PARTICIPANTS

Eric Bettinger, John Taylor, Joshua Aizenman, Uschi Backes-Gellner, Patrick Biggs, Michael Boskin, John Cochrane, Bradley Combest, Steven Davis, Sami Diaf, Eric Hanushek, Michael Hartney, Kevin Hassett, Evan Koenig, David Laidler, Patrick Lehnert, Hans Lueders, Livio Maya, Michael Melvin, Axel Merk, Dinsha Mistree, David Neumark, Radek Paluszynski, Elena Pastorino, Valerie Ramey, Madison Reel, Richard Sousa, Yevgeniy Teryoshin, Eric Wakin

ISSUES DISCUSSED

Eric Bettinger, senior fellow (joint) at the Hoover Institution and the Conley-DeAngelis Professor of Education in the Graduate School of Education at Stanford University, discussed “The Effect of Postsecondary Institutions on Local Economies: A Bird’s-Eye View,” a paper with Patrick Lehnert (University of Zurich), Uschi Backes-Gellner (University of Zurich), and Madison Dell (Tennessee Board of Regents).

John Taylor, the Mary and Robert Raymond Professor of Economics at Stanford University and the George P. Shultz Senior Fellow in Economics at the Hoover Institution, was the moderator.

PAPER SUMMARY

Postsecondary institutions affect the economy of the area around them, but the question is how. In the early 2000s, the United States experienced a rapid increase in both the number of students attending college and the number of branch campuses serving these students. We examine branch campus openings that took place in two states, Tennessee and Texas, that are representative of the underlying patterns in the nation as a whole. We provide estimates of the impacts of these branches campuses on local economic conditions. Because the impacts of these branch campuses could be more localized than county- or state-level data might reveal, we use satellite images to construct otherwise unavailable measures of economic development around these small branch-campus regions. We find a clear positive association. In Tennessee, this effect seems to be driven largely by two-year campuses, while the effect is higher for four-year campuses in Texas. As the location of new branch campuses is likely endogenous to local economic conditions, simple estimates may not reflect a causal effect. For Texas, we are able to use an instrumental variable to estimate causal effects. Our instrument takes advantage of local taxing regulations that likely influence the decision to open a branch campus in certain locations but not the local economic conditions. Using this exogenous variation, we find an even larger positive effect. Given that many states use higher education as a strategy to induce economic growth, particularly in rural areas, this paper contributes some of the first empirical estimates of the impact of campus openings on regional economic activity and offers perspectives on using this approach as an economic development tool.

To read the paper, click here
To read the slides, click here

WATCH THE SEMINAR

Topic: “The Effect of Postsecondary Institutions on Local Economies: A Bird’s-Eye View”
Start Time: November 1, 2023, 12:00 PM PT

>> Presenter: Ready to begin? The speaker today is the famous Eric Benninger, who's going to talk about my favorite topic, the effect of post secondary institutions of local economies. A bird's eye view. Eric, go ahead.

>> Eric Bettinger: So first off, thanks everybody for being here, appreciate you taking time. It's always dangerous when you're giving a talk, especially when it's kind of, you know, it just started at Hoover.

And so this is kind of my first actual event at Hoover that besides a faculty meeting is, and so it's always dangerous. You want to put your best foot forward. And I think this is a good paper, but it's one of those papers where it's right at that phase where we could really use good comments to see ways that we can make this paper better.

I think this paper has asked kind of an important question, but it's one of those where you'll see as we go through it, it's a little messy as we start to go through that, and we hope we've trying to figure out ways to handle that. This question of the impacts of colleges is just a perennial question, right?

We have a large literature that kind of thinks about the private returns. But what I want to talk about today is more kind of the social returns or what happens around the public return that's happening there. And the hard part that we have here is in, at least in us, higher education, there's just very little variation.

You go to these college towns and these colleges have existed for 50, 6100 years, 300 years. And in all the time that we have data and have been really tracking economic outcomes around those, there's just not a lot of variation in the four year campuses and what's there.

One of the things that happens is as you start to think a little bit about in the United States. Particularly between the period 2000 2010, we had a huge run up in terms of the number of students who are attending higher education. That's dropped off a little bit since that point.

I'll show you some numbers here in a second, but we want to try to see if we can use that to try to think about it. I also want you to think a little bit about why it matters. So as you start to think about why higher education might have that impact, the standard stories here, the one that's at the bottom of this slide, which is trying to think about productivity, R and D.

The different papers that have come out around Silicon Valley really focus on this, this kind of technology transfer from higher educational institutions to the local economies. On the other hand, as you start to think about it. Part of it is also that supply of labor that you're producing.

Then as you create institutions, you're creating a willing supply in some sense. Google could go anywhere, but part of the thing that keeps them here in the valley is the supply of students that we produce for them that gives them kind of a good and solid kind of base.

And that's really going to be, I think, the mechanism that kind of goes through the kind of setting that we're going to try to do. So this is where we're going to try to live. We're going to try to live on this a little bit, but we want to try to think a little bit about is how do you start to measure that kind of regional variation in economic activity in a systematic way?

And what we're going to try to do, the reason we call it a bird's eye view, is we're going to use satellite data to help us with that. I'll explain why that's going to be important for us as we go through here. So let me just kind of set this up here.

So the two states we're going to be focusing on are Tennessee and Texas. I'd like to tell you that, you know, we chose strategically. I think it was very convenient. Both of these have data. We had some good connections in both states that might help us with gathering some of these data.

But as we've kind of gone through both of these days kind of are very representative of what was going on in the United States. You look on the left in Tennessee, you see Tennessee has this enormous run up in the nineties, and then it flattens out in the early eighties, beginning of the nineties, has a big run up and then it flattens off.

And then at the end of the two thousands, another big run up. And in Texas, it's just been up and up and up and up in terms of the numbers of students that we have now, this is comparison.

>> Participant 1: Where's California?

>> Eric Bettinger: California is the same thing. Almost every state looks probably more like Tennessee.

California looks a little bit more like Texas. The valley that Tennessee gets isn't as severe.

>> Participant 2: And what colleges?

>> Eric Bettinger: Any post secondary, so any two year, four year colleges are included here now, including private? Including private as well. If I were to draw the four year colleges over the same period of time, it's a flat line.

Between the early eighties and two thousands, there's, in these two states, there's like two new four year colleges, but this number that I have, the dotted line on this graph is the branch campuses that went forward. What happened was, as this enrollment surge was happening over this period of time, there were two ways colleges managed it.

First off, they took their existing colleges and they made them a little bit bigger. So we'll see places like Texas A and M just blowing up to go from, you know, a moderate size college to an extremely large college. But then, on the other hand, what happens is this establishment of these branch campuses, and you can see the numbers here, you know, that's where a lot of that new enrollment was handled at the branch campus level.

And so what we want to try to think about in this paper is we really want to focus on these branch campuses and think about these branch campuses as a type of institution. Look at what the economic impact is of establishing these branch campuses, please.

>> John Cochrane: A lot of the story, a research university starts Silicon Valley, but if we just have a new junior college teaching you accounting, it might be nice for people who are accounting, but it's not going to have the same spillovers.

What question are we asking?

>> Eric Bettinger: I think that's exactly. And so part of like on this slide here, what I'm trying to get at is it's, there's few R and D activities at these branch campuses. These are teaching campuses. So as you start to think about the mechanism that might be going on here, this is about the skilled labor force that you're developing in an area and a more reserve area.

And so the big thing here is going to be fraction of business or helping existing businesses have a deeper and better and more skilled pool.

>> John Cochrane: Farm.

>> Eric Bettinger: Yeah.

>> Participant 3: But are they four year or are they sort of first two years and then you transfer students graduating from school.

Possibly into the labor market.

>> Eric Bettinger: So we have both four year and two year branch campuses now, as you might expect. And I think I have it on the next slide here at the two year campuses, just to get you thinking about what they're doing at the two year branch campuses, there's kind of four things that we put up here.

