WFH Map

Measuring remote work across space and time, using job ads.

Welcome

This page contains selected figures, as well as a portal for researchers to access the data used in our research paper: “Remote Work across Jobs, Companies and Space” (Hansen, Lambert, Bloom, Davis, Sadun & Taska, 2023).


The below figures are interactive and data is available to download


Figure 1: Share of new job vacancies advertising remote work rose dramatically

Note: This figure shows the share of vacancy postings that say the job allows one or more remote workdays per week. We compute these # monthly, country-level shares as the weighted mean of the own-country occupation-level shares, with weights given by the U.S. vacancy distribution in 2019. Our occupation-level granularity is roughly equivalent to six-digit SOC codes. Figures depicts the 3-month moving average. Full screen. Download the data.


Figure 2: Advertised remote work in 2022 is very uneven across occupations

Note: This figure shows the share of new vacancy postings across occupations (using USA data) that say the job allows one or more remote workdays per week. Each bar is an ONET code (scroll over the top to find out which one!). We display the top ten occupations (by count of postings) in each selected group. Full screen. Download the data.


Figure 3: Advertised remote work shows substantial divergence across US cities

Note: This figure shows the share of new vacancy postings across selected US cities that say the job allows one or more remote workdays per week. Full screen. Download the data.


How we measure advertised remote work…

Our team of researchers joined forces with Lightcast, a labour-market analytics firm, and developed a text algorithm capable of measuring offers of remote work (i.e. one or more days per week at home) directly from the raw text of online job vacancy postings.

Our data spans the near-universe of new online job vacancy postings, and currently contains over 250 million documents.

Our approach to measurement has a number of distinct benefits (and some drawbacks). For example:

  • Our dataset is huge, covering five countries, thousands of cities and employers, and hundreds of occupations
  • New vacancies are posted daily, allowing us to measure advertised remote work in near-realtime. We aim to refresh data every three months.
  • The algorithm we developed achieves a 98% accuracy at identifying remote work job ads
  • We measure explicit offers of remote work in new job ads, not the number of employees who work-from-home (for this see survey measures available at wfhresearch.com

These data may be used by researchers and other interested parties. For any specific data requests, media comments, or technical issues, please email Peter John Lambert (p.j.lambert[at]lse.ac.uk).

Blog at WordPress.com.