Lecture 06 Higher Education and Upward Mobility

Lecture 06 Higher Education and Upward Mobility

October 17, 2019 4 By Ronny Jaskolski


– Okay, great. So that brings us to the end of the first major part of the class on
equality of opportunity. Now we’re gonna transition to
the second part of the class on education which is of
course gonna be closely related and build on what we talked
about in the first part because as you all know, education is widely
viewed as one of the most scalable pathways to upward mobility in the United States and elsewhere. We talked about it as an
important correlated factor in the previous lecture, but as we’ll talk about
in these coming lectures, there’s a strong view also that education, it’s not just important in
terms of predictive power and differences in mobility, but it’s really one of the key things that we can potentially change. To the extent we think
things like social capital, family structure, segregation, all these other factors might
matter for longterm outcomes, they may well be as important or maybe even more
important than education, but education has the virtue
that we for a long time thought about how we
might improve education, change educational policies and so forth. It’s more in the policy space. Now that said, there’s growing concern that education no longer
provides a strong pathway to opportunity in the
United States in particular and there are many pieces of evidence. You’ve probably read
about a lot of this stuff in the media that people talk about. So people worry about the
fact that kids in the U.S. nowadays typically perform
worse on standardized tests on average than in many European countries despite the fact that
we actually spend more on our schools per capita than
many European countries do. So that’s just a statement
about the levels. Average performance doesn’t
seem that great in U.S. schools if you look at standardized tests and compare to other places. There’s also concerns about variation within the United States. So there are very sharp differences in the quality of schools
across school districts, across different cities
and we worry in particular because of the local
property tax finance system that kids from disadvantaged backgrounds may not have access to the same quality of
K through 12 education so those first two points
are about the K through 12 elementary education space. Analogously their worries about
the higher education space as well as many of you
will be well aware of so rising costs of college. People worry about a lack of access for low income students
as a result of that and then people also worry that while we have increasing access to certain types of
colleges in particular, there’s a tremendous growth of for profit institutions
in recent years. Maybe those colleges are actually not producing
very good outcomes and not actually that
helpful for the students who are attending them. So lots of I think potential
in the education space, but also lots of concern
and increasingly so that we may not be really
delivering on the idea that education is the great leveler that creates opportunity for everyone. So big picture question
we’re gonna talk about in the next few lectures, how can we try to improve the
education system in America. Traditionally… People have been thinking
about these issues for a long time. Traditionally measuring
the impacts of education systematically was pretty difficult because you didn’t have
really good data on outcomes to see what different types of educational institutions
were really doing. Now as we’ve talked
about in other contexts, administrative data, in this case from colleges
and school districts, are really giving us a much
more scientific understanding in the recent years of
what economists would call the education production function. How do we produce education
and produce good outcomes using education as
effectively as possible? So what we’re gonna do these next lectures is kind of work our way backwards through the education system starting with the higher ed system and then go to the K through 12 system. And so in this lecture in particular I’m gonna focus on a paper with my colleagues called
Mobility Report Cards: The Role of Colleges in Intergenerational Mobility so that’s on your syllabus
and can be a useful reference for what we’re gonna cover here. Okay so the question we’re
asking in this paper, the first question we’ll
tackle in this topic is just a very big picture question, how do colleges shape
income mobility in the U.S. or how does the higher education system affect the level of
intergenerational income mobility in the U.S. So the answer to that question
is not totally obvious and the reason is that in principle higher education can provide
a pathway to upward mobility that is attractive in the
sense that it’s not directly shaped by the neighborhood where
a child happens to grow up. So the great promise of
higher ed in some sense is you may have happened to
have been born in a family in a neighborhood without
a lot of resources. Once you get to college, maybe
that really changes things. Open doors and allows you to break free from whatever circumstances
you happen to be born into. So that is in principle a possibility. However if in practice children
from higher income families tend to attend better colleges, then the higher education system may not actually promote mobility. If the kids who get to
go to the best colleges are all the ones who come
from the best neighborhoods and the highest income families, it doesn’t actually really
change things that much. In fact you can show that
colleges could actually increase the intergenerational
persistence of income that has reduced mobility
across generations if the disparities in
college attendance rates and the types of colleges kids from low and high
income families attend are sufficiently large. Basically you amplify the
