1.7 Transformations and scales of measurement
There are several reasons why data may need to be transformed prior to analysis.
Some parametric and non-parametric analytical procedures (analysis of variance,
Kruskall–Wallis tests of ranks) have underlying assumptions, for example, that the
data are normally distributed, or that variances among samples are not heterogeneous
(i.e. are of the same magnitude). If these assumptions are not met, there is
increased possibility of Type I errors (i.e. increased chance of mistakenly finding
apparent differences among samples, when they are, in fact, similar; Table 1.4).
Although different analyses have different levels of sensitivity to making Type I
errors, depending on the particular assumption being violated (Underwood, 1997a),
it may be necessary to transform data prior to analysis to meet the assumptions of
the analysis (Winer et al., 1991).