Confirmatory Factor Analyses of WISC-IV Scores of Children Diagnosed with ASD

Friday, May 13, 2016: 5:30 PM-7:00 PM
Hall A (Baltimore Convention Center)
L. Peacock, J. Beck and M. South, Brigham Young University, Provo, UT
Background:  Confirmatory factor analyses confirmed the four-factor structure (verbal comprehension [VC], perceptual reasoning [PR], working memory [WM], and processing speed [PS]) of the Wechsler Intelligence Scale for Chidren-4th edition (WISC-IV) in the publisher’s large standardization sample (N = 2,200) of mostly typically-developing children age 6-16. Additional analyses confirmed the validity of interpreting a second-order general intelligence factor.  Given that children and adolescents with ASD frequently have uneven IQ profiles, it is not clear that the four-factor structure is appropriate for this clinical population.

Objectives:  We sought to investigate whether confirmatory factor analyses would validate or invalidate the four-factor structure of the WISC-IV in children and adolescents with ASD.

Methods:  Data used reside in the NIH-supported NIMH Data Repositories, specifically in the National Database for Autism Research (NDAR). WISC-IV data was obtained for children and adolescents who had autism diagnoses confirmed by reported ADOS-2 scores above the diagnostic cut-off (≥7). The sample included 44 participants: 39 males and 5 females age 6-16 years (M 11.30) across a broad range of intellectual functioning (FIQ: M 96.96, range 48-131). We conducted maximum likelihood confirmatory factor analyses of two-, three-, and four-factor models of the ten core WISC-IV subtests. With more than four participants per observed indicator, our sample approaches the recommended five participants per indicator (Bentler & Chou, 1987). The two-factor model included verbal and nonverbal latent factors; the three-factor model included latent factors for verbal comprehension, perceptual reasoning and a combined factor of working memory and processing speed; and the four-factor model separated the working memory and processing speed factors. We also tested a second-order four-factor model which included a general intelligence factor. Goodness of fit indices were calculated for each model.

Results:  All factor loadings in all models were significant (p < .01). The fit values for the two-, three- and four-factor models are shown in Table 1. The  χ² statistics for all models were significant (p < .05), indicating poor model fit; however, the χ² statistic is notably sensitive to sample size. The conservative RMSEA indices indicated poor fit (> .08) for all models. The CFI values indicated adequate model fit (> .90) for all but the two-factor model, and showed the four-factor model to be significantly better than the three-factor model (difference > .01; Cheung & Rensvold, 2002). Of all the fit indices, the AIC is the most appropriate for comparing models (smaller AIC is better). The AIC values reveal the four-factor second-order model as best. Coding consistently had the smallest factor loading (β = .66 and .65 for the first- and second-order four-factor models).

Conclusions:  This research confirms the four-factor model of the WISC-IV core subtests in a small sample of children and adolescents with ASD. Of all the core subtests, Coding should be interpreted most cautiously in terms of how it loads onto its respective composite score (Processing Speed). Future research with larger samples should investigate the validity of the four-factor structure across subgroups based on autism symptom severity.