Using video modeling examples to improve self-assessment and task-selection skills

Video modeling examples seem effective for training self-assessment and task-selection skills, which are key components of self-regulated learning (SRL) skills (Kostons, Van Gog, & Paas, 2012). Kostons et al. used video modeling examples, in which a human model demonstrated and explained how to solve a biology problem by writing out the steps (cf. https://www.khanacademy.org/science/biology/her/heredity-and-genetics/v/introduction-to-heredity), then engaged in self-assessment, and subsequently selected a new learning task. Task selection was based on an algorithm that combined self-assessed performance and mental effort ratings. After training (or no training), students engaged in SRL. Students who had had self-assessment and task-selection training, showed better learning outcomes than students who had not. Nevertheless, for SRL-skills training to become valuable in educational practice, learners need to be able to use the trained skills in different domains and contexts. In other words, transfer of SRL-skills is important. Therefore, we not only investigated whether we could replicate the results from Kostons et al. (2012) that after self-assessment and task-selection training, students learn more from engaging in SRL, but we also added a transfer test, to see if we could find indications of transfer of task-selection training to another context. Because the algorithm used in the original study was highly specific (e.g., a performance rating of 4 out of 5 and effort rating of 3 out of 9 would result in advice to move up 2 steps in task difficulty, either through higher complexity or through less support), and more general rules possibly result in better transfer (Kimball & Holyoak, 2000), we also added a second training condition in which learners were presented with a more generalizable heuristic training (e.g. “if my performance is high and effort is low, I can select a more complex next task”). We hypothesized that the algorithmic and heuristic training conditions would outperform the no training control condition; that is, engaging in SRL would lead to better learning and transfer outcomes after training. Furthermore, we hypothesized that the heuristic condition would show better transfer than the algorithmic condition. In our first experiment, we used other training videos than Kostons et al. (2012), showing a model clicking or typing in an electronic environment rather than writing out the steps. Performance was very low and no differences were found between conditions. To test whether this might have been due to the specific design of the video training, we tested the effects of the training only (without SRL-phase). Indeed, the training proved ineffective. We then changed the videos to more closely resemble those of Kostons et al. and tested the training again, with more success (presumably because these videos guide students’ attention more effectively), and, interestingly, task-selection accuracy was highest in the heuristic condition. Finally, we replicated the very first experiment (i.e., algorithmic, heuristic, or no self-assessment and task-selection training followed by SRL-phase) with the improved training videos and found that both experimental conditions performed better on the posttest than the control condition. However, transfer of task-selection skills now had improved in both experimental conditions, instead of only in the heuristic condition. We suggest that the participants in the algorithmic condition abstracted the heuristic rule from the algorithm by themselves during the self-regulated learning phase. Hence, their transfer is similar to the heuristic condition after using the algorithm during self-regulated learning.