Taxpayers hate to lose: how psychology can help increase collection

Most Americans pay taxes. But “despite the social benefits of taxes, the individual costs of paying them are famously aversive,” wrote Alex Rees-Jones, a Wharton professor of operations, information and decisions, in a briefing paper for the Penn Wharton Public Policy Initiative titled “Addressing Personal-Income-Tax Manipulation with Tools from Psychology.” He advocated that the government use psychological tools to maximize tax collection by tapping into what he called a taxpayer’s “gain/loss framing” in legally taking tax deductions or credits or illegally evading taxes.

When people file their taxes, they often have to make decisions about minimizing losses and enlarging gains. Research shows that people are more likely to seek to reduce their losses than make a gain bigger, even though it might seem like two sides of the same coin. If one were to reduce the ‘loss aversion’ activity of tax filers, Rees-Jones argued, annual tax revenue would increase by $1.4 billion. He recently joined the Knowledge@Wharton radio show on SiriusXM to explain his research. (Listen to the podcast above.)

An edited transcript of the conversation follows.

Knowledge@Wharton: Please summarize your findings.

Alex Rees-Jones: When I started this project, the key thing I wanted to do was test how loss aversion played out in tax contexts. The key finding is that it does seem to happen. But let me give you a little background. This term loss aversion refers to this very widely documented psychological bias that people seem to treat outcomes differently if they’re presented as a gain versus if they’re presented as a loss, or otherwise equal.

So this is a piece of something called prospect theory, which is a really prevalent theory in behavioral economics. It played a major part in Daniel Kahneman winning the Nobel Prize. It’s basically a psychologically motivated theory of how people make decisions.

Knowledge@Wharton: So people hate to lose.

Rees-Jones: They hate to lose. And it’s a little more nuanced than that even. It’s that if you think about the value of a marginal dollar, you care about a marginal dollar more when it makes a loss smaller than when it makes a gain bigger. But they also hate to lose.

At some point during graduate school, I got really interested in trying to think through how that behavior would play out in tax contexts and I thought there was a pretty strong intuition that it would and that it could play out in a couple different stages. But the one that’s most obvious to me is thinking about how people work through their annual tax bill. The bottom line of it is ultimately figuring out if one of two things is going to happen. The thing that happens to most people is you find out that your total tax bill for the year is less than what you actually already paid through employer withholdings. So in a very literal sense you’re going to get a gain. You’re going to fill out your tax return, send it in and then they’re going to send you back a check.

Some people face a loss in the sense that they find out that they actually didn’t pay enough already and they’re going to have to send a check to the IRS. So we have this very natural gain/loss framing, we’re absolutely in the domain of prospect theory. And then the decision I’ll be looking at is thinking about how people try to change the size of that gain or loss through all the things you can do as you fill out your tax forms.

Knowledge@Wharton: And how much energy they put into trying to change it.

Rees-Jones: Exactly. So one dimension is just how much effort you put in, like looking for different credits or deductions you could claim. But there’s also literal tax-incentivized behaviors you can pursue like putting money in a tax preferred retirement savings account. And then of course some people just do the obvious thing of cheating on their taxes and deciding how much to do of that. So let’s just call all these things bundled together manipulation, whether it’s legal or illegal. And the thing I’m trying to think through is if you see the evidence of prospect theory in these manipulation decisions, do you see people do more of this manipulation when they face a loss on tax day relative to a gain?

The bottom line of the paper is if you look at the IRS’s tax records, and I’m looking at a panel going across the 1980s … you see real telltale signs that people are manipulating their bill in that way.

Knowledge@Wharton: What are some of the practical implications of this?

Rees-Jones: There are all these technical considerations in terms of how you’d want to model tax policy. … The most practical thing is thinking about nudge interventions, where lots of us in some capacity or another interact with people and care about whether they pursue or don’t pursue some tax-incentivized behavior.

Employers might care whether their employees are putting money in a retirement savings account as part of their compensation package. We often care about trying to get people to donate to charity, things like that. These are all activities that have some tax incentive behind them. The basic idea of this is that we’re seeing some responsivity of these behaviors to gain/loss framing. If you can control whether something is perceived as a gain or a loss, if I want you to take a certain tax incentive, I want to make sure you feel like it’s making a loss smaller as opposed to making a gain bigger.

And of course there are some activities that are tax-incentivized where we don’t want you to respond to the incentives, like you face some incentives to try to cheat. And in cases where I’m looking at a group of people who I think particularly strongly face those incentives, you might want to make sure they’re thinking about their tax bill framed as a gain to help minimize the degree to which they psychologically feel motivated to pursue that opportunity.

Knowledge@Wharton: So if you want them to take advantage of something like the charitable deduction, then you would present it one way. And if you want to avoid some tax evasion you present it another way. Some examples?

Rees-Jones: You could think about, for example, small business owners. This is a group of people who have the best chance of getting away with tax evasion relative to, say, someone like you or me who directly get a bill from the University of Pennsylvania, and the University of Pennsylvania tells the IRS they paid us. So if we lie, we’ll get caught immediately.

