The analysis of clinical studies: The use of transformations to achieve normality

Abstract

In previous columns it has been necessary to make some strong assumptions about data, to the effect that data are “well behaved,” in order to present certain concepts. For instance, in one of those columns the following remark was made: “Assuming that the pre-test data are not skewed to either scale endpoint, 0 or 10, and the data appear unimodal and symmetric, it is appropriate to focus on differences in mean values utilizing normal theory.” The fact is that many of the common statistical procedures used in data analysis, such as the testing of hypotheses with normal theory, have assumptions about the nature of the data that are often not met, and investigators need to know how to deal with pathologic data. This column will focus on some of the pathologic situations related to departures from normality that occur and that preclude the use of the better-known tests, as well as on some remedies.