Hertz-Picciotto, Irva (1999). What You Should Have Learned about Epidemiologic Data Analysis. Epidemiology, 10(6)
The formal teaching of methods of the analysis of epidemiologic data is not cohesive. In general, this type of instruction occurs under two rather separate rubrics. The first is conceptual, often involving example-based presentations that rely heavily on standard two-by-two tables. The second is statistical, commonly focusing on techniques for fitting linear and logistic models selecting variables, and applying tests of significance. The connection between the two is frequently not transparent and many epidemiologists are not prepared by their training to integrate the fragments when they set out to analyize data. What they learn from epidemilogy courses is somethings overly heuristic, and what the glean from statistics courses cna be overly algorithmic and lacking in context. This splintered state of the teaching of methods for analysis of epidemiologic data can be remedied. I propose a pedagogy for epidemiologic methods that emphasizes the integrative aspects of this philosophical approach. The major elements of this pedagogy are: (1) the interplay of context with methods, (2) experiential leanring, (3) discovery as essentail to data analysis, (4) judgement in scientific training, (5) tools rather than rules, and (6) a subordinate role for technology.