ABSTRACT
In preparation for various types of analysis, an appropriate choice of expression for the measured attributes must be determined. Rescaling, multiplication by and/or addition of a constant, is only the simplest and most trivial form of re-expression. Scaling responses in comparison with one another is meaningful so far as some techniques of linear ordering and fractionation go, but not for all. It depends on whether the different responses are to be used in a single analysis. Lodging is another matter and probably deserves to be scaled to have an overall standard deviation about the same as the median standard deviation of the five logarithmic responses. Few people — if any — can look at displays of two responses separately, and understand what a display of the sum of the responses would show them.
