Summed versus estimated factor scores: Considering uncertainties when using observed scores.

Area of Science:

• Psychological Science
• Psychometrics
• Quantitative Psychology

Background:

• Observed scores are frequently used to represent underlying constructs in psychological research.
• Latent variables (LVs) are theoretical constructs operationalized through measurement models, commonly the common factor model.
• Understanding the performance of different observed scores is crucial for accurate representation of LVs.

Purpose of the Study:

• To evaluate the performance of various observed scores for estimating latent scores and classifying individuals.
• To assess the ability of different observed scores to recover structural relationships among latent variables.
• To account for sampling error and model error in the evaluation of observed score performance.

Main Methods:

• Review of psychometric properties of observed scores within classical test theory and common factor models.
• Conducting a simulation study to examine score performance under varying conditions.
• Analysis of two empirical examples to demonstrate score behavior with different reliability, sample sizes, and model error levels.

Main Results:

• Performance of different observed scores varies significantly depending on the specific application (e.g., estimation vs. structural recovery).
• Reliability, sample size, and model error critically influence the accuracy of latent score estimation and structural recovery.
• Certain observed scores demonstrate superior performance under specific conditions of uncertainty and measurement quality.

Conclusions:

• Provides evidence-based recommendations for selecting appropriate observed scores in psychological research.
• Highlights the importance of considering measurement error and model misspecification when using observed scores.
• Suggests future research directions for advancing the understanding and application of observed scores in latent variable modeling.