Fractionalization and inter-group differences

Fractionalization has been shown to have a detrimental effect on growth, public goods provision, and redistribution. The conventional measure of fractionalization is the Herfindahl index, which calculates the probability that two persons drawn at random belong to different groups. This measure implicitly assumes that all groups are equally distant. In this paper, I argue that a more appropriate measure of fractionalization should take into account that some groups are more different than others, so we need a measure of groups distance. We should then measure fractionalization as the average distance between every citizen, or equivalently the average distance between groups weighted by group size. I present a simple method to estimate these distances from opinion survey data by regressing stated opinions on indicator variables from group and a set of control variables. The coefficients on the group variables can then be interpreted as measures of distance. Finally, I apply the method to US data and show that we get more reasonable measures of fractionalization.