TPC Journal-Vol 10- Issue 3-FULL ISSUE

The Professional Counselor | Volume 10, Issue 3 381 The TKSS was adapted in collaboration with the authors to develop the SCKSS (Olsen, Blum, et al., 2016) to specifically target school counselors and to reflect the updated terminology recommended in the literature (Sugai & Horner, 2009). To update terminology, multi-tiered systems of support (MTSS) replaced Positive Behavior Supports (PBS) throughout the survey. In addition, school counselor replaced teacher to reflect the role of intended participants. Finally, item 6 was updated from “ I know how to access and use our school’s counseling programs ” to “ I know how to provide access and implement our school’s counseling programs ” because of school counselors’ roles and interactions with their own programs. Further, item 6 was adjusted to be an internally oriented question about the delivery of the school counseling program rather than the school counselor’s knowledge of another school service or system in order to assess participants’ perceived mastery of school counseling program implementation rather than their perception of another service not already measured in the SCKSS. A description of the 33 SCKSS items and the means and standard deviations of each item for the current study are located in Table 1. Data Analyses A cross-validation holdout method was used to examine the data–model fit of the SCKSS. Prior to statistical analyses, data were screened for missing data, multivariate outliers, and the assumptions for multivariate regression. Less than 5% of the data for any variable was missing and Little’s MCAR test (χ 2 = 108.47, df = 101, p = .29) indicated missing values could be considered as missing completely at random. Multiple imputation was used to estimate missing values. Although there were some outliers, results of a sensitivity analysis indicated that none of the outliers were overly influential. The assumptions of linearity, normality, multicollinearity, and homoscedasticity suggested that all the assumptions were tenable. The original sample ( N = 4,066) was randomly divided into two sub-samples ( N = 2,033). The first subset was used to conduct exploratory analyses and develop a model that fit the data. The second subset of participants was used to conduct confirmatory analyses without modifications. Exploratory Factor Analysis (EFA). Using the first subset from the sample, an EFA was conducted, using SPSS, to explore the number of factors and the alignment of items to factors. The number of factors extracted was estimated based on eigenvalues greater than 1.0 and a visual inspection of the scree plot. Several rotation methods were used, including varimax and direct oblimin with changing the delta value (from 0 to 0.2). The goal of the EFA was to find a factor solution that was theoretically sound. Confirmatory Factor Analysis (CFA) . The estimation method employed for the CFA was maximum likelihood robust estimation, which is a more accurate estimate for non-normal data (Savalei, 2010). Although the data were ordinal (i.e., Likert-type scale), Mplus uses a different maximum likelihood fitting function for categorical variables. The Satorra-Bentler scaled chi-square difference test was used to determine the best model. The pattern coefficient for the first indicator of each latent variable was fixed to 1.00. Indices of model–data fit considered were chi-square test, root-mean-square error of approximation (RMSEA), standardized root-mean-square residual (SRMR), comparative fit index (CFI), and Akaike information criterion (AIC). Browne and Cudeck (1993) suggested that values greater than .10 might indicate a lack of fit. In this study, an upper 90% confidence interval value lower than .08 was used to suggest an acceptable fit. CFI values greater than .90, which indicate that the proposed model is greater than 90% of the baseline model, served as an indicator of adequate fit (Kline, 2016). Perfect model fit is indicated by SRMR = 0, and values greater than .10 may indicate poor fit (Kline, 2016). Reliability was assessed using Cronbach’s alpha (α). CFAs were used in both the exploratory and confirmatory phases of this study. In the exploratory phase (i.e., using the first subset from the sample), the researchers used the residual estimates and modification indices to identify local misfit. Respecification of correlated error variances was expected because of the data collection method (i.e., counselors responding to a single survey) and similar wording of the items.