Are Connections The Way To Get Ahead? Social Capital, Student Achievement, Friendships, And Social Mobility

>> John Taylor: Thank you all for coming, appreciate it. Good to see you. And it's a great privilege to have Paul Peterson speak to us today. He's the Henry Lee Shattuck Professor of Government at Harvard. He's the director of the Program of Education Policy and Government at Harvard. He got his PhD, maybe I shouldn't say this, in Chicago.

>> Paul Peterson: Is that okay?

>> John Taylor: Chicago, he's written 30 books, that's a hell of a lot. And one of them is with I don't know if he's gonna come, but he's around.

>> Paul Peterson: No, he's traveling as usual.

>> John Taylor: Okay, and anyway, you're a member of the Koret Task Force of K-12 Education right here at Hoover, a Senior Fellow at Hoover.

And the title of the paper, which you're going to focus on, that I hope recirculated is Are Connections the Way to Get Ahead? And you have some subcategories you're gonna cover. I don't know if they're good or bad, but you will find out. This is joint with Angela Dills and Danish Shakeel, so I don't know if there is.

 

>> Paul Peterson: Danish is here.

>> John Taylor: All right, is she here?

>> Paul Peterson: Danny is here. Hi, Danny. On the screen, Angela can't join us, so I want to acknowledge my co-authors who played important roles in helping put this paper together. So, thank you, Danny. And thank you to this workshop, and thank you, John, because I was motivated to write this paper because I attended this workshop three or four years ago.

I'm not sure exactly what the date was, but somebody presented what I'm gonna call the Chetty paper. The Chetty paper was written by 23 people. I'm not gonna mention their names. You can mention a couple. And so, I don't want to privilege any particular. So I'm just gonna call it the Chetty paper, but maybe, we don't know, when you have 23 names on a paper, exactly who wrote what and so forth.

But I'm gonna call it the Chetty paper, cuz it's the first name on there and it uses a lot of his thinking. So, and really, I wanna say at the very beginning that the Chetty team was extremely generous in making available all the data that they used for their paper.

All the data that's at the county level. They had some data at the zip code level that they've not shared, but they made all the county level data available to the research communities. So anybody can go in and do what has been done here. There's no private information that's being exploited here.

It's all publicly available information, and it's to the credit of Facebook and the Chetty team that they have made their data available to others to look at.

>> Paul Peterson: Yeah.

>> Paul Peterson: This paper, that was written by the Chetty team, says that cross class friendships, or what I'm gonna call bridging capital for reasons we'll get to, it really is what's important in life.

If you build a connection with people who are not of your social class but are a little bit above, this is the way you're gonna get ahead. It's more important than anything else, if you're interested in intergenerational mobility. Now, as you know, Chetty has tremendously high-quality data on income of everybody who pays taxes, as of 2015.

So that information is included in the database that is being used here. And this is what their study concludes. The share of high SES friendships among low SES people, is strongly associated with upward income mobility, whereas other forms of social capital are not. Areas with higher shares of such friendships have large, positive, causal, and I want to underline the word causal, effects on children's prospects for upward mobility.

So that is probably the key finding in the paper. And this paper has had a widespread acceptance in the popular media. The New York Times says, this is an expansive new study based on billions of social media connections. It helps to explain why certain places offer a path out of poverty for poor children living in an area where people have more friendships that cut across class lines significantly.

It significantly increases how much they earn as an adult. So, it's just a very good summary of exactly what the study says. And the Brookings Institution, a couple of folks there said, this is sure to have a huge policy impact, because it shows that cross-class connections boost social mobility more than anything else.

More than racial segregation, economic inequality, educational outcomes, and family structure. I mean, these are bold claims. These are not little minor-league statements.

>> John Taylor: Can you go back to that, cuz they have it exactly backwards, though. Go back. Okay, so you're saying cross-class connections boost social mobility more than anything else?

Racial segregation is supposed to boost class social mobility?

>> Paul Peterson: Well, their measure of racial segregation is the degree of racial segregation within a county, and their measure of cross-class relationships is the number of friendships you have with people other than your own social class.

>> John Taylor: Racial segregation that reduce social mobility?

 

>> Paul Peterson: Well, it might retard it, actually, if you look at the data, it has very little effect. If you look at, if you look-

>> John Taylor: Integration is what the word they were-

>> Speaker 3: I see, is that right? When they say segregation, they mean integration.

>> Paul Peterson: No, this, well, they're talking about the degree of racial segregation, probably the right way.

The Brookings people probably should have said, the degree of racial segregation, which is relatively insignificant. And we don't wanna say bad things about Brookings.

>> Speaker 3: Right, well, I'm not trying to say anything. I'm just saying the English language says they're comparing cross class connections, boosting social media more than anything else, including, racial segregation, as if racial segregation possibly boost social mobility.

That's the English.

>> Valerie Ramey: Well, it's an expositional error.

>> Paul Peterson: Yes.

>> John Taylor: They're meant to say, racial integration, is that right?

>> Speaker 3: The variables that they're considering, they're saying racial segregation is a variable.

>> Valerie Ramey: It was a careless way to say-

>> John Taylor: This is gone more than we need to go.

 

>> Speaker 3: I mean, it just seems so ridiculously counterintuitive.

>> Paul Peterson: I would not, I would not want to be responsible for that statement. So it is our argument, not that we're gonna solve this question, but that we're gonna offer a second opinion. He's making causal claims. He only has descriptive data.

We can't improve on that. We can't go beyond descriptive data, though. Maybe you'll have some suggestions as to what to do next. And the study has huge policy implications. We can talk about that downstream, but you can see what they are, because if friendships is what counts, then you really want to create context where those friendships can occur.

And there's a lot of policies out there that are driving that. So the principal concern that we have is that the variable that they are using to measure cross class friendships is endogenous, has an endogenous relationship with social mobility. For the simple reason that as you climb the social ladder, you are going to get more friends from a different social class.

I was pretty far down the social ladder when I was a child. I probably, you people are all famous rich people. So now I have a lot of cross class friendship. So I know Morris Fiorina over there and he's a really upper class kind of guy, so that's so.

He didn't cause me to rise up the ladder. That's the consequence of it. So one of the biggest problems with this whole paper is that something that could easily be a cause of social mobility or a consequence of social mobility is interpreted as cause of social mobility.

>> Speaker 5: I just wanna clarify two things.

I totally get what you're saying. However, there's another component, which is the following, which is, I could have a lot of skill personally. And have highly placed friends who for reasons like we work together can observe my skill. Or I could have no skill with highly placed people and with reasons we work together, they can observe that I have no skill.

