STATISTICAL SIGNIFICANCE TESTING IN THE AMERICAN-JOURNAL-OF-EPIDEMIOLOGY, 1970-1990

Despite persistent controversy over statistical significance testing and the obligation for all researchers to adopt some position on the issue, until now, practices of epidemiologists have not been examined thoroughly. Articles in the American Journal of Epidemiology around the years 1970, 1980, and 1990 were evaluated and divided into infectious disease epidemiology, cancer epidemiology, and cardiovascular disease epidemiology, with a minimum of 25 articles per topic area and time period. Presentation of significance test results and confidence intervals was evaluated in the abstracts, results text, and results tables, as was the degree of reliance on significance testing in the discussion section. Significance testing grew markedly in infectious disease epidemiology and was consistently high in cardiovascular disease and cancer epidemiology. Confidence intervals were virtually absent in 1970 and became ubiquitous (particularly in cancer epidemiology) by 1990, when the most common practice was to provide confidence intervals in results tables and to emphasize statistical significance tests in results text. Reliance on statistical significance tests in interpretation grew steadily in infectious disease, sustained a high level in cardiovascular disease, and declined after 1980 in cancer epidemiology. At present, dual presentation of confidence intervals and statistical test results is dominant.