Comparing the Means of Several Groups

• Published in: The New England journal of medicine Vol. 313; no. 23; pp. 1450 - 1456
• Main Author: Godfrey, Katherine
• Format: Journal Article
• Language: English
• Published: Boston Massachusetts Medical Society 05.12.1985
• Subjects: Statistical analysis
• ISSN: 0028-4793; 1533-4406
• DOI: 10.1056/NEJM198512053132305

This article discusses statistical methods for comparing the means of several groups and focuses on examples from 50 Original Articles published in the Journal in 1978 and 1979. Although medical authors often present comparisons of the means of several groups, the most common method of analysis, multiple t-tests, is usually a poor choice. Which method of analysis is appropriate depends on what questions the investigators wish to ask. If the investigators want to identify which of the groups under study are different from the rest, they will need a different method from the one required if they wish simply to decide whether or not the groups share a common mean. More complicated questions about the group means call for more sophisticated techniques. Of the 50 Journal articles examined, 27 (54 per cent) used inappropriate statistical methods to analyze the differences between group means. Investigators need to become better acquainted with statistical techniques for making multiple comparisons between group means. (N Engl J Med 1985; 313:1450–6.) IN one simple form of experiment or observational study, the investigator compares sets of measurements taken from two groups to decide whether the group means differ. Emerson and Colditz 1 have reported that the t-test, the standard analysis for such an experiment, is the most commonly used statistical procedure in the Journal . When an experiment includes more than two groups, the choice of an appropriate statistical method for comparing group means depends on the experimental design and on the questions asked. Although investigators have many options, few are appropriate, and the most frequently used method, multiple t-tests, is usually a . . .