Why Did Gender Wage Convergence In The US Stall?

>> John Taylor: So we're very happy to have Peter Blair speak to us today. There's a joint paper with Benjamin, I don't know, up there somewhere. And it's great to have you here. You're visiting as a national fellow, unusually, at Harvard, which is great. And I was just so fascinated by this economics lab.

Maybe it's another subject for another day, but today we understand. It's a great topic you're going to talk about. Why did gender wage convergence in the United States stall? It's a great question. We're all interested, so take it away, Peter.

>> Peter Blair: John, thank you so much. Can you all hear me?

Let me know if you can't hear me. In my mind, I speak very loudly, but everyone tells me that I speak very quietly, so that's a bit of the problem. So Ben is here. You want to wave Ben, on the screen? Ben was one of my first PhD students at Clemson.

He's now an assistant professor at St Bonaventure University. And he just had a paper that was revised and resubmit yesterday. So congratulations, Ben. I want to start off the talk by telling you a little bit about some other work. And the reason I do this is, how many of you in here are movie buffs, like, you go to the theaters?

Yeah, they always show you a trailer before they show you the actual movie. And I think that that's really instructive, right? Because this is an amazing opportunity to advertise other research, but it's more than that. I want to tell you about this other project just very briefly, in part because it's probably been one of the pieces of work that has had the most important policy impact of all of the research work that I've done.

So, in the United States, about 60% of workers don't have college degrees. But if we look at job postings, increasingly job postings require college degrees. And all of us in the room knew that, we didn't really learn how to teach as professors when we were in graduate school, but we were thrown into that world, and we learned how to do that through experience.

And that's really how a lot of learning works. And so, along with colleagues at Opportunity at Work, a nonprofit in DC, we tackle this problem of, how do you identify the skills of workers without college degrees who, typically in our profession, have been called unskilled? And so we developed an algorithm that uses data from the Bureau of Labor Statistics to map the skills that a worker has from their current job onto the skills that are required for other jobs.

And by doing that, we identified that there are about 30 million out of the 70 million workers in the United States who don't have college degrees, who have the skills for higher wage work. And we've even renamed how we talk about these workers. We call them skilled through alternative routes other than the bachelor's degree, or STARs.

That's the acronym. So we thought this work was super important. We did this work, and about five months after that, George Floyd happened. And we probably all can remember where we were when we were watching that 9 minutes and 29 seconds of George Floyd being brutally murdered. And many companies in the US said, what can we do to address racial inequality in the United States?

And the traditional answer is, let's just throw money at the problem. And my co-authors and I thought, well, there's more that companies can do. They can hire based on skills and not degrees, because although degree requirements seem to be race neutral on their face, they encode a huge legacy of racial inequality in education.

From the anti-literacy laws of the 1800s that made it illegal for black Americans to learn how to read. And it made it illegal for white Americans to teach black Americans to read, to separate and unequal schools, to all of the other things that have happened since then. And so we wrote this op-ed that was published in the Wall Street Journal, encouraging companies to hire based on skills, not degrees, to hire workers who are skilled through alternative routes, who are STARs.

And guess what? It sparked something. And 70 Fortune 500 companies committed to hiring 1 million workers who are STARs over the next ten years. Since then, I think four states have removed their degree requirements from their state jobs. New Jersey just did it. I saw the announcement, I think yesterday.

Maryland did it. Pennsylvania did it. Arkansas did it. Colorado did it. So red states, blue states. And so this work has had a pretty huge impact. And most recently, the National Ad Council, which does a lot of public service announcements, they picked up STARs as their cause for the next two years.

And so they developed these really high quality ads to encourage companies to tear the paper ceiling based on the work that we have done. And I just want to share this video with you to give you a sense of this. You may have already seen this during the run up to the Super Bowl.

You might see billboards on the highways encouraging companies to tear the paper ceiling. If you can, Marshall, run the clip.

>> Speaker 3: Bosses couldn't see me as a leader.

>> Speaker 4: I've run this place for 20 years, but I still need to prove that I'm more than what you see on paper.

 

>> Speaker 5: As long as I've been able to reach a keyboard.

>> Speaker 3: This is what I do. It's second nature for me, coordinating 100 details at once.

>> Speaker 4: That's the way my mind works. I have a very mechanical brain.

>> Speaker 3: I sold them on my skills.

>> Speaker 6: You gotta be so good, they can't ignore you.

 

>> Speaker 3: My magic is.

>> Speaker 4: Analytics and empathy. That's how I get clients. You have to have the confidence in yourself to show up and defy the odds.

>> Speaker 3: I never got a college degree, and today I'm the CEO of my own company. People want to tell me I'm one in a million, when actually I'm one of millions.

STARs are all around us. It's time for them to shine.

>> Peter Blair: So you can check out more information at tearthepaperceiling.org. So now for the main event after that trailer. So I'm going to talk to you today about a paper, Why Did Gender Wage Convergence in the United States Stall?

This is joint work with my colleague Benjamin Posmanick, who is an assistant professor at St Bonaventure University. This is just a picture of the students in my research group. And I like to highlight them because this work is done collaboratively across students from several universities. And I talked about the STARs work that we have in our pipeline.

Other work that we've done looks at occupational licensing as a job market signal. Other work looking at, is school spending efficient? Rick, you and I have talked a bit about this work, using the housing market to look at what happens to house prices when you start to spend more on teachers.

And then another kind of forthcoming paper looks at why is it that elite colleges like Stanford have not expanded the size of their undergraduate class, even though demand is increasing a ton? And while I'm here at Hoover, I've been thinking a lot about the extent to which universities cultivate human potential.

And I'm currently in conversations with a large VC here to get some data on student companies that they've incubated, where they've encouraged folks to drop out of school to see if these companies that are incubated outside of the university are as productive as companies that are incubated with inside the university.

So you can see, a lot of my work tries to think about the connection between credentialing and labor market access and mobility, both within colleges, but then also outside of the university. And I mentioned Ben. I just want to give Ben a little shout out. Ben does a lot of work on corporate boards, thinking about issues of licensure, gender, the impact of tenure on firm performance.

And Ben is a rock star. So Ben, it's great to work with you, and Ben will know the answer to some of the questions that I don't know the answer to. So it's great that he's here as well. All right, so to give you an outline of the talk, I'll just introduce the puzzle that's at the heart of this paper.

So I'll start there. And then I'll talk a bit about some of the data and some of the descriptive statistics, and then from there I'll set up what we think is going to be our causal analysis for getting a handle on this question of why is it that gender wage convergence stagnated.

And then we'll delve into some of the mechanisms that are at the heart of trying to understand this puzzle, then we look at some heterogeneity analysis, and then we will conclude. And hopefully, you'll have a lot of time for discussion, because this is one of those topics where we all have strong opinions about it, a lot of perspectives about it, it's incredibly important from a social standpoint.

So I hope that we can have a broad-based discussion in terms of how do we make sense of these results and what are some of the implications for public policy. So I think that that's something where we can push it. The other thing that I will say too, is we're in the process of revising this paper.

So it was a split decision at the QJE tiers. And so any suggestions that you have for how we can make the paper stronger for the resubmission would be super helpful. I want you to look at this picture here. I'm gonna laser this. This picture is at the heart of this paper.

