Analysis of source code snapshot granularity levels

Systems that record students' programming process have become increasingly popular during the last decade. The granularity of stored data varies across these systems and ranges from storing the final state, e.g. a solution, to storing fine-grained event streams, e.g. every key-press made while working on a task. Researchers that study such data make assumptions based on the granularity. If no fine-grained data exists, the baseline assumption is that a student proceeds in a linear fashion from one recorded state to the next. In this work, we analyze three different granularities of data; (1) submissions, (2) snapshots (i.e. save, compile, run, test events), and (3) keystroke-events. Our study provides insight on the quantity of lost data when storing data at a specific granularity and shows how the lost data varies depending on previous programming experience and the programming assignment type.