Why Did Gender Wage Convergence in the US Stall? | Peter Q. Blair | Hoover Institution

so we're very happy to have it here uh I respect to us today there's a joint paper with Benjamin I don't know up there somewhere and and it's great to have you here um you're visiting uh as a national I'm usually in Harvard which is great and I was just so fascinated by this economics lab maybe it's not another subject for another day but today we understand it's great topic you're going to talk about why did the gender gender wage convergence in the United States strong it's a great question we're all interested so take it away Peter yeah 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 uh Ben was one of my first PhD students at Clemson he's now an assistant professor at Tim Bonaventure University and he just had a paper that was revised and we submit yesterday so congratulations Ben um 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 and they always show you a trailer before they show you the actual movie and I think that that's really instructive right because like this is an amazing opportunity to like advertise other research uh but but it's more than that I want to tell you about this 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 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 identify 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 skill 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 nine minutes and 29 seconds of George Floyd being brutally murdered and many companies in the U.S said what can we do to address racial inequality in the United States and the traditional answers 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 debris 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 to learn how to read animated 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 do a 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 have stars over the next 10 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 um Pennsylvania did it Arkansas did it Colorado did it so rad 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 um 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 like Billboards like on the highways encouraging companies to tear the paper ceiling if you can Marshal I run the clip yeah bosses couldn't see me as a leader I've run this place for 20 years but I still need to prove that I'm more than what you see on paper yeah this is what I do it's second nature for me coordinating 100 details at once the way my mind works I have a very mechanical brain I sold them on my skills you have to have the confidence to yourself to show up [Music] I never got a college degree it's a day I'm a 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 [Music] you can jack up more information I'll tear the paper ceiling uh for the main event after that trailer so I'm going to talk to you today about a paper why the gender wage convergence in the United States stall this is Joint work with uh my colleague Benjamin posmanik who is an assistant professor it's in bottom Edge University uh this is just a picture of the students in my research group and I like to highlight them because the work this work is done collaboratively across students from several universities and I talked about the Stars work that we have in our pipeline um other work that we've done looks at occupational licensing occupational licensing as a job market signal other work looking at is school spending efficient um Ricky 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 is producing 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 um a large VC here to get some data on company 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 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 and both within colleges but then also outside of the University uh and I mentioned uh 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-prem performance and um 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 line up 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 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 words converge and 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 results 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 tears and so any suggestions that you have for how we can make the paper stronger for the resubmission would be uh super helpful I want you to look at this picture here I'm going to 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 and we're just 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 Raj underage 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 point three percentage points per year and this has been an outstanding puzzle in the labor economics literature for a very long time sure like I'm sorry is this earnings or really a wage rate 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 it's just the cash yes it's not it's not compensation because that's right that's right no no 100 it's mature so the the way to think about this is this is a fact I'm not 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 this is gonna be the question that we're going to try to tackle which is why is it that the rate of General wage convergence stagnated in the 1990s so the the convergence of the reduction in the gender wage Gap during the 1980s is is well understood so there's a decline in unionization there's also a decline in the generic Gap in observable factors like levels of Education labor market experience or reduction in occupational segregation by sex uh and so the question is why is it going to be slower in the 1990s that's the open question and in this paper we're going to argue that the introduction of job protected Family Leave policies is going to cause the stagnation right 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 is that the first explanation is that there's been a convergence in in the occupational distributions in the 1990s this is um by Blau and Khan uh the 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 the other explanation that's been offered is that there was no convergence in overwork but like overtime work between women and men during the 1990s and then the fourth one which we think is is a really interesting idea too is that there's a doc it's been documented by Heinrich kleven at Princeton that the 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 I just I'm probably just confused it sounds like the the last Point goes the other way so the last point if there's a stagnation if there's a stagnation in the mommy penalty that should accelerate wage convergence or am I am I just totally yeah I should let me let me explain that separately so the the penalty that women were facing was declining pretty rapidly and then that decline slowed down in the 1990s that's what kind of applies documents yeah so so if you look at just observable factors controlling from zero 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're kind of 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 