So what are we plotting here? So what we've done is just plot the raw ratio of women's wages to men's wages over time, starting in 1975, all the way up until the present moment. And so these are just the raw differences in wages at the meeting for women and men.

And what you can see is that in the 1970s to the 1980s, this gap is reducing by about one percentage point per year. So there's a steady decline in the gender wage gap during this time, both in the raw gender wage gap, but also if you control for things like education, experience, occupation, etc., right?

And when you look at this period from the early 1990s onwards, there's a marked break in this trend. If you were to draw a straight line through this, the rate of gender wage convergence goes from about one percentage point per year to about 0.3 percentage points per year.

And this has been an outstanding puzzle in the labor economics literature for a very long time.

>> Speaker 7: I'm sorry, is this earnings or really a wage rate?

>> Peter Blair: These are wage rates. This is hourly wages. This is hourly wages. This is for full time. This is for full time white women and full time white men.

 

>> Speaker 8: But it's just the cash?

>> Peter Blair: Yes.

>> Speaker 8: It's not compensation for this?

>> Peter Blair: That's right.

>> Peter Blair: That's right.

>> Speaker 8: Here we go.

>> Peter Blair: No, 100% it's. So the way to think about this is, this is a fact. I'm not saying anything except showing you this fact and just suggesting that this has been an open question for a very long time within the literature.

And this is gonna be the question that we're gonna try to tackle, which is why is it that the rate of general wage convergence stagnated in the 1990s? So the convergence of the reduction in the gender wage gap during the 1980s is well understood. So there's a decline in unionization.

There's also a decline in the gender gap in observable factors like levels of education, labor market experience, a reduction in occupational segregation by sex. And so the question is, why is it gonna be slower in the 1990s? That's the open question. And in this paper, we're gonna argue that the introduction of job-protected family-leave policies is gonna cause the stagnation.

 

>> Peter Blair: So several hypotheses have been offered to explain why gender wage convergence stagnated in the 1990s, and I'll walk you through them on this slide. And on the second slide, I'll try to explain why we think those explanations don't fully get at the puzzle. So the first explanation is that there's been a convergence in the occupational distributions in the 1990s.

This is by Blau and Kahn. The second explanation has to do with the growth of the service sector, which is a place where women have a comparative advantage, and that growth has slowed during the 1990s. The other explanation that's been offered is that there was no convergence in overwork or overtime work between women and men during the 1990s.

And then the fourth one, which we think is a really interesting idea, too, is that it's been documented by Heinrich Kleven at Princeton, that the childhood penalty that women face has slowed in the 1990s. And that this could be an explanation for why gender wage convergence stagnated in the 1990s.

Well, first, it's important to recognize that if you account for changes in observable factors, in fact, you would have gotten more convergence rather than less convergence.

>> Speaker 9: I'm probably just confused. It sounds like the last point goes the other way?

>> Peter Blair: So the last point-

>> Speaker 9: If there's a stagnation in the mommy penalty, that should accelerate wage convergence, or am I just totally-

 

>> Peter Blair: Yeah, let me explain that separately. So the penalty that women were facing was declining pretty rapidly, and then that decline slowed down in the 1990s. That's what documents, yeah. And so, if you look at just observable factors, controlling for observable factors, you'd have actually expected to see faster gender wage convergence.

In part, because women have higher levels of education, higher levels of experience, and so once you account for observable factors, you would expect to see the gender wage gap to have been smaller than it actually is. And then attributing this to a stagnation in the reduction of the mommy penalty, in a sense, leads us to ask the question, well, what caused the childhood penalty to stagnate, right?

And we're gonna argue that it was the introduction of the Family Medical Leave Act, which provided job-protected leave for workers, and that's disproportionately taken up by women.

>> Speaker 7: I'm just a little confused. You're jumping between levels and trends. So let me put it as a factual question. There's a gap between potential experience and actual experience between men and women.

That gap shrank over time, I presume.

>> Peter Blair: Yeah.

>> Speaker 7: Did that gap continue to shrink at the same pace in the nineties as it had been shrinking in the eighties or? You seem to be making a statement about that, but I'm not sure exactly what you're saying.

>> Peter Blair: Yeah, so your question is, if we look at observable factors, they were converging at a certain rate in the 1980s.

Did they continue to converge at the same rate, and could that be the explanation in terms of the rate of convergence of those observables versus thinking about the levels? So we don't have a precise number on that in the paper itself. But what we'll show you when we do a decomposition, is that if you account for these observable factors, you would have seen the reverse of what you actually see in the data.

Which is, given that women are now starting to have even higher levels of education than men, you'd actually expect to see a reverse gender gap based on just levels of education. So it's not the case-

>> Speaker 7: Actually, that's a level point, not a change in slope point.

>> Peter Blair: No, 100% agree.

So if the levels are gonna be going in the opposite direction, we can double check to see if the rates of those observables could explain this. All right, so why might family-leave policies stagnate gender wage convergence? We're just gonna motivate this conceptually. So the first is that, in about 50% of cases, what we see is that when a worker takes a job-protected family-leave, that work is gonna be shifted on to other workers who are at the firm, or it's going to be shifted onto a temporary worker.

We also see from other research, too, that there are examples of situations in which gender neutral policies are passed, and they have had negative or adverse labor market effects on women. So one example of this is a paper that's in the In 2018 that shows that when universities implemented gender neutral tenure clock stoppage policies after the birth of a child.

That you saw that tenure rates for women actually declined relative to what was happening before that. Malika Thomas, in a paper, too, finds that when the FMLA was passed, that there was a reduction in the rate of promotions for women in jobs to managers. Martha Bailey has a 2019 paper where she studies California's paid family leave policy.

And what she finds is that after this policy was passed, that in the long run, that women's wages and employment was substantially lower by about 8 percentage points on both the intensive and the extensive margin, too. And so there's a growing body of evidence showing that offering gender neutral policies that are differentially taken up by women could cause a negative impact on women's labor market progress, John.

 

>> John: If people looked at women more likely to have children versus women less likely to have children, you would think that 50-year-old women, this wouldn't have any effect on them at all cuz they're not gonna actually have any children.

>> Peter Blair: Yeah, so you're saying, looking at women who are past kind of the traditional childbearing age, that you would not expect to see as much of an impact on them.

 

>> John: Right.

>> Peter Blair: Yeah, I think we did, Ben, we did do this analysis. I don't know if we kept this in the paper, but we could definitely bring it back in.

>> Speaker 9: Then they also take up responsibility for their aging parents.

>> Peter Blair: Yeah.

>> Speaker 9: So, even if you didn't find something, that wouldn't necessarily refute the point.

 

>> Peter Blair: Yeah, and the other part, too, is that the family leave policies also covers adoption of a child. And so if you adopt a child, like, later on in life, that, too could be something.

>> John: General point is that gender is a rough correlate of likely to take family leave.

 

>> Peter Blair: Yeah.

>> John: So you would expect that better measures of likely to take family leave would then explain why you're seeing this gender correlate. And I don't know if you can find any better measures, but if you can, that proves your point.

>> Peter Blair: That's a good point. So you think age would be one?

What are some others, John, that you-

>> John: You can think of others.

>> Speaker 7: There's two effects.