And in that world what we will find is social mobility. But it's because of the fact that I have the skill.

>> Paul Peterson: You're anticipating our argument very nicely.

>> Speaker 5: But there's one last step. What if I have the skill but I don't happen to work with the highly skilled people?

 

>> Paul Peterson: Well, we will get to that. So there may be some place for. I'm not saying it has no effect. I'm just saying that the claims that are being made here are extreme claims that this is the most important of a factor. It's not that it's one factor of many, but that it's the most important factor.

So anyhow, so we do have to take into account that we don't have this thing nailed yet. It's only descriptive analysis. We're very much sort of limited by the fact that we are using Chetty's data, and Chetty's data has certain limitations. He's measuring relative mobility, not absolute mobility.

Some of you might think absolute mobility is far more important than relative mobility. That's sort of my own view on it, but that's what we have here, relative mobility. We only have county level data. We don't have individual level data. Maybe county level data is enough. We can discuss that, and we have to assume that social capital is sticky, and we'll come to just how severe that assumption is.

So his measure of intergenerational mobility, this relative measure of mobility is what is the percentile in the income distribution of those born between 1978 and 1983 of parents at the 25th percentile of the socioeconomic distribution. So that's a handful there. But basically, he's looking at the income distribution in 2015 because that's when he has the tax records for the adult population, and he's working now with those between the ages of 25 and 40.

So he knows what their income is in 2015, and he knows which counties people live in. And so he knows in which counties people have an income at a certain percentile. What's the mean percentile by county? So now this is what he has. He shows that the county mean of those who were born into households at the 25th percentile of the SES distribution, it should be at.

It's not at or below. It should be at the 25th percentile of the SES distribution. And he shows that later in 2015, in terms of the income distribution, the mean is 41% for that population. And that has a range. There's 30 percentile point range across counties, and there's a standard deviation of four percentile points, which isn't that large.

 

>> Speaker 3: We've regressed most of the majority of the way to the mean.

>> Paul Peterson: Pardon?

>> Speaker 3: Does the county mean these people that he's measuring in 2015? These are the people who were born in 1978 to 1983.

>> Speaker 5: 35 to 40 now.

>> Paul Peterson: Yeah.

>> Speaker 3: The counties they live in average a mean of the 40th percentile.

And their parents have been in the 25th.

>> Paul Peterson: Yes.

>> Speaker 3: Okay, so that's 16 percentile points out of 25. All right, so it's 64% of the regression of the mean. Okay, if you wanna look at it that way. Right, so that's already something, all right.

>> Paul Peterson: Could be, yeah.

 

>> Speaker 3: All right, I mean, I think if I understand what he's doing.

>> Paul Peterson: Yeah, so we're saying that really it's families, communities, student achievement that's the driving force and that these cross class friendships are less important.

>> Speaker 5: Is it a stupid question to ask who went back to the 25, right?

So today there's somebody the 25th percentile who was at the 25th percentile. Those are old people.

>> Paul Peterson: We know that these people, when they were born, they have selected out those who were at the 25th percentile of the SES distribution. They come from families who they have arranged because they have the tax records for the parents.

But for some reason they're using the SES of the parents, not the income level of the parents. They're using just their standing in the SES distribution and they're selecting out those who were at the 25th percentile of the SES distribution, and they're telling us that-

>> Speaker 5: They're now in the 41st percentile.

 

>> Paul Peterson: At the mean.

>> Speaker 5: And the question is, who's now replaced them? Somebody from the 41st or something from the upper went down. And the question is, is it just old people getting poorer?

>> Paul Peterson: We don't know, we don't know. But that's the problem with relative mobility, is that when somebody goes up, somebody else has to go down.

 

>> Valerie Ramey: Following up. It would be interesting to pluck out the 75th percentile and run their same experiment and see what happens there. Is it a symmetric effect? Is it something else going on?

>> Paul Peterson: That's an interesting thought, and I don't know if the data allows us to do that, but we should.

I don't think we have that information, but we can certainly check to see if we do have that information. That's an interesting thought.

>> John Taylor: There is a hypothesis, some data, etc., suggesting that with a large increase in female college attendance, that the degree of positive assorted mating in society has increased.

So I'm not sure if she did that experiment that we'd get the same 64% now.

>> Paul Peterson: You'd have to get that data at the county level, which maybe you might be able to do.

>> Speaker 3: Can you say precisely what they view as the difference between SES and income?

 

>> Paul Peterson: Well, SES is generally, the standard way of doing SES is to take into account parental education, parental income, and parental occupation. That's the, I think, the original way in which SES was calculated. Then they found out that occupation is very much a function of income and education.

Once you know income and education, you can predict occupation pretty well. So occupation doesn't add much, and it's very difficult to get reliable measures of occupation. So most people now use income and education as a way of estimating SES. Chetty doesn't say in this paper exactly how he constructed that, or probably I should say, I am not exactly sure how Chetty constructed that.

 

>> Speaker 3: It would be good to know what was going on there.

>> Valerie Ramey: The reason not to use income is the same as the permanent income hypothesis. What you think matters to these people is permanent income, and there's a lot of transitory variation income, so you're just gonna get noise there.

 

>> Paul Peterson: Right, and that could be, that's something more with respect to the other side of this equation. What is the income measure there for those ages 25 to 44? Now, you could say the age range is sufficiently constricted so that the income in any one year is a pretty good estimate of permanent income, but it's going to be less than perfect.

Yeah, yeah.

>> Speaker 3: Yeah, I mean, we might prefer consumption or something, but on that basis, but I think the point, Valerie, is I'm suspicious that adding education, weighting current income and education is much better than current income.

>> Valerie Ramey: Don't they use income of the parents, do they? Some measure, I mean, they appealing to the-

 

>> Paul Peterson: I'm not sure that he, he must not have income of the parents, but there's a lot of stuff in those various papers that they've written, which is buried away, and maybe we should be sure exactly what he's doing there if he's reported that out.

>> Speaker 3: So, I mean, it seems to me he has information on the parents, and usually what these people are doing is matching tax records, administrative data, some other survey data or something.

And so the question is what?

>> Paul Peterson: All of his information comes from 2015, but I think he has information on the parents as well as the subjects.

>> Speaker 3: Yeah, you would expect him to have information on the parents somehow linked with Social Security numbers or something like that, or some other way from tax records 25 years ago or something like that, right?

I mean, otherwise, how is he gonna know they were in the 25th percentile?

>> Paul Peterson: All right, so these are all good issues, marching ahead, they do not include state fixed effects in their analysis. And this is a consequential decision.

