Characterizing the time course of decision-making in change detection.

Abstract

We propose a novel modeling framework for characterizing the time course of change detection based on information held in visual short-term memory. Specifically, we seek to answer whether change detection is better captured by a first-order integration model, in which information is pooled from each location, or a second-order integration model, in which each location is processed independently. We diagnose whether change detection across locations proceeds in serial or parallel and how processing is affected by the stopping rule (i.e., detecting any change versus detecting all changes; Experiment 1) and how the efficiency of detection is affected by the number of changes in the display (Experiment 2). We find that although capacity is generally limited in both tasks, architecture varies from parallel self-terminating in the OR task to serial self-terminating in the AND task. Our novel framework allows model comparisons across a large set of models ruling out several competing explanations of change detection.

Unifying approaches to understanding capacity in change detection.

“…Townsend and Eidels (2011) pointed out that several important boundaries, namely the race model inequality of Miller (1982) and the Grice inequality (Grice et al, 1984), can be expressed in similar terms and derived in the same functional space as the capacity coefficient, allowing for direct comparison. Blunden et al (2022) found that capacity was limited in their experiment; consequently, the Grice inequality is the important measure of comparison between the capacity coefficient and the boundaries. The Grice inequality is a lower bound on the limited capacity that can be produced by the UCIP baseline model.…”

“…This synthesis provides a set of qualitative predictions which we test using measures from SFT. In order to account for additional constraints imposed by the sample size model and to compare performance to other theoretical accounts, we generalize the parametric models of change detection accuracy and RTs of Blunden et al (2022).…”

Further tests of sequence-sensitive models in a modified garner task using separable dimensions.

“…Recent work in categorization using combinations of rule-based decisions have shown that many types of separable dimensions (including the stimuli used in Experiments 2 and 4) are processed independently either in serial or in parallel (see, e.g., Little et al, 2011Little et al, , 2013Moneer et al, 2016). For example, to demonstrate that separable dimensions are processed independently, a variant of the SFT double factorial task is used (Blunden et al, 2022(Blunden et al, , 2020(Blunden et al, , 2014Cheng et al, 2018;Fifi c et al, 2010;Fifi c et al, 2008;Little et al, 2011Little et al, , 2013Moneer et al, 2016). In this task, one category of items is defined by a conjunctive decision rule, such that items belong to this category if they are, for instance, brighter AND more saturated than the values given by the rule.…”

Composite Schematic Faces are Processed Analytically

“…For instance, Little, Nosofsky, Donkin, and Denton (2013) contrasted integral and separable dimensions (Garner, 1974) showing that the former were processed coactively but the latter were processed in an independent manner, either in serial or in parallel. A large number of different types of feature combinations have now been explored using this paradigm (Blunden, Hammond, Howe, & Little, 2022;Blunden, Howe, & Little, 2020;Blunden, Wang, Griffiths, & Little, 2015;Little & Lewandowsky, 2012;Little, Nosofsky, & Denton, 2011;Moneer, Wang, & Little, 2016). A natural question is whether this paradigm can be extended to investigate how faces are processed.…”