On the causal priority of latent variables over indicators in SEM: Three main arguments

The article below presents three main arguments in favor of the view that latent variables are always causes of indicators, in the broader context of structural equation modeling (SEM), and in the more specific context of structural equation modeling via partial least squares (PLS-SEM).

Kock, N., & Beauducel, A. (2026). On the causal priority of latent variables over indicators in SEM: Three main arguments. Data Analysis Perspectives Journal, 7(2), 1-9.

Link to full-text file for this and other DAPJ articles:

https://scriptwarp.com/dapj/#Published_Articles

Abstract:

The causal status of latent variables (LVs) is not always clearly presented in structural equation modeling (SEM). However, multivariate determination of causes, measurement error, and sampling error blur the one-to-one relationship of LVs to their indicators. Therefore, the impression that an indicator is the cause of a specific LV may occur because one or more other LVs causing the indicator as well as the specific LV are ignored. If indicators cause the LVs their complete variance may be used for modeling the LVs as in principal component analysis (PCA) and composite-based partial least squares modeling (PLS). In contrast, if indicators are caused by LVs, only the indicator variance caused by the LVs represents the true variance, which should be used for modeling as in factor analysis and factor-based SEM. Therefore, the assumed causal relationship between LVs and their indicators has consequences for model estimation. We discuss the causal relationship between LVs and their indicators.

Video for PLS-SEM workshop discussing the issues addressed in the article, as well as other issues:

Best regards to all!