Matched Individuals

Matched Pairs Design

Experimental design typically centers on observing the relationship between independent and dependent variables within a study. In an experiment, researchers will manipulate the treatment group to compare changes in the control group based on a set of dependent variables. Depending on the type of sampling used, the treatment/control experimental design can result in confounding within the experiment. Confounding is the result of having controls in an experiment that do not adequately rule out alternative explanations to predictions in the study. Confounding can result in spurious relationships between measured variables, and can call into question the validity of a study’s findings. Confounding may not always be obvious in a study’s design, and so researchers should carefully consider controls to be measured and the structure of the sample being used.

For example, consider the following design for a study on experiences of communication apprehension during presentations: A researcher recruits a sample of 200 individuals from an organization (100 males and 100 females) and observes them delivering a presentation to their co-workers in either an online or in-person context. The researcher assigns all 100 men to deliver an in-person presentation and all 100 women to deliver an online presentation. It is determined that there is statistically less apprehension in the in-person context when compared to the online context. The issue with this conclusion is that there may be other confounding attributes, such as gender differences, employee status, or presentation topic, that might also explain the observed differences in apprehension between presentation types.

Confounding controls can be managed during the analysis phase of an experiment post hoc, but a more useful approach is to eliminate confounding during the design phase of an experiment by matching individuals. For the design phase, a matched pairs sample can address many of the issues of confounding by matching individuals on all other attributes with the exception of the treatment/control variable. For example, in the study described above, a matched pairs sample would have equal numbers of males and females in the online and face-to-face presentation contexts to control for potential confounds based on participant gender. Individual cases can be matched on any number of attributes measured within a study, including attributes such as gender, age, marital status, or education. As one matched individual receives the treatment and the other receives the [Page 926]control, a matched pair sample can help eliminate potential confounding variables in the study.

...