>> Valerie Ramey: Why?

>> Paul Peterson: Why, they never say, it is not said in their paper at any point whether or not they should use fixed effects.

I have received comments from their, I shared it with the Chetty team and I received comments back from them in which they said, well, you're using state fixed effects and you're eliminating some of the variation, right? True, but if you don't use state fixed effects, you could be attributing to anything, stuff that could be for any number of factors that can vary across state lines, the most important being the difference between the north and the south and east and the west, the more settled areas and less settled areas.

There's just so many differences across our states that you would have to have a lot of theory to figure out what to do, which you can escape doing by using state fixed effects. And now you're gonna be working with less variation, but at least half of the variation still remains within states.

So half of the variation is between states, half of the variation is within states, roughly speaking.

>> John Taylor: Well, another issue is that if you ignore where people are living, other than identifying them as in a county that has some characteristics, you're ignoring large differences in the cost of living.

So looking at nominal income can be very misleading, unless you have, there's something called regional price parities that deflate-

>> Paul Peterson: Right, but when he's talking about mobility and he's talking about percentile, he's talking about the national percentile. So he's looking at where you stand in the national income distribution, yeah.

 

>> Speaker 5: What's the variation in social mobility coming from? Because I'm gonna play devil's advocate, I assume they would say, no, no, no, wait a minute, states vary in how people are connected. The south has a particular social way of doing things, the north has a different way of doing things, and we want that variation in order to prove our point.

So what is the source of the variation in social mobility?

>> Paul Peterson: Variation in social mobility is the variation in the percentage of those who were from homes at the 25th percentile of the distribution, who are at wherever they are in the income distribution between the ages of 25 to 45.

 

>> Speaker 5: Sorry, I said that wrong. Connections, what's the variation in connections?

>> Paul Peterson: The variations in connections, that is coming from the Facebook data. They have 44, is it billion or million? I've forgotten whether it's millions or billions, but anyhow, it's an incredibly large number of observations on Facebook, which allows them to say, who are you friends with?

And they take the most frequently mentioned friends, and I think you have to have mutual friendships, and their friendship measure is all very cleverly done. So they have a very good measure of friendships, and then they know the social class, not only your social class, or your reported social class on Facebook, but also that of your friends.

 

>> Speaker 5: Further your point, clearly, there's going to be a relation between. So let's say you're in a community which has a lot of social clubs, rotary clubs, local country clubs, whatever.

>> Valerie Ramey: Churches.

>> Speaker 5: Then Facebook, churches, right? And Facebook is gonna be less important for you as a social media than if you come from a city where there are no social clubs.

And that's your only source, which I think supports your point, because

>> Paul Peterson: Maybe, but I'm taking the Facebook data as valid data on friendships. It may not be, this is a standard criticism of this study, is that they're relying on Facebook data, and Facebook is only an indicator of friendships, and it may be a misleading indicator.

We don't know how reliable an indicator it is of true friendships. There's a lot of friendships that are not true, but they've done quite a bit of grinding away at the data to try to make sure. And one of the most interesting things they do is they ask people, were you friends of this person in high school?

So, Facebook knows whether your friends were your friends when you were in high school. So we're gonna be able to isolate friendships that were formed in high school. That's gonna be pretty important for our analysis, because friendships formed in high school are clearly, well, not clearly, but very probably antecedent to social mobility.

So if you take friendships formed in high school as your measure of friendship patterns, then you have probably a pretty good something that you could say is causal, because it at least is happening prior to the formation. Before you have the outcome data on income mobility, however, you gotta remember that the adult friendship pattern is observed in the year 2022.

So they're looking at Facebook data in the year 2022, and they're using it to predict mobility in 2015. So they have to assume a stickiness of their friendship data, so that what you're measuring in 2022 is also a good measure of what happened by 2015.

>> Valerie Ramey: I'm not sure about the argument that you give in their favor of the high school friendships, because if I think about my high school friendships, it was the people on the debate team and the honor society.

There were a few football players in there, but not that many. Cuz high schools, when you start forming, you start specializing in friendships in some ways.

>> Paul Peterson: Well, yes, your friendships in high school could be a function of your family background, of your level of achievement, but at least is antecedent to the outcome, what your income is gonna be between the ages of 25 and 44.

 

>> Speaker 5: You could do better if you had it, which is see what sports teams they play together in high school, cause there you really do think there isn't the IQ effect that you would have with your friendships outside of that.

>> Paul Peterson: Nice idea, we would have to have the Facebook data itself to work with.

But all we get are these precoded variables at the county level. We're not able to go back in and redo his analysis from the bottom.

>> John Taylor: All right, why don't you proceed? Go ahead.

>> Valerie Ramey: Actually, I was on the girls basketball team, so I did have basketball players.

 

>> Paul Peterson: Okay, so here the Chetty team is aware of the potential for endogeneity. And so they have three arguments for why this is a causal relationship and not an endogenous relationship. And the first argument we've already discussed in part, and that is that, well, if you look at the high school friendship patterns, they highly correlate with the adult friendship patterns.

Correct, that is true. And the school ones are antecedent to mobility. So pretty true, probably true, maybe true. And however, they don't use the school variable in their analysis. Once they say that the school variable is a really reliable variable, why don't you use it instead of sort of saying, well, we've now proved that we can use the adult one.

You haven't proved the adult one, you just have proved that there's a sort of a 0.6 correlation between the two, very nice. You really wanna use the best variable, so we're gonna use the school variable. So the second point is, well, we see this mobility correlation with friendship both in predominantly white counties and predominantly black counties.

To me, I can't make any sense of that claim because I don't see that that proves much of anything except that probably this thing that they're talking about is not driven by segregation.

>> Speaker 5: And I think it would have to come to the conclusion that there was no racism during that time period and heaven forbid they come to that conclusion.

 

>> Paul Peterson: So anyhow, it just seems to me to be beside the point, kind of point. And so then finally they say, all right, we see this for those who don't move from their birth county. So if we just limit our analysis of those in the birth county, but they're still measuring this friendship relationship in 2022.

So it can be endogenous for that group just as easily as it can be endogenous for any group. So I don't think any of these arguments that they've offered, there's a lot of statistics thrown at you and a lot of fancy terminology, but when you start really boiling down, you really can't see that they've got much of a defense of the fact that this is an exogenous variable.

And then certainly we've talked about the state fixed effects. I'm gonna show you what happens when you put fixed effects in and when you take it out of, we can come back to that. Now, our theory is that there's all kinds of social capital out there, the most important, probably, is the family.

