Metacognition

Likert scales are ubiquitous in psychological research, but responses are often confounded by systematic preferences for certain response options, regardless of item content. As such, responses to Likert scale instruments reflect two separable processes: expression of a latent psychological state and a decision process about response selection (response style). We present a quantitative model which disentangles response style information from latent state expression, yielding less biased estimates of scale constructs alongside psychologically interpretable response style information. Our approach is grounded in an ordinal-probit model that flexibly estimates response thresholds, capturing a large range of idiosyncratic response style behaviours, and transforms observed ordinal responses into continuous latent scale scores. We apply the model in a hierarchical Bayesian framework and demonstrate that removing the influence of response style generally improves the test-retest reliability and convergent validity of survey instruments and reveals stronger correlations between latent states. Alongside cleaner measures of latent state variables, the model also establishes that response style behaviour is consistent across both time and survey instruments, even when the surveys differ in their response option format and theoretical grounding. Importantly, the model reveals that response style is at least as stable across tasks and time as the latent psychological constructs of interest. These findings have implications for the study of response style as an individual difference in its own right and highlight the many benefits of modelling response style alongside latent state expression.

This is an in-person presentation on July 20, 2026 (15:00 ~ 15:20 EDT).

Self-paced learning, where individuals freely decide how much time to spend encoding each item, is often associated with better long-term memory retention and is further regulated by time pressure. However, the mechanisms underlying these effects remain poorly understood. We hypothesize that the relationship between self-paced learning and memory emerges from people’s optimizing total memory recall under limited cognitive resources during encoding. Unlike participants under a pre-determined presentation schedule, self-paced participants can better allocate and recover cognitive resources according to item difficulty, leading to superior memory retention of these items. When time is abundant, self-paced participants allocate more resources to difficult items; however, when time and resources are scarce, they prioritize easier items instead. These hypotheses were validated via a metacognitive model of memory encoding, where an object-level process uses allocated resources to strengthen memory, and a meta-level process determines how to optimize total recall by adaptively allocating and recovering resources. We discuss the contributions of these results in relation to previous verbal accounts of self-paced learning.

This is an in-person presentation on July 20, 2026 (15:20 ~ 15:40 EDT).

Metacognition comprises two processes: monitoring one’s cognitive states (e.g., confidence judgments) and using these judgments to guide control (i.e., selecting items to restudy). Individuals differ in how confidence guides these selections, yet the underlying reasons remain unclear. Although computational modeling provides powerful tools for understanding metacognitive processes, it has largely focused on monitoring, leaving the mechanisms underlying control - and variability across individuals - underexplored. Moreover, restudying requires choosing among multiple items, suggesting a competitive process that has not yet been formally modeled. We propose a softmax-based sequential choice model in which items compete probabilistically based on confidence-weighted values. We tested the model in two studies using the same learning task. In Study 1 (N = 78), participants freely chose how many items to restudy (1–10); in Study 2 (N = 362), they selected at least 10 of 30 items. Modeled confidence values were reliably recovered across both studies, suggesting that confidence provides a robust signal guiding restudy decisions. Model comparisons and LOOCV revealed that task conditions shape how confidence guides restudy selections. To better understand individual differences, we further examined which confidence signal best captured each participant’s selection behavior. Some participants prioritized extreme-confidence items, others applied threshold-like strategies, and still others relied on absolute or relative confidence. Taken together, these findings offer a computational account of metacognitive control and provide insight into how principles from prominent metacognitive theories, including region-of-proximal-learning and discrepancy-reduction, could be implemented.

This is an in-person presentation on July 20, 2026 (15:40 ~ 16:00 EDT).