E-CIGARETTE FLAVOR RESTRICTIONS’ EFFECTS ON TOBACCO PRODUCT SALES

Abstract

More than 385 US localities and 7 states have implemented restrictions on sales of flavored electronic nicotine delivery systems (ENDS), yet these policies’ effects remain unclear. Utilizing comprehensive data on tobacco product flavor policies linked to retail sales data from January 2018 through March 2023, we estimate the effects of these restrictions. Our findings reveal that ENDS flavor restrictions yield substantial decreases in total ENDS sales, primarily owing to significant declines in flavored ENDS sales alongside nonsignificant increases in unflavored ENDS sales. Further analyses find that ENDS flavor restrictions increase sales of combustible cigarettes, a more harmful product: 11 to 15 additional cigarettes are purchased for every 1 less 0.7 mL ENDS pod sold because of these policies. This uptick in cigarette sales stems primarily from non-menthol cigarettes and includes brands disproportionately used by underage youth. These findings suggest that the public health benefits from reducing ENDS sales via flavor restrictions may be offset by the public health costs of consequent increases in cigarette consumption.

I. Introduction

According to the 2014 Surgeon General’s Report, smoking is the leading cause of preventable disease and death in the United States, leading to more than 480,000 deaths each year (USDHHS 2014). In this context, the rise of electronic nicotine delivery systems (ENDS) has been met with both praise from those who view ENDS as a harm reduction tool that helps smokers move to a less harmful product, and criticism from those who consider ENDS an industry tool to hook a new generation of tobacco users.¹ Quasi-experimental research largely supports the harm reduction argument: An array of work links policies that make ENDS harder to access or more expensive to increased smoking among youth (Friedman 2015; Pesko et al. 2016; Pesko and Currie 2019; Dave et al. 2019; Pesko 2023; Pesko and Warman 2022; Abouk, Courtemanche, et al. 2023), young adults (Friedman and Pesko 2022), pregnant women (Abouk, Adams, et al. 2023), and adults (Pesko et al. 2020; Saffer et al. 2020). However, a demographic shift-share identification strategy suggests limited substitution between ENDS and cigarettes (Allcott and Rafkin 2022). Additionally, high rates of youth experimentation with ENDS raise concerns among some about ENDS inducing youth tobacco use (Committee on the Review of the Health Effects of Electronic Nicotine Delivery Systems et al. 2018; R. Sun et al. 2023).² As current evidence suggests that the health costs of combustible cigarette use (“smoking”) are substantially greater than those of ENDS use (Committee on the Review of the Health Effects of Electronic Nicotine Delivery Systems et al. 2018 Committee on Toxicity of Chemicals in Food, Consumer Products, and the Environment 2020; Goniewicz et al. 2017, 2020), promoting public health requires understanding how ENDS policies affect use of both products. The research presented here contributes to this knowledge base by studying a relatively new ENDS policy: flavor restrictions.

US federal law has prohibited sales of flavored cigarettes other than menthol since 2009, but this policy does not restrict flavors in other tobacco products like ENDS. In the absence of federal law, more than 385 US localities, 7 states, and Washington, DC, have restricted sales of non-menthol-flavored ENDS, with more than 260 localities, 5 states, and DC including menthol in these restrictions.³ Internationally, flavor restriction policies are under consideration in Canada and the UK, among other countries, and have been adopted in China, Hungary, and several Nordic and Eastern European countries (Government of Canada 2021; Triggle 2023; Physicians for a Smoke-Free Canada 2024). Thus, clarifying ENDS flavor policies’ effects on ENDS and cigarette use is critical to inform public health policy in the US and internationally.

Henceforth, we refer to such policies (both limiting flavor sales to specific types of retailers and full prohibitions) as “ENDS flavor restrictions,” implicitly categorizing tobacco-flavored ENDS as “unflavored,” along with truly flavorless products.

In this paper, we match a novel data set on state and local flavor policy coverage to retail sales data, then apply two-way fixed effects to estimate how restrictions on flavored ENDS sales affect per capita volume sales of ENDS and cigarettes, both overall and separately by flavor. Given staggered policy adoption, robustness checks repeat the analyses using de Chaisemartin and D’Haultfœuille’s (2024) approach to ensure estimates are not driven by bias due to heterogenous effects.

While policy makers may pursue flavor restrictions as a means to reduce ENDS use, it is not clear that regulating a subset of these products—flavored ENDS—will decrease overall ENDS sales. Flavor prohibitions can be represented as a leftward shift in the supply curve for flavored ENDS, reducing quantity sold and increasing flavored ENDS prices if demand remains constant. People could potentially still access flavored ENDS through illicit sources, traveling to unrestricted jurisdictions, or self-flavoring ENDS products, but at higher cost and/or inconvenience. Concurrently, demand in the unflavored ENDS market-place may shift rightward depending on substitutability between flavored and unflavored ENDS. The supply of unflavored ENDS could shift rightward as well, if suppliers can quickly convert from flavored to unflavored ENDS production. As both shifts should increase unflavored ENDS consumption, restricting flavored ENDS sales as a means to reduce overall ENDS consumption hinges on the decrease in flavored ENDS sales (because of the leftward shift in its supply curve) being larger than the increase in unflavored ENDS sales (stemming from potential rightward shifts in the supply and demand curves).

Reduced-form analyses estimating ENDS flavor restrictions’ effects on ENDS sales—both overall and by flavor—provide a natural test of these relationships. As these policies do not occur in isolation, however, explicitly controlling for other tobacco policies is critical, including concurrent restrictions on flavors in cigars and menthol cigarettes.

If flavored and unflavored ENDS are not perfectly substitutable, restrictions on flavored ENDS could have unexpected consequences for the cigarette market. These consequences may vary across age groups, depending on their likelihood of using flavored ENDS, their readiness to switch to unflavored options, and their willingness to access flavored ENDS through illicit channels. Consumer willingness to resort to self-flavoring or travel to jurisdictions where such products are legally sold may also vary by age. Consequently, we also analyze the unintended impacts of ENDS flavor restrictions on cigarette sales, both in total and by age group.

Our study has several strengths. The first is that we have assembled a hand-collected, comprehensive database of all US state and local tobacco flavor laws nationwide. We identified almost 500 flavor policies, compiling a list of jurisdictions with flavor restrictions from multiple resources, including two lists updated quarterly by prominent advocacy groups in this space (Truth Initiative 2023; Bach 2023). The text of all corresponding laws and ordinances were then identified through LexisNexis, Municode Library, online searches, and, where necessary, reaching out to local authorities. From these, we coded data on products covered (ENDS, cigarette, cigar), passage dates, legislated effective and enforcement dates, retailer exemptions, and flavors covered (e.g., menthol) for each policy prohibiting or limiting sales of flavored tobacco or nicotine products.⁴ In addition to providing superior data for studying ENDS flavor restrictions, this detailed information is used to control for enactment periods as well as concurrent cigarette and cigar flavor policies, which are important potential confounders often adopted concurrently with ENDS flavor restrictions. Thus, the richness of our data allows us to reduce several otherwise salient sources of omitted variable bias.

A second strength is our application of this high-quality policy data to extremely recent sales data with broad geographic coverage. Specifically, instead of using Nielsen’s store-level retail scanner data obtained through the Kilts Center at the University of Chicago, we purchased Information Resources Incorporated’s (IRI) retail sales data covering 44 states and the vast majority of US residents through March 2023 (Liber et al. 2024). When purchased, IRI’s data included at least two more-recent years of retail sales data than were available through Kilts, ending in March 2023 instead of December 2020. This allowed our effect estimates to capture impacts of policies implemented in 99 jurisdictions after 2020 as well as longer-run responses to more than 100 policies implemented in 2020, for which the Nielsen data available at that time provided less than a year of follow-up. Moreover, the IRI data include only states with sufficient retail sales coverage to project sales to the state level in each period, yielding a sample of 44 states. In contrast, state representativeness was a tangible concern with the Kilts Center’s store-level data, as Nielsen’s state-level sales data set covered only 23 states, suggesting a broader representativeness issue in other states’ underlying store-level data. Consequently, IRI’s data appeared more appropriate for two-way fixed-effects estimation of flavor policies’ effects.

