Effects of g factor on rGMV and of “g-independent” score for each test on rGMV

Although we primarily focused on individual cognitive measures in this study, several previous studies have assessed the correlation of gray matter structures with the general intelligence factor (g factor, a factor that partially explains success in diverse forms of cognitive activity) as well as specific effects that are not explained by g factor⁴².

In order to estimate the effects of g factor as well as task-specific effects independent of g factor on rGMV for each task, we first performed exploratory factor analysis.

Data were analyzed using SPSS 22.0 statistical software (SPSS Inc., Chicago, IL). Promax-rotated factor analysis (unweighted least squares method) of scores for each question in six cognitive tasks was performed. Two factors that showed eigen values higher than 1 were extracted based on the Kaiser–Guttman criterion⁴³.

The factor loadings of each task to the first factor were as follows: RAPM: 0.579, digit span: 0.465, Stroop interference: −0.006, spatial factor of TBIT: 0.685, reasoning factor of TBIT: 0.593, and perceptual factor of TBIT: 0.634.

The factor loadings of each task to the second factor were as follows: RAPM: −0.033, digit span: −0.038, Stroop interference: 0.999, spatial factor of TBIT: 0.018, reasoning factor of TBIT: 0.045, and perceptual factor of TBIT: −0.004.

The correlation coefficient between the first and second factors was 0.034.

As can be seen, Stroop interference showed a rather distinct pattern. This may be due to the nature of Stroop interference in which “more and faster” does not necessarily lead to better performance. Thus, in subsequent g-related analyses. Stroop interference was excluded from the analyses.

To calculate the g-independent score for each test, we followed the method of Karama et al.⁴². We computed five different g scores in order to estimate “g-independent” scores for each test. Each of these scores was based on only four tests instead of the five available to avoid inadvertently controlling for the contribution of the test of interest when controlling for g. For instance, “RAPM”-free g scores were computed from four tasks excluding the “RAPM” test. One regression analysis was conducted for each test, using the appropriate g score as the independent variable and the respective specific test score as the dependent variable. Residuals from these regression analyses were considered to represent “g-independent” specific test scores.

In whole-brain multiple regression analyses, we tested for relationships between calculated cognitive measures and rGMV. In all analyses, the dependent variables were the rGMV signal values at each voxel. Six whole-brain multiple regression analyses were performed, all of which included age and sex as covariates. Additionally, six analyses included the following dependent variables: (1) g factor calculated from the five tasks described above, (2) g-independent RAPM score and g factor calculated from the other four tasks, (3) g-independent digit span score and g factor calculated from the other four tasks, (4) g-independent spatial factor of TBIT and g factor calculated from the other four tasks, (5) g-independent reasoning factor of TBIT and g factor calculated from the other four tasks, and (6) g-independent perceptual factor of TBIT and g factor calculated from the other four tasks. The remaining statistical analyses were performed as described for the main analyses of each score.