Comparaison des modèles probit et logit et application epidemiologique

Abstract

The work presented in this Thesis Magister articulates and highlights primarily a literature review of methods of binary data and focuses on the comparison of Logit and Probit models in the case of a dichotomous response The first part consists of an overview of generalized linear models followed by the development of logistic models and the probit model for a binary response variable depending on several variables (qualitative or quantitative), the functions that define them are recalled and the interpretation of parameters by Maximum Likelihood The second part focuses on the presentation and interpretation of the comparative study of Logit and Probit models. Finally the models mentioned above are applied to data of an epidemiological study to highlight, in a sample of 200 patients, the risk factors of coronary artery disease ; the results are compared and discussed interpreted.