Voting itself, when standing at the polling booth, is ultimately a very individualist and - for the sake of democracy - hopefully an uninfluenced act. Yet, of course we know that during the decision making process, before the actual voting takes place, numerous factors can have an impact on the voters. With regard to the fact that people live and thus are socialized in specific groups (e.g. families, clubs, schools) that can in most cases be located in specific local or regional areas, it is possible to assume people living closer to each other have a higher probability of being affected by the same factors in a similar way. In other words: factors that may shape voting decisions in one area, could play no role in another. Yet, the vast amount of studies analyzing election results and the reasons for certain voting patterns nevertheless assume uniform effects for whole countries. Even when multilevel regression models are estimations which allow for testing varying intercepts and slopes in different (geographical) groups (e.g. states, voting districts), the overall aim of these models is to gain global coefficients (that could vary to some degree between the groups).
In this paper I argue that within election studies these global coefficients are often not very informative. Probably no one who wants to know how the weather will be tomorrow, would be interested in an overall average of the degree for a whole country, but in the local weather forecast for their place of residence. Most election studies nevertheless present just these global effects which may be a product of the composition of the sample, so that in the end, these global effects are not informative for any local context. Based on these findings, I promote to ask our political science questions in a more localized way so that the local, i.e. geographical context forms a central element in research.
In this paper I will present two options of including local context in election studies that operate at a rather low aggregate level. In my example I make use of a newly compiled dataset of election results for the 2017 general election at the municipality level in Germany in order to explain the strong results of the right-wing populist AfD. First, local context will be included in a direct way, measured by distances from the municipalities to relevant geographical markers (in this case e.g. the Eastern border). Second, I introduce the method of Geographically Weighted Regression (GWR) as an option to answer the more localized questions by actively modeling spatial non-stationarity. The resulting coefficient-maps show indeed that most effects (from modernization loser theory to the contact hypothesis) vary significantly and considerably in terms of strength throughout Germany, rendering the interpretation of any global effects very difficult.
