Affiliations: [a] University of California, Los Angeles, USA | [b] New York University, IZA and U.S. Census Bureau, USA | [c] U.S. Census Bureau, USA | [d] Ohio State University, IZA and NBER, USA
Correspondence:
[*]
Corresponding author: Julia I. Lane, Robert F. Wagner School of Public Service, New York University, New York 10012, USA. Tel.: +1 646 208 4990; E-mail: julia.lane@nyu.edu.
Abstract: Characterizing people’s occupations is important for both policy and research. However, as large scale administrative records are increasingly being used to describe labor market activity, it will become important to find new automated approaches to describing occupations. We apply new machine learning techniques to new sources of data and investigate the potential of using algorithms to classify occupations. We find that job titles are both inherently noisy and inconsistent across organizations, but a subset of them can be assigned algorithmically, with little impact on accuracy.
Keywords: Machine learning, occupations, UMETRICS, administrative data
