Education and Development

Carolina Torreblanca

University of Pennsylvania

Global Development: Intermediate Topics in Politics, Policy, and Data

PSCI 3200 - Spring 2026

Agenda

The education miracle
Private vs social returns
Duflo (2001)
From private to social
Education challenges today

Deadlines

Final Project due Sunday May 10 by 11:59pm ET
Wednesday April 22: in-class work session

The Education Miracle

The Central Question

Can education make a person rich?
Does it make a country rich?

Schooling

Most of the World Now Literate

Schooling and Income

Schooling and Fertility

Governments Invest

Our Questions

Does education cause development, or just correlate with it?
If it causes it, why? Through what channels?
Does more schooling make a person richer?
Does more schooling make a country richer?
Does more schooling produce other good outcomes: politics, health, safety?

Three Kinds of Return

Private return: how my schooling raises my wages
External return: how my schooling helps others
Social return = Private + External
Policy should care about the social return; I only care about the private one

Private Return: Becker

School builds productive skills
Skills raise productivity, productivity raises wages
This is the human capital story
Measured by the Mincer wage equation

The Mincer Equation

\[\ln(w_i) = \alpha + \beta \, S_i + \gamma_1 X_i + \gamma_2 X_i^2 + \varepsilon_i\]

\(w_i\): wage of worker \(i\); \(S_i\): years of schooling; \(X_i\): years of work experience
\(\beta\): the return to schooling; percent wage gain per extra year of school
Log wage makes the coefficient a percent: \(\beta = 0.08\) means 8% more per year
Typical OLS estimate across countries: \(\beta \approx 0.07\) to \(0.10\)

Private Return: Spence

Employers cannot see your ability; they can see your degree
School is information to the people who pay you
High-ability types find school cheap; low-ability types find it costly; the diploma separates them
Content may not even matter: the credential does the work

External: Spillovers at Work

The idea: your wage depends on other people’s schooling, not just your own

You are more productive when your coworkers are skilled: teamwork, learning on the job
Firms buy better machinery when workers can run it; that raises everyone’s wages
More educated cities attract better employers
This is a true externality: you benefit from their schooling without paying for it

External: Non-Economic

Educated citizens vote more, follow politics, demand accountability
Educated citizens pay taxes, expand the fiscal base
Educated mothers raise healthier children
Educated populations have lower crime and disease spread

The Market Fails

I decide how much school to get based on my wage gain
I do not factor in: your wages, your votes, your kids’ health, your firm’s productivity
So I under-buy schooling; every student does
The country ends up with less education than is socially optimal
Classic externality: markets alone do not deliver the social optimum
This is why every rich country subsidizes or mandates schooling

The Causal Problem

Is the Mincer \(\beta\) Causal?

Mincer regressions say: one extra year of school, ~8-10% higher wages
But schooling is chosen, not assigned
The kids who get more school are also higher ability, from richer families, in better places
OLS bundles the effect of school with everything else that travels with it

What We Need

Variation in schooling that is not driven by individual choice
Ideally: a shock that gave some people more school, for reasons unrelated to ability
Supply-side expansions are a natural candidate
Duflo (2001) uses exactly this

A Quick DiD Refresher

The 2×2

	Before	After
Treated	\(Y_{T,0}\)	\(Y_{T,1}\)
Control	\(Y_{C,0}\)	\(Y_{C,1}\)

\[\hat\delta = (Y_{T,1} - Y_{T,0}) - (Y_{C,1} - Y_{C,0})\]

DiD Is an Interaction

\[Y_{it} = \alpha_i + \lambda_t + \delta (\text{Treated}_i \times \text{Post}_t) + \varepsilon_{it}\]

\(\alpha_i\): unit fixed effects (level differences)
\(\lambda_t\): time fixed effects (common shocks)
\(\delta\): coefficient on the interaction = the DiD estimate
Parallel trends: without treatment, both would have moved the same way

Continuous DiD

Treatment does not have to be 0/1
It can be a dose: more or less of the treatment across units
Interaction: (dose in your unit) × (are you in the post period?)
This is Duflo’s setup

Duflo (2001)

The Setting

Indonesia, early 1970s
Primary enrollment ~60%, large regional gaps
1973: oil prices quadruple; government has cash
Decision: spend it on primary schools

INPRES

1973-1978: 61,807 new primary schools
Roughly doubles the stock of schools
Allocation across 283 regions
Rule: more schools to regions with fewer existing schools per child

