A measure of adjusted difference between values of a variable

Abstract

Both during academic work and everyday life it is usual to come across such situations, where one has to compute the difference between values of a variable. However, this can be done many ways. For example, the comparison of accuracy values of methods for the solution of a scientific problem is usually done by either looking at how much the absolute difference between the two accuracies in percentage points is, how much their relative difference in percentages is, or how much the relative difference between the two error rates in percentages is. The problem is that these different methods usually give different results, often suggesting contradicting conclusions. To overcome this problem I propose a novel measure to compute an adjusted difference ( ) between values of a variable to represent their difference in a better and uniform way, replacing the collection of the previously used conventional measures, and show how the properties of this new measure are superior to those of the other measures.