Here’s a behind-the-scenes look at our recent @QJEHarvard paper—a project that began in one of my PhD classes at @EconUniMannheim and evolved into a full-scale field experiment on LinkedIn.

RT @JohnHolbein1: Did you know that your siblings can influence whether you vote?. Well, now you do!. Social scientists have long known tha….

RT @QJEHarvard: #QJE Feb 2025, #6, “LinkedOut? A Field Experiment on Discrimination in Job Network Formation,” by Evsyukova (@yu_evsyukova)….

RT @ZEW: 🤔 Gibt es #Diskriminierung beim Aufbau und der Nutzung von beruflichen Netzwerken wie @LinkedInDE? Das untersuchten #ZEW-Ökonomin….

RT @JournoResource: Young Black men fresh out of college may have a tougher time building professional networks on LinkedIn compared to the….

If you are interested in hiring a brilliant applied economist, Felix should be at the top of your list. Felix is certainly one of Mannheim's finest, and I strongly recommend taking a close look at his fascinating work!.

RT @JohnHolbein1: Research has shown that Uber drivers are 2x more likely to cancel a ride if the passenger’s name is one used predominantl….

RT @JPubEcon: Just published in @JPubEcon:. "The long run impact of childhood interracial contact on residential segregation". By @lmerlino….

RT @JPubEcon: Just published in @JPubEcon:. "Personal norms — and not only social norms — shape economic behavior". By @BasicZvonimir (@Uof….

RT @ellliottt: Recently posted on EJM: Postdoc in Economics and Data Science at @ETH_en with me and @sergallet:..

RT @FaHeine: Very interesting study on ethnicity-based discrimination in job network formation using data from a field experiment on Linked….

The project benefited tremendously from amazing feedback from many of you in various seminars and personal meetings. We are grateful for having had the opportunity to talk to so many people about the project and to receive funding from @EPoS224 which made the project possible.

This paper was a team effort in every sense. It took courage to pursue (and to stick to ideas), patience to solve countless challenges, and lots of late nights. Sometimes, the hardest projects are the most rewarding, and this one was no exception!.

Though the core results were apparent early on, it took us almost a full year to address further challenges, rule-out alternative explanations and provide a consistent explanation for our findings (which included an expert survey, a survey of LinkedIn users, etc.).

Once we successfully generated realistic-looking profile pictures that people could not distinguish from real ones (as confirmed by an MTurk experiment), we created authentic fake profiles on LinkedIn and launched the experiment in 2022.

So we created our own images and developed a method that could blend ethnic features—a solution that worked but took time to refine.

3) Obtaining unique AI-generated profile pictures was another challenge. Since LinkedIn uses reverse image searches, existing AI images weren’t an option. Further, we needed to keep facial features constant between Black and White profiles to causally identity the effect of race.

2) Managing 400+ profiles and scraping data from 20,000 users was another challenge (as LinkedIn actively works to prevent scraping). Hiring additional help wasn’t feasible, so we built our own setup to fully automate daily logins, manage connections, etc.

Even with this setup, we kept a low profile to avoid triggering LinkedIn’s algorithms. Many of the design choices also reflect our fear of being blocked (e.g., focusing only on the biggest cities, or largest employers, using common job-titles).

1) Running a field experiment on LinkedIn was very risky. Creating profiles without detection took many attempts, and in the end, required (among other things) unique IPs, email addresses, and phone numbers.