Research linking personality traits to digital records of online behavior in OSNs like Facebook and Twitter has grown rapidly in recent years. Findings indicate that a broad range of traits can be predicted from behavioral residue online with considerable accuracy. In this registered report, we examine the extent to which the accounts a user chooses to follow on Twitter predict individual differences in self-reported anxiety, depression, post-traumatic stress, and anger in a large sample of active Twitter users (NFinal = 661). We combine best practices in open science and machine learning to provide unbiased estimates of predictive accuracy, with an eye towards more interpretable modelling techniques. This novel network approach has several distinct theoretical and practical advantages over more common linguistic analyses, including being subject to less overt impression management efforts and better capturing passive users. Our findings will speak to how individual differences become represented in our social networks.