WISC V Construct Validity: Hierarchical EFA with a Large Clinical Sample

Transcription

1 WISC V Construct Validity: Hierarchical EFA with a Large Clinical Sample The Wechsler Intelligence Scale for Children-Fifth Edition (WISC V; Wechsler, 2014) was published with a theoretical five-factor structure. However, examination of the standardization sample suggested four latent factors (i.e., Visual Spatial and Fluid Reasoning factors merged into the former Perceptual Reasoning factor; Canivez, Watkins, and Dombrowski, 2016, 2017). It is possible that a different structure might emerge from a clinical sample. Accordingly, the present study examined the WISC V factor structure with a large clinical sample using exploratory factor analyses (EFA) of the 10 primary subtests. A large mixed clinical sample (N=2,512) of children who were administered all 10 primary WISC V subtests within outpatient neuropsychological evaluations was randomly bifurcated into EFA (n=1,256) and CFA (n=1,256) samples (Table 1). Extraction criteria examined included eigenvalues>1 (Guttman, 1954), scree test (Cattell, 1966), standard error of scree (SE Scree ; Zoski & Jurs, 1996), parallel analysis (PA; Horn, 1965), and minimum average partials (MAP; Velicer, 1976). Principal axis analysis and promax rotation (k=4; Gorsuch, 2003) with salient factor pattern coefficients.30 (Child, 2006) were conducted and followed by the Schmid and Leiman (1957) procedure to enable examination of variance apportions. Coefficients omega-hierarchical (w H ) and omega-hierarchical subscale (w HS; Watkins, 2013, 2017) were estimated. WISC V descriptive statistics (Table 2) illustrated univariate normality. Factor extraction criteria of visual scree, SE Scree, PA (Figure 1), and MAP indicated one factor should be extracted; while eigenvalues>1 indicated two factors. Factor extraction began with five factors (publisher proposed) and iteratively reduced by one to explore alternate solutions. The five-factor extraction (Table 3) showed the fifth factor contained no salient subtest pattern loadings and was thus inadequate. Matrix Reasoning and Figure Weights had salient loadings on the Perceptual Reasoning factor along with Block Design and Visual Puzzles, not on a separate Fluid Reasoning factor. Congruent with Canivez et al. (2016, 2017) there was no support for five WISC V factors. The four-factor extraction (Table 4) included salient subtest pattern coefficients on singular factors consistently aligned with Wechsler based dimensions (Canivez et al., 2016, 2017). Two- and three-factor extractions (Table 5) produced fusion of meaningful constructs and subtest cross-loadings, which may be symptomatic of underextraction and unsatisfactory solutions (Gorsuch, 1983; Wood et al., 1996). Decomposed variance sources for the WISC V four-factor solution (Table 6), revealed dominance of general intelligence. Residual subtest variance was apportioned to the theoretically consistent group factors and 70.2% of WISC V explained common variance was by general intelligence, while the four WISC V group factors provided between 1.5% 7.1% additional explained common variance. The omega-hierarchical coefficient for the g factor was large (w H =.821), showing strong measurement of true score variance associated with this dimension. Omega-hierarchical subscale coefficients for the four WISC V group factors were low for VC, PR, PS, and WM ranging from and indicating insufficient true score variance for clinical interpretation (Reise, 2012; Reise, Bonifay, & Haviland, 2013). Overall, results did not support the publisher s proposed five-factor model. Instead, a four-factor model appears more appropriate, but is composed primarily of general intelligence test variance, suggesting extreme caution when interpreting scores beyond the FSIQ.

2 Table 1 Demographic Characteristics of the Clinical EFA and CFA Samples EFA Sample (n = 1,256) CFA Sample (n = 1,256) N % N % Sex Male Female Race/Ethnicity White/Caucasian Black/African American Asian American Hispanic/Latino Native American Multiracial Native Hawaiian/Pacific Islander Other Unknown Table 2 Wechsler Intelligence Scale for Children-Fifth Edition (WISC V) Descriptive Statistics for the Clinical EFA and CFA Samples EFA Sample (n = 1,256) CFA Sample (n = 1,256) Subtest/Composite M SD Skewness Kurtosis M SD Skewness Kurtosis Subtests Block Design Similarities Matrix Reasoning Digit Span Coding Vocabulary Figure Weights Visual Puzzles Picture Span Symbol Search Composites VCI VSI FRI WMI PSI FSIQ Note. VCI = Verbal Comprehension Index, VSI = Visual Spatial Index, FRI = Fluid Reasoning Index, WMI = Working Memory Index, PSI = Processing Speed Index, FSIQ = Full Scale IQ. Mardia s (1970) multivariate kurtosis estimate (EQS 6.3) was 4.23 for the EFA sample and 9.71 for the CFA sample. Independent t-tests for mean differences of WISC V subtests and composite scores between the EFA and CFA samples indicated no statistically significant differences with t values ranging from to 1.23 and p values ranging from.2076 to.9880.