The Design

Two dimensions of variation:
- Region: some got many new schools, others got few
- Cohort: kids aged 2-6 in 1974 were fully exposed; kids aged 12+ were too old
Compare: did young cohorts in high-intensity regions get more schooling than young cohorts in low-intensity regions?
This is a DiD with a continuous treatment (intensity × exposure)

Placebo: Older Cohorts

Check cohorts too old to be treated (age 12-17 vs 18-24 in 1974)
Neither group was exposed to the new schools
If intensity still predicts differences, parallel trends unlikely
It does not; the effect turns on precisely at age 12

Duflo Figure 1: Education

Ages 23-13 (unexposed): flat near zero. Ages 12-2 (exposed): rising, significant.

The Main Result

Building schools caused kids to get more education AND earn more as adults
Each year of extra schooling (caused by INPRES) → ~7-11% higher wages
Similar to the OLS Mincer estimate in the same data (~7.5%)

Duflo Figure 3: Education and Wages

Both education (solid) and log(wage) (dotted) turn on at age 12

What It Means

Supply-side access was binding in 1970s Indonesia: more schools → more education
Returns to that extra schooling are economically large
Building schools was a good investment for individuals
But this is the private return; what about everyone else?

The Pivot

Duflo gave us the private return: ~8% per year of schooling
But why do countries invest 4-6% of GDP in education?
Because schooling changes more than just the student’s own life
It changes the lives of the people around them

Quick Check

Your neighbor just finished medical school

Who else is better off because of this?

Her future patients get better care
Her coworkers learn from her
The local tax base grows
Her kids are more likely to finish school too
You, if you ever need a doctor

What Is a Spillover?

A spillover: someone else’s treatment changes your outcome

Private return (Mincer / Duflo):

\[\text{My wage} = \alpha + \beta \cdot (\text{My schooling}) + \varepsilon\]

With spillovers:

\[\text{My wage} = \alpha + \beta \cdot (\text{My schooling}) + \gamma \cdot (\text{Others' schooling}) + \varepsilon\]

\(\gamma > 0\): education has positive externalities
\(\gamma = 0\): no spillover; education is purely private

Why Spillovers Are Hard

Educated people cluster: same firms, same cities, same countries
Your coworkers are educated AND your city pays better AND your firm is more productive
All correlated; none necessarily causal
Clean spillover estimates require exogenous variation in others’ schooling
This is why the spillover literature is much messier than the Mincer literature

Channels of Spillover

How many ways can your education benefit others?

Workplace: educated coworkers → you learn faster; firms invest in better capital
Politics: educated citizens vote more, demand accountability, pay more taxes
Health: educated mothers have healthier children
Safety: more education, less crime
Innovation: educated workers spread ideas; ideas are non-rival

The Virtuous Circle

Economic: educated workers → higher wages → higher productivity → more tax revenue
Political: educated citizens → better accountability → better public goods → better schools
Intergenerational: educated mothers → healthier, better-educated children
Innovation: educated workers → new ideas → spillovers to everyone else
Each channel feeds the others: positive externalities compound

Education Challenges Today

The Learning Crisis

Years of schooling have risen; learning has not kept pace
Rural India: half of grade 5 students cannot read a grade 2 text
In many low-income countries, grade 3 students cannot subtract single digits
Access is now less binding than quality

Why Quality Is Hard

Teachers absent, undertrained, or poorly paid
Curricula designed for the top students, not the median
Parents demand credentials, not skills
Political pressure to expand access, not measure outcomes

The Funding Shock

USAID historically funded ~20% of global basic education aid
2025: US foreign assistance frozen; USAID dismantled
Countries heavily dependent on donor-funded education programs face immediate cuts
Teacher salaries, textbooks, school feeding at risk

Education in the Age of AI

If AI can substitute for cognitive labor, what is the return to schooling?
Optimistic view: AI is a tool; educated users get more out of it; returns rise
Pessimistic view: AI compresses the premium on education; ladder collapses
Open question for developing countries: leapfrog or left behind?
The externalities (politics, health, civic life) do not disappear with AI

Takeaways

Education has many positive externalities
Private return is well identified (Duflo): ~8% per year
Spillovers run through work, politics, health, safety, innovation
Private returns alone do not justify why states fund education; externalities do
The frontier today is learning, not just access