These are the four largest programs, health, so think nursing here, computer and information science, liberal arts. And that's really the students who are trying to go for a transfer. They're going to a two year branch campus hoping to transfer over. And then actually education programs are pretty significant here.

Think a teaching, Aidan. Trying to get some kind of a certificate to get themselves in. So this is really the kind of group where you're seeing a lot of the work here done. If I had to put a fifth one up there, it would be like engineering technology, which is not the R and D on the engineering.

It's really about how to operate that type of machinery. Now there's also four year programs, and the four year programs kind of mirror the same ones in terms of the, again, nursing education, computer science are kind of the big ones that we have there. Please.

>> Participant 4: Yeah, question was whether any college branches, these campuses, closed down at all?

 

>> Eric Bettinger: We don't see a lot of exit from the market across the board, at least in our data that we have. We can't find them. I mean, one of the things we'll talk about here in a second is. These branch campuses are actually sometimes hard to find, because if Texas A&M, as an example, decides to establish a branch campus all the way out in El Paso, they can do that, right?

Now, the hard part is when they establish that branch campus all the way out in El Paso, they report the statistics for Texas A&M altogether. So when you are looking at Texas A&M's data on the national database, like iPads or whatever, you're seeing the aggregation of the main campus and all the branch campuses.

So one of the hard parts in this is when does a campus, a remote campus, really become a branch campus? And our definition on this is as soon as you can complete a degree in that branch campus location, we're gonna count it as a branch campus. So if all it is is a remote location where you can go, it's an extension, you can take some classes, but really you have to get back to the main campus.

We're not gonna count that. We're gonna focus only on those cases where you can attend the campus and complete some type of credential at the campus.

>> Participant 2: In terms of the depth and breadth of the traditional service provided, is a remote location versus a branch systematically.

>> Eric Bettinger: So what's interesting is the biggest source of remote locations that we see in the data is actually high schools.

Dual enrollment programs are becoming much more apparent. And the easiest way to do a dual enrollment program is to actually deputize the English teacher at a high school as a teaching assistant, and have it taught there at the high school itself. So we're gonna throw those out. Those are kind of a different mechanism altogether.

 

>> Participant 2: Because I was familiar with the University of Minnesota system, and the non-Minneapolis-based campuses were informally specializing in catering to different types of student bodies.

>> Eric Bettinger: I think that's right. And part of that gets at also the kind of set of programs that they're going for, like these health programs and the education programs in particular.

In our case here, one of the questions that will kind of come to it, I think, by the end of this, is trying to think a little bit about efficiency versus equity type of thing. Because, like in the case of Tennessee, what you find is the average size of the campus really declines, in part because they're building campuses faster than that enrollment is increasing.

But they're increasing in kind of more and more remote areas.

>> Participant 2: To conclude that, the conversation that I was aware of at the time, in the case of leaders of campuses, was that the was solved towards effectiveness, posturing, segregation, potential student body.

>> Eric Bettinger: Sure. Let's come back to that as we talk more about it, because one of the things as we go through this, I think that's gonna be one of the kind of places where I think there's more work to be done here.

But let's come back to that. Okay, so you have an idea of kind of what these are. So here's our problems with trying to answer this question. The first one is you gotta figure out when all these branch campuses open and close, and this is not an easy task.

This is a lot of sweat equity, trying to call campuses, scour archives on the Internet to try to find out exactly when buildings were commissioned and actually opened up. The second piece here is trying to figure out a little bit about how are you gonna measure economic activity.

Most of our measures of economic activity happen at a level of aggregation, which might be different from what the branch campus is really catering to. And then the third one we're gonna have to deal with is the endogeneity of the location. Is it that you see economic growth about to happen in an area, so you put a branch campus there, or is it that the campus does it?

And so I'll try to kind of walk through those three problems. So on the first two.

>> Participant 8: This NETS, National Establishment Time Series, it's a longitudinal linking of all D&B files, Dun & Bradstreet files, and it includes non-profit and public establishments. And it also has an employment measure at the establishment level.

But these things are geocoded. It's an expensive data set. That's the bad news.

>> Eric Bettinger: I mean, maybe afterwards, I can get the exact details. I mean, the hardest part that we've had to is the paycheck still comes from Texas A&M. And so when you start to think about the employer code that comes with this.

 

>> Participant 8: But you get an establishment location, every establishment is broken out. You've got opening dates.

>> Eric Bettinger: So even if you're working for Google, I can tell which part of Google you're working at?

>> Participant 8: Physically. Physical establishments, yeah.

>> Eric Bettinger: Okay, yeah, that'd be very helpful for us cuz-

>> Participant 8: I use it for very different things, but I think.

 

>> Eric Bettinger: Subject to what Dun and Bradstreet knows. Sure, but I think this issue about the extent to which Dun and Bradstreet establishments actually correspond to physical locations is germane here. It's not clear they're gonna know all of these branch locations.

>> Participant 8: Yeah, I mean, I've done a lot of work that looks pretty.

I mean, I think the thing that they have this whole non-employer established, non-employee establishment thing. But that's irrelevant anyways.

>> Eric Bettinger: I'll tell you, the one that's a little frustrating in this one is we work a lot with the Texas Higher Education Coordinating board. They don't even know when these institutions have started, right?

Yeah.

>> Participant 6: There's a broader issue, right? I'm a little confused about what the underlying economic question is. If it's about the mechanisms by which you develop skills and the workforces that attract certain kinds of employers, then it's not clear why we would even limit our attention to this set of institutions.

Cuz there might be many mechanisms, many vehicles for doing that. Second, if that's really what it's about, don't you need to get into the details of the skills that are being imparted? And into the skills that employers are looking for, and you haven't said anything about it.

>> Eric Bettinger: I haven't, and I'm not going to in some sense, because I think the real research question we're trying to establish is even kind of an antecedent to that.

Which is, you're building these campuses, and our traditional mechanism of R&D is really not happening at these campuses. So I wanna know-

>> Participant 6: That's why I thought we were already away from the R&D side. We are on to the skills transmission side.

>> Eric Bettinger: And what I wanna try to think about here is, there's kind of two things I'm gonna try to think about.

The first piece is, I just wanna ask the question. Do these branch campuses have an effect on the economic impact? The second one I'm gonna try to do is try to figure out measuring that is difficult. And that's what we're gonna try to do in the paper. Just answer the weather.

Now, the question, then, on the skills and the skills balance. That's kinda the place where we're hoping to go next. I mean, one of the things I'm gonna show you is the way we're gonna measure economic activity is I'm gonna look at the satellite data, and I'm gonna build that new buildings that are going up.

 

>> Participant 6: Even if you had exogenous variation where these branches are opening. If you don't know what they're doing, then it seems to me hard to draw much of an inference about the skills transmission. So I think it's fine. It's great that you gotta figure out what's exogenous here, but if you don't know the thing that you're burying exogenously, which is what kind of skills they're imparting.

Then, I don't see how you're gonna make a lot of progress on understanding of interest.

>> Eric Bettinger: Well, I mean, let me give you kind of a different role in this, right? So the first thing that we're doing right now is, I'm just gonna show you kind of what happens around that neighboring area.

And as you think about it, there's both two things, right? The first is you've established it, and so now you just establish more activity happening, and so I'm gonna be able to capture that. What I'm not gonna be able to capture, at least at this point, is whether that new building is a hospital, or a medical center, or a school, right?

Because based on what we know that these schools are doing, those are the biggest professions that we're seeing them produce. So in theory, we should actually see increased build out, if indeed they're kind of having some type of supply, meeting something here. So the hard one here is the second one, the computer and information science and support programs.

And at least in the data I'm gonna show you, I'm not gonna be able to identify the types of buildings those people go in. And in this version, I'm not distinguishing between the types of buildings, but that's our hope to kinda get to the future as we get better data on the satellite part.

Right now, as I show you exactly what we see, you'll see that we can see the build out happen, but it's gonna be difficult for us to figure out why the buildups happen.

