Advertisements for jobs that include subtle ageist stereotypes related to physical ability, communications skills or technological aptitude are putting off older workers from applying. That is the central finding of new research by Ian Burn, Daniel Firoozi, Daniel Ladd and David Neumark, which uses new machine learning language processing tools and an experiment on Amazon’s MTURK platform to test whether people perceive certain phrases in job ads as being ageist.

The study finds strong evidence that they do. For example, when shown a job ad phrase about technological skills – ’You must use software such as Microsoft Office/Excel or Google Sheets’, as opposed to an innocuous job requirement – respondents perceive the phrase as biased against older applicants. Evidence from a related study shows that such language in actual job ads predicts age discrimination by employers.

The researchers conclude that agencies charged with reducing discrimination in the labour market could use machine learning language processing tools to study the text of job ads and flag employers that may be discriminating. They could also issue guidance to employers to avoid language that might discourage older workers from applying, in addition to guidance they already offer employers to avoid explicitly discriminatory phrases, such as an age range for job applicants.

Given ageing populations, lengthening work lives into older ages is a policy imperative. But age discrimination – especially in hiring – is a barrier to achieving longer work lives. This study shows that ageist phrases that employers use in job ads can discourage older workers from applying for jobs. This is an indirect form of age discrimination, compared to simply not hiring a qualified older applicants. But it can be just as harmful to efforts to increase work at older ages.

Age discrimination in hiring is particular important to these efforts for two reasons:

• First, many seniors try to make a transition to part-time or less-demanding jobs towards the end of their careers, but if they can’t get these jobs, they may instead stop working.
• Second, even when there are tough age discrimination laws (as in the United States), stopping discrimination in hiring poses unique challenges, made worse by recent Supreme Court decisions that weaken US age discrimination laws.

And there is compelling evidence of age discrimination in hiring in the United States and other countries. For example, a large US field experiment involving over 40,000 artificial job applications to about 14,000 job postings, found that rates of call-backs for interviews for sales jobs were about 25% lower for men aged 65 versus 30, and about 40% lower for women aged 65 versus 30 (see Neumark et al, 2019).

Because of stiff penalties, companies have an incentive to try to avoid their age discrimination being detected. One way to do this is to write job ads that discourage older workers from applying in the first place.

This study explores what happens when job ad language uses ageist stereotypes. The authors use complex new machine learning language processing tools, which treat job ad text as data, to generate job ad language that varies in whether it includes subtle ageist stereotypes related to physical ability, communications skills or technological aptitude.

They then run an experiment (on Amazon’s MTURK platform) to test whether people in fact perceive these phrases as ageist. There is strong evidence that they do.

For example, when shown a job ad phrase regarding technological skills – ’You must use software such as Microsoft Office/Excel or Google Sheets’, as opposed to an innocuous job requirement – respondents perceived the phrase as biased against older job applicants. The perceived bias was about 25% larger on the scale that the researchers use.

On its own, this evidence does not directly address the actual behaviour or intent of employers that might use these ageist stereotypes in job ads. But in a closely related field experiment, the researchers find that the presence of these stereotypes in real job advertisements predicts age discrimination by employers (see Burn et al, forthcoming).

An implication of this evidence is that new machine learning tools for processing language can be used to identify real-world age-stereotyped language in job ads. Policy-makers and agencies vested with reducing discrimination – in the United States, the Equal Employment Opportunity Commission – could use these tools to study the text of job ads to flag employers that may be discriminating, and investigate further.

They could also issue guidance to employers to avoid language that might discourage older workers from applying, on top of the guidance they already offer employers to avoid explicitly discriminatory phrases like an age range for job applicants.

---

‘Machine learning and perceived age stereotypes in job ads: evidence from an experiment’

Authors:

Ian Burn (University of Liverpool)

Daniel Firoozi (University of California Irvine)

Daniel Ladd (University of California Irvine)

David Neumark (University of California Irvine)

dneumark@uci.edu

---

References

Burn, I, P Button, LF Munguia Corella and D Neumark (forthcoming) ‘Does ageist language in job ads predict age discrimination in hiring?’, Journal of Labor Economics.

Neumark, D, I Burn and P Button (2019) ‘Is it harder for older workers to find jobs? New and improved evidence from a field experiment’, Journal of Political Economy 127: 922-70.