Category Archives: Education

The 2018 John Bates Clark: Parag Pathak

The most prestigious award a young researcher can receive in economics, the John Bates Clark medal, has been awarded to the incredibly productive Parag Pathak. His CV is one that could only belong a true rocket of a career: he was a full professor at MIT 11 years after he started his PhD, finishing the degree in four years before going to Paul Samuelson route through the Harvard Society of Fellows to become the top young theorist in Cambridge.

Pathak is best known for his work on the nature and effects of how students are allocated to schools. This is, of course, an area where theorists have had incredible influence on public policy, notably via Pathak’s PhD Advisor, the Nobel prize winner Al Roth, the group of Turkish researchers including Atila Abdulkadiroglu, Utku Unver, and Tayfun Sonmez, as well as the work of 2015 Clark medal winner Roland Fryer. Indeed, this group’s work on how to best allocate students to schools in an incentive-compatible way – that is, in a way where parents need only truthfully state which schools they like best – was adopted by the city of Boston, to my knowledge the first-time this theoretically-optimal mechanism was used by an actual school district. As someone born in Boston’s contentious Dorchester neighborhood, it is quite striking how much more successful this reform was than the busing policies of the 1970s which led to incredible amounts of bigoted pushback.

Consider the old “Boston mechanism”. Parents list their preferred schools in order. Everyone would be allocated their first choice if possible. If a school is oversubscribed, some random percentage get their second choice, and if still oversubscribed, their third, and so on. This mechanism gives clear reason for strategic manipulation: you certainly don’t want to list a very popular school as your second choice, since there is almost no chance that it won’t fill up with first choices. The Boston mechanism was replaced by the Gale-Shapley mechanism, which has the property that it is never optimal for a parent to misrepresent preferences. In theory, not only is this more efficient but it is also fairer: parents who do not have access to sophisticated neighbors helping them game the system are, you might reason, most likely to lose out. And this is precisely what Pathak and Sonmez show in a theoretical model: sophisticated parents may prefer the old Boston mechanism because it makes them better off at the expense of the less sophisticated! The latter concern is a tough one for traditional mechanism design to handle, as we generally assume that agents act in their self-interest, including taking advantage of the potential for strategic manipulation. There remains some debate about what is means for a mechanism to be “better” when some agents are unsophisticated or when they do not have strict preferences over all options.

Competition for students may also affect the quality of underlying schools, either because charter schools and for-profits compete on a profit-maximizing basis, or because public schools somehow respond to the incentive to get good students. Where Pathak’s work is particularly rigorous is that he notes how critical both the competitive environment and the exact mechanism for assigning students are for the responsiveness of schools. It is not that “charters are good” or “charters are bad” or “test schools produce X outcome”, but rather the conditionality of these statements on how the choice and assignment mechanism works. A pure empirical study found charters in Boston performed much better for lottery-selected students than public schools, and that attending an elite test school in Boston or New York doesn’t really affect student outcomes. Parents appear unable to evaluate school quality except in terms of peer effects, which can be particularly problematic when good peers enroll is otherwise bad schools.

Pathak’s methodological approach is refreshing. He has a theorist’s toolkit and an empiricist’s interest in policy. For instance, imagine we want to know how well charter schools perform. The obvious worry is that charter students are better than the average student. Many studies take advantage of charter lotteries, where oversubscribed schools assign students from the application class by lottery. This does not identify the full effect of charters, however: whether I enter a lottery at all depends on how I ranked schools, and hence participants in a lottery for School A versus School B are not identical. Pathak, Angrist and Walters show how to combine the data from specific school choice mechanisms with the lottery in a way that ensures we are comparing like-to-like when evaluating lotteries at charters versus non-charters. In particular, they find that Denver’s charter schools do in fact perform better.

