[Skip to Navigation]
March 1991

P Values

Author Affiliations

Clinical Epidemiology and Biostatistics Unit Royal Children's Hospital Flemington Rd Parkville, Victoria 3052, Australia

Am J Dis Child. 1991;145(3):250-251. doi:10.1001/archpedi.1991.02160030014006

Sir.—In his article on P values1 published in the April 1990 issue of AJDC, Brown has made some very helpful comments on the meaning of P values and has drawn attention to the difficulties associated with performing repeated statistical comparisons on the same set of subjects. However, twice in his article he does not pay sufficient heed to his own advice.

First, he correctly points out that when a t test is used to compare the birth weights of two groups of infants and the significant P value is set at .05, the random variation in birth weights among the infants would suggest, wrongly, in about one comparison of 20, that an important difference between the two groups in average weight was present. He goes on to consider what happens when a second test, eg, on head circumference, is performed on the same group, and says "The chance of

Add or change institution