Interpretation of student behavior in online learning platforms based on log data is complicated by not being able to directly observe the learner. In this paper, we attempt to identify data patterns that signal either guessing on assessment problems or disengaging from the task for students while working through homework modules in an introductory physics class by contrasting data from the general student population with those who completed homework modules in controlled, observed environments. We found that abnormally short problem solving attempts that were previously modeled as a single guessing or answer copying behavior actually consisted of two different types of "guessing" behavior: rapid and strategic guessing. Both types were associated with lower levels of self-confidence, but had different distribution among proctored and unproctored student populations. More importantly, the fraction of rapid guessing increased significantly after campus closure due to COVID-19, but the fraction of strategic guessing remained constant.