A third strength of our paper involves the use of data on the consumer age distribution of different cigarette brands and flavors to assign each product to one of three distinct age profiles, and assess effects by age group. To the best of our knowledge, this technique has not previously been used in the economics literature.

Our study estimates the effects of restricting flavored ENDS sales on cigarette and ENDS volume sales per capita, using two-way fixed effects and a new estimator by de Chaisemartin and D’Haultfœuille (2024) designed to correct for bias due to dynamic and heterogenous treatment effects. We find that ENDS flavor restrictions yield substantive reductions in ENDS sales driven by decreases in flavored ENDS sales, but also increases in cigarette sales, both overall and for brands disproportionately used by youths. Our top-line estimate is that 11 to 15 additional cigarettes are purchased for every 1 less 0.7 mL ENDS pod sold because of flavor restrictions.

II. Background

In the US, ENDS, cigarettes, cigars, smokeless tobacco, pipe tobacco, and any other product that is derived from tobacco plants or contains nicotine are regulated as “tobacco products” by the US Food and Drug Administration (FDA). The FDA’s jurisdiction over these products imposes pre–market entry requirements, unique decision-making standards, and strict youth access rules. Yet FDA evaluation of premarket tobacco product applications to determine whether specific ENDS products are suitable for the protection of public health began relatively recently: The application deadline for industry submissions was September 9, 2020. The FDA began enforcement actions targeting sales of non-mentholated flavored ENDS cartridges before this, in February 2020. Yet efforts have been made to skirt these rules, including by using synthetic nicotine (Stephenson 2022). Prior to 2024, FDA had authorized only 23 ENDS products for marketing, none of which were flavored, yet flavored ENDS products remained widely available in states that did not prohibit their sales (Le et al. 2023). This reflects both ongoing sales as manufacturers await FDA marketing orders or appeal marketing denial orders, as well as importers and retailers continuing to source flavored products in spite of FDA enforcement efforts.

Evidence from discrete choice experiments points to flavors being important drivers of consumer demand for ENDS (Buckell et al. 2019; Kenkel et al. 2020). Flavored products dominate this market: One study using Nielsen retail scanner data from 2013 to 2019 finds that 38.7 percent of e-cigarette liquid volume is sold unflavored, 21.9 percent is mentholated, and 39.4 percent contains a non-menthol flavor (Cotti et al. 2022). However, the proportion of US ENDS sales that are flavored may be even higher, as those data omit vape shops, which often sell a large variety of flavors.

While economists have long been interested in cigarette flavor restrictions’ effects on use of traditional tobacco products (Carpenter and Nguyen 2021; Courtemanche et al. 2017), impacts of newer ENDS flavor restrictions remain unclear. Several papers in the public health and clinical literature provide early evidence regarding ENDS flavor restrictions’ relationship to tobacco product sales and use (see Cadham et al. [2022] for a review), but few provide causal evidence. Prior difference-in-differences analyses largely lack sufficient pre-period data to test the parallel trends assumption or simply fail to do so (Hawkins et al. 2022; Olson et al. 2022; Kingsley et al. 2019, 2021). Others compare pre-trends but do not run event studies (Dove et al. 2023; Ali et al. 2022).

Three analyses consider broader policies prohibiting sales of flavored ENDS as well as other tobacco products, and thus cannot disentangle the different prohibitions’ effects. Specifically, Friedman (2021) and Gammon et al. (2022) use difference-in-differences and interrupted time series approaches respectively to assess effects of San Francisco’s prohibition on all flavored tobacco product sales, but cannot separate the ENDS flavor ban’s effects from that of prohibiting flavored cigar and menthol cigarette sales. Liber et al. (2023) consider Massachusetts’s emergency ban on all ENDS sales (regardless of flavor) during the 2019 outbreak in the US of e-cigarette or vaping-associated lung injuries (EVALI), which was responsible for 68 deaths and thousands of hospitalizations, and erroneously attributed to ENDS use. While they find increased cigarette sales, those estimates reflect the removal of all ENDS products from the legal market—not just flavored ENDS—as well as EVALI-influenced increases in perceptions of risks from ENDS use (Dave et al. 2020; Kreslake et al. 2022).

Unlike the aforementioned studies, our current paper focuses on estimating causal effects of ENDS flavor restrictions specifically. To the best of our knowledge, there is only one peer-reviewed study, funded by the e-cigarette manufacturer Juul, that uses variation across multiple states to provide plausibly causal evidence for the effect of ENDS flavor restrictions on both ENDS and cigarette sales or use (Chen et al. 2024).⁵ That study evaluates the effect of four states’ flavored ENDS sales prohibitions initially implemented as temporary emergency responses to the EVALI outbreak. The authors find evidence linking these prohibitions to decreases in sales of ENDS refills and increases in cigarette sales, as well as evidence for spatial spillovers between adopters and the nonadopting states with which they share a border. However, estimates from the Centers for Disease Control and Prevention indicate that omitting disposable ENDS from retail sales data analyses would excise a substantial portion of ENDS sales—about a quarter of ENDS unit sales covered by retail sales data sets in January 2020, rising to half of sales by December 2022 (Ali 2023)—raising questions about those findings. In contrast, our current study substantially improves upon this prior work by (1) using data through March 2023, thus making our results generalizable to both the post-EVALI and COVID time periods and reflecting longer-term policy effects rather than very short-term effects; (2) covering both state and local laws across 16 states (compared with four state laws); (3) not limiting ENDS sales analyses to data on refills; (4) using new methods accounting for heterogenous treatment effects; in particular, an estimator developed by de Chaisemartin and D’Haultfœuille (2024); (5) studying the effect of ENDS flavor restrictions on age-disproportionate cigarette sales; and (6) avoiding commercial conflicts of interest among the authorship team, potentially increasing the political reach of this work.

III. Methods

A. DATA

Information Resources Incorporated’s (IRI) retail sales data provide ENDS and cigarette sales information for 44 states—all but Alaska, Delaware, Hawaii, Montana, New Jersey, and North Dakota—covering four-week intervals from January 2018 through March 2023. Passively collected at the universal product code (UPC) level from barcode scanners at mainstream brick-and-mortar retailers, these data include dollar and unit sales, product names, categories, and volumes, which were used to generate aggregate ENDS and cigarette volume sales for each four-week period.

A.1. OUTCOME VARIABLES.

Primary outcomes are per capita volume sales of ENDS (standardized to 0.7 mL units, henceforth “pods”)⁶ and cigarettes (packs) in each state-by-four-week period, overall and by flavor type (tobacco/unflavored, menthol-only flavors, and flavors other than menthol). Cigarette sales data were also stratified by consumer age profiles to separately consider per capita sales of youth-disproportionate products, age-proportionate products, and adult-disproportionate products, using categories calculated and published by Liber et al. (2022) based on data from the 2014 to 2019 National Survey on Drug Use and Health (NSDUH). Specifically, the authors classified products whose share of consumption by persons under age 21 was significantly higher than the remainder of the market’s underage share as “disproportionately young” (e.g., Marlboro Menthol).

Those with significantly lower underage shares were “disproportionately old” (e.g., Virginia Slims Menthol), while products where this difference was not statistically significant were “age-proportionate” (e.g., Camel Regular). As NSDUH did not collect ENDS brand choice data during this period, a parallel analysis of ENDS sales by consumer age-group was not feasible.

A.2. FLAVOR POLICY DATA AND VARIABLES.

We compiled data on the location, passage dates, legislated effective dates, and retailer exemptions for all US state and local policies that prohibit or limit sales of flavored tobacco or nicotine products. Policy location lists were first collated from the Campaign for Tobacco-Free Kids, the Public Health Law Center at Mitchell Hamline School of Law, Truth Initiative, and the American Nonsmokers’ Rights Foundation (Bach 2023; Public Health Law Center 2024; Truth Initiative 2023; ANRF 2024). These lists were expanded based on policies found through news searches, state publications, and local municipal websites.

