A Doubly Winsorized Poisson Auto-model

  • Published : 1998.08.01


This paper introduces doubly Winsorized Poisson auto-model by truncating the support of a Poisson random variable both from above and below, and shows that this model has a same form of negpotential function as regular Poisson auto-model and one-way Winsorized Poisson auto-model. Strategies for maximum likelihood estimation of parameters are discussed. In addition to exact maximum likelihood estimation, Monte Carlo maximum likelihood estimation may be applied to this model.



  1. Journal of the Royal Statistical Society B v.36 Spatial interaction and the statistical analysis of lattice systems Besag, J.
  2. Biometrics v.43 Empirical Bayes estimates of age-standardized relative risks for use in disease mapping Clayton, D.;Kaldor, J.
  3. Journal of the American Statistical Association v.55 Estimation in the truncated Poisson distribution when zeros and some ones are missing Cohen, A.
  4. Statistics for Spatial Data(Revised edn.) Cressie, N.
  5. Journal of the Royal Statistical Society B v.54 Constrained Monte carlo moximum likelihood for dependent data (with discussion) Geyer, C.;Thompson, E.
  6. Journal of the Royal Statistical Society B v.56 On the convergence of Monte Carlo maximum likelihood calculations Geyer, C.
  7. Markov fields on finite graphs and lattices Hammersley, J.;Clifford, P.
  8. Preprint No. 96-30, Statistical Laboratory, Iowa State University The construction of multivariate distributions from Markov random fields Kaiser, M.;Cressie, N.
  9. Statistics and Probability letters Modeling Poisson random variables with positive spatial dependence Kaiser, M.;Cressie, N.
  10. Maximum likelihood estimation in conditionally specified statistical models Lee, J.;Kaiser, M.