The temporal dynamics of partisanship in the electorate fall into three broad categories: While one group of voters steadfastly supports its party of choice, much like the Michigan model predicts, a second one remains completely outside the party system, never announcing an attachment at all. The majority of voters, however, follows a highly characteristic „homing“ pattern in which they pick a party and subsequently alternate between supporting and not supporting it. The reasons behind this behavior are unknown but it has been suggested that the frequency of announcements is both related to political interest and that the “homing” pattern captures an individual's reactions towards day-to-day politics, just as Rational Choice would expect. We propose a simple stochastic process in which the probability of a voter to announce a partisan attachment depends on a mixture of an initial propensity to support a party and the number of times partisan support has been announced before. We simulate the process both as an agent-based model on micro level and, on population level, as a self-remembering Markov chain. We validate results with the German Socio-Economic Panel (SOEP) which holds the longest and most dense measurement of Partisanship worldwide. Our simulated data shows a strong fit for the most important stylized fact empirical distributions in the dataset and the model is able to produce all three types of behavior from a single process. Furthermore, the structure of our model allows us to draw conclusions regarding the theoretical nature of partisanship pointing out how to connect Michigan and Rational Choice models in a unified framework.
