Krapu, T. M. (1977). Locus of control measures: Test-retest reliability. Unpublished masters thesis, University of Missouri-Columbia.
Researchers, who construct instruments for use in their research or for uses such as evaluation and prediction outside of a research framework, are concerned with the validity and reliability of their measures. Unfortunately, another characteristic of a test often is not reported or is even overlooked when reliability data are being reported. This is the tendency for some test scores to increase or decrease in a Lest-retest situation. Since there is no intervening treatment to affect the retest scores, the cause of a change in scores is not immediately apparent. It could be explained by the subjects' expectations of what is expected of them by the experimenter. Also, part of this phenomenon could be seen as a form of practice effect, where cues are taken from the first experience of the test and then are present to affect the second administration. If a study were examining the effect of a treatment, this would threaten the validity of the study in question unless a control group was used. This change of scores over two administrations, when not caused by an intervening treatment, is what will be referred to in this study as a psychometric shift.
Statement of the Problem
Roberts and Griffiths (1937) reported just such a change when the Otis Intelligence Test was administered to ten and twelve year old subjects over a one to one and a half year period. That is, scores increased more over the test-retest period than what was expected from just maturation alone. They believed that some form of practice effect might have accounted for the change.
In the areas of opinion, attitude, and adjustment, Darley (1938) found several instances where test scores seemed to increase over a test-retest period of nine months. He stated that test-retest reliabilities are not always an indication of reliability alone. The stability of a trait or variable can also be reflected in a test-retest correlation and this can be totally unrelated to measurement error. In Darley's report the Minnesota Scale for the Survey of Opinions, the Adjustment Inventory, and the Minnesota Inventories of Social Attitudes all showed changes toward more satisfactory adjustment or greater conservatism in his college population. While these changes occurred over a period of about nine months, one might wonder if other variables or constructs which are being measured by tests don't also fluctuate significantly. Callis (1948) noted a psychometric increase in a measure of teacher pupil attitudes when there was no treatment. Using this natural tendency of the instrument, as measured by a control group, the results of his dissertation were different from what they would have been had the control group's changes not been taken into account.
In the area of personality, Schofield (1948) found changes in scales of the Minnesota Multiphasic Personality Inventory (MMPI) that were consistently toward what would be judged as better adjustment. This also occurred in a test-retest situation.
The Rotter I-E Scale is a measure of perceived internal versus external locus of control. Although satisfactory test-retest reliability coefficients have been reported for the Rotter I-E Scale (Rotter, 1966), means and standard deviations were not reported for the test-retest data. Furthermore, these statistics and the raw data were not preserved (Rotter, 1977). Evidence from the research cited up to this point raises two questions. Does a psychometric shift occur in the measurement of locus of control? Furthermore, if there is a significant shift, then what effect would this have on the results of a study that used a locus of control measure to evaluate a treatment condition when no control group was used? This study was conducted in an attempt to answer these questions.
SUMMARY AND CONCLUSION
This study was designed to see if there is any significant change of scores over a test-retest period of twenty-three days, on two locus of control measures. This study was also designed to determine if control groups are needed in locus of control research.
The Rotter and Nowicki-Strickland measures of internal versus external (I-E) locus of control were administered twice to fifty-five UMC students enrolled in an undergraduate education course. A one group, repeated measures, test-retest design yielded a significant t-ratio at the .05 level for only the Rotter I-E measure. This naturally occurring change in Rotter I-E scores within a no treatment test-retest situation indicates the necessity of a control group in certain types of studies of locus of control. This finding was used as the basis for a reanalysis of a study that was previously completed without the use of a control group.
While the limitations mentioned earlier do make the results of this study tenuous, it should be clear that future studies using a locus of control measure, particularly the Rotter I-E Scale, should formulate their research design to take into account the possible change in scores that occurred in the present study when no treatment intervened.