ERIC Number: ED364552
Record Type: Non-Journal
Publication Date: 1993-Apr
Reference Count: N/A
Alternative Models for Bilevel Factor Analysis.
Mok, Magdalena M. C.; McDonald, Roderick P.
The aim of this paper is to: (1) discuss the characteristics of seven bilevel factor models; and (2) arrive at a list of empirically based recommendations for selecting appropriate bilevel factor models, the assumptions underlying these decisions, and the consequences for interpretation of the Spearman case. The five models considered in the context of empirical data on school culture are: (1) no scaling, loadings constrained across levels; (2) no scaling, distinct loadings permitted; (3) scaling at both levels, constrained equal loadings; (4) scaling at both levels, constrained equal loadings, estimating school-level factor variance; and (5) scaling at both levels, distinct loadings. Analyses suggest that choice among models can depend on whether distinct factor structures across levels can be interpreted meaningfully, whether standardization of variables is necessary, and whether the researcher wants to estimate second level factor variance. If distinct factor structures are interpretable, then isomorphic models 2 and 5 would provide the best fit, and would, in most cases, provide no better fit than the more parsimonious model 4. Parsimony and ease of interpretation favor model 4. Model 1 can provide a direct estimate of school effect not available from other models. Each of the models, except for model 3, provides useful and complementary information. Three figures and two tables illustrate the discussion. (Contains 25 references.) (SLD)
Descriptors: Comparative Analysis, Educational Environment, Elementary Secondary Education, Estimation (Mathematics), Factor Analysis, Foreign Countries, Goodness of Fit, Models, Research Methodology, Scaling, Selection
Magdelena Mok, School of Education, Macquarie University, New South Wales 2109, Australia.
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Sponsor: Macquarie Univ., North Ryde (Australia).
Authoring Institution: N/A