Abstract
BACKGROUND: The objective of this analysis was to compare measurement methods-counts, proximity, mean distance, and spatial access-of calculating alcohol outlet density and violent crime using data from Baltimore, Maryland. METHODS: Violent crime data (n = 11,815) were obtained from the Baltimore City Police Department and included homicides, aggravated assaults, rapes, and robberies in 2016. We calculated alcohol outlet density and violent crime at the census block (CB) level (n = 13,016). We then weighted these CB-level measures to the census tract level (n = 197) and conducted a series of regressions. Negative binomial regression was used for count outcomes and linear regression for proximity and spatial access outcomes. Choropleth maps, partial R(2) , Akaike's Information Criterion, and root mean squared error guided determination of which models yielded lower error and better fit. RESULTS: The inference depended on the measurement methods used. Eight models that used a count of alcohol outlets and/or violent crimes failed to detect an association between outlets and crime, and 3 other count-based models detected an association in the opposite direction. Proximity, mean distance, and spatial access methods consistently detected an association between outlets and crime and produced comparable model fits. CONCLUSIONS: Proximity, mean distance, and spatial access methods yielded the best model fits and had the lowest levels of error in this urban setting. Spatial access methods may offer conceptual strengths over proximity and mean distance. Conflicting findings in the field may be in part due to error in the way that researchers measure alcohol outlet density.