>> Participant 7: So getting back to Steve's question, in some sense, what you're thinking about is you have, say, the state of Texas trying to decide what to do about improving skills.

Several options for doing that. One is expand enrollment at the main campus in Texas A and M or UC, etc. Others go where the people are or might have something that's less expensive, or they're constrained with housing, some other reason they can't expand, so they do this. And the question is, are you able to see any difference in skill composition?

Are there a lot more people doing this IT program relative to what might have projected would have been the case if they had done the alternative, not open that satellite campus or done nothing? Fewer people or fewer sociology majors, maybe, thankfully.

>> Participant 7: And more computer science.

>> Participant 8: Sociology established.

 

>> Participant 7: I think that's something that's really important to think through to see what kind of the value added of what they're doing.

>> Eric Bettinger: Here's the prelates, where we've struggled with this is when you actually look at your Texas A and M degrees, they don't break out which ones came from the branch campus.

So in any of our kind of national reporting or any of the state reporting, what we observe is Texas A and M producing degrees. And I don't know if those extra computer science degrees came from a branch campus location, wherever, or the main campus. So that's gonna be the limitation I have on that front.

 

>> Participant 2: Have you tried.

>> Participant 7: Have you tried to get that unpublished data from them that would be really high value adds.

>> Eric Bettinger: We try to, and I can't, I haven't been able to yet. I mean, the one that we have an ARIA working on right now is trying to go back to the census data, because our unit of observation at the end of the day is gonna be the census tract.

And the hard part is census tracts change over time. But to the extent that you have some continuity there, we can look at whether there's a change in the professional mix and the educational mix at the census tract level. Again, the problem there is, at the level that we're thinking about, we're really only getting ten year intervals.

We wanna see the dynamics that I'm gonna show you, it's really in that first year, four to year ten, that we're really seeing the ramp up and economic activities coming from these institutions.

>> Participant 8: We were just asking the ACS, You can feel the study.

>> Eric Bettinger: Yeah, what I needed at is the census tract level.

 

>> Participant 8: Yeah, In RDC.

>> Eric Bettinger: So that's where we're going next on that front, to try to get at that.

>> Participant 2: And Steve, as a labor economist, of course, I'm sympathetic to questions and mechanisms, but I thought your question is more available, meaning, what are trade like questions? What are the sources of economies of densities and agglomeration?

So at a high level, you're asking, I see, in this type of branch of higher educational institutions, does that contribute in as a way. As you will be able to tell us, to look, the density of local economic activities, which is a big question, some trade urban economies.

 

>> Eric Bettinger: I think that's right, that's very fair, thank you. I need to make you cooperate.

>> Participant 9: But also, I think that's right, I like your question. But to interpret what's going on. Because I'm thinking about suppose they open a branch in Merced like the University of California did.

Suddenly, how much of what, suppose you see an uptick in activity, how much of that is that they train students who stayed there and how much of it is, you brought in a bunch of highly educated people? And their kids go in the public school, and then there are all the direct spillovers just from the people you brought in there as employees at the college.

 

>> Participant 10: Well, if I can just add on, I mean, I'm grappling with how do these two year places compete with community colleges, and what's the impact there? And are these two year colleges, are they resident or commuter colleges? The commuter colleges are probably competing with the local community college or junior college, whatever they call it, in Texas.

 

>> Eric Bettinger: Well, and many of them are actually affiliated. So, instead of using Texas A and M, we'd be talking about, like, Lone Star community college. And Lone Star is an example that I think they have seven branch campuses. So these are, and what we'll see is that a lot of the activities happening here in the suburban level in terms of where these are going.

And I'll show you a little bit more on that mechanism. I think, going back to this question, I think it's a good point. I stake in it to come back to it, to try to think a little bit more about the skills mechanism. I wanna try to think about just the measurement of anything's happening first and then try to get behind that after that.

So let me kind of plunge forward just a little bit here and then, cuz trust me, I told you that this gets a little messy before it gets clean here. So let's start with this first one, acquiring the data on these applications. So I've already told you, we searched everything, and I told you kind of what we threw out on prison based camp branches are another kind of popular one that we threw out.

Whole different purpose. In terms of this measure of economic activity, one of the problems you have in Texas is the counties are big. Now, in Tennessee, the counties are smaller. And so oftentimes you don't get variation, almost as at the county level, not quite. So we got to figure out a better way to do that.

So this is a picture of Chattanooga, what I wanted to show you a Chattanooga here. This is in 1984, and what I want, this is what we're gonna be using to look at. And this is just a time lapse of what happens over the next 40 years in Chattanooga.

And so you can see here this buildup of things that are no longer green, brown, or blue, right? And what it's doing is it's helping us isolate kind of where the areas are, where there's economic growth happening here. And what we're gonna try to do is capture that growth and really think about whether that buildup that we see over that time period is correlated with what we see in terms of branch campus opening.

So what we're gonna do, the problem we have, right, is that kind of county level GDP, we just think it's not there. We could use, like, night light stuff, which is another kind of measure that people have used with satellite measure, we're a little bit worried about it.

We're also worried a little bit about when these campuses open, because a lot of the kind of measures we have don't and the one we're gonna focus on is the Landsat data. The Landsat data begins in 1982. We're gonna focus on 1984 to 2020, so we're gonna basically get a picture of every square kilometer of Texas in Tennessee.

And usually, we get one good picture per year, sometimes more. Sometimes we actually have cloud cover, we don't get anything. But what we're gonna try to do then is look at and show how that evolves over time. So we're gonna take each one of those satellite pictures, and we can basically take about a 1-kilometer square.

And we can basically decompose that into the respective parts that are made up of there. And just to kinda show you a little bit of this, this is a different paper that we wrote here. This is East Germany and West Germany. So the triangle data is the data that's officially reported.

The solid lines here are our estimate based on the satellite data. As you can see that it's not perfect, but it's a very high correlation. And the advantage is, we can drop it back in previous times like in East Germany here, we didn't have data before the wall fell, and the data we had, it was pretty unreliable.

So this is one way to kinda build some time series in terms of economic activity. This one we published in PNAS. So what we're gonna do here, we're gonna take the land satelite data, and then we're gonna take some data that we'll call kinda ground truth data. So think of this as basically we're taking Google maps on the close-up images, and we're gonna train a set of that data to try to identify.

Okay, here's the color we see in the satellite. What's that look like actually, in truth on the ground? And then what we're gonna do is basically create a rule to help us understand the colors we see in the satellite data, how that corresponds to buildup of activity, okay?

 

>> Participant 8: Train that out anymore based on the type of business in the Google data.

>> Eric Bettinger: That's the place we're trying to go.

>> Participant 8: Factories look very different physically in warehouses than strip malls.

>> Eric Bettinger: That's what we're trying to do right now. I think the bigger one for us is parking lots, helipads for hospitals, especially since we're thinking about education and schools.

If we can really distinguish those in the data, it's hard to do. I mean, here's kind of two pictures of Chattanooga in 1984 and in 2020. And so as you look at it in terms of the picture that comes back from the satellite, as you start staring at it, you stare at it long enough, you can start to see the differences, especially kind of in the southeast corner of the map here.

But this is also kind of color coded into the six shades we basically see. Now our one that we're gonna be built up is this built up. And that built up, since we can isolate in such a kind of small area, we can really use as our unit of observation in this study, the kind of census tract level per year.

So every census tract going from 84 to 20, we're gonna do that.

>> Participant 9: I'm noticing another economic activity that seems to go back so the crops seem to become forest.

>> Eric Bettinger: Yeah.

>> Participant 9: So, I mean, is that commercial for us? Because otherwise that would be a decline in economic activity.

 

>> Eric Bettinger: To be honest, I don't know that. What we're gonna do here is I'm gonna be a little agnostic I don't know on this particular map, you're talking about the kind of Northwest quadrant here, how it goes from light green to dark green.