Indeed, reading through Pathak’s papers this afternoon, I find myself struck by how empirical his approach has become over time: if a given question requires the full arsenal of choice, he deploys it, and if it is unnecessary, he estimates things in a completely reduced form way. Going beyond the reduced form often produces striking results. For instance, what would be lost if affirmative action based on race were banned in school choice? Chicago’s tier-based plans, where many seats at test schools were reserved for students from low-SES tiers, works dreadfully: not only does it not actually select low-SES students (high-SES students in low-income neighborhoods are selected), but it would require massively dropping entrance score criteria to get a pre-established number of black and hispanic students to attend. This is particularly true for test schools on the north side of the city. Answering the question of what racial representation looks like in a counterfactual world where Chicago doesn’t have to use SES-based criteria to indirectly choose students to get a given racial makeup, and the question of whether the problem is Chicago’s particular mechanism or whether it is fundamental to any location-based selection mechanism, requires theory, and Pathak deploys it wonderfully. Peng Shi and Pathak also do back-testing on their theory-based discrete-choice predictions of the impact of a change in Boston’s mechanism meant to reduce travel times, showing that to the extent the model missed, it was because the student characteristics were unexpected, not because there were not underlying structural preferences. If we are going to deploy serious theoretical methods to applied questions, rather than just to thought experiments, this type of rigorous combination of theory, design-based empirics, and back-testing is essential.

In addition to his education research, Pathak has also contributed, alongside Fuhito Kojima, to the literature on large matching markets. The basic idea is the following. In two-sided matching, where both sides have preferences like in a marriage market, there are no stable matching mechanisms where both sides want to report truthfully. For example, if you use the mechanism currently in place in Boston where students rank schools, schools themselves have the incentive to manipulate the outcome by changing how many slots they offer. Pathak and Kojima show that when the market is large, it is (in a particular sense) an equilibrium for both sides to act truthfully; roughly, even if I screw one student I don’t want out of a slot, in a thick market it is unlikely I wind up with a more-preferred student to replace them. There has more recently been a growing literature on what really matters in matching markets: is it the stability of the matching mechanism, or the thickness of the market, or the timing, and so on.

This award strikes me as the last remaining award, at least in the near term, from the matching/market design boom of the past 20 years. As Becker took economics out of pure market transactions and into a wider world of rational choice under constraints, the work of Al Roth and his descendants, including Parag Pathak, has greatly expanded our ability to take advantage of choice and local knowledge in situations like education and health where, for many reasons, we do not use the price mechanism. That said, there remains quite a bit to do on understanding how to get the benefits of decentralization without price – I am deeply interested in this question when it comes to innovation policy – and I don’t doubt that two decades from now, continued inquiry along these lines will have fruitfully exploited the methods and careful technique that Parag Pathak embodies.

One final note, and this in no way takes away from how deserving Pathak and other recent winners have been. Yet: I would be remiss if I didn’t point out, again, how unusually “micro” the Clark medal has been of late. There literally has not been a pure macroeconomist or econometrician – two of the three “core” fields of economics – since 1999, and only Donaldson and Acemoglu are even arguably close. Though the prize has gone to three straight winners with a theoretical bent, at least in part, the prize is still not reflecting our field as a whole. Nothing for Emi Nakamura, or Victor Chernozhukov, or Emmanuel Farhi, or Ivan Werning, or Amir Sufi, or Chad Syverson, or Marc Melitz? These folks are incredibly influential on our field as a whole, and the Clark medal is failing to reflect the totality of what economists actually do.

Advertisements

“Designing Efficient College and Tax Policies,” S. Findeisen & D. Sachs (2014)

It’s job market season, which is a great time of year for economists because we get to read lots of interesting papers. The one, by Dominik Sachs from Cologne and his coauthor Sebastian Findeisen, is particularly relevant given the recent Obama policy announcement about further subsidizing community college. The basic facts of marginal college students are fairly well-known: there is a pretty substantial wage bump for college grads (including ones who are not currently attending but who would attend if college was a little cheaper), many do not go to college even given this wage bump, there are probably externalities both in the economic and social realm from having a more education population though these are quite hard to measure, borrowing constraints bind for some potential college students but don’t appear to be that important, and it is very hard to design policies which benefit only marginal college candidates without also subsidizing those who would go whether or not the subsidy existed.

