Referent Indicators in Tests of Metric Invariance

Abstract

Organizations frequently administer surveys and psychological measures to multiple groups (e.g., cultural and demographic groups). However, before making direct cross-group comparisons, researchers need to ensure that the psychometric properties of these measures do not differ by groups. In order to test this hypothesis of measurement invariance, many researchers employ confirmatory factor analytic tests of measurement invariance. These tests require a referent indicator (RI) for model identification. This RI is assumed to be perfectly invariant across groups. Using simulated data, results indicate that inappropriate RI selection may be mildly problematic for scale-level invariance tests and highly problematic for item-level tests. These findings underscore the importance of careful RI selection.