Duflo and Banerjee's Deficient Thinking on Incentives, Part I
Within 2 weeks of sharing the 2019 Nobel Prize in economics with Michael Kremer, MIT economists Esther Duflo and Abhijit Banerjee wrote a long piece for the Sunday New York Times in which they argued that financial incentives are not as important as many economists think. The op/ed is titled “Economic Incentives Don’t Always Do What We Want Them To,” New York Times, October 26, 2019. What I find striking is how badly thought out and how badly backed up their argument is. To be sure, they do back it up somewhat. But their argument overall is weak. Their conclusion about the weakness of incentives is ironic given that they got the prize for their experimental work using incentives in poor countries, work in which they found that incentives seemed to be pretty important. The incentives were not necessarily financial incentives, but they were closely related. More on that anon.
Almost every paragraph of their long piece is interesting and important. That’s why this is a 2-part series.
Start with the last sentence in the first paragraph. They write:
Over the last few decades, this faith [by economists] in the power of economic incentives led policymakers in the United States and elsewhere to focus, often with the best of intentions, on a narrow range of “incentive-compatible” policies.
Notice their word “faith.” They could have chosen the word “confidence.” But by using “faith,” they subtly undercut from the getgo the idea that there is much evidence for the importance of incentives and, instead, claim that the view is faith-based.
Their second paragraph gives their main message:
This is unfortunate, because economists have somehow managed to hide in plain sight an enormously consequential finding from their research: Financial incentives are nowhere near as powerful as they are usually assumed to be.
At least now, they are basing their case on consequential findings from research. Do they back it up? Somewhat yes but mostly no.
Notice their first example:
We see it among the rich. No one seriously believes that salary caps lead top athletes to work less hard in the United States than they do in Europe, where there is no cap.
The typical salary cap is on overall team payroll, not on the amount an individual athlete can be paid, which is what is relevant here. Moreover, even if the salary caps are sometimes on individual players what matters for athlete effort is the size of the caps. Imagine that the NBA caps an individual player’s annual salary at $40 million and a European soccer team has no cap but pays a maximum salary of $30 million. What matters is pay, not whether there’s a cap. So their first example is not evidence at all.
Research shows that when top tax rates go up, tax evasion increases (and people try to move), but the rich don’t work less. The famous Reagan tax cuts did raise taxable income briefly, but only because people changed what they reported to tax authorities; once this was over, the effect disappeared.
This paragraph contains a number of important claims. Whereas for some of their claims, they give links to back up their statements, they give no links in this paragraph. Notice also their subtle slur about people who try to avoid taxes: they mention evasion, which is illegal, rather than avoidance, the non-evasion part of which is legal. Moreover, I’m pretty sure they’re wrong. If they had said that the main effect of higher top marginal tax rates is not that people work less, they might have been on firmer ground. But they didn’t say that. Instead, they said categorically that the rich (by whom they main, in this context, higher-income people) “don’t work less.” The main evidence I know on the Reagan tax cuts was by former Harvard economics professor Lawrence B. Lindsey in some academic papers, a book, and a popular study for the Manhattan Institute. Here’s what I wrote in “Are We All Supply-Siders Now?” Contemporary Policy Issues, Vol. VII, No. 4, October 1989:
Lindsey (1988) shows that 1985 income for taxpayers with an AGI [adjusted gross income] of more than $200,000 was $86.8 billion higher than their baseline income. What were the sources of this added income? Lindsey breaks down the income into four categories: dividends and interests, wages, capital gains, and other income. Interestingly, dividends and interest virtually were identical to the baseline, wages were about 30 percent higher, capital gains were more than 100 percent higher, and other income was nearly 200 percent higher. Because “other income” is made up primarily of business and proprietary income, Lindsey concludes that the main supply-side effect was due not to people working harder but rather to people paying themselves more in cash and less in fringe benefits. In other words, a key to the Laffer Curve is not, as many previous critics and some proponents of the curve had assumed [this included me in 1980], a large elasticity of supply but rather a large elasticity of tax avoidance with respect to tax rates. [italics in original]
Duflo and Banerjee are wrong on two counts. First, a 30 percent increase in wages is strong evidence that they did work harder. Second, although they are right to emphasize the increase in the income that people reported, they are wrong to claim that the effect disappeared.
Duflo and Banerjee continue:
We see it among the poor. Notwithstanding talk about “welfare queens,” 40 years of evidence shows that the poor do not stop working when welfare becomes more generous.
Remember that they were trying to argue that financial incentives are weak. How do they do it? By arguing that welfare doesn’t cause poor people to stop working. The argument, at least the one made by economists–and remember that it’s economists that they’re taking on–is not that increasing welfare payments would cause people to stop working but that increasing welfare payments would cause people to work less.
In their next sentence, Duflo and Banerjee write:
In the famous negative income tax experiments of the 1970s, participants were guaranteed a minimum income that was taxed away as they earned more, effectively taxing extra earnings at rates ranging from 30 percent to 70 percent, and yet men’s labor hours went down by less than 10 percent.
In other words, high marginal tax rates did cause men to work less. To their credit, they could have made a stronger claim. The study they link to shows that husbands reduced their labor hours by only 5 percent. But wives reduced their labor hours by a whopping 21 percent.
Note their next sentence:
More recently, when members of the Cherokee tribe started getting dividends from the casino on their land, which made them 50 percent richer on average, there was no evidence that they worked less.
Notice what Duflo and Banerjee did. They switched from talking about the effects of higher marginal tax rates (explicit or implicit) to talking about the effects of people’s non-wage income increasing without their marginal tax rates increasing. Probably most economists reading this would notice the switch–at least I hope they would. But a large percent of non-economists would probably not notice. Why does it matter? Because basic economic theory says that when one’s dividend income increases, there is an “income effect” on labor hours worked, but the income effect could be quite small or even zero. Because marginal tax rates did not change, the study they cite, even if high-quality, is not a test of the effects of incentives. Recall that what the two MIT economists are trying to establish is that financial incentives aren’t particularly powerful.
In their next paragraph, they write:
And it is true of everyone else as well — tax incentives do very little. For example, in famously “money-minded” Switzerland, when people got a two-year tax holiday because the tax code changed, there was absolutely no change in the labor supply. In the United States, economists have studied many temporary changes in the tax rate or in retirement incentives, and for the most part the impact of labor hours was minimal. Nor do people slack off if they are guaranteed an income: The Alaska Permanent Fund, which, since 1982, has handed out a yearly dividend of about $5,000 per household, has had no adverse impact on employment.
I had been unaware of the Swiss experiment. Assuming the study was well done, score one for Duflo and Banerjee.
But the Alaska story is like the Cherokee story. We would not expect an incentive effect of a yearly dividend. So it’s not evidence at all for their claim about financial incentives.
Next time: Part II.