The Professional Counselor | Volume 13, Issue 3 198 & Sink, 2013). We conducted a principal axis factoring with Promax rotation to determine whether factors were correlated using SPSS version 25.0. We chose parallel analysis (Horn, 1965) using the 95th percentile to determine the number of factors to retain given that previous researchers have acknowledged parallel analysis to be a superior method to extract significant factors as compared to conventional statistical indices such as Cattell’s scree test (Henson & Roberts, 2006). We used stringent criterion when identifying loading and cross-loading items such as items that indicated high (i.e., equal to or exceeding 1.00) or low communality values (i.e., less than 0.40; Costello & Osborne, 2005) and items with substantive cross-loadings (< .30 between two factor loadings; Tabachnick & Fidell, 2019) were removed. To ensure the most parsimonious model, we removed items individually from Factor 1, which has the greatest number of items, to reduce the size of the model while still capturing the greatest variance explained by the items on that factor. Results We screened the data and checked for statistical assumptions prior to conducting factor analysis. Little’s Missing Completely at Random (MCAR) test (Little, 1988), a multivariate extension of a simple t-test, evaluated the mean differences of the 45 items to determine the pattern and missingness of data (Enders, 2010). Given the significant chi-square, data were not missing completely at random (χ2 = 912.062, df = 769, p < .001). However, results indicated a very small percentage of values (< 1%) were missing from each variable; therefore, supporting data were missing at random (MAR; Osborne, 2013). When data are MAR, an FIML approach to replace missing values provides unbiased parameter estimates and improves the statistical power of analyses (Enders, 2010). The initial internal consistency reliability estimates (coefficient alpha) for scores on the original OQ-45.2 model were all in acceptable ranges except for Factor 3 (see Henson & Roberts, 2006): total α = .943, Symptom Distress α = .932 (k = 25 items), Interpersonal Relations α = .802 (k = 11 items), and Social Role Performance α = .683 (k = 9 items). We also conducted Bartlett’s test of sphericity (p < .001) and the Kaiser-Meyer-Olkin value (.950), indicating the data was suitable for conducting a factor analysis. We evaluated multivariate normality of the dataset with Mardia’s multivariate kurtosis coefficient. Mardia’s coefficient of multivariate kurtosis was .458; therefore, we deemed the data to be non-normally distributed (Hu & Bentler, 1995). Confirmatory Factor Analysis We tested the developer’s original OQ-45.2 three-factor oblique model, and because of the results subsequently tested a series of alternative structural models outlined by Bludworth and colleagues (2010). Specifically, the alternative structural models tested included: (a) a three-factor orthogonal model, (b) a one-factor model, (c) a four-factor hierarchical model, and (d) a four-factor bilevel model. Table 2 presents the fit indices results in the series of CFAs. The original three-factor oblique model allowed all three factors (Social Role Performance, Interpersonal Relations, and Symptom Distress) to correlate, but resulted in a poor fit: χ2 (942, N = 615) = 3.014, p < .001; CFI = .779; TLI = .768; RMSEA = .057, 90% CI [.055, .060]; SRMR = .063. We next uncorrelated the factors and tested a three-factor orthogonal model, which also presented a poor fit with worsened fit metrics: χ2 (945, N = 615) = 3.825, p < .001; CFI = .689; TLI = .674; RMSEA = .068, 90% CI [.065, .070]; SRMR = .202. Accordingly, because the factors demonstrated high intercorrelation (rs = .94, .93, .91) in the three-factor oblique model and lack of factorial validity based on the CFA results of both three-factor models, we suspected the OQ-45.2 to be a unidimensional, one-factor model. However, the CFA revealed a poor fit to the OQ-45.2 one-factor model: χ2 (945, N = 615) = 3.197, p < .001; CFI = .758; TLI = .747; RMSEA = .060, 90% CI [.057, .062]; SRMR = .062.
RkJQdWJsaXNoZXIy NDU5MTM1