Gale-Shapley Algorithmic Match Recommendation
RCTReview
Large multi-employer job fairs often suffer from high search and congestion costs, where attendees waste time in random, low-probability encounters. Unstructured interactions prevent firms and jobseekers from identifying mutually beneficial matches within limited event timeframes.
Picture this
Imagine a speed-dating event with hundreds of participants where people wander aimlessly and queue at popular tables. Instead of random mingling, a central coordinator collects basic background profiles from everyone beforehand and uses an automated preference-matching rule to print a personalized schedule of 10 targeted meetings for each person, ensuring attendees meet partners who actually fit their criteria.
What the evidence says
Algorithmic match recommendations increased the probability of actual worker-firm meetings by ~200% compared to non-recommended pairs ($p < 0.01$), whereas randomly suggested match recommendations showed no statistically significant effect ($p > 0.10$).
- Who was studied
- N = 1,007 invited jobseekers and N = 248 invited formal firms (606 jobseekers and 170 firms attending across two two-day job fairs) evaluated via dyadic regressions on 27,778 worker-firm pairings in Addis Ababa, Ethiopia.
- How
- Field experiment randomized trial comparing Gale-Shapley Deferred Acceptance algorithm recommendations (iterated 10 times with synthetic rankings based on education, experience, and wages) against 5 random match recommendations.
What to do
Construct synthetic lexicographic preference rankings using baseline survey variables (education, previous occupation, and advertised wages) and execute a Gale-Shapley deferred acceptance algorithm to schedule targeted meetings at job fairs.
From the source
"Algorithmic recommendations are also shown to have a strong predictive power: matches suggested by our algorithm are about 200 percent more likely take place than non-suggested matches."
Matching_Firms_and_Young_Jobseekers_through_a_Job_Fair_A_Field_Experiment.pdf