Within-Subjects Designs
Between-Subjects Designs
Randomized designs and matched-groups designs are exampes of between-subjects designs. This means that every subject is tested under one, and only one, condition. For example, in a randomized experiment with a treatment condition and a control condition, each subject is tested either under the treatment condition or under the control condition.
Within-Subjects Designs
Sometimes, however, it is desirable to use an experimental design in which each subject is tested under all conditions. This is called a within-subjects design or sometimes a repeated-measures design. For example, the very same subjects might be tested under a quiet condition and a noisy condition to study the effect of noise level on concentration.
Advantages of Within-Subjects Designs
1. Control of Extraneous Variables. Remember that random assignment and matching are intended to create groups that are highly similar to each other. Within-subjects designs go a step further, creating groups that are identical to each other in most ways. The IQs of the subjects in one condition are identical to those of the subjects in the other conditions because they are the same subjects. The same holds true for most other person variables like race, sex, age, and so on. These designs do not control all extraneous variables to the same degree, however. Subjects’ moods, for example, can still differ from one condition to the next. Also, situation variables or task variables (e.g., time of day, temperature in the room) are still free to differ across levels of the independent variable.
2. Efficiency in Terms of Subjects and Time. Within-subjects designs are more efficient in their use of subjects and time. For example, a between-subjects design with three conditions and 20 subjects per condition requires 60 subjects. The same study conducted as a within-subjects design requires only 20 subjects. In addition, the within-subjects version can probably be completed in less time than the between-subjects version.
3. Statistical Efficiency. Within-subjects designs make it easier to detect differences across levels of the independent variable because each subject’s behavior under one condition is compared to that subject’s behavior under the other condition. The best way to see this is with an example.
Imagine that a new treatment increases people’s IQs by one point. The data in the table below are from a between-subjects experimental design (with different subjects in the two conditions). These data show the one-point mean difference, but it is not clear that this difference is reliable—that it is due to the treatment or just to chance.
Between-Subjects Design
|
|||
|
Control |
|
Treatment |
Subject 1 |
100 |
Subject 2 |
109 |
Subject 4 |
98 |
Subject 3 |
113 |
Subject 5 |
108 |
Subject 6 |
101 |
Subject 8 |
104 |
Subject 7 |
105 |
Subject 9 |
89 |
Subject 10 |
99 |
Subject 12 |
99 |
Subject 11 |
107 |
Subject 13 |
106 |
Subject 14 |
100 |
Subject 16 |
112 |
Subject 15 |
90 |
|
M = 102 |
|
M = 103 |
The data in the next table are from a within-subjects experimental design. First, notice that this design uses only 8 subjects whereas the between-subjects design used 16. This is because the 8 subjects are tested under both conditions. Again, there is a one-point mean difference, but here we can see that the difference is shown by each and every subject. It does not seem to be due to chance. This is an exaggerated result, but it should give you some sense of why small effects are easier to “see” when using a within-subjects design.
|
Within-Subjects Design |
||
|
Control |
Treatment
|
Difference |
Subject 1 |
100 |
101 |
1 |
Subject 2 |
98 |
99 |
1 |
Subject 3 |
108 |
109 |
1 |
Subject 4 |
104 |
105 |
1 |
Subject 5 |
89 |
90 |
1 |
Subject 6 |
99 |
100 |
1 |
Subject 7 |
106 |
107 |
1 |
Subject 8 |
112 |
113 |
1 |
|
M = 102 |
M = 103 |
M = 1 |
Are There Disadvantages of Within-Subjects Designs?
Yes. The major disadvantage of within-subjects designs is that they can produce carryover effects. A carryover effect is when having been tested under one condition affects how subjects behave in other conditions. We will consider carryover effects and what to do about them—and say more about the choice of between-subjects versus within-subjects designs—in another section.