To keep the illustration simple, imagine that we developed an educational program consisting of two-hour sessions held once a week for three weeks. During these sessions subjects would be provided positive information about the group in question, such as descriptions of the accomplishments of members of the group, how members of the group are like one's own group, or ways members of that group and one's own group have worked together and helped one another.   This information might be provided in the form of videos, talks by members of the designated group, and by participation in small group discussions.

We also have to decide how to measure prejudice.   We could use a Bogardus social distance or some other scale for measuring prejudice. Whatever measure we use, we have to use it twice - once before the educational program is started and again after it is over.    The first is called the pretest measurement and the second is referred to as the posttest measurement.    By comparing the two levels of prejudice we could see if there was a change in the degree of prejudice among the subjects.

Imagine that we actually carried out this study as described, and, further, that the results supported our hypothesis. We found that the posttest measurement for prejudice was lower than the pretest measurement.   Are we justified in concluding that the educational program caused the decline in prejudice?   No, we are not!   Our inability to draw that conclusion goes to the heart of experimental design.

The design we just described is referred to as a single group design. In this design, you collect data from a group of subjects on one or more selected variables before beginning of the experiment, apply some kind of experimental treatment, and collect data again on the variable or variables measured before. Single group designs do not rule out conditions other than the experimental treatment that could affect the outcome.     With a single group design, one cannot meet the three requirements for establishing a cause and effect relationship.   In the experiment we just described, we safely met the first requirement.   We know that the experimental variable preceded posttest measurement of the dependent variable.   Also, we know that there was some association between the two variables, as indicated by the decline measurement of prejudice following the educational program.   But can we say that nothing else could account for the decline other than the educational program?   To answer this question, let's look at some alternative explanations for why a decline in prejudice could have occurred.

Alternative explanations

The following discussion is based mainly upon the ideas of Donald Campbell and his associates, as reported in their careful analyses of experimental designs (see Campbell, 1985; Campbell and Stanley, 1966; and Cook and Campbell, 1979).   Their analyses show that there are at least seven other reasons why prejudice could have declined among the subjects in our example.    These alternative explanations, which have to be eliminated before one can say that the data support a hypothesis, are known as:

• History  
• Maturation
• Testing effect
• Instrumentation
• Experimental mortality  
• Subject reactivity
• Experimenter bias  

Each of these factors or any combination of them could have affected our results.

History

History refers to any event outside the experiment that could affect the results.   Suppose, for example, that during the three-week period of the experiment a rich person from the group in question gave an exceptionally large and widely announced donation to a local hospital.   Could knowledge of this act have made subjects feel less prejudiced and therefore have been responsible for the decline, rather than the educational program they were in?   Possibly: Our experimental design is flawed in this respect and would prevent us from drawing the conclusion that the hypothesis was supported.