>> Participant 8: Very discreet, they're very straight lines and it's very weird.

 

>> Participant 8: It's a little hard to believe.

>> Eric Bettinger: Well, it's very common.

>> Participant 9: Yeah.

>> Participant 8: That straight line goes from farmland to forest.

>> Participant 9: Yeah, that's one farm.

>> Participant 8: That's a farm, that's huge area.

>> Eric Bettinger: So let me show you what we're gonna do. What we're gonna do is, I'm gonna take every county in the US and I'm gonna take the satellite maps for every county in the US.

And I'm gonna take every county in the US and I'm gonna decompose it into those six surfaces that we can do. And then I'm basically going to predict GDP, use that model to kind of calibrate. Here's our rule, if you have this, this is the conversion of this type of buildup.

Now what I'd love to do is be able to do that at the census track level. Of course, I don't have that data. So what I'm gonna use is I'm gonna use the coefficients from this model and I'm gonna basically standardize it so that we don't have to worry about kind of the intercept here.

And I'm basically gonna standardize it. And in this standardized feature, I'm gonna basically go back and use this formula to compute it at the census tract level. So that's where I'm gonna get my measure here. I'm going to just predict GDP at any given time for any different county.

I'm gonna get those coefficients there on the vector. That LC here is the vector of our land cover categories. I'm gonna get that vector of it, and then I'm gonna use that at a more disaggregated level to predict what the economic activity is in any given year. So here's an example of what we're looking at.

So this is Texas here. I'm gonna show you down at a San Antonio Austin corridor here, and every one of those little divisions here is a different census tract. And I'm gonna be able to identify basically 20 years or 40 years of movement in the satellite maps in each of those.

Now, what I'm going to do is I'm gonna plot out all of my college branch locations. And I'm going to plot those out over time, and I'm gonna basically start to look at that correlation between those. Now, doesn't solve our endogeneity issue. Let's come to that here in a second.

But you can at least see, start to get a picture of where their variation is coming in. We've got this rapid buildup over this period of time. So if you squint hard enough at this map, you'll see these new circles and squares developing, which are campuses that are being built up over this period of time.

 

>> Participant 8: Where does county GDP data come from? That's data.

>> Eric Bettinger: BLS.

>> Participant 8: Yeah, they produce county GDP. They do county levels.

>> Eric Bettinger: Yes, thank you.

>> Participant 8: So, they're collecting data from businesses? Well, they, in a sense, have what you're trying to measure?

>> Eric Bettinger: The disaggregated file.

>> Participant 8: I mean, they must have some output or sales or income, measure one side or the other.

 

>> Eric Bettinger: No, it's a good thing for us to try to dig in. Yeah, payrolls, for sure.

>> Participant 8: Establishment level data. If they're actually surveying, and unless they're just imputing it from the state level.

>> Eric Bettinger: But that's the hard part for us, is whether or not if the economic activity happens in this area as opposed to another.

Our problem also is just thinking about what the local area is. I mean, as you think about some of these campuses, these are large districts, and so if I use the county, I'm not capturing the kind of local cashmen area around a branch campus. And that's really the variation I want.

In Tennessee, like I said, it's kind of Northwest ordinance. So the counties are much more set, much more defined, smaller in nature, a lot closer. But here in Texas especially, it's a little bit of a mess because of the fact that these counties can be very large.

>> Participant 9: As far as I know, there's very municipal data that economists use, and they actually look at retail sales in the cities and look at the effects of various things, all the preparation effects of housing price moves, the bus.

And there are lots of those data sets posted now as part of repetition files.

>> Eric Bettinger: The hard part is, it's gonna be like a suburb of San Antonio. So, I mean, if they think something like San Antonio, and they can decompose it to, like, the suburb of Bernie.

And help us understand economic activity in Bernie versus the greater metropolitan area, that's the direction I want to go here.

>> Participant 8: What about commuting zones? Wouldn't that make the most sense?

>> Eric Bettinger: Well, what we're gonna do here is commute. Kind of average commute in Texas is about 23 miles, so we're gonna focus on a 25 miles radius.

I'm gonna take-

>> Participant 8: Big zone will be big in rural Texas and small and urban.

>> Eric Bettinger: Yeah, absolutely.

>> Participant 8: It's 25 minutes.

>> Eric Bettinger: I mean, I can vary that, I can play with it, ten to 40, it starts to wash out once you get above about 50 miles.

But what I'm gonna do, and the empirical strategy we can think about here is I'm gonna take every one of those little buildings here. And I'm gonna try to identify whether your census tract is within 25 miles of within a commuting of that new branch campus. And I'm going to be comparing whether your particular census tract was essentially being shocked at some point in time over this.

 

>> Participant 8: I get a commuting zone is defined based on most people commuting within it, and it completely partitions the state.

>> Eric Bettinger: Yeah.

>> Participant 8: So then you don't have to do any arbitrary. Cuz 25 minutes, in Houston or miles, I should say there's a huge commute, 25 minutes out in western Texas.

 

>> Eric Bettinger: There's nothing but yeah, 25 miles in Texas is in Houston is your entire day.

>> Participant 11: You know which picture is Tennessee?

>> Eric Bettinger: I don't have a picture of Tennessee here, I didn't put one in this one. So, okay, let me get at, yeah, please.

>> Participant 12: Yeah, are you including the site of the actual campus when you measure the GDP?

Cuz you build a canvas, there's all like new construction. Just be introducing kind of like auto correlation between your different.

>> Eric Bettinger: We're not in part cuz I can't distinguish it enough. I mean, I can tell you which census tract it is, and we've done the results where we just kind of leave out a census tract if they actually had it physically located in there.

So that's about as good as we can do, that kind of that doughnut estimator. And we've done that as one of our robustness tests, and our results are similar. Okay, let me get at the endogeneity then. In Tennessee, I'm not gonna be able to do anything. In Texas, they've got this interesting thing of where these branch campuses come from.

So these branch campuses, they come in two different sizes. The first one, which is the bottom point here, is the Texas higher education coordinating board says, we need one here. And so they basically just say, we're gonna put one here, and they assign a campus to manage it and oversee it.

But the second one is the top one here, is the one that really helps us. So one of the ways in which community colleges form is a group of school districts get together and they say, we will give up some of our taxing authority. And we will form a community college district that can tax and actually generate a community college in this area.

Well, what happens here, these have existed long before our data started, and they charge differential tuition. If you're in the district, it's a different price than if you're out of the district. The marginal price of educating a student is less than that kind of the out of district tuition.

And so one of the things you see is a little bit of a race to be the each when you're in a taxing district, to try to capture those border areas so you don't lose your in seat. And then you might actually be able to attract some people from the other side of that line.

And the gray boxes here are all the taxing districts here. So we have this variation in where they're coming from. And just to show you one more picture, this is kind of zooming in on the Dallas area, 1984 and 2020. If you focus, you'll see that the campuses are kind of, as they form, they almost always start right in the very center of those counties.

And then what happens is, as we go out, they start to move outside of it to where those kind of population areas are developing. And what we find is that we tried to parameterize this in various ways, is that distance from where the taxing border is, is a good indicator for the likelihood that it creates an incentive.

For one of these community college districts to establish a branch campus so they don't lose some of their sample. So that's gonna be our kind of strategy. Let me walk through the three strategies I'm gonna show you today. One is just a conventional difference in differences estimator. I'm just gonna look at fixed effects for the camp, the track, fixed effects for time.

And then I'm just gonna look at did a branch campus open in this area or within a given mileage of that area.

>> Participant 13: You're calling GDP assists, vector of counts.

>> Eric Bettinger: It's this simulated GDP predicted based on the satellite data?

>> Participant 13: This is just Texas, just Texas.

>> Eric Bettinger: We're gonna do this for, I'm gonna show you for both.

I'm only gonna get the kind of instrument for Texas.

>> Participant 15: You actually have an output measure as opposed to just counts of buildings and so on.

