"I Knew it!" Model-Based Dissociation of Prior Knowledge Confounds in Memory Assessments

Computational modeling is a powerful approach for discerning individual differences in memory function. The model-based assessments discussed in this paper rely on estimating an individual's rate of memory decay– a stable and idiographic parameter that the model can capture. However, this paper aims to demonstrate prior knowledge as a confounding factor in these model-based assessments and seeks to parse out the error using Maximum Likelihood Estimations. The metric of individualized memory performance, termed Speed of Forgetting, was significantly lower for facts known beforehand. Still, these facts were identified with 81% accuracy by recovered base-level activation estimations blind to the ground-truth data. A proposal for future model-based assessments to account for prior knowledge is discussed.