Feedback is a powerful instructional tool. However, even after decades of thorough studies, some questions regarding feedback remain unanswered. In particular, it is yet to be determined whether feedback elaboration is indeed helpful to students. In the study presented here, we take a learning analytics approach to investigate the effect of different types of task-level, computer-based feedback in a popular online learning environment for Mathematics (Khan Academy). Specifically, we run a set of randomized controlled experiments on a large scale--with a total of over 26,000 participants--to study the interplay between feedback type (simple vs. two types of elaboration), the time to re-submit an answer after receiving feedback on incorrect submission to current question (referred to as feedback latency), and success on current and subsequent questions. Results suggest a clear effect of feedback elaboration on feedback latency; superior efficiency of symbolic elaborated feedback over verbal elaborated feedback; and superior efficiency of both over simple feedback. We conclude with implications for practitioners.