The emphasis in the preceding chapters has generally been on the modeling of behavioral data such as response probability and response time. As outlined in Chapter 1, models give us a number of advantages over purely verbal reasoning about behavioral data. The current chapter shows how these advantages extend to neurophysiological data, by applying models of cognition to data from techniques such as electroencephalophy (EEG) and event-related potentials (ERPs); magnetoencephalography (MEG); functional magnetic resonance imaging (fMRI); diffusion tensor imaging (FRI); and single-cell recordings.

Historically, many cognitive modelers and mathematical psychologists (and cognitive psychologists more generally) have been critical of the contributions of neurophysiology and neuroscience to theorizing about cognitive processes (e.g., Coltheart, 2006; Lewandowsky and Coltheart, 2012; Page, 2006). These concerns go beyond the methodology of neuroimaging (e.g., Bennett et al., 2009; Vul et al., 2009), and more generally question whether we can learn anything of theoretical import from neuroimaging. One criticism is that some techniques – particularly fMRI – have traditionally been used to infer where cognitive cognitive processing is performed, but this does not necessarily tell us anything about how it was performed (Page, 2006). Another is that neuroscience has been overly concerned with developing taxonomies of processing (Lewandowsky and Coltheart, 2012). In the case of category learning, Newell (2012) argued that evidence for the idea that different memory systems can be used to categorize objects mostly comes from neuroimaging data, and that the behavioral data (e.g., behavioral dissociations) are less clear cut. In the case of recognition memory, neuroscientific data have been argued to be irrelevant to the distinction between recollection and familiarity, or uninformative without a precise specification of the causal (process) mechanisms (Kalish and Dunn, 2012). The general question underlying these different criticisms is whether neural data are able to discriminate between different theoretical explanations for a behavioral phenomenon (Coltheart, 2006).

Regardless of whether those criticisms turn out to be valid, neuroscience has seen a number of changes over the past decade that have addressed at least some of those concerns. One change has been the use of more fine-grained and dynamic methods for analyzing data, such as functional connectivity (e.g., Van Den Heuvel and Pol, 2010) and multi-voxel pattern analysis (e.g., Norman et al., 2006).