Earlier quoted context omitted.
My point is that, if there is in fact a pipeline problem, that probably does not constitute a shift in responsibility away from the hiring company. Saying "you cannot expect to make things better than they are at the input of your pipe segment" is explicitly attempting to shift responsibility away from the hiring company. I simply disagree. I think a company is still largely responsible for the outcomes of its hiring…
Are you saying: 1) the input is not actually biased 2) the company is responsible for making the output less biased than the input (or produce an unbiased output from a biased input) 3) the company is responsible for making sure that they don’t cause the output to be more biased than the input, (but you are not saying that it is responsible for more than that) 3’) The same as 3, except that the amount of bias in the…
Bias exists and is a problem. It is worth spending additional resources to draw additional samples from the underrepresented population in order to offset that bias. With more samples from that population than the previous pipeline stage would provide if you sampled evenly, you can counter some of the bias without adjusting any standards of quality (capability) based on the population being sampled.
If you are a member of the overrepresented population, you will have a smaller chance of being hired with this intervention. But for someone who is hired, their skill level will be independent of what population they belong to. (Women will not be given an easier interview. They will be more likely to be offered an interview in the first place.)
Whether that is "fair" depends entirely on how you define fairness. This procedure stacks the odds against a man. Our current overall system stacks the odds against a woman. Both can legitimately complain about unfairness.
Thus, it's largely irrelevant that the percentages in earlier stages of the pipeline mean that there's no way for the overall balance can reach 50/50 through only later-stage interventions. Where did this magic 50% figure come from? The only point is to improve the percentage from where it is now. The issues that prevent achieving 50% are real, but don't prevent progress anywhere in the pipeline.
I imagine the additional costs of sampling more from a smaller population would rise dramatically the closer you try to push the outcome towards 50%. So companies will have to decide how much they're willing to invest. Fortunately, there is still real value in pushing beyond the status quo even if you don't get to 50%.
(More generally, the magic number is not always 50%. It's the proportion of the URM in the overall population. 50% is roughly the female part of the population in areas advanced enough to have these sorts of jobs available.)