SASS618-OUP.docVIP

Running Higher-order CFA with AMOS AMOS Files: ...\OUPSEM\CFA\Second-order1.amw ...\OUPSEM\CFA\Second-order2.amw ...\OUPSEM\CFA\Second-order3.amw Data Files: ...\OUPSEM\CFA\cfa.xls Why higher-order? Psychometric literature suggests that higher-order factors should be extracted whenever factors are correlated (oblique). Gorsuch (1983): “Rotating obliquely in factor analysis implies that the factors do overlap and that there are, therefore, broader areas of generalizability than just a primary factor.” In scale development, whenever one identifies multidimensional structure of the data and develops subscales, the natural question for next step is whether or not further generalizability can be achieved, and out of the so-developed subscales whether it is possible to develop a “Total” scale combining these subscales? Examples CBCL, a widely employed instrument measuring children’s behavioral problem with several subscales. In addition, there is a Total score. The Total score is valid because higher-order CFA supports a unidimensionality of data out of the first-order factors. In contrast, SPTE (Student Perceptions of Teaching Effectiveness) includes 47 items. A first-order CFA suggests 6 subscales: course organization and design, rapport with students, grading quality, course value, course difficulty, and workload. A second-order CFA suggests two second-order scales: (1) perceived quality index comprised of the 4 first-order subscales; and (2) perceived course demands comprised of the last 2 first-order subscales (Harrison, et al, 2004, Research in Higher Education 45(3): 311-323). Primary features: The path between FF1 and F1 (or F2, or F3) is one-headed straight line. Direct causal: the lower-order factors are determined by the high-order factor. For the purpose of identification: you generally need three 1st-order factors for each 2nd-order factor; also in general you need three indicators per 1st-order factor. Identification: e