This article is only available in the PDF format. Download the PDF to view the article, as well as its associated figures and tables.
To the Editor.
—An EDITORIAL by Schor in the Archives (1981;141: 1263-1264) and an article by Reed and Slaichert in the same issue (1307-1310) dealing with "inconclusive `negative' trials" contain, like many of the studies they criticize, errors in statements of statistical inference.The EDITORIAL asserts that if an observed difference in the mean effectiveness of two drugs, say, is found to be statistically significant (P<.05), then with probability.05 the researcher would be in error in his conclusion that one drug is better than the other. This is not correct. Standard "significance testing" permits no assessment of the "chances" that one's conclusion is wrong. All that can properly be asserted is that if the drugs are really no different, the test used had a 5% chance of mistakenly telling us that they are different. Once the test is carried out and the conclusion of statistical significance is reached, we
Lindgren BW. Statistical Proof. Arch Intern Med. 1983;143(9):1836-1837. doi:10.1001/archinte.1983.00350090214050