trends in e-cigarette and cigarette use.

Note: Authors’ analyses of the YRBS, PATH, MTF, and BRFSS. All samples exclude Washington and Montana, if present, due to temporary flavor bans in these states; and California, if present, due to statewide flavored tobacco ban that went into effect in December 2022. Youth sample is restricted to ages 14–17 for the PATH, and ages 14+ for the MTF and YRBS. Young adult and adult samples are restricted to ages 18–24 and 25+ in all datasets, respectively. Data were adjusted for nationally representative sample weights.

3.1.a. Youth Risk Behavior Survey

The 2023 combined, high school state YRBS (CDC, 2024a) is a self-administered survey conducted in high schools that collects information on health behaviors the Centers for Disease Control and Prevention (CDC) has identified as critical for mortality and morbidity. It provides a biennial representative sample of youth in grades 9–12 for each participating state in odd-numbered years.¹¹ For our analyses, we use the combined data set provided by the CDC, which pools data from all participating states, standardizes variables across the survey years, and provides appropriate weights to make the sample representative at the state level. Our sample period covers e-cigarette use from 2015–2023 and cigarette use from 2013–2023, biennially.¹² The YRBS offers several important advantages. With approximately 150,000 students surveyed in a given year nationally, the combined state data set yields large sample sizes for assessing heterogeneity and improving the precision of the estimates. The YRBS is also one of the few national data sets that is state-representative, which helps to minimize bias in identifying the effects of statewide interventions (such as the ones we study here) that may arise due to potential shifts in the composition of state-specific samples.¹³

3.1.b. Monitoring the Future

As with the YRBS, the MTF survey (Miech et al, 2024) is also a school-based survey, and it is nationally representative of middle school and high school students in the 8^th, 10^th, and 12^th grades. Approximately 45,000 students are sampled each year. In our study, we exclude students under 14 years of age (since smoking prevalence is extremely small for this group) and utilize the restricted version of the MTF with state identifiers from 2014 through 2023. Due to the difficulty with in-school sampling during the pandemic, we exclude 2020 from the analyses. It is important to note that only 50% of the students in the sample were asked about e-cigarette use in the MTF. We adjust the weights following the data user guidelines to account for this.

3.1.c. Population Assessment of Tobacco and Health

The PATH (2024) is a panel study that longitudinally resamples youth ages 12–17 and adults of all ages in multiple waves. It takes place in-home and therefore provides an alternate sample to the school-based MTF and YRBS. We use waves 1 through 7 which cover 2014–2023, where both cigarette and e-cigarette use are measured throughout. We again drop 2020 from the analyses due to challenges with in-home sampling during the pandemic. An advantage of the PATH is that it contains detailed information on the use of flavored ENDS products, which is important for assessing the popularity of these flavors among youth and adults and how e-cigarette users shifted their consumption across banned and non-banned flavors following the restrictions (Table 2). Another advantage of the PATH is its sampling of both youth as well as adults. There are approximately 13,000 youth, 8,000 young adults (ages 18–24), and 16,900 adults (ages 25+) in each year. The samples are refreshed time to time from a shadow sample to maintain sample sizes and representativeness. They also include new sets of youth as they age into the sample, and youth who age out are then included in the adult sample. To be consistent with the analyses using other data sets, we treat the pooled PATH as repeated cross-sections in this study and use the survey weights appropriate for cross-sectional data analysis.

3.1.d. Behavioral Risk Factor Surveillance System

The BRFSS (CDC, 2024b) is a state-representative phone survey of adults conducted on a yearly basis. It covers approximately 24,000 younger adults (ages 18–24) and 410,000 adults (ages 25+) sampled independently each year. One disadvantage of the BRFSS is that the e-cigarette use variable was only included in an optional module that each state could opt into or out of in 2018, and it was not measured at all in 2019, and included again as an optional module in 2020. This limits the consistency of this measure across time. Hence, we only include the years in which the e-cigarette use variable was in the main survey questionnaire (i.e. 2016, 2017, 2021, 2022 and 2023) in our analyses of e-cigarette use in the BRFSS. Past 30-day cigarette use is measured consistently from 2014–2023.¹⁴

3.1.e. Covariates

We account for various additional confounding demographic measures, tobacco regulatory variables, and factors related to the COVID-19 pandemic in our main analyses. Demographic covariates include age, and indicators for race/ethnicity (White/Black/Asian/Hispanic) and sex (male). These measures are obtained separately from each of the four surveys that we employ. Descriptive data on these covariates are shown in Appendix Table A3.

