In this chapter, I turn to the model's expectations for the behavior of non-elites. The central expectation to be tested is that people will vote for candidates from their own tribes in one-party elections and for parties whose leaders belong to their language groups in multi-party elections. As in Chapter 7, I identify and test a range of observable implications of the model using a variety of data sources and analytical techniques.

The chapter is divided into three sections. In the first, I estimate and compare rates of tribal voting in one-party and multi-party elections. These analyses demonstrate that, while tribal identities are not the only motivation for voters' choices in either type of contest, Zambian voters nonetheless vote along tribal lines at measurably higher rates in one-party elections than in multi-party ones. In the second section, I focus exclusively on voting patterns in multi-party elections. First I present evidence to support the central assumption in the model that voters put more emphasis on candidates' party affiliations than on their individual backgrounds. Then I show that this emphasis on candidates' party affiliations leads voters to allocate their support on language group lines.

In the third section, I test the model's implications in a more fine-grained way through a pair of controlled experiments. The first compares the performance across elections of candidates that ran in the same constituencies in back-to-back contests.