The Professional Counselor | Volume 11, Issue 3 273 following each deletion. Oblique factor rotation methods produce two matrices, including the pattern matrix, which displays the relationship between the items and a factor while controlling for the items’ association with the other factors, and the structure matrix, which depicts the correlation between the items and all of the factors (Mvududu & Sink, 2013). Researchers should examine both the pattern and the structure matrices and interpret the one that displays the clearest evidence of simple structure with the least evidence of cross-loadings. Items should display a factor loading of at least ≥ .40 (≥ .50 is desirable) to mark a factor. Items that fail to meet a minimum factor loading of ≥ .40 should be deleted. Cross-loading is evident when an item displays factor loadings ≥ .30 to .35 on two or more factors (Beavers et al., 2013; Mvududu & Sink, 2013; Watson, 2017). Researchers may elect to assign a variable to one factor if that item’s loading is .10 higher than the next highest loading. Items that cross-load might also be deleted. Once again, items should be deleted one at a time and the EFA should be re-computed after each removal. Naming the Rotated Factors The final step in EFA is naming the rotated factors; factor names should be brief (approximately one to four words) and capture the theoretical meaning of the group of items that comprise the factor (Mvududu & Sink, 2013). This is a subjective process, and the literature is lacking consistent guidelines for the process of naming factors. A research team can be incorporated into the process of naming their factors. Test developers can separately name each factor and then meet with their research team to discuss and eventually come to an agreement about the most appropriate name for each factor. Confirmatory Factor Analysis CFA is an application of structural equation modeling for testing the extent to which a hypothesized factor solution (e.g., the factor solution that emerged in the EFA or another existing factor solution) demonstrates an adequate fit with a different sample (Kahn, 2006; Lewis, 2017). When validating scores on a new test, investigators should compute both EFA and CFA with two different samples from the same population, as the emergent internal structure in EFA can vary substantially. Researchers can collect two sequential samples or they may elect to collect one large sample and divide it into two smaller samples, one for EFA and the second for CFA. Evaluating model fit in CFA is a complex task that is typically determined by examining the collective implications of multiple goodness-of-fit (GOF) indices, which include absolute, incremental, and parsimonious (Lewis, 2017). Absolute fit indices evaluate the extent to which the hypothesized model or the dimensionality of the existing measure fits with the data collected from a new sample. Incremental fit indices compare the improvement in fit between the hypothesized model and a null model (also referred to as an independence model) in which there is no correlation between observed variables. Parsimonious fit indices take the model’s complexity into account by testing the extent to which model fit is improved by estimating fewer pathways (i.e., creating a more parsimonious or simple model). Psychometric researchers generally report a combination of absolute, incremental, and parsimonious fit indices to demonstrate acceptable model fit (Mvududu & Sink, 2013). Table 1 includes tentative guidelines for interpreting model fit based on the synthesized recommendations of leading psychometric researchers from a comprehensive search of the measurement literature (Byrne, 2016; Dimitrov, 2012; Fabrigar et al., 1999; Hooper et al., 2008; Hu & Bentler, 1999; Kahn, 2006; Lewis, 2017; Mvududu & Sink, 2013; Schreiber et al., 2006; Worthington & Whittaker, 2006).
RkJQdWJsaXNoZXIy NDU5MTM1