Illustrates how categorization spuriously influences apparent dimensionality inferred from (a) principal components (PC), (b) exploratory maximum likelihood (EML) analysis, and (c) LISREL. Simulated continuous, parallel, unifactor "scores," of differing reliability, were categorized in various ways to creat "items." All forms of categorization spuriously suggested multidimensionality. PC-based indices were more misleading with less reliable data; the reverse was true with inferential (EML and LISREL) indices. Varying item "splits" to create item distribution differences further enhanced these spurious effects. Likewise, multicategory (Likert-type) items were more likely to yield artifacts than dichotomous items using inferential criteria even though the multicategory data were more reliable. Criteria for dimensionality applicable to continuous (scale-level) data are therefore inappropriate for discrete (item-level) data.
ASJC Scopus subject areas