How job ads shape gender and racial segregation in the UK workforce – new study

In the UK, equality, diversity and inclusion (EDI) language in job advertisements (ads) could unintentionally have the reverse effect on attempts to create a more gender-balanced workplace, says a new study led by Lancaster University.

The study also shows that actively using EDI language designed to appeal to racial minorities in job ads is not working.

Workforces with a larger share of women tend to include language associated with family-friendly policies and flexible work arrangements in job ads. Such language, the study suggests, tends to appeal more to female rather than male applicants, which, in turn, intensifies gender segregation in these workforces.

While workforces with a larger share of racial minority workers tend to include more EDI policy pledges and language signalling workplace EDI culture in job ads, such pledges and language have little impact on workforce racial composition, says the research.

The study, ‘Language in job advertisements and the reproduction of labour force gender and racial segregation’, is published in PNAS Nexus – an official journal of the National Academy of Sciences of the United States of America.

The findings provide a labour-market-wide audit of how gender/EDI language in job ads helps shape workforce gender/racial composition, as well as how labour force gender/racial composition influences gender/EDI language in job ads.

This collaborative research, bringing together researchers from universities across the UK, Canada and the USA, is jointly funded by the UKRI Economic and Social Research Council (ESRC) in the UK and the Social Sciences and Humanities Research Council (SSHRC) of Canada, as part of a large-scale project looking at AI and labour market equality.

Working in collaboration across multiple disciplines, the team of sociologists, management scholars and data scientists used cutting-edge natural language processing techniques to analyse 28.6 million job ads in the UK in combination with ONS labour force statistics between 2018 and 2023, making it the most comprehensive and up-to-date study of its kind.

“Understanding and tackling persistent labour force gender and racial segregation are crucial to facilitating equality and diversity in the labour market,” says the lead author, Professor Yang Hu, of Lancaster University.

“Job ads are important because they are the first point of contact between job seekers and employers,” said Associate Professor Nicole Denier of the University of Alberta.

“By signalling characteristics expected of an ‘ideal candidate’, job ads ‘gatekeep’ the labour force and configure its composition by shaping both candidates’ tendency to apply for a job and the criteria used for shortlisting and interviewing.”

The study develops a novel inventory of language in job ads, capturing six dimensions of language related to gender and EDI:

• Explicit gender references such as ‘he’, ‘she’, ‘men’ and ‘women’
• Gendered psychological cues such as ‘caring’ and ‘attentive’ vs. ‘authoritative’
• Gendered work roles and skills, such as ‘soft’ and ‘social’ skills vs. roles involving ‘multitasking’ and high ‘pressure’
• Family-friendly policies (e.g. flexible work, work-family balance) vs. family-unfriendly arrangements (e.g. irregular shifts, long hours)
• EDI policy pledges, such as references to the Equality Act and Stonewall
• EDI cultural references, for example, signalling the workplaces as ‘supportive’, ‘accessible’, ‘diverse’ and ‘inclusive’

Using this newly developed inventory, the study characterised and mapped the gender/EDI language used in job ads across occupations and industries to the gender and racial composition of the corresponding workforce in these occupation and industry groups across the full UK labour market.

The study is the first of its kind to disentangle how language in job ads shapes labour force gender/racial composition and how labour force gender/racial composition shapes language in job ads in both directions.

The findings reveal three distinct ways in which the interplay between language in job ads and labour force composition reinforces or disrupts labour force/gender segregation:

• First, language in job ads could reinforce workforce segregation. For example, the study finds that job ads for workforces with a larger share of women tend to include more family-friendly cues and language signalling workplace EDI culture; in turn, such cues and language contribute to increasing the share of women in the workforce.
• Secondly, language in job ads has the potential to disrupt workforce segregation. For instance, the findings show that while job ads for workforces with a larger share of women tend to include more feminine rather than masculine psychological and work role cues, such cues are found to reduce the share of women in the workforce, thus tilting the gender composition of the workforce toward a more masculine direction.
• Finally, language in job ads may have little impact on workforce composition. The study shows that despite the efforts of workforces with a larger share of racial minority workers to include EDI policies and workplace EDI culture in job ads, EDI language does not seem to have any bearing on racial minority representation in the workforce.

These findings demonstrate both the benefits and limitations of intervening in the language used in job ads to help reduce labour force gender/racial segregation.

“They provide insights that are crucial to mitigating the impact of job ads on labour force gender and racial segregation,” said Professor Hu, “but they also show that ‘window-dressing’ EDI language in job ads is not sufficient in actually creating EDI in the labour market, at least when labour force gender and racial composition is concerned.

“Our study calls for a major rethink on how employers frame their job ads and coming up with meaningful ways of communicating and doing EDI to help reduce gender and racial segregation in the labour market.”

Professor Monideepa Tarafdar, of the University of Massachusetts, Amherst, USA, co-Principal Investigator of the ESRC–SSHRC project and a Visiting Professor at Lancaster University, added: “With the proliferation of large language models, AI-automated text processing tools are increasingly used to help draft and debias job ads. Our research provides a roadmap for building labour market equality into the design of these tools.”

Professor Karen Hughes, of the University of Alberta, added: “Our project, enabled by cross-national funding, also demonstrates the value of collaboration across multiple disciplines to tackle the grand challenges of our time.”

The collaborative research was undertaken by: Yang Hu, Department of Sociology, Lancaster University, UK; Nicole Denier, Department of Sociology, University of Alberta, Canada; Lei Ding, Enze Shi, Linglong Kong and Bei Jiang, Department of Mathematical and Statistical Sciences, University of Alberta, Canada; Monideepa Tarafdar, Isenberg School of Management, University of Massachusetts Amherst, USA and Department of Management Science, Lancaster University, UK; Alla Konnikov, Department of Social Sciences, Concordia University of Edmonton, Canada; Karen D. Hughes, Department of Sociology and Department of Strategy, Entrepreneurship, and Management, University of Alberta, Canada and Diana International Research Institute, Babson College, USA; Shenggang Hu, Department of Statistics, University of Warwick, UK; Bran Knowles, School of Computing and Communications, Lancaster University, UK; Jabir Alshehabi Al-Ani, Department of Computer Science and Data Science, York St. John University, UK; Irina Rets, Institute of Educational Technology, The Open University, UK; Dengdeng Yu, Department of Management Science and Statistics, Alvarez College of Business, University of Texas at San Antonio, USA; Hongsheng Dai, School of Mathematics, Statistics and Physics, Newcastle University, UK.