Bayesian Estimation for Item Factor Analysis Models with Sparse Categorical Indicators.

Area of Science:

• Psychometrics
• Statistical Modeling

Background:

• Item factor analysis (IFA) models are crucial in psychology.
• Item sparseness, caused by small sample sizes or rare behaviors, poses estimation challenges for traditional methods like maximum likelihood (ML).

Purpose of the Study:

• To demonstrate the effectiveness of Bayesian estimation with prior information for IFA models facing item sparseness.
• To offer a reliable alternative to ML estimation in psychological research with sparse data.

Main Methods:

• A simulation study was conducted to compare Bayesian and ML estimation under conditions of item sparseness.
• Bayesian estimation incorporated general prior information applicable across various research contexts.

Main Results:

• Bayesian estimation significantly improved parameter estimate stability and statistical power for IFA models with sparse, categorical indicators.
• ML estimation methods showed a tendency to fail convergence or produce extreme estimates in sparse conditions.
• The proposed Bayesian priors did not negatively impact results when data were not sparse, showing comparability to ML.

Conclusions:

• Bayesian estimation with general prior information is a superior approach for handling item sparseness in IFA models within psychology.
• This method provides a valuable tool for researchers analyzing sparse data, such as in studies on suicide ideation and insomnia.