Why Do People Move ? A Data-Driven Approach to Identifying and PredictingGender-Specific Aspirations to Migrate

Work-related migration has many potential drivers. While current literature has outlined atheoretical framework of various “push-pull” factors affecting the likelihood of international migration,empirical papers are often constrained by the scarcity of detailed data on migration, especially in developingcountries, and are forced to look at few of these factors in isolation. When detailed data is available, researchers mayface arbitrary choices of which variables to include and how to sequence their inclusion. As male and female migrantstend to face occupational segregation, the determinants of migration likely differ by gender, which compounds thesedata challenges. To overcome these three issues, this paper uses a rich primary household survey among migrantcommunities in Indonesia and employs two supervised machine-learning methods to identify the top predictors ofmigration by gender: random forests and least absolute shrinkage and selection operator stability selection. Thepaper confirms some determinants established by earlier studies and reveals several additional ones, as well asidentifies differences in predictors by gender.