What this covers
Esther Duflo speaks with Tyler Cowen about the foundations of her approach to development economics and the empirical case against several conventional economic assumptions. The conversation centers on why RCTs—randomized controlled trials of specific interventions—reveal poverty and growth mechanisms that grand theories miss, and why financial incentives prove far weaker in shaping behavior than standard models predict. Duflo draws on her work with J-PAL, the research organization she helped build, and on concrete examples from graduation programs in Bangladesh and elsewhere to make her case that poverty is not monolithic, that escape routes require both assets and coaching, and that context matters more than universal recipes.
The discussion ranges across education, macroeconomic policy, labor mobility, and the structure of economics itself. Duflo argues that economists lack a transferable set of growth prescriptions—successful countries have followed wildly different paths—and that the field's introductory teaching obscures this messiness by deriving conclusions too directly from stylized assumptions. She examines why the poverty trap persists for some households even after cash transfers, why microcredit borrowers often use loans as commitment devices rather than productive investments, and why Japan's growth slowdown reflected aging demographics rather than policy failure. On labor and incentives, she challenges the standard view that taxes or guaranteed income substantially reduce work effort, citing historical experiments where labor supply proved inelastic. She also reflects on why development economics has become the most gender-balanced subfield in the discipline, how research organizations function best when they grant ownership rather than impose control, and why small structural changes—like protecting uninterrupted seminar time—can improve research culture.
Duflo argues that poverty and development are best understood through rigorous empirical study of specific mechanisms (via RCTs) rather than grand growth theories, and that conventional economic assumptions—especially about the power of financial incentives—are systematically overstated.
- Graduation programs (cash + coaching) produce durable escapes from poverty because coaching, not just assets, provides skills, confidence, and commitment
- Economists have no general, transferable recipe for growth; successful countries follow wildly different and context-specific paths
- Financial incentives are weaker drivers of behavior than economic models assume, as shown by inelastic responses to taxes and income guarantees
Growth rates are unpredictable policy targets; chasing them via debt-financed spending backfires without restoring prosperity.
- Japan's growth slowdown was a natural and expected consequence of an aging population shrinking the working-age share, temporarily masked by total factor productivity growth; the appropriate response was to manage expectations and improve living standards rather than chase growth via debt-financed Keynesian spending that left them in an undesirable macroeconomic situation.
“when growth fell down in Japan, there was a reason for that, which is the population was aging. The working population, as a share of the overall population, was therefore becoming smaller.”
- Politicians should treat growth as a useful 'chief of staff' rather than something to fall in love with, because growth rates are hard to manipulate (especially fluctuations), so when a country fixates on restoring growth instead of asking why it fell, leaders try increasingly desperate measures and back themselves into bad macroeconomic corners—as Japan did with debt-financed Keynesian spending that built up an unnecessary debt level without actually restoring growth.
“I think we should think of growth rate as chief of staff, not something I think we should fall in love with.”
- During the first five-year plan, the Soviet Union could have taken a market-friendly (Leninist) or utopian collectivist path, but the eventual Stalinist form of 'boring, inefficient materialism' was consolidated partly through propaganda use of big construction sites like the Dnieper dam, which served to consolidate Stalin's grip on power and society.
“this is how those economics and the narrative of economics were used to consolidate power and to consolidate grip on society.”
Empirical RCTs reveal that poverty has structural traps and that coaching combined with cash transfers can break them.
- There is a poverty trap with a wealth threshold: households pushed above it grow to become significantly better off and sustain it, while households below it return to their original level of poverty—demonstrated by long-run (10-year) follow-up studies of cash-transfer recipients in Bangladesh.
“there is such a thing with a threshold of wealth above which — if households can go above — they can grow to a point where they are significantly better off. And if they are below it, they go back to wherever they came from.”
- The coaching component of cash-transfer programs is essential because it prevents poor recipients from prematurely liquidating productive assets, provides complementary human capital (e.g., physically teaching people how to go to market and bargain), and builds confidence in people previously excluded from productive activity—as shown by the Ghana 'goat drop' experiment where assets given without coaching did not serve as a springboard for more activity.
“they find very clearly that the coaching makes a big difference, that people who you give just assets to have more assets, but it doesn’t serve as a springboard for more activities.”
