Introduction

Seeking
causal
relationships

Alternative
explanations

Internal validity

The classical experiment

Quasi-
experimental designs

Quasi-
experimental designs

External validity

Further
variations
in experimental design

Strengths & limitations of experimental research

Aids

Internal validity

This term refers to the degree to which the conclusions of an investigation are in fact based solely on the effect of the experimental variable.    For an experiment to be internally valid we have to be able to eliminate all alternative explanations, leaving the effect of the experimental variable as the only plausible explanation for the observed changes in the dependent variable.

How can we achieve internal validity for an experiment?   Simple: our best protection against threats to internal validity is to use the model of the classical or true experiment.

We have barely introduced the concept of internal validity. For additional discussion of this important feature of any experiment, go to Internal Validity. When you view this site, you will see how difficult it is to separate true, causal effects of an experiment from alternative causes that could affect the outcome and, hence, the validity of an experiment. (You may want to review the first section of this chapter to review the requirements for advancing a causal relationship). Earlier, we mentioned the threats to internal validity of the single group design. These are described briefly in Single Group Threats.

The classical experiment

The classical experiment is based on simple but powerful logic.

Logic of the classical experiment

In designing an experiment we want to eliminate the influence of any variable other than the experimental one on the dependent variable.    This is done by using the model of the classical experiment in which we:

1. Create two equivalent groups at the beginning of the experiment;
2. Administer the experimental variable to one group, the experimental group, but not to the other which is retained as the control group; and
3. Measure the dependent variable in both groups before and after the application of the experimental variable. These are the pretest and the posttest measurements.

The logic just described is shown graphically in Figure 9.1. Using this logic, let's redesign the study of prejudice we presented earlier.    Previously we had used only an experimental group.   Following the logic shown in Figure 9.1, we will use an experimental and control group.    (In the next section we will show how to create these groups).   Next, we would measure and compare prejudice levels in both groups before the experimental variable is administered.    If the groups are equivalent, there should be no difference or only a small difference in the results obtained for each group.   Actually, we would expect some difference in average scores between the two groups. The difference could be in favor of one group or the other because of random variations among members in each group.   Any difference, however, should not be statistically significant.   (For a fuller discussion of this point, go to Chapter 19). If a significant difference is found, the experiment should be stopped and not continued until the groups are found to be equivalent.

After equivalence between the experimental and control groups is demonstrated, the experiment can begin.   The experimental variable is administered to the subjects in the experimental group, which in our example means they would participate in the educational program.   Those in the control group would continue with their ordinary daily lives.    After the education program is over a second set of measurements of prejudice would be obtained from all subjects in both groups.   The pretest and posttest data would then be used to test the hypothesis for the study.

For a brief, clear description of the two-group experimental design, visit: Two-Group Experimental Designs. This site also describes how the two-group design protects against threats to internal validity.