Transaction data documenting student course selection during academic planning pose novel opportunities for educational data mining to understand and predict student outcomes. These fine-grain data of student waitlist, add, and drop actions can explain the decision-making processes that lead to mismanaged course selection (e.g., dropping a course late with an institutional penalty). The present study investigates student late drops through the lens of late course enrollment. We document first-of-its-kind empirical differences in student academic planning behavior and their resulting workload. We then leverage records of consecutive student semesters to study the causality of the association between late course enrollment and late drops. Specifically, we apply cross-lagged panel models as a novel methodological lens on large-scale enrollment histories to test the specific directionality of the association over time. Results suggest that late course enrollment (17%) and late dropping (10%) are associated with a higher workload compared to students who do not enroll or drop late. This workload could not be explained by students being forced to take high-workload courses at the end of the academic planning period but rather by students' own volition of taking more courses. Further, we find that students were more likely to drop courses with a high workload late, suggesting that students who drop late are unprepared for their workloads, especially if they enroll late and exhibit less regular course planning activity. Our results align with a causal hypothesis for the link between delayed enrollment and late course dropping, estimating the effect to be about 0.8 SDs of more course drops per SD of late course enrollment. This link is slightly stronger for students who enroll in courses with a higher course load and is stable across years of study. There was no reinforcement of late course dropping and delayed enrollment, bolstering support for a causal hypothesis. We discuss the implications of our findings for academic advising, including potential interventions targeting early planning and academic preparedness for course workload to mitigate over-enrollment related to preference uncertainty during delayed decision-making. All data analysis code used for this study is publicly available (https://github.com/CAHLR/enrollment-procrastination-edm).