Modeling health survey data with excessive zero and K responses

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Zero-inflated Poisson (ZIP) regression is widely used for data with excess zeros.
• Existing models often struggle with data exhibiting peaks at values other than zero.
• These limitations stem from models heavily relying on specific data features.

Purpose of the Study:

• To propose a novel statistical model for count data with excessive zeros or other specific peak values.
• To develop a flexible mixture model combining multinomial logistic and Poisson regression.
• To address limitations of standard ZIP models in capturing non-zero excessive counts.

Main Methods:

• A mixture model combining multinomial logistic and Poisson regression was developed.
• The multinomial logistic component models excessive counts, including zeros and other specific values (K).
• The Poisson component models counts following a Poisson distribution.

Main Results:

• The proposed model demonstrated flexibility in handling excessive counts beyond zero.
• Evaluated on data with peaks at ones and sixes, the K-inflated model showed superior fit.
• The K-inflated and zero-inflated models outperformed standard Poisson and ZIP regressions.

Conclusions:

• The proposed mixture model provides a more flexible approach for analyzing count data with excessive zeros or other specific values.
• This K-inflated model offers improved performance over traditional methods when data exhibit non-zero peaks.
• The findings suggest broader applicability of this flexible modeling strategy in various fields.