SPEAKER 1: Our next speaker is Alex Rees-Jones from our economics department. Interestingly, he has a bachelor's degree from Cornell and a PhD from Cornell.
But he did go away for a while, to the National Bureau of Economics and to Penn for a few years. But he's back. And I'd say welcome back, but it's-- you've been here for a while.
ALEX REES-JONES: Thanks. Yeah, yeah, it's good to be back.
SPEAKER 1: Yeah, but welcome to BEDR, Alex.
ALEX REES-JONES: Yeah. But the nice thing since when I moved away seven years ago is, there's been a fair bit of turnover, and there's many new faces. So it's nice to meet many of yours. I shouldn't say it's a nice thing there was turnover. It's nice that there's new people here.
I'm sorry for anyone who's not.
So I work, broadly, on trying to build behaviorally on ideas into the apparatus economists have for public policy. Like the previous speaker, I was a little daunted by the idea of trying to walk you through like five papers in 10 minutes, so I'm going to do something slightly different and just to be really abstract and tell you how I think about the world and give you a little bit of a hint of how it plays into some papers. But basically, just give you a gist about what I'm all about here.
A couple starting premises that I think are important. If you think about the enterprise that many people in this room are engaged in in either behavioral economics or judgement decision making, I would characterize us as having been extremely successful in influencing how people think about individual decision making. In psychological questions, this literature has been quite influential.
If you think about how influential we've been for affecting the way people think about fundamental economic questions, like econ 101 concepts, our success is a little more modest, I would argue. And that might be a contentious position in this room. But to illustrate what I'm talking about, if you think about just the way economics 101 or intermediate micro is taught now versus before the behavioral revolution, it's pretty much the same.
And it's worth thinking about why that's the case. And I would argue that a big part of the reason why that's the case is that doing what we're advocating for is extraordinarily tricky. So it's very easy for us to be right when we say, oh, existing frameworks aren't great. They're making assumptions that are unrealistic. So we can do that and be right. But proposing an alternative that does everything you want to do in policy analysis-- that's a much more demanding task. But I think behavioral economics is at the stage where that's the task we really need to be fundamentally engaged in.
So the reasons why, I would argue, that's difficult is, in order for this exercise to be useful for policy analysts as opposed to, say, psychologists is, whatever theory we come up with, it has to be predictive, tractable, and measurable. What I mean by that is, first, if we write down a theory, it has to be good at explaining the variance we see happening in the field in response to some policy. We have lots of ideas of specific heuristics and biases that clearly matter in some way or another-- default effects and so on.
But one of the big challenges here is that neoclassical economists made their lives really easy by assuming that people are extremely smart-- because there's typically only one or two ways to be really, really smart. And there's enormously varying ways in which you can be biased or heuristic and things like that. And so if we want to do a good job of explaining the rich variation we see in the world in a world where we think that JDM is really relevant, we have to account not for just one model, but many, potentially. So that's pretty hard.
Something that makes that even harder is that even as we're trying to do this, trying to account for all the richness that exists in the world, we need to ultimately come up with a theory of that that's tractable. And by that, I mean it's something where someone like me could sit down with a pencil and paper and try to figure out what the optimal policy should be given whatever theory you proposed. And if you propose something that's too complicated for someone of my OK but still limited cognitive abilities, then it's not going to be useful for policy analysis.
So we need to balance those two things. And then, even if we get those two things right, ultimately, we have to write the things down in a way where they depend on things that I can measure in the world. And this is a really difficult thing, as well, for behavioral economists, because a lot of the policies-- or, sorry, a lot of the models we've written out were developed to be studied in the lab.
So we've given ourselves the freedom to imagine you can experimentally vary all sorts of things and from that estimate your propensity for loss aversion, your propensity for present bias, et cetera. But now, if we reach the stage where we say, OK, I'm going to run with that and I want to use this to understand labor supply behavior or taxation or things like that, I need to be able to use these models, like consensus statement, where I don't have the same freedom as experimental economists do to measure things.
