Sankhya: The Indian Journal of Statistics

1996, Volume 58, Series A, Pt. 3, pp. 473--490

TESTS FOR OVERDISPERSION IN A REGRESSION MODEL FOR THE TIME DEPENDENT TRUNCATED DATA

By

KALYAN DAS, Calcutta University

and

BRAJENDRA C. SUTRADHAR, Memorial University of Newfoundland

SUMMARY.  A parameter driven regression model for a truncated natural exponential family is considered where the outcomes are time dependent and also overdispersed. In such models, these time dependence and overdispersion are usually caused by an unobservable correlated noise process. Similar to Gurmu and Trivedi (1992), a score test for examining the overdispersion has been considered. An asymptotic test based on the quasilikelihood approach has also been developed. The latter test is much simpler to compute and is recommended for use in practice.

AMS (1980) subject classification.  Primary 62H15.

Key words and phrases. Correlated as well as truncated data, overdispersion, partial score test, quasi-likelihood test, asymptotic chi-square distribution.

Full paper (PDF)

This article in Mathematical Reviews.