Measuring the level of homework answer copying during COVID-19 induced remote instruction

This paper examines the prevalence of rapid answer copying among university students completing online homework for an introductory level calculus-based physics course taught remotely during the COVID pandemic. We first compared the attempt duration distribution of 26 problems, between 42 students who self-reported as having completed the homework by themselves against the rest of the class. Significant differences were detected for 3 out of 26 problems. We then identified abnormally short problem attempts indicative of potential rapid answer copying, by fitting the attempt duration distribution of each problem with finite-mixture models, using mixtures of either normal or skewed distributions. We detected a significantly smaller fraction of short attempts from self-reporting students on only 3 out of 26 problems and found no statistically significant difference in percentage correct of short attempts between the populations. In conclusion, our analysis did not find evidence indicating widespread rapid answer copying among students. We also explored differences in learning behavior between the two populations by applying process mining to the event logs of one of the homework learning modules, which reveals that some students may have copied answers after spending a longer time or using multiple attempts on a given problem. However, this form of answer copying is also unlikely to be prevalent since the percentage correct on normal attempts is also similar between the two populations on most problems.

Figure 1

Schematic illustration of the design of two online learning modules.

Figure 2

Density distribution of log (base 10) attempt duration of the assessment component of module E08.

Figure 3

Histogram and FMM fit profile for log attempt duration of module 1D03. The black line represents the cutoff originally determined by the algorithm based on a two-component fit, and the red line represents the adjusted cutoff (35 s).

Figure 4

Histogram and best FMM fit profile for log attempt duration (base 10) of modules E02 (1-component skewed normal) and E03 (two-component normal). Red line indicates the cutoff for short attempts estimated from the FMM fit.