So trying to overcome these three things is basically the necessary thing you need to do if you're trying to build this into policy. I, as well as many other people, are trying to work on equity seriously right now and basically convert what we've done successfully in the lab to something that satisfies these goals enough to be able to use them seriously for policy questions. And the general template for that will always be starting with something-- I should have looked back earlier. I didn't realize it wasn't across the whole screen. Whatever.
You start out with a model that we've already worked with a fair bit, potentially a set of psychological models that we think are important. You're going to model the integration of those people into a market, into whatever environment you're trying to study. In that market, there will fundamentally be policy levers available. And you're going to think about how the social planner should think about them. And that involves really, formally writing out the things that you care about as a policy designer, the things that you don't want to happen, and thinking about how you want to optimize your choice of policy instruments given the behavioral ways people act.
So that's super abstract. To be a little more specific, let me go through a couple examples of domains that I've worked in a bit. Let's talk about taxation for a second.
So almost all of the work in this domain I've done has been thinking about optimal tax policy if people are a little more JDM-influenced than tax policy designers have typically thought. So people in public finance go through a lot of work to make you think that what they're doing is really complicated. And it's generally not. Normally, everything balances out to really small sets of trade offs that you're just trying to be very specific about.
So high-level trade off: you, as a policy designer, care about getting money to the government. The government's going to use money for various good things like building roads, having an army, redistributing, et cetera. But the whole reason that this is a tricky problem is, in order to raise money to do that, you're going to have to tax people. And taxing people distorts their decisions. And so you have this question of how to optimally trade off the need for money versus the need to not mess up people's decisions too badly.
Typical framework for asking that question is to say, well, let's assume that people fully optimize against whatever policy I design. There's all sorts of reasons to worry or to think that that won't be terribly reasonable for taxes, because in this room, or, indeed, in rooms full of tax economists, most people don't have sufficient knowledge or inclination to fully optimize against really complex tax systems.
So that's the main idea, thinking that this is a domain with a lot of ambiguity about what policy you face, about what you should do. And where ambiguity exists, that's where there's a lot of room for biases and heuristics to take hold. So rather than doing the typical trade off exercise, where you're thinking about how to optimize the collection of revenue versus the need to not distort decisions of optimizers, you can do the exact same type of exercise and imagine that you're going to be distorting the decisions of people who are subject to heuristics and biases that you've measured in some way or another in these markets.
So I've done several studies along that type of line. The interesting thing that comes out of it is often the integration of people with JDM proclivities changes the way you would think about public policy in these domains and often makes things that previously seemed impossible somewhat possible. So for example, if you're just totally ignoring taxes, it's easier to raise the money you need for the government. There's less distortion. In a lot of environments, you can actually get more redistribution done, et cetera. So there's a real fundamental change in the way you look at the problem if you start allowing for JDM to be relevant in this domain.
In the interest of time, I'm going to gloss over other environments. But I'm going to point out that the other domains I'm working-- looking at have this same kind of mechanism design framework where you have some policy goal. You design a policy, but you're worried about people optimizing against it. And you're going to change how you do that calculus if you think people are biased when they optimize against it.
That's the gist of what I'm trying to do. I understand I've been really abstract. I can talk with you all about details at extreme length if you want. But otherwise, I open the floor to questions.
AUDIENCE: Can I ask a question?
ALEX REES-JONES: Please.
AUDIENCE: [INAUDIBLE] so I'm a psychologist. Why do economists start with models, not just talk to people and ask them how they're actually going about the decisions that you want to model?
ALEX REES-JONES: So I'm very much on board with that for answering questions about psychology. And going back to how I started this, if your goal is not to answer questions about psychology but to answer questions-- what tax rate should I set or things like that, then what I need is not a gist of the process that you're using, but rather, something that pretty directly maps a proposed tax rate into how well-off or poorly off I think society will be.
And so I think when you're going through developing those models, the way you should begin is exactly with what you're talking about, and more generally, the approach that psychologists use as opposed to economists. But one end user of psychology are people doing these model-based exercises where I say, I'm going to take this really rich world you painted based on data like you just described and try to make it simple enough that I can answer more direct numerical questions-- understanding that it's going to be imperfect, but understanding that ultimately, someone out there is going to have to choose a number. And you want to give them as much guidance as possible.
