Using Twitter Data to Estimate the Relationship between Short-term Mobility and Long-term Migration

Abstract

Migration estimates are sensitive to definitions of time interval and duration. For example, when does a tourist become a migrant? As a result, harmonizing across different kinds of estimates or data sources can be difficult. Moreover in countries like the United States, that do not have a national registry system, estimates of internal migration typically rely on survey data that can require over a year from data collection to publication. In addition, each survey can ask only a limited set questions about migration (e.g., where did you live a year ago? where did you live five years ago?). We leverage a sample of geo-referenced Twitter tweets for about 62,000 users, spanning the period between 2010 and 2016, to estimate a series of US internal migration flows under varying time intervals and durations. Our findings, expressed in terms of 'migration curves', document, for the first time, the relationships between short-term mobility and long-term migration. The results open new avenues for demographic research. More specifically, future directions include the use of migration curves to produce probabilistic estimates of long-term migration from short-term (and vice versa) and to nowcast mobility rates at different levels of spatial and temporal granularity using a combination of previously published American Community Survey data and up-to-date data from a panel of Twitter users.