Kernel density smoothing of composite spatial data on administrative area level

Abstract Composite spatial data on administrative area level are often presented by maps. The aim is to detect regional differences in the concentration of subpopulations, like elderly persons, ethnic minorities, low-educated voters a political party or persons with certain disease. Thematic collections such maps different atlases. standard presentation Choropleth where each unit represented single value. These can be criticized under three aspects: implicit assumption uniform distribution within area, instability resulting map respect change reference and discontinuities at borderlines areas which inhibit detection clusters. In order address these problems we use density approach construction This does not enforce local distribution. It depend specific choice system there no displayed A estimation procedure densities Kernel estimates. However, estimates need geo-coordinates units disposal as have only access aggregates some system. To overcome this hurdle, statistical simulation concept. interpreted Simulated Expectation Maximisation (SEM) algorithm Celeux et al (1996). We simulate observations from current consistent aggregation information (S-step). Then apply estimator simulated sample gives next estimate (E-Step). concept has been first applied for grid rectangular areas, see Groß (2017), display minorities. second application demonstrated so-called “change support” (Bradley 2016) problem. Here (2020) used SEM recalculate case numbers between non-hierarchical systems. Recently Rendtel (2021) spatial-temporal clusters Corona infections Germany. present modifications basic algorithm: 1) introduce boundary correction removes underestimation kernel borders population area. 2) recognize unsettled lakes, parks industrial computation density. 3) adapt percentages important especially voting analysis. evaluate our against several means register known addresses. empirical part results 2016 election Berlin parliament. contrast show new possibilities reporting results.