Worrying about normality
John L. Moran, Patricia E. Solomon · Critical Care and Resuscitation · 2002
Some recent debates have highlighted the recurring problem of which statistical approach to use in the analysis of non-normally distributed data and/or small unequal datasets. Received wisdom suggests that, when comparing two independent groups in the presence of one or both of the above conditions, the t-test may be unreliable and the Wilcoxon-Mann-Whitney (WMW) test is preferable. When considering the use of statistical tests such as the t-test and the WMW test, an important and frequently overlooked assumption is that of independence of observations. It is important to realize that the WMW test gives no more protection against false-positive inference than the t-test; the WMW test, in its normal approximation, being equivalent to the t-test on the ranks of the original variables. Moreover, the interpretation of the null hypothesis being tested with the WMW test is not easily described. (non-author abstract)