I produced a hot spot analysis using point data and another aggregating the points to census tracts before running the Getis G-Ord* statistic. The results are somewhat contradictory—with some statistically significant points lying outside statistically significant census tracts. Looks to be a MAUP issue, given the variability in sizes for census tracts. We’re only interested in areas with excess incidence of disease (we defined as features with p<0.05), so it’s particularly important to capture the statistically significant points in our resulting maps. Am thinking of running the hot spot analyses (using the integrate/collect events method) on point data, visualizing the results via IDW, and selecting census tracts containing statistically significant points as target areas for intervention. Have others done this before? Are there limitations with this approach? We still need census tracts to report out sociodemographic data.
