Racial demographics and cigarette tax shifting: evidence from scanner data

The death rate from lung cancer among black men is 21% higher than white men (American Cancer Society 2015).

2016 National Healthcare Quality and Disparities Report. Agency for Healthcare Research and Quality, Rockville, MD. http://​www.​ahrq.​gov/​research/​findings/​nhqrdr/​nhqdr16/​index.​html.

The IRI is one of two major vendors of supermarket scanner data (along with A.C. Nielsen).

For instance, using Nielsen Homescan data, Harding et al. (2012) concluded that cigarette taxes are passed through less to consumer price if consumers live closer to a lower-tax border. Using data from the Current Population Survey (2003 and 2006–2007), DeCicca et al. (2013) found that price search behaviors—that is, buying cartons instead of packs—significantly reduce the shifting of cigarette excise taxes to consumer prices.

A potential concern with this comparison is that smoking rates are lower among Hispanics. According to Centers for Disease Control and Prevention (2015), smoking rates for Hispanics (11%) are lower than whites and blacks (both roughly 18%). By definition, lower levels of smoking lead to larger elasticity estimates (in absolute values). Thus, even if the price elasticity of demand appears larger among Hispanics, the price responsiveness may not vary much between whites and Hispanics.

Blacks are more likely to smoke cigarettes that are mentholated and stronger, and extract significantly more nicotine per cigarette than whites (Adda and Cornaglia 2006; Caraballo 1998; Perez-Stable et al. 1998).

Maclean et al. (2013) suggested that living in ethnic enclaves may allow more opportunity to obtain untaxed cigarettes. For instance, a large tax increase in New York City led to pervasive illegal cigarette markets in black and Hispanic communities (Coady 2013). Using focus groups study, Shelley et al. (2007) documented large-scale purchasing of untaxed cigarettes in Central Harlem in New York City—a phenomenon known as the “5 dollar man.”

For related discussions, see Berry et al. (1995), Nevo (2001), Petrin (2002), Goolsbee and Petrin (2004), Smith (2004), Ellickson and Misra (2008), and Jia (2008).

Note that retail stores have substantial flexibility to alter cigarette prices through deals with tobacco producers (Bloom 2001; Toomey et al. 2009).

Since 2008, IRI has made available to researchers detailed transaction-level data spanning 30 categories of products, including cigarette product. See Bronnenberg et al. (2008) for further details on the data.

We computed the monthly prices by averaging weekly prices over a month. This does not undermine the virtue of disaggregate level information because relatively little variation in prices is observed within a month.

See Besley and Rosen (1999), Hanson and Sullivan (2009), Sullivan and Dutkowsky (2012), and DeCicca et al. (2013).

In some states, localities impose excise taxes on cigarettes in addition to the state tax. With a few exceptions [e.g., Chicago and Cook County (Illinois) and New York City (New York)], however, the local taxes are relatively small (DeCicca et al. 2013).

For confidentiality reasons, IRI does not report the data collected from markets where only a few chains have dominant market shares (because chain names would be easily identified). Out of the available data, we exclude the stores (in Maine, New Hampshire, and Vermont) for which the geographic market is ambiguously defined by IRI. This reduces our sample to 1687 stores in 28 states.

Summary statistics of key demographic variables are similar between included and excluded states.

Lovenheim (2008) and Harding et al. (2012) computed the distance from each census tract to a road crossing into the lower-tax state. Callison and Kaestner (2014) used Google Maps API to measure the distance to the nearest lower-tax border. In our paper, border locations are measured by a set of geographical coordinates of the line segments that make up the state borders (see Holmes (1998) for further details). The estimates for our distance measure are consistent with the findings of previous literature.

In addition, although Seattle is larger and wealthier than Tacoma (e.g., more than three times larger in population), the pass-through rate in Seattle ($1.00) is close to the pass-through rate in Tacoma ($0.99). This is consistent with the fact that the shares of black population in the two cities are relatively similar (i.e., 8% in Seattle and 11% in Tacoma).

Alternatively, we could include store fixed effects rather than chain fixed effects. In our data, however, price variation across stores within a chain for each UPC (in each city and month) is much smaller than the price variation across chains—suggesting that the chain fixed effects better capture pricing behavior. The average coefficient of variation (ratio of standard deviation to mean) is 7.05% for variation across chains but only 0.29% for variation across stores of the same chain. Using the Nielsen scanner data, Hitsch et al. (2017) also found that retail prices and promotions are relatively homogeneous across stores belonging to the same retail chain.

