Some model misspecifications in logistic regression model

Correct specification of the model is the most important assumption for the logistic regression model, as for all models. It means that the model has the correct functional form, does not include irrelevant variables and has all the relevant variables. Previous studies show that misspecification may cause undesirable results such as biased logistic regression coefficients, inefficient estimates, invalid statistical inferences and less efficient test statistics. In this thesis, the effects of misspecification on asymptotic relative efficiency of various coefficients of determination are investigated. Misspecification types include using wrong functional form of explanatory variable, categorizing continuous explanatory variable and omitting the covariate. Unlike linear regression model, there is not only one coefficient of determination in logistic regression, which makes the results of this thesis more important. Simulation studies using bootstrap method and an application on agricultural data about land consolidation have been carried out to examine the efficiencies of these measures. Keywords: Asymptotic relative efficiency, coefficients of determination, land consolidation, logistic regression, misspecification.