matched control group

Definition
The randomized matched control group design is presented below (M stands for matched):
M R T Post
M R C Post

Notes

This especially strong design is similar to the randomized posttest control group design but distinguished by the use of matching prior to random assignment of subjects. A researcher may choose this design if the sample size is too small (perhaps less than 40 per group) to reasonably ensure group comparability after random assignment. (Suter 2012, pg. 9)

Subjects are first rank ordered on a variable closely related to the posttest. Then one of the two highest (the two forming a matched pair) is randomly assigned to T or C, and the remaining participant is assigned to the other group. The next highest matched pair is similarly assigned, and this continues until the lowest two matched subjects are assigned randomly. (pg. 9)

After assignment is complete, the two case-by-case matched groups formed with this technique are nearly identical on the matched variable, and probably comparable on other variables as well.

Less is left to chance when using matching prior to random assignment. The choice of the matching variable is crucial, for nothing is gained if it is not related to the posttest. For this reason, the matching variable is often a pretest version of the posttest measure. ( A pretest is probably more highly correlated with a posttest than with any other measure. )

Example

A researcher planned to test whether a new method for teaching reading called Read Now! was more effective than one currently in use. To this end, 60 first graders ' reading skills were assessed with a pretest, then rank ordered from most advanced to least advanced. Pairs were formed by coupling the two most advanced, next most advanced, and so on, until the two least advanced were coupled. One member of each pair was randomly assigned to Read Now! while the other was retained as a control.

The two groups, now nearly identical ( on average ) in their pretreatment reading ability, were exposed to treatment or control instruction for 12 weeks, followed by a posttest measure of reading achievement. Any posttest difference could hardly be attributed to pretreatment reading differences because they were the same on average. If other extraneous influences ( e.g., teachers ' skill ) are controlled, then the researcher is entitled to conclude that the manipulation ( the true independent variable, in this case the new teaching method Read Now! ) probably caused the difference in the outcome.