Novel Applications of Statistical and Machine Learning Methods to Analyze Trial-Level Data from Cognitive Measures

Many cognitive tasks and measures can benefit from trial-level analyses including Item Response Theory models as well as other Bayesian and Machine Learning models. Specifically, this dissertation focuses mainly on task-based measures of metamemory and how within-set variability as well as item-level characteristics can improve the inferences researchers make about these measures.First, a clustering analysis of judgements of learning across a task is examined in order to detect different participant strategies on a metamemory task and whether strategy use differs by age. Second, the benefits of using item response theory models to analyze both individual and item-level differences in metamemory tasks are discussed, and applications to multiple datasets are provided. Third, an extended, hierarchical item response theory model was applied to the Child Risk Utility Measure, a tablet-based lab measure used to measure risk taking in preschool aged children. Finally, multiple Bayesian logistic based regression models (including a cumulative logit model, logistic regression model, and zero-one-inflated beta regression model) are applied to the metamemory task described previously to demonstrate the benefits of performing item-level analyses especially as it pertains to differences in the variability of judgements of learning in addition to mean differences between groups. Item or trial-level analyses have many benefits when applied to cognitive tasks and measures and can provide deeper insight into observed effects.