Solving unobserved heterogeneity with latent class inflated Poisson regression model

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Traditional Poisson regression models struggle with inflated count data and over-dispersion.
• Unobserved heterogeneity is a key driver of these modeling challenges.

Purpose of the Study:

• To propose a novel Latent Class Inflated Poisson (LCIP) regression model.
• To address unobserved heterogeneity causing inflation and over-dispersion in count data.

Main Methods:

• Development of the Latent Class Inflated Poisson (LCIP) regression model.
• Evaluation of model estimation performance via simulation studies.
• Application to Behavioral Risk Factor Surveillance System (BRFSS) data.

Main Results:

• The proposed LCIP model effectively handles inflated data and over-dispersion.
• Simulation studies confirmed the model's robust estimation performance.
• The LCIP model demonstrated a better fit for inflated counts than standard Poisson and zero-inflated Poisson models.

Conclusions:

• The LCIP regression model provides a significant improvement for analyzing count data with inflation and over-dispersion.
• Incorporating a latent class variable enhances the model's ability to capture complex data structures.
• The model proves useful for real-world datasets like the BRFSS, characterized by excessive values.