advantages potentially that high income kids had to begin with if there’s a lot of stratification in the higher education system. So it’s really an empirical question for which you need a lot of data to figure out how colleges are shaping intergenerational mobility
and how potentially we could change admissions policies or other types of policies
to amplify the impacts of the higher ed system on mobility. So in order to make progress it’s useful to have the following
conceptual framework in mind. The effect of the higher education system on mobility across generations fundamentally depends upon three factors. You can write this out mathematically or you can just see the logic of it. It’s quite straightforward. So the way to think about
mobility is there are a set of inputs so who is showing
up at every college in America so take Harvard students for example. If we think about mobility at Harvard, one of the things that matters is what is the income distribution of the people who are coming in. So what are the parent income distribution of students at Harvard
look like for example? If there are very few low income students at a given college, then that college cannot
be in some accounting sense contributing a great deal to mobility because if we think of mobility as rising from the bottom of the income
distribution to the top, if there are no low income
kids in certain colleges, obviously they can’t be having
a big impact on mobility. So that’s the starting point,
the inputs into each college, the parental income distributions. Okay so then to think about mobility, it’s not just about who comes in the door. It’s also about how they end up doing after they go out the door. And so the second thing you wanna look at are students’ earnings outcomes conditional on parent income by college so is it the case that at
Harvard or at any given college lots of kids come from low
or middle income families and end up rising up. How do kids from low
income families at Harvard end up doing five, 10
years after graduation? So that’s the second thing that matters. The third thing that matters again comes back to the keyoshoe of correlation versus causality. We need to understand the
portion of the variation in the student’s earnings outcomes that we see across colleges that is due to colleges’ causal effects. So we’re gonna see… It will not surprise you that
outcomes of Harvard students look great in terms of earnings. Is that just because of
who gets into Harvard or is that because of some
causal treatment effect of attending Harvard? If it’s all about selection, then changing which students
attend which colleges may not actually have a
very big impact on mobility if there’s no actual treatment effect of attending a different institution. Okay and so that third piece
is also gonna be critical in terms of understanding
how this all plays out in the aggregate system. So what we’re gonna do is talk
about the empirical evidence on each of these three
elements in some detail and then we’re gonna tie them all together to think about how higher
education affects mobility and in particular what would happen if we had different
types of admissions rule so hypothetically if we
moved to an admission rule where it was based say
purely on SAT scores or if there was class based
affirmative action for example, giving an advantage for kids
from lower income families. We’ll be able to simulate out how that would affect
segregation at colleges and how that will affect income ability once we have these three inputs. But the first thing we’re gonna do is concentrate on measuring
these three things as well as we can with modern data. Okay so how do we do that? So in the paper that I’m discussing here, we estimate these three sets of parameters for every college in America using data covering all
college students in the U.S. from 1990 to 2013 so that’s
about 30 million students. Okay and so how do we get that data? We combine information from three sources to construct an anonymized data set. So the first is you need data on income. You’ll be familiar with where we get that. It’s the same source as what
we’ve been talking about in the previous lecture so we’re gonna get parental income and student
income from income tax records. Then we need to know where everybody in America
is going to college. So how do you figure that out? Turns out that the IRS receives records from every college in the
U.S. like Harvard for instance will report the social security numbers of all students who
attend Harvard to the IRS in order to have a double reporting system when parents claim tax
credits or deductions when they send their kids to college. Okay so the IRS wants to
make sure that parents are not making up the
fact that they’re sending their kid to a college and so they have a double reporting system where the colleges also report which kids are attending colleges and so you end up having this information. We also supplement that with data from the department of education
on who got a Pell Grant because there are some coverage issues at the bottom of the income distribution if you just use what’s called
the IRS 1098-T tax data so you can put those things together. All this as you can imagine
is a fair bit of backend work with various government agencies, but you can put those things together and basically get a complete roster of students at every college in America. And then the third piece of
information we’re gonna use is we have data on everyone’s SAT scores from the College Board which
is gonna be very useful in figuring out the
causal effects of colleges as you’ll see when we
get to the third part. Now one important note before
I get into the analysis, you’ll see why this ends
up being quite important, all of the statistics we’re
gonna show you for colleges are based on whether you
attend a college or not, not necessarily whether you complete. That is whether you graduate. So at Harvard that distinction
is not that important. Almost all kids who attend
