Looking for opportunities – On aggregation in random utility models for migration

This paper considers a random utility model (RUM) in which migrants view locations as aggregates of large numbers of alternatives from which they can freely choose. The best alternative is more likely to be found in a location where they are many and diverse. This predicted effect of size and dispersion contrasts with models considering regions or countries as atomistic units of choice, or where the outcome obtained within a location is stochastic and more dispersion implies more uncertainty. The coefficient on size equals 1 in an ideally specified RUM model including all relevant control variables and appropriate nesting of locations such that residual correlation between alternatives within locations is small. Only then intuitive spatial properties hold: there is zero predicted net migration between otherwise similar regions of different size, and migration flows scale proportionally when aggregating locations. Imposing proportional scaling also constrains how measures of size corresponding to distinct sets of alternatives (e.g. the number of houses and jobs in a location) should be combined. Lastly, assuming normally distributed returns from individual alternatives within locations, the coefficient on the variance should be close to 0.5 in the suggested framework. The approach is showcased and key predictions are tested in a study of internal migration and urbanisation in Ethiopia.