Online Invited Seminar Talk
Predicting Global Migration with Social Media Data: Potential and Limits
Laboratory of Digital and Computational Demography
Online Invited Seminar Talk, March 09, 2021
Jordan Klein from Princeton University discussed the potential for and the limits to using Facebook data to predict migration on a global scale.
Estimating international migrant stock, or the number of people living in countries other than those of their birth, is of great importance to demographers and policymakers. To meet these needs, the United Nations publishes estimates of global international migrant stock biennially, using data collected by member states and statistical methods to fill in the gaps. Consequently, these estimates reflect the great degree of heterogeneity in the methods the countries of the world use to collect migration data, the frequency with which these data are collected, and their overall quality. Simultaneously, social media platforms, in particular Facebook, are ever more ubiquitous, even in parts of the world without high quality migration data. The feasibility of using Facebook data to predict migrant stock has been explored in previous work. While its utility is most clear in certain parts of the world with higher quality migration data and Facebook penetration rates, it is less so across a wider spectrum of countries with greater heterogeneity of migration data quality and Facebook usage. Using UN migrant stock estimates as a “ground truth”, I build on this previous work and explore the potential for and the limits to using Facebook data to predict migration on a global scale. I examine: 1) whether models using Facebook data improve predictive accuracy compared to baseline models using previous years’ migrant stock, 2) which machine learning methods are most accurate, 3) how these findings and overall predictive accuracy vary when considering countries with different levels of migration data quality, and 4) implications for limits to prediction when the data used as a “ground truth” is itself only an estimate of the ground truth.
Jordan Klein is a PhD student in demography at the Princeton University Office of Population Research. At Princeton, his research has focused on the following two themes: 1) How human population health changes over time in different parts of the world, particularly through infectious disease and environmental factors, and how we can measure, model, and explain these changes. 2) The feasibility and ethics of using “big data” (eg digital trace data) for conducting demographic estimation, particularly among vulnerable populations. Before coming to Princeton, he earned a BS in international relations and biology and an MPH in epidemiology and biostatistics from Tufts University, and worked at the World Health Organization and Abt Associates, a U.S.-based policy research firm.