2021-12-29-兰德-利用社交媒体数据_即时播报_全球的国际移民（英）_162页_6mb

Summary

This study developed a methodological tool to 'nowcast' international migrant stocks using real-time data from the Facebook Marketing API combined with official migration data from EU member states and US states. The research aimed to provide timely estimates to address the significant time lag in conventional migration statistics and their potential underestimation.

Key objectives were:
(i) Collect real-time data from Facebook and official migration statistics from 2010 onwards
(ii) Develop a Bayesian model combining Facebook data and official sources to nowcast migrant stocks

Core methodology involved:

• Using the 'Lived In' status on Facebook as a proxy for migration history
• Analyzing data from 2019-2021 with advanced statistical techniques
• Applying Bayesian Gaussian process models to combine multiple data sources while accounting for varying quality and time resolution

The approach demonstrated promising results but important limitations exist:

• Facebook penetration varies globally, and data may reflect social media trends rather than migration patterns
• The approach was effective but requires future extensions with longer time series data overlaps for validation

Potential applications include early warning systems for migration trends during crises like the COVID-19 pandemic and more responsive migration policy-making.

Future work suggestions include:

• Extending data time series further
• Incorporating more demographic detail using Facebook's capabilities
• Improving model design through hierarchical approaches