Date of Award
Master of Science (MS)
Robert D. Pritchard
One potential problem associated with the use of the Pearson Product Moment Correlation Coefficient (r), is that various types of errors may contaminate and thus influence the value of r that is observed. This study was undertaken to determine whether or not two specific classes of error have a significant impact on the magnitude of the correlation observed between two measuring instruments. One source of error arises from the context in which values for the variables to be correlated are obtained. A second source of error arises from the moods of the subjects who are used to generate values for the variables to be correlated. This study attempted to reduce the influence of the context by using a different context for every observation (i.e., randomize contextual stimuli). To reduce the influence of mood on the magnitude of r, the time interval between the administration of the two instruments was increased, in the hope that the increased time interval would reduce the probability that the subjects would experience the same mood on both instrument administrations. For this study, it was specifically hypothesized that a randomization of contextual stimuli, coupled with a separation in time between the administration of two measuring instruments would be associated with a reduction in the amount of correlated error. To test this hypothesis, one group of subiects (high error group) was administered an internal-external locus of-control scale and a se lf-esteem scale in the same context and at the same point in time. A second group of subjects (low error group) was administered the same two scales in differing contexts and at different points in time (i.e., a two day delay separated the completion of the two scales). For this low error group, a randomization of contextual stimuli was achieved by selecting one experimenter and one room from a pool of nine available experimenters and six different rooms for each administration of both scales. Correlations between the two scales were then computed separately for each group. Sample correlations of ,371 and .246 were observed for the high and low error groups respectively. Using a one-tailed test of significance, these correlations were not found to be significantly different from one another. Analysis of the manipulation checks indicated a partial, but weak impact on the subjects due to variations in the experimenters and rooms. Furthermore, while the low error group reported a greater frequency of changes in mood over the high error group, these differences were not statistically significant. It was concluded that while some practical importance may be associated with the way the two sample correlations would be interpreted, the results of the experiment may be regarded as inconclusive. Future tests, with stronger manipulations are called for in the assessment of the impact of correlated errors on the correlation coefficient.
Mayo, Robert Joseph, "An Investigation of the Effects of Correlated Error on the Magnitude of the Correlation Coefficient" (1974). Open Access Theses. 1244.