Control Groups

David Dufty

Why do we need a control group?

We need it for comparison: a control group lets us see what would have happened without our experimental manipulation or intervention. An experiment without a control group is often meaningless.

For example, what if I want to test cognitive behavioral therapy (CBT) on a group of patients. I want to know if CBT alleviates depression. Let's say I collect some data from some people who are depressed.

Before my treatment: Their average Beck Depression (BDI) score: 28

They do CBT

After my treatment. 8 weeks later their BDI average: 16

Their depression seems to have gone down, on average, at least according to the Beck scale. But so what? Maybe they simply got better anyway. Maybe, for some of the people, the depressive episode ran its course.

Was it the treatment that reduced depression or time? To answer that, we need to know what normally happens to someone who is depressed. Specifically, we need to know what our group would have looked like 8 weeks later if we had not intervened. Maybe we made no difference at all.

The only way of doing this is to have a second group of people, exactly the same as the first in all respects, except they don't get the treatment. This is called the control group. Those people are in the "Control condition".

The group you give the therapy to are called the "experimental group". They are in the "experimental condition."

Now, if your experimental group of depressed patients is less depressed than the other group after 8 weeks, it's fair to say your treatment helped. But if the two groups have the same average depression, then we can conclude that actually, the treatment didn't do anything. This is true, even if your patients are less depressed. Why? Because the control group patients are also less depressed, which tells us that people seemed to just get less depressed over the eight weeks- it had nothing to do with your therapy.

So you can see that a control group is necessary to find out if a therapy, or other intervention, actually works. Just because things change, doesn't mean that you caused the change.

People's scores can change for a variety of reasons: regression artifact, spontaneous recovery, maturation, and environmental influences (for example, if the economy improves, people tend to be less depressed).

Different kinds of designs

Post-test only - this is where you only measure your dependent variable once, after you've done your experimental manipulation.

Post-test only is always between-subjects. In other words, you have two or more groups, and at least one of those groups must be a control group.

For example, Cialdini et al (1978). Investigated tactics of used-car salesmen. He call students and asked them to participate in a study, then after they agreed, he told them that the study is at 7am. The control group got told about the 7 am start up-front: he didn't wait for them to agree. The point of the study was to look at commitment to an action, and the cost (or effort) of the action.

As used car salesmen know, you're better off getting the commitment first, then increasing the cost. Getting commitment first (before mentioning the 7 am start) increased the number of people who were willing to participate : 56% to 31%.

An excellent example of the use of control groups comes from Siegal and his research into the true nature of addiction. Siegal's view of addiction was influenced by behaviorism. He believed that addiction, tolerance, withdrawal and craving could all be explained by basic, low-level associative processes.

In one of his most celebrated studies, he got a group of rats addicted to morphine and then observed their reaction to overdose under different conditions. His theory was that tolerance to drugs is an associative phenomenon, and as such, was cue dependent. Take away the cue, take away the tolerance.

Every day, the rats would be taken to a special place- a room with a distinctive color, to get their morphine. Siegel believed that the rats would be more likely to survive an overdose if they got the overdose in a place that they associated with morphine. Similarly, drugs such as alcohol and heroin affect humans more if taken in unusual locations.

This required two sets of rats: one to get the overdose in the regular dosage room, and one to get the overdose in a novel location ( a room of a different color).

There was a third set of rats - the control group. These rats got the overdose, but never recieved morphine on a daily basis. They were necessary to see what effect the overdose would have on "normal", non-addicted rats.

On the last day, all the rats recieved 15 mg of morphine, a massive dose. All but one rat in the control group died from the overdose. In the other two groups, the death rate from overdose was in keeping with Siegel's theory. In fact, the death rate doubled for rats given overdose in a novel location.

Siegel et al (1988).

group 1: morphine in new, unfamiliar room (64% mortality)

group 2: morphine in familiar room (32% mortality)

group 3: they never had morphine before (96% mortality)

This demonstrated the powerful effect of context on drug tolerance and risk of overdose.

The strength of the study comes in large part, from the solid design: two experimental groups to compare the effect of context as well as a control group.

© Copyright 2007 David Dufty