Raj Chetty on Teachers, Social Mobility, and How to Find Answers to Big Questions

Raj Chetty — economist at Harvard and director of Opportunity Insights — joins Tyler Cowen in Ep. 23 to trace the arc from his Tamil family’s history through Milwaukee’s stark inequalities to the research programme that made him, in Cowen’s assessment, the most influential economist in the world today.

Key ideas

1. Kindergarten teachers shape adult earnings through non-cognitive skills. A high value-added kindergarten teacher raises students’ test scores in the short run, but the gains disappear in middle school — and then re-emerge as higher earnings in adulthood. Chetty’s best explanation: the teacher is not transmitting academic content that carries forward, but discipline, social skills, and the ability to get along with others — attributes the labour market prices directly but standardised tests do not.
2. Upward mobility varies enormously by place, and the driver is socioeconomic integration. Rates of upward mobility (the chance of a child born into the bottom quintile reaching the middle or top) are persistent across generations and diverge sharply across cities. Atlanta grows fast but has low mobility; Iowa and Utah consistently produce good outcomes for low-income families. The strongest correlate is low segregation: in small towns, children from different income groups attend the same schools and participate in the same activities, producing cross-class contact that big-city self-segregation forecloses.
3. Relative mobility has been flat for forty years; absolute mobility has collapsed. The share of children earning more than their parents — the arithmetic of the American dream — has fallen from roughly 90% for children born in 1940 to around 50% for those born in the 1980s. This collapse is driven by rising income inequality, not by reduced economic growth: the same growth distributed as in 1970 would restore most of the gap.
4. Big datasets demand conceptual tricks, not just computation. Chetty’s research advantage is not access to data alone — it is finding the statistical move that lets an imperfect dataset answer a question it was not designed to answer. For the fading American dream paper, no US dataset links parents’ and children’s incomes back to 1940; the key insight was a method for estimating intergenerational comparisons without that link.
5. Social capital and cross-class contact are likely causal, not merely correlated. Chetty’s planned research using Facebook network data aims to move from correlation to causation — exploiting the fact that children befriend classmates in the same grade but not adjacent grades, creating quasi-random variation in social exposure that can be used to test whether the people you grow up with actually determine your life trajectory.

Content

Personal origins and the research programme

Chetty grew up in a Tamil family that had made its own cross-generational mobility: his grandfather, a Gandhi-era freedom fighter, backed women’s education at a time when it was rare; his parents — a physician and an economist — were among the first in their respective families to receive higher education. Moving to the United States as a child, he passed through Milwaukee’s University School — a private school on a 150-acre campus — while riding buses through some of the most segregated neighbourhoods in the country. That early sight of how thinly the accident of family position determines opportunity became, in retrospect, the through-line of his career.

Social mobility refers to the ease with which people born into lower-income families move up (or, symmetrically, those born into higher-income families move down) across their lifetimes — a kind of economic elevator whose speed and accessibility the research tries to measure. The research draws on administrative data — tax records, school registers, social security files — which cover entire populations rather than samples, allowing patterns too subtle or localised for conventional surveys to emerge. The tools are quasi-experimental methods: situations where the real world produces variation in who is treated (by a teacher, a neighbourhood, a policy) that is close enough to random to support causal inference, rather than mere association.

The kindergarten teacher results

The starting point is a long-run follow-up of Project STAR, a Tennessee randomised experiment in the 1980s that assigned roughly 12,000 children to kindergarten classrooms of varying size and teacher experience. Chetty and co-authors tracked those children into adulthood and found that assignment to a high value-added teacher — one whose students score better on tests than would be predicted — translated into meaningfully higher adult earnings, lower teenage pregnancy rates, and higher rates of home ownership, even though the intermediate test-score gains had entirely faded by third grade.

Value-added, in this context, means a statistical estimate of a teacher’s contribution to student learning beyond what prior test scores and student background would predict — roughly, how much better a given class does compared with a similar class under a different teacher.

The paradox — effect disappears, then reappears — points away from academic skill transmission. Chetty’s best guess is that effective kindergarten teachers instil non-cognitive skills: discipline, the capacity to sit still, social competence. Subsequent surveys of the STAR children confirm that those in better kindergarten classrooms did better on non-cognitive measures even in later grades, long after their academic results converged.