The naive thought might be “why should we subsidize college in the absence of borrowing constraints? By revealed preference, people choose not to go to college even given the wage bump, which likely implies that for many people studying and spending time going to class gives negative utility. Given the wage bump, these people are apparently willing to pay a lot of money to avoid spending time in college. The social externalities of college probably exist, but in general equilibrium more college-educated workers might drive down the return to college for people who are currently going. Therefore, we ought not distort the market.”

However, Sachs and Findeisen point out that there is also a fiscal externality: higher wages equals higher tax revenue in the future, and only the government cares about that revenue. Even more, the government is risk-neutral, or at least less risk-averse than individuals, about that revenue; people might avoid going to college if, along with bumping up their expected future wages, college also introduces uncertainty into their future wage path. If a subsidy could be targeted largely to students on the margin rather than those currently attending college, and if those marginal students see a big wage bump, and if government revenue less transfers back to the taxpayer is high, then it may be worth it for the government to subsidize college even if there are no other social benefits!

The authors write a nice little structural model. People choose to go to college or not depending on their innate ability, their parent’s wealth, the cost of college, the wage bump they expect (and the variance thereof), and their personal taste or distaste for studying as opposed to working (“psychic costs”). All of those variables aside from personal taste and innate ability can be pulled out of U.S. longitudinal data, performance on the army qualifying test can proxy for innate ability, and given distributional assumptions, we can identify the last free parameter, personal taste, by assuming that people go to college only if their lifetime discounted utility from attendance, less psychic costs, exceeds the lifetime utility from working instead. A choice model of this type seems to match data from previous studies with quasirandom variation concerning the returns to college education.

The direct benefit to the government from higher tax revenue from a subsidy policy, then, is the cost of the subsidy times the number subsidized, minus the proportion of subsidized students who would not have gone to college but for the subsidy times the discounted lifetime wage bump for those students times government tax revenue as a percent of that wage bump. The authors find that a general college subsidy program nearly pays for itself: if you subsidize everyone there aren’t many marginal students, but even for those students the wage bump is substantial. Targeting low income students is even better. Though the low income students affected on the margin tend to be less academically gifted, and hence to earn a lower (absolute) increase in wages from going to college, subsidies targeted at low income students do not waste as much money subsidizing students who would go to college anyway (i.e., a large percentage of high income kids). Note that the subsidies are small enough in absolute terms that the distortion on parental labor supply, from working less in order to qualify for subsidies, is of no quantitative importance, a fact the authors show rigorously. Merit-based subsidies will attract better students who have more to gain from going to college, but they also largely affect people who would go to college anyway, hence offer less bang for the buck to government compared to need-based grants.

The authors have a nice calibrated model in hand, so there are many more questions they ask beyond the direct partial equilibrium benefits of college attendance. For example, in general equilibrium, if we induce people to go to college, the college wage premium will fall. But note that wages for non-college-grads will rise in relative terms, so the net effect of the grants discussed in the previous paragraph on government revenue is essentially unchanged. Further, as Nate Hilger found using quasirandom variation in income due to layoffs, liquidity constraints do not appear to be terribly important for the college making decision: it is increasing grants, not changing loan eligibility, that will do anything of any importance to college attendance.

November 2014 working paper (No IDEAS version). The authors have a handful of other very interesting papers in the New Dynamic Public Finance framework, which is blazing hot right now. As far as I understand the project of NDPF, essentially we can simplify the (technically all-but-impossible-to-solve) dynamic mechanism problem of designing optimal taxes and subsidies under risk aversion and savings behavior to an equivalent reduced form that essentially only depends on simple first order conditions and a handful of elasticities. Famously, it is not obvious that capital taxation should be zero.

Advertisements
%d bloggers like this: