Risk ratio and odds ratio are two measures that are often encountered in medicine when we are evaluating the impact of an exposure factor on binary outcome. This thesis contains definitions, properties and methods for estimating both measures. If control over the influence of confounding factors is not possible in the sampling process itself, which almost certainly happens in observational research, their effect must be taken into account in the statistical analysis. This is usually done using regression models, and consequently, in the core of the work, I concentrate on the basic theory of generalized linear models, which is necessary for estimating both measures. When estimating odds ratio, the choice of a logistic model is self-evident. Unfortunately,
there is no model for estimating risk ratio that is as mathematically perfect as logistic model is for estimating odds ratio. However, since risk ratio is usually the parameter of primary interest, a lot of effort has been invested in finding the most appropriate approach to estimating it. The two most common approaches to estimating risk ratio are log-binomial regression and Poisson regression. In the second part of this thesis the former model is briefly mentioned, while the latter is described in greater detail.
|