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 uh and we're going to argue that it was the introduction of the Family Medical Leave Act which provided job protected leave for for workers and that's disproportionately taken up by by women I'm just a little confused you're you're jumping between levels and trends so uh I want to just let me put it as a factual question there's a there's a gap between potential experience and actual experience between men and women that Gap shrank over time I presume did that Gap continue to shrink at the same Pace in the 90s as it had been shrinking in the 80s or that's you seem to be making a statement about that but I'm not sure exactly what you're saying 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 versus I'm thinking about about the levels what I'll show you so we don't have data we don't have like a precise number in that in the paper itself but we can 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 are now starting to like have even higher levels of Education than men you'd actually expect to see like a reverse a reverse gender gap based on like just levels of education so it's not the case that's a level Point not a not a change in slope point no 100 100 agree so like if the levels are going to be going in the opposite direction we can we can double check to see if the if the if the rates of of those observables could explain us all right so why might [Applause] family leave policy stagnate gender wage convergence we're just going to motivate this conceptually so the first is that 58 in in about 15 of of cases what we see is that when a worker takes a job protected family leave that work is going to be shifted on to other workers who are at the firm or it's going to be shifted onto a temporary worker uh 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 very uh one example of this is a paper that's in the ER in 2018 that shows that when universities implemented uh 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 Monica Thomas in a paper two finds that when the athlete 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 unemployment was substantially lower by about eight percentage points on both the Intensive and the expensive margin too and so there's a growing body of evidence showing that you know offering gender-neutral policies that are differentially taken up by women could could cause a negative impacts on women's labor market progress John give people a bit um women more likely to have children versus women less likely to have children you would think that 50 year old women wouldn't this wouldn't have any effect on them at all because you know they're actually having children yeah so you're saying like looking at women who are like with women who are past kind of like the traditional like child-bearing age that you wouldn't not expect to see as much of an impact on on them right yeah I think we did Ben we we did do this analysis I don't know if we kept this in the paper but we could we could definitely bring it back in yeah well but then they also take up responsibility for their aging parents yeah so if even if you didn't find something that wouldn't necessarily refute the the point yeah and the other part too is that the Family Life policies also cover as adoption of a child and so if um you adopt a child like later on in life that too could be something but General point is is that yeah gender is a a rough correlate of likely to take family leave yeah so um you would expect that better measures of likely to take family leave would then explain why you're seeing this gender and I don't know if you can find any better measures but if you can proves your point that's a good point so you would think age would be one what are some others John that you um you can think of it's not just about there's two effects no I mean like I this is actually really useful because I think we it would be helpful for us to explore some of this heterogeneity one of the challenges is and then you brought this up is that there's there are these kinds of like General equilibrium effects that in a sense it's like it's always like difficult to think about like which individual characteristics you should look at to make predictions about how people are going to to respond but age is one what are others one obvious one is whether they already have children okay family leave policy also also subsidized but you can definitely take care of kids rather than spending more time in the labor market which can have permanent effects yeah your earnings over your entire career so even 50 year old women who are beyond the child varying 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 more these these benefits more than men yeah one of the things that we're going to find so we do do some heterogeneity by whether you have a child or not and what you see is that mother is 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 like much more stagnation after the implementation of these policies and for single women and so that that checks out they were converging much more quickly in the absence of this policy but they start to converge a lot a lot more slowly after these policies are implemented yeah the show the relationship between that sharp line and some of these policies yeah so I'm going to show you that that's going to be that's so let me get a question if you also check the effects of uh this policy on fertility or like did this because the subsidized families or having kids we do check whether it actually had a positive effect on having on them we can do that I want to 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 stagnates what caused this that's what we're focused on and what we're going to show you by using um state level policies that were passed before this is that you see a similar pattern of gender wage convergence followed by gender waste stagnation even for these policies that ruled 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 like why the gender wage convergence stagnate now whether or not the F family impacted fertility and a lot of these other things they're like 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 the takeaway message to be that we have an anti-afamily agenda it just turns out that this particular policy is implicated in this in this question so you described it it would be a level effect not a growth rate effect and it should be quantitatively reasonable so if you know one-fourth of women are going to take uh one a month of annually leave and that's a week a year per woman so then wages should go down by by 150 seconds in a level effect and then to keep growing you build up to some enormous this is like 10 15 to 20 large points at the end and it's a growth effect not a level effect so so there's there's work um for example if you look at the minimum wage literature where that's another context where you might expect to see like 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 um wages are also like 