>> Eric Bettinger: So where we're getting our count, our measures, we're predicting GDP based on this kind of countywide formula.

>> Participant 15: Keep using the term GDP, which by you mean gross county product.

 

>> Eric Bettinger: Gross county product, yeah.

>> Participant 15: You're saying that you're getting a proxy for that by taking some physical things, is that correct?

>> Eric Bettinger: Yeah.

>> Participant 15: So if you don't have issues of value, prices, all that are out here.

>> Eric Bettinger: No, I mean, basically what I'm literally doing is just trying to predict what the log GDP, kind of standardized would be in that particular zone.

 

>> Participant 15: So you have a measure of GDP that you're using at the county level.

>> Eric Bettinger: At the county level.

>> Participant 15: GDP county level, take at the track level.

>> Participant 14: Take it down to the track level.

>> Participant 15: Okay, so there's value, so there's your prices here that have to deal with because they differ.

If you look at Texas, one of those branches in Dallas or Houston suburbs is going to be twice as expensive as outdoor west Texas panhandle or West Texas. So that kind of makes a difference.

>> Eric Bettinger: I mean, the one thing that I'll be able to do here is I'm gonna actually be able to, like in our empirical specifications when we get down to our data.

Is I'm gonna actually put in a fixed effect for the census tract they're in, right? So if there's a systematic thing going on here, it'll pick up that. And so what I'm really gonna be looking at is that kind of variation from their norm and whether it's changed once it's happened, once they've read it.

And let me just, let me plunge through and just show you two quick other ones, just so we have all the things. I'm gonna do, kind of standard now in the difference and differences is this kind of heterogeneity robusty, difference to differences. What I like about this is I'm gonna be able to show you the kind of before it was established and after it is.

So if you think about like a granger causality, it's almost trying to help us really track what's happening, whether it helps us get a little bit of causality, but it's not perfect. And then the last one I'm gonna show you is I'm gonna actually do this. My first stage is gonna be when that branch campus is opening, and then something that helps me measure the kind of what the distance are to the nearest campuses around there.

 

>> John Cochrane: Yeah, go back to the first one. I think that was the simplest, so I'm just struggling with. So branch campus either opens or doesn't in a place like once, and then we've got a time dummy and the census tract dummy. It boggles my mind, what variation is left once you have a time dummy, census tract dummy.

 

>> Eric Bettinger: So what we're really looking is once it's there, the average afterwards, is it higher in this simple specification?

>> Participant 8: Time is common.

>> Eric Bettinger: So if we go-

>> Participant 8: Time is common.

>> Eric Bettinger: So the treatment effect is coming from the variation of when it was established. So if I take one area and it's established ten years in, then this dummy variable here looks, after those ten years, are we going to see a change?

Thank you. I like this one better because this one really kind of focuses it, trying to say, okay, let's allow for a differential effect in every different year before and after. So if we establish ten years in, I'm going to then look at the dynamics over the next ten to 20 years.

 

>> John Cochrane: And I'm sorry, how does that handle the endogeneity of the opening of the campus?

>> Participant 17: This next one does.

>> Eric Bettinger: So my therapy instance-

>> John Cochrane: These are not correct for endogeneity.

>> Eric Bettinger: This one here, you can make an argument that you might be able to do it, especially if we don't have selection.

And one of the things that, when I show you this, the graph of this, it's looks pretty much like the selection. This is picking up a lot of the selection and I'll show you this because, you can think of a specification check of looking at the lag 20 that you're doing.

Because that's showing us what the economic activity was the year before it was open. And what I'm gonna show you is that looks flatlined, across our data. We just don't see any differential-

>> Participant 8: Pre transcessor.

>> Eric Bettinger: Yeah, it's actually a beautiful graph, it's like my favorite graph, this whole paper.

So you can argue this one, right?

>> Participant 8: If there's not selection, is there, like political favor thing here? This was the Merced story, right? Why'd it go? Merced should be unrelated to economic activity.

>> Eric Bettinger: That's open, I mean, it's definitely unrelated, or at least it looks like it in that your view is.

 

>> John Cochrane: They are not putting campuses in places where you want to put campuses, namely places that are.

>> Eric Bettinger: I think that they're thinking about it, and we'll try to get at the other one. But what I'm saying is, when you look at the data itself, if that growth, it might be that there's a projection of growth that hasn't as of yet.

 

>> Participant 9: The way macroeconomists, deal with this is you put lags of the economic activity in as controls, because that's how people think, in VARS, we put lags in only for identification. But no, we're trying to control for pretrained.

>> Eric Bettinger: Exactly.

>> Participant 2: But there is, in fact, Justice Shapiro at Harvard and Carl, trying to account for all of these, there's pretrained corrections of these versions of different.

If you run into that, they're exactly trying to control both for persistent pretrend effects at the RV level. And it's a very cute thing you can do in this data and of course, if the effects are robust.

>> Eric Bettinger: Correct.

>> Participant 2: It survives it all.

>> Participant 9: But you want lagged white hats in there.

 

>> Participant 2: That's the idea it's part of, yeah.

>> Participant 8: I'm curious about this addiction off of physical pictures, right? So if I were taking satellite images of Detroit, I bet, I think a lot more people live in Detroit than live there because houses don't come down for a really long time.

 

>> Eric Bettinger: Yeah.

>> Participant 8: And economic activity fluctuates more than residents.

>> Eric Bettinger: I mean, so one of the things that we're gonna favor, right, is a place where we're getting better pictures of suburban areas, and that's where a lot of the branch campus is. So if you think of it almost like, yeah, I don't wanna use compliers here, cause Blyers is really about complying with the treatment.

But it's almost like in terms of thinking about the cleanliness of our measure, we still get variation in Houston, right? But Houston's mostly built out. And so you don't see, over this period a lot of variation there. The variation in why that we're really going to be identifying off, is when it's really happening in the suburbs, which is where those branch campuses are being.

We tried to think a little bit about kind of gentrification and how to think about that. We never figured out a way because, like, in Detroit. Detroit's a good example because a lot of the actual branch campuses that are formed in Detroit, have actually formed in old factories and old businesses.

Take an old Walmart, build some partitions, and suddenly you have a classrooms and you have a branch campus. And so that's a place we don't have, we're not gonna be able to do that part.

>> Participant 18: You do some sort of just urban rural separation. I imagine a lot of the stuff that's happening in Texas, for instance, in rural areas would be like oil and things like that would have nothing to do with.

Whereas you keep talking about Dallas and Houston and places like that, or suburban, rural Tibet.

>> Eric Bettinger: Exactly, beginning period. I mean, the returns to higher education were probably negative in Texas at the beginning of this.

>> Participant 19: So I wanted to come back to the issue of what's the net effective of all this?

You've got this different ways you're trying to see if there was a boost, from four to after there's a shift, etcetera. Are you able to think about or try, have you tried to disentangle how much that's a net add for the state or country or versus subtraction? Less in Dallas, there's less in San Antonio?

 

>> Eric Bettinger: Not yet, we haven't done that. I mean, right now we're, we're living off that variation across areas, right? So if it's a zero sum game and we're just moving economic activity.

>> Participant 19: Much harder to identify.

>> Eric Bettinger: Yeah, so that part I don't have here.

>> Participant 19: Well, that's fine, I've seen some.

 

>> Eric Bettinger: Not at all. So let me get a start on the results here. And so if you look at the results here, I mean, essentially what you see is, kind of positive significance. The magnitudes in these ones, I'm going to try to interpret those here after I show you all of them because remember, we've standardized this metric.

It's like a standard deviation unit of log GDP. And so right now I want to just focus on the positive or negative and the significance levels here. And so we're getting things going. And right here, at least in this one, it's all happening through the two year branches.

That's where the lion's share in Texas, you're seeing it in both the types of branch campuses there are. This is the place where I love the picture. So this is the results when you're looking before and after, blue or before it was established, red is once it comes into play.

