Surprise Maps for Exploratory Data Analysis

08-18-2021 05:00 PM
Status: Open
New Contributor III

Hi! I would like to suggest the addition of surprise maps to ArcGIS Pro's exploratory data analysis tools.

They are a tool to identify interesting areas that can account for both counts and rates. Traditional choropleth maps and normalized maps may fail to identify areas for further investigation, but this technique, which is based on Bayesian surprise can identify these interesting areas.

The method is described in the 2017 paper "Surprise! Bayesian Weighting for De-Biasing Thematic Maps" by Correll and Heer.

Description of attached figure: Fig. 1: Choropleth maps of (a) event density, (b) per-capita event rates, and (c) Bayesian surprise for “mischief” (a class of property crime) in Canada. Which province or territory is safest? The density of crimes (Fig. 1a) in the southern provinces suggest that they are less safe; however, this is due to the much larger populations in those provinces. Normalizing to a per-capita rate (Fig. 1b) gives the opposite impression. A Surprise Map (Fig. 1c), using both population density and a de Moivre funnel as models, finds the provinces that stick out: Ontario and Quebec have crime rates lower than expected given their population. The seemingly high per-capita rates in Nunavut accord with the higher variability that can arise from a smaller population.