Bias On LinkedIn - An Experiment
Many would consider a photo on a CV to be outdated, leaving the reviewer unnecessarily vulnerable to bias. Some organisations take this further by ‘blinding’ applications, removing all demographic information before shortlisting. There’s good reason to believe these steps will lead to better, fairer recruitment - which is also good reason to be cautious about using LinkedIn as a hiring tool.
“LinkedIn’s profile design leaves the platform vulnerable to biased hiring decisions. Most of the organisations using it have not explored debiasing strategies to mitigate that risk.”
This perspective should not be controversial. ‘Audit studies’ on CV screening have revealed stark racial, gender and social class discrimination through information as seemingly harmless as a name, picture and a few words in an ‘About’ section. All of which are given particular prominence on a LinkedIn profile. Our project was partly inspired by a large-scale field experiment with 900,000 LinkedIn candidates where Black, Hispanic, and Asian applicants were 8-13% less likely to receive a call-back compared to White applicants.
The Experiment
To illustrate how managers may inadvertently use the information given prominence on LinkedIn’s profiles, we designed a hypothetical hiring task with 800 managers. Each would shortlist four candidates from a list of eight and we were interested in how gender and race influenced their decisions. However, to build on the field research above and overcome the issue of ‘social desirability bias’ in lab experiments (where underrepresented candidates are selected at higher rates than in real-world studies), we introduced a third variable of how candidates presented themselves in the ‘About’ section.
This meant our attention wasn’t on the number of women vs. men or Black vs. White candidates shortlisted, but on how these characteristics influenced the selection of an agentic candidate (someone presenting as self-confident and ambitious in the ‘About’ section) vs. a communal candidate (someone presenting as relationship-focused and prosocial). We predicted that gender and racial stereotypes relating to these traits would influence managers’ decision to shortlist candidates for interview, but not necessarily in ways you would expect.
The Design
Participants viewed eight applicants for a project manager job for a fictional company, and were asked to select four to progress to the interview stage. Our candidates were designed to replicate a LinkedIn profile as below:
The eight-person list included an intersectional mix of four women and four men, four Black candidates and four White candidates, and four ‘agentic’ candidates and four ‘communal’ candidates. Lilian, above, is an example of an agentic profile and Emily below is an example of a communal profile:
To isolate the impact of race, gender and presentation style from factors such as education and experience, we used a technique called counterbalancing. This means that for some managers in our experiment, Emily became the agentic candidate working at ‘First Bank’ and Lilian became the communal candidate working at ‘Telecommunications Global’. As a result, any difference in managers’ selection at the group level could be isolated to our variables of interest.
This technique was used across all eight candidates. We also counterbalanced the candidate’s name and style of the photo to account for differences caused by these details.
Our Key Findings
FINDING 1: Agentic candidates were significantly favoured over communal candidates in all demographic groups
Agentic candidates were significantly favoured over communal candidates (57.43% vs 42.57%; p=<.01). Despite being evenly represented in the eight-person list, 400 more agentic candidates (+14.86%) made it onto the interview list. The agentic advantage - which could also be reframed as a communal penalty - was present across all demographic groups.
The Agentic Advantage
FINDING 2: The agentic advantage / communal penalty was significantly affected by gender and race
The strength of the Agentic Advantage changed depending on the candidate’s gender and race. The preference for assertive, confident profiles was strongest among White Women (+20.7%) and Black Men (15.6%), and weakest for White Men (11.2%) and Black Women (10.6%). This could be reframed by saying that communal, relationship-focused profiles were comparatively penalised among White Women and Black Men and preferred among Black Women and White Men.
The Agentic Advantage within Demographic Groups
Implications and Actions
The Agentic Advantage, independent of race and gender.
There is nothing discriminatory about the preference for agentic candidates per se, but it seems at odds with the values of collaboration and purposeful work so heavily promoted by corporate organisations. This agentic preference may also lead to poor hiring decisions: Communal behaviours are increasingly identified as predictive of higher performance, particularly in leadership positions during times of change and crisis. Confronting the value of this general preference is our clearest next step: how many organisations have an unknown bias towards agentic candidates that is in direct odds with their hiring strategy and that may inadvertently favour men (who are perceived to exhibit more agentic behaviour as a group) over women (who tend to exhibit more communal behaviour as a group)?
“How many organisations have an unknown bias towards agentic candidates that is in direct odds with their hiring strategy, and which inadvertently favours men over women?”
The importance of intersectional research.
We predicted that candidates who presented themselves in counter-stereotypical ways - specifically women as agentic and men as communal - would be favoured by hiring managers. To put it bluntly, we thought there would be an assumption that female candidates display communal behaviour by default, so an agentic presentation would be more desirable in a LinkedIn profile.
We were right - but only for White candidates - and this was humbling for us as a research team. In line with prior research on stereotypes, we expected an interaction with race, but not so significantly. The Agentic Advantage appeared strongest for White Women, then Black Men, then White Men and finally Black Women. It's a stark illustration of the complexity and significance of challenges brought about by intersectionality.
The need for action
While organisations may not have control over the design of the LinkedIn profile, there are many ways they could mitigate bias through the way they present those profiles to their hiring managers. These need to be tested and shared, and that is only possible with more research.
“LinkedIn has a profound influence on our ability to create diverse workplaces, and organisations invest £billions in the platform as a hiring tool. How much attention is paid to ensuring the decisions we make contribute to a fair, meritocratic and high-performing workforce?”
We finish, as always, with a call for organisations to ask themselves hard questions. Do you want to use LinkedIn to make better, fairer decisions? Are you willing to contribute to the research at the same time? We are here to help.