Estimation of odds ratios in Logistic Regression models under different parameterizations and Design matrices

摘要

In this article an alternate method is proposed for the estimation of odds ratios in logistic regression based only on the predicted logit values without the necessity of knowing the type of parameterization and the design matrix used in the statistical model. The new method which is developed is illustrated with several examples based on both artificial and real data sets. These examples are also used to show by comparison the invariance property of deviance, predicted logit values and odds ratios when the parameterizations which are used imply equivalent design matrices. The disadvantages of the standard method either computational or interpretive are both discussed as well as the advantages of the alternate method which facilitates the comprehension of the results obtained in logistic regression analysis.