Blind sifting has been iterative in its adoption over the past decade, becoming increasingly common over the last couple of years. Its aim is to avoid any possibility of bias – conscious or unconscious – from being able to determine gender or ethnicity from a person’s name and is just one of many good practices that recruiters should be adopting in order to embed good practices within their organisation.
But does the recruitment industry need to take a closer look at the way language is being analyzed during the recruitment process, especially when it comes to influencing invites to interview?
For HR practitioners, the use of name blind system allows for the anonymisation of name and other irrelevant data so that managers are responsible for running recruitment processes fairly and without prejudice. It helps to facilitate independent panel screening and can even make your recruitment process accessible to disabled candidates. At a time where equal opportunities are important, it also means that businesses can guarantee minority applicants are satisfying minimum criteria for interviews and more accurately identify the sources that generate high quality minority applicants.
Blind recruitment might start with a person’s name and contact emails but can them extend to all personal details (address/contact info) or parts of a candidate’s work history e.g. line manager, employee number. When a narrative or covering letter are important scoring criteria points, a CV or other attachments can be hidden until the candidate is officially shortlisted etc. Such engagement allows a business to maintain virtual relationships with communities that would not normally consider an industry or their organisation.
Oleeo partnered with University College London to conduct the first large-scale statistical linguistic analysis of male and female CVs across multiple job sectors (financial services, information technology, management consulting and retail and buying). The results we found prove how small, but statistically significant patterns, such as resume length, readability and use of certain words can easily lead to gender identification and influence a recruiter looking to achieve gender balance subconsciously. Even with blind recruitment methods applied, the redundant encoding of gender in machine learning representations could still be an issue which could lead to disparate impact on the gender minority.
In a study of over 200,000 CVs from candidates around the world applying to leading global businesses, it was evident that women are more likely to use words such as volunteer, assistant, organise and social, while equivalently qualified men are more likely to use stronger sounding words such as engineer, analyst, investment and leadership. The result is that even if the CVs are anonymised and the candidate’s gender is removed, there’s a potential for unconscious bias to creep in to the recruitment process due to the language being used.
The analysis looked at the lexical, syntactic and semantic differences in the text that distinguish male and female CVs and identified the top 10 words for each sector and gender and found that 90% of the top 10 words men used in male CVs are powerful proper nouns and nouns. In contrast, just 68% of the top 10 terms on female CVs are such words. The top 10 terms are:
Financial Services
Male: equity, portfolio, investment, capital, analyst, finance, market, stock, interests, technical
Female: organise, event, volunteer, assistant, social, student, marketing, community, department, plan
Information Technology
Male: php, c, software, Linux, c++, computer, have, developer, engineer, network
Female: volunteer, event, assistant, organise, analyse, plan, student, social, conduct, excel
Management Consulting
Male: engineering, sport, investment, finance, analyst, club, cost, financial, technology, technical
Female: volunteer, assistant, event, social, organise, write, community, student, communication, research
Retail and Buying
Male: football, play, sport, business, club, technology, computer, mobile, it, leadership
Female: art, child, volunteer, shop, assistant, assist, social, design, organise, create
The study also identified slight differences in CV length, readability and use of certain words, which can also easily lead to gender identification. Analysis of the average number of words and unique words used by male and female candidates across all sectors showed that female candidate CVs tend to be longer and use a greater variety of words.
Academics have concluded that the research suggests the differences are so significant that blind recruiting alone is not enough to reduce gender predictability. Employers need to use intelligent algorithms which have learnt to disregard proper nouns and other gender identifiable language. Only then will recruiting really become truly fair and transparent.
So can this be mitigated? Intelligent selection techniques can ease this burden by providing gender de-biasing methodology which can be applied to successfully remove gender redundant encoding and lower disparate impact scores applied during machine based hiring predictions. This makes it suitable for use in automated hiring screening processes where not causing disparate impact is paramount.
Technology works because algorithms can replicate your collective decision making, reducing the influence of bias by individuals or process. It’s not just Oleeo saying this. The Confederation of British Industry has described “name-blind” recruitment was one way to remove “criteria that could unintentionally bias managers, and give under-represented groups confidence that their application will be fairly considered”.
Fair consideration is the first step to true inclusion. How a company then promotes its values to demonstrate this is the next step. Once applications are blind sifted, the recruiter must consider if they can fairly run interviews alone and continue the momentum. Panel interviewing may help avoid doubt and showcase a more transparent commitment to equal opportunities. The power really does lie in the recruiter’s hands to see these values through. Are you ready?
