An empirical study of motor performance data in an analysis of variance model under the generalized Behrens-Fisher problem : Krutchkoff's robust omnibus test K /
| Main Author: | |
|---|---|
| Other Authors: | , , |
| Format: | Thesis Book |
| Language: | English |
| Published: |
1993.
|
| Subjects: | |
| Online Access: | ProQuest, Abstract Link to OAKTrust copy |
| Abstract: | The purpose of this study was to examine, through Monte Carlo methods, the robustness efficacy of Krutchkoff's K statistic in the evaluation of motor performance data with unequal error variances in both balanced and unbalanced conditions for a two-factor factorial analysis of variance design. Twenty-eight different conditions involving group size, group error variance, and the pairing of each were tested using Monte Carlo simulation software. For logical relevancy, an original set of human motor performance (reaction time response) data was obtained and resampled to form the test data. From this test data, parameters were taken which were used as the simulation parameters. Ten thousand iterations were run and test size (actual Type I error rate), conditional rejections (with respect to the other test), and probability level means were collected. Test sizes for the F-test and the K-test were given, rejections decisions were compared with a McNemar (non-parametric) Test for Change, and probability level means and ninety-five percent confidence intervals were established. Results indicated that the F-test performed as well as or better than the K-test when all assumptions were met, but the F-test was biased when cell variances were unequal. The K-test was adequate under situations where assumptions were met and was unbiased in all situations where cell variances were unequal. A problem was noted with the K-test interaction effect test size: extremely conservative values were reported (in the range of 0.009 to 0.012 with nominal alpha at 0.05). The cause of this phenomenon was not determined. The author concluded that, while the F was indeed biased as expected, the K could not be unequivocally recommended due to the difficulties with the interaction effect test size. A researcher should be aware of the error variances in a given problem, and the K-test was suggested as an auxilary test when it is known that the assumption of homoscedasticity is not met. |
|---|---|
| Item Description: | Vita. "Major subject: Kinesiology." |
| Physical Description: | xvi, 134 leaves : illustrations ; 28 cm |
| Bibliography: | Includes bibliographical references. |