A spatial interaction model of vote dispersion

How do votes disperse through a territory? Studies of spatial voting patterns have largely focused on the influence of local factors on voting. The “Friends and Neighbors” model (Key (1949)) explains the advantage of candidates running for office in the locality with which they are associated (Arzheimer and Evans (2012, 2014): Collignon and Sajuria (2018); Horiuchi et al. (2018); Jankowski (2016); Hunt (2020); Munis (2021)), and the “neighbor” effect helps to explain why votes spread. More recent studies have found that the dispersion of votes decreases with distance (Put et al. (2020); Arzheimer and Evans (2012)). However, we know little about how spatial patterns of voting emerge or the mechanism behind the neighbor effect. We argue that this effect depends on the neighbors’ access to information about a candidate, which is constrained by the way information flows. Although scholars have argued that information is a relevant driver explaining the dispersion of votes (Bowler et al. (1993); Arzheimer and Evans (2012); Evans et al. (2017); Campbell, Cowley, Vivyan, and Wagner (2019)), no research has examined the relevance of the network through which information flows. We propose that a spatial interaction model (Wilson (1971)) allows us to predict where this information flows or the voting pattern that will form. Taking advantage of a quasi-natural experiment in Brazilian legislative elections in 1974 and 1978, we show that votes spread through areas of influence created by a hierarchy of cities based on the flows of exchanges among them, including information. We then use our spatial interaction model to predict voting patterns in the elections of 1978 using data from the 1974 elections. Our findings show that the spatial interaction model results fit the data quite well and can help predict spatial patterns of voting.