Each law, bylaw, regulation, or ordinance was then reviewed to determine the date of passage, legislated effective date, and enforcement date (if the policy states one), as well as which products were affected—ENDS, cigarettes, cigars, hookah, smokeless tobacco—and whether the law restricted menthol flavors for each of those products. Given extensive variation in flavor policies nationwide, additional variables describe retailer exemptions, product exemptions, and other notes (e.g., if a policy had different effective dates for menthol restrictions versus other flavors, this fact and corresponding dates would be given in one of the notes variables). For county-level policies, we also noted whether the policy covered the entire county or only unincorporated areas therein.⁷ Repeal and end dates were also included where relevant, along with indicators for temporary policies passed as emergency actions in response to the 2019 EVALI outbreak. Where sources contradicted each other and primary sources did not clarify (i.e., the text of the legislation/ordinance), local authorities were contacted for clarification. (See Online Appendix for further details.)

To calculate the proportion of state residents covered by each type of flavor restriction, policy data were matched to local and state population sizes from the 2020 census (US Census Bureau 2022). The proportions of residents covered by restrictions on sales of flavored ENDS, flavored cigars, and menthol cigarettes were calculated for each state-by-date.⁸ These values were then averaged across each four-week period in our sales data to generate average coverage for each sales period by state.

A.3. COVARIATES.

Policy covariates adjust for other flavor restrictions, non-flavor tobacco policies, and policies likely to impact the price or availability of key substitutes/complements. Specifically, as some ENDS flavor policies also restrict sales of flavored cigars or menthol cigarettes, covariates adjust for the average proportion of state residents covered by each policy type, over each four-week period. Given evidence of anticipatory responses to tobacco product taxes (i.e., stocking up before a tax increase takes effect), we also calculate the average proportion of state residents in the enactment period—between a law’s passage and effective dates (henceforth, the “interim period”)—separately for flavored ENDS, flavored cigar, and menthol cigarette sales restrictions.⁹

As 46 jurisdictions prohibited all ENDS sales regardless of flavor at some point in our analytic period (see Online Appendix Table A1), a covariate also captures the proportion of a state’s population covered by prohibitions on unflavored ENDS sales. Where all ENDS sales are prohibited, this unflavored ENDS sales prohibition variable and the ENDS flavor restriction variable both equal 1, such that the unflavored ENDS sales prohibition coefficient estimates the added impact of prohibiting all unflavored ENDS sales over and above the flavor restriction’s effect. Adding those two coefficients gives the estimated impact of prohibiting all ENDS sales (flavored and unflavored).

Additional covariates are calculated across the four-week sales period as above when the underlying data were available by date or week, and linked based on the month (year) at the start of this sales period when underlying data were provided by month (year). Further tobacco policy covariates calculated as averages across each four-week period include state tax rates for cigarettes, cigars, and ENDS (Centers for Disease Control and Prevention 2023). As our tobacco-21 and clean air data are based on population coverage on the first of the month, the proportion of residents covered by state or local laws prohibiting sales to those under 21 years old, comprehensive smoke-free worksite laws, and vape-free worksite laws are based on the month each four-week period began (Colston 2022; Centers for Disease Control and Prevention 2023; Seidenberg et al. 2024; ANRF 2023). To address effects of policies targeting tobacco product substitutes or complements, additional policy covariates include the proportion of each sales period covered by state medical and recreational cannabis legalization, and annual beer tax rates.

Finally, environmental controls address state socioeconomics (state median household income by year, seasonally adjusted monthly unemployment rates), seasonality (total heating degree days during each four-week period [NCEI 2021]), and potential effects of both the COVID-19 pandemic (state deaths per capita during the sales period, the proportion of that period with nonessential business closures in place [Raifman et al. 2022]), and the 2019 outbreak of vaping-associated lung injuries (state EVALI deaths by the start of the month [Liber et al. 2023]; proportion of the period through March 1, 2020, where a state’s EVALI-era ban on flavored ENDS sales was blocked or stayed by courts, separately for each affected state: New York, Michigan, Oregon, and Utah). See the Online Appendix for further details on covariate construction and underlying data.

B. METHODS

Given widespread familiarity with two-way fixed-effects (TWFE) methods, our first analyses use multivariable linear regressions to estimate TWFE specifications assessing how the proportion of a state covered by ENDS flavor restrictions averaged over each four-week sales period $\left(Fl{v}_{st}\right)$¹⁰ relates to ENDS and cigarette sales per capita $\left(Y_{st}\right)$:

$$Y_{st}={\beta }_0+{\beta }_{1}Fl{v}_{st}+\mathit{\lambda }\overrightarrow{{\mathit{X}}_{\mathit{s}\mathit{t}}}+{\mathit{\gamma }}_{\mathit{s}}+{\mathit{\delta }}_{\mathit{t}}+{\epsilon }_{st}$$ (1),

where $s$ indexes state and $t$ indexes each four-week period. Sales period fixed effects $\left({\mathit{\delta }}_{\mathit{t}}\right)$ absorb national time trends in the outcome, while state fixed effects $\left({\mathit{\gamma }}_{\mathit{s}}\right)$ adjust for time-invariant differences between states. Time-varying covariates $\left(\overrightarrow{{\mathit{X}}_{\mathit{s}\mathit{t}}}\right)$—that is, the tobacco policy covariates, other policy covariates, and environmental covariates described previously—are added as controls. Standard errors are clustered by state, the level of policy variation assessed here (Abadie et al. 2017).

Robustness checks drop the first 12 months of the COVID-19 pandemic—defined as sales periods overlapping March 11, 2020 (when the World Health Organization declared the outbreak a pandemic) through March 10, 2021—and population weight the full covariate analyses. As prior work finds evidence of spatial spillover effects from state flavor restrictions and the temporary 2019 prohibition on all ENDS sales in Massachusetts (Chen et al. 2024), the baseline specification is repeated with two additional covariates to assess cross-border spillovers: one indicator for states bordering those with a statewide ENDS flavor restriction that do not themselves restrict flavored ENDS sales, and a second for those bordering Massachusetts during its temporary statewide ban on all ENDS sales—flavored and unflavored—in late 2019.

Given concerns about bias in TWFE analyses with staggered implementation, our preferred approach applies de Chaisemartin and D’Haultfœuille’s (2024) estimator, which is robust to biases stemming from dynamic or heterogenous treatment effects (henceforth “DCDH [2024]”). We use a binary exposure variable capturing state-level flavor restrictions in effect for the entire four-week period $\left(BinFl{v}_{st}\right)$ alongside a continuous variable capturing the proportion of state residents subject to incomplete flavor policy coverage, whether due to local policies only or a state law coming into effect partway through the four-week period (i.e., $FlvU{1}_{st}=Fl{v}_{st}$ if $Fl{v}_{st}<1$, and 0 otherwise):¹¹

$$Y_{st}={\beta }_0+{\beta }_{1}BinFl{v}_{st}+{\beta }_{2}FlvU{1}_{st}+\mathit{\lambda }\overrightarrow{{\mathit{X}}_{\mathit{s}\mathit{t}}}+{\mathit{\gamma }}_{\mathit{s}}+{\mathit{\delta }}_{\mathit{t}}+{\mathit{\epsilon }}_{\mathit{s}\mathit{t}}$$ (2).

We use DCDH (2024) as our preferred estimator for equation 2 because it both allows controls for area-level time-varying variables and does not require monotonicity in the exposure (e.g., treatments are reversible), which is useful in our situation as several flavor policies were repealed or passed as temporary measures in response to the 2019 outbreak of vaping-associated lung injuries.