>> Peter Blair: No, no, no, I mean, this is actually really useful because I think it would be helpful for us to explore some of this heterogeneity. One of the challenges is, and Dan, you brought this up, is that there are these kinds of general equilibrium effects that, in a sense,

>> Peter Blair: It's always difficult to think about which individual characteristics you should look at to make predictions about how people are going to respond.

But age is one, what are others?

>> Speaker 13: One obvious one is whether they already have children.

>> Peter Blair: Okay.

>> Speaker 7: Family leave policy also subsidized-

>> Speaker 9: But you can definitely ask.

>> Speaker 7: Taking care of kids rather than spending more time in the labor market, which can have permanent effects on your earnings over your entire career.

So even 50-year-old women who are beyond the childbearing age can be affected by having passed through that policy earlier in their lives. Men can, too, obviously. But one might expect women to take up these policies, these benefits more than men.

>> Peter Blair: Yeah, one of the things that we're gonna find, so we do do some heterogeneity by whether you have a child or not.

And what you see is that mothers, the gender wage gap among women who are parents is a lot larger than for women who are single. And we see a much more rapid rate of gender wage convergence among married workers before the passage of these policies and much more stagnation after the implementation of these policies and for single women.

And so that checks out. They were converging much more quickly in the absence of this policy, but they start to converge a lot more slowly after these policies are implemented.

>> John Taylor: Show the relationship between that sharp line and some of these policies.

>> Peter Blair: Yeah, so I'm gonna show you that.

That's gonna be so let me get a question.

>> Speaker 11: Did you also check the effects of this policy on fertility? Or because they subsidized families or having kids, did you check whether it actually had a positive effect on having a number of kids?

>> Peter Blair: We can do that.

I wanna make sure to frame the exercise that we're doing accurately. So this is not a policy evaluation of the family Medical Leave act. What we're trying to do is to say that there's a puzzle, which is that if you look at the gender wage gap over time, it was converging and then it stagnates.

What caused this? That's what we're focused on. And what we're going to show you by using state level policies that were passed before this, is that you see a similar pattern of gender wage convergence followed by gender wage stagnation, even for these policies that rolled out across the United States prior to the implementation of this federal policy.

And so I want you to keep that in focus, which is why did gender wage converge and stagnate? Now whether or not the FMLA impacted fertility and a lot of these other things, they're very important. But we try to be really clear in terms of making sure that what you take away from this paper is that we're laser focused on trying to answer this really big puzzle in libre economics, because we don't want the takeaway message to be that we have an anti FMLA agenda.

It just turns out that this particular policy is implicated in this question.

>> John: As you described it, it would be a level effect, not a growth rate effect, and it should be quantitatively reasonable. So if one-fourth of women are gonna take a one month of annual leave, and that's a week a year per woman, so then wages should go down by 152nd in a level effect, and then keep growing.

You build up to some enormous. This is 1015 to 20 log points at the end, and it's a growth effect, not a level effect.

>> Peter Blair: So there's work, for example, if you look at the minimum wage literature, where that's another context where you might expect to see level effects.

So policy diffusion takes time, so that's one thing. And so there could be some latency or some dynamics in terms of that. Wages are also sticky. There's downward rigidity in wages. And so, for example, like, if a worker becomes more costly to you today, it might be very difficult for you to cut their wages.

There was a paper, I think, that was presented, this was at the NBR labor studies meeting, where they were doing some surveys in Europe, trying to see, like, why is it that firms are more likely to fire people than to reduce wages?

>> John: So if we've got to the level, shouldn't we go back to growing?

And does it make sense that it's 20 log points?

>> Peter Blair: In terms of the point estimate, so we could benchmark this to what Martha Bailey finds in her paper. So, if you look at what she finds in her paper, looking at the California policy, she finds that ten years out, women's wages are about 8% lower.

And when you look at the impacts that we're finding here, so if you think about 20 log points over 20 years, you're looking at a comparable estimate to what Martha finds in our papers. I think that what we're finding here is within the ballpark.

>> Speaker 13: I'll just say also to John's question, this gentleman's, I don't know your name, point about fertility and marriage even might be relevant because that could accumulate over time.

 

>> Peter Blair: Can you say more?

>> Speaker 13: In a sense, once you have a child in the household that you wouldn't have had.

>> Speaker 7: To work less for the rest of your life.

>> Speaker 13: You're going to be dealing with that kid for at least 18 years and it may have effects beyond the first year.

 

>> Peter Blair: That's a good point. So in some sense you're saying that some of the-

>> Speaker 13: In having children that you wouldn't have had and it changes the way you live your life.

>> Peter Blair: Yeah, that's it, yeah. Thank you very much. That's a very helpful point. All right, so let me describe for you what we're going to see in the FMLA.

So the FMLA is passed in 1993, and this is going to guarantee 12 weeks of unpaid family leave to workers. And this is going to be for the birth or adoption of a child. It could be for your own illness, for the illness of a loved one that you need to take care of.

And there are going to be certain requirements in terms of, for the most part, it's going to cover workers who are full time, full year more or less. This is going to be a gender neutral policy, but that's going to have much higher take up by women than men.

This policy is also going to be preceded by similar policies in 12 states and also the District of Columbia. And we're going to be leveraging some of that pre-FMLA variation in our causal research design. Let me describe for you the approach that we're going to take in the paper.

First, what we're going to do is we're going to show you that the trend break in gender wage convergence is going to occur around 1993, which is the FMLA year. That's going to be purely descriptive. And then secondly, what we're going to do is use the state variation in laws before the FMLA to show you that even with these laws that pass prior to the FMLA, we see a similar pattern of gender wage convergence followed by gender wage stagnation just using those pre-FMLA states.

The second thing that we're going to do is we're going to combine both that state level variation and the federal level variation, which is going to treat the remaining 38 states to show you that quantitatively and qualitatively, we get a very similar pattern of results in terms of the impact of these family leave policies on gender wage convergence.

And the state level variation is going to be important because there are a lot of things happening federally in the early 1990s. And so you might be concerned about policy endogeneity, that welfare reform happens in 1996, the EITC gets changed. There's a whole lot of other things doing happening at the same time.

And so the state level variation really is going to give us a handle on, just generically, what happens when you pass family leave policies. We're also going to test the extent to which alternative explanations like the EITC or welfare reform could impact our results. We're going to directly control for when welfare reform waivers are going to be passed at the state level in our main analysis.

And then for the EITC, we're going to split our sample into workers who don't have college degrees and workers who have college degrees with the understanding that the EITC is going to be a lot more binding for workers without college degrees than for workers with. And we're going to show you that a lot of the action is really for workers with college degrees.

And then we're going to conclude with some heterogeneity analysis and also a description of differences in leave taking by gender between women and men using some survey based results. Right, so literature review. I talked a bit about some of the prior work that shows, you know, the negative labor market impacts of even gender neutral policies for women.

That's not just in the United States. There's even a recent study that was published in the AEJ applied looking at Sweden that showed that when they implemented paid family leave, that this increased the gender wage gap and reduced women's labor supply.

>> John Taylor: The timing is clear?

>> Peter Blair: Pardon?

 

>> John Taylor: The timing is clear. You look like a particular event here for the US. Is the timing the same in Sweden?

>> Peter Blair: No, no, the timing is not this. So in Sweden, it was the extension of an existing policy from I think three months to twelve months. And so it's like, yeah, this is a very completely different timing.

