Bayesian Gaussian distributional regression models for more efficient norm estimation.

Area of Science:

• Psychometrics
• Statistical modeling
• Psychological assessment

Background:

• Psychological test scores are typically normed scores relative to a reference population.
• Norm estimation often requires large normative samples to accurately model predictor relationships (e.g., age) with score distributions.
• This requirement poses a significant burden in terms of sample size and cost.

Purpose of the Study:

• To investigate the efficiency and robustness of Bayesian Gaussian distributional regression for norm estimation.
• To determine if incorporating prior information can alleviate the need for large normative samples.
• To assess the impact of prior type and misspecification on norm estimation accuracy.

Main Methods:

• A simulation study was conducted using Bayesian Gaussian distributional regression.
• Prior information was incorporated to estimate new norms.
• Simulation parameters included prior type, prior misspecification (age-dependent vs. independent), and sample size.

Main Results:

• Using a fixed effects prior led to more efficient norm estimation compared to a weakly informative prior, provided prior misspecification was not age-dependent.
• The proposed Bayesian method, with reasonable prior information, achieved comparable norm precision with smaller normative samples.
• The method demonstrated robustness to certain prior model deviations.

Conclusions:

• Bayesian Gaussian distributional regression offers a more efficient approach to norm estimation for psychological tests.
• Incorporating prior information can significantly reduce the required sample size for high-quality norm development.
• This method facilitates cost-efficient norm creation, benefiting test developers and users.