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 um there was a paper I think that was presented this was at the NBA labor studies meaning 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 and so if we've got to the level shouldn't it go back to growing and is it at all does it make sense that it's 20 20 log points in terms of the point estimate so we could 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 10 years out women's wages are about eight percent lower and when you look at the the impacts that we're finding here so if you think about you know like 20 log points over 20 years you're looking at a comparable estimate to what Martha finds in our paper so I don't think that I think that what we're finding here is within the ballpark I'm going to say also to John's question this gentleman's I don't know your name uh point about fertility and and marriage even might be relevant because that could accumulate over time can you say more in a sense once you have a child in the household that you wouldn't have had to work last for the rest you're going to be dealing with that kid for at least 18 years and it may have effects beyond the first it's a good point sometimes you're saying that some of the ideas into having children that you wouldn't have had and it changes the way you live your life yeah 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 FM at least the F me is passed in 1993 and this is going to guarantee about 12 weeks of unpaid family leave to workers and this is going to be for um 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 they're 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 like much higher take up by women uh 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-family variation in in our causal research design let me describe for you the approach that we're going to take in the paper so 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 F family year and 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 athlete 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 weight 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 Leaf policies on 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 in the early 1990s and so you might be concerned about policy endogeneity that their you know welfare reform happens in 1986 the ITC 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 on just generically what happens we're also going to test the extent to which alternative explanations like the eitcol for reform could impact our results we're going to directly control for when uh 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 are 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 and for workers with and we're going to show you that a lot of the action is really for workers with 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 AJ applied looking at Sweden that showed that when they implemented a paid family leave that this increased the gender wage Gap and reduced women's labor Supply the timing is clear pardon the timing is clear you look like a particular event here for the U.S is it timing the same in Sweden no no the timing is not this is the timing is this so in Sweden it was the extension of an existing policy from I think three months to 12 months and so it's like yeah it's like very completely different like 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 that policy is borne by the employer or by the state because that could have an impact I have to double check I don't know if Ben if you remember offhand for the for the ginger doll paper I I don't have that off the top of my head um unfortunately okay but we can get back to you in that because you're absolutely right in terms of like the incidents will matter a ton and in a lot of ways when you think about the alpha Malay The Firm has to keep the job open for the worker and so a lot of the incidents is borne 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 can have a situation in which firms have an incentive to engage in statistical discrimination and that's that that almost like takes us to the conclusion in terms of what are some of the 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 cut off so firms that are smaller than 75 are Exempted from the effling in the publicly available data we don't have access we have like 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 like competitive costs 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 you know a hundred thousand or more so you think thinking confirm that's like 75 to 100 people versus firm that's 500 right is the impact you said to the individual going 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 um whereas for firm one slot is quite significant yeah I guess it would be Citrus paribus I mean because that that at that level they're just they're they're yeah so serious power boost like that that that seems reasonable yeah John do you know how much medical leave the average woman and the average man take how many months a year 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 that's conditional in the breath of a child so the eight percent wage Gap is the average woman is taking eight percent of her time off more than the average man is doing is that is that roughly right can you say that again John you got an eight percent wage Gap the average woman to be taking eight percent of the days off that year more than the average man is that roughly how much they take off so are you saying that like if if statistical discrimination were happening like efficiently in our sense like this it was kind of like it was actually Fair the the that women were being compensated in terms of the time like would this line up yeah we did we did we did we did we did do that exercise and we have something that kind of gets a little bit there at the end when we look at what happens to the difference in earnings over time so um in the first like two years that the policy is in place you virtually see like not a lot of over time you see that the difference in earnings grows to about four thousand dollars over a period of about 10 years and so we could do the exercise where you deflate Bible how many women are taking leave and like how does this matter how does this map into kind of like is this action in a sense like is this actually fear yeah the word fair but does it line up with the Actuarial yeah Ben we did do that we did do that exercise do you remember off the top of your head then um the aren't you talking about earnings you're not wages right yeah you know average person yeah but the average price no wage rates let me clarify this right now because when you talk about wage and most of us who at least do this kind of research would think that would be the hourly wage it is the hourly wage yes it is the hourly wage are not included here no they're not going to be friends just just the effect of that's right because remember like what we're after is we're trying to what we're after is we're