DCDH (2024) offers unique flexibility in handling non-absorbing treatments and complex treatment paths compared with other available estimators.¹² The DCDH (2024) estimator can be used for both binary, absorbing treatments and nonbinary treatments that may increase or decrease multiple times. It is designed to account for the effects of both current and past treatments, even if these effects vary across locations or over time. This estimator relies on the assumptions that there is no anticipation of treatment effects and that the treated and control groups would have followed parallel trends in the absence of treatment. It computes the average difference-in-differences (DID) estimates for groups whose treatment status changes (whether increasing or decreasing), comparing their outcomes with those of similar groups whose treatment status remains unchanged. The method also includes placebo tests to provide suggestive evidence regarding the parallel trends assumption. A dynamic event study framework meanwhile tracks and aggregates the effects of entering, exiting, and reentering treatment over multiple periods, something that other estimators designed for monotonic treatment paths cannot handle as effectively. The DCDH (2024) estimator chooses control groups by comparing the outcome evolution of units whose treatment status does versus does not change. Ensuring that these control groups have the same baseline treatment status as the treated groups maintains comparability, in order to isolate the effect of the treatment itself. For further reference, please see Section 3.3 in de Chaisemartin and D’Haultfœuille (2023).

To provide suggestive evidence on the parallel trends assumption, we add leads and lags of the continuous treatment to the TWFE continuous treatment analysis, and apply DCDH (2024) to generate event studies that are robust to dynamic and heterogenous treatment effects in assessing binary state-level flavor policies. We specify our DCDH estimation such that pre- and post-period coefficients in our DCDH event studies have the same general interpretation as in a TWFE event study for purposes of evaluating parallel trends (Roth 2024). Critically, these are best interpreted as the impact of a state policy in effect for the full four-week sales period, since the binary ENDS flavor restriction variable equals 0 for policies adopted in the middle of a given period. Thus, controlling for partial coverage (i.e., $FlvU{1}_{st}$) is vital here, to absorb the impact of state policies in effect for only a portion of a given sales period.

Robustness checks repeat the DCDH analyses without the first 12 months of the COVID-19 pandemic, and with population weights. To ensure that a single state’s unobserved state-specific policy details are not driving our findings, we also reestimate our DCDH analyses sequentially dropping states with at least 50 percent of their populace covered by flavor restrictions at any point in the analytic period, one at a time.

Next, the full variable specification is repeated for each product by flavor, to consider whether observed changes in ENDS sales are consistent with a flavor restriction driving these effects. These analyses also clarify whether effects on cigarette sales are specific to menthol cigarettes and thus might be addressed through a menthol cigarette sales ban.

To clarify whether cigarette sales responses are likely to be limited to a particular age group, cigarette sales analyses are repeated separately for sales of youth-disproportionate, age-proportionate, and adult-disproportionate products, as described above.

Finally, as DCDH analyses do not allow for multiple treatment variables, exploratory TWFE regressions assess how policy effects differ between ENDS flavor prohibitions versus limitations that include at least one retailer exemption (e.g., for adult-only tobacco retailers or smoking bars). Specifically, an exploratory analysis separates our continuous flavor restriction variable into the proportion of a state covered by each type of policy. This analysis focuses on cigarette sales outcomes as the vast majority of cigarette sales are covered in IRI’s data (Liber et al. 2024). As retail scanner data coverage of ENDS sales appears to be closer to 50 percent of volume sales—likely because these data sets do not cover online sales or specialty stores like vape shops—estimates of flavor limitations’ effects on ENDS sales could overstate market-level changes if consumers respond by shifting their sourcing in accordance with the law (e.g., from IRI-covered convenience stores towards adult-only specialty shops not included in IRI’s data).

IV. Results

A. DESCRIPTIVE ANALYSES

Figure 1 depicts changes in per capita cigarette and ENDS sales from January 2018 through March 2023 for the 44 states included in our data. These show broad declines in cigarette sales with seasonal increases in the summer, alongside steadier increases in ENDS sales, dominated by menthol and non-menthol-flavored ENDS. More specifically, summary statistics in Table 1 show that the average person purchased an average of 2.635 packs of cigarettes per four-week period, 69 percent of which were non-menthol. These included 0.703 packs of products disproportionately purchased by youth, 0.632 packs disproportionately purchased by adults, and 1.3 packs of products that were fairly proportionately purchased across ages, based on the consumer age distributions reported in Liber et al. (2022). Cigarette sales per capita declined 30 percent across the analytic period.

FIGURE 1.

Per capita cigarette and ENDS sales, January 2018–March 2023. This figure plots average per capita cigarette and ENDS sales by flavor, based on IRI retail sales data by state and four-week periods for January 2018 through March 2023. Cigarette volume sales are measured in packs, while ENDS volume sales are in 0.7 mL units. Short vertical dashes along the bottom indicate per capita sales of cigarettes and ENDS with unknown flavors, in light gray and dark gray, respectively.

TABLE 1.

Summary statistics, means/[standard deviations]

	Full sample (1)	Jan. 8–Feb. 4, 2018 (2)	Feb. 27–Mar. 26, 2023 (3)	t-test: col. 2 = 3 (4)
ENDS sales/capita	0.482 [0.372]	0.138 [0.064]	0.793 [0.519]	p < 0.001
Tobacco or unflavored ENDS sales/capita	0.100 [0.063]	0.048 [0.021]	0.113 [0.059]	p < 0.001
Menthol ENDS sales/capita	0.223 [0.159]	0.046 [0.027]	0.313 [0.175]	p < 0.001
Non-menthol-flavored ENDS sales/capita	0.157 [0.246]	0.044 [0.024]	0.357 [0.399]	p < 0.001
Unknown-flavor ENDS sales/capita	0.002 [0.017]	0.00001 [0.00001]	0.010 [0.051]	p = 0.20
Cigarette sales/capita	2.635 [1.166]	2.918 [1.190]	2.041 [0.904]	p < 0.001
Menthol cigarette sales/capita	0.823 [0.435]	0.893 [0.429]	0.624 [0.344]	p = 0.002
Tobacco cigarette sales/capita	1.811 [0.862]	2.025 [0.907]	1.417 [0.683]	p < 0.001
Unknown-flavor cigarette sales/capita	0.0002 [0.001]	0.001 [0.002]	0.00001 [0.00003]	p = 0.07
Youth-disproportionate cigarette sales/capita	0.703 [0.367]	0.790 [0.383]	0.490 [0.247]	p < 0.001
Age-proportionate cigarette sales/capita	1.300 [0.553]	1.437 [0.567]	1.025 [0.412]	p < 0.001
Adult-disproportionate cigarette sales/capita	0.632 [0.406]	0.691 [0.408]	0.526 [0.385]	p = 0.06
Proportion of state restricting flavored ENDS sales	0.091 [0.262]	0.021 [0.079]	0.150 [0.345]	p = 0.02
Proportion of state prohibiting flavored ENDS sales	0.056 [0.210]	0.0003 [0.001]	0.104 [0.290]	p = 0.03
Proportion of state limiting flavored ENDS sales1	0.036 [0.158]	0.020 [0.079]	0.046 [0.192]	p = 0.44
Proportion of state prohibiting unflavored ENDS sales	0.002 [0.030]	0.00003 [0.0002]	0.002 [0.009]	p = 0.27
Proportion of state restricting flavored cigar sales	0.068 [0.208]	0.057 [0.184]	0.090 [0.263]	p = 0.51
Proportion of state restricting menthol cigarette sales	0.022 [0.120]	0.001 [0.009]	0.052 [0.212]	p = 0.13
$N$	2,917	40	44

Note: ENDS sales are measured in 0.7 mL units based on their liquid volume, to standardize measurement across product sizes. Cigarette sales are measured in standard 20-cigarette packs.

¹“Limitations” on flavored ENDS sales refers to policies that restrict sales but allow some type of retailer exemption (e.g., for smoking bars, 21 + tobacco retailers).

