
Why statistical gender differences matter in recruitment
In the aftermath of the “Google Memo” Sam Bowman wrote an interesting piece on understanding statistical averages. However I wanted to elaborate on why this matters a lot when hiring people, especially in the talent focused tech sector.
Looking at Sam’s graph, it’s easy to see that a difference of a 0.5d in some random attribute doesn't really make us very different, ~most people are the same. However if you try to hire the best of the best, the effect becomes more and more important. If you aim to hire at the ~0.25d mark in talent, you would have a 50/50 distribution of men and women. However when you move higher in the talent tree, the distribution quickly widens. If you’re looking to hire above 1d (~top 15% of the candidates) you’re looking at a 65/35 distribution. If you’re trying (like many high tech companies are) to hire above 2d (~top 2% o the candidates), you’re already looking above 80/20! I.e. when you go to the extremes of the distribution, even small differences have big impact without bias.
I think the key thing for companies to do is implement blind hiring processes, i.e. people picking candidates for first round interviews should not see information about gender, race, ethnic background, etc. of candidates. This will protect us all from basic biases and allow companies to truthfully evaluate (“does ratio of hired people match the interviews”) if their hiring processes are equal or not.