trying to understand the extent to which these policies contributed to gender wage stagnation right so this we're starting with this big puzzle we're asking can we Leverage The variation in these policies to see if like after the implementation of these policies generically we see that the rate of gender wage convergence is going to stagnate and that's an empirical question so let me start let me continue on by telling you a bit about the data that we're going to use and then giving you a sense of the descriptive results before moving on to some of the causal analysis we're going to use data from the asex supplement of the CPS and we're going to cover from uh 1976 to 2016 so this will tell you when we started the paper in terms of the sample selection it's going to be the standard things that you do it's going to be people who are of working age um and we're going to focus on full-time workers in part because the puzzle at the heart of this paper has to do with the the stagnation of gender wage convergence for women who are working full-time and we're also going to winterize the data just to keep the middle 99 but like our results are not going to 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 in the levels of experience Etc okay so for our descriptive Mouse we're going to start off like very simply by just running a mincer wage regression where we're going to take the sorry we're going to take the log of the implied hourly wages of a worker in a given time period and we're going to regress that on on indicators for for gender interactive with race and we're also going to include uh coefficients for each demographic group in each time period and so the way to think about these betas is beta 1 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 uh estimates of these betas over time which are going to capture the way in which the gender wage Gap has evolved over time once we account for things once we account for things like education um experience occupation Etc so this is going to be remember that first figure is there any parallel research on the labor Supply good so what I have a I have a slide here where we can show you that if you look at the composition 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 going to be driven by by selection this is just another question that I'm asking about this is not casting any Shadows on okay two defensive but isn't it an equally interesting question about the effect of these programs on subsequent hours of work it's a super interesting question the the question that we are trying to get at with this paper is the gender wage Gap stagnated um during the 1990s and it's we're exercising they're related they're related because there's a 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 yes so that that's the sense in which the labor Supply the 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 and 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 I say so the model that you have in your mind is that 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 like less valuable in a dynamic sense than that maybe but there's I'm thinking even simpler like 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 uh originally that they get compensated for but that alters their their pattern of specialization the household yeah that's that persists throughout the rest of their career I think that strikes me as I think a potential hypothesis that can account for your facts so that could be like a mechanism that could be a mechanism a way to a way to test that is to look at the labor Supply responses all right 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 yeah okay we can now I appreciate that um and I I appreciate you pushing Us in that direction it's it's we have another paper where we look at um what's happening in the part-time segment of the labor market and we've we've decided to kind of split those two papers but I guess you're pushing us to think about this in a more in a more integrated way um there there is as you mentioned there is work on this so Heinrich Levin 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 lose about 20 of their their earnings drop by about 20 after the birth of of a child um in terms of the extent to which this is going to affect specialization there is work so Mary Bertrand like shows that what happens with MBS 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 um career paths and different trajectories and so what so in a sense like what you'd need is that's been happening for time and Memorial and so what you need is the interaction of that with these policies which intensifies the in a sense it provides more Liberty for women to take this path but that in turn kind of creates a lot more 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 but but it's more than just specialization as much as you just have less investment for at any given age they have less experience and less investment and that's a form of specialized investment in the market versus the household you could do without the men you know but you're right you're right there how am I doing that but it gets stronger if there's specialization between you okay okay so a lot time off really going to do that and we look at there isn't an outburst of child you know fertility didn't jump around this isn't about fertility along forever okay um so they so these plans are doing it I appreciate I appreciate y'all um pushing Us in in this direction because we have and that that's 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 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 0.2 percentage points too and so just even descriptively it's not just the case that um there's something magical happening with the covariates that's explaining why we're seeing this appear in the Raw um in the Raw gender gaps all right so just wait with me yes so we it looks very similar for 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 have to you kind of have to pick your your battles okay yeah did you ever thought on that job you're for sure the first chart was white women but it looks pretty similar if we yes the first chart is white women who's the author's prerogative here you define This research to deal with hourly wages of white women and that's what we're going to learn about yes and I think white women are quite like 70 of the labor market and if you go further back um yeah but we do have we do have in the paper companion results for black women as well too but the companion results we'd love to see is hours of work yeah I was working okay not not earnings earnings is the product of hours of working engagement yeah and so then I think if we 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 phobage women I think it's important to situate this within the context of this literature so for 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 uh part-time in a sense like that's going to be like very endogenous to like the family situation and