ABSTRACT
A useful rule of thumb when no such extra information is available is provided by the"~-~" rule, as proposed by Champernowne in an unpublished manuscript and as formally derived in Cowell and Mehta (1982): Approximate the unknown true Gini index by ~(h(x) + ~Gu(x). As shown by Cowell and Mehta, this rule gives the exact Gini index when the observations within each group are split at x(i) and spread evenly across the subintervals (ai--l, :xUl) and (x(i), a;) such that the group mean is preserved (the "split-histogram technique"'), but it gives a good approximation to the true Gini index also for many other disturbances.
