Please use this identifier to cite or link to this item:
|Title:||A Statistical Rank Test For Analysing Biomedical Data|
|Authors:||Magee, Alexander Robert|
|Advisor:||Dunnett, Charles W.|
|Keywords:||Medicine and Health Sciences;Medicine and Health Sciences|
|Abstract:||<p>In the analysis of biomedical data, a question commonly asked by researchers involves the determination of the "best " or "worst" member of a group of results and an associated measure of the probability that this member is the "best" or "worst". Commonly, analysis of variance is suggested as the test of choice. Unfortunately, this test does not exactly answer the original question and further testing must be done to satisfy the question completely. This thesis presents a non-parametric rank test which directly answers the question of "best" or "worst".</p> <p>Before applying this test to biomedical problems, the probability tables associated with this test are expanded and the methods used are presented and discussed. An analogous parametric test is then described and compared in performance with the non-parametric test throughout the remainder of the thesis. Power curves for both the nonparametric and parametric test are developed for several population distributions and the results compared. The three areas of application are; chromosome frequencies in the culture of human melanoma tissue; scoring patterns among evaluators of letters of applications to medical school; and the determination of outliers when relating vital capacity to ventilatory response.</p> <p>It was found that except for cases where the number of objects was less than 10, the parametric test has equal or greater power than the non-parametric test when analysing continuous data, regardless of the population distribution. For less than 10 objects, the non-parametric test had greater power regardless of population distribution. Subsequent to analysis in the three areas cited, it was concluded that the two tests agreed very highly in selecting extreme deviates although the non-parametric, test was consistently more conservative in its probability measure. The problem of ties was found to weaken the power of the non-parametric test as did the ranking procedure itself but its ease of application and superior power with small sample sizes is a distinct advantage. The robustness of the parametric test is obvious throughout the examples. A method of selecting data values which are second or third most extreme was tested and it became obvious that the data must be displayed to show its distributional characteristics before this type of analysis could be carried out or interpreted.</p>|
|Appears in Collections:||Open Access Dissertations and Theses|
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.