A new factor analysis model for factors obeying a Gamma distribution.

Area of Science:

• Statistics
• Data Analysis

Background:

• Traditional factor analysis assumes normally distributed factors, unsuitable for nonnegative data.
• Nonnegative data is prevalent in various scientific fields, necessitating alternative models.

Purpose of the Study:

• To develop a novel factor analysis model for nonnegative data using Gamma distribution.
• To evaluate the parameter estimation and information extraction ability of the new model.

Main Methods:

• Constructed a new factor analysis model with Gamma-distributed factors.
• Employed Maximum Likelihood Estimation (MLE) via an Expectation-Maximization (EM) algorithm.
• Utilized the Metropolis-Hastings (M-H) algorithm within Markov Chain Monte Carlo (MCMC) for the E-step.

Main Results:

• The new Gamma factor model was applied to real and simulated nonnegative data.
• The model's information extraction ability was assessed using a defined true loading matrix.
• The Gamma factor model outperformed traditional models in information extraction for nonnegative data with an equal number of factors.

Conclusions:

• The proposed Gamma factor analysis model provides a more practical approach for analyzing nonnegative data.
• This model offers enhanced information extraction compared to traditional methods in specific contexts.
• The study validates the utility of Gamma distribution in factor analysis for nonnegative datasets.