AUDIENCE: So you're building, in an engineering sense, a model that predicts well. But again, my question is, why not have a better foundation?
ALEX REES-JONES: I'm in favor of good foundations and opposed to bad foundations.
AUDIENCE: That's what behavioral economics is all about, is trying to convince economists to improve our foundation.
SPEAKER 1: Yeah, I mean, going back to the extreme--
AUDIENCE: Hang on. I feel dumb, having asked that question.
ALEX REES-JONES: No, no.
Not to be too flippant about it, but the-- the existing foundations, going back to this-- we're just saying, we need a model of how individuals respond. Let's assume they're really smart, because that give us something to work with. And the exercise I'm talking about here is saying, let's not run with that foundation. Let's use, as a foundation, the JDM literature, but try to make it so that that literature is as close to-- as tractable as the original proposition as possible, understanding that-- the ideal is to have something that's as rich as the world is subject to the constraint that someone at the treasury can work with it. And that's the trade off we're trying to fumble towards.
AUDIENCE: So I guess I'm wondering how you think about-- for policy recommendations, you-- part of the reason people start with models is because zero is very easy to estimate. It's zero. There's zero bias-- that's easy to incorporate into a model. And so for things like the extent to which you ignore taxes, that's going to be on a spectrum.
And so we don't know exactly where and how much are they ignoring. And the issue is, experiments aren't really good at identifying effect sizes, necessarily-- how much, on average, do you ignore your taxes? So where does that sort of recommendation come from? How do you go about getting to those estimates that are tractable and generalizable?
ALEX REES-JONES: So I've run studies along the lines of what you're describing, of trying to measure, at the individual level, effect sizes of how much you attend to taxes. I agree with your characterization that most people have historically viewed experiments not to be geared towards estimating effect sizes, but to get at things more qualitatively.
I think some of the most interesting work being done in these domains now is actually thinking about experiments differently and saying, look, that statement was true back in the day, where we were doing, like, 30 student lab experiments, where obviously, you're not going to measure effect sizes precisely. But nowadays that it's actually relatively feasible to use experimental methods with 10,000 people rather than 30, it actually becomes not crazy to say, we can estimate relatively precisely the distribution of effect sizes here-- as long as my experiment is close enough to the world that there's not a crushing external validity problem. And there is often a crushing external validity problem.
But I view this as not-- this is something that can be worked on. I mean, you design an experiment very differently if you want it to mirror the world closely enough to be able to used in one of these type of applications versus if you just want to get at the gist of a qualitative idea. And we can design experiments that way if we want to. We just haven't tried to very much historically.
Anything else? Please.
AUDIENCE: I've run a bunch of economic tax experiments.
ALEX REES-JONES: I'm aware.
AUDIENCE: I'm aware of [INAUDIBLE] for the IRS. What we finally did was run, with a separate group, a focus group first to try to understand how people are thinking in terms of a particular tax compliance question [INAUDIBLE]. And that's been very helpful at getting some of the process issues that Jay-- or heuristics that people use. So anyway, so you can do that if you're experimenting, as sort of a pilot that takes the form of a focus group.
ALEX REES-JONES: Yeah. And to tie a few of these questions together, I've never-- I mean, I've run some of these many thousand person experiments trying to get-- that are fundamentally aimed at measuring effect sizes in these models. But I've never run one of those without running several internal experiments at a very small level with focus groups beforehand, trying to do what Jay was suggesting and, as you're bringing up, trying to understand what's going on before you take it super literally and run with it at scale.
I think I've probably taken more than my fair share of the time, but thank you all very much.
We've received your request
You will be notified by email when the transcript and captions are available. The process may take up to 5 business days. Please contact firstname.lastname@example.org if you have any questions about this request.
Alex Rees-Jones, associate professor of economics, presents current research findings regarding behavioral economics and human decision-making Sept. 3, 2019 as part of the BEDR Workshop Showcase. Sponsored by the Behavioral Economics and Decision Research Center at Cornell University.