Reliable and timely estimates of migration flows are necessary to guide policy decisions and to improve our understanding of migration processes. However, obtaining these estimates remains an elusive goal. We propose an approach to combine geo-located Twitter data for over 2 million users (2010-2016) with data from the American Community Survey (ACS) in order to estimate US internal out-migration flows at state-level. We leverage the correlation structure for state-level biases in Twitter data by proposing a Bayesian hierarchical spatial model that smoothes out bias across space. We show that Twitter-based estimates can be combined with ACS estimates to improve short-term predictions of internal migration flows or to address problems of data sparsity.
Presented in Session 145. New Data Sources and Sampling Strategies for Identifying Migrants