Not necessarily. If the pool of talent is exactly the same(let's say 1000 women and 1000 men) and your interviewing process is perfect at determining engineer quality, but you are hiring 80 women and 20 men, then you'll hire the top 2% of men and the top 8% of women. Again, just assuming that the talent is exactly the same, the women you've hired will still be worse.

In reality, the pool of female talent is very shallow. Again assuming the talent levels are exactly the same, there's just not enough highly-skilled women to go around, e.g. if you're trying to hire the top 1% of engineers, and you have a pool of 1 million men and 100,000 women, then you're looking at 10000 men and 1000 women as your potential hires. If the top 1000 women are already taken, then you're forced to relax your interview requirements for women. In fact, if the pool is completely free for your usage, but you're looking to hire 10000 men and 10000 women, you'll be hiring the top 1% of male talent and down to the top 10% of female talent, which should lead to a difference in output.

This doesn't mean that "sexism" isn't real or that it's not a problem women face in the workplace. The question however, is how to measure the effectiveness of "anti-sexism" practices employed, if you cannot measure the effect of "sexism" itself, as it is entangled with all the other, confounding factors. You can't optimize performance if you can't measure the effects your solutions have on it.

Not at all. If the skill requirement is lower for women to get into the same position as men, is it not rational that, on average, those women will perform poorer than the men?

Imagine they did the same thing for university entrance: men need all prerequisite courses and a 3.8gpa, whereas women need only 75% of prereqs and a 2.9gpa. It only makes sense that the women are going to score lower in exams/flunk out, since their initial starting point was, on average, further behind.

That wasn't said.

There are certainly good and bad developers no matter what gender.

If you have stricter hiring policies for men and e.g. only the top 10% of male candidates land a job while the less strict hiring procedures for females get the top 40% females a job, then you created an imbalance in skills between the genders in your company (and only there), and this is what might have happened here

No, there's no biased assumption that women are worse developers, just one about relative numbers and relative skill distribution.

If the number of qualified female developers is less than the number of qualified male developers, then any company which manages to achieve hiring parity has either lowered the qualifications it demands from women developers, or increased the incentives it offers to them relative to other firms (enabling it to capture a higher share of high-skill female developers). Since the latter approach is obviously discriminatory and quite possibly illegal (not to mention a blatant violation of the principle of equal pay for equal work), the former is more likely.

The result of the former is that the median female developer at that company will have lower skill than the median male developer at the same company, even if women in the population at large have the same or better development skill.

The incentive strategy could enable a firm to secure the best female talent, but besides the ethical and legal implications it would also make things much worse for all the other firms out there: the best female talent being off the market, those firms would be forced to either settle for lesser-skilled women or accept a disproportionate share of male developers, neither of which would be appealing.

Naturally, increasing the supply of women developers to match that of male developers (or reducing the supply of male developers …) would improve the situation, but that's a horse of a different colour.

It's more like... if 1 good female developer gets a CS degree for every 10 good male developers, you would expect the workforce to be 10% female. However, if you artificially force 50% female representation you'll have to start drawing upon bad female developers since the rate of good developers is currently only 10%. You first have to increase the output of good female developers from schools before you can scale up their representation in the workforce, not the other way around.

You can still lower the bar below "hiring straight out of college in droves".

Hiring more self taught and bootcamp people is often advocated as a way to increase diversity, and a lot of roles don't require a college degree (plus a lot of self taught and bootcamp folks are very good). However, on a large enough dataset, they will be weaker candidates. So if you use those sources to improve diversity, you'll end up with a correlation between diverse hires and more junior devs. Which screws up with every other metrics you'll look at.

If the bar is the same, but there are fewer female applicants than male applicants, then there will necessarily be fewer females accepted (assuming standards are applied equally).

Until the number of women in our field are on par with men, efforts to create an artificial balance will mean lowering the bar.

The Github results were based on the assumed gender of submitters based on things like having female anime character avatars and "female sounding" usernames. I would posit that methodology as highly flawed.