Clustering Students' Metacognitive Beliefs: Comparing The Results Of K-Means And K- Medoids Algorithms

Self-testing is a powerful method of increasing learning and fostering long-term memory retention (Roediger & Karpicke, 2006a; Roediger & Karpicke, 2006b; Roediger & Butler, 2011). As a study strategy, self-testing has the potential to help students learn and, as a consequence, improve academic performance (Hartwig & Dunlosky, 2011; McAndrew et al., 2015). However, the rates of self-testing usage among undergraduates varies widely. Research linking self-testing and academic performance has found conflicting results as to which strategy increases students’ grade point averages (GPA). While previous research raised important questions about self-testing usage and its relationship to academic performance, what students actually believe about self-testing remains unknown. Is self-testing an effective strategy to remember information? Is self-testing an easy to use strategy to remember information? Examining students’ self-testing beliefs has the potential to reveal what students actually believe about effective strategies, such as self-testing, when they implement these strategies during their studying. Examining students’ beliefs presented a unique analytic opportunity to explore individual differences in study patterns. Rather than using a variable-centered method, cluster analysis was employed to discover groups of distinct, self-testing study profiles. This dissertation examined students’ reported self-testing beliefs and their relationship to reported academic performance in two analyses. Analysis 1 focused on identifying self-testing study profiles in 266 undergraduates using K-means cluster analysis. Analysis 2 focused on identifying self-testing study profiles in 266 undergraduates using K-medoids cluster analysis. The relationship between study profiles and reported academic performance was also examined in both analyses. Study profiles in both analyses showed differences between students’ reported self-testing beliefs and behaviors. Comparing study patterns from both studies revealed methodological differences between K-means and K-medoids cluster analyses. Differences in reported study beliefs, due to the clustering algorithm applied, also changed the theoretical interpretation of the results. Two recommendations were made based on the methods used to identify study profiles and the self-testing belief patterns discovered