Anonymization: A Step Towards Reducing Bias in Hiring?

In the last five years of running a recruiting software company, I’ve encountered a difficult and intellectually challenging problem that many people face: How to reduce bias in the hiring process. There are many forms of bias that occur and at a high-level, the major buckets are conscious and unconscious bias. The former is typically the more recognizable and more reprehensible. However, I’d argue that the latter might actually be more damaging. There’s no real way to quantify that but I feel that so many of us carry around biases about different types of people that the collective impact of the things we aren’t aware of in ourselves might be larger than the more overt feelings of “I don’t want to hire him or her because he or she is X…”

Tackling unconscious bias in hiring is really hard. In large part this is because a lot of what people have been taught to do in hiring is to pattern match. If I see X and Y and Z (or don’t see X or Y or Z), that’s a good candidate. This is natural and not always bad. After all, part of making good decisions in life is strengthening one’s ability to properly pattern match. And trying to resist any pattern matching is most likely futile.

The thing to be wary of though are false patterns. An example of this can be found in Malcolm Gladwell’s book Blink where he covers the topic of blind auditions. You can read more detail on it here but the summary version is that when orchestras started putting up screens in audition rooms prior to the 1980s, more women ended up being hired for orchestras. In effect, the faulty pattern match that was happening when visual stimuli (e.g., whether someone was male or female) was broken. The lack of the irrelevant gender data made the judges focus on what was actually relevant, the performance itself.

I believe there’s an opportunity for something similar to happen in the corporate world today. It would be the equivalent, at least the closest we can get, to the blind auditions described in Blink. And it’s actually a fairly simple (if not easy) thing to do.

The idea would be to keep as many irrelevant details of a job applicant “anonymized” as long as possible in the hiring process. As soon as details are introduced that allow faulty pattern matches to occur, bias is allowed to creep in (be it conscious or unconscious). The longer that those details can be held back, the larger the effect on minimizing bias.

The obvious starting point is name. A study that I’ve referenced often in conversations with media is the ground-breaking study that was conducted 15 years ago where two professors sent out 5,000 “fictional” resumes for various job ads they found in newspapers. In this study, they compared the callback rate for the applicants to measure a key variable: Would “white sounding” names get more callbacks than “black sounding” names? They kept the credentials the same on different resumes and only changed the names (e.g., from a white-sounding name like Emily Walsh or Brendan Baker to an African-American-sounding name like Lakisha Washington or Jamal Jones).

The conclusion: Applicants with white-sounding names are 50 percent more likely to get called for an initial interview than applicants with African-American-sounding names.

If that doesn’t make you cringe a little, put yourself through a simple thought exercise. Picture the person you love most (e.g., your spouse, child, etc.) and imagine them applying for a job one day and not receiving a callback simply because of the race, gender or ethnic implications of their name. I have a three year old daughter and the notion that one day someone might see her name and (again, consciously or unconsciously) say “You know, we really need a man for this job…” pisses me off.

So the answer here could be to work to keep someone’s name out of the hiring process as long as possible. This is difficult but not impossible with modern technology. At some point the recruiter or hiring manager needs to know. But if that moment can be delayed, even if simply through the initial screening process where so many people are weeded out via faulty pattern matches, that could be significant.

There are other forms of anonymization that should be considered as well. Education could be looked at. If you are hiring a recent college graduate, the educational institution they attended might be a relevant data point in your process. But what if someone graduated from college 10 years ago? This is another potential data point that could produce a faulty pattern match and inject bias into the process.

Where someone lives is another great example. How many recruiters have looked at a resume, seen someone with an address that implies that they live in a lower-income area and let that implication bias their decision? Few would admit to it but it happens all the time.

However, there’s a potential flip side to anonymization. It’s possible that anonymization could have the opposite effect of what’s intended. For companies that are working to emphasize diversity in their hiring (more and more these days thankfully!), anonymization removes data from the hiring process that could be helpful. For example, if a company is looking to increase its number of female engineers, not seeing the names on the resumes could prevent recruiters from being able to give talented female engineers a shot at making it through the interview process.

There’s a study I came across in my research out of France where it was found that anonymizing resumes actually led to fewer minority candidates being hired.

Anonymizing resumes widened the interview gap between non-minority and minority candidates by 10.7 percentage points, from 2.4 percentage points in the standard procedure to 13 percentage points in the anonymized procedure. At the hiring stage, the recruitment gap widened by 3.7 percentage points.

This is the potential dark side of anonymization. It helps prevent people from doing the wrong thing but at the same time could also help prevent people from the right thing.

So what is the answer? I’m not fully sure. I do believe some anonymization efforts hold great promise, especially in situations where there has been a lot of historical bias against candidates on the basis of race or gender. However, I think there are important factors to consider in the equation. This post is only scratching the surface of what is a very complex and nuanced topic. I plan to revisit the topic again and I’m looking forward to having a dialogue with others who are passionate about this topic and see the impact for positive change.