PAIRED EXPERIMENTAL DESIGNS


When comparing two treatments in an experiment, is it always better to form pairs of similar subjects (or to observe the same subject under both conditions) than to use two independent groups? No. Paired experimental designs are only useful when you can form pairs on the basis of a variable that is related to the one you are studying. If you pair your subjects based on shoe size when you are studying responses to a new drug, the paired design actually makes it less likely that you can identify a true difference when it exists. The two members of a pair are not alike in any way that matters. In this case, using a paired t test makes it more difficult, statistically, to detect true differences than just treating the two groups as independent samples.

A paired design is a good way to eliminate some of the differences between subjects in two groups so you can focus on the particular difference that you are testing. But you need to keep some things in mind:

  • If the effect of a treatment does not wear off quickly, you must make sure that enough time passes between treatments so that one wears off before another begins. Otherwise, you will not know whether the first or the second treatment is causing the results during the second observation.

  • You should also be aware of the learning effect. You encounter it when a subject's response improves merely by doing the same thing again. For example, if you give subjects the same test twice, they may do better the second time, regardless of what else has happened .

You must pay attention to both the timing and the sequence of administering the treatments. You may also want to include a control group that receives no treatment but undergoes the same measurements that the treatment groups undergo.




Six Sigma and Beyond. Statistics and Probability
Six Sigma and Beyond: Statistics and Probability, Volume III
ISBN: 1574443127
EAN: 2147483647
Year: 2003
Pages: 252

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net