Why Formative Latent Variables Fall Short in Psychology Research

A recent study questions whether formative latent variables can truly capture psychological concepts. Researchers revisited a 2005 paper that warned about bias in path estimates when formative models are treated as reflective ones.

The new analysis used simulations to test the original claims. It found that the bias in unpaired path coefficients was not caused by model misspecification. Instead, scaling differences between variables explained the distortion. Once those differences were adjusted for, the bias nearly vanished.

The study also challenged the need for formative indicators. These indicators were meant to define the latent variable. But the research showed they were redundant. The latent variables got their identity from reflective indicators used for model identification, not from the formative ones.

Removing one or all formative indicators had little effect on path estimates between latent variables. This suggests that formative indicators may not play the role researchers assume.

The core issue is identity. Formative latent variables, whether first-order or higher-order, rely on downstream reflective indicators for their meaning. This creates a problem. If the identity comes from reflective rather than formative indicators, the latent variable cannot represent psychological concepts as intended.

This has big implications for measurement and index construction. Researchers should rethink how they build and interpret formative latent variables in psychological models.