Over the same period, the average person purchased 0.482 standardized (0.7 mL) units of e-liquid, with mean sales rising from 0.138 in the first sales period (beginning January 8, 2018) to 0.793 in the last (beginning February 27, 2023). Indeed, ENDS sales per capita increased 475 percent from January 2018 through March 2023. By the end of the study period, 39 percent of ENDS sales were menthol-only flavors while 45 percent were non-menthol-flavored products, with 14 percent tobacco or unflavored. We were unable to identify about 1 percent of ENDS sales’ flavors in the last sales period, versus 0.2 percent on average across the entire sample.

Over the course of our analytic sample, the average percentage of state residents covered by a state or local ENDS flavor restriction rose from 2.1 percent in January 2018 to 15 percent in March 2023. California, District of Columbia, Massachusetts, Maryland, Montana, New Jersey, New York, Rhode Island, Utah, and Washington had state-/district-wide policies restricting or prohibiting flavored ENDS sales for at least one full four-week period during our analytic period, although Montana, New Jersey, and DC are not covered in our sales data. Plotting within-state changes in ENDS flavor policy coverage over time separately for states that did versus did not adopt a state-level policy in our analytic period, Online Appendix Figure A1 shows that most of our identifying variation occurs within a relatively narrow window (mid-2019 to mid-2020).

The 15 percent average percentage of state residents covered by an ENDS flavor policy in March 2023 reflects two distinct versions of these policies: ENDS flavor prohibitions (10.4 percent) versus less restrictive ENDS flavor restrictions that allow certain retailer exemptions, such as for smoking bars or 211 tobacco retailers (4.6 percent). Notably, coverage in March 2023 does not capture temporary flavor policies adopted in late 2019 in response to the US outbreak of vaping-associated lung injuries, as those were stayed, expired, or replaced with permanent policies before April 2020.

Across the entire study period, states averaged 6.8 percent and 2.2 percent of residents covered by flavored cigar and menthol cigarette sales restrictions, respectively, while 0.2 percent were subject to prohibitions on sales of unflavored ENDS as well as flavored (see Table 1).

B. MAIN RESULTS

Figure 2 presents two-way fixed-effects estimates using equation 1 for how the proportion of state residents covered by ENDS flavor policies relates to ENDS and cigarette sales per capita. Groups of covariates are sequentially added across specifications to clarify each group’s influence on estimates of the effect of ENDS flavor policies on outcomes. We first show that the effect of ENDS flavor restrictions on ENDS and cigarette sales is fairly consistently estimated regardless of controls. With full controls, we find that 100 percent coverage leads to a 0.386 reduction in ENDS volume sales per capita (p < 0:01), and a 0.259 increase in cigarette pack sales per capita (p < 0:01). Results are similar with larger point estimates and overlapping confidence intervals when omitting the first year of the COVID-19 pandemic: 20.501 and 0.354 respectively, both with p < 0:01. Population weighting the full control analysis yields slightly smaller estimates with confidence intervals overlapping the unweighted version. Coefficient estimates for prohibitions on all ENDS sales (i.e., regardless of flavor) are in the same direction as ENDS flavor restriction coefficients for all specifications, reinforcing that ENDS and cigarettes are economic substitutes (Online Appendix Table A2).

FIGURE 2.

Two-way fixed-effects estimates of ENDS flavor restrictions on ENDS and cigarette sales per capita. Coefficients and 95% confidence intervals are estimates of the relationship between the percentage of a state subject to restrictions on flavored ENDS sales and per capita volume sales of ENDS or cigarettes, as indicated on the left above in bold, based on IRI retail sales data by state and four-week periods for January 2018 through March 2023. Estimates come from multivariable linear regressions of two-way fixed-effects specifications, with covariates added sequentially to clarify their contribution to potential confounding in more naïve specifications. Each estimate is generated by a different regression specification in all cases. The “Basic Covariate Specification” adjusts for state and date fixed effects as well as a range of other tobacco product flavor restriction variables. The second specification—“+ Tobacco Control Covariates”—adds additional nicotine and tobacco policy controls to the basic specification; “+ = All Policy Covariates” further adjusts for policy covariates affecting likely tobacco product substitutes or complements; and “+ Environmental Covariates” further adds environmental controls (e.g., state socioeconomic indicators, deaths from COVID-19). See the Online Appendix for details on all specific covariates in each specification. Robust standard errors are clustered by state. Online Appendix Table A2 gives output in tabular form. ** p < 0.01, * p < 0.05.

Given evidence of spatial spillover effects from state flavor restrictions during the outbreak of vaping-associated lung injuries (Chen et al. 2024), Table 2 repeats the baseline specification with two additional covariates adjusting for states that do not themselves restrict flavored ENDS sales but border a state with a statewide ENDS flavor restriction, and states bordering a state that bans all ENDS sales (i.e., Massachusetts in late 2019). Coefficients on these covariates are statistically nonsignificant for both outcomes, with varying signs. As their inclusion also appears to inflate the ENDS flavor restriction coefficient, we take the more conservative approach and omit these covariates in subsequent analyses.

TABLE 2.

Cross-border spillovers from ENDS flavor restrictions, coefficient/(standard error)

	ENDS volume sales per capita		Cigarette packs sold per capita
	(1)	(2)	(3)	(4)
Proportion of state restricting flavored ENDS sales	−0.386a (0.104)	−0.419a (0.112)	0.259a (0.083)	0.294a (0.084)
	0.001	0.001	0.003	0.001
Border state has ENDS flavor restriction		−0.107 (0.058)		0.117 (0.073)
		0.070		0.117
Border state prohibits all ENDS sales		0.002 (0.182)		−0.121 (0.071)
		0.991		0.097
Proportion prohibiting unflavored ENDS sales (${\beta }_2$)	−0.312a (0.083)	−0.310a (0.086)	0.124 (0.096)	0.116 (0.100)
	0.001	0.001	0.205	0.251
Proportion of state restricting flavored cigar sales	0.756 (0.450)	0.878 (0.457)	0.321 (0.314)	0.190 (0.333)
	0.100	0.061	0.313	0.572
Proportion of state restricting menthol cigarette sales	−0.957a (0.315)	−1.030a (0.314)	0.097 (0.235)	0.176 (0.243)
	0.004	0.002	0.682	0.473
Fixed effects	State & date	State & date	State & date	State & date
$N$	2,917	2,917	2,917	2,917
Adjusted $R^2$	0.785	0.787	0.978	0.979
Mean (Y var.)	0.482	0.482	2.635	2.635

Note: Multivariable linear regressions use 2018-March 2023 IRI retail sales data for four-week periods by state to estimate ENDS flavor restrictions’ relationship to per capita volume sales of cigarettes and ENDS, adjusting for all covariates in our baseline specification (see Online Appendix Table A2, columns 4 and 10) as well as potential cross-border spillovers from states’ restrictions on flavored ENDS sales and prohibitions on all ENDS sales. Robust standard errors are clustered by state.

^ap < 0.01.

Using the main specification including all controls and without dropping the first year of COVID-19, the ratio of these coefficients suggests 13 additional cigarettes sold (20 cigarettes per pack × 0.259 packs / −0.386 pods) for every 1 less ENDS pod sold. Accounting for unequal shares of total ENDS and cigarette sales captured in the IRI data,¹³ as in prior work (Cotti et al. 2022; Allcott and Rafkin 2022), we calculate an adjusted substitution rate of approximately 11 cigarettes sold for every 1 fewer cigarette-pack-equivalent ENDS pod sold.

Figure 3 shows DCDH average treatment effects (ATEs) for state policies. Estimated effects are similar to TWFE results in Figure 2 and statistically significant in all but two specifications. Full covariate specifications indicate an adjusted rate of approximately 15 additional cigarettes purchased for every 1 less ENDS pod sold owing to state ENDS flavor restrictions (20 cigarettes per pack × 0.258 packs / −0.293 pods × 0.84 adjustment factor).

FIGURE 3.

