Counterbalancing Controlled Experiments (page 2)
You want to settle an argument among friends about the following question: Does loss of sleep reduce a person's physical strength? You decide to test this hypothesis: Sleeping only four hours in a night instead of eight hours will increase the time it takes a person to run 100 meters the following day. Working with people is much more complicated than working with plants or animals, but let's see how it might work as an example of counterbalancing.
Determining the Effect of Loss of Sleep on Physical Strength
You will need eight friends who agree to work with you; they are the subjects of your experiment. It would be better if all the subjects were of the same sex and close to the same age. Sex and age should be constants, as nearly as you can make them.
Your independent variable will be the differences in the amount of time your subjects have spent sleeping the night before they do the test run. Your dependent variable will be the time it takes each person to run 100 meters. You will need a stopwatch and note-taking materials.
For the first night, randomly divide your subjects into two groups. Group 1: All four subjects will sleep eight hours. They will go to bed at 10:30 P.M. and get up at 6:30 A.M. Group 2: All four subjects will sleep only four hours. They will go to bed at 2:30 A.M. and get up at 6:30 A.M.
We are assuming for this example that all have agreed to run 100 meters out at the school athletic field, and to do their best running against a stopwatch.
The next morning, as soon as you can, get all of your subjects together at the school athletic field. With your stopwatch ready, have each person do a 100-meter run and record the time for each. Their running times are the dependent variable in this experiment.
Now for the counterbalancing part of this experimental design (also called the cross-over design): On the second night, the Group 1 subjects (who got eight hours sleep the night before) sleep only four hours. The Group 2 subjects sleep eight hours. (See table 5.1.)
When all the running trials are finished, compute the average, or mean, timings of all eight runs for each of the eight-hour sleepers and for each of the four-hour sleepers. Then try to determine if there is a meaningful difference. In this experiment, counterbalancing is used to take care of differences that may already exist between the two groups. It would be practically impossible to find two groups of people who are nearly matched in running ability and other characteristics. Therefore, if you want to make a fair test using only a controlled experiment, you need a counterbalancing plan as well.
This running experiment shows how the counterbalancing principle can help to "wash out" differences among experimental subjects. There is no way, obviously, to wash out all of the differences among the eight people in this example.
It would be hard to do such a counterbalancing plan in many controlled experiments, even ones with simple subjects like popcorn plants. The problem there would be that, once you have done an experiment with one variable, for example, a fertilizer additive, you cannot switch the subjects so as to get a fair test.
When to Use Counterbalanced Experiments
A simple experiment, neither uncontrolled nor counterbalanced, may be good enough for testing a hypothesis about fairly simple physical things. But for experiments in biology and in the ways people and animals behave it is difficult to hold constants steady so that only the variables change. For these experiments we need better experimental plans—controlled experiments with counterbalancing.
Will this pet mouse eat this dried pea? A simple, uncontrolled experiment will serve: The mouse eats the pea or it does not.
Can mice live healthfully on dried peas? Now things are not so simple, and you would want a controlled experiment: a control group that is fed a good normal diet and an experimental group (as evenly matched as possible with the control group) that is fed only dried peas.
Is there more fighting among mice living in crowded conditions than in conditions that are less crowded? Here it is better to use a counterbalanced plan. You would need carefully matched groups in two cages or pens, one cage or pen large enough for the number of mice and the other cage small and cramped. Keep everything else constant—food, water, litter, ages, sex, light, and temperature. After a few days of observing and recording of any aggressive behavior in each group, switch each group to the other cage and observe again for an equal amount of time.
The experiment with people that tests the effects of loss of sleep on running speed has the problem of prejudice, especially on the part of the subjects. You, the experimenter, are aware of the prejudice problems. How about the subjects of the experiment, the ones doing the sleeping and running? Do they have prejudices about the hypothesis? Surely some of them would have. And, if so, how would their prejudices affect the results? Some of them might run harder after the short sleep just to show that they were "tough;' or because they don't believe their running speed will be affected by a little loss of sleep. Others might be prejudiced in the opposite way. This prejudice among human subjects is a difficult problem. Professional scientists often deal with such prejudices by using experimental designs called "blind" and "double-blind," which we describe in the next chapter.