Rethinking the Relationship of Working Memory and Intelligence - A Perspective based on Process Overlap Theory

Abstract

Psychometric theories of intelligence typically assume a causal relationship between latent cognitive abilities (general intelligence) and observations of performance (test scores). However, such assumptions are not necessary conditions of the observed covariance structure of test scores, and may have caused over interpretations, if not misinterpretations, of the latent factors of intelligence. Process Overlap Theory (POT; Kovacs & Conway, 2016) proposed a computational framework of intelligence that does not rely on the latent common cause assumption and applies a sampling mechanism of cognitive processes. We re-analyzed data from Kane et al. (2004), which included tests of verbal and spatial working memory and verbal, spatial, and fluid reasoning. We tested latent variable models and network models to compare the traditional common cause approach to POT. As predicted, traditional models overestimated domain-general covariance in working memory and reasoning. In contrast, network models revealed distinct patterns of domain-general and domain-specific covariance across working memory and reasoning. We then simulated Kane et al. (2004) using a POT algorithm and observed the same pattern of results. We argue that cognitive models of working memory and reasoning overemphasize domain-general mechanisms because psychometric models overestimated domain-general covariance. Working memory and reasoning have more domain-specific overlap than current models suggest.