Of the 156 randomly selected schools that were invited to participate in the 2005–2006 California Student Tobacco Survey, 135 schools that agreed to be surveyed were eligible for an ancillary study about retail tobacco marketing in school neighborhoods. Using the state’s tobacco retailer licensing records and ArcGIS (v9.3, ESRI, 2009), we identified 726 tobacco retailers within 0.8 km (1/2 mile; straight-line distance) of the surveyed schools. The 44 schools without any tobacco retailers within this distance were excluded from the analytic sample (n = 91).
In school neighborhoods with six or fewer tobacco retailers, we observed all of them; in 31 neighborhoods, we randomly selected 6 or 50%, whichever yielded the larger number. Trained coders completed observations in 407 stores (M
= 4.5 per school neighborhood, SD
= 2.9, completion rate = 94.9%) between September and October 2006. Because business classification data were not available with the retailer licensing records, the coders used standard definitions to categorize stores according to type: convenience with or without gas, gas station (only), liquor, small market, supermarket, pharmacy/drug store, and other. All cigarette advertisements were counted and categorized by flavor (menthol, nonmenthol, or both) and by brand (Marlboro, Newport, Camel, and other). Newport was an exclusively menthol brand at the time these data were collected. For the menthol and nonmenthol variety of each brand category (Marlboro, Newport, Camel, and other), coders noted the presence of any advertised promotion (multipack discount, other discount, or gift with purchase). Because collecting price data for the menthol and nonmenthol varieties of the three major brands was cost prohibitive, coders recorded the lowest pack price for Marlboro (nonmenthol), Newport (menthol), and Camel (nonmenthol). Coders indicated whether the price was discounted (e.g., a multipack discount or other sale price) and recorded the number of packs received for the advertised price. Prices for cartons were not recorded because the majority of smokers, particularly menthol smokers, purchase cigarettes by the pack (Fernander, Rayens, Zhang, & Adkins, 2010
For analyses, we computed menthol share of voice for each store, defined as the proportion of all cigarette advertisements in a store that featured any menthol variety. Observed prices were converted to the price of a single pack before sales tax. For the current study, analyses of brand-specific data regarding advertised promotions and prices focused exclusively on Newport and Marlboro, the two most popular cigarette brands among U.S. youth.
To account for clustering of stores within school neighborhoods, multilevel models (HLM6.0) estimated each of the following outcomes as a function of neighborhood demographics: menthol share of voice, the presence of an advertised promotion for Newport and Marlboro (nonmenthol and menthol), as well as the lowest pack price for Newport and Marlboro (nonmenthol only). Enrollment data described the proportions of racial/ethnic groups and the proportion of students eligible to receive free or reduced-price lunches, a common measure of school socioeconomic status (Education Data Partnership, 2010
). We used enrollment data for race/ethnicity because schools were the primary sampling unit. In addition, our previous research observed high correlations between those variables measured by school enrollment data and by census data for the ½-mile radius from the school (Henriksen et al., 2008
). The total number of tobacco retailers in each school neighborhood was obtained from the geocoded licensing data. Neighborhood data for population density (residents per square mile) and proportion of residents ages 10–17 were obtained from Census 2000 and weighted in proportion to tract area. All multilevel models included a random intercept and adjusted for store type, treating convenience stores as the reference category because they were the most prevalent store type. All numeric predictors were centered at the mean. The predictors that represent percentages were scaled to equate a one-unit increase with an increase of 10 percentage points; population density was scaled to represent an increase of 1,000 residents per square mile. Linear outcome multilevel models were estimated using restricted maximum likelihood and robust SE
s. For models of dichotomous outcomes, such as the presence of an advertised promotion for specific brands, hierarchical generalized linear population average models were estimated.