3 Table 3 Exploratory Factor Analysis of the 10 Wechsler Intelligence Scale for Children-Fifth Edition (WISC V) Primary Subtests: Five Oblique Factor Solution with Promax Rotation (k = 4) for the Clinical EFA Sample (N = 1,256) General F1: PR F2: VC F3: PS F4: WM F5 WISC V Subtest S P S P S P S P S P S h 2 SI VO BD VP MR FW DS PS CD SS Eigenvalue % Variance Factor Correlations F1: PR F2: VC F3: PS F4: WM F5 F1: PR F2: VC.716 F3: PS F4: WM F Note. WISC V Subtests: SI = Similarities, VO = Vocabulary, BD = Block Design, VP = Visual Puzzles, MR = Matrix Reasoning, FW = Figure Weights, DS = Digit Span, PS = Picture Span, CD = Coding, SS = Symbol Search. PR = Perceptual Reasoning, VC = Verbal Comprehension, PS = Processing Speed, WM = Working Memory. S = Structure Coefficient, P = Pattern Coefficient, h 2 = Communality. General structure coefficients are based on the first unrotated factor coefficients (g loadings). Salient pattern coefficients (.30) presented in bold.

4 Table 4 Exploratory Factor Analysis of the 10 Wechsler Intelligence Scale for Children-Fifth Edition (WISC V) Primary Subtests: Four Oblique Factor Solution with Promax Rotation (k = 4) for the Clinical EFA Sample (N = 1,256) General F1: Perceptual Reasoning F2: Verbal Comprehension F3: Processing Speed F4: Working Memory WISC V Subtest S P S P S P S P S h 2 Similarities Vocabulary Block Design Visual Puzzles Matrix Reasoning Figure Weights Digit Span Picture Span Coding Symbol Search Eigenvalue % Variance Promax Based Factor Correlations F1: PR F2: VC F3: PS F4: WM F1: Perceptual Reasoning (PR) F2: Verbal Comprehension (VC).738 F3: Processing Speed (PS) F4: Working Memory (WM) Note. S = Structure Coefficient, P = Pattern Coefficient, h 2 = Communality. General structure coefficients are based on the first unrotated factor coefficients (g loadings). Salient pattern coefficients (.30) presented in bold.

5 Table 5 Exploratory Factor Analysis of the 10 Wechsler Intelligence Scale for Children-Fifth Edition (WISC V) Primary Subtests: Two and Three Oblique Factor Solutions for the Clinical EFA Sample (N = 1,256) Two Oblique Factors Three Oblique Factors WISC-V Subtest g 1 F1: g F2: PS h 2 g 1 F1: PR F2: VC/WM F3: PS h 2 SI (.765).031 (.528) (.635).781 (.809) (.473).658 VO (.745).065 (.533) (.631).809 (.814) (.460).667 BD (.730).028 (.503) (.820) (.596).080 (.530).676 VP (.641).263 (.574) (.862).009 (.649) (.506).743 MR (.458).791 (.745) (.732).135 (.617).048 (.491).547 FW (.744).047 (.522) (.725).217 (.629) (.428).544 DS (.735) (.447) (.544).577 (.690).185 (.538).498 PS (.819) (.514) (.489).411 (.595).300 (.552).410 CD (.565).362 (.578) (.436) (.444).774 (.754).569 SS (.516).744 (.758) (.495).007 (.493).727 (.764).585 Eigenvalue % Variance Factor Correlations F1 F2 F1 F2 F3 F1 F1 F2.667 F2.751 F Note. WISC V Subtests: SI = Similarities, VO = Vocabulary, BD = Block Design, VP = Visual Puzzles, MR = Matrix Reasoning, FW = Figure Weights, DS = Digit Span, PS = Picture Span, CD = Coding, SS = Symbol Search, g = general intelligence, PS = Processing Speed, WM = Working Memory, h 2 = Communality. 1 General structure coefficients based on first unrotated factor coefficients (g loadings). Factor pattern coefficients (structure coefficients) based on principal factors extraction with promax rotation (k = 4). Coefficient, P = Pattern Coefficient, h 2 = Communality. General structure coefficients are based on the first unrotated factor coefficients (g loadings). Salient pattern coefficients presented in bold (pattern coefficient.30)

6 Table 6 Sources of Variance in the Wechsler Intelligence Scale for Children-Fifth Edition (WISC V) 10 Primary Subtests for the Clinical EFA Sample (N = 1,256) According to an Exploratory Bifactor Model (Orthogonalized Higher-Order Factor Model) with Four First Order Factors General F1: Perceptual Reasoning F2: Verbal Comprehension F3: Processing Speed F4: Working Memory WISC V Subtest b S 2 b S 2 b S 2 b S 2 b S 2 h 2 u 2 Similarities Vocabulary Block Design Visual Puzzles Matrix Reasoning Figure Weights Digit Span Picture Span Coding Symbol Search Total Variance Explained Common Variance w w H /w HS Relative w H PUC.800 Note. b = loading of subtest on factor, S 2 = variance explained, h 2 = communality, u 2 = uniqueness, w = Omega, w H = Omega-hierarchical (general factor), w HS = Omega-hierarchical subscale (group factors), H = construct reliability or replicability index, PUC = percentage of uncontaminated correlations. Bold type indicates highest coefficients and variance estimates and consistent with the theoretically proposed factor.

7 Random Data WISC-V Clinical Sample N = 1, Eigenvalue Factor Figure 1. Scree plots for Horn s parallel analysis for the 10 WISC-V Primary Subtests with the clinical EFA sample (N = 1,256).