Overdispersion and Poisson Regression

This article discusses the use of regression models for count data. A claim is often made in criminology applications that the negative binomial distribution is the conditional distribution of choice when for a count response variable there is evidence of overdispersion. Some go on to assert that th...

Full description

Saved in:  
Bibliographic Details
Authors: Berk, Richard (Author) ; MacDonald, John M. (Author)
Format: Electronic Article
Language:English
Published: 2008
In: Journal of quantitative criminology
Year: 2008, Volume: 24, Issue: 3, Pages: 269-284
Online Access: Volltext (lizenzpflichtig)
Volltext (lizenzpflichtig)
Journals Online & Print:
Drawer...
Check availability: HBZ Gateway
Keywords:
Overdispersion
Count data
Negative binomial distribution
Poisson regression
Description
Summary:This article discusses the use of regression models for count data. A claim is often made in criminology applications that the negative binomial distribution is the conditional distribution of choice when for a count response variable there is evidence of overdispersion. Some go on to assert that the overdisperson problem can be “solved” when the negative binomial distribution is used instead of the more conventional Poisson distribution. In this paper, we review the assumptions required for both distributions and show that only under very special circumstances are these claims true.
ISSN:1573-7799
DOI:10.1007/s10940-008-9048-4