Knowledge@Wharton: Your paper notes that there’s evidence to show that small businesses tend to be involved in illegal tax evasion more than the average.

Rees-Jones: Yes. So trying to control incentives for those types of groups to go after, this is a key thing you might do. And then whenever you’re interacting with them, when you’re sending them materials from the IRS or interacting with them not as the IRS but some third-party organization, to the degree to which you can frame the taxes they pay not as a loss relative to, say, not paying any taxes, but as a gain relative to what some other company like them might have paid — or what they might have expected to pay, etc. — the key prediction of this theory is that you should then expect them to cheat a little bit less.

Knowledge@Wharton: Let’s say you have a retail store. You’re saying that if you gave the owner information that said, “Of all the retailers in your zip code or your state, X percent tried to evade taxes.” How would that actually look in practice?

Rees-Jones: I want to be clear. I don’t have that exact data, although that feels a lot like a lot of work of, say, Robert Cialdini on social proof, for example. … So yes, exactly the combination of my results with this existing literature would suggest that exactly the type of thing that you’re suggesting might be effective.

For example, if you’re trying to decide which people to go after to try to get them to give an extra dollar to charity, looking at people who are facing a tax loss might be a group of people who are particularly susceptible to marketing materials around tax season.

Knowledge@Wharton: So, if around the time when people were filing taxes a message came to the right taxpayer that said, “you can lower your tax bill if you give to charity” … that could nudge people to give.

Rees-Jones: Exactly. But there’s a little bit of a disconnect in timing for a number of tax provisions, charitable giving being one, where the moment where you think a person might be most motivated to find these manipulation opportunities is, let’s say, on April 14th as they frantically try to beat the deadline. But for many rules what you’re documenting on April 14th is what you did in the previous calendar year. And if I did something today I wouldn’t claim it as a tax reduction today, I would claim it next April.

Ideally, if you want to them pursue that activity, you’d want to make it so that they can act on that and have it immediately influence their tax bill. … And some motives like that influence why we allow people to put money in types of tax preferred retirement savings up till April.

Knowledge@Wharton: What would the wording actually say – if you’ve targeted someone who’s more likely to cheat on their taxes?

Rees-Jones: The settings where I have wording actually worked out are more for the tax incentives than the tax disincentives. For small businesses the bottom line is just, I think, as much as possible they don’t view their reference point as paying zero taxes at all and view any tax as a loss.

In the cases where we as third parties, not as the IRS, have the most chance to act on these nudge opportunities are cases where we’re trying to run a charitable-giving drive or get our employees to put money in a retirement savings account. The employer can only send this to the, say, core of employees who actually face a loss — “You know, around this time of year you’re about to send $1,000 to the IRS. But note that if you immediately give this money to charity based on your margin tax rate you would actually send this lesser amount” and present the numbers for them and let them make the decision themselves. Do I really want to give a certain amount of money to the IRS versus do I want to give it to charity? Do I want to put it in my retirement savings account?

Knowledge@Wharton: So again with small businesses. Maybe people think: “Most of my peers are probably not paying very much in tax and I don’t want to be the one paying more,” or “I just don’t want to pay more period.” But if they got a note saying, “People in your peer group pay between” and you give some dollar range, then you’re saying that their prime motivation wouldn’t be to pay zero tax, their prime motivation would be, maybe, to be something closer to their peers, is that it?

Rees-Jones: Potentially, yes. There’s a very interesting paper by Ricardo Perez-Truglia and co-authors that was looking at nudge-type interventions to collect late taxes from taxpayers — people who it’s already been determined owe the IRS money and they haven’t paid yet. And they sent out peer information and they found in some contexts that they have these backlash effects where they said, “Here’s information on all these other people who are not paying their taxes.” Normally those types of peer information provision interventions are meant to persuade you to do it. But it seems like a lot of people got those letters and said, “All sorts of people aren’t paying their taxes, now I definitely shouldn’t.” They had negative effects that were not intended — in some cases — that’s not always the case. But that’s the flavor of the result we’re talking about.

Knowledge@Wharton: Another issue is you note that more than three quarters of taxpayers actually have too much money withheld, either because they see it as a forced savings or they don’t want to get a surprise bill at the end. But you’re saying that’s not a great way to handle things.

Rees-Jones: It’s been observed long before me that it’s not great for taxpayers that so many are over-withheld. That it’s costly for them. Here’s the way to think about it. If I’m doing my withholding perfectly, let’s say if I did nothing differently I would have a balance due of zero when I fill out my tax return next year. But today I withhold an extra $50. Then in April when I fill out my tax return I’ll get $50 back, so what I did was I gave the IRS a $50 loan with a zero percent interest rate. And that’s a bad deal for me because normally when I give people $50 I get interest on it. And so I’ve let the IRS collect interest instead of me.

