Title: Exploring Dimensions of Partisan Bias in Large Real-life TextsNak Won Rim, Doctoral Student, Department of Psychology, University of ChicagoPeople evaluate topics and objects differently based on their political beliefs. Several political and social psychology theories have suggested different dimensions where the evaluations based on political belief differ. For example, moral foundations theory proposes that liberals and conservatives have different moral frameworks – something a conservative evaluates as immoral might be moral to liberals, and vice versa. This work tests these theories in a large-scale, out-of-lab context by utilizing language models (word2vec) trained on large corpora (Reddit comments and news articles). Specifically, we use a combination of the dimension-based approach and a classifier to test (1) if these dimensions can distinguish the models trained on liberal corpora and models trained on conservative corpora above chance, (2) whether these dimensions achieve significantly better accuracies than random dimensions. We show that partisan (democrat vs republican) and morality (moral vs immoral) dimensions have significant classification accuracy in both tests and both corpora. Overall, our results show that morality is a dimension where people with different political beliefs differ to the point that we can detect the division in real-life texts people produce outside labs.