Harvard end up graduating, but at most colleges in the United States, especially two year colleges,
completion is a huge deal. Many, many kids don’t end up completing at lower ranked schools
and that ends up driving some of the effects that we’ll see. Okay so any questions on
the basic setup of the data before we start talking about results? Questions on what we’re trying to do here? Yeah? No, no, you… In fact you cannot pay the government to get access to these data. There’s complicated backend set
of bureaucratic arrangements as you can imagine to get this sort of data and work with it, but you work with it in collaboration with people at the government and internally at government
agencies, but it’s not… You can’t buy everyone’s tax records by paying the government
as you can imagine. Yeah, anybody else? Okay so let’s start with the first piece, parent income distributions by college. Alright so what are we
actually measuring here? So we’re gonna define parent income as average pretax household income during the five year period when the kid is between ages 15 and 19. So you can measure
incomes at various points of a child’s life. It doesn’t actually end
up mattering that much. What we decided to do
here is measure income around the time that kids are applying to and attending college. That seemed sensible when you’re looking at
college attendance issues. Like what we did in the earlier
studies we talked about, we’re going to focus on
percentile ranks so we’re… Exactly as like we did before. We’re gonna rank parents
relative to other parents with children in the same birth cohort. So if take all the kids
who were born say in 1980, there are about 3.5 million of them, rank their parents from one to 3.5 million based on their incomes. Convert that to percentiles. Okay give you a sense
of dollar magnitudes, we’re gonna talk about
people in the bottom fifth, the top fifth, the top 1%. Just wanna give you a feel
for what those numbers are in terms of dollars so this is just plotting
the empirical distribution of income in our data. And so you can see the 20th
percentile is $25,000 a year. The 80th percentile is $111,000 a year so that puts you in the top
fifth, being above $111K. To be in the top 1% you
have to be making more than about half a million
dollars a year on average over this five year period
that we’re looking at. There’s been as I showed
you in earlier lectures a significant increase
in inequality over time so this is for kids born
in the 1980 birth cohort. If we look at kids born say
in the 1995 birth cohort or say your year, that 99th percentile cutoff
would be more like $650,000. So it’s gone up by about 150K. We use different… We use the percentile
cutoff that’s relevant for your birth cohort to account for these changes over time. Okay so the first thing
to say before we start to get into the data
about specific colleges, the first simple point
is that just whether you go to college or not is strongly related to
your parents income. So this is just plotting
college attendance rates versus parent income percentile. Near the top of the distribution virtually all kids of the
highest income parents go to college like a 90%, 95% rate, but then if you get to
kids born to families at the bottom of the income distribution, it’s something like 30% go to college and this is going to college by the time you’re in your early 20s. So some people will go to college later on in their life and so forth. It’s not a terribly huge number, but the point is at least the
traditional way you’d think about traditional college attendance ages, there is a dramatic gradient here. Something like a 65 percentage point gap between kids from the
lowest income families and the highest income families. So that’s just whether
you go to college or not. What I think is actually more important and actually drives some of
the issues I’m gonna talk about in an even stronger way is
that the types of colleges low and high income kids go
to tend to be quite different and so that’s what we’re
gonna talk about next. To get into that, I first
wanna just do a poll to get a sense of people’s
perceptions of these issues before showing you the
data for specific colleges and so given that we’re here at Harvard, let me ask you the following question. How much more likely do you think you are to attend Harvard if your
parents are in the top 1% so earning more than
$500,000, $600,000 a year versus the bottom 20%? So there are your choices. Twice as likely, five times, it goes on. Give you a minute to fill that out. What did you say? 100+ times, there is no… Yeah, there’s… It’s one of these I can tell you that. Alright it looks like we’ve
got a distribution there that centered around 60
or 80 times which is… That’s a pretty high ratio. So let’s look at what the data
actually have to say here. Alright so this is the
parent income distribution of students at Harvard
first just by quintiles. I’m gonna get to the top 1% in a second. So this is just very simple tab. Take all the students who
are classes of 2002 to 2004 at Harvard so this obviously
a little bit earlier, actually around the time that I was a student here in the college. But as I’ll get to in a second, this has not changed
tremendously over time. So 3% of students come
from the bottom fifth, 5.3% from the second quintile,
70% from the top quintile so obviously if you were drawing uniformly from the U.S. income
distribution random sampling, you’d have 20% in each of these. So perhaps not surprisingly,
you have significantly more high income kids than low income kids. 15.4% of kids are from the top 1% and so what that means, answer to your question
is that your probability of attending Harvard is 103 times higher if you’re from the top
1% than the bottom 20%. So how does that work out? The top 1% is a group
that’s 20 times smaller than the bottom 20%, but the number of kids who
are at Harvard from the top 1% is five times more than the number of kids from the bottom 20%. So five times 20 get you to about 100. So Harvard is enormously skewed