James Coleman was at the University of Chicago when I was there. And so I breathe the same air as, I have breathed the same air as James Coleman, so of course, this is the best. Then, of course, there's Putnam, who's at Harvard, he's not nearly as good as Coleman, is this recorded?

He's very, very good, almost as good as Coleman, sorry, let me correct. And then there's this. Really unknown product of the Joint Economic Committee that was released in 2017. And that has a lot of data on social capital, and we're relying a lot on the work that they have done to measure social capital, and then there's trust in government, which we're calling political capital.

And then we have some data on student achievement that's taken here from the Stanford Data Education Archives, or it's called Education Data Archives, SEDA. And then, of course, we have this bridging capital, which Granovetter in a famous article in 1973 said, cross-class friendships form bridges to the outside world and are a tool for social mobility.

So that's basically what Chetty does, is he draws on Granovetter's theoretical insight from the past. So then we look at these things, so those are the variables that we use in our model, and that's the theory that we're relying upon to generate the variables. And all of the indexes that we're using have been used by others, developed by others, except we've modified it in order to get the right time frame.

Then we throw in a bunch of control variables, and the control variables we use are pretty much the same as what Chetty does. Median household income, degree of racial segregation, the Gini coefficient for the county, the percent black in the county, and then they have a measure of third grade math, which isn't very good.

And we use a better measure, grades 3-8, the SEDA data, and we're gonna turn it into one of our main variables instead of using it as a control variable and not that that makes any difference in a regression. So the percent single parent family, we're gonna use our family capital index, which is a three variable construct, but it's not much different.

So then they use weighted data, we use unweighted data, it's not gonna make much difference whether you weight the data or not.

>> John Taylor: Why do you use unweighted?

>> Paul Peterson: Why do we use.

>> John Taylor: Why do you make a difference here?

>> Paul Peterson: Well, every observation is an independent observation on the degree of social mobility in that county, why do you wanna say, we're gonna weight by population, by the size of the disadvantaged population?

So they're gonna weight by the size of the disadvantaged population, so they're gonna give more weight to observations that are based upon large counties with lots of low income people. But it seems to me every county is of equal value in estimating a relationship. So I just don't know why they wanted to use weighted data, but I'm quite willing to check it out and see if that's a key assumption, and it turns out it's not a key assumption.

 

>> Speaker 3: Is the variable, the percentage of people who move or, in which case, I think.

>> Paul Peterson: The dependent variable.

>> John Taylor: What are we trying to look at.

>> Paul Peterson: Dependent variable is the percentile in the income distribution in 2015 of those who were born between 1978 and 1983 and were at the 25th percentile of the SES distribution parents were.

Yeah, it's a very good question to keep asking because you got to keep your mind focused on exactly what it is you're trying to explain here is the county variation. In some counties, you see a lot of people moving up the ladder as so defined, and you find other counties where there's not much movement, but that's what the measure of social mobility is, yeah.

 

>> Speaker 6: Just very briefly, what's the measure of racial segregation? Do you know why they controlled for that? I mean, it's not obvious to me why you need to control for that.

>> Paul Peterson: Well, they make a big deal of it in their article, their article goes on about racial segregation at great length.

And in fact, mainly their point is friendships count, not the degree of racial segregation, and I don't know why they think it's so important, pardon?

>> Speaker 6: To compare the coefficient of the control variable?

>> Paul Peterson: Very small, the racial segregation, if you look at it in a bilateral relationship, it's not very strong, and if you look at it in a regression equation, it's not very strong.

But it's a very politically, captures a lot of attention, yeah.

>> Speaker 5: What is the measure? They look at the county and they figure out, is there more concentration of black areas and white areas, is that what they're doing?

>> Paul Peterson: Well, racial segregation within a county would be the degree to which black people are concentrated in certain zip codes within the county as compared, yeah.

 

>> Speaker 6: Geographic information.

>> Paul Peterson: Yeah.

>> Speaker 7: I have a question, so I'm thinking, going back to grad school days, whenever one wants to measure the impact of social connections and outcome, there is the usual reflection problem. So it is really the group of individuals other than me affecting my own outcome or through my interaction with them, I affect them mine, and therefore the outcome.

What is their approach to this, or they ignore any of this? They want to know imagine whether.

>> Paul Peterson: It seems to me that looking at it at the county level solves some of that problem, because you're really looking at a pattern of social connections that exists within a county.

Is that pattern of connection such that it facilitates social mobility for those who are at the 25th percentile of the distribution. And so that could be because others are moving down or because you're moving up, or it could be interactions or all kinds of complex, but at least you're picking up the fact that it's a set of social interactions is not an individual characteristics.

So I actually think that's not a bad way of measuring social capital at the aggregate level rather than at the individual level.

>> Speaker 7: So the question, sorry to follow up, I mean, we're gonna get there, you mentioned that from a policy debate standpoint, when you go to the determinants of the value of social capital, I mean, these questions come back to the forefront.

There are individuals who are exceptionally good also at forging connections in addition to having many other skills that are valuable later in life from an income and social mobility point of view.

>> Paul Peterson: Yeah, okay.

>> Speaker 7: You mentioned.

>> Paul Peterson: Hold that, let's, we'll come back to that, so for we our measure of family capital is the density of two parent households, we have three indicators of that, but we wanna see the more two parent families in a county, the more family capital.

If you go back to Coleman, he talks about the importance of the family, not just in terms of parental income, parental education, but also the parental interactions with the child and that social engagement within the family. And he shows that's a very important component of why the family is so important for downstream consequences of the individual, really a very powerful theoretical argument and simply measured by this one indicator which you can look at from different kinds of perspectives.

 

>> Speaker 6: Just a teeny point, in Murray's book where he goes into a lot of detail on this, he makes a big fuss that it's not simply two parent households. It's two married parents with genetically connected to the kid, and every other form is a junk.

>> Paul Peterson: Yeah, well, not junk, but is inferior.

And so I would agree with what you're saying. There's a new book out by Melissa Kearney that's very persuasive on that point. I think any two parents are better than one. But I agree with you that if you have two biological mother and father married together, that's the best you can do, right.

 

>> Speaker 7: And it's predictive of stability of marriage.

>> Speaker 5: It's amazing, that jump. I don't know this, but in that book there's a jump that doesn't make any sense. You go from two married, but not genetically connected jump.

>> Valerie Ramey: Well, the famous stepmothers take their stepchildren to the dentist less often than biological mothers do with their ones.

 

>> Paul Peterson: All of this is true.

