Logistic regression and some of its peculiarities

Abstract

One major aim of this dissertation was to investigate the existence or lack of relationship between the use of health care services and the intake of certain medicines with a number of other variates in particular body mass index. To achieve this, one has to start by studying in some depth logistic regression since our dependent variables were discrete. We also wanted to explore how each explanatory variable affects the probability of an event occurring with the idea of consulting recent research work related to problems arising in empirical studies within this area. Logistic regression has statistical limitations and a number of problems related to the method of estimation. In particular we mention lack of robustness, biasedness and problems which arise within the rare events context. Related to such concerns, a number of issues about sampling design arise and were also investigated. Furthermore, various techniques to improve parameter estimates, intended for a case-control design were studied. Finally these techniques were applied to analyse the relationship between body mass index, gender and age with the use of health care services, in particular consultation with doctors and the use of particular medicines, using data obtained from the department of health information. There is also an attempt at running through a number of experiments which use the case-control methodology.