This is just to say that when you look at even a country that has a lot of social protections existing, like a fundamentally different labor market, you see similar impacts of these types of policies on the earnings and employment of women.

>> Speaker 8: That policy is borne by the employer or by the state?

Because that could have an impact.

>> Peter Blair: By the employer, I would say. I have to double check. I don't know, Ben, if you remember offhand for the et al paper.

>> Benjamin Posmanick: I don't have that off the top of my head, unfortunately.

>> Peter Blair: Okay, but we can get back to you on that because you're absolutely right in terms of, the incidence will matter a ton.

And in a lot of ways, when you think about the FMLA, the firm has to keep the job open for the worker. And so a lot of the incidence is born by the firm and not by the state. And so a potential design flaw with the way that this policy is designed is that because the incidence is on the on the firm and not on the state, you could have a situation in which firms have an incentive to engage in statistical discrimination.

And that almost like takes us to the conclusion in terms of what are some of the policy lessons in terms of policy design, would you expect, therefore, for a specific individual, that the impact would be much larger in a small firm than in a large firm? There is going to be a cutoff.

So firms that are smaller than 75 are exempted from the FMLA. In the publicly available data, we don't have access. We have a very crude access to the firm size and it doesn't line up over time. And so we've kind of struggled to exploit that variation. But given the way that the policy is designed, you would expect to see more of an impact on these larger firms because they're required to comply.

Now, it could also be the case that there could be spillover to smaller firms, because if you're competing against a larger firm that is required to offer this benefit, then you might offer this benefit, but then the cost might be higher to you, even though that's not compliance costs.

It could be competitive costs.

>> Speaker 16: But amongst the large firms, because there's going to be diversification of the dates at which different people will have their pregnancies, would you not expect the impact for a given person is much larger for a small firm that is a large firm.

So cutting off above 75 versus one that has 100,000 or more.

>> Peter Blair: So you're thinking firm, that's 75 to 100 people, versus firm that's 500 is the impact. So you said to the individual going?

>> Speaker 16: Well, so even though the proportions, say, of females are going to be the same, just take that as a stipulation, because there's going to be diversification across sort of the date of when.

And therefore, you're not keeping n slots open for all of those people, you're keeping only the ones that are currently. Whereas for a firm, one slot is quite significant.

>> Peter Blair: Yeah, I guess it would be I mean, because at that level they're just. Yeah, so like that seems reasonable.

 

>> Speaker 16: Yeah.

>> Peter Blair: John.

>> John: Do you know how much medical leave the average woman and the average man take? How many months a year?

>> Peter Blair: Yeah, we can tell you. So the discrepancy is going to be for non-family related leave, there's virtually no difference between men and women, but for the birth or adoption of a child women take about 36 more days on average of leave than men.

 

>> John: That's conditional on the birth of a child. So the 8% wage gap is the average woman is taking 8% of her time off more than the average man is doing. Is that roughly right?

>> Peter Blair: Can you say that again, John?

>> John: 8% wage gap. The average woman to be taking 8% of the days off that year more than the average man.

Is that roughly how much.

>> Peter Blair: So are you saying that if statistical discrimination were happening efficiently in a sense like this, it was actuarially fear that women were being compensated in terms of the time, would this line up?

>> John: Yeah.

>> Peter Blair: We did do that exercise. And we have something that kinda gets a little bit there at the end when we look at what happens to the difference in earnings over time.

So in the first two years that the policy is in place, you virtually see not a lot of. Over time, you see that the difference in earnings grows to about $4,000 over a period of about ten years. And so we could do the exercise where you deflate by, well, how many women are taking leave?

And like, how does this map,

>> Peter Blair: How does this map into kind of is this act, in a sense this is actuarially fair?

>> John: Yeah, I'm not necessarily wanting to use the word fair, but does it line up with the actuarial costs?

>> Peter Blair: Ben, we did do that exercise, do you remember off the top of your head, Ben, the.

 

>> Speaker 17: Aren't you talking about earnings and not wages?

>> Peter Blair: Right?

>> John: Yeah. Average person, yeah, but the average person-

>> Speaker 7: No wage rate.

>> John: The whole point of the paper is they pay women less on average CUZ the women take eight days off.

>> Speaker 8: Let me clarify this right now, because you talk about wage.

Most of us who at least do this kind of research would think that would be the hourly wage.

>> Peter Blair: It is the hourly wage.

>> Speaker 8: Yes, it is the hourly wage.

>> Peter Blair: No, they're not gonna be.

>> Speaker 8: It's just the effect of.

>> Peter Blair: That's right, because remember, what we're after is we're trying to understand the extent to which these policies contributed to gender wage stagnation, right?

So we're starting with this big puzzle, we're asking, can we leverage the variation in these policies to see if after the implementation of these policies generically, we see that the rate of gender wage convergence is gonna stagnate. And that's an empirical question. So let me continue on by telling you a bit about the data that we're goNNA use and then giving you a sense of the descriptive results before moving on to some of the causal analysis.

We're gonna use data from the ASEC supplement of the CPS and we're gonna cover from 1976 to 2016. So this will tell you when we started the paper. In terms of the sample selection, it's gonna be the standard things that you do it's gonna be people who are of working age.

And we're gonna focus on full time workers, in part because the puzzle at the heart of this paper has to do with the stagnation of gender wage convergence for women who are working full-time. And we're also gonna winsorize the data just to keep the middle 99%. But our results are not gonna be very sensitive to this windsorization decision.

These are just the descriptive statistics. There isn't like a ton that's interesting here, except, I mean, you can see the raw gender wage gap in this, the difference in the levels of experience, etc. Okay, so for our descriptive knowledge, we're gonna start off like very simply by just running a mentor wage regression where we're gonna take the-, sorry.

We're gonna take the log of the implied hourly wages of a worker in a given time period and we're gonna regress that on indicators for gender interactive with race. And we're also going to include coefficients for each demographic group in each time period. And so the way to think about these betas is beta1 is going to capture in time period t, what is the difference in the controlled gender wage gap between women and men.

What we'll do next is present a visual where we show you estimates of these betas over time, which are gonna capture the way in which the gender wage gap has evolved over time. Once we account for things like education, experience, occupation, etc. So this is going to be, remember that first figure,

 

>> Speaker 8: Is there any parallel research on the labor supply?

>> Peter Blair: I have a slide here where we can show you that if you look at the gender composition of our sample in full time, like it's not going to be changing dramatically during this period. And so we don't think that the results are gonna be driven by selection.

 

>> Speaker 8: This is just another question that I'm asking about. This is not casting any shadows on-

>> Peter Blair: Okay.

>> Speaker 8: On what you've done. You're too defensive. But isn't it an equally interesting question about the effect of these programs on subsequent hours of work?

>> Peter Blair: It's a super interesting question.

The question that we are trying to get at with this paper is the gender wage gap stagnated during the 1990s. And it's-

>> Speaker 7: They're related cuz there's a literature that shows well-educated women offer suffer sizable penalties throughout their career by taking off time to raise children in their 30s.

 

>> Peter Blair: Yes.

>> Speaker 7: So that's the sense in which the labor supply decision around the time of fertility and early childhood raising is very closely related to your question. And it would help us understand potentially why these effects persist for years and years accumulate over time. Which I don't think the statistical discrimination model pushes you in that direction, that's part of what John was suggesting earlier.

