Aggregated spatial intensity as a method for estimating point-level exposures within area-level units: The case of tobacco retailer exposure in census tracts. (June 2022)

• Record Type: Journal Article
• Authors: Brooks, Madeline M.; Siegel, Scott D.; Corrigan, Anne E.; Curriero, Frank C.
• Is Part Of: Spatial and spatio-temporal epidemiology. Volume 41(2022)
• Publication Date: 2022-06
• Subjects: Choropleth map -- Kernel density estimation -- Spatial intensity -- Tobacco retail -- Exposure; Epidemiology -- Statistical methods -- Periodicals; Epidemiology -- Periodicals; 614.4072
• DOI: 10.1016/j.sste.2022.100482
• Languages: English
• ISSNs: 1877-5845

Abstract

Highlights: Choropleth maps of counts or rates can misrepresent the spatial distribution of point-level events when events are located on or near the boundaries of area-level units used for aggregation. We propose a new method, termed aggregated intensity, to generate area-level estimates of event exposure that retain information from the spatial point-level resolution of events. Using kernel density estimation to calculate the spatial intensity of events, then aggregating these results to area units, produces area-level expected counts of events that are initially unconstrained by unit boundaries. Aggregated intensity (i.e., expected counts) differs most from observed counts for area units with many events on or proximal to their borders. Aggregated intensity may facilitate more spatially realistic measures of event exposure. Abstract: Background: Aggregating point-level events to area-level units can produce misleading interpretations when displayed via choropleth maps. We developed the aggregated intensity method to share point-level location information across unit boundaries prior to aggregation. This method was applied to tobacco retailers among census tracts in New Castle County, DE. Methods: Aggregated intensity uses kernel density estimation to generate spatially continuous expected counts of events per unit area, then aggregates these results to area-level units. We calculated a relative difference measure to compare aggregated intensity to observed counts. Results: Aggregated intensity produces estimates of event exposure unconstrained by boundaries. The relative difference between aggregated intensity and counts is greater for units with many events proximal to their borders. The appropriateness of aggregated intensity depends on events' spatial influence and proximity to unit boundaries, as well as computational inputs. Conclusions: Aggregated intensity may facilitate more spatially realistic estimates of exposure to point-level events.