What is Experimental Design?
Experimental design refers to the way we assign experimental units to different groups in an experiment.
The three most common designs are:
- Completely randomized design
- Randomized block design
- Matched pairs design
Suppose researchers recruit 400 individuals for an experiment to find out if a new drug causes weight loss. Here is how we would perform this experiment using the three different experimental designs.
Completely randomized design
In a completely randomized design, researchers randomly assign subjects to different groups. This is the simplest experimental design.
In this case, researchers may randomly assign 200 subjects to the treatment group and 200 subjects to the control group. The subjects in the treatment group take the new drug once per day for 30 days. The subjects in the control group take a placebo (a harmless sugar pill that looks like the real drug but actually does nothing) once per day for 30 days.
The benefit of randomly assigning subjects to groups is that it controls for the effects of confounding variables, which are variables that we don’t account for in the experiment.
For example, we don’t account for exercise or diet, which are both variables that may affect weight loss. But since we randomly assigned patients to groups, we can assume that these confounding variables will affect both groups roughly equally. This means any differences in weight loss can be attributed to the new drug.
Randomized block design
In a randomized block design, researchers first split up subjects into “blocks”, and then randomly assign them to a treatment.
In this case, researchers may split up the 400 people based on gender – 200 subjects in a male group and 200 in a female group. Then, researchers randomly assign subjects to a the new drug group or the placebo group.
The benefit of using this design is that we explicitly rule out gender as a possible cause of weight loss. In a completely randomized design, we assume that the genders are roughly spread out equally among the new drug and the placebo group. But in a randomized block design, we make sure that the genders are spread out equally among the new drug and placebo group.
This design is considered to be better than a completely randomized design since it explicit controls for a confounding variable.
Matched pairs design
A matched pairs design is a specific type of randomized block design where one subject is paired with another subject based on one or more blocking variables.
In this case, researchers may pair subjects based on gender and age. For example, one 35-year-old male will be paired with another 35-year-old male.
Since we have 400 total subjects, we would have 200 total pairs:
The benefit of this design is that we can control for two confounding variables – weight and age.
Note: It’s possible that we could pair subjects together based on even more than two variables. For example, we could pair subjects based on weight, age, and blood pressure. It can be difficult to pair subjects based on more than two variables, though, because it can be hard to find enough subjects to participate in a study who “match” on more than two variables.