in terms of representation towards kids from very,
very high income families and that’s typical actually
so I’m showing you the data for Harvard just because it’s of interest given where we are, but if we now look more
generally at Harvard and its peer institutions so look at other elite private colleges, this is showing you the same kind of data that I showed you on the previous slide, but just horizontally. So bottom quintile is that
dark red bar on the left, top 1% on the far right and it’s the shares of
each of those groups. You can see it’s basically the same at Stanford, Harvard,
Yale, Princeton, et cetera. Now one thing that’s somewhat interesting is if you look at MIT and Cal Tech, in particular MIT and Cal
Tech have significantly lower top 1% shares as you can see than Harvard, Stanford, Yale and so forth. Those are also the two colleges that don’t really have legacy admissions so that I think is part of different feel in terms of what the
admissions rules look like at these institutions, but broadly the big picture pattern is the representation of low income kids at all of these institutions
is pretty modest. So to show you that now in
a little bit more detail, let’s pool together a set of colleges that we’re gonna call the Ivy Plus, a little bit arbitrary, but
basically take the Ivy League plus Stanford, MIT, Duke and Chicago and now show the income distribution. Sorry, the percent of students
coming from each percentile of the parental income distribution rather than the five quintiles
as we were doing before so just more granularity. And what I think is interesting about this is I had expected before we
started looking at these data that you’d see a pretty steady increase once you were at the 75th,
80th, 85th percentile, but it’s really not the case. It really only shoots up
when you’re at the very, very upper tail of the income distribution. If you’re relatively affluent, you have slightly higher odds of being at elite private
colleges in America, but it’s not dramatically
different from the middle class. It’s really only when you’re
in the half a million dollar type range that you see dramatic changes in kids’ chances of
attending these colleges. Okay so we’ve been focused on
the elite private colleges. Let’s now expand that to
look at other colleges and broaden the discussion. So this is the data I showed
you before for Harvard. Here’s UC Berkeley where I
was previously a professor and you might’ve thought
intuitively UC Berkeley, state flagship school, one of the most famous state
schools in the United States. Intuitively you think
of a place like Berkeley as having maybe much more access for low and middle income kids and you can see that that’s somewhat true. Berkeley does have fewer kids from high income families
than Harvard does, but it does not have a
tremendously large number of kids actually from low and
middle income families and that’s true more broadly of many elite state flagship schools. Now a school that looks fairly different is the State University
of New York Stony Brook where there you see that the green bars, they all hover actually
around the 20% mark. You’re seeing an almost equal draw from the income distribution
so that looks pretty different. And then here’s a fourth one,
Glendale Community College, a really big community
college in Los Angeles where you see the opposite pattern. A greater share of students
coming from low income families than from higher income families so opposite sign gradient. In many community colleges
you’ll see a larger share of low income students than
higher income students. Okay so what you see, I’m giving you four examples here. We have this data for 2,500
different colleges in the U.S. It’s all publicly available. You can look at it online. And there are a bunch of tools now to explore this type of
data in an intuitive way, but the big picture I think point of this is that there are really sharp differences in parental income
distributions across colleges. To put it differently, there’s
significant segregation across colleges in America by income. Just like we talked about earlier, there’s significant segregation across neighborhoods in America. Okay so coming back to what
I was saying at the beginning to motivate this discussion, if you think about higher
education as providing ladders to upward mobility and that’s particularly
likely to be the case if access to higher education institutions is similar for kids from
different backgrounds, at least at first glance
the state has addressed that you’re not in that
world where you have kids from very different backgrounds
attending the same schools. And so to try to quantify
that a bit more precisely, it’s useful to compare
the degree of segregation across colleges to the
degree of segregation across neighborhoods. So when I show you that data, you can clearly reject the view that there’s no segregation
across colleges. Clearly it’s not a uniform
income distribution across all colleges, but just how much segregation is there? Is it large, is it small? Should we be worried about it? It’s not totally clear. And so there’s been extensive discussion. We all experience in kind
of our day to day lives how segregated American cities are so you know that there are
affluent neighborhoods, there are less affluent neighborhoods. If you’re in an affluent neighborhood, you don’t come into contact
in many American cities with lower and middle
income people that much. There’s a common perception I think, maybe many of you had this view when you were coming to college, that college is a place where
you’re likely to meet people from much more diverse backgrounds
than where you grew up, in particular much more
diverse backgrounds, but we’re focused on
here socioeconomically. There’s this idea that there’s
gonna be greater interaction between kids from different
socioeconomic backgrounds than in the neighborhoods