Now in both cases, what you get at the beginning period. Now Tennessee is a little bit messy over there, but with Texas, you get this nice cluster around there. No clear pre trend going on here. And then once it happens, it starts to go up. You must not have 20 year leads and lags for every school because some happen.

 

>> Participant 8: So have you tried a balance thing to see if it looks the same?

>> Eric Bettinger: I haven't yet.

>> Participant 8: Shortened the window, choose only those that have all to minus ten to plus ten or whatever.

>> Eric Bettinger: Yeah, that's one that we can easily do, we haven't done it yet.

I mean, part of that is reflected in the shaded bars. Clearly when we get over to the tails, those standard.

>> Participant 8: Composition changing it's not.

>> John Cochrane: Do you take out the direct, because of the community college itself, I mean, the state is gonna spend $10 million and add $10 million in GDP right there.

 

>> Eric Bettinger: So this is where we can actually exclude the actual place where they're located and so you essentially think of it like a doughnut estimator, right? Then I'm just gonna take out the treated area and just focus on the census tracts around that one.

>> John Cochrane: So we take out the area that has the actual campus in it.

 

>> Eric Bettinger: Exactly.

>> John Cochrane: How much of that is directly spelled like the houses of the people who live there?

>> Eric Bettinger: That's fair.

>> John Cochrane: Taco stands and all the other stuff.

>> Eric Bettinger: That's fair, I mean, one of the things we could also think about, which I don't think we've done yet, and my co authors there, I should be acknowledging at the beginning.

Ushi and Patrick, you can see. You guys can wave here. Patrick and Ushi, everybody can see you.

>> John Cochrane: You took out the university itself.

>> Eric Bettinger: We took out the university itself.

>> John Cochrane: We're not just directly.

>> Eric Bettinger: But I might be able to take the contiguous, right?

>> John Cochrane: Yeah.

 

>> Eric Bettinger: Okay, so make it slightly bigger Donut.

>> John Cochrane: You also know how much the state spent on the university. So, to these questions of value and, is a university parking lot from another parking lot.

>> Eric Bettinger: Yeah, I mean-

>> John Cochrane: They'll subtract out. Here's $10 million box.

>> Eric Bettinger: I mean, the hardest part that, is I don't always see those numbers in terms of university.

 

>> Participant 20: This comparison between Texas and Tennessee is interesting in multiple respects. The numbers are much bigger and more persistent in Texas. So, one interpretation, one possible interpretation is they're just better at Texas, at predicting where growth will happen. Another interpretation is they're offering more useful skills on the margin, at these new units in Texas and Tennessee.

The fact that there's such a big difference between these two states suggests there's probably lots of. Even if you interpret this causally, there's probably lots of true heterogeneity within states. Back to the issue of what are we really measuring here? If you take your exogenous assumptions required to interpret this causally, at face value, this is some average treatment effect over treatments that we don't know what they are.

 

>> Eric Bettinger: That's right.

>> Participant 20: That's another way to state the concern I was trying to express earlier.

>> Eric Bettinger: No, I think that's valid, in the case in Tennessee, that as you kind of. Just as you start to unpack those numbers, I mean, Tennessee was a lot more about access than it was about accommodating new enrollments.

And so one of the things that happens in Tennessee is you start to think about, especially in the longer period of time, what you're really trying to do is establish some. It's more of an equity agenda that you're trying to establish that every student is within a certain mileage of the campus.

And so some of those campuses, it's not surprising to me that the effect doesn't persist and is smaller across the board.

>> Participant 8: Isn't that better variation for your experiment then, in some sense, then?

>> Eric Bettinger: And the fact that you still get these kind of positive numbers at least for the first decade, for Tennessee, I mean, is suggesting.

 

>> John Cochrane: Are the numbers comparable? You said they were scaled by standard deviations, sort.

>> Eric Bettinger: They're standardized. I mean, these are essentially kinda a standard deviation unit.

>> John Cochrane: So they're standardized by something different.

>> Participant 8: Of the state's GDP or both combined.

>> John Cochrane: They are standardized by this by the.

Patrick, you might have to help me out here. We're standardizing those within the actual county themselves.

>> Participant 8: It may not be comparable.

>> Patrick Lenhert: So, the prediction model is going on for all the US. So, we're predicting this outcome for all the US. So we're standardizing across all the US.

 

>> Participant 8: Okay, so it may not be bigger.

>> Participant 3: Numbers are comparable.

>> Participant 8: Yeah, okay.

>> John Cochrane: Yeah, now they are.

>> Participant 20: Did you look at trends and projections, populations in the two states?

>> Eric Bettinger: Texas is growing, Tennessee, isn't it growing?

>> Participant 20: Especially recently, but probably not attractive.

>> Eric Bettinger: I don't know that offhand.

Let me just get that. This is where the census data or the ACS data will help us towards just trying to think about what those numbers look like in terms of. Partly it's that from looking at that local area, but also it's partly at Texas. The university Texas system has to decide what to do.

 

>> Participant 20: They project they're gonna have a lot of enrollment, they don't have enough room. They're actually preventing kids from coming from those areas to Austin, by building something there. So, I just, I think that's something you probably wanna look at.

>> Eric Bettinger: And going back to one of the earlier comments, I mean, one of the things that we don't have here and I don't have a way to do it.

We might find one, but I mean we haven't yet, is there's two things going on, right? You have a recapture of your students who previously were going there, but now want something more convenient. So, those are students who are already going to college, the treatment that they have.

And then we have a second group that might be empowered to go to college because now there's a campus nearby. And so, I haven't been able to back out those numbers to try to figure out some estimate.

>> Participant 20: There was a big move in California by Jerry Brown in his second term, to devote resources to have more people, go to two year colleges and transfer to UC in the Cal state system.

That's in a sense kind of, maybe a quasi exogenous or at least.

>> Eric Bettinger: Well, and remember.

>> Participant 20: And I've never seen an analysis of that as possible to look at California as well.

>> Eric Bettinger: Yeah, that would be, I mean, across the state, they're not the only state that have done that.

I mean, almost every state, somewhere in the kind of two thousands started to advocate that because they couldn't, they didn't have the capacity. Yeah, well, it was also a capacity issue. It's a good way to mask dealing with a capacity issue as well. In this case here, part of what we're really identifying off of here are these ones that are popping up in suburban areas, and we're seeing that economic growth come in that suburban area.

But it's very highly correlated, as you can see here, with the likelihood that that was. So, again, I like your analysis of the either guests very well, or there's really transfer skills. If we can back that out, it'd be better.

>> Participant 21: Is there a way where you can measure what skills they're producing at those colleges and then compare to the demand?

Cuz it looks like we've got one state that is, has a demand that is being met in one state that is really providing access to students. And that's.

>> Eric Bettinger: Our hope is that at some point in time that we'll be able to distinguish some types of buildings, because education and health, tend to have different footprints.

We haven't figured out yet how to do that, whether we can train the satellite data to show us that level of detail. You focus on a smaller set of schools.

>> Participant 22: You might just be able to do either, a Yemenite, go onto their websites and see what departments they've got or what they're offering.

Or do some kind of survey or something like that to see what kinds of targets they've got.

>> Eric Bettinger: Yeah, you'll see, as I put, when I show my kind of last slide here of the next things it's trying to get, see if we can get more disaggregated data within the actual branch campuses themselves.

Let me just show you the iv here to put that up there. The iv looks good, this is only Texas. Stronger magnitudes, also less precision, as you might expect, to try to reverse it and try to say, well, how do you do it? How do you go back to go from these standardized coefficients, to something that's more interpretable?

The hard part here is you don't really have the constant at the census tract level. So, there's a couple different ways. A conservative estimate might be to just take the constant and basically uniformly distributed across the census tracts. Another way is to do it according to where the buildup area is.

So it's just kind of a weighted average. And so if you do that, this is just taking our, these are coefficients that I've showed you in these different, three different strategies here. And then on the right and the left are kind of reverting things back to log GDP instead.

