The dependability of general-factor loadings: The effects of factor-extraction methods, test battery composition, test battery size, and their interactions

To understand the extent to which the general-factor loadings of tests are inherent in their characteristics or due to the sampling of tests, the number of tests in the correlation matrix, and the factor-extraction methods used to obtain them, test scores from a large sample of young adults were inserted into independent and overlapping batteries of varying sizes. Principal factors analysis, maximum-likelihood estimation, and principal components analysis yielded general-factor loadings for each test. Generalizability theory analyses revealed that the characteristics of the tests consistently contributed the largest percentage of variance. Variance attributable to the factor-extraction method and its interactions was sizeable when principal components analysis was included in the analysis but negligible when it was excluded. Psychometric sampling error produced sizeable variance components in some analyses, and its effects were magnified when test batteries diminished in size. When results from principal components analysis were excluded and when the effects of psychometric sampling error were reduced, general-factor loadings were highly dependable across varying conditions.