- Thom makes a good point about pairing, and there is one thing more to emphasize, which is not completely obvious.
On Tue, 29 Jul 2003 13:35:42 +0100, Thom Baguley <[EMAIL PROTECTED]> wrote: > Paul R Swank wrote: > > > > there are three basic situations for paired data: 1) repeated measures, that > > is, repeated obseevations on the same subjects over time; the pretest > > posttes design is one example of this and longitudinal studies; 2) matched > > data as in Karen's Case-Control designs in health sciences data; 3) paired > > data such as sibling pairs, husband-wife pairs, etc, where the subjects are > > naturally paired in that they may be more similar to each other than to > > other people, but not matched by the researcher. Of course, you can have > > combinations of these situations. > > A good summary, but slightly misleading when you say "more similar", it is > just as appropriate to used paired tests when the expectation is that they are > less similar than chance (i.e., any situation where the two halves of the pair > are not considered independent). An example, might be two animals from the > same litter - we might find that some variables such as how much milk they > drink are negatively correlated (i.e., when one pup drinks more this may > reduce the amount available to the other pup). ... If the correlation is negative, the paired test is proper and necessary, even though the paired test is apt to have a larger error term than the t-test that ignores pairing. That is, the paired test will have less power; the unpaired test will have more power, but it will be *too much* power, and that means that it would be an improper choice for a test. It is improper for our research logic, to have "too much" power, unless we are in an unusual circumstance, such as, invalidation. Even then, it would be wiser to use the proper, paired test and openly shift the p-level, if that is the end to be served. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
