Dependent variables
We constructed four dependent variables: total per capita milliliters of e-liquid, total per capita standardized packs of cigarettes, total per capita cigar units, and total per capita smokeless tobacco ounces. For e-cigarettes we focused on products that contained e-liquid and excluded hardware, batteries, and starter kits with no e-liquid (3.8% of the data). We standardized cigarettes to packs of 20 units, cigars to total pieces of large cigars, tipped cigars, cigarillos, and little cigars, and total ounces for smokeless tobacco products. We defined smokeless tobacco products as snus, moist snuff, chewing loose-leaf, chewing plug twist, dry snuff, and nicotine powder pouches. We excluded smokeless tobacco products where ounce information was not available (0.7% of the data). We further categorized our dependent variables into flavor categories. E-cigarettes were categorized as fruit, menthol, tobacco, and other flavors. Cigarettes were categorized as tobacco and menthol flavors. Smokeless tobacco products were categorized as fruit, mint, and tobacco flavors; we excluded menthol (0.01% of the data) and other (5.3% of the data) flavored products. Cigars were categorized as fruit, tobacco, and other flavors. The category “other flavors” included flavors such as spices, desserts, coffee, and alcoholic beverages. For cigars, menthol (1.2% of the data) is included within other flavors. We utilized data aggregated to 4-week periods from the 4-week period ending on October 12, 2013 for e-cigarettes, cigarettes, and cigars to the 4-week period ending on September 5, 2020. For smokeless tobacco, we only had access to data from the 4-week period ending on November 8, 2014 to the 4-week period ending on September 5, 2020.
Flavor restrictions
We estimated the impact of various flavor restrictions. In November 2019, MA passed its law ending the sale of all flavored tobacco products, including flavored e-cigarettes, menthol cigarettes and flavored cigars. The law immediately ended the sale of flavored e-cigarettes and starting June 1, 2020, the sale of flavored combustible cigarettes and other tobacco products were restricted to licensed smoking establishments [17]. To account for these different implementation dates, we created two dichotomous variables. The value “1” was used to indicate the presence of flavoring restrictions for e-cigarette products, starting with the 4-week period ending November 30, 2019. The second dichotomous variable was used for cigarette, cigar, and smokeless tobacco products and used the value “1” to indicate the presence of flavoring restrictions, starting with the 4-week period ending on July 11, 2020.
In January 2020, NJ signed legislation to impose a permanent ban on flavored e-cigarette products, which became effective on April 20, 2020 [18, 19]. For NJ the value, “1” was used to indicate the presence of flavor restriction regulations, starting with the 4-week period ending May 16, 2020. The states of VA and PA had no flavoring restrictions and were used as our control states.
Control variables
We controlled, at the state level, for various population and policy variables. Included in our analysis were population percentages by race/ethnicity (non-Hispanic White, non-Hispanic Black, non-Hispanic Others, Hispanic), and sex. The analysis also controlled for inflation adjusted median household income (September 2020 dollars) and state unemployment rates [20,21,22]. We also controlled for minimum legal sales age and state tobacco 21 policies. These data were identified from the Centers for Disease Control and Prevention (CDC) State Tobacco Activities Tracking and Evaluation (STATE) System [23]. To indirectly control for excise and wholesale percentage taxes, we used Nielsen data to construct inflation adjusted weighted average prices for one milliliter of e-liquid, a standardized pack of 20 cigarettes, one cigar piece, and one ounce of smokeless tobacco (September 2020 dollars). Lastly, we also used Nielsen data to construct and control for various product market shares. For our e-cigarette regressions we controlled for the market share of disposable products; for cigars we controlled for the market share of cigarillos and large cigars; and for smokeless tobacco we controlled for the market share of snus and moist snuff.
To account for the novel coronavirus (COVID-19) pandemic, we used cumulative cases and deaths from the CDC in each state [24]. Finally, to account for time-specific factors, the analysis controlled for 4-week period time fixed effects.
Statistical analysis
For all our analysis we used difference-in-differences (DID) regression models to evaluate MA full product sales flavor prohibition, and NJ e-cigarette flavor restriction on fruit-, tobacco-, menthol-, and other-flavored e-cigarette; menthol and tobacco cigarettes; fruit-, mint-, and tobacco-flavored smokeless tobacco sales; and fruit-, tobacco-, and other-flavored cigars sales [25, 26]. We used data from VA and PA as control states in our models. Our dependent variables were log per capita total e-liquid milliliters sold, log per capita total standardized cigarette packs sold, log per capita total cigar units sold, and log per capita total smokeless tobacco ounces for each of the above-mentioned flavors. For all models we ran a parallel trends assumption tests on the pre-treatment data. The analysis was conducted using Stata version 17.