>> Valerie Ramey: There's a reason for those Grimm's fairy tales.

>> Paul Peterson: Cinderella is real.

>> Speaker 3: Maybe we would show a stronger relationship if we had a more refined measure.

>> Paul Peterson: So going on to community capital, yeah.

>> Richard Sousa: In this community capital, I don't care anything about education, particularly higher education.

Is that in there somewhere?

>> Paul Peterson: This is a good point, Richard. We built our model thinking we're gonna match Chetty as closely as possible. And he left education out. I don't know why he left education out, but about a month ago, I said, he's left out education. We've got a stick in there.

So we've done a robustness check. But once you've done all the other stuff, you throw in parental education, it doesn't make much difference. We're not getting a lot different.

>> Richard Sousa: So much of parental education, more as a college or university in the county. In an example, let me give you an example.

Cal Poly is in San Luis Obispo County. The next county up is San Benito County, no higher education. There are community colleges, but no college in that community. I would think, and I think I've seen data that say kids tend to end up close to where they go to college.

And particularly when you talk about social mobility, one of the most important things, obviously, is college education.

>> Paul Peterson: I like that idea. You could almost add that to the model and see whether or not you're gonna get. I said you could add it to the model. It sits out there in space, and you can just add on to it, yeah.

All right, so moving on to community capital. We are using the JEC data, the Joint Economic Committee, because they start talking about religious congregations as being an important base for social capital. So we don't wanna exclude other forms of community organization, so forth. But anyhow, that's an indicator that has all of these factors built into it.

And then for political capital, we use voter participation rates, online census participation rates. That turns out to be pretty insignificant in almost every estimation, so it's not worth a lot of discussion. And then we use student achievement as measured by the CTA data, yes, Michael.

>> Michael: Just a quick question on the student achievement data.

I know the SEDA data also has a growth measure. I'm assuming you're using the proficiency measure, but I wondered if you'd also-

>> Paul Peterson: No, we just use the standard level measure. We're not looking at the growth measure.

>> Paul Peterson: Because we're saying, what's the level of cognitive ability as observed at that point in the life cycle?

So this is the equation, it's very simple. There's nothing fancy about it. It's just sort of saying all these variables, our particular variables, and we do not include in our preferred model this endogenous variable, the adult bridging variable. We say that's endogenous. We're gonna see how much we can explain simply by using the school variable.

The school bridging variable, the friendship. Pardon?

>> Speaker 10: You have adult bridging in the equation.

>> Paul Peterson: It's in there. Yes, it's in that one because we start with that, but we drop it out. Yeah, we do it both ways, with it and without it. But you're right, our preferred model does not include that.

 

>> Speaker 6: Could do an instrumental variable because you might believe that school bridging is predictive.

>> Paul Peterson: Yes, we do include the school bridging one, but not the adult one. We have both adult and the school. Well, I'll show you the results. You'll see what we get. So our model is similar to Chetty's in that it has the same dependent variable, the same control variables, except for the little adjustments.

I've talked about the same regression model, the same counties. Chetty excludes small counties because he says he can't get a good measure of racial segregation in small counties. So he's dropped about, I think he ends up with 1800, which is about half of all counties. So that's actually a very interesting restriction.

And you might argue that we should go back and expand it to all counties and drop the racial segregation variable, see what happens. But we haven't done that. So the differences between the models is we've added these additional explanatory variables and we've included state fixed effects. But we're trying to keep it as close to what Chetty did as possible.

Okay, but we do have to say that social capital is sticky. There's a lot in the literature that talks about the stickiness of social capital. We can talk about that literature more later on if you want. But basically, these are the variables and here's the dates. So for the family structure, we're looking at the data in 1980.

Cuz the group that we're talking about were born 78-83. So we wanna get the family. We're assuming that the family has its biggest impact early in life. And so we are trying to get. We have to work with census data. So the choices between 1980 and 1990, we're going with 1980 for that.

But for the community variable, we're assuming adolescence is the time when the community is playing a greater role. So we're using data that's around 1990. And we're doing that for the political data as well. The achievement data is what's really requires the sticky assumption because we can't get achievement data before 2009 and that's too late.

We really would like to have that in 2000, in 1990. And so we're about 20 years off.

>> Speaker 7: I can ask you.

>> Paul Peterson: I wish we could get the earlier data, but we can't.

>> Speaker 7: I'm wondering conceptually, do you really need stickiness? Or if you think about a persistent is cumulative, the social capital building process.

So you maybe don't observe the same social connection 20 years later, but those who you forge in an important point in life, in your personal development, have an implication.

>> Paul Peterson: But you have to assume that the counties are gonna stay relatively in the same relationship in the hierarchy over time.

 

>> Speaker 7: When I put it away-

>> Paul Peterson: But like in Italy, the southern Italy has no social capital, northern Italy has a lot of social capital, it's been that way for 400 years.

>> Speaker 7: Well, they have their own social capital, I would say.

>> Speaker 7: No, I'm not actually, some argue it's very resilient.

 

>> Paul Peterson: It is very resilient. So that could be an issue, here's our results.

>> Speaker 7: One more question, going back to the IV question. Aren't they worried about the observational data? One argument for an IV or control function type of approach is imagining this could be an error-invariable type of model.

And so just to clean the measurement error on the right-hand side variable, you may wanna control for something else.

>> Paul Peterson: So what would you do for an IV?

>> Speaker 7: I have, but only at the county level, there's nothing individual level. I'm wondering if any of the characteristics of the parents in the tax data somehow aggregated to be predictive of the right-hand side variable at the county level.

If the first stage relies on variables that are available at a more granular level than the geographical level of the county. That could be predictive of the county level outcome that you're including on the right side.

>> Paul Peterson: So something that will predict the county level on each of these variables.

Variables, yeah.

>> Speaker 7: That would be a valid IV.

>> Paul Peterson: What would that be?

>> Speaker 7: In principle, characteristics of the parents.

>> Paul Peterson: Characteristics of the parent? Education, achievement of the parents.

>> Speaker 5: Cuz they're still not perfect, genetically related.

>> Paul Peterson: Doesn't that have a direct impact on the dependent variable too?

It can only have an effect on the independent variable without having any direct effect on the dependent variable.

>> Speaker 6: Can you think of some average to the parent that is unrelated to IQ?

>> Speaker 7: That's not variable at the county level, yes.

>> Paul Peterson: What I was trying to do was think of some exogenous event that destroyed social capital, like sort of the China shock or something like that.

And see whether or not-

>> Speaker 5: It happened in Poland, if you wish to try to get Polish data. Did the communism completely destroy the social capital in the country? It could, but that's a very different project.

