[Skip to Navigation]
July 1990

Statistical Interpretation of Multiple Comparisons and Sample Size

Author Affiliations

Division of Pulmonary Medicine Children's Hospital of Philadelphia 34th and Civic Center Boulevard Philadelphia, PA 19104

Am J Dis Child. 1990;144(7):751-752. doi:10.1001/archpedi.1990.02150310017014

Sir.—Caution must be exercised in interpreting the results of the study by Miller and Strunk1 regarding the deaths of children that were due to asthma. Even if the control group was perfectly matched with the study group, statistical analysis of the results should take at least two factors into account: (1) The greater the number of comparisons undertaken, the more likely it is that a "significant" low P value will be found on the basis of chance alone.2 In the study by Miller and Strunk, at least 30 comparisons were made between the control and study groups. Six variables were noted to be different between the two groups at P<.05, but because of the problem of multiple comparisons this does not mean that the groups actually differed in these six variables. Methods to correct for multiple comparisons have been developed.2 For example, the conservative Bonferroni correction

Add or change institution