Biased latent variable mean comparisons due to measurement non-invariance: a simulation study

Abstract

Making valid comparisons of latent variable (LV) mean scores 1 across groups is anything but a trivial task as it relies on stringent measurement invariance conditions.Many authors (e.g., Little, 1997;Meredith, 1993;Van de Vijver & Leung, 1997) have firmly stated that in order for such comparisons to be valid, all factor loadings as well as all indicator intercepts need to be invariant across groups (i.e., they are required to exhibit scalar invariance, also referred to as full score invariance across groups; Van de Vijver & Leung, 1997, p. 144).In contrast, some authors have promoted the use of less stringent measurement invariance conditions.For instance, according to Alwin and Jackson (1981) the equality of all factor loadings (i.e., an invariance condition referred to as "metric invariance" across groups) would 1 In this study, we deal only with the comparison of LV means across groups.Cross-group comparisons of structural relationships between observed or LVs are not considered.Such comparisons require only the factor loadings of LV indicators to be identical across the groups (i.e., metric invariance across groups) involved in the comparison (see, for instance, Van de Vijver & Leung, 1997, De Beuckelaer et al., 2007).RT8233X_C005.indd119 7/7/10 11:02:11 PM RT8233X_C005.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 39 citations exist upstream.