Whereas these decisions about labor supply today affecting future wage rates, has the potential at least to explain what you find.

>> Peter Blair: I see, so the model that you have in your mind is that a woman has a child, drops out of the labor market. And as a firm, you recognize that all of the money that you spent in investing in that person is less valuable, in a dynamic sense.

 

>> Speaker 7: I'm thinking even simpler, think about a Becker model of specialization in the household. If you subsidize spending time child rearing during a critical stage of your career, that can alter your specialization patterns of the woman and of the man too in the marriage for the rest of their careers.

And so auxiliary implications of this view, which are not coming out of the statistical discrimination view. Is that if you look at married men and women five, six, ten years down the road, the men may actually be working more hours as a consequence of this policy. The women may be working less, not just during the 36 days they take off originally that they get compensated for, but that alters their pattern of specialization in the household.

 

>> Peter Blair: Yeah.

>> Speaker 7: That persists throughout the rest of their career. That strikes me as an intentional hypothesis that can account for your facts.

>> Peter Blair: So that could be like a mechanism.

>> Speaker 7: That could be a mechanism.

>> Peter Blair: I appreciate that.

>> Speaker 7: The way to test that is to look at the labor supply responses not just contemporaneously, but several years down the road.

You have the staggered introduction of these FMLA programs by state, which allows at least some possibility of actually teasing that out.

>> Peter Blair: Yeah, okay I appreciate that, and I appreciate you pushing us in that direction. We have another paper where we look at what's happening in the part time segment of the labor market, and we've decided to kind of split those two papers.

But I guess you're pushing us to think about this in a more integrated way. There Is as you mentioned, there is work on this, so Heinrich Cleven at Princeton has a great paper documenting the childhood penalty over time and across state and even across countries, where you're absolutely right, women generically, their earnings dropped by about 20% after the birth of a child.

In terms of the extent to which this is gonna affect specialization, there is work, so Mary Betran shows that what happens with MBAs after they graduate right away is that you see that there's virtually no gender wage gap. But then around the time of childbirth, you start to see women then go in, like, very different career paths and different trajectories.

So in a sense, like, what you'd need is, that's been happening for time immemorial, and so what you need is the interaction of that with these policies, which intensifies, in a sense, it provides more liberty for women to take this path, but that in turn, kind of creates a lot more occupation.

 

>> Speaker 7: Well, it subsidizes both men and women, to take the path, but there's reasons to think that women will be more responsive to that particular subsidy than men.

>> Speaker 17: But it's more than just specialization, isn't it? You just have less investment for any given age, they have less experience and less investment.

 

>> Speaker 7: That's a form of specialized investment in the market versus the household.

>> Speaker 17: You could do without the men.

>> Speaker 7: But you're right.

>> Peter Blair: How am I doing on time, John?

>> John Taylor: About an half hour.

>> John: So, 366 days of time off, really gonna do that, and we look at, there isn't an outburst of fertility didn't jump around.

 

>> Speaker 7: This isn't about fertility. Wait, do we know to be about fertility?

>> Peter Blair: Okay, so these points, Julian, I appreciate you all pushing us in this direction, and that's been one of the challenges that we have when we're presenting this paper, because we really want to focus on what's happening with the gender wage gap.

But then the policy itself is, like, really interesting, whereas we've been treating it, in a sense, instrumentally, in terms of giving us some traction on what's happening with kind of like, this macro trend. All right, so, looking at this picture, so we've accounted for experience, for occupation, and what you see is a very similar picture of gender wage convergence that's happening at a rate of about one percentage point per year before the policy, and then after the policy, it's about 0.02, percentage 0.2.

And so just even descriptively, it's not just the case that there's something magical happening with the covariates that's explaining why we're seeing this appear in the raw gender gaps, all right?

>> Speaker 7: Just white women.

>> Peter Blair: Yes, it looks very similar for black women as well. Now we could put all women together, but if you put all women together, then, like, what does the race dummy mean so, you kind of have to pick your, your battles.

Yeah, did you ever thought on that, John?

>> John Taylor: Your first chart?

>> Peter Blair: The first chart was white women, but it looks pretty similar, yes, the first chart is white women.

>> Speaker 8: Here's the author's prerogative here, you've defined this research to deal with hourly wages of white women, and that's what we're gonna learn about.

 

>> Peter Blair: Yes, and I think white women are what, like 70% of the labor market, if you go further back, but we do have in the paper, companion results for black women as well too.

>> Speaker 8: But the companion results we'd love to see is hours of work.

>> Peter Blair: Hours of work, okay.

 

>> Speaker 8: Not earnings, earnings is the product of hours of work and wage.

>> Peter Blair: Yeah, and so then I think if we can do that, and if we do that, then we should, like, broaden it to include, like, people working at all different levels of hours, because right now we focused on full time, full wage women.

I think it's important to situate this within the context of this literature, so for a lot of the work looking at the gender wage gap, there's a focus on full time, full year women, in part because the labor supply decisions of women are a lot more flexible. And when you start to look at women who are working part time, in a sense, like, that's gonna be like very endogenous to like the family situation and so on and so forth.

So we are really anchoring our approach in this paper to how this puzzle has been framed in the literature, so, in a way that's not so much the author's prerogative as it is as trying to say, this is a question in labor economics that we're trying to get traction on.

And using the definitions, using the sampling frame that has historically defined this question.

>> Speaker 16: But even if you're restricting, as you do, to full time, as you define as 35 hours a week or more, you mentioned that one of the explanations out there is differential overwork, and the impact both exogenously and endogenously on male versus female choice to overwork.

And therefore, it does seem that at least within this data set again, leaving the other one aside, that would be an important thing to be controlling for and differentiating.

>> Peter Blair: Yeah.

>> Speaker 18: So you've made a persuasive case that hiring a woman after this enactment of this law should produce a less valuable employee in the eyes of the firm.

Firm's gonna make less of an investment, so that's gonna create a wage gap, I don't understand why the wage gap is now going to grow. Whatever the forces were that were causing convergence before, has it changed those forces of convergence, as opposed to just changing the level at which it's a reason why there would be a gap in.

 

>> Peter Blair: This is a fair point, and we get this question a lot, and so I appreciate you all pushing us in this, and this is something that it would be helpful for us to leave this seminar with an awesome explanation for when we do the paper. So we've approached this from the standpoint of saying, let's leverage the variation that we do have in these policies, which I'm gonna show you in just a slide, and let's see what happens empirically.

 

>> Speaker 8: It's a finding, not a theory.

>> Peter Blair: Right, it is a finding, now, we have theories of how the labor market should work, that we have an efficient labor market, that costs should automatically be updated into wages. But then we also, that runs up against the reality that there's nominal wage rigidity, where it's hard to cut people's wages, for example, then there's this other feature of, if you think about a lot of bonuses and promotions, that's in proportion to where you're starting wages.

And so, even if you have a drop in the wage level, if growth rates are tied to that initial level, you can have that drop being carried forward in terms of being carried forward. In addition to this Malika Thomas work, looking at the FMLA shows that when the FMLA is implemented, that women are promoted at lower rates too.

