[The use of statistical methods in 2 medical journals with high impact factor]

This paper intends to contribute to the discussion about the use of statistical techniques within the current biomedical and health investigation. We hypothesized that statistical methods necessary to go ahead with most high level research were rather small. Two high impact factor journals about epidemiology and health sciences were selected: New England Journal of Medicine and American Journal of Epidemiology. Both represent what we have called "successful research". The 2386 original and special articles which were published along the 1986-1990 period were carefully examined. Each statistical technique was classified into one of four complexity levels. Each paper was eventually classified on the same grounds. The frequency of each method was used within each level was then accounted. Statistical cited literature was also registered for each article. Finally, a descriptive examination of this information was developed. The results show that techniques widely dominant are those classified as from lower levels. Only the 12% for New England Journal of Medicine, and 17% for American Journal of Epidemiology belong to the most complex type. It was shown that current biomedical research dismisses, to a large extent both, complex statistical techniques and specialized statistical literature. In general, it limits itself to resources no more ahead than elementary multivariate procedures. It was concluded that simpler statistical methods are useful enough not only to understand what is currently published in journals like the ones studied but to produce high level results as well.