tending to score closer to the mean is known as ―regressing toward the mean‖ (sta- tistical regression). Likewise, those with very high abilities would also have a greater tendency to regress toward the mean—they will score lower on the posttest than on the pretest. Thus, those who are at either end of the continuum with respect to a variable would not ―truly‖ reflect the cause-and-effect relationship. The phenomenon of statistical regression is thus yet another threat to internal validity.
Mortality
Another confounding factor on the cause-and-effect relationship is the mortality or attrition of the members in the experimental or control group or both, as the experiment progresses. When the group composition changes over time across the groups, comparison between the groups becomes difficult, because those who dropped out of the experiment may confound the results. Again, we would not be able to say how much of the effect observed arises from the treatment, and how much is attributable to the members who dropped out, since those who stayed with the experiment could have reacted differently from those who dropped out. Let us see an example.
Example 7.2 A sales manager had heard glowing reports about three different training pro-
grams that train salespersons in effective sales strategies. All three were of 6 weeks‘ duration. The manager was curious to know which one would offer the best results for the company. The first program took the trainees daily on field trips and demonstrated effective and ineffective sales strategies through practical experience. The second program trained groups on the same strategies but indoors in a classroom type of setting, lecturing, role playing, and answering question from the participants. The third program used mathematical models and simulations to increase sales effectiveness.
The manager chose eight trainees each for the three different programs and sent them to training. By the end of the fourth week, three trainees from the first group, one from the second group, and two from the third group had dropped out of the training programs due to a variety of reasons including ill health, fam- ily exigencies, transportation problems, and a car accident. This attrition from the various groups has now made it impossible to compare the effectiveness of the various programs.
Thus, mortality can also lower the internal validity of an experiment.
IDENTIFYING THREATS TO INTERNAL VALIDITY
Let us examine each of the possible seven threats to internal validity in the con- text of the following scenario.
An organizational consultant wanted to demonstrate to the president of a company, through an experimental design, that the democratic style of leadership best enhances
156 EXPERIMENTAL DESIGNS
the morale of employees. She set up three experimental groups and one control group for the purpose and assigned members to each of the groups randomly. The three experimental groups were headed by an autocratic leader, a democratic leader, and a laissez-faire leader, respectively.
The members in the three experimental groups were administered a pretest. Since the control group was not exposed to any treatment, they were not given a pretest. As the experiment progressed, two members in the democratic treatment group got quite excited and started moving around to the other members saying that the par- ticipative atmosphere was ―great‖ and ―performance was bound to be high in this group.‖ Two members from each of the autocratic and laissez-faire groups left after the first hour saying they had to go and could no longer participate in the experi- ment. After 2 hours of activities, a posttest was administered to all the participants, including the control group members, on the same lines as the pretest.
History Effects. The action of the two members in the participative group by way of unexpectedly moving around in an excited manner and remarking that participative leadership is ―great‖ and the ―performance is bound to be high in this group‖ might have boosted the morale of all the members in the group. It would be difficult to separate out how much of the increase in morale was due to the participative condition alone and how much to the sudden enthusiasm dis- played by the two members.
Maturation. It is doubtful that maturation will have any effects on morale in this situation, since the passage of time, in itself, may not have anything much to do with increase or decrease in morale.
Testing. The pretests are likely to have sensitized the respondents to the posttest. Thus, testing effects would exist. However, if all the groups had been given both the pre- and the posttests, the testing effects across all groups would have been taken care of (i.e., nullified) and the posttests of each of the experi- mental groups could have been compared with that of the control group to detect the effects of the treatment. Unfortunately, the control group was not given the pretest, and thus, this group‘s posttest scores were not biased by the pretest—a phenomenon that could have occurred in the experimental groups. Hence, it is incorrect, on the face of it, to compare the experimental groups‘ scores with those of the control group.
Instrumentation. Since the same questionnaire has measured morale both before and after the treatment for all members, we do not expect instrumenta- tion bias.
Selection Bias. Since members have been randomly assigned to all groups, we do not expect selection bias to exist.
Statistical Regression. Though not specifically stated, we can assume that all the members participating in the experiment were selected randomly from a
FACTORS AFFECTING EXTERNAL VALIDITY 157
normally distributed population, in which case, the issue of statistical regression contaminating the experiment does not arise.
Mortality. Since members dropped out of two experimental groups, the effects of mortality could affect internal validity.
In effect, three of the seven threats to internal validity do apply in this case. The history, testing, and mortality effects are of concern and hence the internal validity will not be high.