This comes back to the point about trajectories that you mentioned, if a woman gets put on a different career trajectory, that can also tilt down their wage path too.

>> Speaker 19: Just adding to what his, the point was, you said it manifests itself 30 years, 20, 30 years out, that changes the slope, not necessarily just a shift.

Because you have somebody coming into childbirth one year, and there's only a small number percentage of women going into that status each year. So then instead of a shift, it leads to a change in the slope.

>> Speaker 20: The slope is controlled for a number of children, so you're gonna end up controlling-

 

>> Speaker 19: No, but-.

>> Speaker 20: Your regressions are level effects. You could just say the graphs are pretty, my regression's a level effect.

>> Peter Blair: So I'll show you what happens to the wage levels as well too in the remaining 20 minutes. Let me show you what I've done first. What we're going to do is we're going to focus on, so you might be worried that there's a lot of things happening in 1993.

And to get away from that we're going to leverage the fact that several states passed job protected leave policies for after women had children and even after fathers had children as well too. And we're gonna leverage these subsample of states to look at what happens to gender wage gaps in those contexts.

These are gonna be the states, so we have Massachusetts in 1972 and then all the way up to Vermont in 1992. And so we're gonna be focusing on these fraternity policies, which in large part many of them, so for example, New Jersey, the law looked identical to what would happen in the FMLA, twelve weeks, job protected leave and so on and so forth.

And so you can think about these as being precursors to the Federal policy. We're gonna run a standard event study design here where we're gonna regress the log of a worker's wages on dummies that are gonna be event time dummies. So if the law is passed in calendar year, why we're gonna use that calendar year as being event time zero.

And we're going to look at all of these event study coefficients relative to the event year before the passage of the policy, so relative to tau minus 1, we're going to interact these with race and gender. We're going to show you primarily the results for white women. The results for black women are in the paper.

So if you're interested in seeing those, we have coverage there. The other thing that I should mention, too is several states are gonna have welfare waivers and we're gonna directly control for that in the specification, too. So you should think about this design as being in a sense purging out some of the contributions from these welfare reforms that you might think could also impact our results.

So these are the state level estimates just looking at those twelve states plus the District of Columbia that passed policies prior to the FMLA. And this is necessarily gonna be noisy because we don't have a ton of observations of variation. But what you can see here, qualitatively, is that prior to the implementation of this policy, you have this steady gender wage convergence.

And after the policy you start to see the stagnation. And we're gonna quantify this in terms of fitting a trend line through these points. That's going to allow for a different slope and a different intercept before and after these policies to quantify what was the rate of gender wage convergence prior to this policy?

What was the gender wage, the rate of gender wage convergence after this policy? And is that difference going to be statistically significant? Hold on one second. We're also going to weight inversely by the standard error of each of these point estimates so that noisier estimates receive less weighting.

 

>> Speaker 8: There are five observations post event which continue the growth.

>> Peter Blair: I'm going to show you these results. I'm going to quantify this. So right now we're running ocular regressions in our brain. Let's run some actual regressions, okay. All right, so the second thing I'm gonna show you now is going to be what happens when we include both the state and the federal variation.

And I promise you those regressions are going to come in just two slides. So we have an additional 38 events that are all happening in 1993, and so we're gonna include those in our event study analysis. And this is what we get when we look at both the state and the federal variation.

So if you had a policy that was passed prior to the FMLA, we're going to use that year as your tau equals zero year. If you had the FMLA as the first time that a family leave policies pass, we're gonna use 1993. And so what you can see here is a similar pattern of gender wage convergence followed by gender wage stagnation after the policy.

And this is gonna be a lot more precisely estimated because we have more events, we have more observations. And so to compare these two, what we do is we regress the wage gaps at each of those time periods on a trend that is before the FMLA and then also on a trend that's post the FMLA.

We also allow for the levels to change before and after the FMLA two. And so think about this as fitting a piecewise linear regression onto those patterns that we saw beforehand. Now let's focus in on this event time trend. So this is gonna tell us what's the rate of gender wage convergence prior to the policy.

In the first column this is using just the state variation in the states that had policies prior to the FMLA. In the state and federal variation, we're gonna have both the 1993 variation that covers those 38 states plus the 12 states and the District of Columbia beforehand. So what you can see is that prior to the implementation of these policies, the rate of gender wage convergence was about 0.7 percentage points per year.

And then after these policies, the rate of gender wage convergence is about 0.2, right, and so there's a statistically significant drop in the rate of gender wage convergence. If you look at the state and the federal variation, you see a very similar picture of gender wage convergence, prior stagnation thereafter.

And this rate of gender wage convergence post period is gonna capture what that rate is, and so this is gonna be, I think, 0.03 percentage points after the policies. Your question, sir, did you have a question?

>> Speaker 8: No, you resolved my objection.

>> Peter Blair: Okay.

>> Speaker 8: My ocules, so to speak, were just fooled by sampling variation.

When you killed the sampling variation by extending the sample, then you cleared up, you validated your assumptions.

>> Peter Blair: Thank you I appreciate that. Thank you. All right, good.

>> Speaker 7: That is good, but now when you bring in the extra data, you're leaning very heavily on the assumption that nothing else happened around the same time that's not controlled forward or regressive and that had effects that filtered forward in time.

I'm not saying I have a solution to that problem.

>> Peter Blair: Yeah, I think that's right. So like the, in a sense, like the concern with just using the federal variation, if you just thought it was only the affiliate and you didn't have any of the state level variation, would be all of the other policies that are happening at the federal level.

You bring in the state variation and only focus on that and you can see, sorry, you can see it's more noisily estimated, but this pre trend is statistically significant. The post trend drop is statistically significant. This difference here is statistically significant. And the rate of gender wage convergence is really small.

And if you look quantitatively, at these point estimates, too, they're very similar, the confidence intervals are overlapping, too. And so what's striking about this is that even leveraging similar policy variation prior to the federal policy variation, we could have learned something that could have indicated that perhaps implementing this policy at the federal level could have had these impacts.

And again, we're just hands on the table reporting what we find when we run this. All of the theoretical questions about, well, why is it that the labor market is adjusting in this way? Completely valid. And our hope is to make sure that we can assuage some of those concerns, at least, like, leverage them in terms of explaining why we think these magnitudes make sense.

And so this is, again, we're restricting to full time, full year workers. This regression is the same event study type setup where we regressed As a dummy for like is a person a woman or not? To see if the composition of the sample of this already selected sample is going to be changing around the time of this event.

Because you might be worried that, well, what's really driving this result is like selection into who is in our sampling frame because this isn't even selection into the labor market because again, full time, full work. And it doesn't seem that around the event time or even far away from the event time, that there's differential selection by gender into our sampling frame that is correlated with this remarkable break in trend that we saw beforehand.

Okay, all right, so what we're gonna do next is try to think about, well, how much traction do we get in explaining the reduction in the rate of gender wage convergence, like using the variation that we have in our setup? And in order to implement this, we're going to do a decomposition that's motivated by the work of Blau and Kahn, as well as this paper by Juhn et al.

And fundamentally, if you think about the data generating process in this context as saying you have some outcome wage of worker I and time period t, that's gonna depend on observable characteristics x. The observable prices of those characteristics beta, which are gonna be time varying, and then on some residual which you can break into two components, into like a variance component and then an idiosyncratic component.

