Econometric Features of Models Used in Aid Allocation Studies

Sammanfattning: Research of aid allocation deals with a dependent variable, received aid expressed in absolute or relative terms, that is equal to zero for numerous observations. This is because donor countries tend to target specific countries for their allocation, leaving the rest of the countries without any development assistance. The special characteristic requires non-linear methods with censoring, truncation or selection bias of the data. Using a panel covering Swedish aid recipient countries between 1998-2016, three commonly used models in aid allocation are examined; a multiple linear regression on a truncated data set, Heckman’s two step model and the Tobit model. The models are estimated with a set of political and altruistic variables that are frequently used as explanatory factors to aid allocation, with share of Swedish aid as dependent variable. With around 13 % of the total observations below threshold, the models yield similar parameter estimations. The results from Heckman’s two step model and the multiple linear regression are almost, but not exactly, the same. This can partly be explained by the information each model has of the dependent variable, and partly by a small selection bias. In general, the parameters have approximately the same impact on the dependent variable. However, the estimations in the Tobit model are slightly different from the other models. Countries in sub-Saharan Africa, the educational level and the expected lifespan in the recipient country have a significant effect on the share of Swedish aid received according to the Tobit model, but not in the other models. This is mainly explained by the different estimation methods in the models.