Tobacco regulatory variables are at the state-year level and include combustible cigarette taxes, and indicators for the adoption of statewide e-cigarette taxes are available in a CDC data set called STATE, internet sales bans on tobacco products come from the Public Interest Law Center, 2024. Since counties in some states implemented local flavor bans prior to 2020, we include an indicator for those states (Chen et al. 2018; Truth Initiative 2020, 2022, 2024; Campaign for Tobacco-Free Kids 2025). This indicator does not turn on for the six treatment states with statewide bans unless they had separate county-level bans prior to 2020.

Three of the four years of our post-treatment period (2020–2022) coincide with the COVID-19 pandemic. Therefore, we account for a measure proxying the impact of the pandemic in each state: COVID-19-related deaths per 100,000 population from the National Center for Health Statistics (NCHS) 2023 and 2023. Higher death rates may have increased awareness of the extent of the pandemic and discouraged the use of tobacco products. We also account for actions states took to curtail the pandemic’s spread: the issuance of stay-at-home (shutdown) orders by the state’s governor. These orders may have had spillover effects on tobacco participation.¹⁵

We also control for the state governor’s political party affiliation of during the pandemic years (2020–2022) as a complementary measure of COVID-19 policy stringency. While the Hale et al. (2021) index captures formal stay-at-home orders, implementation and enforcement of these and other policies often varied by political leadership. Democratic and Republican governors frequently approached pandemic restrictions differently, even when formal policies appeared similar on paper. Our indicator identifies states with a Democratic governor in at least two of the three pandemic years (National Governors Association 2025a, 2025b). The variable equals zero for all states in pre-pandemic years and in 2023, effectively creating an interaction between pandemic timing and states with Democratic governors.¹⁶

We created a state-year level covariate data set covering 2015 through 2023 and merged these variables to the PATH, MTF, YRBS, and BRFSS using the state of residence and survey year of the respondents in each survey. Summary statistics of these variables are presented in Appendix Table A4, broken down by treatment status of the states.

4. Methods

We leverage the quasi-natural experiment provided by the enactment of comprehensive flavor bans in six states (MD, MA, NJ, NY, RI, and UT) to provide plausibly causal estimates of the effects of these bans. First, we start with the following standard difference-in-differences (DID) model:

$$E_{ist}=\mu +{\beta \mathrm{*}FBAN}_{st}+X_{ist}\mathrm{*}\theta +Y_{st}\mathrm{*}\delta +{\gamma }_s+{\tau }_t+{\epsilon }_{ist}$$ (1)

Here, $E_{ist}$ denotes various dichotomous tobacco use outcomes for person $i$, residing in state $s$, at year $t$. ${FBAN}_{st}$ is an indicator for when, and which states, enacted the comprehensive e-cigarette flavor ban. The six states with permanent e-cigarette flavor bans adopted these bans between late 2019 and mid-2020. These adoption dates are reasonably proximate and minimally staggered such that 2020 can be defined as the treatment initiation year. The issues associated with potential biases in DID models due to staggered adoption periods are thus not empirically relevant. $X_{ist}$ is a vector of demographic characteristics including age, sex, race, and Hispanic ethnicity. $Y_{st}$ is a vector of the state-year level variables defined in Section 4.1.e. The terms $\mu$ and ${\epsilon }_{ist}$ denote the intercept and the disturbance term, respectively.

In addition, all models include fixed effects for each state (${\gamma }_s$), which accounts for any stable unmeasured heterogeneity across these areas (for instance, differences resulting from unmeasured cultural factors or sentiment towards tobacco use) and for each year (${\tau }_t$), which captures unobserved secular trends in tobacco use outcomes impacting the full sample. The parameter of interest, $\beta$ captures the average causal effect of the flavor bans on tobacco use realized over the post-treatment period.