We also controlled for the number of stores operating within varied thresholds of the radius as a proxy variable for price competition. The estimates were insignificant, however, and the coefficients of tax shifting barely changed (not reported). This indicates that the chain fixed effects capture most of the competition effects.

Another possibility is that sellers might offset anti-smoking regulations by lowering prices (Keeler et al. 1996).

The baseline assumption underlying our empirical model is that there are time-invariant unobserved effects on cigarette prices (across products, chains, and states) that are correlated with the observed regressors such as demographics, distance to the lower-tax state, and taxes. To deal with the potential endogeneity, we use the fixed effects model. Thus, the key identifying assumption is that only the observed regressors affect the expected value of cigarette prices once the fixed effects are controlled for. Section 3.3 shows that our results are robust to various specifications of the fixed effects model.

As indirect evidence, the cross-city variation in prices (measured by the coefficient of variation) is 4.7% overall while the within-city variation is only 2.6%. (The cross-city variation in prices is restricted to the same state.)

The estimates of pass-through rate in columns 2 and 3 are statistically different from one another. In addition, all the coefficients for chain dummies are statistically different from zero.

Adding only demographic variables without border effects has little impact on tax shifting—that is, the pass-through rate does not change significantly from that in column 3. This result does not imply that the demographic variables are unrelated with the pass-through rate, but simply means that time, state, chain, and UPC fixed effects sufficiently control for store pricing behaviors that are associated with cigarette taxes and demographic profiles (Harding et al. 2012). We show in Table 4 that racial demographics indeed have a significant impact on pass-through rates.

Our sample cities do not include Native American reservations that sell lower price cigarettes.

As an alternative measure of racial composition, we have computed the Herfindahl index as the degree of diversity, using eight racial/ethnic categories: white, black, Asian, Hispanic, Native American and Native Alaskan, Hawaiian and Pacific Islander, mixed, and other. The diversity index shows the probability of two randomly drawn people in the city to be of different racial/ethnic groups, ranging from 0 (complete homogeneity) to 1 (complete heterogeneity). The correlation between the diversity index and the share of white population is \(-0.56\). Not surprisingly, we obtained a similar result: that tax shifting decreases with the index of diversity (not reported). As is well known, however, the diversity index does not take into account the composition of specific racial groups, so the index gives the same score whether the city has 70% white and 30% black or 30% white and 70% black.

The pass-through estimate of 0.963 is obtained by 16.168 − 0.329 \(+\) 0.010 \(\times \) (mean of log distance to the lower-tax border) − 0.975 \(\times \) (mean Black share) − 1.238 \(\times \) (mean Hispanic share) \(+\) 1.880 \(\times \) (mean Asian share) − 0.610 \(\times \) (mean Native share) − 2.502 \(\times \) (mean Mixed share) \(+\) 9.237 \(\times \) (mean Other share) − 1.367 \(\times \) (mean of log median income)—based on column 3 (Table 4). Note that population shares of non-black minority groups and income levels are held constant at the mean level.

The median household income and the distance from the lower-tax border are held constant at the sample mean values.

Demographics are collected at the level of Zip Code Tabulation Area (ZCTA) from the US Census Bureau.

We also used the population shares of minority groups at the census tract level. The results (not reported) show that the potential for price search behavior across census tracts (with a population size ranging between 1200 and 8000) also leads to biased estimates of the magnitude of the relationship between local racial composition and cigarette tax shifting.

The intensity of tobacco counter-marketing media campaigns is measured by the gross rating points (GRPs). GRPs are defined as total reach (i.e., the total number of households that could potentially be exposed to an ad campaign) multiplied by frequency of exposure to the ad (i.e., the number of times households in a given media market are exposed to the ads in a given time frame). GRPs are averaged across media markets in each state for 2010. See the Tobacco Control State Highlights 2012 (issued by the Centers for Disease Control and Prevention) for more details on the variables.

Most of the demographic variables in our data have substantial variations across markets within each state.

Because smoking rates by city are not available, we construct a measure of smoking rates based on cigarette sales by city using the scanner data. Specifically, we define the pre-tax smoking rates as the monthly sales of cigarettes (in pack) across stores in a city, prior to tax change, divided by the total monthly potential of cigarette consumption. The potential is assumed to be one pack per day per capita for 2-mile-radius population (over age 18) of a store (provided by the IRI).

Approximately 90% of African-Americans smoke menthol cigarettes (Giovino et al. 2015).