Research work · 2003
Statistical methods in psychiatric epidemiology 1: a statistician's perspective
Michael E. Dewey
Practical Psychiatric Epidemiology
Abstract
In this chapter we shall look at methods of statistical analysis used in psychiatric epidemiology. We shall focus on the issues which arise in trying to make sense of a small real dataset. We assume that readers are already familiar with the concepts of confidence interval, means, correlations, and odds ratios. What makes statistical analysis in psychiatric epidemiology different? We have given more space to methods dealing with measures than would be usual in a general text on epidemiology. This is quite deliberate. What makes psychiatric epidemiology different is the emphasis on measurement. By contrast most outcomes in medical statistics were historically binary (usually dead vs. alive). This is beginning to change (note for instance the increased interest in measuring quality of life almost everywhere). Of course psychiatry as a branch of medicine has used the concept of diagnosis freely, and so naturally we also include methods for handling such binary outcomes. We start by discussing methods for predicting an outcome, whether a measurement or a binary outcome. We then discuss a group of methods used for exploring the relationship between groups of variables where there is no single outcome.