We estimate the effects of flavor bans for the three age groups separately, and for e-cigarette use and cigarette use in the past month. As noted above, the three age groups studied are youth (ages 14–17), young adults (ages 18–24), and adults (ages 25+). We estimate equation (1) at the individual level, but cluster bootstrap standard errors at the state level. We report p-values at 1, 5, and 10 percent levels.

To draw a more explicit focus on the validity of the counterfactual design, we also use the synthetic difference-in-differences (SDID) estimator (Arkhangelsky et al. 2021; Clarke et al. 2023). The SDID estimation bridges strengths from both panel data DID and synthetic control (SC) methods, while providing various additional strengths and modeling flexibility. Specifically, in its basic form, a consistent causal effect of the flavor restrictions on a given tobacco use outcome ($Y_{st}$, in a given state $s$ at time period $t$) can be derived by estimating:

$$({\widehat{\beta }}^{SDID},\widehat{\mu },\widehat{\gamma ,}\widehat{\tau })=\begin{array}{c}argmin\\ \beta ,\mu ,\gamma ,\tau \end{array}\{\sum _{s=1}^n\sum _{t=1}^t(Y_{st}-\mu -{\gamma }_s-{\tau }_t-{FBAN}_{st}\beta {)}^2{\widehat{\omega }}_i^{SDID}{\widehat{\lambda }}_t^{SDID}\}$$ (2)

In equation (2), the causal effect of the treatment, that is the average treatment effect on the treated (ATT; represented above by ${\widehat{\beta }}^{SDID}$), is estimated from a two-way fixed effects (TWFE) model with optimally-chosen weights ${\widehat{\omega }}_i^{SDID}$ and ${\widehat{\lambda }}_t^{SDID}.$¹⁷ In contrast to the standard TWFE DID estimation, however, which relies on the “parallel trends” assumption, SDID more flexibly reweights and matches pre-treatment trends by selecting a weighted set of control units that minimizes the trend differences in the pre-exposure periods. Specifically, optimal unit-specific weights ${\widehat{\omega }}_i^{SDID}$ are chosen to align pre-treatment trends across outcomes in the untreated vs. treated states, subject to a regularization parameter which prevents overfitting while increasing the variance and uniqueness of the weights. SDID also introduces and optimally chooses time-specific weights ${\widehat{\lambda }}_t^{SDID}$ to further remove bias from unobserved shocks and improve precision. These considerations serve to improve the robustness and precision of the SDID estimator, in addition to making the model more flexible in generating credible counterfactual comparisons (Arkhangelsky et al. 2021). As with our estimates of the ATT based on the TWFE DID, given by equation (1) and with one exception noted below, we construct standard errors for the SDID estimates by cluster bootstrapping at the state level. We report the same three levels of p-values that we report on the DID models.

SDID estimation requires a balanced panel which, given the state-level policy variation and the cross-sectional nature of the data we are using, requires aggregation to the state and year level. This necessitates excluding the states which do not appear in every year. SDID has been shown in some contexts to outperform two-way fixed effects models based on having superior power and insensitivity in power to selection of the pre-treatment period by the analyst (Dench et al. 2024). In this case, this advantage may be balanced against the need to drop some states from some analytic samples.

In order to assess the plausibility of the pre-policy parallel trends assumption between the treatment and control states, we rely on event study plots. In the context of the DID analysis, we use the reference period that is closest to the treatment year, and control groups are all weighted equally. The SDID event studies are time series plots of the conditional difference between the treatment group and the control groups. Because the average pre-period differential between the treatment group and control group is subtracted from each period’s differential, the “reference group” is the average over the entire pre-policy period. We report 95 confidence intervals around the point estimates for the event studies based on the bootstrapping method.

BRFSS and PATH DID models include all six treatment states. YRBS DID models exclude RI because it is not in the combined high school state YRBS in our sample period. The data use agreement that we signed with the Inter-university Consortium for Political and Social Science Research (ICPSR) to access restricted MTF files with state identifiers prevent us from identifying the treatment states in DID MTF models. Unlike individual-level DID models, SDID models require balanced and aggregate state-level data. BRFSS SDID models include all six treatment states. PATH SDID models include all treatment states except RI. YRBS SDID models are limited to one treatment state, MD, and standard errors and confidence intervals are computed by the placebo method in this case.

