Tigran SOGOMONIAN, The University of Manchester, United Kingdom
Despite millions of people moving between states in India each year, no apparatus other as decennial census exists to track bilateral inter-state migration in a timely way. This was costly during the first COVID-19 pandemic lockdown, when state attempts to co-ordinate emergency relief for millions of displaced internal migrants was fraught by a lack of recent, detailed data on labour migrants.
To fill this void in migration estimates, we used georeferenced social media data from Twitter to create a synthetic dataset of bilateral inter-state migration rates in India over the period 2015-21. We used a sample of 2.98 million geotagged tweets across 17.6 thousand Twitter users as the source data and leveraged earlier migration rates from the IMAGE-Asia project as prior information within a Bayesian inferential framework.
By harnessing the spatial and temporal granularity inherent in social georeferenced data, we were able to analyse spatial migration patterns for both long-term and temporary migrants in India. We found that Twitter-derived estimates yield qualitatively valid results that fit conventional gravity models and exhibit spatial migration patterns similar to those previously observed in India. This also allowed us to model the effects of the non-pharmaceutical interventions on inter-state mobility in 2020.
The data, however, carries significant sampling bias and Twitter-derived estimates of migration rates cannot yet be generalised to absolute numbers of migrant flows. We suggest ways to resolve this bias and make the case that georeferenced social media data has useful potential for estimating internal migration for recent periods or in poor data situations.
Mots clés : social media data|internal migration|India|Bayesian inference|data integration
A104871AW