The methods used in nonparametric covariance analysis

Abstract

Aim: Nonparametric covariance analysis (ANCOVA) methods are used when the assumptions of parametric ANCOVA are not met and/or the dependent variable has bivariate/ordinal scale. In the nonparametric ANCOVA methodology, Quade, Puri & Sen and McSweeney & Porter methods are known as Ranked ANCOVA methods. However, commonly used programs do not have module(s) for applying these methods. The objective of this study is to introduce the ranked ANCOVA methods, to apply it in a web-based program developed by the authors and to present the advantages of these methods. Material and Methods: The theoretical features and application steps of the Ranked ANCOVA methods are defined and a web-based program for the application of each method has been established. The application of each method on this program with the help of simulated data taken from the health field study, where the effect of cigarette smoking on biochemical tests was examined has also been included. Results: Although there is no specific module in the widely used statistical programs for the methods described in this study, it is shown on a clinical study constituted with simulated data that these methods can easily be applied and the results of the methods are given. Conclusion: The use of parametric methods for factorial models leads to an increase in Type-I error rate and a decrease in test power in many studies, where the sample size is limited and/or the dependent variable does not have normal distribution. To reduce this error, we recommend using the methods suggested in the study. These methods are also expected to reach widespread use thanks to the web-based program. © 2018, Duzce University Medical School. All rights reserved.