so on and so forth and so like to 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 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 yeah and the impact both exodusly and endogenously on male versus female choice to overwork yeah 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 yeah yeah so you've made it persuasive case that um you know hiring a woman after this enactment of this law you know should should produce a less valuable employee in the eyes of the firm that firms going to make less of an investment so that's going to create um you know a wage Gap I don't understand why the wage Gap is now going to grow you know if what what whatever the forces were that were causing convergence before hasn't changed those forces of convergence as opposed to just changing the level at which uh yeah it's a reason why why there would be you know a gap in you know but I don't see why I mean like this is this is this is a fair point and we got we got 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 when we do the paper so we've approached this from the standpoint of saying um let's Leverage The variation that we do have in these policies which I'm going to show you in just a slide and let's see what happens in Pure Clean right now it's a finding not a not a yes not a theory 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 um example then there's this other feature of if you think about a lot of bonuses and promotions that's in proportion to where your 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 um getting carried forward in addition to this Malita Thomas's work looking at the action Malaysia is that when the F family is implemented that women are promoted at a lower rates too and so this could also this comes back to the point about computer trajectories that you mentioned if a woman gets put on a different career trajectory that can also tilt down um that can also tilt down their wage packs too and just adding to what his the point was you know if you said it 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 but they're you know and and there's only a small number percentage of women going into that into that status each year so then it could lead to instead of a shift it leads to a change in the slope because the slope is controlled for a number of children so you're gonna control it but your regressions are Level effects you could just say the graphs were pretty my regressions are Level so I'll show you what happens to the wage levels as well too in the remaining tournaments 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 1983 and to get away from that we're going to leverage the fact that several States passed a job protected leave policies for um for after women had children and even after fathers had um children as well too and we're going to leverage these sub-sample of states to look at what happens to um gender wage gaps in those contexts this these are going to be the states that we have Massachusetts in 1972 and then you all the way up to Vermont in 1992 and so we're going to be focusing on these maternity policies which in large part many of them so for example New Jersey the law looked identical to what would happen in the afternoon 12 weeks job protected leave and so on and so forth and so you can think about the federal policy this is being precursors to the federal policy we're going to run a standard event study design here where we're going to regress the log of our workers wages on Dummies that are going to be event time dummies so if the laws passed in calendar year why we're going to use that calendar year as being event Time Zero and we're going to look at all of these event study uh coefficients relative to the event year before the passage of the policy so relative to Tau minus one are we going to interact these with 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 we have coverage there the other thing that I should mention too is they're going to be several states are going to pass um uh several states are going to have welfare waivers and we're going to directly control for that in the specification too so you should think about this 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 all right so these are the state level estimates just looking at those 12 States plus the District of Columbia that pass policies prior to the FMLA and this is necessarily going to be noisy because we don't have a ton of of observations of variation but what you can see here qualitatively is that prior to the implementation of this policy you have the steady gender wage convergence and after the policy you start to see the stagnation and we're going to 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 of 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 linked to we're also going to wait inversely by the standard error of each of these Point estimates so that noisier estimates receive less waiting all right there are five observations post event which which continue the growth and then and then I'm going to show you I'm going to show you 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 going to show you now is going to be what happens when we include both the state and the federal variation and I promise you those those regressions are going to come in in just two in two slides so we have an additional 38 events that are all happening in 1993 and so we're going to include those in our event study analysis and this is what you're 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 equal zero year if you had the F family as the first time that a family leave policy is passed we're going to use 1993. and so what you can see here is a similar pattern of gender wage convergence followed by gender weight stagnation after the policy and this is going to be a lot more precisely estimated because we have more events we have more um observations and so to compare these two what we do is we regress the 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 after Malaysia and so think about this as fitting a piecewise linear regression onto those patterns that we saw beforehand now let's focus in on um this event time Trend so this is going to tell us what's the rate of gender wage convergence prior to the policy and 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 going to have um 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 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 like a very similar picture of gender wage convergence prior stagnation thereafter and this rate of gender wage conversions most period is going to capture like what that rate is and so this is going to be I think .03 percentage points after the policies now let me your questions objection okay so to speak we're just fooled by sampling variation yeah when you kill the sampling very variation by extending the sample of then you cleared up okay you validated your assumptions thank you I appreciate that thank you all right good um that is good but you'll