Abstract
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.
Similar content being viewed by others
Notes
The canonical link function is the log of the expected value of the response variable.
A wide variety of other examples could serve just as well. For example, the observational units could be neighborhoods, the response variable could be the number of burglaries in a given month, and among the predictors could be median household income and the proportion of the male population between 15 and 25 years of age.
The Poisson processes are different because they have different values for λ, conditional on age.
The variance of a sum of random variables is a linear combination of their variances and covariances (Freund 1971, p. 195). If those covariances are positive, the variance of the sum is inflated. If those covariances are negative, the variance of the sum is deflated. The Poisson distribution assumes that all of the covariances are zero.
This follows from the assumptions about ɛ i .
Without normalization, the model’s constant would not be identified.
As θ → ∞, one is back to the Poisson distribution. If θ < 0, there is underdispersion.
In order to have sufficient observations to get a good reading on the conditional variance, ranges of conditional means are used.
References
Berk RA (2003) Regression analysis: a constructive critique. Sage Publications, Newbury Park, CA.
Bottcher JB, Ezell ME (2005) Examining the effectiveness of boot camps: a randomized experiment with long-term follow up. J Res Crime Delinq 42:309–332
Box GEP (1976) Science and statistics. J Am Stat Assoc 71:791–799
Braga A (2003) Serious youth gun offenders and the epidemic of youth violence in Boston. J Quant Criminol 19:33–54
Cameron AC, Trivedi PK (1998) Regression analysis of count data. Cambridge University Press, Cambridge
Cook RD, Weisberg S (1999) Applied regression including computing and graphics. Wiley, New York, NY
Cox DR (1955) Some statistical models related with a series of events. J R Stat Soc B 17:406–424
de Leeuw J (1994) Statistics and the sciences. In: Borg I, Mohler PP (eds) Trends and perspectives in empirical social science. Walter de Gruyter, New York, pp 131–148
D’Unger A, Land K, McCall P, Nagin D (1998) How many latent classes of deliquent/criminal careers? Results from mixed Poisson regression analyses of the London, Philadelphia, and Racine cohorts studies. Am J Sociol 103:1593–1630
Duwe G, Kovandzic T, Moody C (2002) The impact of right-to-carry concealed firearm laws on mass public shootings. Homicide Stud 6:271–296
Freedman DA (1987) As others see us: a case study in path analysis. (with discussion). J Educ Stat 12:101–223
Freedman DA (2005) Statistical models: theory and practice. Cambridge University Press, Cambridge
Fruend JE (1971) Mathematical statistics, 2nd edn. Prentice Hall, New York
Greene WH (2003) Econometric analysis, 5th edn. Prentice Hall, New York
Heckman J (1999) Causal parameters and policy analysis in economics: a twentieth century retrospective. Q J Econ 115:45–97
Lattimore PK, MacDonald JM, Piquero AR, Linster RL, Visher CA (2004) Studying frequency of arrest among paroled youthful offenders. J Res Crime Delin 41:37–57
Manski CF (1990) Nonparametric bounds on treatment effects. Am Econ Rev Papers Proc 80:319–323
McCullagh P, Nelder JA (1989) Generalized linear models, 2nd edn. Chapman and Hall, New York
Morgan SL, Winship C (2007) Counterfactuals and causal inference: methods and principle for social research. Cambridge University Press, Cambridge
Nagin DS (2005) Group-based modeling of development. Harvard University Press, Cambridge, MA
Nagin DS, Land KS (1993) Age, criminal careers, and population heterogeneity: specification and estimation of a nonparametric, mixed Poisson model. Criminology 31:501–523
Osgood W (2000) Poisson-based regression analysis of aggregate crime rates. J Quant Criminol 16:21–43
Osgood DW, Chambers JM (2000) Social disorganization outside the metropolis: an analysis of rural youth violence. Criminology 38:81–115
Parker K (2004) Industrial shift, polarized labor markets and urban violence: modeling the dynamics between the economic transformation and disaggregated homicide. Criminology 42:619–645
Paternoster R, Brame R (1997) Multiple routes to delinquency? A test of developmental and general theories of crime. Criminology 35:45–84
Paternoster R, Brame R, Bachman R, Sherman L (1997) Do fair procedures matter? The effect of procedural justice on spouse assault. Law Soc Rev 31:163–204
Piquero AR, MacDonald JM, Parker KF (2002) Race, local life circumstances, and criminal activity. Soc Sci Q 83:254–270
Poole EE, Regoli RM (1983) Violence in juvenile institutions. Criminology 21:213–232
Ripley BD (2006) Stochastic Simulation. John Wiley and Sons, New York
Sampson RJ, Laub JH (1997) Socioeconomic achievement in the life course of disadvantaged men: Military service as a turning point, circa 1940–1965. Am Sociol Rev 61:347–367
Stucky TD (2003) Local politics and violent crime in US cities. Criminology 41:1101–1135
Acknowledgments
We are indebted to David Freedman, David McDowall, and the anonymous reviewers for their helpful suggestions. All errors and omissions remain those of the authors.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Berk, R., MacDonald, J.M. Overdispersion and Poisson Regression. J Quant Criminol 24, 269–284 (2008). https://doi.org/10.1007/s10940-008-9048-4
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10940-008-9048-4