Possibilities of comparing more than two group means in a response variable in the case of failure of the assumptions of analysis of variance

Eva Kotlebová, Faculty of Economic Informatics, University of Economics in Bratislava, Slovak Republic
Martin Sliacky, Swis Re, Slovak Republic

Pages: 3 – 17


When using analysis of variance in practice, the assumption of homoscedasticity (equality of variance in the compared files) appears to be the most problematic. In such case (if the other assumptions are met), it is possible to use Welch analysis of variance where the ranges of individual samples are replaced by weights, in the calculation of test statistics, including sample variances. The paper compares the advantages and disadvantages of classical and Welch analysis of variance by means of error sizes of I and II types at different input values of file parameters, the outcome of which are the recommendations for choosing a suitable method for different combinations of sizes of the compared files and their variabilities.

Issue for download
PDF (1.9 MB, 169 downloads)