And so when you look at these magnitudes, you know the magnitudes when you get down to Texas and the iv, the magnitudes are pretty good. As you start to think about these other metrics that we have here, the conservative estimate is, we're talking about kind of a 4% growth that kind of persists, for about a decade.

And then starts to drop off.

>> Participant 2: Can I just ask a clarification question, what are the two columns conservative?

>> Eric Bettinger: So the conservative is, I'm assuming that the kind of county level constant is uniformly distributed. So in the rural areas, I'm kind of giving them a little bit too much credit for producing it.

The liberal estimate is, I'm basically saying that all the economic activity corresponds to the satellite data that we have. And so that's kind of our best guess at trying to figure out how to convert these into log GDP numbers. Again, there's both good and bad here. When you start to think about in the Tennessee, if you're telling me that two year campuses, that's where the action was in Tennessee, are giving you a bump of about 2%, 4% when it's built, that seems reasonable to me.

When you get up to the IV numbers, the numbers, especially the kind of pooled number, a little larger than I would have expected, a lot larger than I would have expected. So other things we've done just to kind of, we've talked about the doughnut, one of our showed you.

I showed you here a 25 miles radius, I can do a 10 mile, 40 miles, same things. When you get the smaller and smaller radius, it gets harder and harder to find a things that are happening, at least in Tennessee, we lose a lot of power. So, and then this is really become my last slide of the group, just to tell you where we're going in this.

Understanding that two and four year difference, I think that gets to Steven's. Steve's word about, the kind of skill differential and where we're doing things, we made the mistake of using a 2020 census tracts. And problem with 2020 census tracts, right, is they're somewhat endogenous, so if the economic activity has developed.

I put 1990, but it's actually 1980, that we're trying to move back and use those and recalibrate it. Using the ACS and the census data, try to get education levels, the nature of occupations in those areas, that's the next one that we're also doing. And then trying to extend that database in the college grants for both their locations and also their functionality.

So, I think that's my last one here, I think everything else is just kinda summary of what we've done here. And so I'll just leave this slide up here, this is where we're at, and just kind of look forward to some discussions here for the rest of the time we've got.

 

>> Participant 20: Two things, one is that I'm really concerned looking at the tables, that you've got influential observations because the convo data, it's not uncommon to have a year over year GDP change for a county of 100% and a decline of 50%. And so like when we think of GDP.

So I just looked up 2021 and it's 81.8% was the max for Cote County, Texas -35% Chateau County, Montana were the bigs in 2021. So it's not the GDP that we're used to, right, there's these real weird. And so I'm kind of worried, given the scale of your parameters, that there's like some, county Texas put a community college.

And so I just think that,-

>> Participant 8: You might run sort of a more robust for the prediction of GDP from the physical stuff. The predictions shouldn't have those fluctuations, as long as the prediction regression.

>> Participant 9: Although I noticed when you showed that East Germany, the prediction had more variation than the official data.

Now, I don't know if that's before but the thing about the prediction should be smoother than the actual,

>> Eric Bettinger: The volatility that we're seeing, that's a good point.

>> Participant 22: But my second thing is just a real quick observation is there's gotta be data on the for profit. A million for profits closed, right, because they crack down on them.

And so if you've got a place where a for profit closes, then maybe that's where they stick a branch and it might give you a nice instrument or something if the for profits close for exogenous reasons. But there were a million of them that closed all around.

>> Eric Bettinger: I mean, the hardest part that I have with that one is, I mean, during this period of time, the bride calls me on the phone and they're, look, here's our plan, what we're trying to do.

I mean, they basically were tracking, help wanted in prison and in hospitals. And so they pop up a campus until they see that the help wanted basically drop and then close the campus. So a lot of the opening and closing of some of these for profits are much more endogenous and systematic.

 

>> Participant 23: Bankruptcy was related to explicit department of education policies where they blacklisted some companies.

>> Eric Bettinger: Yeah, the question is where we can figure out all our campuses.

>> Participant 8: I just get two issues, one is, my guess is, you sort of talked about suburban campuses versus the ones that weren't suburban, I'm kind of interested in the differences for two reasons.

One is you might think that in these far from urban areas, that's a population that isn't very mobile, so those might be more people more likely to stay once trained. Whereas if I come to the main campus, I might disperse even to other states, right, so that's an interesting comparison.

I also struck what Michael said about, if people go to Plano instead of Austin, then you get fewer people coming to Austin. And if the glomeration is convex, which I don't know if we think it is, but at least other policies seem to believe it is, that trade off is quite interesting.

The suburban campuses shouldn't detract from that because presumably suburban Austin is gonna lead to the same elaboration of the rest of Austin.

>> Eric Bettinger: That's right.

>> Participant 8: But you put it 100 miles away and it's convex, it could be worse than zero sum.

>> Eric Bettinger: I mean, so a different way is going back to this point about the urban and the rural creating a distinction between.

I think at one point we ran, we were trying to actually think about the distance to the four year and returning Jack, to those models. Cuz part of the getting at that, we think that the agglomeration economies are actually coming because of that R&D that's being generated from the four year campuses.

And so one of the things that we could do here is actually measure what their kind of starting distance is from the main campus and interact that part to see if we're getting that. Because what we should see, right, if we're closer to that campus, then it should actually change the nature of the effect, cuz now we're picking up a different effect, hopefully, or contributing to something else.

Agglomeration, that's a nice idea like that, yeah.

>> Participant 15: So you could be a lot more disaggregated on your outcome side. And instead of your GDP proxy, you could look at measures of establishment counts or even employment, our detailed industry sector, and there's at least two ways you could do that.

One is the county business patterns data is highly granular at the establishment count, county level by year. Those are high quality data, they're limited. Or you could go to the nets, which I think are a little lower quality, but then you could define the areas as you saw fit cuz you know the zip code of the location in the nets.

And you could also get a measure of employment from the nets as well. So, yeah, so then you could tell us whether increases in economic activity that you are finding are in concentrated in sectors where you think the skills transmission to the local labor force is actually plausible.

Or if it's coming from something else entirely unrelated, which would kinda push you towards just endogeny.

>> Eric Bettinger: So in some sense, if I could run the same regression that we have here. But run it at the kind of a sector level in terms of child's detail.

>> Participant 15: Like four digit industry, very detailed.

There's hundreds of them.

>> Eric Bettinger: Yeah.

>> Participant 8: Leave out the education sector.

>> Participant 15: Leave out the education and give us some sense of where this is showing up.

>> Eric Bettinger: Okay.

>> Participant 15: And if you could in turn get crude information on what skills they're actually imparting at these campuses, that would allow you to construct even more.

 

>> Participant 2: You mentioned, Eric, that you had occupation information, that's a good proxy.

>> Participant 9: Yeah.

>> Eric Bettinger: No, the census has that data.

>> Participant 2: So you may not know the major about the graduating body.

>> Eric Bettinger: Yeah.

>> Participant 2: Especially, if you have it at that fine level.

>> Eric Bettinger: At that final, yes.

 

>> Participant 2: Yeah.

>> Participant 20: And I think one thing you may be grappling with is particularly this post COVID era, the remote learning. So you're gonna be dealing with a lot of, I think, dampening of the local economic effects. Because these people are taking these classes remotely. And I think that may have some effect and dampening effects.

The other thing that I thought was interesting is I thought when looking at Texas versus Tennessee. I thought the Tennessee data was pretty much what I would have predicted. And that is, in the year before the new campus comes on. You would expect to see a lot of construction activity, all this other stuff.

And the year after this is what you see in Tennessee, it goes down because all those construction workers are gone. You did not see that in Texas, which kinda surprised me a little bit.

>> Participant 20: Yeah, and I don't know how much that had to do with what they were building.

And maybe in Texas they were taking over the Walmarts, whereas in Tennessee they were actually building. But I find the Tennessee data to be much more what I would have predicted.

>> Eric Bettinger: Yeah.

