Analysis of Clustered Count Data with Excessive Zeros

• Activity Number: 82
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): On Yee Tang*+
• Companies: University of Hong Kong
• Keywords: clustered data ; count data ; frailty ; multiple imputation ; Poisson regression ; zero-inflated

Abstract

In many medical or public health investigations, the count data encountered often exhibit an excess of zeros, and very frequently this type of data are collected on clusters of subjects or items. Poisson regression model with frailty is adopted to analyze this type of clustered zero-inflated count data. The noncentral chi-square distribution with zero degrees of freedom is proposed to model the random effects, which not only account for the subject specific heterogeneity, but also the dependency among subjects within a cluster. The use of this special distribution can provide more flexibility on the relationship between the covariates and the random effects. A simple multiple imputation approach is proposed for the parameter estimation of this model. The proposed methodology are motivated and illustrated by the public health survey conducted in Indonesia, which employs a multilevel cluster sampling scheme on the number of days of missing primary activities due to illness in a four-week period and the number of days in beds due to illness in a four-week period.