199 The Professional Counselor | Volume 13, Issue 3 Table 2 Goodness-of-Fit Indices for the Item-Level Models of the OQ-45.2 χ2 df p χ2/df CFI TLI RMSEA 90% CI SRMR One-Factor 3021.300 945 .000 3.197 .758 .747 .060 [.057, .062] .062 Three-Factor (orthogonal) 3615.060 945 .000 3.825 .689 .674 .068 [.065, .070] .202 Three-Factor (oblique) 2839.335 942 .000 3.014 .779 .768 .057 [.055, .060] .063 Four-Factor (hierarchical) 2839.335 942 .000 3.014 .779 .768 .057 [.055, .060] .063 Four-Factor (bilevel) 2363.263 900 .000 2.626 .829 .812 .051 [.049, .054] .054 Note. N = 615. χ2 = chi-square; df = degrees of freedom; χ2/df = relative chi-square; CFI = comparative fit index; TLI = Tucker-Lewis Index; RMSEA = root-mean-square error of approximation; 90% CI = 90% confidence interval; SRMR = standardized root-mean-square residual. We proceeded to test the OQ-45.2 as a four-factor hierarchical model. In this multidimensional model, the three first-order factors (Social Role Performance, Interpersonal Relations, and Symptom Distress) became a linear combination to sum a second-order general factor (g-factor) of Psychological Distress (Eid et al., 2017). Results evidenced an unacceptable overall fit to the data: χ2 (942, N = 615) = 3.014, p < .001; CFI = .779; TLI = .768; RMSEA = .057, 90% CI [.055, .060]; SRMR = .063. Last, we examined a four-factor bilevel model. In this model, the g-factor of Psychological Distress has a direct effect on items, whereas, in the hierarchal model, it had an indirect effect on items. Therefore, the items in the four-factor bilevel model load onto both their intended factors (Social Role Performance, Interpersonal Relations, and Symptom Distress) and the g-factor (Psychological Distress). Nevertheless, although the four-factor bilevel was cumulatively the best fitting OQ-45.2 factorial model, the results still yielded a poor fit: χ2 (900, N = 615) = 2.626, p < .001; CFI = .829; TLI = .812; RMSEA = .051, 90% CI [.049, .054]; SRMR = .054. Overall, all models demonstrated a significant chi-square (p < .001); however, this result is common in larger sample sizes (N > 400; Kline, 2016). Because the chi-square statistic is sensitive to sample size and model complexity, researchers have recommended using other fit indices (e.g., RMSEA, CFI) to determine overall model fit (Tabachnick & Fidell, 2019). Nevertheless, the levels of the CFI values (ranging from .689 to .829) and TLI values (ranging from .674 to .812) were low, and far below the recommended referential cutoff (> .90; Tucker & Lewis, 1973). Although the models’ RMSEA values were within the recommended range of .05 to .08 (Browne & Cudeck, 1993), and the majority of SRMR values were below .08 (Hu & Bentler, 1999), these were the only fit indices that met acceptable cutoffs. We further examined outer loadings for the 45 items within the factorial models and identified that all models had outer loadings (ranging from 5 to 14 items) below the 0.5 cutoff (Tabachnick & Fidell, 2019). When CFA produces low factor loadings and poor fit indices, researchers should not be constrained to the original specified number of factors and should consider conducting an EFA (Kline, 2016). Hence, we elected to conduct an EFA to explore the factor structure with this population.
RkJQdWJsaXNoZXIy NDU5MTM1