now you're leaning much heavily here 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 for in the regression and that had effects that filtered forward in time I'm not I'm not saying I have a solution to that problem yeah I think I think that's right so like the in a sense like the concern with just using the the federal variation if you just thought it was only the FMLA 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 only focus on that and you can see sorry you can see like it's more noisily estimated but like this you know pre-trend is statistically significant the post-trend dropped is statistically significant this difference here statistically significant and the rate of gender wage convergence is like really small and if you look qualitatively at these quantitatively at these Point estimates to they're very similar the confidence intervals are like overlapping too and so what's what's striking about this is that even leveraging similar policy variation prior to the 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 like this is just we're just hands on the table reporting like what we find when we run this all of the the theoretical questions about like 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 um 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 a full-time full-year workers uh this regression is the same event study type setup where we regressed the dummy for like is a person a woman 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 you know 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 um this remarkable break-in Trend that we saw beforehand okay all right so what we're going to 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 of Blau and Khan as well as this paper by doing it all and fundamentally if you think about the data generating process in this context of saying you have some outcome wage of worker I and time period T that's going to depend on observable characteristics X the observable prices of those characteristics beta which are going to be time varying and then on some residual but you can break into two components into like a variance component and then an idiosyncratic component so we're going to think about this uh 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 this decomposition and then what we can do is we can look at the differences in in the gender wage Gap in the pre-period from 1976 to 1992 and we can see how much of that convergence in the gender wage Gap so the gender wage Gap reduces by about 19 log points how much of that is due to differences in observed access 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 of 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 well you can see in this First Column is that the reduction in the gender wage Gap a lot of that is going to be driven by um differences in in The observed axes but then also there's going to 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 they're just just the observed X's is education then what's the measure of experience it's a potential experience trying to construct some out okay so then so then presumably the the potential the actual experience differential which is shrinking that's going to show up in one of the other Rose yeah so this yes that's right that's right um when we look at what's Isaac take a stab it quantifying the difference in actual experience because you know you know the universe of women and you know what share of them are working can you speak up a little bit Isaac so you know the universe of women you know in any given year what percent of them are working how much in an aggregate level and then you just roll that forward and then you can sort of yeah you could treat it as a synthetic panel exactly and then impute the mean hours worked today yeah based on what their synthetic panel unit is yeah we can we can 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 11 and we can look at well which which components are substantially different like almost like a difference in difference in a way and what you can see is that the major the thing that really stands out in a very glaring way is the fact that the unobserved 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 is 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 reduce is being reduced at a much slower rate and so that's kind of like what this decomposition is 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 um 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 how do you think so the way that we got 94 is if you go back to let's see if you go back to here right so like after the implementation of these policies you see a reduction in the rate of generate 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 um 14.74 percentage points and then that relative to this 15 this this 15.6 percentage points it's gonna be 94 percent so the price effect actually goes the wrong direction uh since it's positive the unobserved price effect of the yes the Observer crisis yes exactly yeah so in a sense it's almost like those um let's see yeah so what's happening what's happening during this time period is like women are getting like higher levels of education but then like the returns the returns to those skills are those the returns of those skills that actually push the gender wage Gap in the opposite direction it made it worse 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 like 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 for child care 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 with what's happening with the wage levels these uh 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 downwards and then after 1993 you start to see men's wages like going upwards women's wages by contrast were going up and they seemed to be like unaffected in terms of the path and so even before running a regression what this is telling you is that what's 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 if if the work of women is going to be pushed off to men that you increase men's wages you pay them in efficiency uh it could be consistent with that it potentially could be consistent with other stories too which we're happy to like hear because that would be something that we can include in our paper you're smiling John you have some suggestions which is funny go ahead 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 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 um what are the actual adjustments in the labor market that's driving the overall gender of each Gap here are the results from 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 out so about 10 years later you see it's increased by about a dollar Isaac this is maybe very minor I noticed these progressions you're controlling proactivation there's an argument that you don't want to do that and that like which occupation you and sort of sort of so in in particular various stories of how this sort of plays out yeah would be women are left like to end up managers