Monthly Archives: November 2013

Fractionalization and flawed data

Ethnic and cultural fractionalization have become popular concepts in the social sciences. Intuitively, one may hypothesize that more fractionalized countries are more prone to ethnic conflicts and civil war, which is why there has been a stream of papers published on this topic. However, one may question the data that such research uses.

The popular fractionalization dataset that Alesina et al. have developed was supposed to ameliorate many of the problems that previous data have suffered from. Nevertheless, this dataset also seems to have major drawbacks . First of all, like most fractionalization datasets, it lacks a time dimension. Estimating correlation at one fixed point in time naturally yields less powerful results than an analysis that estimates the relationship over time. The question “Does ethnic fractionalization increase the probability of civil war?” is likely best answered by instead directly approaching the question “Does increased ethnic fractionalization over time increase the probability of civil war?”

A second problem is that many countries in the dataset are coded in terms of “citizenship fractionalization” rather than ethnic fractionalization, and there is no consistency or rationale for this coding. For example, Finland – according to the data – is considered more fractionalized than Sweden because the 6 percent Swedes in Finland are counted as an ethnic minority, whereas all Western Europeans in Sweden are considered part of the ethnic majority. So Swedes in Finland increase the fractionalization in Finland, but Finns in Sweden decrease the fractionalization in Sweden. Or in other words, if a Danish family moves to Finland, it will increase Finland’s fractionalization, but if the same family moves to Sweden, it will decrease it.

Simon Hedlin