This paper studies how ideological bias affects the transmission of information on social media. We exploit a novel database combining administrative and Twitter data from a population of French politicians over a two-year period, and study how messages flow (i.e. get re-posted and liked) within the sample. Our data show that the network is divided into five distinct  communities (`blocks') with internally homogeneous political ideology. We aim at quantifying two biases which may affect information cascades: the `identity' bias against messages originating from different political blocks, and the `topic' bias related to the message content. Our preliminary findings suggest that identity and topic bias are strong yet heterogeneous across political blocks, and that information cascades are based on political affinity and ideological distance.