MIT Sloan PhD Student Zach Tan Wins Lucea Award

IWER April Newsletter: Generative AI and the Future of Work

Generative AI and Worker Productivity



Gendered Language in Job Postings Has Little Effect on Applicant Behavior, New Research Finds


Many contemporary U.S. employers seek to create gender-neutral job postings when recruiting. In fact, a number of software products are now available to help companies remove language that may have masculine or feminine connotations from job descriptions. The goal is to attract and hire a diverse pool of talented candidates. For example, a company that wants to encourage more women to apply for jobs in a traditionally male-dominated industry may try to craft gender-neutral job descriptions for its open positions.

But how much difference do such practices make in reality? Not as much as has been believed, suggests new research by MIT Sloan Professor Emilio J. Castilla and Hye Jin Rho, an assistant professor at the School of Human Resources and Labor Relations at Michigan State University. In a new article published in the journal Management Science, Castilla and Rho find that tweaking the language of job postings in an attempt to make them more gender-neutral has negligible practical effects on men and women’s likelihood of applying for the jobs. Castilla, who is NTU Professor of Management and a Professor of Work and Organization Studies at MIT Sloan, is Co-Director of the MIT Institute for Work and Employment Research (IWER), and Rho earned her doctorate from the IWER PhD program at MIT Sloan.

Castilla and Rho conducted two studies as part of the research for their Management Science article titled “The Gendering of Job Postings in the Online Recruitment Process.”  In the first study, they analyzed a large data set from an online recruiting platform based in North America that included close to 300,000 job postings over two years along with, in many cases, information that indicated the gender of the recruiters associated with the postings. The researchers also had access to anonymized information about the interactions via the platform of more than 485,000 unique job seekers who viewed the postings and who, collectively, sent, more than 590,000 inquiries to recruiters on the platform to learn more about particular jobs.

Castilla and Rho used a machine learning-based word embedding technique to analyze the occurrence of stereotypically masculine words (for example, “assertive” and “determined”) vs. stereotypically feminine words (for example, “cooperative” and “interpersonal”) in the job listings in this data set and gave each job listing a score based on how stereotypically feminine or masculine the overall language use was. The researchers then analyzed the effects of that score on applicant inquiries (while controlling for other factors). They also studied the effect of the gender of the recruiter, when it could be determined from the data set, on job seeker behavior.

Castilla and Rho found that women were slightly more likely to apply to jobs with more feminine wording, but that the differences, while statistically significant, were very small.  The effects of the gender of the recruiter on applicant behavior were “negligible,” the authors write.

(Pictured: Emilio J. Castilla, left, and Hye Jin Rho, right)

Castilla and Rho followed up their field study with a separate field experiment where job seekers on Amazon Mechanical Turk viewed a listing for a job whose language the researchers varied to in some cases be more stereotypically masculine and in others more stereotypically feminine. The researchers also varied the gender of the person applicants would contact for more information about the job. As in the first study, Castilla and Rho found that the effects of tweaking the language were so small as to have little practical impact on job-seekers’ likelihood of applying for the position.

“Our findings reveal that both the language used when posting jobs and gender of the recruiters have no effects that matter in practice on how women and men behave during recruitment,” Castilla and Rho conclude. This outcome, Castilla and Rho write, was “unexpected.” However, the authors note that their findings have significant practical implications, as their research suggests that the increasingly popular approach of trying to “degender” job listing language may not have the effect employers hope for.

“We caution that the practice of simply altering the language of job descriptions may not necessarily help organizations address diversity issues,” the authors write. Instead, Castilla and Rho conclude, “we encourage scholars and employers to propose and test better solutions” for achieving gender diversity in their applicant pools.