This IAP AP101 Brief Report was sent to me for posting by Philippe Golay. It is reproduced "as is" with only minor editing. This is a guest blog/brief report. Figures included should be possible to enlarge by double clicking on them.
If other folks have completed research related to this blog, and would like to make brief post reports, please contact the blogmaster @ email@example.com
University of Geneva, Switzerland
The fourth edition of the French Wechsler Intelligence Scale for Adult (WAIS-IV) was recently released (Editions du Centre de Psychologie Appliquée – ECPA, 2011). The French WAIS-IV was standardized on a representative sample of 876 people in France ranging in age from 16 to 79. However, for some subtests (Letter Number, Figure Weights and Cancellation), normative data were restricted to 730 participants (and from 16 to 69 years only). In the French WAIS-IV manual, confirmatory Factor analyses were reported, and models with 1, 2, 3 and 4 factors were presented. CFAs supported a factorial structure with 4 factors. Surprisingly, no models based on the Cattell-Horn-Carroll (CHC) theory were reported in the technical manual of the French WAIS-IV. Thus, the main goal of this VERY brief report is to provide a preliminary independent examination of the factor structure of the French WAIS-IV according to the CHC theory. Analyses were conducted on the basis of the subtest inter-correlation matrix and the standard deviations reported in the French manual (p. 50). We used the Akaike Information Criterion (AIC) to compare models.
In the first step, models based on the four-factors solution were tested: four-correlated factors (VCI, PRI, WMI, PSI) and a hierarchical model with four factors and one general factor. We also tested modified versions of the basic 4 factor models because they were suggested and reported in the technical manual. This variant included correlated error terms for Digit Span and Letter Number Sequencing, a cross-loading for Figure Weight on the WMI factor and a cross-loading for Arithmetic on the VCI factor. The model fit of both four factor models (with or without g) was greatly increased as a result. We also tested a bifactor model, in which all subtests scores directly load onto a general factor and also onto one first-order group factor. Results indicated that the bifactor “WAIS-IV” model fits better the data than the other WAIS-IV models.
In a second step, we tested a couple of CHC-based models. We retained a model (fig.1) in which Arithmetic loads both on Gsm and Gf but does not include a cross-loading for Figure Weight on the Gsm factor. This model was better than the basic four-factors WAIS model but slightly less adequate than both modified four-factors solutions. Finally, we tested a bifactor CHC-based model (fig.2). This model with 5 uncorrelated group factors and a first order g factor showed the best fit to the data. The results are summarized in figure 3.
These preliminary results indicated that CHC-based interpretation of the French WAIS-IV is also a valid alternative. Furthermore, bifactor models showed better fit to the data than their higher-order counterparts. This challenges a rather implicit but nevertheless strong assumption that the relationship between the general factor and each subtest is only mediated by the broad abilities.