Protocol - Tobacco Retailer Density/Proximity - Known Residence

Collectively, this measure for tobacco retailer density/proximity includes the following components:

1. Valid data sources providing the location of tobacco product retailers are required.
2. Density requires a definition of "neighborhood" or other spatial unit that is then linked with census-based information about neighborhood characteristics, such as land area, roadway miles, or population size. Density is typically a ratio measure (retailer count divided by neighborhood attribute) and the "correct" choice for the denominator depends on the research question. The Young-Wolff et al. protocol used land area for 500-meter and 1 kilometer person-centered buffers defined by roadway network (that is, as far as one can walk/drive on all roads going in any direction for the specified distance. The protocol computes density for each neighborhood by dividing the count of retailers by the land area (e.g., retailers per km²). Alternatively, one could compute density by dividing the count of retailers by the population in a spatial unit (e.g., retailers per 1,000 people).

3. Proximity of tobacco retailers may be computed for residence, schools, or other locations (e.g., distance between two retailers). The Young-Wolff et al. protocol measures the distance from a known residence to the nearest tobacco retailer in roadway miles and the latter contains a discussion about data confidentiality.

Assuming information on current address (see PhenX Demographics - Current Address) has been collected for a study respondent, then via geocoding it is possible to link the address of a study participant to a measure of tobacco retailer proximity (distance to nearest retailer) and to a measure of density for a defined neighborhood, however defined.

For any density/proximity measure, the WG suggests using GIS software, such as ESRI ArcGIS version 10.1 (ESRI, Redlands, CA). Investigators without such software or expertise may employ a third party vendor to compute these measures for a nominal cost. Multiple steps are required:

• Obtain address data for licensed or likely tobacco retailers: Where there are state or local tobacco retailer licensing requirements, the investigator may obtain retailer addresses from the appropriate licensing authority. When licensing is not required or unavailable to researchers, address lists for likely tobacco retailers may be obtained from commercial vendors (e.g., Dun & Bradstreet), along with some determination of whether or not they sell tobacco products, or investigators may use on-the-ground assessments to identify tobacco retailers in communities.
• Geocode the latitudes and longitudes of addresses for tobacco retailers and participants’ residences (and/or schools and workplaces). Mapping rates of 90% or greater are typical, but the mapping rate depends on the individual data set and one would expect lower rates in rural areas. When geocoding residential address data to a random shift may be employed to avoid incidental disclosure for shared data.
• Define neighborhood: Egocentric neighborhoods (also referred to as "egocentric buffers" and "egohoods") are defined by a radius around a particular location, such as a residence, and these definitions are preferred by both the Young-Wolff and Duncan protocols. Network-based data better captures the travel distance necessary to obtain tobacco products from retailers nearest to participants’ residence. The appropriate distance (400m, 500m, 800m, 1km) depends on the research question. Street-network buffers excluding highways and ramps are created by using software similar to ESRI’s ArcGIS 10 Buffer tool, ArcGIS 10 Data and Maps, and ArcGIS Network Analysis Extension. According to the Duncan protocol, when residential address data are unavailable, alternative definitions of neighborhood are administrative units, such as census block group, tract, zip code tabulation area, city or county.
• Extract census data to characterize each neighborhood: Use data from decennial census or intercensal estimates to compute the land area (or other attribute, such as roadway miles, population size). When buffers overlap multiple tracts, buffer characteristics are weighted in proportion to tract area inside the buffer.
• Compute density: Use software (such as ArcGIS Spatial Join tool) or third-party vendor to calculate the count of tobacco retailers in each neighborhood, and compute retailer density by dividing by the count of retailers by the area attribute of interest (e.g., acres or roadway miles or population size).
• Compute proximity: Use ArcGIS Closest Facility tool (or comparable tool in alternate software) to determine the distance between two points, such as the roadway distance from each residential address to the nearest tobacco retailer.