DCDH estimates of ENDS flavor restrictions’ effects on ENDS and cigarette sales per capita. Average treatment effects (ATEs) and 95% confidence intervals are estimates of the relationship between a binary indicator for state restrictions on flavored ENDS sales and per capita volume sales of ENDS or cigarettes, as indicated on the left above in bold, based on IRI retail sales data by state and four-week periods for January 2018 through March 2023. Estimates were generated using de Chaisemartin and D’Haultfœuille’s (2024) did_multiplegt_dyn program, which produces a difference-in-difference estimate robust to potential bias from dynamic and heterogenous treatment effects in the context of staggered binary treatment variables. Each estimate above comes from a different regression specification, with covariates added sequentially to clarify their contribution to potential confounding. See the notes to Figure 2 for details on covariates in each set listed above. In addition to covariates noted there, all of these regressions adjust for a continuous measure of partial ENDS flavor policy exposure (i.e., below 100%). Robust standard errors are clustered by state. ** p < 0.01, * p < 0.05, + p < 0.10.

C. IDENTIFICATION

Balance checks in Table 3 confirm the need for covariates adjusting for both tobacco and nontobacco policies in quasi-experimental analyses of the effects of flavor restrictions: Beyond state and date fixed effects, significant predictors include several other flavor restriction variables and prohibitions on unflavored ENDS sales, as well as cigar tax rates and worksite indoor smoking restrictions. Thus, ENDS flavor restrictions appear to be adopted in places that are more aggressive in regulating tobacco in other ways, highlighting the importance of controlling for these variables in our regressions to reduce potential sources of confounding.

TABLE 3.

Balance checks for continuous and binary ENDS flavor policy exposure, coefficient/(standard error)

	Proportion of state restricting flavored ENDS sales, continuous (1)	State restrictions on flavored ENDS sales in effect for the full period, binary (2)
Proportion of state prohibiting unflavored ENDS sales	0.281a (0.041)	0.195 (0.126)
	0.000	0.128
Proportion of state prohibiting flavored ENDS sales if <1		−1.403a (0.498)
		0.007
Cigar flavor policies	0.781b (0.359)	1.248 (0.628)
	0.035	0.053
Menthol cigarette policies	−0.203 (0.398)	−0.630 (0.636)
	0.613	0.328
Interim %: Restricting flavored ENDS sales	−0.435a (0.089)	−0.372a (0.123)
	0.000	0.004
Interim %: Restricting menthol cigarette sales	0.529b (0.247)	0.031 (0.660)
	0.038	0.963
Interim %: Restricting flavored cigar sales	−0.593 (0.605)	0.093 (1.021)
	0.332	0.928
ENDS tax rates, $/mL	0.026 (0.021)	0.025 (0.021)
	0.216	0.238
ENDS tax rates, percentage of cost	0.003 (0.003)	0.003 (0.003)
	0.430	0.419
Cigar tax rates, percentage of cost	0.010b (0.004)	0.010b (0.004)
	0.026	0.022
Cigarette tax rates, $/pack	−0.066 (0.102)	−0.060 (0.101)
	0.523	0.557
% Requiring smoke-free worksites	−0.143b (0.071)	−0.134b (0.055)
	0.049	0.019
% Requiring vape-free worksites	−0.016 (0.041)	−0.022 (0.035)
	0.692	0.528
% Covered by state or local tobacco-21 laws	0.009 (0.036)	0.006 (0.037)
	0.809	0.861
Medical cannabis legal	0.043 (0.057)	0.042 (0.057)
	0.448	0.464
Recreational cannabis legal	0.114 (0.062)	0.114 (0.061)
	0.072	0.068
Beer tax	0.632 (0.406)	0.647 (0.414)
	0.127	0.125
Unemployment rate	0.008 (0.010)	0.008 (0.009)
	0.420	0.408
Heating degree days	0.000 (0.000)	0.000 (0.000)
	0.487	0.440
Median household income	0.000 (0.000)	0.000 (0.000)
	0.166	0.118
COVID-19 deaths/100,000	−0.001 (0.001)	−0.001 (0.001)
	0.115	0.106
Nonessential businesses closed	0.027 (0.052)	0.021 (0.055)
	0.606	0.698
EVALI deaths by state, from CDC website at 1st of month	0.010 (0.021)	0.013 (0.022)
	0.627	0.547
EVALI-era ENDS sales restriction blocked, Michigan	−0.044 (0.031)	−0.036 (0.031)
	0.156	0.252
EVALI-era ENDS sales restriction blocked, Oregon	0.008 (0.057)	0.015 (0.057)
	0.895	0.792
EVALI-era ENDS sales restriction blocked, Utah	−0.518a (0.111)	−0.508a (0.110)
	0.000	0.000
EVALI-era ENDS sales restriction blocked, New York	−0.445a (0.068)	−0.440a (0.071)
	0.000	0.000
Constant	−0.888b (0.367)	−0.889b (0.332)
	0.020	0.011
Fixed effects	State & date	State & date
$N$	2,917	2,917
Adjusted $R^2$	0.777	0.773
Mean (Y var.)	0.091	0.072

Note: Multivariable linear regressions estimate the relationship between ENDS flavor restrictions and all other covariates included in the main regressions (i.e., Figure 1 and Online Appendix Table A2) for our analytic sample. All covariates are indicated above. Robust standard errors are clustered by state.

^ap < 0.01

^bp < 0.05.

Next, we further assess the plausibility of ENDS flavor restrictions meeting the parallel trends assumption by evaluating whether adoption is correlated with time-varying unobserved factors through event study modeling. Repeating the full covariate analyses in Figure 2 with leads and lags on the continuous ENDS flavor policy coverage variable does not suggest differential pre-trends (see Online Appendix Figure A2). However, given staggered adoption of these policies, DCDH event studies offer a more compelling test of that assumption. Figures 4 and 5 plot DCDH event studies with each unit on the $x$-axis measuring the number of 4-week periods since the policy went into effect and the $y$-axis plotting point estimates and 95 percent confidence intervals for the estimated policy effect on the outcome variable at that point. Figures 4A and 5A are unweighted, while 4B and 5B are population weighted. Both use the same covariates as in the full covariate specifications in Figure 3. Pre-implementation point estimates are close to 0 in all cases, and pre-trends appear quite flat, particularly for ENDS sales (see Figure 5).

FIGURE 4.

DCDH event study of state ENDS flavor restrictions on ENDS sales per capita. Applying an approach robust to heterogenous and dynamic effects (de Chaisemartin and D’Haultfœuille 2024) to IRI retail sales data covering four-week periods from January 2018 through March 2023, this analysis assesses the relationship of state-level flavor policies in Massachusetts, California, Maryland, New York, Rhode Island, Utah, and Washington to ENDS sales per capita, relative to the other 37 states for which we have IRI data. The binary exposure variable captures state flavor policies in effect for the entire four-week period. Since the exposure of interest is a binary full coverage indicator, less extensive policy coverage is adjusted for as a covariate equal to the proportion of state residents covered by flavor policies where coverage is only partial (i.e., <1), along with date and state fixed effects. Both analyses adjust for all variables in our full covariate specification (i.e., with all covariates listed in Online Appendix Table A2, columns 4 and 10), with robust standard errors clustered by state. Average treatment effects estimated here are −0.293 (SE: 0.113; p < 0.01) for panel A and −0.284 (SE: 0.157; p = 0.07) for panel B. A: Unweighted specification. B: Population-weighted specification.

FIGURE 5.

DCDH event study of state ENDS flavor sales restrictions on cigarette sales per capita. Applying an approach robust to heterogenous and dynamic effects (de Chaisemartin and D’Haultfœuille 2024) to IRI retail sales data covering four-week periods from January 2018 through March 2023, this analysis assesses the relationship of state-level flavor policies in Massachusetts, California, Maryland, New York, Rhode Island, Utah, and Washington to cigarette sales per capita, relative to the other 37 states for which we have IRI data. The binary exposure variable captures state flavor policies in effect for the entire four-week period. Since the exposure of interest is a binary full coverage indicator, less extensive policy coverage is adjusted for as a covariate equal to the proportion of state residents covered by flavor policies where coverage is only partial (i.e., <1), along with date and state fixed effects. Both analyses adjust for all variables in our full covariate specification (i.e., with all covariates listed in Online Appendix Table A2, columns 4 and 10), with robust standard errors clustered by state. Average treatment effects estimated here are 0.258 (SE: 0.114; p = 0.024) for panel A and 0.158 (SE: 0.127; p = 0.214) for panel B. A: Unweighted specification. B: Population-weighted specification.