or sort of end up in different occupations and social condition on that yeah so this yeah so if we don't control for occupation in some sense if you think that there's discrimination on the extensive margin this is understating the extent to which um which that's happening right like I think you can tell stories about to go the other way as well or okay yeah so just yeah that's fair but for consistency in the previous regression we controlled for occupation fix the facts so if we didn't hear then you might wonder if we're trying to hide the ball but arguably the previous progression we also might not want to do it I'm just deserve this in general that's fair that's fair so you previously made the argument that it's I was taught menu 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 well women are entering the labor market and women are getting lots of Education there's a lot more competition for for for these jobs that men in a sense had a monopoly on beforehand there's less occupational segregation so you think their wages are going down it because if their wages can go down why can't women so I'm just trying to use the same thing no that's a fair point that's a fair point I mean like I'm showing you data I'm I'm not gonna I'm not I'm saying like one of the reasons why you might see firms being unwilling to to to cut wages might be nominal wage rigidity now what I'm showing you in the previous page were real wages so I should look at nominal wages and we can say we're nominal weight just declining and if we see for example that nominal wages we're declining then your point completely holds that in a sense nominal wage rigidity should hold for both men and for women or it's being driven by something else I'm asking that is what is that something else that maybe we should be controlling for yeah it's a weekend it's informative either way yeah I think I think that's a fair point and we can look at we can look at the nominal ages yeah but those those were real wages um 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 all right can I have seven minutes this is good making good time all right so now something you might be concerned with is that you know 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 uh the results so what we do is we look at um what we do is we look at um 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 you know just to go back to the previous conversation so in a sense like they're not facing differential competition in this segment of the labor market which makes sense because what's happening during this period is like women are getting more college degrees Etc and so where you expect to see that 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 do you see in the society as a whole that women were married to men who are college educated are more likely to take this policy than not can you say that again please women who are married to men who are college educated right are they more likely to take this family leave than not we can check that you seem to be showing an intersectionality yeah between the education we can talk and the impact yeah we can we can check that if we have the 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 yeah 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 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 a leave for is it for the birth or the adoption of a child or is it for some other reason and then we're gonna we're gonna 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 is there are there gender differences so on the extensive margin so we're just going to regress whether individual I and time period takes leave and we're going to include indicators for race and gender we're going to control it for age and we're also going to control like what year you're in and also we're going to control for um whether someone is married we also do the analysis not controlling for whether someone's married because you might think that that's endogenous which it is um but it's not going to quantitatively like affect the results in a meaningful way and so if you look at um everything is going to be relative to to white men so what you can see is that relative to white men um 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 extensive margin now if we look at the length of leaves I'm sorry this the units are are for the white females they're 3.6 percentage points more likely than white men but to take leave in any given year that's right okay that's right so it's a big these are Big differences and for black black females they're huge yeah and in terms of the so now we 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 both white women and black women take on average 11 days more leave for like 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 like medical related stuff that's not related to the adoption of 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 like 12 weeks of job protected on leave and so this is this is pretty substantial all right um You should wrap up I have I have I have two minutes so I'm just going to share the last set of slides so in terms of hydrogenity 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 with without 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 as you might expect I'm going to show you a back of the envelope calculation um 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 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 in each year of the policy for a worker who is working uh 2 000 hours a week right so this is effectively someone who's working full-time and if you look at these policies in the in the black triangles these are the wage impacts for men relative to the manufacturer of the pre-leave um rate of of wage growth and in the circles this is what happened for women what you can see is that for men over time like after the implementation of these policies like 10 years out white men are about three thousand dollars like better off relative to this pre-leave rate of of of wage growth whereas for white women it's like they're worse off by about like oh like 200 but that's not statistically different from zero and so effectively you have a situation in which um this policy is kind of like transferring um money to men without necessarily harming women so you could think about this as like um rent extraction from from maybe employers uh in in conclusion we started out with this puzzle which was the rate of gender wage convergence stagnated in the 1990s we don't 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-based stagnation in the shadow of of these policies and we think that this paper sheds some like 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 of policies thank you

Channel: Hoover Institution

Views: 3,671

Keywords: Gender Wage Convergence, Why Did Gender Wage Convergence in the US Stall?, federal family-leave policies, wage gap, men vs. women

Id: RhwE8AatDlU

Length: 76min 26sec (4586 seconds)

Published: Tue May 09 2023