TPC Journal V8, Issue 2 - FULL ISSUE

180 The Professional Counselor | Volume 8, Issue 2 following goodness-of-fit indices were reported: chi-square absolute fit index (CMIN), comparative fit index (CFI), root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), goodness-of-fit-index (GFI), and normed fit index (NFI). Two HLR analyses were computed to examine the predictive validity of the CMHPCS for both faculty member and student participants (research question 3). Previous investigators found group demographic differences in college students’ willingness to utilize mental health services by age (Eisenberg et al., 2016) and their willingness to make peer-to-peer referrals to resources by gender (Kalkbrenner & Hernández, 2017). Based on these findings, gender and age were entered into the first regression model as predictor variables. Participants’ composite scores on the knowledge, fear, and engagement scales of the CMHPCS were entered into the second regression model as predictor variables. The criterion variable was participants’ referrals to the counseling center (1 = has not made a referral to the counseling center, or 2 = has made referrals to the counseling center). Results After screening the data, descriptive statistics were computed on the faculty and student samples to examine unusual or problematic response patterns, missing data, and the parametric nature of the item distributions. Missing values analyses revealed that less than 2% of data was absent from faculty participants and less than 1% of data was absent from student participants. Both data sets were winsorized and missing values were replaced with the series mean (Field, 2018). Skewness and kurtosis values for items were largely within the acceptable range of a normal distribution (absolute value < 1) for the sample of faculty members and the sample of students (see Table 1). The findings are presented in three phases of analyses that correspond to the three research questions, respectively. Phase 1: Exploratory Factor Analysis A PFA was conducted using the sample of faculty members to establish the initial dimensionality of the CMHPCS (research question 1). The inter-item correlation matrix revealed low-to-moderate correlations among items ( r = .17 to r = .69). The Kaiser-Meyer-Olkin Measure of Sampling Adequacy (KMO = .81) and Bartlett’s Test of Sphericity ( B [153] = 1375.91, p < 0.001) provided further evidence that the data set was factorable. The oblique rotated PFA (direct oblimin, ∆ = 0) revealed a 5-factor solution based on the Kaiser criterion (Λ > 1.00). Seventy percent of the total variance in the correlation matrix was explained by these five factors. The scree plot, parallel analysis, and meaningful variance explained (at least 5% for each factor) that a 3-factor solution was the most parsimonious with the least evidence of cross-loadings (see Table 2). Five items displayed commonalities < .30 and were consequently removed from the analysis. The first factor accounted for 31.6% of the variance (Λ = 4.74), the second factor comprised 12.5% of the variance (Λ = 1.89), and the third factor accounted for 11.8% of the variance (Λ = 1.78). Redundant items that were highly correlated, and thus conceptually interrelated, were deleted. The inter-item correlation matrix was reproduced and indicated that item 8 (“I am aware of resources in the community for mental health”) and item 15 (“I am aware of the university resources for mental health”) were statistically and conceptually similar, suggesting that these items were measuring the same construct. Item 8 was subsequently removed, as the content of item 15 was more closely related to mental health services on campus. The PFA was recomputed and a final 3-factor solution (see Table 2) comprised of 12 items was retained. These 12 items were renumbered in chronological order.