Sick or Healthy, Dead or Alive: A Century of Progress in Analyzing Dichotomous Disease Data

The APHA Statistics Section was founded only shortly after Pearson proposed his chi-square test for independence in two-way contingency tables in 1904. Since then a core task of public health statistics, to understand the relationship over time between the presence of an exposure and the development of disease, has been immeasurably aided by increasingly sophisticated tools for analyzing dichotomous responses. This brief historical survey will skip along a timeline pointing out linkages between these tools and problems in interpreting public health data to which they were addressed or which soon benefited from their application. Developments touched upon, by necessity very briefly, will include rate adjustment; the odds ratio in case-control studies; Mantel-Haenszel methods; logistic regression; marginal modeling by weighted least-squares and generalized estimating equations; Cox proportional hazards models (which view survival data as ordered clusters of dichotomies); publication of Fleiss' Statistical Methods for Rates and Proportions; spatial models for disease; and generalized linear mixed models. The thread will be growing power to explore and exploit interconnections among variables to ask increasingly refined questions of ever larger and more complex data structures.