Groß, Marcus Rendtel, Ulrich Schmid, Timo Bömermann, Hartmut Erfurth, Kerstin
Year of Publication:
Map-based regional analysis is interested to detect areas with a large concentration of certain populations. Here kernel density estimates (KDE) offer advantages over classical choropleth maps. However, kernel density estimation needs exact geo-coordinates. In a recent paper Groß et al. (2017) have proposed a measurement error model which uses local aggregates for kernel density estimation. Their algorithm simulates "exact" geo-coordinates which reflect the information on the aggregates. In this article we suggest two extensions of this approach. First, we consider boundary constraints, which are usually ignored in the KDE framework. This concerns not only the outer limits of a municipality but also unsettled regions within a city like parks, lakes and industrial areas. Without a boundary correction standard KDEs underestimate the density in the vicinity of boundaries. Here we propose a modification of the original algorithm which uses rescaled kernel functions. Regional maps often display local percentages, for example, voters for a special party among all voters in each voting district. Here we derive a smooth representation of percentages which is based on the ratio of two densities. Again, the original algorithm is modified to cope with the estimation of a ratio of two densities. Our empirical examples refer to voting results from Berlin. It is shown that the proposed methodology reveals a lot of regional insight which is not produced by standard choropleth maps.
Regional Analysis Choropleths Kernel Density Estimation Geo-Coordinates Open data