Let's say, for the sake of argument, that you do not trust randomization to give you an experimental and control group that are initially equivalent (at least on average) in post-prandial blood sugar measured in the office. (I'm referring to the hypothetical situation I set up in the Posttest-Only Control Group Design discussion.)

This would actually put you in good company, which is one reason why pretest-posttest approaches have been so popular.

Anyway, you decide that you want to obtain initial measurements before you start your experiment. After you complete your experimental treatment, you again check everyone's blood sugar level. Then you subtract your post-experimental measurements from each person's pre-experimental measurement to see how much better the experimental group is doing.

Even though you are looking for a drop, or loss, in the experimental group measurements, this approach has traditionally been called a gain score. As I mentioned on the previous page, many experimenters use a special type of t-test, called the "repeated measures" t-test. This test also has a number of other names in the literature. Some call it the "dependent measures" test, some the "correlated measures" test, some the "matched measures" test.

Here is a chart of hypothetical pre and post measurements.

Return to the discussion of the Pretest-Posttest Control Group Design.
Return to the discussion of experimental designs.