Through use of confirmatory maximum likelihood factor analysis in conjunction with the LISREL V computer program devised by Jöreskog and Sörbom, an evaluation was made of the validity of higher-order ability constructs in structure-of-intellect tests all containing semantic content and operations of cognition or evaluation. The hypothesized first-, second-, and third-order factors were all reproducible with every one of the estimated factor loadings being significant beyond the .01 level. Although the first- and second-order factors were shown to be reproducible and statistically separable, the high intercorrelations among the six first-order product factors and the two second-order operations factors would suggest from a practical standpoint that the single third-order factor of semantic content would constitute a plausible alternative for accounting for much of the covariance among the test variables.
Maximum likelihood factor analysis with rank-deficient sample covariance matrices
