While college completion is predictive of individual career happiness and economic achievement, many factors, such as excessive alcohol usage, jeopardize college success. In this paper, we propose a method for analyzing large-scale, longitudinal social media timelines to provide fine-grained visibility into how the behaviors and trajectories of alcohol-mentioning students differ from their peers. Using propensity score stratification to reduce bias from confounding factors, we analyze the Twitter data of 63k college students over 5 years to study the effect of early alcohol usage on topics linked to college success. We find multi-year effects, including lower mentions of study habits, increased mentions of potentially risky behaviors, and decreases in mentions of positive emotions. We conclude with a discussion of social media data's role in the study of the risky behaviors of college students and other individual behaviors with long-term effects.