>> Participant 24: I also think the scale is a little bit deceptive, too, because the scales are different.

And so Texas has a wider range than they're showing. And so you tricked me at first, but then when I looked more closely.

>> Eric Bettinger: Sorry.

>> Participant 24: That's okay.

>> Participant 2: I'm sorry, I just wanted to add something about occupation. So there is the ONAD specification. So now there are very granular type of classifications of occupations by the general human capital intensity requirement of it.

So whether they're routine, cognitive.

>> Eric Bettinger: Sure.

>> Participant 2: And map them back into the type of bundles of skills that you must have acquired to qualify for.

>> Eric Bettinger: Do we have that data and enough to disaggregate that.

>> Participant 2: All I know they all came to campus for a while last spring.

So he's engaged into a hundred years long census.

>> Eric Bettinger: Okay.

>> Participant 2: So they're trying to go back in time and try. So it's more granular time wise in geography class.

>> Eric Bettinger: Okay.

>> Participant 2: But he has a lot of.

>> Participant 25: So Eric, combining most recent comment and the spirit of where I've been asking.

Look at NAICs, four digits in these areas. You can also compare them to what's going on in the other areas in Texas, kind of a spirit of a bargain.

>> Eric Bettinger: Okay.

>> Participant 25: That would be a very informative shifting, replacing people industry by industry. Is it shifting people from one to another?

Combine that with some sense of projected or predicted popular would have happened otherwise. I think that would all be extinguished. And I also wanna say that while you're getting lots of comments here. Once said every good piece of research raises more questions.

>> Participant 25: So he viewed himself as a historian of economic thoughts.

You've raised a lot, got me thinking about a lot of things. So I think kudos for that. You've heard about 6 or 10 or 12 different things that the kind of research you're doing might be a piece of information that would be useful in answering your question. And I kind of agglomerate those, get a list of those as a motivational thing at the beginning.

And you're hearing around the table that you don't have a complete answer to those things.

>> But I think it's a piece of answers to a lot of different puzzles that would be useful for you in your research to generate interest from a variety of people interested in different of those things.

 

>> Participant 25: Appreciate that. Thank you.

>> John Cochrane: I've a question, two issues. One issue is that, is this just stimulus? If we had put a amusement park in there, would that have had the same.

>> Eric Bettinger: Yeah.

>> John Cochrane: And cross-sectional stimulus?

>> Eric Bettinger: You might be able to actually identify the music.

 

>> John Cochrane: And on that same thing, I wonder, I guess census has the numbers. It would be lovely to see. And the mechanism has to be this produces more people with higher education.

>> Participant 8: One alternative mechanism is just people move in who have higher educations, but this lowers the cost of getting it.

Now, you don't necessarily want a fraction cuz in Texas especially, there's a lot of people moving into Texas who have no education. But is the raw number of people with higher education increasing? That would be really confirmatory that this is the mechanism for something else.

>> Eric Bettinger: Yeah, I mean, like I said, the different things we've talked about so far and that's one that I've got.

RA is trying to chase down right now.

>> Participant 15: So I think it would be a good study comparatively to look at the number of people who leave an area, get an education and don't come back. Because I think that's the alternative that this applies. When I go out of state and I meet a significant other, do other things, chances of me going back to my home state after I've got that four six year education is much smaller.

If I do all those activities in my early twenties in my home county.

>> John Cochrane: These are places people and kinds of degrees where people are working part time, leaving at home and so forth. So the argument that this substantially lowers the cost of them getting a degree by staying local is probably strong.

But, yeah, I entirely agree with you.

>> Participant 15: The hardware of going back and trying to figure out how to disaggregate that because like in the census data. I mean, doesn't actually if you live in the same house. It's much more cuz my family is from rural Utah, and it's two hours ago to Salt Lake and that's kinda what everybody did.

And the people who did well like they didn't come back. So there was this brain drain from the rural to the urban that.

>> Eric Bettinger: What part of Utah were you?

>> Participant 15: The Uinta Basin, Vernal, Roosevelt area. But so like I saw in my cousins, I got a 20.

 

>> Eric Bettinger: Yeah.

>> Participant 15: The ones that were kinda like smart and did well. They're the ones who didn't come back.

>> Eric Bettinger: Yeah.

>> Participant 15: And that's just the argument is that putting the university there gets the brain drain from other places.

>> Eric Bettinger: But I think they're trying to get at that.

I mean, they might not be able to go all the way back to 84, but the coordinating board. Well, some of the data centers actually can tell me what high school they went to and what the zip code is that they're working at now. Now, whether I can disaggregate that and try to look at the difference between.

I might be able to do that at the high school level. Okay, you pre-post what the division is between students, that's possible.

>> Participant 8: Doesn't Tennessee have all that stuff linked to employment data too? I thought it's one of those states where people linking education and bonding and welfare.

 

>> Eric Bettinger: So one of our Toronto, I don't know if she's on or not, is Madison Dillon. She just finished her PhD here last year. And she now actually works for the state. So I'll ask Madison if there she has it. I mean, most definitely, like these ERCs that the Texas run, it's more capable.

And given that I feel better about my identification in Texas and Tennessee, I'd rather put my eggs in that basket.

>> Participant 7: I remember my dad helped start campus, Georgia Tech, Savannah, which ended up closing that it was an expansion campus. But there were all kinds of government estimates that were floating around about how much money that would generate.

It would just be interesting to see if they're doing any of these kinds of projections, if they actually wanted.

>> Eric Bettinger: When I was at the coordinating board in July and I was kinda talking about some of these initial results, just having conversations about them. I mean, they were arguing that or saying that as they consider applications.

Now, in that kinda second category where people are giving them numbers and throwing around numbers. I would think of it more like a straw man that I'm gonna tear down in the papers. Is there a literature?

>> Participant 8: I see you recall a couple papers on getting a land grant college, whatever the identification strategy was that sort of these kinda one time effects on local communities that you could just compare magnitudes to?

 

>> Eric Bettinger: Yeah, I can go and do that, I haven't.

>> Participant 8: Remember seeing something recently. I don't know this literature.

>> Eric Bettinger: I mean, the hard part is a lot of that, I don't know if I believe in a lot of that.

>> Participant 8: I think there's a couple that had some clever identification.

 

>> Participant 2: I can't remember.

>> Participant 8: A lot of schools were from a long time ago, yeah.

>> Participant 2: Yeah, they were a long time ago.

>> Participant 8: I seem to recall some papers with some identification strategy.

>> Participant 8: Yeah, I remember as well, I just remember being, yeah, you're hard to identify.

 

>> Eric Bettinger: Yeah, we're gonna end there. And if you've got other comments or questions, please come on up and talk to me. And more generally, I'm kind of new to the Hoover Community, so. I love to meet people and start to interact. Trying to move my life from all in the Ed School to having a little bit more of it over here.

 

>> Participant 2: So I appreciate it.

>> Eric Bettinger: I will tell you that the faculty meeting here, when we did appointments. And then the next day going to the Ed School were two different experiences.

>> Eric Bettinger: But I thank you so much. I really appreciate your time today.

 

Show Transcript +

Upcoming Events

Thursday, November 21, 2024
Language Model Detectors Are Easily Optimized Against
The fourth session of Challenges and Safeguards against AI-Generated Disinformation will discuss Language Model Detectors Are Easily Optimized… Annenberg Conference Room, Shultz Building
Wednesday, December 4, 2024
Ideas-Uncorked
Health Care Policy In The New Trump Administration And Congress
The Hoover Institution in DC hosts Ideas Uncorked on Wednesday, December 4, 2024 from 4:45–6:00 pm ET. The event will feature Hoover Institution… Hoover Institution in DC
Friday, December 6, 2024
Emerging Technology and the Economy
The Hoover Institution cordially invites you to attend a conversation with President and CEO of the Federal Reserve Bank of San Francisco, Mary C.… Shultz Auditorium, George P. Shultz Building
overlay image