Research Design Notation

Post date: Aug 26, 2012 4:39:0 PM

This notation is mostly used in the context of studies in which interventions (treatments) are performed on some groups of subjects. For simplicity, we deal only with the case of two groups.

Design 1. Post-test only randomized experiment

Each row is a group. The R indicates a randomized assignment of each subject to the group. The X indicates an intervention or treatment of some kind (such as being given a drug). The O indicates measurement of the dependent variable. So in this example, the subjects are randomly assigned to one of two groups. One of these groups gets the treatment, and the other does not. Then, after a certain amount of time, an outcome is measured for each person in all groups.

Design 2. Pre-Post randomized experiment

If the group sizes are small, randomized assignment doesn't guarantee that the resulting groups are really the same. In that case it would be better to measure the dependent variable before the treatment as well. This design lets you compare the outcomes for the treatment and control groups, and you can also compare how much change there has been in each group. If both groups change, something complicated is going that needs investigating. Finally, the pretest scores for the two groups can be compared as a check on the random assignment procedure. One problem with this design is that pretesting could cause people to respond differently. In such cases you need a more complicated 4-way design in which you some half of both the treatment and control groups receive a pretest and half do not (look up Solomon 4-group design). Of course, sometimes it simply doesn't make sense to do a pretest because it would give away what the study is about. 

Design 3. Pre-Post non-randomized quasi experiment

The N indicates a non-random assignment to a group. As you can see, a baseline measurement is made of the dependent variable in both groups, and then an intervention is applied to one group. Then the dependent variable is measured again. We refer to this as a quasi-experiment (or natural experiment) because of the lack of randomized assignment to groups. An example is when we compare the political attitudes of people who go to college with those who didn't. The pre-post design is a good one because it controls for a person's pre-existing attitudes. So if the college group comes out more liberal, even controlling for the initial attitudes, it seems like the college experience might be the cause. Unfortunately, people who choose to and are able to go to college may have been different all their lives -- long before the first measurement of attitude. 

Design 4. Post-only non-randomized non-experiment

If the difference between the two groups is defined by X, this is your basic correlational study, whether X was an intervention or not. For example, many medical studies will compare cancer rates for, say, smokers and non-smokers, and show that smokers are more likely to be get cancer. Obviously, they will try control for other differences between the smokers and non-smokers, but they likely differ in hundreds or thousands of ways, so there is no way to control for all of these things. 

Design 5. Post-test only non-experiment

Here there is just one group, and everyone receives the treatment. Then an outcome is measured. This is largely useless for research purposes, but may be useful in other contexts. For example, in education we give students a final exam to see whether they have mastered the material. We don't really care whether they gained their level of mastery from the class or not, we just want to make sure they have it. Note that it doesn't matter whether X is a deliberate intervention, such as a training class, or something that happened and is studied after the fact. For example, we might study soldiers returning from active duty (X) to see if they have psychological problems (O).