A Goal Activation Account of Confidence Judgments

Kevin Zish

Advisor: Tyler H. Shaw, PhD, Department of Psychology

Committee Members: Patrick McKnight, Greg Trafton

Online Location, #2084
May 07, 2020, 01:30 PM to 03:30 PM

Abstract:

Retrospective confidence judgments are used to evaluate the quality and quantity of evidence collected in memory regarding a decision. Recent work suggests that the role of confidence judgments is to determine when continued evaluation of a choice should stop (Desender et al., 2018, 2019; Dotan et al., 2018). For the field of human factors, a clear understanding of how confidence judgments are formed is important whenever the correct decision is unclear such as in real-world tasks including navigation, visual search with baggage screeners, and medical diagnosis. We believe that current theories and models of confidence need to be improved to provide a simpler and broader explanation for the relationship between confidence and memory. For example, current theories of confidence posit dozens of features of memory that inform a confidence judgment, making predictions about how confidence changes with novel manipulations complex. In addition, current models of confidence are good at predicting how a cognitive system would form a confidence judgment for lab-based tasks with discrete trials. However, these same models do not specify how a confidence judgment is formed for more complex tasks, such as procedural tasks, which require that a specific step must be retrieved at a specific time. In this dissertation we demonstrate that the feature of memory used to make a confidence judgment may be goal activation as defined by ACT-R. We begin by investigating how confidence changes in procedural tasks which already have a strong theoretical explanation grounded in goal activation. For Study 1, we demonstrate the existence of confidence carryover which is a novel finding that current models of confidence do not specify. For Study 2, we formalize confidence carryover mathematically with two different procedural tasks. For Study 3, we show that the predicted outcomes of two competing models of confidence judgments can be produced using the same stimuli. The results of Study 3 add a strategic component that is not addressed by any model of confidence judgments.