Plus they essentially admit that anonymizing the data makes it less useful and their claimed solution is to use a larger sample size. Which is like offering "buy more fuel" as a solution to poor fuel economy. It's no solution to the efficiency loss (at any given sample size it's still worse) and you may not always be able to get a larger sample size.
None of which addresses the problem I identified anyway, which is related to identifying the cause of the disparity once one is discovered, which is already almost intractably hard even without anonymized data.
The much better solution is to identify specific instances of discrimination and address them (and the mechanism of discrimination they represent) regardless of what the statistics say because, again, aggregate statistics can both say that something is wrong when it's not and that nothing is wrong when it is, and the only way to tell is by looking at the individual cases.
https://arstechnica.com/tech-policy/2009/09/your-secrets-liv...
Plus they essentially admit that anonymizing the data makes it less useful and their claimed solution is to use a larger sample size. Which is like offering "buy more fuel" as a solution to poor fuel economy. It's no solution to the efficiency loss (at any given sample size it's still worse) and you may not always be able to get a larger sample size.
None of which addresses the problem I identified anyway, which is related to identifying the cause of the disparity once one is discovered, which is already almost intractably hard even without anonymized data.
The much better solution is to identify specific instances of discrimination and address them (and the mechanism of discrimination they represent) regardless of what the statistics say because, again, aggregate statistics can both say that something is wrong when it's not and that nothing is wrong when it is, and the only way to tell is by looking at the individual cases.