Recent theory on ethnic identity suggests that it is constructed and highly influenced by contextual factors and group leaders' strategies. Therefore, while the notion that social cleavages stabilize electoral politics is well established, which cleavages matter and why remain open questions. This article argues that in new democracies the effects of diffuse ethnic cleavages on electoral politics diverge depending on the amount of information they provide to their constituency. The information provision, in turn, depends on how well the cleavage lends itself to the formation of ethnic parties. A formal model of voting stability is developed and empirically tested using data on electoral volatility in all new democracies since 1945 and on individual voting in democratizing Bulgaria. The results show that in the sample examined only identity that centers on language jump-starts party-system stabilization, while race and religion do nothing to stabilize the vote in early elections.