The event studies also suggest the need for analyses robust to dynamic effects: ENDS sales per capita fall increasingly over time since the state policy went into effect, while cigarette sales per capita rise increasingly over time. However, growth in flavor policies’ effects on cigarette sales may simply reflect a proportional response to their expanding effect on ENDS sales. Plotting the ratio of the effect estimates in Figure 4 and 5—increases in cigarette sales to decreases in ENDS sales—across the post-period reveals a trend that flattens over time (see Online Appendix Figure A3). The slowly increasing effect of ENDS flavor restrictions on reduced ENDS sales may be due to localities allowing retailers time to sell off existing stocks of flavored ENDS products, delayed supply-side responses (e.g., if manufacturers respond to flavor restrictions by developing new nicotine-product substitutes over time), and slow-moving consumer responses (e.g., learning self-flavoring techniques, identifying alternative sources of flavored ENDS, and shifts in perceived risks of ENDS use in response to these policies (Abouk, Courtemanche, et al. 2023; Abouk, Adams, et al. 2023).

D. SENSITIVITY

As shown in Figures 2 and 3, omitting sales data periods that overlap the first 12 months of the COVID-19 pandemic does not alter our findings. To confirm that a single state’s unobserved policy details or context does not drive our effects, we repeat the DCDH analyses while sequentially dropping each state in our data that had a state-level ENDS flavor policy in place for at least one full four-week sales period. Estimates are similar regardless of which state is dropped, including for California and Massachusetts, which cover 90 percent of state and local policies prohibiting both flavored and unflavored ENDS sales (see Figure 6). This evidence that ENDS flavor policy effect estimates are not sensitive to unobserved factors in any one state’s flavor policy further attests to the robustness of our findings.

FIGURE 6.

DCDH estimates of ENDS flavor restrictions’ effects on ENDS and cigarette sales per capita, sequentially omitting states. Average treatment effects (ATEs) estimate a binary indicator for state restrictions on flavored ENDS sales’ relationship to ENDS and cigarette sales per capita, sequentially dropping states with 100% flavor restriction coverage at any point in the analytic period, to clarify whether findings are driven by a single outlier. All analyses use the full covariate specification in Figure 3. See corresponding figure notes for further details. ** p < 0.01, * p < 0.05, + p < 0.10.

E. EXTENSIONS

We extend our main findings to study heterogeneity in the effect of flavor restrictions by product flavor type, on cigarette sales by consumer age distribution, and by policy type; that is, prohibition of flavored ENDS sales versus sales restrictions that include retailer exemptions. All but the last of these analyses use DCDH. As DCDH does not allow for multiple exposures, TWFE analyses of equation 1 are used to consider the relative impacts of prohibitions versus restrictions on flavored ENDS sales.

Figure 7 repeats analyses separately by product flavor type. ENDS flavor restrictions yield declines in menthol and non-menthol-flavored ENDS sales, but nonsignificant increases in tobacco/unflavored ENDS sales, consistent with the policies reducing access to the former but not the latter. At the same time, these policies yield increases in total cigarette sales, a finding that holds for tobacco-flavored cigarettes but not menthol. This appears to be due to menthol cigarette sales restrictions: The cigarette sales increase is significant for both tobacco and menthol cigarettes when limiting consideration to states with no state or local menthol cigarette sales restrictions during our analytic period.

FIGURE 7.

DCDH estimates of ENDS flavor restrictions’ effects by flavor type. Average treatment effects (ATEs) and 95% confidence intervals are estimates of the relationship between a binary indicator for state restrictions on flavored ENDS sales and per capita volume sales of ENDS or cigarettes overall and by flavor, as indicated on the left above in bold, based on IRI retail sales data by state and four-week periods for January 2018 through March 2023. Estimates were generated using de Chaisemartin and D’Haultfœuille’s (2024) did_multiplegt_dyn program, which produces a difference-in-difference estimate robust to potential bias from dynamic and heterogenous treatment effects in the context of staggered binary treatment variables. Each estimate comes from a different regression specification. All analyses use the full covariate specification in Figure 3. To clarify policy effects in states with versus without restrictions on menthol cigarette sales, subsample analyses limit consideration to states with no residents covered by state or local menthol cigarette sales restrictions during the period of analysis. Robust standard errors are clustered by state. ** p < 0.01, * p < 0.05, + p < 0.10.

Figure 8 shows estimates for the effect of ENDS flavor policies on sales of cigarette products used by younger versus older smokers. Youth use flavored ENDS more than adults (Kaplan et al. 2023), which could lead to larger substitution for youth-disproportionate brands from ENDS flavor restrictions. At the same time, adults who vape are more likely to be current or former smokers, potentially inducing greater substitution towards cigarettes in this age group if flavor policies impede their access to preferred ENDS products. Repeating cigarette sales regressions separately by consumer age profiles shows significant, positive relationships between ENDS flavor policies and either total or non-menthol cigarette sales in all age profiles. Critically, tobacco-flavored cigarettes account for 47 percent of the increase for youth-disproportionate products, 58 percent for adult-disproportionate products, and the entire increase for age-proportionate products, suggesting that menthol cigarette sales bans would not be sufficient to eliminate substitution towards cigarettes in response to flavored ENDS sales restrictions.

FIGURE 8.

DCDH estimates of ENDS flavor restrictions’ effects on cigarette sales per capita by consumer age distribution. Average treatment effects (ATEs) and 95% confidence intervals are estimates of the relationship between a binary indicator for state restrictions on flavored ENDS sales and per capita volume sales of cigarettes by consumer age distribution and flavor, as indicated on the left above in bold. Cigarette sales come from IRI retail sales data by state and four-week periods for January 2018 through March 2023. Estimates were generated using de Chaisemartin and D’Haultfœuille’s (2024) did_multiplegt_dyn program, which produces a difference-in-difference estimate robust to potential bias from dynamic and heterogenous treatment effects in the context of staggered binary treatment variables. Each estimate comes from a different regression, all of which use the full covariate specification in Figure 3. To clarify policy effects on menthol cigarette sales for brands within each consumer age profile, that analysis is run on both the full sample and then again limiting consideration to states with no residents covered by state of local menthol cigarette sales restrictions during the period of analysis. Robust standard errors are clustered by state. ** p < 0.01, * p < 0.05, + p < 0.10.

Regressions in Figure 9 separate the continuous flavored ENDS restriction variable into two separate variables capturing the proportion of residents covered by prohibitions of flavored ENDS sales versus policies that limit flavored ENDS sales to specific types of retailers via retailer exemptions (e.g., adult-only tobacco retailers, smoking bars).¹⁴ As the latter policies often limit ENDS sales to adult-only tobacco retailers, which are not covered in IRI’s (or Nielsen’s) retail sales data, these analyses consider cigarette sales outcomes only.¹⁵ Prohibitions and limitations on flavored ENDS sales yield similar increases in total cigarette sales per capita (both with p < 0.05). However, only flavor prohibition coefficients are significant when considering cigarette sales by consumer age-groups.

FIGURE 9.

Two-way fixed-effects estimates for effects of ENDS flavor prohibitions versus policies limited to specific types of retailers. Point estimates and 95% confidence intervals reflect the relationship between the percentage of a state subject to ENDS flavor restrictions and bans on flavored ENDS sales and cigarette sales per capita, as well as per capita cigarette sales by product consumers’ age profiles. On average, 61% of the population covered by ENDS flavor policies in our data are subject to true bans, while 39% face laws that allow some retailer exemptions. Cigarette sales come from IRI retail sales data by state and four-week periods for January 2018 through March 2023. Estimates here come from a single multivariable regression for each outcome (indicated on the left above in bold). All analyses use the full covariate specification in Figure 2. Robust standard errors are clustered by state. ** p < 0.01, * p < 0.05, + p < 0.10.

