Sunday, April 30, 2006

What does the WISC-IV measure? CHC-organized CFA study

[Double click to make image larger.....sorry for the poor quality. It is the best I can do at this time]

The latest issue of School Psychology Review has an excellent article on confirmatory factor analysis (CFA) of the WISC-IV norm data, by a research team led by Tim Keith. As I've mentioned before, I always read anything written by Tim Keith, particularly if it involves structural equation modeling (SEM) of cognitive batteries. In the language of the PGA golf tour.....Tim is "dah man" in my book for good analyses (with SEM/CFA methods) in the arena of applied psychometrics.

Currently a pdf copy of this journal issues is not no link to the article is provided. Below is the formal reference citation, journal abstract, select important comments/conclusions by the researchers, one editorial comment by this blogmaster, and an invitation to anyone at Psych Corp to provide a guest blog post response.
  • Keith, T., Fine, J., Taub, G., Reynolds, M. & Kranzler, J. (2006). Higher-Order, Multi-Sample, Confirmatory Factor Analysis of the Wechsler Intelligence Scale for Children—Fourth Edition: What Does it Measure? School Psychology Review, 35 (1), 108-127.
  • The recently published fourth edition of Wechsler Intelligence Scale for Children represents a considerable departure from previous versions of the scale. The structure of the instrument has changed, and subtests have been added and deleted. The technical manual for the WISC-IV provided evidence supporting this new structure, but questions about consistency of measurement across ages and the nature of the constructs measured by the test remain. This research was designed to determine whether the WISC-IV measures the same constructs across its 11-year age span and to explicate the nature of those constructs. The results suggest that the WISC-IV indeed measures consistent constructs across ages. The scoring structure of the test was not supported in these analyses, however. Comparison of theory-derived alternative models suggests a model more closely aligned with Cattell-Horn-Carroll theory provides a better fit to the WISC-IV standardization data than does the existing WISC-IV structure. In particular, it appears that the WISC-IV measures crytallized ability (Gc), visual processing (Gv), fluid reasoning (Gf), short-term memory (Gsm), and processing speed(Gs); some abilities are well-measured, others are not. We recommend that users regroup the Perceptual Reasoning tests, and Arithmetic, to better reflect the constructs measured by the WISC-IV. Specific suggestions are also provided for interpretation of WISC-IV scores.

Select Author Comments/Conclusions
  • The present research did find overall support for the invariance of the constructs across the instrument’s 11 age-differentiated groups. The theoretical and scoring model contained in the WISC-IV technical manual was not supported, however.
  • For the WISC-IV, CHC theory provided an alternativeand likely a more valid understanding of construct measurement than did the structure that guided its development and the ensuing scoring structure. Results from these analyses indicate that CHC theory provided a better fit to the standardization data thandid the instrument’s four factor theoretical model. Specifically, the findings from this research indicate that a model separating PRI into measures of fluid reasoning and visual-spatial processing fit the data better than a model with a single PRI factor.
  • The Word Reasoning subtest, although designed in part to measure fluid reasoning, measures a verbal factor, comprehension-knowledge. The results also indicate that the Picture Completion subtest measures both crystallized intelligence and visual-spatial processing. As hypothesized by the WISC-IV’s developers, the Matrix Reasoning subtest is mainly a measure of fluid reasoning. However, the Symbol Search subtest measures not only processing speed, but also the CHC factor visual-spatial processing.
  • The results also indicate that the Arithmetic subtest is a complex measure of cognitive abilities; it measures primarily fluid reasoning, but may also measure working (or shortterm) memory and crystallized intelligence. These findings concerning Arithmetic are consistent with previous findings on the WISC-III and continue the debate concerning what this subtest measured in previous versions of the WISC (Keith & Witta, 1997; Kranzler, 1997). They also suggest that the WISC-IV Arithmetic subtest, like previous versions, provides an excellent measure of g.
Comment from blogmaster
  • I take issue with the last statement above, namely, that the Arithmetic subtest may be a good measure of Gf and possibly g. As appropriately pointed out the Keith et al (in their limitations section), a significant limitation of this study is the lack of other construct indicators from important CHC domains. When cross-battery CFA studies with prior editions of the Wechsler have included indicators of Gq (primary math achievement tests), Arithmetic always demonstrates strong loadings on Gq...not Gf. For example, a recent joint CFA study of the WISC-III and WJ III (Phelps, McGrew, Knopik & Ford, 2005) found, when evaluated within the context of a relatively "complete" set of CHC indicators, the WISC-III Arithmetic test is primarily a measure of Gq (.60+ loading) , with some possible low Gs variance (.20+ loading). Additional support for this interpretation was provided in Woodcock's (1990 [warning....the Woodcock article is 3.8 MB in size] seminal (and first) CHC/Gf-Gc cross-battery organized empirical publication where the Wechsler Arithmetic test, across 8 different CFA analyses, demonstrated a median Gq loading of .753. Most of the studies that continue to suggest that Arithmetic may be a great indicator of Gf/g are typically within-Wechsler CFA studies, studies that do not include an appropriate set of Gq factor indicators.
Invitation for guest blog post response
  • Given that my WJ III authorship status may be viewed as biasing my reading of the Keith et al article, I hereby extend an open invitation to anyone from Psych. Corp. to send me (in Word format) comments and or concerns they have about the original Keith et al study and/or my post. Is anyone from Psych. Corp reading this blog? I would present your comments "as is" as a guest blog post.

Technorati tags:


Anonymous said...

Thank you for pointing out this very important article. It seems like what was done with the CAS was done again with the WISC-IV. The CFA seemed to concur with what most who are in the know on CHC theory knew (the suspected broad ability relators have been embedded in the templates for some time).

I think your twelve step program needs a sequel!

Joel said...

Is it possible that the nature of the Arithmetic subtest has changed since the WISC-R and WISC-III were published? I recall that the WISC-IV manual said that the working memory load of the Arithmetic subtest was deliberately increased.

Kevin McGrew said...

Thanks for the thoughts Joel. If the working memory load of Arithmetic has "been increased", then this would need to be empirically verified in EFA/CFA studies that include enough indicators (at least 3 per factor) to include the latent factors of Gq, Gsm, Gf, and Gs. I would further argue that the Gsm broad factor would need to represented by separate MS (Memory Span) and Working Memory (MW) narrow factors. Similarly, it would be ideal if the Gf factor had enough indicators of Induction (I), gen. seq. reasoning (RG; aka deduction), and quantitative reasoning (RQ). Only when all of these factors are indicated (with sufficient indicators in the measurement model) will someone finally be able to answer some of the "still lingering" questions about Arithmetic. This sounds like a nice masters/PhD thesis for someone. Until I see the empirical data that supports the "increased working memory load" of Arithmetic, I don't buy the extant EFA/CFA studies to date do not support this. Show me the beef (data)?