where they grew up. So is that actually the case? So we can assess that in the
following very simple way with these data. Let’s think about the following
simple summary measure of the amount of segregation. So what we’re doing here
is asking if you’re a kid whose parents are in the bottom quintile, so you yourself come
from the bottom quintile, what fraction of your peers
in the college that you attend come from the bottom
quintile, the second quintile, the third quintile, the fifth quintile. So again if kids were uniformly
mixed across all colleges, this would be 20% in all the bars. So what is happening here as you can see, you have in your college much fewer peers if you’re from a low income family. You have many fewer peers who
are from high income families than who are themselves
from low income families. Alright and so what’s the
simple logic for that? If you’re a kid from a low income family, as we saw earlier you’re
much more likely to say attend a community college or some sort of lower ranked institution. Many other low income kids attend those institutions as well. Many fewer high income kids do. And so as a result you are less
exposed to high income kids in your college if you’re
from a low income family than you are to other low income kids. So that’s the data for colleges. Now you can do the exact same exercise for the neighborhood in which you grew up and that gives you these bars here basically asking who are you exposed to in your precollege neighborhood. If you’re a low income kid, what fraction of your childhood neighbors are low income, middle
income, high income? And the big picture
point I wanna make here without dwelling too much
on those small differences is that those two things
look basically the same. So if you look at the fifth
quintile for instance, in both your college and your precollege
residential neighborhood, about 10% of the people you’re around are gonna be from the
top quintile on average. And so it doesn’t look like
the degree of integration across colleges is actually
tremendously different than the degree of integration
across neighborhoods. A different cut at the data that makes basically the same point, look at now the income
distribution of the peers of kids from the top quintile so this
is looking at the flip side. Suppose you are from a
relatively high income family. What fraction of your
peers in your college in your childhood neighborhood come from different parts
of the income distribution? So if you’re from a high income family, you tend to be around people
from high income families in both your childhood
neighborhood and your college and again those rates
look roughly the same. The degree of segregation
of high income people looks about the same across neighborhoods as it does across colleges. I think most striking is if
you zoom in in particular on places like Harvard, if we go to the Ivy Plus group that we were looking at before, you see I think an interesting fact which is if we focus now on
kids from high income families, that’s the majority of kids as
we saw at Ivy Plus colleges, kids from high income families, what fraction of their
peers in their college and what fraction of their peers in their childhood neighborhood are themselves from high income families? You can see that you
actually have fewer peers from low and middle income
backgrounds in college than you did in your own
childhood neighborhoods so you are less likely
to come into contact with kids from low and middle
income parents here at Harvard than you were in your
childhood neighborhood. That’s what you see directly in these data so it’s the opposite of
reducing segregation. There’s actually amplified segregation at elite private colleges relative to the childhood neighborhoods where those kids grew up. Yeah? No, it’s true even without the top 1% so if you look at the top 1% in particular we have those calculations in the paper. You see a more exaggerated
version of this of course in the top 1%, but even among
kids from the top 1% here, you are more likely to be
exposed to kids from the top 1% here at Harvard than you were in your
childhood neighborhood. So it continues throughout
the income distribution. It’s not a fact just driven by the top 1%. Yeah? Yep. Yeah. Perfect question, trends
in income segregation. That’s the next thing
that I’m gonna talk about. Exactly as you said,
the preceding estimates are based on kids born
between 1980 and 1982, attended college in the early 2000s. As you noted there have
been substantial changes in the higher education
system since that time and in particular substantial
changes in financial aid at places like this and
tuition policies more broadly. So some favorable like the
increase in financial aid in places like Harvard, but then at many state schools significant increases in tuition, reductions in financial aid so things have gone in different
direction different places. So very natural question, how has income segregation across colleges changed in recent years? What can we learn from that? Are these data still relevant today? Can we get some lessons about what might matter going forward? Okay so let me show you a
few different things on this. I’ll come to the Harvard
specific data in a second. So let’s first start at the broader level. So think about those
measures of segregation that I was showing you. What fraction of peers come
from the top income quintile for kids who are themselves
from the top income quintile? So this is that right most bar on the chart that I was just showing you and so the blue series measures the amount of
segregation basically in precollege residential neighborhoods and then the red series is for
segregation across colleges. And the point I wanna make here is both of these things
look basically flat. Up to the point where we
have data, about 2011 or so, there’s no real appreciable change in the amount of isolation of
kids from high income families from other backgrounds either
