It's trading off the false positive vs. false negative rate in the hiring signal. The question is what your assumptions are about the sources of error in that signal.
If you assume that the variance consists exclusively of false negatives due to discrimination, then the extra interviewers will work as Google claims and no bars will be lowered.
If you assume that the variance consists of randomly distributed error that's the same for all interviewees, then the second interviewer is just some statistical sleight-of-hand that's, on average, equivalent to lowering the bar.
Of course it's possible that the truth is in between.
>Of course it's possible that the truth is in between.
In this case, isn't the finding the truth pretty straightforward? Pull the performance reviews of the second chance hires. Check to see if there's a clear correlation in review and find a pattern.
Are we assuming that Google, which specializes in using data, does not bother to mine its own data with regards to its hiring practices?
If you assume that the variance consists exclusively of false negatives due to discrimination, then the extra interviewers will work as Google claims and no bars will be lowered.
If you assume that the variance consists of randomly distributed error that's the same for all interviewees, then the second interviewer is just some statistical sleight-of-hand that's, on average, equivalent to lowering the bar.
Of course it's possible that the truth is in between.