- The graduation/coaching program is already substantially scalable, as evidenced by Bangladesh's enormous program reaching hundreds of thousands of families and by Indian state governments now mainstreaming it into their programs with support from the original microfinance institutions—so the scalability question is partly already answered in the affirmative.
“in Bangladesh, the program is huge as it is. It’s enormous. It reaches hundreds of thousands of families as it is. So to some extent, we already have the answer to how scalable it is because it’s already scaled.”
Field experience, decentralized research infrastructure, and empirical rigor correct economists' ideological assumptions and expand who can conduct serious research.
- Although RCTs risk centralizing authority in a few well-funded, well-connected institutions, J-PAL has deliberately countered this by building infrastructure (200 staff and 1,000 surveyors in India serving dozens of researchers) and raising money to redistribute to roughly 400 often-junior affiliated researchers, so that someone with drive and sweat equity—even outside top institutions—can run an RCT without their own funding.
“People can get in with no funding of their own, in part because one of the things we are doing as a network is raising a lot of money to redistribute to other people widely.”
- Development economics is the most gender-balanced subfield of economics—with arguably more women and more powerful women than men—because the field's infrastructure lets researchers work safely in groups even in dangerous environments, so the dangers to women in some emerging economies do not actually deter them.
“I think development economics is the most gender-balanced subfield of all of economics. In fact, if anything, there are more women and more powerful women in development than men.”
- Effective management of research organizations works by giving people a clear vision of the destination and then maximal ownership over how to get there—accepting that they will make correctable mistakes—because ownership of a piece of the project generates greater effort and creativity than micromanagement.
“you want to give people a pretty clear vision of where we’re heading and then give them a lot of ownership for how we are getting there.”
- Talented students interested in development should spend a year or two working on a project in a developing country, because only through field exposure do they learn how wrong their intuitions and preconceptions are—confronting what seems obvious against something entirely different is the only way the lesson sticks.
“It’s only through this exposure that you can learn how wrong most of your intuitions are and preconceptions are.”
- Introductory economics teaching is misleading because its conclusions follow too directly from the assumptions baked in—e.g., a simple supply-demand diagram with a horizontal line trivially 'shows' minimum wage is bad and redistribution perverts incentives—making it seem ideological and disconnected from real people's lives.
“if you want to show that minimum wage is bad, then it’s very easy to write a supply-demand curve and the horizontal line. Low and behold, minimum wage is bad”
- Rock climbing is analogous to empirical economics: you form a theory about what route might work by looking at the wall, then test it against reality where it sometimes fails (the data), and it is fundamentally social and collaborative—requiring you to trust and attend to a partner—rather than competitive.
“You look at the wall, and you have a decent idea of what might work. Think of it as a theory or an intuition. But then when you try it out, sometimes it works, sometimes it doesn’t. That’s kind of the data going in a way.”
- Financial incentives are overrated: people are pretty inelastic to taxes at all income levels, and receiving guaranteed money does not lead them to stop working or go on vacation—as shown by the 1970s negative income tax experiments where labor supply barely responded even when income was taxed away at roughly a 50 percent marginal rate. This does not mean people ignore all incentives, since sufficiently strong (often non-financial) incentives almost always work.
“the response to financial incentives is less than what typically we assume it is, both in terms of people seem to be pretty inelastic to taxes at all levels of income”
- Small structural rules can improve economics' interruptive culture (which disproportionately bothers women and minorities): MIT's rule giving seminar speakers the first 15 minutes uninterrupted has improved the experience for both speaker and audience—an example of incremental improvement rather than a single magic-bullet reform.
“to institute the rule that someone who gives a seminar gets the first 15 minutes uninterrupted. And that has, for our seminars, the development seminars, I think it’s actually improved the experience for everyone”
Education and human capital matter for development, but their effects are tangled with selection and reverse causation; unpacking actual learning is essential.
- Even if human capital were not a major cause of growth, investing in education remains a low-risk bet because in the worst case people end up educated, which is a good end in itself rather than merely a means to an end.
“worse comes to worst, people would be educated, which seems to be a good end in itself. It’s not just a means to an end.”
- Human capital broadly defined—especially education—plays a big role in growth, but some of the apparent effect is correlation rather than causation (countries good at expanding education tend to be good at other things, and fast-growing countries generate incentives to invest in education), and the policy question only becomes tractable when 'education' is unpacked into actual learning, cognitive versus noncognitive skills, and how those skills are produced.
