Percentages and other derived variables (and their problems) - If you divide one variable by another, and use that number (the ratio, such as x/y) as a variable in statistical analyses, you may open up a wide range of unexpected problems. A major justification people give for using ratios is to factor out size, however, ratios do not actually factor out size in a consistent manner among variables. A fairly complicated demonstration of this is given on P. 27-32 of ref. 1. Furthermore, ratios can produce spurious correlations, compound the measurement error and thus decrease both accuracy and precision and obscure the relationship between the variables in question. More details are given also in ref. 5, p. 17-19. Multivariate techniques allow a more robust description of variation, and should be used whenever possible instead of ratios.