Saturday, June 01, 2013

The Challenge: Prove that Gender Discrimination in Labor Markets Exists


SleeZee Lyers in the comments to my earlier post on the gender gap in wages asks this question:

Regarding hidden discrimination, I would think that in the 50 years since the Equal Pay Act of 1963, that if such hidden sex associated wage discrimination as you hypothesize existed, that you would be able to find testimony to that effect from retired managers, retired executives, retired HR employees.
Surely someone must know and be ready to talk!
...
Occam's Razor isn't the be all and end all, but given a choice of personal choice / no discrimination or discrimination hidden by thousands for 50 years, I'd say the burden is on you to demonstrate that discimation.

This post is my answer to that Occam's Razor argument, though I wish to preface it with the fact that I believe the earnings differences reflect many reasons:  Choice based on societal expectations about what is appropriate for women and men,  gendered differences in family responsibilities, gendered preferences (whether innate or societally molded or both) and discrimination of various types.  Thus, there is no reason to go for just one explanation, such as choice.

To return to the main point:  That the burden is on me to demonstrate that gender discrimination exists in the labor markets:

First, there are fields of studies which do exactly thatThe audit studies are one group.  These consist of using trained actors, in this case men and women, to go out and apply for jobs in some industry.  The actors are coached to say all the same things and they are provided with equally good resumes.  The studies usually randomize the order in which they visit the firms and do other stuff to guarantee that the results make sense.  The studies then measure call-back rates and other measures to see whether the female and male job applicants, otherwise the same, are treated the same. 

The classic study in this field is a 1990s study about server job applications in Philadelphia restaurants. It demonstrates some discrimination against female applicants to server jobs at that time and in that place.

The other important example of studies which have demonstrated the impact of gender discrimination is the classical orchestra study.  Musicians audit to get employed by orchestras.  A simple change in auditing rule:  introducing a screen so that the evaluators cannot observe the appearance of a musician but only his or her musical talent increased the probability that a female musician would be hired by an orchestra.

A further group of studies which can be used to study possible discrimination in hiring are the correspondence studies where various evaluators are asked to judge an application.  Some evaluators get the application with a female name, others get the exactly same application with a male name.  Given that the actual application is the same for both names,  in the absence of any discrimination we would expect the average evaluations of the candidates to be the same.

This is sometimes the case in such studies, but not always.  A recent study in this field shows that science faculty evaluated fictional female applicants to a laboratory manager position more severely than the fictional male applicant.  In other words, being called "John" rather than "Jane" caused the same application to be treated less harshly and also resulted in higher estimated salary offer.

Both male and female evaluators treated "Jane" worse than "John," by the way.  Thus, what these studies find is probably a societal and unconscious gender bias, not some kind of explicit discrimination by either men or women.  Other studies in this field have also found that female evaluators are usually no less discriminatory than male evaluators.

Correspondence studies about gender do not always show discrimination just against women in gender studies.  What seems to matter here is whether a job is regarded as somehow "belonging" to men or somehow "belonging to women."  Women are judged more harshly in traditionally male-dominated occupations (such as science and in writing plays), men are judged more harshly (in at least some studies) in traditionally female-dominated occupations (such as secretarial work). 

Most of this appears to be something the evaluators are unaware of.  In other words, they are not explicitly singling out applicants with female or male names. 

But note that whatever the causes for this might be, the likely effect this tendency has is to keep occupations more gender-segregated:  Men are more likely to be hired in traditionally male occupations and more likely to be offered a higher starting salary, whereas the reverse applies to women in traditionally female occupations.  That the latter occupations pay much less is, however, important to remember in this context, because the benefits the applicants accrue from being treated as "typical" for their occupations are smaller for women than for men, on average.

Second, the existence of discrimination can also be measured from court cases which decide for the plaintiff in gender discrimination cases.  Such cases have appeared in the years since the 1960s and are too numerous to list here.  A few examples:  The AT&T case, the Price-Waterhouse v. Hopkins case and the Lily Ledbetter v. Goodyear Tire&Rubber Co case.

It is more difficult to study the existence of any possible gender discrimination in long-term labor contracts, because we cannot force actors to keep on acting roles over time and because it is much harder to control for individual differences in skills etc. under that setting.  The multiple regression techniques which studies us are a way around that.  If we could establish and measure all the variables which are non-discriminatory but which affect earnings, we could create studies where whatever gender difference we have been unable to account for after controlling for all those other variables would clearly be due to men and women being treated differently just on the basis of their gender.  But in reality there are always variables we don't have data about.  This means that the unexplained residual even in good studies could be an overestimate of discrimination.

At the same time, some of the variables which are included in the "neutral" category could themselves have a partially discriminatory background.  For instance, in my earlier post I noted that if women don't get promoted into certain jobs then the fact that they are not in that job category terribly often might not be a "neutral" part of the explanation.  That would require that occupations are simply chosen in the same way by both men and women.

This post is most likely a partial one.  It probably should include a discussion of the different concepts of discrimination (including institutional discrimination etc.), but I think I have written enough for the time being.