A Causal Framework for the Comparability of Latent Variables

Abstract

Measurement invariance (MI) describes the equivalence of measurement models of a construct across groups or time. When comparing latent means, MI is often stated as a prerequisite of meaningful group comparisons. The most common way to investigate MI is multi-group confirmatory factor analysis (MG-CFA). Although numerous guides exist, a recent review showed that MI is rarely investigated in practice. We argue that one reason might be that the results of MG-CFA are uninformative as to why MI does not hold between groups. Consequently, under this framework, it is difficult to regard the study of MI an interesting and constructive step in the modeling process. We show how directed acyclic graphs (DAGs) from the causal inference literature can guide researchers in reasoning about the causes of non-invariance. For this, we first show how DAGs for measurement models can be translated into the path diagrams used in the linear structural equation model (SEM) literature. We then demonstrate how insights gained from this causal perspective can be used to explicitly model encoded causal assumptions with moderated SEMs, allowing for a more enlightening investigation of MI. Ultimately, our goal is to provide a framework in which the investigation of MI is not deemed a “gateway test” that simply licenses further analyses. By enabling researchers to consider MI as an interesting part of the modeling process, we hope to increase the prevalence of investigations of MI altogether.

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 — 31 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.