A Potential Problem with Random Assignment
Random assignment makes it likely that subjects in different conditions are highly similar to each other. In some cases, though, “likely” may not be good enough.
For example, imagine that a researcher is testing the effectiveness of a new drug to treat schizophrenia. Imagine also that she has good reason to think that the age of the patient might be related to the effectiveness of the drug (e.g., it might work better on younger patients). She could randomly assign her subjects to treatment and control conditions, but if she does, she risks ending up with two groups of very different-aged people (especially if her sample is small). This could be very bad. Imagine that the treatment group ended up consisting of older subjects who than the control group. In other words, age has become a confounding variable because it differs systematically across conditions. If the researcher finds no difference between her treatment and control groups, is this because the drug does not work or is it because the people who got the drug were older and the drug does not work well on older people?
A matched-groups design is used when there is an extraneous variable that the experimenter thinks might be related to the dependent variable … and the experimenter wants to be sure it does not become a confounding variable. Also, the experimenter must be able to measure this extraneous variable before conducting the experiment. Here is how it works.
The experimenter measures the extraneous variable of interest, which is called the matching variable (e.g., age). The experimenter then creates groups of subjects who have the same value or level of that matching variable. For example, in an experiment with two conditions, the experimenter might create a pair of 60-year-olds, a pair of 57-year-olds, a pair of 52-year-olds, a pair of 37-year-olds, and so on. Then the experimenter would randomly assign one member of each pair to one condition and the other member to the other condition (e.g., by flipping a coin). Note that this results in two groups that are identical in terms of their ages.
Sometimes there are very few subjects with identical values on the matching variable. Imagine, for example, that you have subjects with the following ages: 60, 59, 57, 57, 52, 48, 47, 45, …. Here you could take the two oldest and randomly assign them to the two conditions, take the next to oldest and do the same, etc. The two groups will not be identical in terms of their ages, but they will be pretty close.
Randomized Design vs. Matched-Groups Design
Is the matched-groups design “better” than a randomized groups design? Not necessarily. The problem is that matching is a lot more trouble than random assignment. You have to measure the matching variable first (which may not be easy to do), then create your matched pairs and assign your subjects to conditions, and only then conduct the experiment. So it would be perfectly reasonable in many cases to decide to use a randomized groups design rather than a matched-groups design. On the other hand, if you are conducting a study that costs lots of money, takes lots of time, or requires special subjects that are hard to find (e.g., schizophrenic people), and if there is an obvious extraneous variable that is related to your dependent variable, then it is probably worth the trouble to use the matched-groups design.