Response variables that are scored as counts and that present a large number of zeros often arise in quantitative health care analysis. We define a zero-in flated Poisson model with fixed-effects in both of its equations to identify respondent and health-related characteristics associated with health care demand. This is a new model that is proposed to model count measures of health care utilization and account for the panel structure of the data. Parameter estimation is achieved by conditional maximum likelihood. An application of the new model is implemented using micro level data from the 2004-2006 Survey of Health, Ageing and Retirement in Europe (SHARE), and compared to existing panel data models for count data. Results show that separately controlling for whether outcomes are zero or positive in one of the two years does make a difference for counts with a larger number of zeros.
|Place of Publication||Tilburg|
|Publication status||Published - 2011|
|Name||CentER Discussion Paper|
- Count Data
- Zero-In ated Poisson Model