Analysis of surveillance data: A rationale for statistical tests with comments on confidence intervals and statistical models

In the examination of differences between subgroups in surveillance data, whether through simple counting or through sophisticated statistical modelling, the comparison is not between simple random samples from two or more populations. The rationale for statistical tests rests on an appeal to a model of random permutation of demographic and disease factors for the observed population during the surveillance period. The testing evaluates chance as a possible explanation for the observed results. In the analysis of internal structure in a surveillance data set, statistical tests produce a conceptually simple result that lends itself to concise presentation and flexible interpretation. Tests limit emphasis on probabilistic manipulation and on parameter estimates. They cannot stand alone, and thus encourage descriptive presentation of observations. In contrast, statistical models and confidence intervals emphasize parameters rather than distributions and compete with the data for limited space.