So, we're gonna think about the standard deviation as being a measure of the observed of the price of unobserved skills theta I. So that's kind of like the insight behind this decomposition. And then what we can do is we can look at the differences in the gender wage gap in the pre period from 1976 to 1992, and we can see how much of that convergence gender wage gap?

So the gender wage gap reduces by about 19 log points. How much of that is due to differences in observed x's? So like women having like more levels of like gaining in terms of education, gaining in terms of experience, how much of that is going to be driven by changes in the observed prices of those skills?

How much of that is going to be driven by differences in the unobservables? And so, the gap effect you could think of as being like differences in the unobserved skills of workers. That also is going to include things like any kind of discrimination that workers might face in the labor market, which the econometrician is not going to observe.

What you can see in this first column is that the reduction in the gender wage gap, a lot of that is gonna be driven by differences in the observed X's. But then also there's gonna be reduction in these unobservable skills as well, too. The way that this has been interpreted in the literature is that there's a reduction in gender discrimination that's happening during this period too.

And you might think about this as when you allow women into the labor market, they have access to educational opportunities. Firms learn over time that women are very similar to men and so there's no need to penalize them when they're in the labor market.

>> Speaker 7: Just the observed X's is education.

Then what's the measure of experience? Is it potential experience.

>> Peter Blair: That's potential experience.

>> Speaker 7: Or you're trying to construct some of. Okay, so then, so then presumably the potential, the actual experience differential, which is shrinking, that's gonna show up in one of the other rows.

>> Peter Blair: Yeah, so this.

Yes, that's right. That's right. When we look at what's Isaac?

>> Isaac: Stab at quantifying the difference in actual experience, because you know the universe of women and what share of them are working.

>> Peter Blair: Can you speak up a little bit, Isaac?

>> Isaac: So, you know the universe of women, you know in any given year, what percent of them are working, how much at an aggregate level, and then you just roll that forward and then you can sort of.

 

>> Speaker 7: Yeah, you could treat it as a synthetic panel.

>> Isaac: Exactly.

>> Speaker 7: And then impute the mean hours work today based on what their synthetic panel unit is.

>> Peter Blair: Yeah, we can do that. Yeah, when you look at the difference in terms of. So one way to think about this is the change in the gender wage gap in the, between 76 and 1992 is a reduction of 20 log points.

The reduction between 1993 and 2015 is eight log points, and that difference is eleven. And we can look at, well, which components are substantially different, almost like a difference and difference in a way. And what you can see is that the thing that really stands out in a very glaring way is the fact that the reduction, the gender wage gap coming from a reduction in differences and unobserved skills between women and men was pretty dramatic in the pre period, and it's pretty muted in this post period.

And so, a lot of what's happening here is that this gap effect or the difference in unobserved skills is being reduced at a much slower rate. And so that's kind of like what this decomposition is telling us now. What we can do is we can say, let's look at the estimated reduction in the rate of gender wage convergence that we estimated from our event study estimate and then see if we were to use that reduction of, I think it was like 0.6 percentage points per year, and we rolled that forward across all of these years.

Like, how does that compare to this slower convergence in this gap effect, right? And if you do this, you can kind of explain, like close to about 94% of the reduction in the gap effect between.

>> Speaker 8: How do you get 94?

>> Peter Blair: Say that again.

>> Speaker 8: How do you get 94?

 

>> Peter Blair: So, the way that we get 94 is if you go back to, let's see, if you go back to here, right? So, after the implementation of these policies, you see a reduction in the rate of gender wage convergence of about 0.7. This is 0.67 percentage points. And then if you roll that forward by, I think it's 20 years, then you get this 14.74 percentage points.

And then that relative to this 15.6 percentage points is gonna be 94%.

>> Speaker 8: So, the price effect actually goes the wrong direction since it's positive.

>> Peter Blair: The unobserved price effect.

>> Speaker 8: Yes.

>> Peter Blair: The observed prices. Yes, exactly. Yeah, so in a sense, it's almost like those. Let's see.

Yeah, so what's happening during this time period is like women are getting higher levels of education, but then the returns to those skills are those, the returns of those skills that actually push the gender wage gap in the opposite direction?

>> Speaker 8: Made it worse.

>> Peter Blair: Yeah, okay. What we do next is we think about, we run this regression in levels to try to understand, like, what's happening to the wage levels of women and men.

Is it that the wages of women are being decreased and that's causing the gender wage gap to stagnate? Or is it that the wages of men are going up because now men are being seen as substitutes to women who are stopping out of the labor market for childcare reasons?

And so, if you look just at the raw wages. So this is, we've not done any controls, anything like that. We just want to show you what's happening with the wage levels. These blue dots here are going to be the wages of men. And this vertical red line here is going to be 1993, which is the year of the FMLA.

So what you can see is that men's wages were trending down, and then after 19. 1993, you start to see men's wages going upwards. Women's wages, by contrast, were going up, and they seemed to be unaffected in terms of the path. And so even before running a regression, what this is telling you is that what's driving the stagnation is an increase in men's wages rather than a decrease in women's wages in terms of the levels.

Now, this could be consistent with a story in which there's nominal wage rigidity, and so it's hard to cut the wages of women. But if the work of women is going to be pushed off to men, that you increase men's wages, you pay them an efficiency wage. It could be consistent with that.

It potentially could be consistent with other stories, too, which we're happy to hear because that would be something that we can include in our paper. You're smiling, John. You have some suggestions?

>> John Taylor: This is funny. Go ahead.

>> Peter Blair: Okay, so we can run a similar event study instead.

Like, what we do is we run this in levels as opposed to running this in logs. The reason why we ran the first specification in logs is because the literature is really focused on the gender wage gap in terms of percentages. This decomposition in terms of the levels is going to help us to understand what are the actual adjustments in the labor market that's driving the overall gender wage gap.

Here are the results for men. And what you can see quantitatively, or qualitatively, rather, is that you saw that men's wages were trending down right up to the implementation of these policies. And then afterwards, you see that men's wages start to trend up. So about ten years later, you see it's increased by about a dollar, Isaac.

 

>> Isaac: This is maybe very minor. I noticed these regressions you're controlling per occupation. There's an argument that you don't want to do that and that which occupation you end sort of. Sort of. So, in particular, various stories of how this sort of plays out would be women are less likely to end up managers or women sort of end up in different occupations and sort of conditioning on that.

 

>> Peter Blair: Yeah. So if we don't control for occupation in some sense, if you think that there is discrimination on the extensive margin, this is understating the extent to which that's happening, right.

>> Isaac: I think you can tell stories where it goes the other way as well, where you're not obviously wanting to.

 

>> Peter Blair: Okay, yeah. So yeah, that's fair. But for consistency, in the previous regression, we controlled for occupation fixed effects. So if we didn't here, then you might wonder if we're trying to hide the ball.

>> Isaac: But arguably, in the previous regression, you also might not want to do that.

Just reserve just a general comment.

>> Peter Blair: That's fair. That's fair.

>> Speaker 16: So you previously made the argument that it's I was taught, menu of costs as a graduate student, that it's difficult to cut wages. So what is causing the decline in men's wages pre? Are they shifting to different jobs?

Are they changing?

