| In traditional generalized linear model (GLM) theory, covariates are assumed to be non-stochastic. However, in real-life applications, covariates are generally stochastic in nature. Oral (2002), Oral and Gunay (2004), as well as Oral (2006) have shown that in a single covariate binary GLM, when the covariate is in fact stochastic, traditional non-stochastic modeling gives inaccurate and inefficient results. For a single covariate binary GLM they utilized modified maximum likelihood (MML) methodology to propose new estimators and test statistics. In this study, we generalize the methodology given in Oral and Gunay (2004) and Oral (2006) to the case when there is more than one covariate. We derive MML estimators and study their properties. We also study the effect of link misspecification in binary regression. |