It’s been appreciated for a while that the way to think about the cost of this prevalent over-withholding is the fact that a lot of taxpayers are losing out on these interest costs. And there’s been some very nice work people such as [University of Chicago professor] Damon Jones calculating how much that is. I join this debate by saying that there’s actually another thing that over-withholding is doing, it’s influencing psychologically people’s incentives to manipulate their tax bill in some ways that are probably good and in some ways that are probably bad. So it’s probably good that gain framing is making it so people like to cheat less or seek out opportunities to cheat less. It’s not clear if we actually want to dissuade people from pursuing tax-behaviors. And it’s probably doing that too.

So a bottom line that I get to in the paper is that if you actually calculate up the dollar amount of how much less manipulation you see due to having so many people in the gain domain, it’s of comparable quantitative importance to the dollar amount of foregone interest that we were already talking about.

Knowledge@Wharton: So could you restate that? That’s a little bit complicated.

Rees-Jones: When we’re thinking about the social cost of having so much over-withholding, the thing we’re normally thinking about was these foregone interest payments that people were giving to the IRS. And I’m saying if you calculate up the amount of money that’s moving around just from less manipulation occurring, that amount of money is comparable or is a magnitude to these interest payments that we were previously focused on.

Knowledge@Wharton: From the government point of view you wouldn’t necessarily want to encourage people to change that behavior because it’s free money for them.

Rees-Jones: Yes. So we already knew very well that it’s great for the government if people give them too much money. And now we have another reason why that’s even more true than before.

Knowledge@Wharton: Did anything surprise you in this research?

Rees-Jones: Most surprising, even to me, although I was hoping to find this, is the size of the effect that I ended up documenting. The key thing I was trying to do technically in the paper was find a way to estimate the amount of extra manipulation you’d face in the loss domain compared to the gain domain. The estimate I get to is that on average people seem to reduce their tax bill by $34 more if they’re in the loss domain compared to the gain domain. So $34 by itself is not necessarily a big deal.

Knowledge@Wharton: And when you say a loss gain domain versus a gain domain, you mean what exactly?

Rees-Jones: I’m imagining if you had to pay the IRS $100 on tax day versus if you got a $100 refund. And the implication of the estimates is you would reduce your tax bill overall for the year by $34 more under the loss situation than under the gain situation. And so $34 is not a big deal. When you start multiplying $34 by the number of taxpayers we have it becomes a big deal pretty quickly.

A way to get at quantifying that, one thing you can do is ask how much would it change total tax collection if everyone manipulated as much as people in the loss domain? The answer is that it would reduce total tax collection by close to $4 billion, which is plenty of money, particularly when you’re thinking about this literature on psychological effects on prospect theory — often when we’re studying psychological effects we’re studying numbers much smaller than that. You rarely see direct demonstrations of this psychology playing out in that kind of scale.

Knowledge@Wharton: How exactly does your research differ from what else has been done in this area?

Rees-Jones: There’s a long line of research thinking through the implications of prospect theory for taxation. I would characterize it in a couple different bins. There’s a bunch of work that’s just pure works of theory. My contribution is primarily empirical. I’m trying to think through how to actually measure this in field data.

On the same point, among existing empirical estimates where we’re trying to actually see it play out, most of that evidence is from small-scale experiments or surveys where I take people and put them into some economic game or survey question that would reveal if they think about things in a loss averse way. But I’m not actually looking at their tax data, necessarily. So that leaves open the question if it actually translates through to what they really do on tax day.

There are a couple of studies that do directly inform this type of question, where we look at specific types of credits or deductions and they show some evidence consistent with people pursuing them more when they face a loss than a gain. Now this is where I jump in. The key thing is if your goal was to think about the total aggregate consequences of prospect theory it doesn’t work to just look at one item at a time because there are just so many things you can do to manipulate your tax bill. There are so many credits and so many deductions, and so many ways you can cheat. So you’ll really understate the importance of loss aversion of prospect theory if I look at one thing at a time.

The approach that I build in the paper is one where anything people are doing to manipulate will get bundled together and calculated to get to that $34 number. So the key way I differ is in being able to provide this estimate of the total quantitative impact of the psychology, which these papers have not done in the past.

Knowledge@Wharton: If you were talking to someone from Congress, what would you say? “Here are a couple things that might make better public policy?

Rees-Jones: The largest implications of this research and most of the research I do on psychology and taxation is not necessarily on how to do nudges and things like that, and how to present things to the public. But instead, it’s about how to actually do things when you’re modeling the consequences of tax reforms — policy changes — rather than operating fully with the typical economic model where we assume that people are fully informed about the tax system, perfectly rational, etc., and using that to forecast their response.

If we can build in some structure about how we know people come to understand the tax system, what biases we know play out, and build that into our forecasting apparatus, I think that helps us figure out and better forecast what’s going to happen after we change taxes, which types of tax changes we should expect to be beneficial versus not, and so on.

Knowledge@Wharton: Is that to say, build in some things that take into account more ideas from behavioral economics?

Rees-Jones: Exactly, yes. There’s a lot of people who care a lot and do very serious work trying to think about how people understand their taxes, and to forecast what they’re going to do in any given year. There are branches of government that focus on this a lot. And the time is right for building more ideas for behavioral economics into this exercise, and use what we know from psychology to really put a bit of of structure on this problem to improve our forecasting accuracy.