How to remove gender bias from an academic job search?

Question

I overheard some professors discussing the next round of hiring at my university and several were concerned about gender bias possibly playing an issue in the hiring process. An initial thought is to just go through CV's and black out an applicant's name. However, an academic job search makes it much more difficult to do this. Not only do applicants have CV's, but there are also typically 3 letters of recommendations as well as publication lists (which might reveal the identity of an individual if the paper is known by some of the hiring panel).

Some suggestions that were thrown out involve trying to find a way to scan through 2-300 applicant materials and black out/replace gender pronouns and names. From browsing StackOverflow, there is apparently quite a bit of difficulty with this from a programming perspective. Still, it seems the most efficient way to remove as much gender bias as possible in the process but doesn't seem to be widely used.

What are some of the best ways a hiring panel can remove gender bias from the application process?

This question could also potentially extend to ways to generally remove other forms of bias, such as ethnic bias.

Answer 1

Bias exists at many points of the hiring process. You suggested blinding the search committee to applicant gender but, as you point out, this is extremely difficult to do perfectly and completely broken by even small failures. For obvious reasons, blinding will also not be particularly relevant after you start interviewing candidates. I like the other suggestions to provide training to sensitize the committee to issues of gender bias.

Beyond that — and if your university policies allow it — you might also decide now (i.e., before the search) to interview at least one male and at least one female candidate. This way, you will give the best male and female candidates a full chance to convince you that they are right for your department. This ensures that at the top person of each gender makes it through the earlier stages of the process where gender bias may very well play its biggest role. At the interview stage, blinding would not have worked anyway.

This kind of policy is unusual but not unheard of. The most famous example I know if is the Rooney Rule in the US National Football League which requires that all teams interview minority candidates for head coaching and senior football operation jobs. Although this is sometimes cited as an example of affirmative action, it does not mandate any preference or quota to candidates within the pool of those being interviewed. If you're doing it right, it does mean that the very best candidates from under-represented groups will always have an opportunity to show their stuff at the final round.

If you found out that best person from the under-represented groups is really not as good as the best person from the over-represented group, at least you'll know that you gave the best member from each group a full hearing.

Update: I will point out that this answer basically assumes that all of your candidates will present as either male or female. As a result, is it very limited in the case of non-gender conforming candidates. These candidates may also be subject to even greater discrimination and this approach will not solve (and could even aggravate) those problems.

Answer 2

I think there are two approaches. The approach your question focuses on is blinding of the panel to the gender of the applicant. Doing this, may actually increase gender bias. By blinding the search panel to the gender of the applicant, it becomes very difficult for things like maternity leave to be taken into consideration. The better way to remove, or limit gender bias, is to provide training to the search panel about gender bias in academia and help them become aware of any biases they might have.

Answer 3

One simple and extremely effective step is to start tracking metrics on the candidate pool at every stage of the process. Let's say you're looking at how your department hires assistant professors. Then you might track:

1. What percentage of Ph.D. holders in the field are women?
2. What percentage of the applications you receive are from women?
3. What percentage of the short-listed candidates are women?
4. What percentage of the interviewed candidates are women?
5. What percentage of the offers made are to women?
6. What percentage of the accepted offers are taken by women?
7. What percentage of the professors who advance toward tenure are women?

Now you've got actual data on what your pipeline looks like and can look for where the leaks are. If the fraction of females in the pool changes significantly at any particular stage, then that's where to focus your energy. Likewise, if the base fraction in the field is lower than you want in your institution, you can use your metrics to decide where to try to enrich the pool with good candidates. Obviously, the same approach can be applied for other disadvantaged groups as well.

I personally think this type of approach is a critical addition to the toolbox of addressing bias, because it lets you scientifically study your institution's process. You may discover things that surprise you. For example, the colleagues who I learned about this from discovered that the later stages of the hiring pipeline they were dealing with were actually OK, but that the percentage of women applying in the first place was much lower than the percentage of women in the field. That meant (to everybody's surprise) that the problem was primarily in the way that positions were being advertised and recruited for, rather than in the interviews themselves, and so that was the process that fixes were targeted at.

Answer 4

As a former programmer, I can confirm that it would be quite difficult to do this purely in a programmatic way. Take my name, for example. From my name, you can probably guess that I am male. If you had a computer program looking at my resume, how would you remove the word "Evan"? Keep in mind that it's easy for a human to know that that's my name, but difficult for a computer to know. Computers are very fast, but also dumb - they'll do exactly what you tell them to do, no more, and no less. Sometimes this is problematical with programming when whet you're telling them isn't what you think you're telling them, but that's another discussion entirely.

If you want a truly blind resume / cover letter / letter of recommendation review, you'd need either a manual approach needing humans who aren't involved with the hiring committee and won't report what they saw to black out the names and the pronouns used in all of these documents. Ideally, they'd be retyped with gender neutral terms such as "that person" instead of "he / she", so you won't have someone trying to squint under a blacked out ink trying to read it. People are curious - it'll happen if you let them.

There's also a hybrid approach: use computers do to the easy part, and people do the hard part. The easy part would be getting rid of all the he / she / him / her words in documents and replacing them with a gender neutral word / phrase of your choosing. The hard part would be doing the same thing with names.

The reason this is hard for computer is simple: how does the computer program doing the parsing know that what it's seeing is a name? It doesn't, unless you have some way of telling it. In an ideal world, word documents or PDFs would have metadata with a field clearly identifying that this is a name, but I doubt such a feature exists - at least in Word. PDFs probably do support this, but again, whether or not Adobe Acrobat supports this isn't the issue, but rather the issue is whether or not the appropriate metadata is embedded in the document.

Answer 5

A way to implement gender neutral CVs for the first round of selection would be to ask applicants to give a preliminary gender neutral file along their full application. This short CV would be asked to contain only information that does not permit to guess the gender of the applicants, at least not easily (e.g. publication lists with names replaced by the number of author and the position of the applications, etc.)

This can only be used to a very first round of selection, and letters of recommandation would have to be only used in the subsequent round, or the recommandants should be asked to make them gender neutral and without the name of the applicant.