at the neighborhood level or the college level. So in the aggregate
you’re not really seeing much of a change. Let’s now break that data down
and look at specific colleges so this is now coming
to answer your question. So in particular around 2006-2007, there were a series of changes
at elite private colleges. Many of you might have heard about this. Harvard effectively became free for kids from low and
middle income families. Stanford followed suit the
next year and so forth. And what you can see is now we can plot various different statistics here, but if you look at in this case the fraction of kids
from the bottom fifth, you look at the red line for Harvard, you see a little bit of an
uptick around 2006-2007, but it’s not really a very large change. If you look at Stanford,
it looks totally flat. I’m showing you the data here
for kids from the bottom 20%. You can also look at other statistics like fraction of kids from the bottom 60% which might be more relevant
because some of the changes were targeted specifically
at the middle class. There also… You can debate was there a
small increase, was there not. What is very clear from the data is there definitely was
not a dramatic change. Where you see significantly larger changes in the American higher education system are you look at places like
UC Berkeley or Stony Brook which I showed you had a fairly
equal income distribution at the beginning of the period. Stony Brook has a steady
decline in the fraction of kids from low income families. You see a little bit of that at Berkeley, see quite a bit of that at
Glendale Community College. So what’s happening in some
of these state institutions is as government support basically, funding for state
institutions is being shrunk, many state universities like Stony Brook increasingly have pressure
to increase tuition and admit more out of state students who tend to be higher income and so they’re serving a
smaller and smaller share of low income students. Now there are many other trends in the higher ed system as well. There are more kids now attending, low income kids attending
for profit institutions. There are more low income kids attending two year community colleges, at least certain two
year community colleges so all of it on net basically leaves the amount of segregation roughly unchanged in
the system as a whole. There are significant
changes at specific colleges as you can see, but coming
back to your question at the elite private colleges our sense is the changes in financial
aid actually did not have an enormous impact by themselves. Now that said… So the way I interpret that
is if you want to increase the representation of low income students, you might need to do more in addition to changing financial aid. It’s hard to imagine financial
aid is totally irrelevant. Obviously if the college
was completely unaffordable, it’s hard to imagine you’d have a lot of low income students, but I think of it as like
it’s not sufficient by itself and so we’ll talk a little bit later in subsequent lectures about what else might make a difference, but that’s what the data have to say about the financial aid change by itself. Okay so let me stop here
to take some questions. Anybody else? Yeah, in the back there, yeah. Yeah. Yeah so you might worry
about that so we’ve also… It was easier for technical data reasons to do zip code than census trakt which would’ve maybe
been more natural measure because it’s a little bit finer. We know from other work that census trakt would give you something extremely similar so I’m not worried about
that for that reason, but you’re right. If you only have the data on zip code you might worry about that. Yeah. Anybody else? Okay great, so what we’re gonna do now is turn to the second part of the picture on students’ earnings outcomes. Okay so what we’ll do is measure kids’ individual
earnings in their mid 30s so one thing I should note just as a point of distinction, so in everything we’ve been doing so far on the parent side we’ve
using household income so both earners if there
are two earners in a family. We do it that way because
we think total resources might matter for kids’ outcomes. When we look at the kids in contrast, we are going to look
at individual earnings excluding your spouse. Now we also construct studies
including your spouse. The reason we did it this
way is because turns out that rates of marriage also vary pretty sharply across colleges and they also vary significantly across, vary significantly with parental income. And so you will start to see patterns that are driven in significant part by differences in marriage
rates rather than differences in a child’s own earnings if you use household income measures in the early 30s or the mid 30s. Okay so just to be clear
on what we’re doing. Okay and so again with the kids, we’re gonna define percentile
ranks by ranking kids relative to other kids
born in the same year. Let me once again just give you
a sense of the distribution, give you a sense of the magnitudes. So when we’re looking at the
income distribution as a whole for individual income for kids at age 34, getting into the top fifth means earning more than about $60,000 a year. Getting into the top
1% means earning about more than $200,000 a year
when you’re 34 years old. And the median is $28,000 a year so keep those numbers in mind as we talk about the different quintiles. Okay so to look at
students’ earnings outcomes I’m gonna go back to the
charts that I was showing you of parent income distributions
which are the bars here and I’m gonna add these lines and I’m giving you an example here with Columbia and Stony Brook. And the lines show you
the fraction of kids who reach the top fifth of
the income distribution. Sorry, that should be labeled, but just to be clear each dot there is telling you the fraction of children in a given parent income quintile who end up reaching the top fifth. Okay so that upper left