Because we rely on a set of studies and alternate samples, we further construct and report an aggregate of the separately estimated treatment effects for each age group and outcome using a fixed effects method of aggregation (Hedges 1998). We do so to provide a convenient summary of our estimates and to highlight patterns across different data sets. This “meta-analysis” involves taking a weighted average of the estimates across the alternate data sets, where the weights are the inverse of the squared standard error of each estimate, normalized to add up to one. The standard errors for these aggregate estimates are obtained by taking the inverse of the square root of the sum of all these weights. This approach requires the assumption that each data set is estimating the same target parameter from an underlying population (i.e., youth, young adults, adults) but with different samples. We believe this is a plausible assumption in our case.

5. Results

5.1. Main findings on the impact of flavor bans on e-cigarette and cigarette use

We report our main findings in Table 3. The accompanying event study plots of statistically significant estimates are shown in Figures 2 through 4. The remaining event study plots are shown in the Appendix Figures A2–A4. In Table 3, we present the estimated treatment effects of the statewide flavor bans on our key outcomes – e-cigarette and cigarette use – across the three age-defined sub-populations, across the four datasets, and for both the DID and SDID estimation. In our discussions, we emphasize the results based on our preferred method, SDID, because of the improved matching process, though our conclusions and overall pattern of findings are not materially different from the standard DID estimates. We report the overall ATT estimates across all data sources for each age group and outcome in Table 4.

Table 3.

Results by dataset: Dichotomous e-cigarette and cigarette use in the past 30 days

	Panel A: Youth		Panel B: Young Adults 18–24		Panel C: Adults 25+
	YRBS		BRFSS		BRFSS
	DID	SDID	DID	SDID	DID	SDID
E-cigarette	0.008	0.000	−0.032 ***	−0.025 **	−0.003	−0.001
SE	(0.021)	(0.035)	(0.012)	(0.012)	(0.003)	(0.004)
p-value	0.694	0.995	0.009	0.040	0.330	0.890
N	607,350	90	107,074	225	1,751,597	225
Mean Y	0.192	0.199	0.158	0.160	0.049	0.049
Cigarette	0.016 **	0.016	0.035 ***	0.035 ***	0.000	−0.001
SE	(0.008)	(0.024)	(0.006)	(0.007)	(0.003)	(0.003)
p-value	0.040	0.501	0.000	0.000	0.933	0.743
N	757,955	90	214,379	460	3,545,771	460
Mean Y	0.076	0.083	0.121	0.119	0.167	0.166
	PATH		PATH		PATH
	DID	SDID	DID	SDID	DID	SDID
E-cigarette	−0.001	−0.025	−0.009	−0.039	0.000	0.002
SE	(0.012)	(0.024)	(0.028)	(0.060)	(0.011)	(0.016)
p-value	0.914	0.304	0.746	0.513	0.967	0.880
N	43,010	296	54,722	296	117,939	296
Mean Y	0.069	0.081	0.186	0.203	0.060	0.058
Cigarette	0.022 **	0.024	0.016	0.005	−0.003	−0.004
SE	(0.009)	(0.021)	(0.023)	(0.045)	(0.009)	(0.031)
p-value	0.014	0.268	0.480	0.919	0.745	0.892
N	43,108	296	54,880	296	118,581	296
Mean Y	0.036	0.039	0.207	0.213	0.202	0.202
	MTF
	DID	SDID
E-cigarette	−0.019	−0.011
SE	(0.016)	(0.020)
p-value	0.247	0.585
N	112,887	288
Mean Y	0.137	0.129
Cigarette	0.017 *	0.011
SE	(0.009)	(0.020)
p-value	0.052	0.562
N	194,752	288
Mean Y	0.053	0.052