>> Paul Peterson: Very different.

>> Paul Peterson: So in any case-

>> Speaker 3: There was a really deep recession of 1982.

I don't know if that 1982 deep recession that was differentially impacted different counties, you're saying?

>> Paul Peterson: Absolutely.

>> Speaker 3: You get unemployment rate by county, for example.

>> Paul Peterson: If we could see, I mean, that's what I'm looking for something like that, that's might be. Unemployment rate peaked in 10.8% in December of 1982.

And undoubtedly there's a wide variation by county because it was much stronger in Rust Belt counties, I believe, than in the average county, things of that sort. So maybe the unemployment rate by county, it might not be a strong enough instrument, but it might be something you think of.

 

>> Valerie Ramey: So what is it instrumenting for?

>> Paul Peterson: The 82 recession.

>> Valerie Ramey: No, no, no, that's the instrument.

>> Paul Peterson: It's instrumenting for some of these things that were some of the characteristics.

>> Valerie Ramey: But your exclusion restrict for something to be a valid instrument, it must be uncorrelated with anything else that can affect the left hand side variable.

 

>> Speaker 7: Okay, so what about the NLSY? Because of course we need to think about data that is available and is granular enough at that geographical level. So that rules out PSID, that rules out, see, but the NLSY has very accurate measures of young adults characteristics, AFQT scores.

>> John Taylor: And that is more correlated to with achievement and conditionally uncorrelated with social mobility.

I mean people with low and high AFQT scores make friendships.

>> Valerie Ramey: Or the ad health survey about who's friends in high school.

>> Speaker 6: I've got an instrument for you, as long as there's enough variation in the county, the variation in the schools of amount of sports played. If you believe that-

 

>> Valerie Ramey: Where would you measure it?

>> Speaker 6: Well, I'm assumed every school will list the sports programs they have in the county. So you could, and if there's variation in that. And if you assume that your friends on the sports team is not highly correlate to your IQ, it is a little bit, but that would say that would definitely a shock to friendship.

 

>> Valerie Ramey: Yeah, see, a shock to friendship.

>> Speaker 6: You need a shock to friendship, right. You need a shock to friendship, uncorrelated IQ. But that would be, how many sports teams there were, or the sports participation, maybe that's a sports participation in the county.

>> Speaker 3: I would like to see a shock to family life and to see whether or not, because our big variable here is the family.

 

>> Speaker 7: No, no.

>> Paul Peterson: Certainly a deep on comment would be a shock to the family. Another thing, that-

>> Valerie Ramey: High school basketball team at my high school.

>> Richard Sousa: Title IX.

>> Paul Peterson: High school basketball player.

>> Valerie Ramey: Title IX, which made things equal for women, came in the 70s and I was on the first girls' basketball team at my high school.

And Betsy Stevenson has some nice work looking at the effect on later life outcomes of title IX. But anyway-

>> Speaker 3: We could use Title IX as an instrument maybe.

>> Speaker 6: Yes, I remember when Raj came and gave it an early version of a paper here at Hoover, and in that, he seemed to regard social mobility as an unalloyed good.

So let's set aside the fact that for everybody that moves up, somebody's got to go down. But then the question was, if it's an alloyed good, what are the policy recommendations? One that was clear in the data he presented was, move to Canada, because their social mobility was quite clearly better than was the case in the United States.

If family is the important variable here, looks as if it might be, does that mean choose your parents better? Or is there some policy intervention that we could recommend that would improve kids' chances to in life-

>> Paul Peterson: Well, thinking on social mobility, relative social mobility, is that you would not want a society where you had total randomness, because then parents would not invest in their children's future.

We want parental investment in their children's future. So if it were completely random, that would not be Pareto optimal. But at the same time, if you said, we want a caste system where there's no social mobility at all, nobody from the bottom can move up the ladder. Everybody's gonna stay where they are, as we have had in societies in the past, and we think maybe a lot of that.

Well, that's not something we would want. So we want some, but not too much. And what the optimum level is, I have no idea. So what you do observe in industrialized societies is that it doesn't vary that much from one country to the next. If you look at all the European societies, you compare the United States, the degree of relative social mobility is roughly the same, maybe the Danes are ahead of the rest of us, but it's pretty similar.

So you might even sort of say, there's an equilibrium that societies tend towards in terms of relative social mobility. But for the purposes of this paper, we're just sort of saying, okay, some people think we should have more social mobility. If that's what you're interested in, then the question is, how do you get it?

Do you get it through friendship, creating more of these cross-class friendships? Or do you get it by creating family life where you have more two-parent families and you have community supports, yeah?

>> Speaker 3: I wanna come back to these beautifully exogenous, to this macroeconomic shocks that they have had a deep effect on this in many dimensions.

I mentioned unemployment, Valerie doesn't like that, but it doesn't think it's a. In this period, we had horribly high inflation in the first two or three years of what you're talking about, continued high inflation for the rest of the 78 to 83 period. We had this horrible recession following a deep, but brief recession, and there was a great increase in migration because of that, which would clearly disrupt family ties and all this stuff that goes on in the schools, right?

Well, they were born then, so there may not have been school. But it was a big disruption and differentially affecting different counties and people moving from many out-migration from some, in migration to others. So I'm wondering if there's any way for you to advantage of that.

>> Speaker 6: Yeah, I wanted to follow up.

Michael's moved on. But, I mean, you said invest in the children, and this is perhaps one way in which the family is important. Now, that's a policy area where the community can, in fact, do things to invest in children when the family may not. And that would equally apply whether or not you've moved down or up, because if you've moved down and that is a consequence of the fact that the family failed to invest in the children, then that's something that perhaps society can actually work on.

I think what we mustn't lose track of here, and I've been involved with this in China a little bit, is what we really want is everybody to move up. So we're looking at a situation in which you want income per head to go up for everybody. And then within that, I agree with you, social mobility is a limited good, but we've gotta be aware of the fact that for people who are affected adversely by social mobility, and that is they're moving downwards, then there's gotta be remedies for them too.

 

>> Paul Peterson: Yes, no, I don't disagree that absolute mobility is probably more important than relative mobility. And I also agree with you that the community can affect the family, and this is the measures of the density of two-parent families in a county. So that's picking up two things, one, family's willingness to stay together and community supports for families to stay together.

So that family variable is being measured at the aggregate level. And we know from other work that there is an effect of the community on whether or not families are likely to stay together. So that's right. Now in terms of the endogeneity or trying to solve that problem, that period, 1980, if we can find a big variation across counties, using that as an instrument, certainly strikes me as something to explore.

