Abstract
To find the best model for different link functions in Generalized Linear Models has been a challenge to researchers and users. Lindsey and Jones (1988) provided some procedures to suggest the techniques for choosing among the generalized linear models applied to medical data. They employed Akaike's Information Criteria (AIC). In the present study, an attempt is made to compare the performance in choosing the best GLM using the alternative procedures such as: deviance. Simulation from different distributional assumptions are performed to demonstrate the comparison mong the alternative approaches for various link functions. In addition to existing methods based on distributional assumptions, an alternative approach is suggested based on the quasi likelihood approach where no distributional assumptions is needed. The suggested methods for comparison of models are also applied to the same data set which Lindsey and Jones (1998) used in their study; data on T4 cells/mm3 in blood samples from 20 patients in remission from Hodgkins disease and 20 other patients in remission from disseminated malignancies. The performance of deviance appears to be satisfactory for this data.