Sunday, April 20, 2014

Yes, the Pay Gap Persists

Mark Perry and Andrew Biggs, economists at the American Enterprise Institute, argued recently that no pay gap exists between men and women after you control for the different choices they make. The story of a woman who is paid less than a man she is exactly alike, they claim, is false.

I took issue with this argument in two posts. When I actually ran the numbers, I found a persistent pay gap on the order of 4 percent to 10 percent, accounting for a battery of things -- frankly, everything that I could think of, and everything labor economists usually consider -- occupation, work experience, education, race, marital status, children, union membership, geographic location, and weekly hours. And I also wrote that it's probably wrong to take all these things as unaffected by pressure or discrimination.

Perry responded last week in a post I somehow missed. It's worth a follow-up. "What if gender discrimination could be completely eliminated, would there still be a gender wage gap for reasons not related to discrimination against women by employers?" he writes. He argues that the pay gap might persist because of gender differences in risk tolerance. Men might take on higher-risk jobs, such as in oil drilling, and receive what's called a compensating wage differential. Perry also suggested that the gender gap might persist because professional athletes and musicians are paid well and tend to be men.

Sadly, his argument makes no sense. Let me explain why.

1. My regression has "fixed effects" for occupation. This means that it fully accounts for any occupation-level compensating differentials for risk. So everything Perry and Biggs write about men dying in forestry, or what have you -- yeah, my analysis accounts for that. That's what a fixed effect is.

2. My analysis is of workers paid hourly wages. Professional athletes and musicians are not hourly workers. So this can't possibly explain the pay gap.

Look, I understand why Perry and Biggs have to respond to me and Matthew Yglesias, who riffed on my data analysis for Vox, where I am also a contributor. They misrepresented the research consensus on the gender pay gap in a major newspaper, and I called them out on it. 

I would agree with them that the 23-percent number reflects more than discrimination. But if they are going to try to explain away the pay gap, they're going to need to try a bit harder than this. I would know: That was the point of my original post, to see if I could entirely eliminate the pay gap with controls. I couldn't. I easily could make it small, as it surely is, but in the process of spending hours on the data analysis, I grew convinced it was real.

6 comments:

  1. The unemployment rate among men is higher so this may represent the choice of lower pay for enhanced employment and be the result of self discrimination.

    ReplyDelete
    Replies
    1. For the period studied -- 1990 through 2013 -- the gap is not much to consider: http://research.stlouisfed.org/fred2/graph/?g=xU0

      Delete
  2. Read the book "give and take". Studies show that woman and men receive similar starting offers .... but men are more likely to ask for more than initially offered. And men are more likely to negotiate raises.

    ReplyDelete
  3. Hey Evan-

    I think you need to look at paths of experience rather than just "fixed effects" by occupation. The idea of a difference-in-means comparison is that the control group serves as an approximate counterfactual for the treated group. Studies show that women (especially professionals) tend to suffer more disruptions in their careers, which in turn inhibits the formation of human capital and/or networking. There's some research from Booth, I think, that looks at the career paths of women coming out of Harvard professional schools.

    Your stronger point, I think, is that some of these things should not be controlled for, since they are in fact a result of gender structures (which is likely the major confounder that you're only catching by proxy in your regression), so in a sense they are "choice variables" instead of exogenous shocks.

    ReplyDelete
    Replies
    1. I agree, but a big limitation of the CPS data is that it's not a panel. I'm actually working with someone at Princeton on PSID data to look into this, which would address exactly your 100% correct point about "paths of experience."

      Delete
  4. Sowell has also looked at some career path data.

    I agree Perry's rebuttal did not have strong points. And definitely agree that some of the control variables are themselves influenced by gender roles and discrimination.

    I have seen analyses similar to yours that whittle the pay gap down to mid single-digits when most relevant variables are controlled for. Still, you can never control for everything...quality of education, risk tolerance within occupations, etc... I still find it hard to swallow that a somewhat meaningful (5-10%) competitive advantage is hanging out there for nearly every business if they simply hired women across their organizational structure. Maybe, but would be very surprised.

    ReplyDelete