Four- and five-factor models of the WAIS-IV in a clinical sample: Variations in indicator configuration and factor correlational structure.

Abstract

A growing body of research supports the validity of 5-factor models for interpreting the Wechsler Adult Intelligence Scale-Fourth Edition (WAIS-IV). The majority of these studies have utilized the WAIS-IV normative or clinical sample, the latter of which differs in its diagnostic composition from the referrals seen at outpatient neuropsychology clinics. To address this concern, 2 related studies were conducted on a sample of 322 American military Veterans who were referred for outpatient neuropsychological assessment. In Study 1, 4 hierarchical models with varying indicator configurations were evaluated: 3 extant 5-factor models from the literature and the traditional 4-factor model. In Study 2, we evaluated 3 variations in correlation structure in the models from Study 1: indirect hierarchical (i.e., higher-order g), bifactor (direct hierarchical), and oblique models. The results from Study 1 suggested that both 4- and 5-factor models showed acceptable fit. The results from Study 2 showed that bifactor and oblique models offer improved fit over the typically specified indirect hierarchical model, and the oblique models outperformed the orthogonal bifactor models. An exploratory analysis found improved fit when bifactor models were specified with oblique rather than orthogonal latent factors. (PsycINFO Database Record

Contributing Towards a Cultural Neuropsychology Assessment Decision-Making Framework: Comparison of WAIS-IV Norms from Colombia, Chile, Mexico, Spain, United States, and Canada

“…Subsequently lower factor loadings were found for Gv/Gc, Gsm, and Gs. Comparing factor loadings of our best-fitting models to similar models fit with the United States standardization data show that highest factor loadings across higher-order and bifactor variants were found most consistently in Gc subtests for the Chilean WAIS-IV, and were more variable (ranging across Gc, Gv, Gf) in United States Models (Gignac, 2014; also see Niikesla et al, 2013 andStaffaroni et al, 2018). While few conclusive determinations about underlying construct validity differences can be made from these comparisons, the results from this study contribute another (albeit small) piece of information to the accumulating literature suggesting stronger cross-cultural differences in nonverbal tests and dispelling the myth of intelligence test universalism (Brickman, Cabo, & Manly;Cores et al, 2015;Daugherty, Puente, Fasfous, Hidalgo-Ruzzante, & Pérez-Garcia, 2017;Flores et al, 2017;Gasquoine, 1999;O'Bryant, Edwards, Johnson, Hall, Gamboa, & O'jile, 2018;Olson & Jacobson, 2014;Rivera Mindt et al, 2019;Rivera Mindt, Byrd, Saez, & Manly, 2010;Rosselli & Ardila, 2003).…”

“…This finding was not fully anticipated since the AR cross-loading is not only part of the final United States WAIS-IV reported model , AR cross-loading = 0.08) but also reported in several other studies in North American and European studies (e.g., , AR cross-loading = 0.22; van Aken et al, 2017, AR cross-loading = 0.32). At least one other study, however, showed a negative cross-loading of AR on Gc in a clinical population (Staffaroni et al, 2018) and it is possible other studies encountered, but did not report, a similar finding. While this may be a sample-based result (i.e., AR contributes to a verbal factor in some clinical or cultural samples but not in others), it more likely suggests that an alternate factor structure may better describe the data (e.g., an AR cross-loading on Gc is may be erroneous and inconsistent finding, or CHC model may better describe the data than a Wechsler model; Staffaroni et al, 2018).…”

“…At least one other study, however, showed a negative cross-loading of AR on Gc in a clinical population (Staffaroni et al, 2018) and it is possible other studies encountered, but did not report, a similar finding. While this may be a sample-based result (i.e., AR contributes to a verbal factor in some clinical or cultural samples but not in others), it more likely suggests that an alternate factor structure may better describe the data (e.g., an AR cross-loading on Gc is may be erroneous and inconsistent finding, or CHC model may better describe the data than a Wechsler model; Staffaroni et al, 2018).…”

“…Oblique versions of all models were tested first before adding model complexity with higherorder and bifactor structures. This step was also important given that oblique models are part of reporting in the Wechsler manuals (e.g., fit statistics for the final model are reported for both oblique and higher-order variants; and they relate substantially to how the Wechsler scales are often used in clinical practice (analysis of performance on index scores and even subtest scores; Staffaroni et al, 2018). While all oblique models converged normally and generally showed improved fit indices with increasing model complexity, all arithmetic cross-loadings onto a verbal factor (VCI/Gc) resulted in a negative factor loading.…”

“…while an argument could be made for parsimony (less factors being more favorable due to a high risk of over-factorization; Frazer & Youngstrom, 2007), CHC theory has substantially prevailed and numerous arguments in favor of CHC models have been well presented throughout the literature (e.g., Benson et al, 2010;Staffaroni et al, 2018;Weiss et al, 2013b).…”