Is there a K in capacity? Evaluating the discrete-slot model of visual short-term memory capacity

Visual short-term memory (VSTM) is a fundamental cognitive structure that temporarily maintains a limited amount of visual information in the service of current goals. There is active theoretical debate regarding how limits in VSTM should be construed. According to discrete-slot models of capacity, these limits are set in terms of a discrete number of slots that store individual objects in an all-or-none fashion. According to alternative continuous resource models, the limits of VSTM are set in terms of a resource that can be distributed to bolster some representations over others in a graded fashion. Hybrid models have also been proposed. In the current thesis, I present a series of experiments that leverage different analytic approaches towards model assessment with the aim of evaluating candidate models of VSTM structure. In Experiments 1-3 I fit theoretical ROCs derived from a suite of models to two standard VSTM tasks: a change detection task in which participants had to remember simple features and a rapid serial visual presentation task in which participants had to remember real-world objects. I assessed the performance of each model via cross-validation. Cross-validation analysis provides insight into how well each model generalizes to new samples of data and therefore goes beyond modelling approaches that only involve assessing model fit. To gain a fuller understanding of the nature of limits in VSTM, I also evaluated the ability of these models, as well as a recently proposed hybrid model, to jointly model the two tasks. As part of this analysis, I tested a new variant of a hybrid model, which includes a fixed item capacity but differentially engages an independent attention-like resource that determines the resolution of memory representations (Adam, Vogel, & Awh, 2017). The cross-validation and the joint modeling analyses revealed support for pure continuous-resource models, as well as evidence that performance across the two tasks cannot be captured by a single model of capacity or resource. The purpose of iii Experiment 4-6 was to use a different analytic approach towards model evaluation, which overcomes some of the limitations of fitting models to empirical ROC curves. This alternative analytic approach is based on ranking judgments and involves only minimal assumptions about how memory states are translated to ranks. These experiments corroborated the results of Experiment 1-3, revealing that the qualitative pattern of results was only consistent with the unequal variance signal detection model. In the discussion I provide an interpretation of the unequal-variance signal detection model that aligns it with the broader idea that items can be encoded with variable precision into VSTM