ABSTRACT
When data are not normally distributed or when levels of an IV do not have homogeneous variance, then it is often inappropriate to use a parametric statistical test. However, it is sometimes possible to transform the data so that it is more normal or so that variances are closer to each other. In addition, when looking at the relationship between two variables (bivariate data), if there appears to be a relationship between them but one that is non-linear then one of the variables can be transformed to produce a more linear relationship. To transform data is to apply the same mathematical formula to each of the values in a set of data. You may think that this appears like fiddling with the data to get the answer which you want. However, as long as you make the transformation in order to put the data into a form which would allow a parametric test or a linear test to be conducted, then it is perfectly legitimate. What is not legitimate is to try one transformation, run a statistical test on the data and then go on to try another transformation if you do not achieve statistical significance.
