[Replying simultaneously to Peter Frank and to the edstat list.] You don't supply any indication of where your interests lie, nor why you have asked the question. So, in a correspondingly vague mode of response: Any design of the kind called "repeated measures designs" by psychologist (and some others). That is to say, any design in which each individual case (for psychologists, that would be "each subject") encounters all levels of one or more design factors. In a certain standard notation for analysis of variance designs (which I presume, in the absence of evidence to the contrary, is what you wish to refer to), a "completely randomized design" for the one-way case would be represented as R(A) where "R" = the formal factor corresponding to the individual replications (cases, or subjects, or objects, or whatever) and "A" is the (usually manipulated) design factor, sometimes called the experimental treatment factor. Similarly, R(AxB) would represent a two-way design. BxR(A) would be a "repeated-measures" design, showing R crossed with the repeated measure B, but nested within the treatment factor A. When B is a factor of two levels and represents measurements made before and after some experimental intervention, this would be a pre/post (or before/after) design. However, the logic of the design does not require that the levels of B differ in time: one could as well consider measurements made simultaneously (as of temperature at different locations on an object or person, or of skin resistance measurements made at different locations on a person) as well as in temporal sequence.
One can also have designs of the form BxCxR(A) or CxDxR(AxB), for designs with more than one repeated-measures factor. Of course, when a repeated measures factor represents time, a thorough design would ALSO have a factor "Order" (say), to control for whatever influence the first measure might have on the second (or any later) measure merely because of being first. This is not an issue when dealing with pretest-posttest designs, because in that case it is not usually possible to reverse the temporal order (and one must use other means if controlling for the temporal effect is desired), and one is usually interested first in discovering WHETHER there is a detectable effect on the response variable between pretest and posttest. If there is one, explanations can be worried about later; if there is not, such worries are moot in any case. On Wed, 23 Jul 2003, Peter Frank wrote: > Which other study designs yield paired data other than before/after > experiments? ----------------------------------------------------------------------- Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
