On 1 Dec 2002 06:44:13 -0800, [EMAIL PROTECTED] (Brad Branford) wrote: > Hi, > > Actually, the reason for the test is to determine which is the better > segment to target. A colleague of mine contended that with a mailing > of 25000 and a response rate of approximately 0.5%, one can expect > about 125 responses which is not sufficient to determine how the test > fared against the control.
Is there any special reason for the control? - if there is, then you might want that group to be larger. And you could use limits defined by Dunnett's test, which specifies that the N for control versus N-for-each experimental has the ratio of sqrt(m) where m is the number of experiments. If the groups serve as mutual control, then you can forget a special control; divide the total by 5 instead of 6. If you get 125 responses, that will seem to be "more" than getting 100 responses, by the simple two-group comparison. The Ns you need will depend on -- How drastic do you hope or expect the differences to be? -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
