Control Group

In experimental designs, a control group is the “untreated” group with which an experimental group (or treatment group) is contrasted. It consists of units of study that did not receive the treatment whose effect is under investigation. For many quasi-experimental studies, treatments are not administered to participants, as in true experimental studies. Rather, treatments are broadly construed to be the presence of certain characteristics of participants, such as female gender, adolescence, and low socioeconomic status (SES), or features of their settings, such as private schools or participation in a program of interest. Thus, the control group in quasi-experimental studies is denned to be those lacking these characteristics (e.g. males, respondents who are older or younger than adolescence, those of high and medium SES) or absent from selected settings (e.g. those in public schools, nonparticipants in a program of interest). Control groups may alternatively be called “baseline groups.”

In a true experiment, control groups are formed through random assignment of respondents, as in between-subject designs, or from the respondents themselves, as in within-subject designs. Random assignment supports the assumption that the control group and the experimental group are similar enough (i.e. equivalent) in relevant ways so as to be genuinely comparable. In true experimental studies and between-subject designs, respondents are first randomly selected from the sampling frame; then they are randomly assigned into either a control group or an experimental group or groups. At the conclusion of the study, outcome measures (such as responses on one or more dependent variables, or distributions on survey items) are compared between those in the control group and those in the experimental group(s). The effect of a treatment (e.g. a different incentive level administered to each group) is assessed on the basis of the difference (or differences) observed between the control group and one or more experimental group.

Similarly, in within-subject designs, respondents are randomly selected from the sampling frame. However, in such cases, they are not randomly assigned into control versus experimental groups. Instead, baseline data are gathered from the respondents themselves. These data are treated as “control data” to be compared with outcome measures that are hypothesized to be the result of a treatment after the respondents are exposed to the experimental treatment. Thus, the respondents act as their own control group in within-subject designs.

Control groups are often used in evaluation studies that use surveys, and they are also relevant to methodological research on surveys. Research that examines the effects of questionnaire design, item wording, or of other aspects of data collection often uses a classical “split-ballot” design or some variant. In these studies, respondents are assigned at random to receive one of two versions of a questionnaire, each version varying on a single point of question order, wording, or presentation. In practice, these studies often depart from the conception of presence versus absence that typically marks the contrast between treatment and control [Page 147]groups. Researchers may present a variation of an item to both groups, for example, as opposed to administering the item to one group and not to the other. Nevertheless, these lines of survey research rely on the control group—either literally or by extension—as a necessary support for claims about the causal effects of the items, procedures, or programs being studied.

...