How Do I Remove Bias From My Recruitment Process?

Published by:
Andy Babbage
November 29, 2018
 min read

As humans we are all susceptible to unconscious bias.

In an attempt to fill in gaps of missing information or to make a snap fight or flight judgement, the brain ‘helps’ us by using a whole array of mental shortcuts to make fast decisions (check out our infographic for 20 of the big ones in recruitment).

This can be really useful if you are in survival mode but is less useful for considered, deliberate decisions, such as recruitment.

Unconscious bias is detrimental to recruitment processes, as it not only adds noise but also tends to work against the recruitment of minorities and underrepresented groups into your organisation (if you haven't already, read this first: Is my recruitment process biased?). The good news is that bias can be controlled for in many parts of the process. Doing so will improve the quality of hiring decision and naturally promote diverse recruitment - here’s how:

Unconscious bias training doesn’t work

Most people’s first stop when arming themselves and their teams against bias, is unconscious bias training (it's estimated that $8bn is spent annually on this type of training). The assumption is that by understanding the unconscious processes that are going on in our brains, we will be able to detect when it is happening and consciously stop ourselves from doing the problematic behaviour.

Unfortunately it just doesn’t work like that. These biases are unconscious so are incredibly hard to stop. If you’ve just recognised a situation where a bias could pop up it probably has already, and the noise that comes along with that bias has already entered into your view of the situation or decision.

Rewiring the brain is hard, and in fact the most recent research on the effectiveness of unconscious bias is mixed at best.

As the study below shows, training can somewhat reduce bias... but only for around 8 weeks.
Does training reduce bias? (chart)

And some studies indicate that it can even backfire.

If we look at the chart below, we can see that training has the potential to actually reduce the representation of some minority groups.

Effect of training on management diversity (chart)

The key is not to try and change humans - if this is possible at all, it would take more time and money than you've probably got to spare.

The way to tackle bias is to design systems and processes that prevent bias from creeping in in the first place.

What sounds like a more effective dieting tactic - buying ice cream and attempting to resist eating it or simply removing all of the ice cream from your house?

Removing bias from recruitment through process design

In Professor Iris Bohnet’s book ‘What works: Gender equality by design’ she brings this very view that debiasing people’s minds has proven to be difficult and expensive.

She instead suggests that behavioural design is the solution...

It's far more effective to debias a process than it is to debias humans themselves.

Bias isn't a boogeyman to be defeated. It's an unavoidable part of being human that we have to design around to ensure outcomes are fair and objective.

Removing bias from recruitment doesn't have to blow your budget.

There are proven, evidence-backed methods you can use to remove bias and make your hiring process more ethical and data-driven.

1. Anonymize names of candidates

Anonymous recruitment is one of the best known methods of removing bias.

It's been proven to be a highly effective at mitigating bias at the screening stage and is a critical component of a fair and diversity-optimized hiring process.

The most basic version of anonymization involves manually removing someone’s name and photo from their resume.

This means that you are forced to review their experience and education without being subject to the types of bias that someone’s name and appearance can trigger (affinity bias, stereotype bias etc.).

As the study below shows, just these two pieces of information can have a dramatic effect on someone's chances of receiving a callback.

Callback rates by name and photo (chart)

So, by simply anonymizing this information, you can see how outcomes may begin to improve.

However, this is just the tip of the iceberg...

2. Consider removing education and experience too

A genuinely debiased process will require you to remove more than just names and photos.

This includes address, hobbies, name of university and names of previous companies.

Wait, aren't university names important?

Well, whilst this could be an indicator of intelligence, it could also be an indicator of what socio-economic background someone is from.

In most cases, it adds more noise than actual useful information.

We know that here in the UK, around 80% of Oxbridge offers go to students from the top two socio-economic groups.

And whilst there will always be outliers, it's safe to assume that someone's socio-economic circumstances will play a significant role in where they go to university - or if they attend university at all.

We also know that those who attend top universities and private schools tend to bag the top jobs ands therefore the most impressive looking experience.

So, if we want to improve diversity, it's fairer to test for skills learned through education and experience rather than for education and experience themselves.

Education and experience are proxies for job performance - which is what we ultimately want to assess.

And when we look at the research - we can see that education and experience aren't very predictive of future job performance.
Predictive validity of assessment methods (chart)

3. Test for skills upfront using work samples

So, if we're not using flawed signifiers like education and experience to make assumptions about candidates' ability, then how should we be assessing them?

If we look back at the predictive validity chart, 'work samples' are ranked as the most predictive assessment method.

Here at Applied, we use 3-5 work sample questions to anonymously screen candidates, instead of a CV/cover letter.

Work samples are designed to simulate a role by asking candidates to perform small parts of it.

Instead of asking someone how they handled a given task in the past, you simply ask them to either perform that task, or explain their approach to doing so.

Work sample example

By posing questions hypothetically, you can hone in on someone's potential, rather than experience.

True, work experience may well equip someone with the skills and knowledge to give the best possible answer...

But since we're focussed on removing bias from recruitment, we'd rather test for these directly.

Why? Because we want the best possible person for the job, regardless of how or where they acquired their skills.

