Statistical Modelling 1 (2001), 65–80
Models for zero-inflated count data using the Neyman type A distribution
Melissa Dobbie
Centre for Mathematics and its Applications,
School of Mathematical Sciences,
Australian National University,
Canberra ACT 0200 Australia
eMail:
melissa.dobbie@anu.edu.au
A H Welsh
Faculty of Mathematical Studies,
The University of Southampton,
Highfield, Southampton S017 1BJ UK
Abstract:
We explore the possibility
of modelling zero-inflated count data using the Neyman
type A distribution. We extend three parameterisations
of the Neyman type A distribution
to allow their parameters to depend on covariates. We develop
models which relate counts of Leadbeater's possum to various
habitat variables to
illustrate the methodology. Half-normal plots are constructed
for each model to explore
the quality of the fit. We then formally compare the Neyman type A models
using the method of Cox to test non-nested hypotheses.
Finally, we compare each of the Neyman type A models with a model from
a competing family, the conditional Poisson model .
Keywords:
contagious distributions;
covariate adjustment; Neyman type A
distribution; non-nested hypothesis testing; parameterisation; zero-inflated
counts.
Downloads:
Software in
zipped file
back