Note: Authors’ analyses of the YRBS, PATH, MTF, and BRFSS. All samples exclude Washington and Montana, if present, due to temporary flavor bans in these states; and California, if present, due to statewide flavored tobacco ban that went into effect in December 2022. Youth sample is restricted to ages 14–17 for the PATH and ages 14+ for the MTF and YRBS. Young adult and adult samples are restricted to ages 18–24 and 25+ in all datasets, respectively. Table presents difference-in-differences (DID) and synthetic difference-in-differences (SDID) estimates of each outcome, bootstrapped standard errors clustered at the state level, p-values, sample sizes, and weighted means of the dependent variables. DID models use individual-level data. BRFSS and PATH DID models include all six treatment states. YRBS DID models exclude Massachusetts and Rhode Island because these states are not in the YRBS in our sample period. The confidentiality agreement that we signed with the Inter-university Consortium for Political and Social Science Research (ICPSR) to access restricted MTF files with state identifiers prevents us from identifying the treatment states in DID MTF models. SDID models require balanced, aggregate state-level data. BRFSS SDID models include all six treatment states. PATH SDID models include all treatment states except Rhode Island. YRBS SDID models only include Maryland. The confidentiality agreement that we signed with the Inter-university Consortium for Political and Social Science Research (ICPSR) to access restricted MTF files with state identifiers prevents us from identifying the treatment states in SDID MTF models. All models control for age, race/ethnicity (White/Black/Asian/Hispanic), sex (male), cigarette tax, e-cigarette tax (binary), internet sales ban (binary), county-level flavor bans within state (binary), state-year COVID-19 stay-at-home order index, state-year COVID-19 death rates (per 100,000 population), and an indicator for Democratic state governor for the majority of the time between 2020–2022 (binary). Data were adjusted for sample weights.

^***p < 0.01;

^**p < 0.05;

^*p < 0.1.

Figure 2. Difference-in-differences event study plots for youth.

Note: Authors’ analyses of the YRBS, PATH, and MTF. All samples exclude Washington and Montana, if present, due to temporary flavor bans in these states; and California, if present, due to statewide flavored tobacco ban that went into effect in December 2022. Sample is restricted to ages 14–17 for the PATH, and ages 14+ for the MTF and YRBS. Figure presents the event study analysis of the synthetic difference-in-differences (SDID) estimates from Table 3. Shaded area represents 95% confidence interval around the point estimates based on the bootstrapping method. YRBS SDID models only include Maryland as the treatment state. PATH SDID models include all treatment states except Rhode Island. The confidentiality agreement that we signed with the Inter-university Consortium for Political and Social Science Research (ICPSR) to access restricted MTF files with state identifiers prevents us from identifying the treatment states in SDID MTF models. All models control for age, race/ethnicity (White/Black/Asian/Hispanic), sex (male), cigarette tax, e-cigarette tax (binary), internet sales ban (binary), county-level flavor bans within state (binary), state-year COVID-19 stay-at-home order index, state-year COVID-19 death rates (per 100,000 population), and an indicator for Democratic state governor for the majority of the time between 2020–2022 (binary). Data were adjusted for sample weights.

*** p < 0.01; ** p < 0.05; * p < 0.1.

Figure 4. Synthetic difference-in-differences event study plots for young adults 18–24.

Note: Authors’ analyses of the PATH and BRFSS. All samples exclude Washington and Montana, if present, due to temporary flavor bans in these states; and California, if present, due to statewide flavored tobacco ban that went into effect in December 2022. Sample is restricted to ages 18–24 in all datasets. Figure presents the event study analysis of the synthetic difference-in-differences (SDID) estimates from Table 3. Shaded area represents 95% confidence interval around the point estimates based on the bootstrapping method. PATH SDID models include all treatment states except for Rhode Island. BRFSS SDID models include all treatment states. All models control for age, race/ethnicity (White/Black/Asian/Hispanic), sex (male), cigarette tax, e-cigarette tax (binary), internet sales ban (binary), county-level flavor bans within state (binary), state-year COVID-19 stay-at-home order index, state-year COVID-19 death rates (per 100,000 population), and an indicator for Democratic state governor for the majority of the time between 2020–2022 (binary). Data were adjusted for sample weights.

Table 4.

