Code tracing involves simulating at a high level the actions the computer takes when executing a program. Given that students experience difficulties learning this fundamental skill, research is needed on how to effectively teach it. We report on two studies that investigate the pedagogical utility of various notional machine representations used to explain the mechanics of program execution. In study 1 (N = 44), we compared instruction using a concrete computer representation to an abstract table representation. In study 2 (N = 49), we tested if fading between representations improved learning over only providing one representation. The instruction in both studies was embedded in basic tutoring systems we implemented that served as testbeds for the present research. On average students did learn in each study, as evidenced by pretest to posttest gains, but the type of representation did not significantly affect learning;
Bayesian statistics provided substantial evidence for this null result. We discuss potential explanations for our findings and suggest future research directions.