Fitting One Factor to Two Traits: Population Solutions and Fit Indices
When a scale contains items measuring two related traits, researchers often fit a one-factor model to assess whether the items can be treated as a single dimension. Even when the model fits well, it can be unclear how strongly the fitted factor represents each trait. We address this problem by deriving the best one-factor approximation to a two-trait population under maximum likelihood. For two equally long item blocks with a common population loading, we obtain an exact threshold for the trait correlation. At and above the threshold, the unique minimizing covariance gives both blocks equal loadings. Below it, two distinct covariances fit equally well, one favoring each trait. At zero correlation, each solution reproduces one block exactly and assigns zero loadings to the other. When equally long blocks differ in loading strength, we determine when a second local minimum favors the more weakly measured trait. The solutions also yield exact population fit indices. With loadings of .7 and 4 to 16 items, CFI reaches .95 at trait correlations of .78 to .84. In simulations, changing which block was listed first could change which trait the fitted factor favored.
Authors
- Aleksandar Tomašević (ORCID: https://orcid.org/0000-0003-4863-6051)
Publication Details
- Journal
- PsyArXiv (OSF Preprints)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.31234/osf.io/k9bx3
- Primary Topic
- Psychometric Methodologies and Testing
- Type
- preprint