Testing for zero-inflation in count series: application to occupational health

Abstract

Count data series with extra zeros relative to a Poisson distribution are common in many biomedical applications. A score test is presented to assess whether the zero-inflation problem is significant to warrant the analysis by the more complex zero-inflated Poisson autoregression model. The score test is implemented as a computer program in the Splus platform. For illustration, the test procedure is applied to a workplace injury series where many zero counts are observed due to the heterogeneity in injury risk and the dynamic population involved.

Keywords:

• occupational health
• random effects
• score test
• workplace injuries
• zero-inflation

Acknowledgements

This research is supported by a project grant (ID 425510) from the National Health and Medical Research Council of Australia. The test procedure has been implemented as an Splus computer program available from the corresponding author.