Abstract
In separate neuroimaging studies, the effects of stimulus similarity and prototype representation have been sought in neural coding for faces and objects (e.g., Jiang et al., 2006, Neuron, 50, 159–172; Loffler et al., 2005, Nat Neurosci, 8, 1386–1390; Panis et al., 2010, JOCN). Stimulus similarity is characterized as a graded recovery from neural adaptation with ever greater dissimilarity between a pair of stimuli. A prototype effect, in contrast, is a larger absolute response to a stimulus which is distant from the center of a stimulus space. While intellectually distinct, these effects are highly confounded in measurement in standard neuroimaging paradigms and can be mistaken for one another. Stimuli which are more distinctive are less subject to adaptation from perceptual neighbors. Therefore, a putative prototype effect may simply result from greater adaptation of prototypical stimuli by other stimuli in the experiment. Conversely, stimulus pairs which are the most perceptually distant from one another, and therefore expected to show the greatest recovery from adaptation, are disproportionately drawn from the extremes of the stimulus space. Therefore, a putative neural similarity effect may be created by an underlying prototype representation. In a standard design (5 morph levels and a target; event-related presentation) these effects can be 90% confounded. These effects may be dissociated properly in the context of a counter-balanced experimental design (e.g., continuous carry-over; Aguirre, 2007, NeuroImage). A linear model may then be used to estimate simultaneously the otherwise confounded effects of similarity and prototype. We will show how this may be done both with event-related potential (Kahn, Harris, Wolk, Aguirre, 2010, JOV) and fMRI data, and consider the interpretation of prior reports in the literature in light of these observations.