What can individual differences tell us about the specialization of function?

Abstract

Can the study of individual differences inform debates about modularity and the specialization of function? In this article, we consider the implications of a highly replicated, robust finding known as positive manifold: Individual differences in different cognitive domains tend to be positively intercorrelated. Prima facie, this fact, which has generally been interpreted as reflecting the influence of a domain-general cognitive factor, might be seen as posing a serious challenge to a strong view of modularity. Drawing on a mixture of meta-analysis and computer simulation, we show that positive manifold derives instead largely from between-task neural overlap, suggesting a potential way of reconciling individual differences with some form of modularity.

Keywords:

• Individual differences
• Functional specialization
• Cognitive architecture
• Positive manifold

Notes

¹ Modularity and domain specificity are two closely related but not identical notions. One could, for example, have domain-specific knowledge (e.g., about the rules of chess) in an architecture like Newell and Simon's General Problem Solver, in which the rules of inference are fully domain general. Conversely, a single computational algorithm (e.g., for performing spectral analysis) could be modular in its implementation but be used in multiple domains (e.g., in both vision and audition). Although in most accounts modules are presumed domain specific, and vice versa, we do not here consider the terms to be fully interchangeable.

² As noted later, however, positive manifold (which is true at the population level, or between individuals), sometimes also referred to as psychometric g, is conceptually distinct from g as in general intelligence (a hypothesized construct within an individual): The former is a statistical phenomenon, and the other is a hypothesized cognitive ability or resource.

³ Thomson, working over 80 years ago, was not focused on modularity issues, and the particular instantiation of overlapping resources that he considered—in terms of “bonds” in the brain—might conceivably be cached out in a very unstructured view of the mind, in which computational problems were solved anew from a completely shared pool of neural resources. Here, we consider a more structured instantiation of his general viewpoint, in which the neural resources at issue might correspond to larger units (e.g., brain regions or populations of neurons) and be configured in systematically structured ways.

⁴ In some cases, the meta-analysis reported activation of multiple Brodmann areas for the same proportion of studies. When this happened, we chose the Brodmann area that would result in the greatest number of possible overlapping areas for the domains across studies. This served to extend the range of possible overlap and explore a greater number of possible scenarios. Sensitivity analyses were done, selecting these areas at random, and the results did not change (although the range of possible overlapping areas was restricted).

⁵ The choice of item error variances was selected according to those observed in the literature; however, there is substantial variability in these values. As a result, we fit the same set of models but varied the magnitude of error variances. In all cases, the magnitudes of the parameter estimates and the R-square of the g-factor in the fitted models we present below remained unchanged.

⁶ The chi-square value provided somewhat less evidence of good fit to the data; however, the chi-square value is known to be an imperfect indicator of model fit that is especially susceptible to large sample size and large correlations (Kline, 2010), both of which are present in our simulated data.