blue dot for Columbia, that tells you that about 60% of kids from low income
families at Columbia end up reaching the top fifth
of the income distribution. At Stony Brook that number’s like 50%. I think most interesting about this chart is not necessarily that level, but the fact that that line
is almost perfectly flat across the income distribution. So basically what that’s telling you is that kids from low income families and kids from high income
families at Columbia had essentially the same outcomes. And the same thing is true
at Stony Brook as well and the same thing is true more generally at all of the colleges we’ve looked at. There’s some gradient,
but it’s not very large. Okay and so why is that so important? The key lesson that you see there, what it implies is that most
of the gap in the outcomes that we see between children from high and low income
families in the United States so kind of starting fact that
kids from high income families tend to do much better than
kids from low income families, most of that can be explained by differences between colleges, that is differences in
the colleges they attend rather than within colleges. That is to say if you take kids from low and high income families attending the same college, they have much more similar outcomes. Maybe not identical outcomes, but much more similar outcomes. And so why is that important from the mobility perspective? It raises the possibility
that reallocating students across colleges could potentially
have a significant impact on intergenerational mobility. Imagine for a moment that
we had found the converse. Suppose the gap in outcomes
by parental income was large even among Harvard students or even among students at Columbia. Then if that were true, then simply changing the admissions rules and reallocating students across colleges, you’d be surprised if
that would have any effect if even at a given college, low income kids end up having significantly different outcomes, you’re not necessarily gonna do anything by changing admissions rules. The fact that those lines look pretty flat raises the possibility that changes in the
higher education system could potentially have a big impact on intergenerational income ability. So we view that as encouraging in terms of suggesting there could be a pretty significant role
for higher education, potential changes in higher
education admissions policies for having an impact on mobility. Okay is that clear? Any questions on that? Okay so what we can then
do is combine the data on parents’ incomes and students’ outcomes in order to characterize
colleges’ mobility rates. So specifically think about
the following question, at which colleges in America
do the largest number of children come from poor families and end up say in the top fifth or end up in the upper middle class. Okay so if you think about mobility as rising from the bottom to the top, at which colleges in America just in a descriptive accounting sense, where are we seeing
that happening the most where you’ve got lots of kids
coming in from poor families and ending up doing quite well? So to make that precise quantitatively, I’m gonna focus on two
statistics in particular, what I call the top quintile outcome rate. That’s just the fraction of students who reach the top quintile. That number is 51% at Stony Brook and then I’m gonna focus
on what you might think of as low income access
or the representation of low income students at a given college which is the fraction of
parents from the bottom quintile and at Stony Brook that number is 16%. So we can combine these two numbers to construct what we call a mobility rate so we’re gonna define a
college’s mobility rate as the fraction of its students who come from the bottom quintile and end up in the top quintile. So for students who have
taken a statistics class, this is a joint probability. What’s the probability that you both come from the bottom quintile and end up in the top quintile? That mobility rate can be
computed very simply as follows. It’s just the product of
the fraction of students who come from the bottom quintile. The product of low income access
and the top quintile rate, the fraction of those students who end up reaching the top quintile. So let me just walk through an example that I think will make that clear. So remember at Stony Brook
we saw that 16% of students come from families in the bottom quintile. Of those students, 51% make
it to the top quintile. So what does that mean? 8.4% of them started out
in the bottom quintile and ended up in the top quintile, just the product of those two numbers. You might notice if you’ve
taken a statistics class, this is basically Bayes’ rule. That’s what you’re seeing here. And so that mobility rate, 8.4%, how do we interpret that number? It means that if I talk
to 100 kids at Stony Brook and talk about their experience, 8.4 of them on average will have come from a
relatively poor family and will have ended up when we’re measuring
their earnings at age 30 near the top of the income distribution, the top fifth of the income distribution. So the mobility rate gives you some sense of the role a college is playing in terms of promoting income ability. I wanna be careful about saying promoting because we haven’t shown anything about causal effects here, but at least in a descriptive sense it’s kind of where mobility
is happening if you like. Okay so I talked about the
mobility rate at Stony Brook and we talked about the data for Columbia. So I’m gonna use this visual device to show you the data for
all colleges in the U.S. and give you a sense of what
mobility rates look like. So what we’re doing here is plotting the top quintile outcome rate, what fraction of kids reach the top fifth versus the low income access measure, the percent of parents who
are in the bottom quintile. So remember at Stony Brook for example we had 16% of kids coming
from the bottom quintile and about 50% of them
