Periodic Reporting for period 4 - ESEARCH (Direct Empirical Evidence on Labor Market Search Theories)
Berichtszeitraum: 2022-10-01 bis 2023-09-30
Unemployment is a major economic and social issue of our society. Our project provides new empirical evidence on the job search strategies of workers. Our overall objective is to enhance our understanding of the source of frictions on the labor market, and to provide guidance for policy-makers.
We reached three main conclusions:
1. Providing personalized job recommendations to unemployed worker creates more matches on the labor market. We design, implement and test new machine-learning job recommender systems developed on the main online job board in Sweden (plastbanken.se). We find that job seekers follow their personalized job recommendations, and they benefit from it. Their employment increases during the 6 months following their first exposure to the recommender system (by 0.6 percent). The job recommender system helps to solve coordination failures on the labor market, occupational mismatch, and inattention bias during job search. The job recommender system generates only limited congestion on popular jobs. Overall, our project created around 500 extra matches in the Swedish labor market during one year.
2. Online job search leaves very useful data on online job boards, that can be used to analyze recent labor market evolution in real-time. During the COVID crisis, we provided and analyzed new real-time indicators of the state of the labor market based on online job search and vacancy posting activity. These indicators are particularly useful in uncertain times.
3. Gender gaps in commuting time between work and home contribute significantly to gender inequality in pay. We contrast how men and women search on the labor market, especially the type of jobs they are looking for. We find that women are looking for jobs closer to their home than men, and are willing to accept lower wages through that process.
Our results guide policy-makers, who design public job boards and transportation policies, and more broadly policies addressing sources of inefficiencies in the labor market.
We reached three main conclusions:
1. Providing personalized job recommendations to unemployed worker creates more matches on the labor market. We design, implement and test new machine-learning job recommender systems developed on the main online job board in Sweden (plastbanken.se). We find that job seekers follow their personalized job recommendations, and they benefit from it. Their employment increases during the 6 months following their first exposure to the recommender system (by 0.6 percent). The job recommender system helps to solve coordination failures on the labor market, occupational mismatch, and inattention bias during job search. The job recommender system generates only limited congestion on popular jobs. Overall, our project created around 500 extra matches in the Swedish labor market during one year.
2. Online job search leaves very useful data on online job boards, that can be used to analyze recent labor market evolution in real-time. During the COVID crisis, we provided and analyzed new real-time indicators of the state of the labor market based on online job search and vacancy posting activity. These indicators are particularly useful in uncertain times.
3. Gender gaps in commuting time between work and home contribute significantly to gender inequality in pay. We contrast how men and women search on the labor market, especially the type of jobs they are looking for. We find that women are looking for jobs closer to their home than men, and are willing to accept lower wages through that process.
Our results guide policy-makers, who design public job boards and transportation policies, and more broadly policies addressing sources of inefficiencies in the labor market.
The first important step of our project was to gather data on both online job search and labor market outcomes at the worker level. This was made possible thanks to a partnership with the Swedish Public Employment Service (Arbetsformedlingen) and their online job board (platsbanken.se). The unusually rich data allowed us to design a new job recommender system based on an item-to-item collaborative filtering algorithm. Within our partnership with platsbanken.se we implemented the recommender system online. From April 2021 to March 2022, users of the online job board platsbanken.se viewed personalized vacancy recommendations. We designed a randomized control trial to assess the effectiveness of the job recommender system, implemented at the national scale in Sweden. Analyzing the data of the trial, we reached the positive conclusion that the job recommender system created extra matches on the labor market. The job recommender system solved an array of inefficiencies in the search and matching process without creating extra congestion. We disseminated those results in workshops at the Swedish Public Employment Services, and in 26 academic seminars and workshops.
In 2020, thanks to the same data on online job search, we provided new real-time indicators of labor market tightness during the COVID-19 crisis. The indicators are based on the number of clicks that vacancies receive on online job boards. We published their evolution and interpreted them in a peer-reviewed academic article “Job Search During the COVID 19 crisis”. The results were also presented in numerous academic seminars and workshops. They gained the attention of economists at OECD, of various European Public Employment Services, and of the French Council of Economic Analysis (CAE).
We conducted a thorough empirical analysis of gender differences in job search in French administrative data. Among the most robust gender differences, we observed that women search for jobs closer to their residence than men, and have lower wage demands. We showed that women are willing to accept larger wage reduction than men in order to work closer to their home. This commute channel explains 10 percent of the gender wage gap. The results of our study are published in a peer-reviewed academic journal “Gender Differences in Job Search: Trading off Commute against Wage”. We presented the study in numerous seminars and conferences.
In 2020, thanks to the same data on online job search, we provided new real-time indicators of labor market tightness during the COVID-19 crisis. The indicators are based on the number of clicks that vacancies receive on online job boards. We published their evolution and interpreted them in a peer-reviewed academic article “Job Search During the COVID 19 crisis”. The results were also presented in numerous academic seminars and workshops. They gained the attention of economists at OECD, of various European Public Employment Services, and of the French Council of Economic Analysis (CAE).
We conducted a thorough empirical analysis of gender differences in job search in French administrative data. Among the most robust gender differences, we observed that women search for jobs closer to their residence than men, and have lower wage demands. We showed that women are willing to accept larger wage reduction than men in order to work closer to their home. This commute channel explains 10 percent of the gender wage gap. The results of our study are published in a peer-reviewed academic journal “Gender Differences in Job Search: Trading off Commute against Wage”. We presented the study in numerous seminars and conferences.
Our project made important progress beyond the state of the art.
In labor economics, it shows for the first time the feasibility and more importantly the effectiveness of machine-learning job recommender system delivering highly personnalized vacancy recommendations. This is based on a national-scale randomized control trial, the golden standard of program evaluation. Computer scientists showed also great interest in the recommender system design and its evaluation as well. Along the way, we developed new experimental design methods suited to such demanding evaluations and beyond the state of the art.
Our project provided the first estimate of the contribution of gender differences in commute valuation to gender inequality in pay. This makes a contribution to labor economics beyond the state of the art. It is based on new methodology to estimate willingness to pay for job attributes.
In labor economics, it shows for the first time the feasibility and more importantly the effectiveness of machine-learning job recommender system delivering highly personnalized vacancy recommendations. This is based on a national-scale randomized control trial, the golden standard of program evaluation. Computer scientists showed also great interest in the recommender system design and its evaluation as well. Along the way, we developed new experimental design methods suited to such demanding evaluations and beyond the state of the art.
Our project provided the first estimate of the contribution of gender differences in commute valuation to gender inequality in pay. This makes a contribution to labor economics beyond the state of the art. It is based on new methodology to estimate willingness to pay for job attributes.