CitationSavitz, David A. (1992). Editorial: Measurements, Estimates, and Inferences in Reporting Study Results. American Journal of Epidemiology, 135(3), 223-224.
AbstractStatistical calculations constitute the basic epidemiologic study results, yielding such entities as odds ratios, rate ratios, kappa statistics, and linear regression coefficients. There is little controversy over how to perform these calculations. However, the manner in which results are to be presented and the inferences to be made from these statistics are subject to some disagreement. The adequacy of the calculated statistic as an approximation for the quantity of direct interest must be scrutinized. For example, in case-control studies, depending on the methods by which the control group was selected, the stability of exposure
prevalence, and the prevalence of disease, the accuracy of the odds ratio as an estimate of the rate ratio or risk ratio can be evaluated (1, 2). Substantive questions about causal inference from those parameters—i.e., whether a risk ratio estimate of 2.0 is a valid indicator that exposure doubles disease risk—must be answered primarily on the basis of nonstatistical criteria.