CFA of WISC-V: A five factor CHC battery

There are dueling factor study articles regarding the WISC-V in the research literature. Here is the take of Reynolds and Keith, who, IMHO, tend to do some of the best factor structure research in intelligence testing.

The five factors look like clear Gc, Gv, Gf, Gwm and Gs CHC factors.

Abstract

The purpose of this research was to test the consistency in measurement of Wechsler Intelligence Scale for Chil-dren-Fifth Edition (WISC-V; Wechsler, 2014) constructs across the 6 through 16 age span and to understand the constructs measured by the WISC-V. First-order, higher-order, and bifactor confirmatory factor models were used. Results were compared with two recent studies using higher-order and bifactor exploratory factor analysis (Canivez, Watkins, & Dombrowski, 2015; Dombrowski, Canivez, Watkins, & Beaujean, 2015) and two using con-firmatory factor analysis (Canivez, Watkins, & Dombrowski, 2016; Chen, Zhang, Raiford, Zhu, & Weiss, 2015). We found evidence of age-invariance for the constructs measured by the WISC-V. Further, both g and five distinct broad abilities (Verbal Comprehension, Visual Spatial Ability, Fluid Reasoning, Working Memory, and Processing Speed) were needed to explain the covariances among WISC-V subtests, although Fluid Reasoning was nearly equivalent to g. These findings were consistent whether a higher-order or a bifactor hierarchical model was used, but they were somewhat inconsistent with factor analyses from the prior studies. We found a correlation between Fluid Reasoning and Visual Spatial factors beyond a general factor (g) and that Arithmetic was primarily a direct indicator of g. Composite scores from the WISC-V correlated well with their corresponding underlying factors. For those concerned about the fewer numbers of subtests in the Full Scale IQ, the model implied relation between g and the FSIQ was very strong.

Click on images to enlarge.










- Posted using BlogPress from my iPad

Why full scale IQ scores are often much lower (or higher) than the part scores? Dr. Joel Schneider on the "composite score extremity effect"

Bingo.  There is finally an excellent, relatively brief, explanation of the phenomena of why full scale IQ scores often diverge markedly from the arithmetic average of the component index or subtest scores.

This composite score extremity effect (Schneider, 2016)  has been well known by users of the WJ batteries.  Why....because the WJ has placed the global IQ composite and the individual tests on the same scale (M=100; SD=15).  In contrast, most other cognitive ability batteries (e.g., Wechslers) have the individual test scores on a different scale (M=10; SD=3).  The use of different scales has hidden this statistical score effect from users.  It has always been present.  I have written about this many times.  One can revisit my latest post on this issue here.

Now that the WISC-V measures a broader array of cognitive abilities (e.g., 5 index scores), users have been asking the same "why does the total IQ score not equal the average of the index scores?"  Why?  Because the five index scores are on the same scale as the full scale IQ score...and thus this composite score extremity effect is not hidden.  A recent thread on the NASP Community Exchange provides examples of psychologists wondering about this funky test score issue (click here to read).

As per usual, Dr. Schneider has provided intuitive explanations of this score effect, and for those who want more, extremely well written technical explanations.

The WJ IV ASB 7 can be downloaded by clicking here.  Although written in the context of the WJ IV, this ASB is relevant to all intelligence test batteries that provide a global IQ score that is the sum of part scores.

Kudos to Dr. Schneider.

Click on image to enlarge

New Wechsler Flynn Effect study by Weiss et al. (2015)

Click on image to enlarge

WISC-V expanded index scores: Verbal Expanded Crystallized (Gc) and Expanded Fluid (Gf) index scores and tables

Click on image to enlarge for easier reading

 

Somehow I seemed to have missed this development on my research literature monitoring radar.  More comments and a link to the technical report can be found at Joel Schneider's excellent blog.