Six questions with Seema Jayachandran

Seema Jayachandran is the Kenneth J. Boudreaux '65 Professor of Economics and Public Affairs at Princeton University. Her research focuses on environmental conservation, gender equality, and other microeconomic topics in developing countries. 1. In your new JEL review with Alessandra Voena, you focus on power between spouses, and flag multigenerational and polygynous households as a promising direction for future research. Have economists made the household too small? In many settings, a woman is bargaining not just with her husband but with parents, parents-in-law, siblings, children, and wider kin. How would taking those actors seriously change either our models of household power or the interventions we design to increase women’s agency? Let me start by defending simple models. A two-person model offers insights that also apply to more complex families, and it is often the better tool for revealing and explicating them. But, yes, in many places other family members have a say in what a couple decides. In north India, a woman often needs permission from her mother-in-law to leave the house, as your work has highlighted. In parts of Africa, a man may have several wives who compete for power. I’ll speculate on three ways this complexity might matter, focusing on in-laws. The first is obvious: interventions that aim to change husbands’ minds, through information or persuasion, could target other family members too. In a project of mine in Uganda that gave health information to fathers, men’s mistaken beliefs often came from their mothers. They’d urge their wives to feed the newborn sugar water, for example, because that’s what their mother said was right. Changing the father’s mind might suffice, but it’s worth testing whether targeting mothers-in-law works better. I’ve seen quite a bit of recent work in that vein. Second, I’d expect larger family units to be further from Pareto efficiency. Two spouses can plausibly strike the deals that make both better off. With three or four or five people involved, each with their own goals, that seems much harder. Third, a woman’s in-laws and husband might differ in how much they want the marriage to stay intact. Interventions improve a woman’s well-being within her marriage partly by strengthening her position were she to leave – her outside option. If her husband wants to stay married, he has to cede some power to her. When in-laws have a say, it also matters how much they want to keep her in the family. They might value the marriage more than their son does if they are elderly and rely on her care, or if a divorce would bring shame on them. They might value it less if a new marriage means another dowry or if they just don’t like their daughter-in-law. Depending on which case holds, in-laws’ involvement could amplify or dampen how much these interventions help married women. 2. Your recent work also makes me wonder whether economists sometimes confuse our measures with the concepts we care about. Your new Malawi paper finds daughter preference using mothers’ life satisfaction in a setting where conventional fertility-based measures can miss it; your work on women’s agency has similarly questioned standard survey measures. You've argued, for example, that DHS questions should ask about influencing rather than making decisions. What do researchers get wrong in measuring son preference, and what other concepts are we still particularly bad at measuring? Lack of clarity about “son preference” is one of my hobbyhorses. “Son preference” is used to mean both a desire to have sons and the tendency to provide more to sons than to daughters, and those don’t always go hand in hand. I tried to make that point in my 2015 Annual Review article. And as a measure of the desire to have sons, researchers often use the share of sons in a person’s ideal family composition, from the Demographic and Health Survey questions. But when people mostly care about having at least one son, which is the case in patrilocal and patrilineal societies like India, someone will want 100% sons if they want 1 child and 50% sons if they want 2 children. Their son preference hasn’t changed; their ideal number of children has. I emphasize that point in my AEJ Applied paper on fertility decline and missing women. The new Malawi paper makes a different point that several of our measures of son preference, such as stopping rules, have more signal at lower fertility, so they might not work as well in some parts of sub-Saharan Africa. Regarding other concepts we’re bad at measuring, I’m not convinced survey questions on aspirations are useful. A typical question asks, “Ideally how far would you like to go in school?” Asking about a world without constraints is fine for learning whether high school or college is a “bad” or a “good” in someone’s preferences. But in many contexts, we’re past girls’ education being a bad, and the question is how much people value it. In my work evaluating a gender equality curriculum in India, we found no effect on girls’ stated educational aspirations but some movement on a revealed preference measure that gets at the intensity of their desire to go to college. I generally think economists overstate how much better our small-stakes revealed preference measures are than stated preferences, but in this case, our revealed preference measure was more informative, not because it had real stakes but because it captured the intensity of the preference -- how much effort girls were willing to put in to go to college. 