>> Peter Blair: Well, women are entering the labor market and women are getting lots of education. There's a lot more competition for these jobs that men, in a sense, had a monopoly on beforehand. There's less occupational segregation.

>> Speaker 16: So you think their wages are going down? Cuz if their wages can go down, why can't womens?

So I'm just trying to use the same.

>> Peter Blair: I see what you're saying. No, that's a fair point. That's a fair point. I mean, like, I'm showing you data. I'm saying, one of the reasons why you might see firms being unwilling to cut wages might be nominal wage rigidity.

Now, what I was showing you in the previous page were real wages. So I should look at nominal wages. And we can say, were nominal wages declining? And if we see, for example, that nominal wages were declining, then your point completely holds that, in a sense, nominal wage rigidity should hold for both men and for women.

 

>> Speaker 16: Or it's being driven by something else. And the reason I'm asking that is what is that something else that maybe we should be controlling for?

>> Peter Blair: Yeah.

>> Speaker 16: It's informative either way.

>> Peter Blair: Yeah, I think that's a fair point. And we can look at the nominal wages.

Yeah, but those were real wages. Okay. And so this is looking at the wages of women. So you can see that women's wages were increasing leading up to this event. And then afterwards they kind of like, stagnate, and then towards the end, they start to uptick a little bit.

And so a lot of the action is really happening on the wages of men where there's this reversal of fortunes happening. Men's wages were declining prior to the implementation of these policies. And now men's wages are increasing, but we don't see, sorry, a negative impact on the wages of women.

I have seven minutes. This is good. Making good time. All right, so now something you might be concerned with is that there are other policies that are passed during this time, too, for example, like the earned income tax credit. Maybe this is something that's driving the results. So what we do is we look at these wage level regressions for workers with and without college degrees under the presumption that the EITC should have a bigger impact on workers who don't have college degrees versus workers who do have college degrees.

And let's see, let me show you those results. And so this is looking at men. So remember, a lot of the action was looking at what was happening with men's wages where they were declining before the policy, and then they start to increase after the policy. If we look at men without college degrees, this is on the left, we see a pretty flat pre trend.

So men without college degrees actually were not experiencing substantial declines in their real wages, right. So, just to go back to the previous conversation, so in a sense, they're not facing differential competition in this segment of the labor market, which makes sense because what's happening during this period is women are getting more college degrees, etcetera.

And so where you expect to see the competition happening is for men with college degrees. And that's precisely what you see in this figure here, where it's the real wages of men with college degrees. Those were the ones that were declining. But after the implementation of these policies, the wages of both sets of men go up, but the wages of men with college degrees go up by a lot more.

And so if this was a story of the EITC driving this, we would have expected to see the reverse happen, where it was the wages of men without college degrees going up by a lot more than the wages of men with college degrees. Okay.

>> Speaker 16: Do you see in the society as a whole that women who are married to men who are college educated are more likely to take this policy than not?

 

>> Peter Blair: Can you say that again, please? Women who are married to men who are college educated.

>> Speaker 16: Right. Are they more likely to take this family leave than not?

>> Peter Blair: We can check that.

>> Speaker 16: You seem to be showing an intersectionality between the education we can check and the impact.

 

>> Peter Blair: Yeah, we can check that when we look at the usage data. All right, so I'll turn now to looking at the usage data. So what we've shown you so far is that these family leave policies are passed. We see a stagnation in the rate of gender wage convergence, and that's being driven by higher wages for men, not necessarily a wage cut for women.

What we do now is we combine some survey data from the Department of Labor, and in this survey data, we get to see who's taking leave, we get to see what reasons are they taking leave for? Is it for the birth or the adoption of a child or is it for some other reason?

And then we're going to harmonize these data and we're going to show you results for two margins. So what's the likelihood that you take leave on the extensive margin or that you take leave that's family related or that's non family related? And then also on the extensive margin, conditional on taking the leave, how long do you take?

And are there gender differences? So on the extensive margin. So we're just going to regress whether individual I, in time period t takes leave and we're going to include indicators for race and gender. We're going to control for age, and we're also going to control for, like, what year you're in.

And also we're going to control for whether someone is married. We also do the analysis, not control. Controlling for whether someone's married, cuz you might think that that's endogenous, which it is, but it's not gonna quantitatively affect the results in a meaningful way. And so if you look at everything is going to be relative to white men.

So what you can see is that relative to white men, white boat, white and black women are more likely to take any leave and then they're going to be more likely to take family leave and also non family leave too. If you look at black men, there's not a substantive difference between white men in terms of leave taking.

And so a lot of the action is happening on women taking more leaves on the extensive margin. Now if we look at the length of leaves-

>> Speaker 7: I'm sorry, just the units are for the white females, they're 3.6 percentage points more likely than white men to take leave in any given year.

 

>> Peter Blair: That's right.

>> Speaker 7: Okay.

>> Peter Blair: That's right.

>> Speaker 7: So it's big. These are big differences. And for blacks, black females are huge.

>> Peter Blair: Yeah, and in terms of the, so now we've run a similar specification where we're now focusing on the intensive margin, which is the length of leave taken.

And what you can see is that overall both white women and black women take on average 11 days more leave for overall. But if we focus in on the family leave, this is where the big differences emerge. If you look at non-family leave, there actually are no differences in leave taking between women and men for medical related stuff that's not related to the adoption or the birth of a child.

But it's really all of the action is happening in the family leave. And so this is 36 days. So this is like seven weeks more of leave. And remember, the policy itself is 12 weeks of job protected leave. And so this is pretty substantial. All right-

>> John Taylor: You should wrap up.

 

>> Peter Blair: I have two minutes, so I'm just going to show the last set of slides. So in terms of heterogeneity, we can look at what happens with mothers and we can see that with mothers the rate of convergence is much stronger than for women without children. And also the drop in the rate of convergence after the passage of these policies is a lot more dramatic, as you might expect.

I wanna show you a back of the envelope calculation. I literally have two slides. Where, what we do is we take the point estimates that we have for the rate of gender wage convergence before and after the leave policies in terms of the wage levels. And we project out the pre leave wage rate as a kind of a counterfactual to then look at the difference in the wages that women earn relative to what they would have earned in the absence of these policies.

And then we consider, well, what are the wage impacts of these policies and each year of the policy for a worker who is working 2000 hours a week, right? So this is effectively someone who's working full time. And if you look at these policies in the black triangles, these are the wage impacts for men relative to counterfactual the pre-leave rate of wage growth.

And in the circles, this is what happens for women. What you can see is that for men, over time, like after the implementation of these policies, like ten years out, white men are about $3,000 better off relative to this pre leave rate of wage growth, whereas for white women, it's like they're worse off by about $200.

But that's not statistically different from zero. And so effectively, you have a situation in which this policy is kind of like transferring money to men without necessarily harming women. So you could think about this as like rent extraction from maybe employers. In conclusion, we started out with this puzzle, which was the rate of gender wage convergence stagnated in the 1990s.

We don't know why, based on observable factors like this is not what's driving it. And then what we showed you is that by leveraging variation in family leave policies across states, that we see a pattern of gender wage convergence followed by gender wage stagnation in the shadow of these policies.

And we think that this paper sheds some light on a very important question in economics, which is why the gender wage convergence in the United States stall in the 1990s. And we think it's because of the implementation of these family leave policies. Thank you.