Meta-analysis estimates of the impact of flavor bans on e-cigarette and cigarette use across different age groups

	Panel A: Youth		Panel B: Young Adults 18–24		Panel C: Adults 25+
	DID	SDID	DID	SDID	DID	SDID
E-cigarette	−0.005	−0.014	−0.028 **	−0.026 **	−0.003	−0.001
SE	(0.009)	(0.014)	(0.011)	(0.012)	(0.003)	(0.004)
Cigarette	0.018 ***	0.017	0.034 ***	0.034 ***	0.000	−0.001
SE	(0.005)	(0.012)	(0.006)	(0.007)	(0.003)	(0.003)

Note: Authors’ analyses of the YRBS, PATH, MTF, and BRFSS. The outcome variables are dichotomous indicators of use in the past 30 days. We use the fixed effect method of meta-analysis to combine estimates from Table 3 whereby estimates are calculated by taking the weighted average of the estimates for each population and for each outcome. The weights are equal to the inverse of the standard errors squared, normalized to add up to one. The standard errors for each estimate are computed as the inverse of the square root of the sum of the non-normalized weights.

^***p < 0.01;

^**p < 0.05;

^*p < 0.01.

For youth (shown in Table 3, Panel A), we do not find any statistically significant effects of the bans on e-cigarette participation, for any of our datasets, using either the DID or SDID estimator. However, each data set in Table 3 does show a significant positive effect of flavor bans on cigarette use by youth, but only in the DID models. We also observe these patterns in the meta-analysis estimate shown in Table 4 (Panel A), where we find a significant increase in cigarette use of the magnitude 1.8 percentage points (pp). The DID event study plots show reasonable evidence of parallel trends in the pre-period for the YRBS and the MTF (Figures 2b and 2f), but a non-negligible violation for the PATH (Figure 2d). Furthermore, while the SDID point estimates are similar to the DID estimates in both Tables 3 and 4, they are not statistically significant. Given the mixed results between our DID and SDID models, and considering the pre-trend violations in some specifications, we conclude there is no consistent, robust evidence that statewide flavor bans significantly affected youth e-cigarette or cigarette use at the extensive margin.

Next, we explore effects for young adults (ages 18–24) using data from the BRFSS and the PATH (Table 3, Panel B). The BRFSS estimates of the ATT show a significant decrease in e-cigarette use and substitution into cigarette use in both the DID and the SDID models. The estimates indicate a 2.5 to 3.2 pp reduction (DID and SDID, respectively) in e-cigarette participation and a corresponding increase in cigarette participation of 3.5 pp. A causal interpretation of these estimates is strongly supported by the balanced trends in both the DID (Figure 3) and SDID (Figure 4) event study plots. It is important to note, however, that the BRFSS e-cigarette estimates leverage data from only five years (two pre- and three post-periods), as we do not have reliable information from 2018 to 2020.

Figure 3. Difference-in-differences event study plots for young adults 18–24.

Note: Authors’ analyses of the PATH and BRFSS. All samples exclude Washington and Montana, if present, due to temporary flavor bans in these states; and California, if present, due to statewide flavored tobacco ban that went into effect in December 2022. Sample is restricted to ages 18–24 in all datasets. Figure presents the event study analysis of the synthetic difference-in-differences (SDID) estimates from Table 3. Shaded area represents 95% confidence interval around the point estimates based on the bootstrapping method. PATH SDID models include all treatment states except for Rhode Island. BRFSS SDID models include all treatment states. All models control for age, race/ethnicity (White/Black/Asian/Hispanic), sex (male), cigarette tax, e-cigarette tax (binary), internet sales ban (binary), county-level flavor bans within state (binary), state-year COVID-19 stay-at-home order index, state-year COVID-19 death rates (per 100,000 population), and an indicator for Democratic state governor for the majority of the time between 2020–2022 (binary). Data were adjusted for sample weights.

Despite finding strong effects in the BRFSS, our PATH analysis for young adults reveals no statistically significant impact on either cigarette or e-cigarette use. While the PATH point estimates are sizable and its event studies suggest patterns similar to those in the BRFSS, the estimates lack precision due to large standard errors. This imprecision likely stems from the smaller sample sizes in the PATH – approximately half the size of the BRFSS sample for e-cigarette analyses and only a quarter of the size for cigarette analyses. The resulting statistical power limitations prevent us from detecting effects that may actually exist. Meta-analysis estimates shown in Table 4, Panel B provides support for this argument and confirms the findings of a significant decrease in the use of e-cigarettes (−2.8 to −2.6 pp, based on DID and SDID respectively), accompanied by substitution to cigarettes (3.4 pp).

