I posted this on LinkedIn a couple of years ago, but some recent posts made me think it was worth reposting on a platform that people actually read.

No, you don't need to hire more women

You can't solve any problem until you understand the problem that you're trying to solve and diversity and inclusion (D&I) is no different. I was Director of Studies for Computer Science at Murray Edwards (an all-women Cambridge college), have been Chair of the Microsoft Cambridge D&I Committee and sat on the D&I Council for Microsoft Research (worldwide), so this is a topic that I find myself discussing a lot.

A lot of the D&I-related conversations that I've had over the last few years have begun with someone telling me that their group needs to hire more women (or members of some other under-represented group - feel free to mentally substitute any other such group as you read this post) and asking me how to do it.

The number of women in an organisation is very rarely the underlying problem. It is a trailing indicator of an underlying problem, a spot health check, not an optimisation goal. If hiring more women is really the most important requirement, it's easy to solve: walk into any unemployment office and you'll find around half of the people there are women looking for jobs. Of course, most won't have the skills that you need (that, after all, is why you have a hiring process involving CVs, interviews, and so on) and hopefully that gives you a hint that just hiring people because they are women isn't actually the right solution.

It's very easy to set up metrics about number of women in each organisation and drive evaluation of culture based on that. This can often make inclusion worse for your company. Imagine being a woman in an all-hands meeting when someone in a leadership position puts up a graph of the number of women in the org and congratulates the leadership on the fact that it's going up. Your first thought will probably be something along the lines of 'was I hired just to meet some quota?' Your second (more worrying) thought may be 'do all of my co-workers think I was hired to meet some quota?' Now, you're immediately second-guessing your own competence and expecting other people to think you're underqualified.

So why should a company care about the number of women in a group? If just hiring more women doesn't solve the problem, that suggests that what we really want to do is hire and retain the most qualified people; if a particular group is underrepresented, that may be because your hiring and retention favours or disadvantages some people for reasons other than competence. If the best candidates are self-deselecting before you even get them to interview, that's a problem. If the best candidates are being filtered out because HR doesn't really understand the job, or because your hiring process magnifies implicit biases, that's a problem. If the best people are leaving because of your team culture, that's a problem.

When I've talked about D&I, I've often been approached by people afterwards saying that D&I is great, that helping disadvantaged people is nice, but that they need to focus first on business impact. This misses the point. Companies don't engage in D&I activities to be nice or to help people. Companies engage in D&I activities because hiring and retaining the best people has a greater business impact then hiring and retaining the best out of an arbitrary subset of the candidate pool. It's important to keep that in mind with diverse hiring: you are not doing diverse candidates a favour by hiring them, they are doing you a favour by allowing you to benefit from their skills and unique perspectives.

Various studies have shown that teams with diverse perspectives do better. It's easy to focus on a single dimension here but a team of male, rich, white, Eton-educated, Oxford PPE graduates will not get much benefit if they start hiring female, rich, white, Eton-educated, Oxford PPE graduates. Diversity of viewpoints comes from a large number of axes, including education, interests, gender, ethnicity, and so on. Optimising for a single dimension will not give you the desired results.

Even though the root problem for your company is not the number of women that you employ, that statistic is still an easy metric to give us a quick culture health check. In the last few years, the number of women graduating from computer science degrees in the UK has remained at around 20%, so at first glance you should expect an organisation that hires computer science graduates to be about 20% female.

That high-level stat doesn't tell the whole story though. As a middle-class white boy, there are a lot of conversations I never had. No one told me I shouldn't be interested in computers because they're a girl's thing. No one called me a race traitor for being interested in mathematics because it's not a white thing. No one told me 'boys can't code'. No one ignored me as a possible candidate for extra classes in a STEM subject because I was a boy. In my time at Murray Edwards, I heard stories like these from countless (female) STEM students about their time at school.

Any woman who even made it into the first year of an undergraduate computer science programme overcame far more obstacles than someone like me. By the age of 18, they've already shown a passion for the subject that let them push through these barriers. The fact that many will have left the field in spite of their aptitude is a separate problem that schools need to solve. As an employer, are you more interested in the candidates who care deeply about the subject, or the ones that coasted through looking for a well-paid job? If it's the former, then you should probably expect more than 20% of your candidate pool to be women. A lot of under-represented groups are far less under-represented in the top 10% of a field than in the field as a whole. That still doesn't mean that's the metric that you should optimise for, just a suggestion of where your ballpark culture health check should be.

So why is your group less than 20% female? It might be simply a small group. For a team of five people, assuming that 20% of the qualified candidate pool is female and that you hire at random from that pool, you have around a 33% chance of being an all-male team. If you're hiring for a particularly rare skill set, there's a good chance that this will be higher: you're relying on candidates being available on the job market at the same time that you're hiring. The same probabilities work with respect to the available candidate pool: if there are only three qualified candidates on the job market at any given time, there's a >50% chance that they'll all be male. Groups that can hire speculatively (bring in competent people as they become available, rather than needing to hire someone this month) have a big advantage here, by being able to hire the most competent people when they're available.

Does your hiring process favour a particular group? I'm not going to go into detail here because there's a staggeringly large amount of research on this topic. Whoever designs your company's hiring process needs to read a decent selection of this research and consciously design the process to minimise implicit bias. If no one has done this for your company then there's a very good chance that implicit bias is the dominant factor in hiring outcomes. This isn't limited to decisions made by humans. Amazon famously tried to use machine learning for hiring based on their current employee profiles and it learned that being male correlated strongly with being a good hire, so used that as the key metric.

Do your culture or your HR policies favour retention of a particular group? The biggest single improvement that you can make for retaining women is, somewhat counter-intuitively, to improve paternity leave. If you offer six months maternity leave and six weeks paternity leave, then a mother in your team will be four and a half months behind a father. Worse, every manager of a team will have a higher expectation that women on their team may disappear with short notice for longer than men. There are lots of other subtle ways that team culture can favour groups, such as promoting people who speak a lot in meetings and so on.

Gender breakdown isn't the only misleading metric. A lot of gender pay-gap reporting is nonsense because it shows that men and women of the same grade are paid the same, but doesn't account for promotional velocity or the relative expertise of people at a particular grade. If you're using any such metric then you need to be very careful that you treat it as a diagnostic indicator, not as an optimisation goal.

Having a particular group under-represented in your workforce is almost certainly a symptom of an underlying problem but if you try to treat the symptom without treating the cause then you will fail.