Estimating Regression Effects in Com Poisson Generalized Linear Model

Authors: Vandna Jowaheer, Naushad A. Mamode Khan

Abstract:

Com Poisson distribution is capable of modeling the count responses irrespective of their mean variance relation and the parameters of this distribution when fitted to a simple cross sectional data can be efficiently estimated using maximum likelihood (ML) method. In the regression setup, however, ML estimation of the parameters of the Com Poisson based generalized linear model is computationally intensive. In this paper, we propose to use quasilikelihood (QL) approach to estimate the effect of the covariates on the Com Poisson counts and investigate the performance of this method with respect to the ML method. QL estimates are consistent and almost as efficient as ML estimates. The simulation studies show that the efficiency loss in the estimation of all the parameters using QL approach as compared to ML approach is quite negligible, whereas QL approach is lesser involving than ML approach.

Keywords: Com Poisson, Cross-sectional, Maximum Likelihood, Quasi likelihood

Digital Object Identifier (DOI): doi.org/10.5281/zenodo.1075883

References:

[1] G. Bates and J. Neymann, " Contribution to the theory of accident processes", Publications in Statistics Vol 1, 215-253, 1952, University of California.
[2] N. Breslow and X. Lin , " Bias correction in generalized linear mixed models", Biometrics Vol 82, 81-91, 1995.
[3] R. Conway and W. Maxwell, " A queuing model with state dependent service rates", Journal of Industrial Engineering, Vol 12, 132-136, 1962.
[4] C. Dean and R. Balshaw, " Efficiency lost by analyzing counts rather than even times in Poisson and overdispersed Poisson regression models", Journal of American Statistical Association, Vol 92, 1387- 1398, 1997.
[5] D. Firth and L. Harris, " Quasi-likelihood for multiplicative random effects", Biometrics, Vol 78, 545-555, 1991.
[6] S. Guikema, " Formulating informative, data(based priors for failure probability estimation in reliability analysis", Reliability Engineering and System Safety, Vol 92, 490-502, 2007.
[7] S. Guikema and J. Gofelt, " A flexible count data regression model for risk analysis", Risk analysis, Vol 28, 213-223, 2008.
[8] J. Kadane, G. Shmueli, G. Minka, T. Borle and P. Boatwright, " Conjugate analysis of the Conway Maxwell Poisson distribution", Bayesian analysis, Vol 1, 363-374, 2006.
[9] J. Klein, " Semi parametric estimation of random effects using Cox model based on the EM algorithm", Biometrics , Vol 48, 795-806, 1992.
[10] D. Lord , S. Washington and J. Ivan, " Poisson, Poisson(Gamma and Zero inflated regression models of motor vehicle crashes: Balancing statistical fit and theory", Accident analysis and Prevention, Vol 37, 35- 46, 2005.
[11] T. Minka, G. Shmueli, J. Kadane, J. Borle and P. Boatwright," Computing with the Com Poisson distribution", Technical report CMU statistics department, http:// www.stat.cmu.edu/tr/tr776/tr776.html, 2003.
[12] D. Moore, " Asymptotic properties of moment estimators for overdispersed counts and proportions", Biometrics Vol 73, 583-588, 1986.
[13] G. Shmueli, T. Minka, J. Borle and P. Boatwright, " A useful distribution for fitting discrete data", Applied Statistics, Journal of Royal Statistical Society, 2005.
[14] B. Sutradhar and K. Das, " A higher order approximation to likelihood inference in the Poisson mixed model", Statist, Prob, Lett, Vol 52, 59- 67, 2001.
[15] R. Wedderburn, " Quasi-likelihood functions, generalized linear models and the Gauss Newton method, Biometrics, Vol 61, 439-447, 1974.