Related Experiment Videos
Construct validity in health behavior research: interpreting latent variable models involving self-report and
Raymond F Palmer1, John W Graham, Bonnie Taylor
1Department of Biobehavioral Health, Pennsylvania State University, University Park, Pennsylvania 16802, USA. palmerr@uthscsa.edu
Journal of Behavioral Medicine
|December 5, 2002
Summary
Latent variable models can yield misleading results when combining diverse data sources. Careful specification of residual covariances is crucial for accurate interpretation in health research.
Area of Science:
- Psychometrics
- Health Research Methodology
Background:
- Latent variable models (LVMs) are used to assess common variance among indicators, often to mitigate measurement error.
- Interpreting latent constructs becomes complex when indicators originate from disparate measurement domains (e.g., self-report vs. biochemical data).
Purpose of the Study:
- To demonstrate how model fit can be misleading when residual covariances are improperly specified in LVMs.
- To highlight the critical need for careful data specification when defining latent variables from mixed-domain indicators.
Main Methods:
- Utilized simulated data to evaluate the impact of residual covariance specification on LVM interpretation.
- Assessed model fit using conventional criteria.
Main Results:
- Even models with "good fit" can lead to erroneous interpretations if residual covariances are not appropriately handled.
- The choice of indicators and potential biases significantly affect the definition and predictive power of latent variables.
Conclusions:
- Findings underscore the importance of meticulous indicator selection and residual covariance specification in health-related LVMs.
- Improper LVM specification can dramatically alter the conceptualization and observed effects of latent variables in health research.