Poisson regression is useful in modeling count data. In a study with many independent variables, it is desirable to reduce the number of variables while maintaining a model that is useful for prediction. This article presents a variable selection technique for Poisson regression models. The data used is log-linear, but the methods could be adapted to other relationships. The model parameters are estimated by the method of maximum likelihood. The use of measures of goodness-of-fit to select appropriate variables is discussed. A forward selection algorithm is presented and illustrated on a numerical data set. This algorithm performs as well if not better than the method of transformation proposed by Nordberg (1982).
Famoye, Felix and Rothe, Daniel E.
"Variable Selection for Poisson Regression Model,"
Journal of Modern Applied Statistical Methods: Vol. 2
, Article 11.
Available at: http://digitalcommons.wayne.edu/jmasm/vol2/iss2/11