TPC Journal V7, Issue 4 - FULL ISSUE

340 The Professional Counselor | Volume 7, Issue 4 The maximum likelihood estimation successfully converged in 40 iterations. The correlations table was consulted for multicollinearity; only one relationship was above .90. The relationship between war-related threat to self and intrapersonal religious commitment was .93. The fit indices indicated an acceptable fit of the data to the hypothesized structure. The CFI was .94, SRMR was 0.056, and RMSEA was .050, with a 90% confidence interval of .042–.058. All items loaded statistically significantly ( p < .05) on the theorized latent variables and no modifications were warranted based on the values calculated (see Table 1). The squared multiple correlations, an indicator of reliability of items, seemed acceptable, except for some items loading onto the optimism factor. These results were unsurprising considering the low Cronbach’s alpha of the instrument in this sample. The theoretical model. The 5-factor solution hypothesizing the directional relationship successfully converged in 29 iterations. Using the maximum likelihood estimation, evidence from the model suggested that the data did not fit the model as expected (CFI = .932, SRMR = 0.062, RMSEA = 0.052). Although all parameters within the model indicated statistically significant t-values, one of the paths linking two latent constructs was non-significant. The standardized path coefficient from religious commitment (F3) to growth (F5) was not significant ( t = 1.87, se = 0.25, p = 0.06). Further, inspection of the squared multiple correlations table indicated that R-square values relating to the negatively worded optimism items (3, 7, and 9) were weak (< .25). Revised model. To look for the best fitting model, the Wald test and the Lagrange multiplier tables were consulted. The Wald test provides information on parameters that can be dropped to improve the model. The Lagrange multipliers provide information on parameters to be added. Experts caution researchers to ensure that data-driven model modifications do not capitalize on chance characteristics of the sample data, as they have the tendency to produce a final model that is not generalizable to the population or to other samples (O’Rourke & Hatcher, 2013; Schreiber, 2008). Researchers are therefore encouraged to identify parameters that could be dropped from the model without significantly affecting the model’s fit, as it is generally safer to drop parameters than to add new parameters when modifying models (O’Rourke & Hatcher, 2013). The Wald test suggested the intrapersonal variable within the religious commitment factor be dropped. Even though that suggestion was deemed statistically feasible, it was not theoretically feasible. Furthermore, because of the problems associated with the negatively worded items in the optimism scale, the errors associated with those items were allowed to covary. When the three errors were covaried, the model was reanalyzed. The maximum likelihood successfully converged in 19 iterations. The revised model fit the data well (CFI = .953; SRMR = 0.049; RMSEA = 0.044). All path coefficients were nontrivial and statistically significant (i.e., t > |1.96|). Figure 3 depicts standardized path coefficients for the revised model.