A second study using 2.5 million observations from New York City refined the experience finding: novice teachers underperform, teachers with two or three years of experience do better, and after that the profile flattens. The implication for pay policy is that the current seniority-based salary schedule — where pay rises continuously with years of service regardless of performance — is poorly designed. Chetty favours tying bonuses or retention incentives to value-added or principal evaluation, neither of which currently plays any role in most US school systems.

On the structural question of whether teaching quality fell as high-skilled women gained better labour market options, Chetty finds the anecdotal account plausible and consistent with the data: earlier cohorts of teachers, recruited from a pool where talented women had fewer alternatives, appear to have been stronger.

Geography, segregation, and what Iowa does

The geographic research identifies four predictors of high upward mobility that are robust across places: better primary schools, lower income segregation, higher social capital, and stronger family stability. Of these, Chetty views integration — actual cross-income contact — as the most likely causal mechanism, and the one where the coming Facebook data should allow the cleanest test.

The intuition is tractable: in a large city, an affluent household can self-segregate entirely — private school, gated community, separate recreational infrastructure. In a small Iowa town, a single school district, a single set of youth activities, and limited residential differentiation force children of different backgrounds into daily contact. Cowen presses whether this is just restating that Iowa is a nicer place; Chetty concedes that what causes Iowa’s integration remains an open question, but argues the mechanism — exposure and contact — is distinct from the outcome (good life chances) in a way that gives it traction as a target for policy.

The absolute-vs-relative distinction matters here. Relative mobility — the chance of moving up the percentile ranking — has held roughly constant for forty years. Absolute mobility — the chance of earning more than one’s parents in real terms — has fallen sharply, from around 90% for the 1940 birth cohort to around 50% for the 1980s cohort. The divergence tracks rising inequality: a growing economy that concentrates gains at the top still grows, but delivers the experience of stagnation to everyone else.

Designing research around clarity and causal credibility

Asked to articulate what distinguishes his group’s output from work by others with access to comparable data, Chetty points to two things: the conceptual move that allows the data to answer the intended question, and an obsessive commitment to clarity. Papers are iterated until the chain of calculation appears, in retrospect, obvious — an aesthetic of simplicity that is, as he describes it, anything but simple to achieve. The production function, as Cowen frames it, is multiplicative: question selection, data access, conceptual advance, execution, communication, and team management all have to be done well; weakness in any one undermines the product.

The tax salience research — one of his most cited early papers, showing that consumers respond less to prices when the sales tax is displayed separately rather than embedded — grew out of a similar observational habit: noticing that friends had no idea what tax rates they actually paid, and asking whether the simplest possible version of the question (a sales tax in a grocery store) would reveal the same inattention. It does.

Overrated/underrated and the India interlude

In the programme’s rapid-fire segment, Chetty rates Utah underrated for upward mobility — Salt Lake City consistently ranks near the top of his data on multiple dimensions — and attributes it tentatively to community cohesion and the social infrastructure of the LDS Church. He rates A. R. Rahman underrated for fusing Carnatic and Western musical influences, Tamil cinema overrated for formula, and checklists underrated as a tool for decomposing complex research problems into manageable sub-tasks.

On India: Chetty roots the tradition of South Indian mathematical prodigies partly in role-model effects — hearing about Ramanujan from childhood, he suggests, shapes what subsequent generations of children believe is possible for someone who looks like them.

Future directions

At the time of recording, Chetty had recently moved from Harvard to Stanford to be at the heart of Silicon Valley, and his planned next step was using Facebook’s social network data to test whether cross-income connections are causally responsible for mobility differences. The research design exploits the grade-level structure of friendship networks — children befriend classmates in the same year-group but rarely those a year above or below — which generates quasi-random variation in the income composition of a child’s social world. The longer-term aspiration is a recipe for raising upward mobility in specific cities, informed by whatever causal levers the network data reveals, and potentially reshaped by technologies — autonomous vehicles, remote work — that could alter the geography of daily life and with it the geography of integration.

Related

• What Makes Economies Grow — theme; Chetty anchors the people-and-place camp — growth that fails to lift its people, with cross-class integration as the mobility mechanism
• Raj Chetty — guest
• Joe Studwell — fellow guest on Conversations with Tyler; complementary treatment of economic development and the mechanisms behind convergence
• What Makes a Great Investor — theme; Chetty’s framework for deploying research effort (stable, persistent signals over noisy short-run variation) parallels capital allocation logic