How to impute migration flows time series using a Bayesian Additive Mixed Model?'

Speaker(s):

Andrea Aparicio-Castro, University of Manchester

Abstract:

Data on migration flows usually suffer from a high level of missingness. Several studies have imputed missing migration flows by using auxiliary predictors. However, auxiliary variables highly correlated to migration often require additional imputations of missing observations. We propose a Bayesian additive mixed gravity model that imputes missing data on migration flows and overcomes the dependence of the imputation process on incomplete and unreliable auxiliary data. Our proposed model relies on a gravity model extensively used in estimating migration flows. Our model uses widely available data on the population size of origins and destinations and a measure of the distance to predict migration volumes. The additive mixed part of our model allows (i) modelling the non-linear relationship between flows and time and (ii) capturing the non-uniformly-spaced trends of flows. We illustrate the use of our model with administrative data on migration reported by South American countries from 2000 to 2019.