How we change what others think, feel, believe and do
Two parallel experiments are set up, identical in all respects except that only one includes the treatment being explored by the experiment.
The people in both groups should be similar. Ideally these are selected and assigned randomly, though in practice some groups come as one (such as school classes) or are selected on a pseudo-random basis (such as people on the street).
The control group may have no treatment, with nothing happening to them, or they may have a neutral treatment, such as when a placebo is used in a medical pharmaceutical experiment.
A common form of experiment is to apply the treatment and measure the results, for example a training course is followed by testing their knowledge as compared to a control group who are not given the training.
In design notation, with randomized selection and assignment to treatment and control groups this is:
Pre-test and Post-test
A problem with the post-test only is that there is no direct indication of what actual change was found in the treatment group. This is corrected by measuring them before and after the treatment. The control group is still useful as additional factors may have had an effect, particularly if the treatment occurs over a long time or in a unique context.
In much research, random assignment is not easy or not possible and other means are used in a quasi-experiment. Note that the control group 'saves' the research from being a less credible non-experiment.
In design notation, non-equivalent selection with pre-and post-tests is:
A researcher wants to discover the effects of a new teaching method. A class is selected and randomly divided into two groups, one of which is taught by the conventional method and one by the new method. Both groups are then measured with a standard ability test.
It is usually important to ensure control groups are as similar as possible to treatment groups. Random assignment to both groups of a pre-selected random set of subjects is the best method, but is not always possible. When assignment is in a group, designs such as switching replications can be used to check the similarity of the two groups.
Whilst control groups are particularly useful in many situations, much research does not need or use control groups, for example when researching public attitudes to government policy.
And the big