Using a Prequestion-Based Protocol to Monitor Metacognition in Undergraduate Chemistry: Confidence Predicts Correctness across Content Domains

While prequestioning, a strategy that involves asking students questions before instruction, has been shown to enhance subsequent learning, the metacognitive information generated during the prequestioning process has received little attention. This study investigates whether three self-reported metacognitive variables (Confidence, Knowledge, and Nervousness) predict item-level correctness and whether their predictive value is stable across content domains. Data were collected over 14 consecutive lectures (L) in a first-year undergraduate chemistry course at University College Dublin, yielding approximately 5800 item-level observations from 114 students. At the end of each lecture, students responded to eight multiple-choice items via Poll Everywhere and rated their Confidence, Knowledge, and Nervousness on 0–100 scales. Logistic regression models were fitted separately for each predictor within each lecture, with effect sizes reported as odds ratios per standard deviation (OR/SD). Confidence was a significant positive predictor of correctness in every lecture (OR/SD ranges: 1.69–3.86), indicating a metacognitive signal that was stable across all five chemistry content domains examined. Knowledge showed a closely parallel total-effect pattern (OR/SD: 1.57–2.99), though it did not add unique predictive information beyond Confidence. Nervousness functioned as a significant negative predictor in lectures covering Atomic Structure, Chemical Bonding, Stoichiometry, and Acids and Bases (L1–L11) but was no longer a statistically significant predictor in lectures on Organic Chemistry and Biochemistry (L12–L14). In this block, the OR/SD estimates remained negative (0.78–0.80) and failed to reach significance. Because response rates declined substantially in L12–L14, a plausible and empirically supported explanation for the Nervousness attenuation is reduced statistical power, although a content-driven mechanism cannot be ruled out. The results suggest that prequestions paired with brief metacognitive scales can serve as a practical, real-time monitoring tool, with Confidence providing a robust early warning signal across all content domains and Knowledge serving as a descriptive complement.