Time management is crucial to success in online courses in which students can schedule their learning on a flexible basis. Procrastination is largely viewed as a failure of time management and has been linked to poorer outcomes for students. Past research has quantified the extent of students' procrastination by defining single measures directly from raw logs of student activity. In this work, we use a probabilistic mixture model to allow different types of behavioral patterns to naturally emerge from clickstream data and analyze the resulting patterns in the context of procrastination. Moreover, we extend our analysis to include measures of student regularity--how consistent the procrastinating behaviors are--and construct a composite Time Management Score ("TM"). Our results show that mixture modeling is able to unveil latent types of behavior, each of which is associated with a level of procrastination and its regularity. Overall, students identified as non-procrastinators tend to perform significantly better. Within non-procrastinators, higher levels of regularity signify better performance, while this may be the opposite for procrastinators. [For the full proceedings, see ED593090.]