Turning to the estimates for adults 25 and over, we find no extensive margin effects of flavor bans on either e-cigarette or cigarette use, regardless of estimation method or data source (Table 3, Panel C). The point estimates for this age group are also consistently small with narrower confidence intervals, providing more compelling evidence of null effects in this demographic. As expected, the meta-analysis estimates presented in Table 4, Panel C is in line with this conclusion.

5.2. Extensions (Details reported in Appendix A)

5.2.a. Other tobacco use outcomes

In this section, we explore whether the bans affected past 30-day use of other tobacco/nicotine products beyond e-cigarettes and cigarettes, as well as the dual use of e-cigarettes and cigarettes (Appendix Table A5).¹⁸ The associated event study plots are presented in Appendix Figures A5–A7 for SDID, and Figures A8–A10 for DID estimates. We do not find any noteworthy flavor ban spillover effects on other nicotine product or dual use for youth. Although we find some evidence of increases in either variable at the 10% confidence level for some specifications, none of these results are robust to both DID and SDID estimation. We therefore interpret these effects as suggestive at best.

Among young adults, for whom we found a significant decrease in e-cigarette use and an increase in cigarette use, we also find evidence of substitution into other nicotine use based on the BRFSS analyses. The findings from the DID and SDID models reveal that the use of other nicotine products increased by around 1.5 to 1.8 pp in this group. These results are further supported by evidence of parallel trends in the event study analyses. Considering the overall mean of 5% prevalence for these products in our study period, these effect sizes are significant. Turning to the PATH, we again do not find any meaningful effects.

For adults 25 and over, we find a slight reduction in other nicotine use in the BRFSS DID specification. This result, however, is not robust to the SDID estimation or the findings from the PATH. We therefore conclude that we do not find evidence of any significant spillover effects on this demographic.

5.2.b. Sensitivity to inclusion of covariates

Although the covariates can improve the match between the control and treatment groups, in Appendix Table A6, we present results from our main analyses without controlling for any covariates. This acts as an additional robustness test. For youth, as was the case in Table 3 which included covariates, we find null results for e-cigarette use. For cigarette use, however, DID estimates from all data sets demonstrate an increase in cigarette use, by 1.6 to 2.1 pp depending on the data source. On the other hand, these results are only robust for the PATH SDID model. We also find some suggestive evidence of other nicotine and dual use in some specifications, but only MTF dual use is consistent for both the DID and SDID estimators. All in all, the results without covariates for the youth do not divert significantly from our main findings.

For young adults, we find that our main results of significant negative effects for e-cigarettes, and significant positive effects for cigarettes and other nicotine using the BRFSS are robust to the exclusion of covariates. For adults 25 and over, we find significant declines for both e-cigarettes and other nicotine use, in both DID and SDID models using the BRFSS. Thus, the adult models without covariates do suggest stronger negative effects for e-cigarettes than the main models with covariates.

5.2.c. Heterogeneous effects by sex and race/ethnicity

Third, we assess the heterogeneity in the estimated treatment effects across various demographic sub-populations. Results by sex and race/ethnicity are presented in the Appendix Tables A7 and A8, respectively. Among youth, it is difficult to discern heterogeneity that is consistent or credibly supported across datasets and estimation methods. For young adults, using the BRFSS, there is more consistent evidence of stronger substitution effects into smoking and other nicotine use among whites and Hispanics. The effects are largely similar across gender, except we observe that males are driving the substitution effects of other nicotine use. We again do not find much evidence for this group using the PATH. Among adults 25 and over, where we had overall found no impact on their use of e-cigarettes or cigarettes in relation to the flavor bans, unpacking the estimates by gender and race/ethnicity continues to show no economically or statistically significant impacts on their tobacco use. We recognize that the estimation of heterogeneous treatment effects across these sub-populations is a noisy endeavor, and we view these patterns as suggestive rather than conclusive.

5.2.d. Additional Robustness tests

Finally, we also examine the effects of dropping one treatment state and check for interactions between the COVID-19 restrictions and flavor bans. These results are discussed in more detail in Appendix B. We find that excluding any one state generally does not change the sign and significance relative to our main results shown in Table 3 but does change the magnitudes of the coefficients. The Covid results suggest that the pandemic restrictions may have muted the impact of flavor bans on adults.

