I’d like to correct a big misunderstanding (source: worked at the EEOC for a number of years):
Scott says: “(here “the applicant pool” is an abstraction, often but not always the same as the general population, which is poorly defined and which bureaucracies can interpret however they want. It’s definitely not the same thing as the actual set of qualified applicants to the business!)”
This is simply not true. Companies are required to track and maintain records of candidates. When the EEOC considers a hiring discrimination case, they obtain this data and can use it to see if there is a statistically significant difference in hiring rates between applicants who are in the protected class and applicants outside the protected class. Ideally, the expert the EEOC uses can account for job-related characteristics of applicants (previous experience) and the characteristics of the job applied to.
(This is part of why there should not be hard quotas, job-relevant characteristics are sometimes correlated with protected class-status.)
If the company does not have quality applicant records, the EEOC needs some benchmark to compare the share of protected class members to. Usually, this is the share of the protected class within the geographic vicinity of the firms locations who work in the firm’s specific industry. This is obviously imperfect, but here are a couple of relevant points:
(1) In almost all the cases I was involved with, we heard directly from former employees or often HR personal from the company about specific issues, and there was substantive anecdotal evidence that discrimination of some form was happening. Usually, this anecdotal evidence is pretty serious, and I think it’s reasonable that this shifts your priors. If there are very large differences between the share of workers in the census and at the firm in the protected class, it seems reasonable to say the company should be able to explain this.
(2) The company really should be keeping track of its applicants! If they aren’t, or they don’t give the data they have (illegal) the EEOC has to do something.
(3) The shortfalls I saw were almost always pretty large. We aren’t in a situation where, oh, the Census shows 30% of men are servers, but in your restaurant its only 27%! It was more like: the Census shows 30% of men are servers, you have 3 male servers across all of your 20 locations. The court uses the same cut-off for statistical significance on proportions tests as most research papers (.05 p-value) anyway.
Are there reasonable criticisms of these methods? Of course. But we had to try to reach the truth the best way we could, or at least to do a thorough job of analyzing the data and then let the judge/jury decide from there (Though most cases ended with mediation).
The issues with using disparate impact are mitigated by the fact that the disparate impact measure has to be job-related. It would be inappropriate to have a test for whether a person could lift 50 .lb boxes for a computer engineer role, but you probably should have a pre-employment test for that if the job is a construction worker. Yes there are issues with this measure, and yes it comes down to argument and precedent, but it’s probably better to use this imperfect method than allow firms who want to discriminate an easy get out of jail free card with spurious requirements.
There are always going to be tradeoffs in the way you set rules, (I personally don’t know how I feel about background checks, its seems reasonable for employers to screen on this, but then again, people deserve second chances), but I think this idea of the EEOC as incompetent/SJW crusaders just does not match my experience at all. It’s easy to make caricatures when you only focus on the extreme downsides of any tradeoff.
“but it’s probably better to use this imperfect method than allow firms who want to discriminate an easy get out of jail free card with spurious requirements”
Do you have an alternative suggestion of what to do? So let's say we have some plausible anecdotal evidence of discrimination or there is a big discrepancy between the share of some protected class at the firm and the share at other firms in the same industry/location.
Competitive markets take care of heavy discrimination because firms are punished with high labor costs. For discrimination whose costs are smaller than a firm’s current competitive “moat” we simply accept that the costs of fixing it are not worth it.
We do this for all sorts of problems. With enough authoritarian power I could ensure that nobody in the US was obese, for example. But nobody thinks it’s worth it to send troops into everyone’s house weekly for weigh-ins and pantry audits.
We’ve all just come to reflexively assume that even low-level discrimination is worse than nuclear holocaust because anyone born in the last 50 years has been raised with that message.
I’d like to correct a big misunderstanding (source: worked at the EEOC for a number of years):
Scott says: “(here “the applicant pool” is an abstraction, often but not always the same as the general population, which is poorly defined and which bureaucracies can interpret however they want. It’s definitely not the same thing as the actual set of qualified applicants to the business!)”
This is simply not true. Companies are required to track and maintain records of candidates. When the EEOC considers a hiring discrimination case, they obtain this data and can use it to see if there is a statistically significant difference in hiring rates between applicants who are in the protected class and applicants outside the protected class. Ideally, the expert the EEOC uses can account for job-related characteristics of applicants (previous experience) and the characteristics of the job applied to.
(This is part of why there should not be hard quotas, job-relevant characteristics are sometimes correlated with protected class-status.)
If the company does not have quality applicant records, the EEOC needs some benchmark to compare the share of protected class members to. Usually, this is the share of the protected class within the geographic vicinity of the firms locations who work in the firm’s specific industry. This is obviously imperfect, but here are a couple of relevant points:
(1) In almost all the cases I was involved with, we heard directly from former employees or often HR personal from the company about specific issues, and there was substantive anecdotal evidence that discrimination of some form was happening. Usually, this anecdotal evidence is pretty serious, and I think it’s reasonable that this shifts your priors. If there are very large differences between the share of workers in the census and at the firm in the protected class, it seems reasonable to say the company should be able to explain this.
(2) The company really should be keeping track of its applicants! If they aren’t, or they don’t give the data they have (illegal) the EEOC has to do something.
(3) The shortfalls I saw were almost always pretty large. We aren’t in a situation where, oh, the Census shows 30% of men are servers, but in your restaurant its only 27%! It was more like: the Census shows 30% of men are servers, you have 3 male servers across all of your 20 locations. The court uses the same cut-off for statistical significance on proportions tests as most research papers (.05 p-value) anyway.
Are there reasonable criticisms of these methods? Of course. But we had to try to reach the truth the best way we could, or at least to do a thorough job of analyzing the data and then let the judge/jury decide from there (Though most cases ended with mediation).
The issues with using disparate impact are mitigated by the fact that the disparate impact measure has to be job-related. It would be inappropriate to have a test for whether a person could lift 50 .lb boxes for a computer engineer role, but you probably should have a pre-employment test for that if the job is a construction worker. Yes there are issues with this measure, and yes it comes down to argument and precedent, but it’s probably better to use this imperfect method than allow firms who want to discriminate an easy get out of jail free card with spurious requirements.
There are always going to be tradeoffs in the way you set rules, (I personally don’t know how I feel about background checks, its seems reasonable for employers to screen on this, but then again, people deserve second chances), but I think this idea of the EEOC as incompetent/SJW crusaders just does not match my experience at all. It’s easy to make caricatures when you only focus on the extreme downsides of any tradeoff.
Thank you for providing this perspective.
+1
“but it’s probably better to use this imperfect method than allow firms who want to discriminate an easy get out of jail free card with spurious requirements”
Disagree pretty hard on this.
Do you have an alternative suggestion of what to do? So let's say we have some plausible anecdotal evidence of discrimination or there is a big discrepancy between the share of some protected class at the firm and the share at other firms in the same industry/location.
Competitive markets take care of heavy discrimination because firms are punished with high labor costs. For discrimination whose costs are smaller than a firm’s current competitive “moat” we simply accept that the costs of fixing it are not worth it.
We do this for all sorts of problems. With enough authoritarian power I could ensure that nobody in the US was obese, for example. But nobody thinks it’s worth it to send troops into everyone’s house weekly for weigh-ins and pantry audits.
We’ve all just come to reflexively assume that even low-level discrimination is worse than nuclear holocaust because anyone born in the last 50 years has been raised with that message.