Matched Groups

When Matching Groups Are Appropriate

The hallmark of a true experiment is random assignment of participants to groups. This means that every person taking part in the study has an equal chance of being assigned to any of its conditions. This is the opposite of putting people in groups in such a way that there could be a pattern or selection bias, either on the part of the researcher or the participants themselves. Imagine a professor wishing to conduct an experiment on whether a live or a prerecorded review session results in students performing better on an upcoming exam. The professor recruits 100 students to participate in an experiment in which she predicts that students who experience the immediacy of a live review will earn higher exam scores on average than those who are exposed to a prerecorded video of a review session. The students are given a time to report to a certain building on campus, at which point half will be placed in a room where they will receive a live review session, and the other half will be placed in a different room where they will view the review video. After this manipulation of the type of review they receive (the independent variable), all students will take the same paper-and-pencil exam, with their scores on the exam representing the dependent variable.

The ideal way to design this experiment would be to randomly assign the students who have agreed to participate to either the live or the [Page 923]recorded condition. This could be done by putting all of their names in a hat and drawing names for each condition at random, or by flipping a coin for each participant, with heads representing one condition and tails representing another. The point is that if random assignment is used, there should be no pattern in who ends up in which condition. By contrast, suppose the professor does not use random assignment. Perhaps, she asks the first 50 students to arrive at the assigned time to fill up one classroom, which represents the live review condition, and then sends the students who show up later to the other classroom to receive the prerecorded review. In this case, random assignment has not occurred, and it cannot be assumed that the individual differences across the participants are evenly spread among the two conditions. For example, maybe the people who show up a few minutes early to an assigned meeting time on campus are better and more conscientious students, while those who are a couple minutes late are less concerned with grades and schoolwork. By having all of the former in the live review session, and the latter in the recorded condition, if it turns out that the live review students do better on the exam on average than the other group, the professor will not be able to determine if their higher scores are due to the independent variable—the format of the review session—or if it is because there were better students in that group to begin because random assignment was not used. The same thing could occur if the professor allowed the students to split themselves into conditions (self-selection bias): Perhaps the better students are friends with each other and will naturally gravitate to the same condition instead of being distributed across both conditions as they would be if random assignment were used.

...