Formative versus reflective measurement models in nursing research: a secondary data analysis of a cross-sectional...

This study aimed to empirically verify the impact of measurement model selection on research outcomes and their interpretation through an analysis of children's emotional and social problems measured by the Pediatric Symptom Checklist (PSC) using both reflective and formative measurement models. These models were represented by covariance-based structural equation modeling (CB-SEM) and partial least squares SEM (PLS-SEM), respectively.

This secondary data analysis evaluated children's emotional and social problems as both reflective and formative constructs. Reflective models were analyzed using CB-SEM, while formative models were assessed using PLS-SEM. Comparisons between these two approaches were based on model fit and parameter estimates.

In the CB-SEM analysis, which assumed a reflective measurement model, a model was not identified due to inadequate fit indices and a Heywood case, indicating improper model specification. In contrast, the PLS-SEM analysis, assuming a formative measurement model, demonstrated adequate reliability and validity with significant path coefficients, supporting the appropriateness of the formative model for the PSC.

The findings indicate that the PSC is more appropriately analyzed as a formative measurement model using PLS-SEM, rather than as a reflective model using CB-SEM. This study highlights the necessity of selecting an appropriate measurement model based on the theoretical and empirical characteristics of constructs in nursing research. Future research should ensure that the nature of measurement variables is accurately reflected in the choice of statistical models to improve the validity of research outcomes.