Posttest-Only Control Group Design

Considerations for Causal Inference

The three commonly referenced conditions that must be met in order to infer causality are (1) temporal precedence of proposed cause and effect, (2) covariation of proposed cause and effect, and (3) elimination of alternative explanations for the effect. For the first condition to be met, the treatment or intervention must occur before differences between groups on the outcome variable (i.e., the effect) are observed. The posttest-only control group design is able to meet this condition because the treatment or intervention is administered before the potential group differences are observed. For the second condition to be met, it must be possible for the researcher to determine what happens both when the treatment or intervention is present and when it is absent. Again, the posttest-only control group design is able to meet this condition because it includes at least one treatment or intervention group and one control group. For the third condition to be met, potential alternative explanations for differences in group outcomes must be accounted for or ruled out. The posttest-only control group design may, but does not necessarily, meet this condition.

To minimize such alternative explanations, groups are formed using random assignment, so that any differences between groups that are observed before the administration of treatment or intervention will be due to randomness. Ideally, there will be no preexisting group differences due to random assignment and/or matching of some sort. The determining factor is how the groups are formed; if they are formed using random assignment with matching (in which the researcher creates pairs of participants, one from the treatment group and one from the control group, who have comparable important characteristics beyond group membership) or without, then the posttest-only control group design is able to meet this condition. If the groups are formed using a nonrandom mechanism (e.g., convenience, self-selection of participants into groups, criterion-based inclusion, or already-existing groups), then the posttest-only control group design cannot meet this condition. The way groups are formed for comparison is of crucial importance for inferring a causal relationship between the treatment or intervention and the observed differences between groups. For the types of research questions that lend themselves to posttest-only control group design, causal inference is commonly the desired outcome, so random assignment with or without matching is recommended.

...