3. Is there something about gender or development economics that you believed pretty strongly, say, ten years ago and no longer believe? What changed your mind? I used to think job creation didn’t need to be a policy objective separate from economic growth and poverty reduction. But seeing high youth unemployment and underemployment in several places where I work, and the unhappiness and unrest that causes, I’ve changed my mind. (But I haven’t become as single-minded about jobs as your employer has!) 4. You’ve combined machine learning with qualitative interviews to develop better measures of women’s agency. Generative AI potentially takes that much further; for example, by making it possible to collect or analyze much richer qualitative data at scale. Where do you think AI could genuinely improve how development economists measure difficult concepts such as agency or norms? And what are you most skeptical of? In our research, we are often interested in nebulous or loaded constructs that a couple of close-ended questions can’t measure well. For example, in one study I’m working on, we want to know whether people regret a decision, but we are worried they will be reluctant to say so explicitly; not admitting regret might be psychologically protective. Yet you can get a pretty good sense of someone’s regret if they talk more expansively about the decision and its consequences. Without AI, conducting semi-structured interviews and coding up the responses for statistical analysis uses human experts at both stages. The research I’ve seen so far hasn’t found AI to be very good at coding open-text interview responses. A recent World Bank working paper replicates our machine learning approach for measuring women’s agency in Tunisia. The authors also compare human and AI coding, and AI performs poorly. But I’m optimistic that AI will become very good at coding. I don’t know if that will happen within a year, but I bet it will happen within 3 years. And even if AI is not better than human coders, it will be cheaper and therefore more feasible at scale. I’m more skeptical that chatbots can replace humans at the interview stage, at least on sensitive topics. People sometimes shut down when asked a particular question. That can be because they don’t understand the question and are embarrassed about that. Or maybe the topic is fraught, and they don’t want to talk about it. How to proceed depends on why they clammed up, and a perceptive interviewer can usually tell from body language or verbal cues. I’m less sure that AI will develop that emotional intelligence. 5. You’ve worked on a remarkably wide range of questions, from gender inequality to environmental conservation. What advice would you give young development economists about choosing questions, especially how to tell the difference between a question that is merely publishable and one that is genuinely worth spending several years of your life on? Looking back, what helped you learn which questions were worth pursuing and which ones to walk away from? Is there a project you deliberately walked away from? When I grow up, I hope to know which questions are worth pursuing! I think it’s a fool’s errand to identify *the* most important problem; there are plenty of them. That’s freeing, because it means you can work on the questions you’re personally excited about. “Worth spending several years on” is mostly person-specific. They’re the questions that make your mind buzz with ideas, that you want to talk about with your colleagues and friends. Beyond that, the questions I’m happiest I pursued addressed what I thought was a blind spot in how researchers or policymakers were thinking about a topic. To use a baseball analogy, it’s about wins above replacement. If I didn’t write this paper, probably no one else would in the next couple of years, and that would be a shame. Looking back, the projects that met that ex ante criterion are the ones that have had the most impact. That means I drop some paper ideas because I think someone else would write an equally good or better version if I walked away. But most of the projects I’ve walked away from were randomized experiments where the implementing partner’s goals and the research team’s goals were too far apart. I recently abandoned a project where the partner had developed a very cool but expensive intervention. Making it leaner wasn’t a priority for them, and I didn’t want to evaluate an intervention that I didn’t think would ever scale. In most of these cases, the signs that I should cut bait were there early, so I should probably walk away sooner, but hope springs eternal! 6. You’ve written for both academic and public audiences. Has writing for a general audience changed the way you think about your own research? Has there been a finding from your own work that became clearer to you through trying to explain it publicly? What do economists tend to get wrong when trying to communicate evidence to policymakers or the public? The project I’ve written and talked about the most is on paying people not to deforest, and doing so has been very illuminating. For example, I’ve learned that many people have a strong moral intuition that you shouldn’t reward people for not harming others. You can punish them for doing the harm, but you shouldn’t reward them for avoiding it. For most economists, taxing a behavior and rewarding avoidance of it seem similar; it’s the same relative price change. But most people see the world very differently. Understanding “normal people’s” vantage point has helped me in talking to policymakers and NGOs about follow-on projects. Sharing the research with a broader audience has also sparked ideas for follow-on work. Almost universally, people view payments for conservation as a win-win: giving someone $100 for keeping their forest intact makes them $100 richer and protects the forest. The fact that the payment is offsetting opportunity costs is not intuitive. That prompted me to write a short paper illustrating that idea with data. I show that, in our Uganda experiment, payments had no effect on the economic well-being for those who had to stop deforesting to qualify. Poverty only fell among those who were going to keep their forest intact anyway, where the payments produced no environmental gain. It’s terrific how much more development economists are writing and speaking to broader audiences today than 20 years ago, through blogs like this one, VoxDev, podcasts. I love it, and I think many people are very good at it! I’ll share one thought. I’ve found that what sticks with broader audiences is mostly the narrative of why the problem exists and why the solution seemed promising. The summary of the findings can be short. People will remember one number at most, and more likely only something coarser like “it backfired” or “it sort of worked” or “it worked really well.” The evidence on mechanisms and other analyses that make the paper “more than just an impact evaluation” and so get emphasized in economics seminars won’t matter to most people.