Last update:
08-Mar-2007

The simplest type of epidemiological analysis, which is based on crude data, applies when it is not necessary to take into account any factors beyond the exposure (or determinant) and the outcome of interest (e.g. disease onset). Although it is not unusual to see data presented solely in crude form, typically the researchers need first to explore more complicated analyses, using stratification or multivariate regression to evaluate the role of other factors. This review describes the statistical models used to analyze crude epidemiological data. Researchers pay more attention to statistical estimation than to statistical significance testing. This review concentrates on formulas for obtaining confidence intervals for measures of effect, although formulas for deriving p values are also included. The formulas presented here give only approximate results and are valid only for data sets with sufficiently large numbers. More accurate estimates can be obtained using what are called exact methods, but in this case calculations are more difficult than in approximate methods. It is difficult to determine a precise threshold of data above which it can be said that the approximate results are good enough and below which exact calculations are needed. Fortunately, even for studies with modest numbers, the interpretation of results rarely changes when exact rather than approximate results are used to estimate confidence intervals.

Key words: Confidence interval, Measures of effect, Source population, Statistical model, Study population.