iiinova — where your digital self keeps evolving

deankarlan.bsky.social @deankarlan.bsky.social · 24/06/2026

More evidence wouldn’t have saved USAID. Read my latest installment, a response to Todd Moss, in the “RCTs Can’t Do Everything” (but they can do a lot) franchise of blog posts and thought pieces. deankarlan.substack.com/p/more-evide...

deankarlan.substack.com

More evidence won’t save development, but it will help the poor

Revisiting an old conversation

010

deankarlan.bsky.social @deankarlan.bsky.social · 17/06/2026

Thanks Oliver for hosting, was great to talk with you and share these thoughts. Hope they resonate!

010

Reposted by @deankarlan.bsky.social

Oliver Hanney @olihanney.bsky.social · 16/06/2026

🆕 Moving billions towards evidence @deankarlan.bsky.social joined me on Ideas in Development to discuss how he and his team moved ~1.7 billion in funding at USAID, the growth of embedded labs and how they work, and the two key gaps between research and policy.

123

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

See www.nber.org/data-appendi... for a reply from BBBGH that we hope others find as useful as we did, thank you BBBGH team for laying out the heterogeneity issue so eloquently and clearly. We agree!

031

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

The more useful question: what investments are actually available in a community, what do they cost, and what do they return? Without that mapping, we can't say whether a tipping point exists, where it sits, or how big a push someone needs. Working paper: www.nber.org/papers/w35253

nber.org

The Sisyphean Pursuit of Evidence for Poverty Traps

Founded in 1920, the NBER is a private, non-profit, non-partisan organization dedicated to conducting economic research and to disseminating research findings among academics, public policy makers, an...

110

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

We are NOT arguing that poverty traps don't exist. We almost certainly believe they do exist for some families in some settings. We are arguing that finding clean, generalizable evidence for one specific tipping point is much harder than the field has assumed.

110

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Maybe a tipping point is hiding in less-poor populations? We look at long-running rural surveys covering all of rural Bangladesh and all of rural Ghana, tracking wealth over 8 years. No tipping point appears in either country.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

What is “geography” capturing? A lot. Six of eight low-cow-price branches cluster in Rangpur's northeast — Bangladesh's poorest region, hit hard by river erosion. High-cow-price areas have more education, more electricity, more NGO presence, and higher assets at baseline.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Strongest evidence that this is geography (or geography-esque), not personal wealth: even families in the CONTROL group — who got no cow at all — accumulated assets faster if they lived in a high-cow-price area. A household wealth tipping point can't explain that.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

When we split villages into high-cow-price and low-cow-price groups and look within each, the tipping-point pattern disappears. The apparent "trap" was driven by mixing different villages, not by different family trajectories.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Local cow prices explain about four times as much of who lands above versus below the line as families' own starting assets do. Nearly half of households are above simply because their village has high cow prices.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

That variation is much bigger than the variation in pre-transfer household assets. So whether a family ends up "above" or "below" the line depends mostly on where they live, not on what they had.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Issue #2 (and the bigger one) is geography. The measure of household wealth combines (a) what each family owned before, plus (b) the median cow price in their local BRAC branch. Cow prices vary a lot across branches.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Issue #1: The S-curve depends critically on the log transformation. BBBGH use log(x/1000+1); levels (x), log(x+1), and log(x/100+1) give qualitatively different pictures. The divisor isn't innocuous.

101

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

But, two problems with how the tipping point is identified for the poverty trap interpretation in BBBGH.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

So back to BBBGH. First, absolutely critical: we replicate the average treatment effects just fine from Bandiera et al.’s earlier QJE article. BRAC’s program rocks. And kudos to BRAC for repeatedly iterating & testing (e.g., nber.org/papers/w34309, at AER:I pending data check).

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

This doesn't prove traps don't exist. Maybe samples are too small, or the transfers didn't land people near the tipping point. Or there is too much unmodeled heterogeneity. But across seven other contexts using the same approach as BBBGH, we do not find such evidence.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Result: in none of the seven do we find the tipping-point pattern. Either everyone is rising on average, or everyone is falling, or wealthier starters do slightly worse. None show the "below = sliding, above = rising" signature.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

We start by asking: do we see the same pattern in OTHER Graduation programs? The model has been tested in Bangladesh, Ethiopia, Ghana, Honduras, India, Pakistan, and Peru. Seven studies. Same approach.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

Their finding: 11 years later, families who landed just above a wealth threshold were thriving. Just below it, they were sliding back. The line sat right around the price of a cow. Striking. Influential. Widely cited.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

The most influential recent evidence comes from Balboni, Bandiera, Burgess, Ghatak and Heil (2022) - BBBGH - studying BRAC's "Graduation" program in Bangladesh. The program gives ultra-poor families a productive asset - usually a cow - plus training and support.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

This idea shapes how we think about foreign aid, cash transfers, and big-push programs. If traps are real, one well-designed shove can change a life. If they aren't, gains may fade once the support stops.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

The "poverty trap" idea: there's some level of wealth where, below it, you keep sliding back; above it, you keep rising. A tipping point. Push someone over it, and life changes. Stay below, and you're stuck no matter how hard you work.

100

deankarlan.bsky.social @deankarlan.bsky.social · 25/05/2026

New working paper with Amol Singh Raswan and Chris Udry: "The Sisyphean Pursuit of Evidence for Poverty Traps." A central idea in development economics is that poverty can trap people. We went looking for the cleanest evidence. Here's what we found – and didn't.

24224

deankarlan.bsky.social @deankarlan.bsky.social · 19/05/2025

because the abstract should be the highlights! wastes time to come up with two sets of highlights. and they rejected me when i tried to just repeat the abstract in bullet form. hence decided to have fun with it instead.

040

Reposted by @deankarlan.bsky.social

BSPA @bescipol.bsky.social · 09/04/2025

Join us for our annual online conference 5/1-5/2! Learn how to apply behavioral science in your own backyard, explore ways to collaborate, and hear from @nickkristof.bsky.social, Dean Karlan, Kevin Volpp, and many more. Register today! bit.ly/4iWXKoY #behavioralscience #BehSci #BehEcon

001