March 1991

More on the P Value

Author Affiliations

The University of Utah Department of Pediatrics Eighth Avenue and C Street Salt Lake City, UT 84143

Am J Dis Child. 1991;145(3):249. doi:10.1001/archpedi.1991.02160030013003

Sir.—Brown's discussion1 on the P value continues his elegant series of lucid dissertations on medical statistics. A brief clarification is needed, however, concerning the issue of multiple statistical tests and the risk of type I error. Such corrections to compensate for multiple tests on the same data set are important during "data dredging" expeditions, explorations of a data set in search of statistical differences that were not anticipated by the original study design or hypothesis. As Brown points out, there is one chance in 20 that any such comparison will be significant at the P=.05 level. The corrected P values also protect against the type I error, which can occur when "in examining the treatment means we notice a combination that we did not intend to test but which seems unexpectedly large."2

In contrast to the data dredger is the investigator who designs a prospective study

First Page Preview View Large
First page PDF preview
First page PDF preview