This paper makes use of a gravity model to investigate the determinants of global forced migration. We find that omitting zero counts results in parameter estimates that underestimate covariate effects on refugee counts. We compare the pooled regression, which does not account for unobserved heterogeneity, and the fixed-effects method, which does not identify the effect of time-invariant covariates, and cannot shed much light when covariates vary little. We propose the pre-sample mean generalized method of moments (PSM-GMM) estimator, which accommodates the zeros, accounts for unobserved heterogeneity, but does not have the drawbacks of the fixed-effects methods when covariates exhibit little variation. In addition, using recently developed methods to estimate standard errors that are robust to dyadic correlation, we find evidence suggesting that previous findings based on underestimated standard errors mostly remain valid after properly adjusting standard errors; in particular, conflict and civil liberties in the source country and proximity are found to be significant determinants of forced migration. However, some covariates previously found to be significant, such as sharing a common language or having a colonial relationship, lose significance. In addition, we find a significant positive influence of the level of civil liberties in the destination country on the number of refugees it receives, a mechanism not explored before. © 2016, Springer Science+Business Media New York.

Gravity model Pre-sample mean generalized method of moments Refugee migration

[No Keywords available]