Can states reduce the risk of violent political conflict by simply refusing to collect or publish data on their ethnic makeup and change? This study addresses a neglected aspect of the ethnic conflict literature and provides the first systematic empirical study of the significance of recording ethnic affiliation in censuses for the risk of armed conflict. A general empirical regularity noted in the ethnic heterogeneity and civil war literature is that ethnicity is associated with a somewhat higher risk of conflict in bipolar societies. However, few quantitative studies focus on how changes in the relative strength of groups may affect the risk of civil war. Some recent literature indicates that differential growth may destabilize heterogeneous democracies internally. In democratic societies, political power is distributed according to popular support in elections. A changing balance between groups may thus alter the distribution of power in regimes where ethnic, linguistic or religious divisions to a certain extent determine voting behavior, and this may potentially lead to political instability and ultimately civil conflict. We argue that the relationship between differential growth and instability and violence may be even more important in semidemocracies with electoral systems, but with weak and inconsistent political institutions. We start from the premise that, for differential growth to become a potential driver of instability and violent conflict, information about such change has to be recorded with a national census and actually published. In a cross-national time-series study we investigate whether countries publishing identity data from censuses are at a greater risk of experiencing low-intensity armed conflict. We find that the effect of publishing data on group size is indeed mediated through political institutions. In countries with stable institutions, publication of population identity data is associated with a lower risk of conflict, whereas unstable institutional arrangements are associated with an increase conflict risk when publishing such data.