Despite unprecedented access to information, partisans increasingly disagree about basic facts that are backed by data, posing a serious threat to a democracy that relies on finding common ground based on objective truths. We examine the underpinnings of this phenomenon using drift diffusion modeling (DDM). Partisans (N = 148) completed a sequential sampling task where they evaluated the honesty of Democrat or Republican politicians during a debate based on fact-check scores. We found that partisans required less and weaker evidence to correctly categorize the ingroup as more honest, and were more accurate on trials when the ingroup candidate was more honest, compared to the outgroup. DDM revealed that such tendencies arise from both a prior preference for categorizing the ingroup as more honest (i.e., biased starting point) and more precise accumulation of information favoring the ingroup candidate compared to the outgroup (i.e., biased drift rate). Moreover, individual differences in cognitive reasoning moderated task performance for the most devoted partisans and maintained divergent associations with the DDM parameters. This suggests that partisans may reach biased conclusions via different pathways depending on their depth of cognitive reasoning. These findings provide key insights into the mechanisms driving partisan divides in polarized environments, and can inform interventions that reduce impasse and conflict.