 

>> John Taylor: Outer type instruments, but what you use is the China shock is exactly that. But whether it violates other features we would like for an instrument-

>> Paul Peterson: Well, yeah, it's gotta have an effect on the dependent variable via the independent variable without having any directive.

>> John Taylor: Yeah, absolutely, may I explore that.

But anyway, I have a more expositional question, how are we to think about the relative size of these coefficients you've got? Obviously you've got these things being quite significant.

>> Speaker 3: Why don't you explain this table a little bit?

>> Paul Peterson: Well, I would say the correlation between family and social mobility is pretty high, and the correlation between the school bridging variable and mobility is fairly low.

 

>> Speaker 3: Could you explain what this regression is, what the data is? Could you explain a little, we don't know, go ahead, tell us a little about the equation?

>> Paul Peterson: Well, the equation looks at each of these variables and controls for all of the control variables that we listed.

And so that all of control variables, together with these interesting social capital variables, account for 82% of the variance in social mobility within states. Because this includes state fixed effects, I haven't shown you the size of the state fixed effects. So what you see here is controlling for all of the control variables, which included family income, percent black, Gini coefficient.

So then the question is, of the remaining variance, how much is accounted for each of these variables? Now you can unpack this and look at, well, to what extent is achievement a function of family? So we're looking at the effects of achievement after controlling for the effects of family and community.

So they're all given equal opportunity to influence the outcome variable. And it turns out that the connections variable is pretty modest, it's not zero, but it's pretty modest, yeah, in the back?

>> Speaker 11: What is one unit increase?

>> Valerie Ramey: Yes, that's what I was gonna ask. No, no, what does it mean to have something be 0.44 for family?

Does that mean that you add a parent and you get 0.44? I mean, you can't compare these coefficients unless you have standardized them.

>> Paul Peterson: Well, these are beta coefficients.

>> Valerie Ramey: Okay.

>> Paul Peterson: The percent of the variation, these are not b coefficients.

>> Valerie Ramey: So that's why John was asking you that cuz we weren't clear on that.

 

>> Speaker 7: It's an ANOVA decomposition.

>> Paul Peterson: Yeah, these are all normalized.

>> Speaker 7: Good.

>> Paul Peterson: Yeah.

>> Richard Sousa: So, Paul, I wanna actually ask you about the unit of analysis here a little bit. So I don't know the Chetty paper, I mean, I know of it, I haven't read it. But I presume he's got really granular data.

I know you only have county-level data, so there's some ecological inference issues potentially going on, but you have what you have, has he done analysis at the individual level? We talk a lot about counties, whether state fixed effects should be included. I'm sort of, what's so special about a county, other than that you have the data aggregate to a county?

Why not something like commuting zones, which is supposed to be more of a measure about kind of, how people live their everyday lives, who they bump elbows with. That already the county is just. I'm fascinated by that, again, I know that's the data you have, but does Chetty talk about this at all?

If we could get-

>> Paul Peterson: He has zip code data, and what he says, you have to take his word for it, but there's no reason for him to mislead us. He says he gets the same results, not significantly different, when he does his analysis at the zip code level.

So, I'm assuming that that's true. He didn't share the zip code data. He claims that he's gotta respect privacy. This is, maybe he does, I mean, he's got all these tax records, so there are.

>> John Taylor: A lot of restrictions.

>> Paul Peterson: Yeah, so that's probably correct. So, we have to use the county, I don't think you'd get wildly different results if you used commuter zones.

You've got 700 commuter zones. You got 1300 counties. My guess is you're not gonna get wildly different results by. If you do it by commuter zone, though, we could combine counties into commuter zones.

>> Richard Sousa: Well, just one brief suggestion. You might think about putting a dummy variable in for whether the county and the school district are perfectly aligned, because obviously in the south, as you know, the school districts are the counties.

But in other parts of the country, those things overlap pretty radically. And if we think some of what's going on has to do with what the local schools have provided or not, it might be good to use that.

>> Paul Peterson: Sure, I don't disagree with that, and I don't disagree with the fact that, all of these measures are sort of imperfect proxies for what we really wanna get at that.

But I still think it's terribly informative to see how significant each of these variables, no matter how imprecisely measured, turn out to be. The community variable is a little disappointing. And maybe that's because we don't have a very good measure of-.

>> John Taylor: Why don't you show me the tables, the other two tables?

 

>> Paul Peterson: You wanna see some more tables?

>> John Taylor: Just two.

>> Paul Peterson: So, here we're trying to predict adult bridging capital, and treating that as the dependent variable. What affects, and here school bridging variable shows up as a very good predictor of adult bridging is what Chetty said, and we find exactly the same thing.

But the interesting thing here, is that the kids who are smarter in school are more likely to be able to have cross-class friendships as adults, which is exactly what I would have thought, that if you're a smart kid, you're gonna have cross-class friendships as an adult. So, and actually, the family's not so important in this estimation.

And the family turns out to be much more important for social mobility than for cross-class friendships.

>> Speaker 12: It seems to me the interaction term is really what you're interested in, I mean. I don't think any of us believe that if you're not very smart, having smart, having rich friends is gonna make any difference.

It's the people who are smart, in the lowest classes, having high-class friends. It makes a difference. So why aren't we looking at that interaction variable?

>> Paul Peterson: We could, that's an interesting idea. All right, now here we have the adult bridging capital now included in the equation. That's what you've been waiting for.

Yeah, so now the school bridging capital falls apart, which doesn't mean that it's not a causal variable, but it's all absorbed by the adult one, which is endogenous, in my view. And, the family variable, interestingly enough, just survives, even when you put the bridging variable in, and the others survive at a lower level.

So, okay, so you don't care about the policy conclusions. Right, you know what they are. You know that. You know what they are.

>> Speaker 12: You're missing a very important one. Change the bloody tax laws. Make it that it's advantageous to get married and have children.

>> Speaker 7: And not-

 

>> Speaker 12: I mean, it's nuts.

>> Paul Peterson: They sustain dual parent families.

>> Speaker 12: So okay, but our tax laws, I mean, we're not encouraging it, and we know this is.

>> Paul Peterson: That's very important for tax laws to be doing that. Yeah, so this is our conclusion, but we want to see my robustness checks.

John is very interested in the robustness checks. Here's the weighted data. This is what they prefer. But we use our model, and we lose ground on the family and gain on the achievement, school bridging disappears. The others disappear.

>> Speaker 6: Why is it negative? Who knows? Who knows? I don't like this weighted data, because you're looking only.

