What explains when citizens in transitions exhibit stable preferences and consistent voting behaviour, and when they do not? Studies of democratization have found higher party system volatility in new democracies, partly due to the learning process of political participation. The underlying premise is that as voters participate in several election rounds, their preferences and party choices stabilize. Yet, not all voters exhibit inconsistency. Our original survey data from Tunisia and Egypt (collected in 2011-2015) find that some voters exhibit consistent political behaviour --- both reflecting vote choices that are consistent with their political beliefs, and preferring similar parties over time--, while others do not. We examine two competing explanations to determine which voters are consistent: first, that those who voted in elections prior to the transition were “taught” democracy, and thus developed stable preferences (cf. Lindberg 2009); and second, that in polarized political contexts, it is those on the extremes who are most politically consistent. Interrogating these hypotheses allows us to contribute to understanding the extent to which ‘democratization by elections’ is possible, and also to identify those voters who defy polarisation and vote across sharp cleavages.