The invariance properties of the Mutual Information index of multigroup segregation

In the context of educational segregation by ethnic group, it has been argued that rigorous pair wise segregation comparisons over time or across space should be invariant in two situations: when the ethnic composition of the population changes while the distribution of each ethnic group over the schools remains constant (invariance 1), or when the size distribution of schools changes while the ethnic composition of each school remains constant (invariance 2). This paper makes three contributions to this literature. First, it presents a testing strategy for choosing between the two properties. Second, it argues that both properties have strong implications, and that there are reasons to defend that the overall segregation index need not satisfy either one. In particular, the contrast between invariant segregation indices and the Mutual Information segregation index that violates both properties is illustrated with a number of examples. Third, nevertheless, it is shown that pair wise segregation comparisons using this index can be expressed in terms of (i) changes in the ethnic composition of the population, (ii) changes in the school size distribution, and (iii) changes in a third term that is invariant 1 or invariant 2. These decompositions can be used to reach the analogous ones obtained in Deutsch et al. (2006)