“Some of it is correlation, rather than cause and effect, that countries that have been good at increasing education have been good at doing other things. Also, countries that have grown fast also generate an incentive to invest in education.”
- The French education system is very good at producing and reproducing a very small elite of highly educated people, but current PISA and TIMSS results show France at the median overall with high heterogeneity—excellent results at the top, very poor at the bottom—and the strongest OECD correlation between socioeconomic class and math scores, indicating the historically excellent, inclusive system has been allowed to crumble.
“the current PISA and TIMSS results show France exactly at the median in terms of the scores, but with a lot of heterogeneity, with excellent results at the top and really bad results at the bottom.”
- Concern about single parenthood is wrongly coded as a right-wing issue; the data show it is bad for both children and the single mother, and the real distinction is framing—social conservatism treats it as a moral failing while progressivism treats it as the changeable outcome of life conditions (such as economic shocks eroding social norms in hard-hit towns).
“what would make it a social-conservatism issue is to approach it from a kind of moral failing way. And what makes it a progressive issue is to approach it as the outcome of conditions in life that are potentially changeable.”
Structural barriers and psychological factors—asset depletion, lost meaning from expertise, weak savings instruments—explain why poor households and displaced workers remain trapped.
- US geographic mobility has fallen (to about half its 1948 level, with decline starting in the 1990s) because economic shocks are geographically clustered—when a region's clustered industry is hit, all local businesses suffer, real estate prices collapse, and people become underwater on their homes—and because moving means abandoning social networks and entering a different, often demeaning sector where one starts from scratch and loses the meaning derived from long-held expertise.
“When a region is hit by a shock, that hits not just one company when everybody else is doing fine, but all the companies because everybody was doing furniture or clothing”
- Only a small minority of borrowers have very high (50-60%) returns on investment, and crucially these high-return individuals are not necessarily the ones with access to capital—a mismatch between investment opportunity and capital access ('misallocation') that explains why average returns and growth are lower than the high individual rates would suggest.
“there is a mismatch between the investment opportunity and who has the money. And that’s what my colleagues call misallocation.”
- Economic models should incorporate the fact that people derive meaning from their jobs—but mainly when they have held a job a long time and grown in expertise and rank—so that the prospect of starting from scratch in a demeaning new sector is a real deterrent to mobility, not captured by standard wage-comparison reasoning.
“they are more likely to be deriving meaning from their job if they’ve held this job for a long time and if they’ve grown in expertise and in ranks in doing this particular job.”
- Most microfinance borrowers do not invest in productive activity; instead they use the loan to acquire a durable good and repay from regular income, effectively using the credit officer's repayment pressure as a commitment device to save—a rational, if expensive, response to an environment lacking savings instruments and full of temptations.
“So the microcredit officer serves as a commitment once you borrow the money. The incentive of the credit officer is very strongly to make sure that everybody reimburses.”
Context-specific institutions and policies drive growth; no universal prescriptions exist because successful countries follow radically different paths.
- Economists have not found a transferable set of prescriptions—even context-dependent ones—that reliably produce growth, because successful countries follow wildly varied, context-specific paths (Ireland needed low taxes, South Korea needed industrial policy, Ethiopia and China carved out trade niches differently), so scientific generalization collapses into the uninformative statement that 'good things are good.'
“the good policymaking is so varied that at the scientific level, it becomes very quickly “Good things are good.””
- France's biggest companies are all old because, unlike the US where bankruptcy is easy to navigate and carries little stigma (producing churn that yields both many failures we never see and some great new companies), the French system discourages failure and is therefore less conducive to new firms emerging.
“in the US, failing is quite all right; bankruptcy is not really a problem... and there is not much stigma associated with failing, and so there is more churn in the system.”
- In France parents follow a sensible shared 'railroad' of child-rearing without agonizing, whereas in the US child-rearing has become an occasion to make a statement about personal identity—so much so that a parenting choice can almost predict one's political views—leading American parents to feel they must invest enormously and to feel inadequate when they cannot.
“In the US, child-rearing is one more occasion to make a statement about your identity... And somehow it almost can predict what you’re going to think about Donald Trump.”