reaching the top quintile. That’s how you add that up. Columbia had fewer kids
from low income families, slightly better outcomes in terms of higher levels of earnings. Okay so those are two colleges. Now we can introduce all
the colleges in the U.S. with all of these dots. Each college is a dot on this plot. So first thing you see is there’s a strong negative correlation between these two things. Kids from low income families so colleges that serve
a lot of low income kids tend to have poorer outcomes. So what is a concrete way to say that? What are these colleges that
are near the bottom here? Those are a lot of community colleges or it might be for profit colleges, less selective colleges basically
or non-selective colleges which tend to have a high
representation of kids from low income families
as we saw in the first part of what we discussed. And they also tend not to have very high levels of earnings outcomes. Now to be clear that could be
either because those colleges don’t actually provide
very good instruction, don’t actually maybe connect
you to good networks. They don’t have a great treatment effect or it could be about selection. Obviously the types of kids who
are going to those colleges, the backgrounds they have coming in are very different than the types of kids who are sitting in the room here today so there’s obviously an important selection component as well. So we’re gonna talk about
that in the next section when we talk about causal effects, but for now just as a fact in the data, you see that low income
kids tend to go to colleges where you have relatively
poor earnings outcomes and then if you look at the places with very limited access
for low income kids or small representation
of low income kids, places like Columbia or Harvard, you see much higher outcomes. So we can look at a few different
clusters of colleges here so you can look at the
Ivy Plus institutions. They are all concentrated on
the upper left of this chart. Relatively few low income kids. Some of the best earnings
outcomes that you see. Another group that’s interesting is the state public flagships. So as I was saying earlier, we had expected going into this that the state public
flagships placed at places like Berkeley would be some of the highest mobility rate places where you’d have lots of kids from low income families
and quite good outcomes. They do have very good outcomes as you can see Berkeley, University of Michigan
Ann Arbor and so forth, but in terms of access
they actually don’t look that much dramatically higher in terms of their representation
of low income kids than the elite private colleges. And so what then are the colleges that have the highest mobility rates? Those are the colleges that
are in the upper right. They’re like the upper right
outliers on this chart. They both manage to have
a lot of low income kids and they have pretty good
outcomes for low income kids. So what are those places? So this is a list of the
top 10 colleges in America by bottom to top quintile mobility rates and so you can see here that the places on this list are not necessarily the
places you would think of as some traditional like
U.S. News & World Report type best colleges in America. They’re more mid-tier public institutions so you take like Cal State L.A. We talked about Stony Brook, the City University of New York. In all of these places you have something like a mobility rate of
8% or even nearly 10% at Cal State L.A. so that
means one out of every 10 kids at Cal State L.A. is coming
from a low income family and ending up doing quite well. So this is again this joint probability so this is not saying for a given child your outcomes look the
best at Cal State L.A. It’s saying Cal State L.A.
has relatively good outcomes and lots of low income kids which is why there’s a lot of mobility
happening at Cal State L.A. at some level. And so you can see all of those, the colleges that have
the highest mobility rate have that kind of mid-tier public school sort of flavor to them more or less. If you look at Harvard, there’s a mobility rate of 1.8% so why does Harvard rank relatively low like other Ivy Plus colleges on this list? It’s just very simple logic that if you have relatively
few low income kids, obviously you can’t have a lot of mobility defined in this way
because you’re just working with a small pool of kids to begin with so even if you have the best outcomes, you’re multiplying that by
a relatively small number. And so more broadly there are
these really sharp differences in mobility rates across colleges and I think the interesting
agenda is to do two things. Figure out… If you want to increase mobility through the higher education system, you wanna figure out
perhaps how you increase the representation of low income kids at colleges that have the best
outcome so increase access at places like Harvard for instance or University of Michigan
Ann Arbor, et cetera. And then second coming
back to this chart here, you have a set of colleges
that are near the bottom here where you have quite a few low
income students to work with, but you don’t have very
high mobility rates because the outcomes just don’t
look that great at present. And so a second different
problem to try to tackle is how do you improve
the outcomes in colleges where you don’t see very
good outcomes at the moment and typically that’s on
margins like trying to get kids not to drop out or do better
in the college and so forth and that I think is
also a very challenging and interesting problem. And so what we’re gonna do… And so what I’m gonna do now is since we’re essentially up on time, I’m gonna leave the
causal effects of colleges to a subsequent lecture. In the next lecture we’re
gonna have a guest lecturer by Tim Renick from Georgia State who’s been doing a lot of great work, predictive analytics work on trying to improve outcomes there and then we’ll come back to
the causal effects of colleges. So that’s it for today. (applauds) Thanks.