You're weighting the places where there's a lot of poor people. Why are you doing that?

>> Valerie Ramey: So there's a wonderful paper by Gary Solon and co-authors called Why Are We Waiting With The Weigh? It's just fantastic, short read. You'll never make a mistake again if you read that carefully.

 

>> Paul Peterson: Do you agree with me?

>> Valerie Ramey: Yeah, yeah, no, I mean, that's why every time I try to make a decision, I go back to that paper.

>> Paul Peterson: All right, so I wanna show you no state fixed effects. So, here our school bridging variable becomes more important. I think there's a lot of difference between the South and the North, perhaps in the friendships that are established across class lines, and maybe in other parts of the country as well.

So the school bridging variable, I think, is contaminated here because there's other variables out there that differ across states that are affecting it. To me, the interesting thing is that the family variable survives, the political variable, for some strange reason, becomes important. This is a bad model. I mean, this is not the right model.

I'm just showing you this model because Chetty wants to see this model, and if I wanna get this paper published, I have to show it, but I don't like it, unless you give me a good reason to use it.

>> Speaker 7: Two-quick comments. The first comment is that, IV is a clean way, but if we can leverage the vast body of evidence about how achievement works, what's the technology to produce achievement in kids and younger adults?

You could model a first step as, that was the control function suggestion earlier. If we have confidence from time use data, that the use of certain inputs and the time used in child rearing at those stages in children's life are predictive of achievement, their census data and time use at the family level.

That could form the basis of a proper first step through which you could predict achievement and clean the variable that personally is the one I'm most worried about. They're taking it as exogenous is a bit of a stretch. And without an IV, you could predictively account for any variation in achievement.

It is correlated in the average term of your main outcome equation. The second common suggestion, there are two very good paper, very many excellent papers by Jim Heckman and co-authors, but there are two in particular that have become very contentious lately. It is Cameron-Heckman, and they made the point in the JP in 2008 and 2011.

That if you look at various measures of constraint, at the family level that prevent completion of grades from middle school to high school. Even lifting constraints through income support type of programs, nothing helped as much as the family characteristics. They were established wisdom, and this paper fell out of favor for a healthy debate that we're having now.

And I'm wondering what your thoughts are, whether Chetty relates his findings to those, cuz I think that important trail fell cold for reasons that you can imagine.

>> Paul Peterson: Right, we do-

>> Speaker 7: And the point was made very, very persuasively.

>> Paul Peterson: Yeah, you're saying we should use some predictor variables for achievement and include them in there.

 

>> Speaker 7: There's 15 years of literature that tells us what's the technology to produce achievement in children, and most of those variables are available in time use data now.

>> Paul Peterson: So good suggestion.

>> John Taylor: There is an interesting parallel on the argument about fixed effects, totally changing your interpretation. In this case, it's country fixed effects by Daron Acemoglu et al criticism of Michael Kremer's work on institutions and growth.

And it's an easy read, it's a short criticism, especially the first part, which is mostly about fixed effects, which Kremer doesn't include need.

>> Paul Peterson: So I'm not sure what to do with that suggestion.

>> John Taylor: Well, I think it's a good explanation of the trouble you can get in of not using fixed effects.

And in another

>> Paul Peterson: Eliminate some of the variation.

>> John Taylor: I understand all that, I'm just saying, if you're looking for another citation of another case among famous people where an art debate was settled by including fixed effects, there it is, that's all.

>> Speaker 11: Yeah, that's great.

>> Speaker 12: Do you have a conclusion?

 

>> Paul Peterson: Yeah, so backing up here, it's very broad.

>> Valerie Ramey: I like that.

>> Paul Peterson: Do you want the policy?

>> Speaker 6: Okay, could you leave that up there for long enough for me to read past it's too long.

>> Valerie Ramey: Are you trying to hide something in that previous one?

>> Speaker 6: The got some, you put policy up.

 

>> Paul Peterson: Very likely it's not who you know.

>> Speaker 3: They wanna end tracking and all that sort of stuff. They wanna end tracking. They wanna just not let talented people maximize their talent.

>> Paul Peterson: Yeah, if you want to do what Chetty wants you to do, you want to eliminate all this separate learning tracks in school, you want to abolish Stanford.

You want to get rid of merit-based entrance requirements for high schools. You wanna make sure that everybody is in contact with everybody else all the time. That's the way you get cross-class friendships. If that's the way you wanna maximize it. If you really think that it's important that people have learning opportunities, you might not wanna do those things.

And you might want to support community institutions and help keep two-parent families together, yeah.

>> Speaker 5: I will say that, another way of looking at this is we kind of tried the first one, right? I mean, and you can say, well, how well the country work out after Johnson versus before Johnson.

 

>> Paul Peterson: That's a fairly.

>> Speaker 5: That's a very broad thing. But this idea that, well, we don't have any evidence other than the granular evidence.

>> Speaker 7: What are your thoughts, Paul?

>> Paul Peterson: What are my thoughts?

>> Speaker 7: What do you take home?

>> Paul Peterson: About the policy implications? Well, I think we've got the right explanation.

I think we may not be right on every detail, but I think the broad, you'd like to have causal data, but if descriptive data that's well thought through can be pretty powerful, too. And my thought is that we've got a lot of evidence here that early investments in life, as Huckman would say, have downstream consequences.

And if we're gonna wait for adult friendships to form to solve our social mobility problem, if it is a problem, and to some extent I think it is, well, then, indeed, we should attend to basics, not to sort of ephemeral things like who your friends are in a given afternoon.

 

>> John Taylor: It would be interesting to take a look at what if you were focused on absolute mobility, and to what extent there's substitutes or complements, relative mobility.

>> Paul Peterson: My guess is, if we put in absolute mobility, we would get something similar. I don't think the relative versus absolute mobility by county, just making a wild eye guess at it, where you have relative mobility, you're probably also having absolute mobility.

 

>> Speaker 3: Well, it would be good to know.

>> Paul Peterson: Yeah, we could, I mean, that's a tractable problem, one could do that.

>> Speaker 3: I mean, a really big question, which Jonathan's trying to get at with his pre and post the great society stuff. Is to what extent, if we're interested in mobility, absolute and relative mobility, not just as a county level, but for people are substitute to compliment, which is kind of in some sense the deepest and most important question, right?

Should we be concerned that China became much more equal when a quarter of the population was listed out of abject poverty? That's the next stop. Would we have better off if we hadn't created a bunch of tech billionaires, etc?

>> John Taylor: Paul, thank you.

>> Paul Peterson: We will try.