We do not compare estimates for ENDS flavor policies that include versus exempt menthol-flavored ENDS because of concerns about confounding from Maryland’s policy. Specifically, by the end of our sample period, Maryland accounted for 65 percent of the population covered by menthol-exempting ENDS flavor restrictions, but its policy did not apply to refillable open-system devices, which are primarily sold in retailers not covered by IRI. This unusual policy detail could drive resulting estimates. However, estimates of ENDS flavor policy effects are similar to each other regardless of whether they include menthol, and are consistent with our main analyses (see Online Appendix Table A3).

V. Discussion

This research has four key findings. First, ENDS sales fall and cigarette sales rise as a greater percentage of state residents are subject to policies restricting flavored ENDS sales. Effects are in the same direction for policies prohibiting all ENDS sales (i.e., flavored and unflavored), consistent with substitution. Second, ENDS flavor restrictions reduce sales of flavored and menthol ENDS, but not tobacco/unflavored ENDS. Third, ENDS flavor restrictions raise cigarette sales, with 81 percent of the increase in cigarette sales coming from tobacco-flavored cigarettes, suggesting that eliminating menthol cigarette sales would not stop substitution of cigarettes for ENDS following ENDS flavor restrictions. Finally, ENDS flavor restrictions’ relationship to cigarette sales holds across cigarette product age profiles, particularly for non-menthol cigarettes, including among brands disproportionately used by underage youth.

These findings are consistent with flavored ENDS policies encouraging substitution from ENDS towards combustible cigarettes, aligning with results from 16 of 18 studies using observational data and causal methods to assess cigarette use following adoption of minimum legal sales age laws for ENDS, ENDS tax rate increases, and advertising restrictions (Pesko 2023). In other words, policies making ENDS more expensive, less accessible, or less appealing appear to incentivize substitution towards cigarettes.

This study’s strengths include combining state-of-the-art quasi-experimental design with novel data on state and local ENDS flavor policies to assess ENDS flavor policies’ relationships to ENDS and cigarette sales across 44 states. We improve upon the prior literature by Chen et al. (2024) by using new methods accounting for heterogenous treatment effects, considering temporary and lasting state and local laws across 16 states (four of which had temporary laws), assessing standardized ENDS volume sales without limiting ENDS sales analyses to data on refills or cartridge counts, studying the effect of ENDS flavor restrictions on age-disproportionate cigarette sales, and conducting this free of commercial conflicts of interest, potentially increasing the findings’ credibility and usefulness for policy making.

Our study makes important contributions to current policy debates, as policy makers in the US, UK, Canada, Ireland, and other countries consider ENDS flavor restrictions. More or less, the FDA has signaled their intention to implement a de facto national ENDS flavor prohibition in the United States by authorizing zero non-menthol-flavored ENDS and only four mentholated ENDS products; however, flavored ENDS continue to be sold through enforcement discretion or by illegally bypassing the FDA entirely. This study’s finding that laws restricting flavored ENDS sales lead to 11 to 15 additional cigarettes sold for every 1 less 0.7 mL ENDS pod sold (equivalent to a pack of cigarettes), suggests that increased cigarette sales could partly or more than fully offset any public health benefit of reducing ENDS sales through such policies.¹⁶

Many economists argue that regulation is justified in situations of externalities and internalities. In the case of cigarettes and other combustible tobacco products, there are clear externalities associated with secondhand smoke exposure and internalities associated with time-inconsistent preferences. The most direct policy approaches to reducing these market failures from combustible tobacco are banning indoor smoking and raising the costs of youth use (such as through minimum legal sales age laws or aggressive enforcement against retailers selling to kids). For ENDS, the case of externalities is likely far smaller as they contain fewer toxicants than cigarette smoke and those toxicants are typically present in secondhand aerosol in much lower levels (Committee on the Review of the Health Effects of Electronic Nicotine Delivery Systems et al. 2018). ENDS could continue to have internalities though, particularly if youth do not anticipate the impact of nicotine addiction on future use.

Consistent with this framework and evidence that ENDS are less harmful to users than combustible cigarettes (Committee on the Review of the Health Effects of Electronic Nicotine Delivery Systems et al. 2018 Committee on Toxicity of Chemicals in Food, Consumer Products, and the Environment 2020; Goniewicz et al. 2017, 2020; Liber 2022; Song et al. 2020), some leading public health scholars have advocated for regulating tobacco products proportionate to their risk (Balfour et al. 2021). This approach would avoid giving more harmful combustible products a competitive advantage over less harmful alternatives. Operationalizing this in the context of flavors could mean more flavors being available in ENDS than cigarettes. There is some precedent for allowing greater flavor availability for less-risky nicotine products: For example, FDA-approved nicotine replacement therapy gum is authorized for sale in flavors of fruit chill and cinnamon surge, in addition to a variety of menthol/mint-related flavors (Haleon Group 2022). Policy makers could also consider pairing a policy that allows more flavors in ENDS with restricting all nonmedicinal tobacco product sales to adult-only stores, to reduce internalities among time-inconsistent (or uninformed) youth while allowing adult access to flavors in order to promote smoking cessation. Given evidence that ENDS use is more effective for smoking cessation than nicotine replacement therapy (Hartmann-Boyce et al. 2022) and high rates of “accidental quitting” among smokers that try ENDS without intending to quit cigarettes (Kasza et al. 2021), these approaches avoid an inequitable trade-off inherent in banning flavors in ENDS, which prioritizes youth over the 11.2 percent of US adults that smoke (Schiller and Norris 2023).

As an alternative to reducing flavors in ENDS, another way to increase the regulatory distance between ENDS and cigarettes is to bar menthol in cigarettes. Yet despite the FDA having been engaged in the federal rule-making process to end sales of mentholated cigarettes for several years now, that rule now appears to be permanently on hold (Becerra 2024).

Our research has several limitations. First, as our estimates of ENDS flavor restrictions’ effects are specifically for mainstream brick-and-mortar retail stores, impacts could be different for online stores or specialty vape shops. Given our adjustment for the portion of ENDS and cigarette sales not covered in our data, our results may underestimate flavor restrictions’ effects on ENDS sales if vape shops are equally as compliant with flavor restrictions as traditional retailers. Since IRI’s data cover the vast majority of cigarette sales in sample states, this limitation concerns ENDS more than cigarettes (Liber et al. 2024). Similarly, these data do not capture illicit market sales. We are unaware of an alternative source that does.

Second, if people travel to buy ENDS in unrestricted jurisdictions, then flavor restrictions’ effects on ENDS use will be smaller than estimated by sales data. Concerns about cross-border spillovers are attenuated, however, by our use of state-level sales data omitting DC, focus on state flavor policies, and robustness checks controlling for non-restriction states that share a border with a state restricting flavored ENDS sales. Third, we cannot infer health effects, distinguish changes in extensive/intensive margin smoking, or discern dual-use of cigarettes/ENDS from sales data. Even so, an increase of 11 to 15 cigarettes purchased for every 1 less ENDS pod sold due to ENDS flavor restrictions raises distinct concerns that unintended costs of such policies may counteract or even outweigh their benefits from reduced vaping.

VI. Conclusion

In this study, we find that ENDS flavor policies reduce flavored ENDS sales as intended, but also increase cigarette sales, including sales of non-menthol cigarettes, menthol cigarettes in areas where menthol cigarette sales are not themselves restricted, brands disproportionately used by underage youth, and brands disproportionately used by adults. As cigarettes are much more harmful than ENDS, the high rate of substitution estimated here suggests that, on net, any population health benefits from policies prohibiting or restricting flavored ENDS sales are likely small or even negative in the current US context.