LinkedIn ESG Research Panel

Challenge pressure

A quantitative research team needed Stata-ready panel data on more than 300 companies' LinkedIn activity, with employee posts tagged by their environmental, social and governance (ESG) content.

The situation

The team works in Stata and needed a panel: the same 300+ companies observed across multi-month windows. The signal it wanted sits in what employees post and share on LinkedIn, in a form no statistics package reads directly.

What we built

Five phases run in order: company profiles, their posts, the engagements on each post and the profiles of the people who engaged. The pipeline resumes after an error, so multi-day runs go unattended. A scoring pass then matches each post against environmental, social and governance dictionaries with N-gram matching, and the results are shaped into Stata .dta panel files with pandas.

What changed

The team received a panel dataset of more than 146 MB as Stata .dta files, covering 300+ companies over multi-month windows. The dashboard on this page, built later from the delivered posts file (1,395,138 rows, 20 fields, 450 MB, aggregated in full), shows the scorer at work: 108,376 posts carry a non-zero ESG score, and 19,185 of them score on two or more of the three dimensions.

Constraints and decisions

The scorer is deliberately conservative: it leaves 92.2% of posts at zero rather than guessing, so only 7.8% carry a score. Social is the most common signal (73,949 posts), governance follows (35,815) and environmental is the rarest (19,421). Posting volume peaked at 302,590 posts in 2021 and fell 10% in 2022. Employee posts name real people, so this page shows aggregates only.

Technology stack

PythonSeleniumNLPpandasStata