Do algorithmic job recommendations improve search and matching? Evidence from a large-scale randomised field experiment in Sweden
Department d'Economie et Sciences de la Décision
Intervenant: Roland Rathelot (CREST-ENSAE)
We design a job recommender system that recommends job ads to Swedish job seekers. The job recommender system is hosted on the largest online job board in Sweden, and it is based on a collaborative filtering machine-learning algorithm. We use a two-sided randomized experiment to evaluate how job seekers respond to job recommendations (clicks, applications, job finding, earnings), and whether employers fill their vacant jobs at a faster rate. We find that job seekers increase the number of job ads they view on a given day, with a larger treatment for recommended vacancies. They are also more likely to apply to vacancies that they were recommended. However, job seekers are not more likely to be hired in companies corresponding to the recommended job ads: the treatment effect on employment is on average zero.
Joint work with Thomas Le Barbanchon & Lena Hensvik