6. Discussion

Our study addresses a key challenge facing policymakers: how to regulate e-cigarettes in ways that both help adult smokers while preventing youth uptake and use. This balancing act has created varied approaches across the country. Our analysis provides important evidence leveraging the quasi-natural experiment of statewide flavor restrictions, helping to inform future policies that can better navigate these competing public health goals and avoid unintended consequences.

When a sample of current e-cigarette users were asked how they might respond to comprehensive flavor restrictions in nicotine vaping products, three modal responses emerged: 1) quit e-cigarette use; 2) not change their use of e-cigarettes; 3) substitute into cigarette use (Romm et al. 2022). Each of these scenarios has important implications for public health. We provide new and comprehensive evidence informing these scenarios and assessing how the statewide flavor bans affected youth, young adults, and adults with respect to their actual use of e-cigarettes and cigarettes.

We find evidence of a meaningful decline in e-cigarette participation, on the order of about two to three percentage points, among young adults ages 18–24; however, this decrease was offset by substitution into smoking. For youth, some of our analyses seem to suggest a similar pattern, including potential substitution into other tobacco/nicotine product use as well, though pre-existing trends and sensitivity of these estimates across data sets and methods make us cautious in attributing a causal interpretation; we therefore cannot rule out that the bans had little to no effect on adolescents’ cigarette or e-cigarette use. Turning to adults ages 25+, we do not find any evidence of impacts associated with the bans.

One implication of these results is that the statewide restrictions are still being circumvented, even if more comprehensive in scope compared to the federal ban. Support for this interpretation of the findings comes from the PATH, which shows that a substantial fraction of youth and young adult e-cigarette users continue to report using banned flavors even after the bans were implemented. Survey and anecdotal evidence point to various ways that e-cigarette users are able to bypass the restrictions: through online purchases, purchases from illicit sources, cross-border purchases, purchases from non-compliant retailers in the state, and users adding their own flavors to the e-liquid in vaping devices (Romm et al. 2022; Chen 2023; Yang et al. 2023a; Rich 2022). A more comprehensive federal ban on flavored e-cigarettes could be more effective in reducing flavored and overall e-cigarette use by shutting down some of these circumvention channels, for instance by deterring cross-border purchases or by enforcing retailer compliance. However, other alternatives to obtain flavored e-cigarettes may remain (black market, self-made flavorings) and may continue to moderate the effectiveness of further nationwide restrictions unless directly addressed.

A key challenge for any analysis of the recent statewide flavor bans, adopted over late 2019-mid 2020, is that these bans coincided with the beginning of the COVID-19 pandemic. While we were cautious in drawing causal interpretations in conjunction with evidence of balanced trends pre-policy adoption or pre-pandemic, as well as controlling for COVID-19 deaths and shutdown restriction intensity, we cannot rule out potential confounding bias arising from more complex interactions between the bans and the pandemic and from any heterogenous impact of the pandemic related shocks (economic, health, social distancing, school and business closures) across the various treated and control states. Nevertheless, we provide some evidence from two robustness checks that suggest that the interaction between the COVID-19 policies and the flavor bans have not been substantial.

Given the recency of the statewide bans, the treatment effects we estimate capture changes over a post-policy window of two to three years. Observing effects as additional years of data become available would be fruitful for assessing behavioral changes in tobacco use that may take further time to materialize; extending the post-policy window can also help further disentangle the confounding effects of the pandemic (which would be expected to fade over time) from any persistent direct effects of the bans.

These caveats notwithstanding, we note that the substitution effects into smoking, which we find among the young adult population, can generate substantial costs. During the sample period, the e-cigarette participation rate and the smoking participation for 18–24 year olds was about 16% and 12% respectively (BRFSS data from Table 3, Panel B). In this period, we find that the flavor bans decreased the e-cigarette participation rate by 2.5 pp and increased the cigarette participation by 3.5 pp. Given the standard error of these estimates, this is approximately an even swap from e-cigarettes to cigarettes. However, cigarettes are known to be more dangerous to health than e-cigarettes. Thus, there is a net negative effect on health for this age group. Such unintended consequences underscore the need to account for not only outcomes directly targeted by such restrictions but also potential spillovers into non-targeted outcomes for a more complete calculus of the potential costs and benefits of such policies.