Modeling Temporally Correlated Multivariate Counts with Excess Zeros

• Activity Number: 317
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: SSC
• Author(s): Gary Sneddon* and Tariqul Hasan and Renjun Ma
• Companies: Mount Saint Vincent University and University of New Brunswick and University of New Brunswick
• Keywords: compound Poisson ; discrete observations ; correlation ; time series ; quasi-likelihood ; zero-inflated

Abstract

Multivariate discrete time series counts with excessive zeros frequently occur in in a variety of disciplines. An appropriate analysis of such data has to account for serial correlation over time, correlation between the multivariate responses, and excessive zeros. Our proposed model uses a serially correlated random effect series shared by each set of responses to account for a common temporal trend. A key feature of our model is the incorporation of a compound Poisson random effect series into a Poisson model for each response to capture excessive zeros and additional variation. We will discuss parameter estimation for this model and illustrate the method in a numerical example.