I would do contrasts. The first two would compare the three controls (then combine them) and the last would compare the experimental to the average of the three controls. A liberal would then test the latter at the .05 level (assuming an experimentwise error rate of .05) since it is a planned comparison. The conservative would test it at .025 to control for the type one error rate. This process would have more power than Dunnett's test.
Paul R. Swank, Ph.D. Professor, Developmental Pediatrics Medical School UT Health Science Center at Houston -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of John Mercer Sent: Tuesday, June 24, 2003 9:46 PM To: [EMAIL PROTECTED] Subject: Re: t-Test vs Dunnett's Test In article <[EMAIL PROTECTED]>, [EMAIL PROTECTED] (Karl L. Wuensch) wrote: > Dunnett's test represents an attempt to control familywise error in the > situation where each of several means is contrasted with a single reference > mean. The test statistic is computed the same as the usual t (possibly with > pooled error), but the function relating t to p is different, and dependant > on the number of groups. So would this be the proper choice if I have a single experimental group and three control groups? I am in this situation, with very large standard deviations caused by uncontrollable factors. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . ================================================================= . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