How to create your own work samples:

  • Identify skills: what are the 6-8 core skills needed for this job?
  • Identify tasks: what tasks will this person be doing in the role?
  • Create rubric: what does a good answer look like? (we'll go into more detail on this later)

4. 'Chunk' applications to compare like-for-like

Rather than reviewing a candidate’s application in full, this method involves ‘chunking’ up the application into different sections, and then comparing each section for all candidates next to each other.

By cutting up applications into sections, you are stopping your brain from being affected by the halo effect that could come from the association of a top tier university or the confirmation bias of a really great start to an application.

Halo effect explained

Also, by comparing each section immediately between all candidates, you are better able to develop a fairer internal mental model of what poor, average and great applications look like. Your brain will have no choice but to judge that section of that application on its merits alone, not with the baggage or noise of any other aspects of that candidate’s full application.

Here's what this looks like in practice:

Applied screening process

5. Randomize the order you review answers to minimize ordering effects

The brain is a funny old thing - it will do some really clever things to ‘help’ you out with the processing of information, but it is also predictably fallible in several other situations.

This is why we recommend randomizing the order in which the sections of candidate applications are reviewed each time.

This is not only because your brain will start to guess and develop patterns as you review sections of applications and try and stitch them together based on order, but also because there are a whole host of order effects that can cause our brains to act weirdly.

The first one is the ‘hangry’ effect, where research has shown that as we get tired and hungry and our blood sugar drops, we get harsher with our assessments.

Just take a look at this study of Israeli judges.

They made harsher decisions as the day went on - the spikes we can see in favourable decisions were following the judges’ breaks.

Hangry effect study (chart)

This is usually great news for Andy Aardvark and bad news for Zippee Zugman.

Poor Zippee will usually be reviewed last, so will have been copping the brunt of this down-marking for most of their lives.

Another one is the Nobel Laureate effect, where if we assess an exceptional candidate, we all then mark down the following three, no matter what they say or do.

By randomizing the order of the reviews, you are at least averaging out these effects so the same person is not affected each time.

6. Score candidates using a rubric

Since human judgment can be easily warped and our ability to accurately recall events is less than perfect, it's essential to give yourself a scoring rubric.

For each question you ask candidates, have a simple 1-5 star scale to score them against.

Data-proofing your judgment means that at the end of the process, you can make decisions based on your original criteria, and not 'gut feeling'.

Example review guide

7. Use crowd wisdom for more accurate scores

Left to an individual, the odds of getting the right person for a role usually comes down to a coin flip. This is because as individuals we are subject to a whole host of context that makes us as subjective as heck.

Plus, as individuals we also tend to love to recruit in our image, after all that’s what we know best - who wouldn’t want another me in the team after all?

The solution is simple, get more than one person to review each candidate and harness the wisdom of the crowd to average out people’s subjectivity.

Our research shows that 3 reviewers is ideal, with the marginal gains rapidly dropping off as you add more reviewers after this.

Getting more people involved in recruitment is also a great way to distribute load, get alternative perspectives and get buy-in from stakeholders early.

Below you can see how we use review panels here at Applied.

Example review panel structure

8. Remove unconscious interview bias by adding structure

When you talk about removing bias from recruitment, most people start to object when it comes to interviews.

After all, you are sitting in front of the candidate, how can you stop interview bias from creeping in?

Short of getting them to speak behind a sheet as the Boston Symphony orchestra famously did to increase the representation of female musicians, we will all be swayed by what we see and interpret from the spoken and unspoken interaction we have.

This is true, but it is not a reason to throw your arms up in the air and give up.

Interviews can still be vastly improved from what they are in many organizations (a fireside chat).

When you make your interviews structured, they become one of the most predictive forms of assessment.

Structured interviews are exactly as they sound; you set out the questions that will be asked and also how each answer will be assessed. This prevents the interview from going down a rabbit-hole of gut, rapport or fit, all of which can be coveralls for ‘they just weren’t like me’.

Whilst screening work samples have to be writing-based due to their anonymous nature, you can use your interviews to simulate tasks that can't be testes via writing.

Instead of asking candidates to tell you about at time when they did something, you can actually role-play the task.

Take a look at how we do this Applied...

Turning interview questions into work smulations

This sounds like a lot of work, why should I bother?

We hear you - when we started doing this many years ago we thought the same thing.

But it’s worth it - by following these methods you will not only hire the best people, but you’ll also expand your talent pool and improve diversity.

Organizations using these practices tend to see:

  • Up to 4x attraction and selection of ethnically diverse candidates
  • 3x as many suitable candidates
  • 93% retention rate after one year
  • 9/10 average candidate experience rating (including unsuccessful applicants)

It’s not all hard work though, the Applied Platform speeds up and automates all of the practices above.

We’re all about empowering HR and Talent functions to use data and the latest research to hire amazing talent that add to your company’s culture.

The research is clear - traditional hiring is antiquated and biased. Talent is lost and candidates from underrepresented backgrounds are disproportionately overlooked.

Make no mistake - good intentions alone won't change these outcomes.

The only way to tangibly improve diversity and make a genuine impact is to tweak the hiring process itself.

Applied is the essential platform for debiased hiring. Purpose-built to make hiring empirical and ethical, our platform uses anonymized applications and skill-based assessments to identify talent that would otherwise have been overlooked.

Push back against conventional hiring wisdom with a smarter solution: book in a demo