A control group in an experiment does not receive the treatment. Instead, it serves as a comparison group for the treatments. Researchers compare the results of a treatment group to the control group to determine the effect size, also known as the treatment effect.
A control group is important because it is a benchmark that allows scientists to draw conclusions about the treatment’s effectiveness.
Imagine that a treatment group receives a vaccine and it has an infection rate of 10%. By itself, you don’t know if that’s an improvement. However, if you also have an unvaccinated control group with an infection rate of 20%, you know the vaccine improved the outcome by 10 percentage points.
By serving as a basis for comparison, the control group reveals the treatment’s effect.
Related post: Effect Sizes in Statistics
Using Control Groups in Experiments
Most experiments include a control group and at least one treatment group. In an ideal experiment, the subjects in all groups start with the same overall characteristics except that those in the treatment groups receive a treatment. When the groups are otherwise equivalent before treatment begins, you can attribute differences after the experiment to the treatments.
Randomized controlled trials (RCTs) assign subjects to the treatment and control groups randomly. This process helps ensure the groups are comparable when treatment begins. Consequently, treatment effects are the most likely cause for differences between groups at the end of the study. Statisticians consider RCTs to be the gold standard. To learn more about this process, read my post, Random Assignment in Experiments.
Observational studies either can’t use randomized groups or don’t use them because they’re too costly or problematic. In these studies, the characteristics of the control group might be different from the treatment groups at the start of the study, making it difficult to estimate the treatment effect accurately at the end. Case-Control studies are a specific type of observational study that uses a control group.
For these types of studies, analytical methods and design choices, such as regression analysis and matching, can help statistically mitigate confounding variables. Matching involves selecting participants with similar characteristics. For each participant in the treatment group, the researchers find a subject with comparable traits to include in the control group. To learn more about this type of study and matching, read my post, Observational Studies Explained.
Control groups are key way to increase the internal validity of an experiment. To learn more, read my post about internal and external validity.
Randomized versus non-randomized control groups are just several of the different types you can have. We’ll look at more kinds later!
Related posts: When to Use Regression Analysis
Example of a Control Group
Suppose we want to determine whether regular vitamin consumption affects the risk of dying. Our experiment has the following two experimental groups:
- Control group: Does not consume vitamin supplements
- Treatment group: Regularly consumes vitamin supplements.
In this experiment, we randomly assign subjects to the two groups. Because we use random assignment, the two groups start with similar characteristics, including healthy habits, physical attributes, medical conditions, and other factors affecting the outcome. The intentional introduction of vitamin supplements in the treatment group is the only systematic difference between the groups.
After the experiment is complete, we compare the death risk between the treatment and control groups. Because the groups started roughly equal, we can reasonably attribute differences in death risk at the end of the study to vitamin consumption. By having the control group as the basis of comparison, the effect of vitamin consumption becomes clear!
Types of Control Groups
Researchers can use different types of control groups in their experiments. Earlier, you learned about the random versus non-random kinds, but there are other variations. You can use various types depending on your research goals, constraints, and ethical issues, among other things.
Negative Control Group
The group introduces a condition that the researchers expect won’t have an effect. This group typically receives no treatment. These experiments compare the effectiveness of the experimental treatment to no treatment. For example, in a vaccine study, a negative control group does not get the vaccine.
Positive Control Group
Positive control groups typically receive a standard treatment that science has already proven effective. These groups serve as a benchmark for the performance of a conventional treatment. In this vein, experiments with positive control groups compare the effectiveness of a new treatment to a standard one.
For example, an old blood pressure medicine can be the treatment in a positive control group, while the treatment group receives the new, experimental blood pressure medicine. The researchers want to determine whether the new treatment is better than the previous treatment.
In these studies, subjects can still take the standard medication for their condition, a potentially critical ethics issue.
Placebo Control Group
Placebo control groups introduce a treatment lookalike that will not affect the outcome. Standard examples of placebos are sugar pills and saline solution injections instead of genuine medicine. The key is that the placebo looks like the actual treatment. Researchers use this approach when the recipients’ belief that they’re receiving the treatment might influence their outcomes. By using placebos, the experiment controls for these psychological benefits. The researchers want to determine whether the treatment performs better than the placebo effect.
Blinded Control Groups
If the subject’s awareness of their group assignment might affect their outcomes, the researchers can use a blinded experimental design that does not tell participants their group membership. Typically, blinded control groups will receive placebos, as described above. In a double-blinded control group, both subjects and researchers don’t know group assignments.
Waitlist Control Group
When there is a waitlist to receive a new treatment, those on the waitlist can serve as a control group until they receive treatment. This type of design avoids ethical concerns about withholding a better treatment until the study finishes. This design can be a variation of a positive control group because the subjects might be using conventional medicines while on the waitlist.
Historical Control Group
When historical data for a comparison group exists, it can serve as a control group for an experiment. The group doesn’t exist in the study, but the researchers compare the treatment group to the existing data. For example, the researchers might have infection rate data for unvaccinated individuals to compare to the infection rate among the vaccinated participants in their study. This approach allows everyone in the experiment to receive the new treatment. However, differences in place, time, and other circumstances can reduce the value of these comparisons. In other words, other factors might account for the apparent effects.
Thank you very much Jim for your quick and comprehensive feedback. Extremely helpful!! Regards, Arthur
Thank you very much Jim, very interesting article.
Can I select a control group at the end of intervention/experiment? Currently I am managing a project in rural Cambodia in five villages, however I did not select any comparison/control site at the beginning. Since I know there are other villages which have not been exposed to any type of intervention, can i select them as a control site during my end-line data collection or it will not be a legitimate control? Thank you very much, Arthur
Jim Frost says
You might be able to use that approach, but it’s not ideal. The ideal is to have control groups defined at the beginning of the study. You can use the untreated villages as a type of historical control groups that I talk about in this article. Or, if they’re awaiting to receive the intervention, it might be akin to a waitlist control group.
If you go that route, you’ll need to consider whether there was some systematic reason why these villages have not received any intervention. For example, are the villages in question more remote? And, if there is a systematic reason, would that affect your outcome variable? More generally, are they systematically different? How well do the untreated villages represent your target population?
If you had selected control villages at the beginning, you’d have been better able to ensure there weren’t any systematic differences between the villages receiving interventions and those that didn’t.
If the villages that didn’t receive any interventions are systematically different, you’ll need to incorporate that into your interpretation of the results. Are they different in ways that affect the outcomes you’re measuring? Can those differences account for the difference in outcomes between the treated and untreated villages? Hopefully, you’d be able to measure those differences between untreated/treated villages.
So, yes, you can use that approach. It’s not perfect and there will potentially be more things for you to consider and factor into your conclusions. Despite these drawbacks, it’s possible that using a pseudo control group like that is better than not doing that because at least you can make comparisons to something. Otherwise, you won’t know whether the outcomes in the intervention villages represent an improvement! Just be aware of the extra considerations!
Best of luck with your research!