The Factor Structure of Cognitive Functioning in Cognitively Healthy Participants: a Meta-Analysis and Meta-Analysis of Individual Participant Data

Abstract

Many neuropsychologists are of the opinion that the multitude of cognitive tests may be grouped into a much smaller number of cognitive domains. However, there is little consensus on how many domains exist, what these domains are, nor on which cognitive tests belong to which domain. This incertitude can be solved by factor analysis, provided that the analysis includes a broad range of cognitive tests that have been administered to a very large number of people. In this article, two such factor analyses were performed, each combining multiple studies. However, because it was not possible to obtain complete multivariate data on more than the most common test variables in the field, not all possible domains were examined here. The first analysis was a factor meta-analysis of correlation matrices combining data of 60,398 healthy participants from 52 studies. Several models from the literature were fitted, of which a version based on the Cattell-Horn-Carroll (CHC) model was found to describe the correlations better than the others. The second analysis was a factor analysis of the Advanced Neuropsychological Diagnostics Infrastructure (ANDI) database, combining scores of 11,881 participants from 54 Dutch and Belgian studies not included in the first meta-analysis. Again, the model fit was better for the CHC model than for other models. Therefore, we conclude that the CHC model best characterizes both cognitive domains and which test belongs to each domain. Therefore, although originally developed in the intelligence literature, the CHC model deserves more attention in neuropsychology.

Keywords: Cattell-horn-Carroll model; Clinical neuropsychology; Cognitive functioning; Factor analysis; Meta-analysis of individual participant data; Meta-analytic SEM; Neuropsychological tests.

Fig. 1

PRISMA diagram

Fig. 2

CHC 2 model for the twelve tests included in study 1. For each combination of latent variables, the correlation is given

Fig. 3

CHC model for the ten tests included in study 2. For each combination of latent variables, the correlation is given for the meta-analytic data in roman type, and for the ANDI data in italic type

Fig. 4

Bivariate raw and partial correlations between Trail Making Test B and Letter Fluency, plotted for different studies. The studies are ordered by the size of the correlation

Fig. 5

Bivariate raw and partial correlations between Trail Making Test B and Story Recall Delayed Recall, plotted for different studies. The studies are ordered by the size of the correlation