What Forms an Occupational Position? Predicting Workers’ Occupations through Bridging Occupational Theory and Machine Learning with Linked Employer-Employee Data

Occupations are central to understanding inequality. This paper asks: Can we predict occupations when data is missing? We combine Israeli administrative data from 2022—covering demographics, education, employment, and family characteristics (but not occupation)—with matched survey data that supplies 3-digit occupational codes for a sub- sample of workers. Using XGBoost decision trees, we classify individuals into occupational groups. We also assess whether the relationship between traits and occupations changes across contexts by examining how prediction accuracy varies by gender, ethno-nationality, education, and year (2022, 2008, 1995). Finally, we measure how much occupational distinctions—whether observed or predicted—explain variance in earnings, to evaluate the importance of including occupation in models of income inequality.

Research led by Prof. Tali Kristal. Additional Authors: Dan Manor, and Daniel Yahalom, University of Haifa