AI summary by TwinMind

Edward 'Ted' Miguel | Open Science: Assessing How to Do Good Better | Talks at Google

Talks at Google

42m 20s

Introduction and Background

  • Ted Miguel: Professor of developmental economics at UC Berkeley
  • Founder of Berkeley's Center for Effective Global Action (CEGA) - founded 15 years ago
  • Mission: improve research quality informing poverty alleviation and global development work
  • Personal motivation: immigrant family background, early exposure to global poverty drove career focus

Open Science Philosophy and Methodology

  • Found development economics field lacking data-driven, scientific approach when starting career
  • Decisions driven by ideology rather than hard evidence and data
  • Committed to bringing best possible data and statistical methods to global poverty debates
  • Tools introduced: randomized control trials (RCTs), machine learning, new data sources
  • Open science principles from start:
    • Share data openly with other scholars (rare 20 years ago in economics)
    • Publish all results including unexpected/null results
    • Avoid publication bias - publish regardless of donor/government preferences
    • Share results regardless of statistical significance
  • Combination of rigorous methods + open science ideals = influential work

Deworming Study - Major Success Case

  • First research project, started in grad school with co-author Michael Kremer (Chicago)
  • Worked with NGO providing school-based intestinal worm treatment in Kenya
  • Context: billions globally have intestinal parasites causing nutritional/health problems
  • Treatment: simple pill taken once/twice yearly, costs <$1/year
  • Study design: randomized order of scale-up across schools - one of first RCTs in development economics
  • Short-term results (few months): kids receiving treatment attended school more, healthier, worms eliminated
  • Long-term follow-up (25 years):
    • Kids earn ~10% more in adulthood
    • More likely to move to cities, get more schooling
    • Incredibly high rate of return given $1 treatment cost
  • Policy impact: hundreds of millions of kids/year now receive deworming drugs globally
  • Programs in India, Nigeria, Pakistan, Ethiopia, Kenya explicitly cite this research
  • Data shared publicly for scientific community critique and building

Rural Electrification Study - Null Results Example

  • Started 12-13 years ago with Kenya government's major rural electrification effort
  • Kenya early adopter compared to neighbors
  • Randomized evaluation with Kenya Power company
  • Expected transformative impacts on economy, children's schooling (study at night)
  • Results: null effects across economic outcomes, agricultural productivity, entrepreneurship, schooling
  • Reasons discovered:
    • Median number of appliances was zero
    • Typically only had light bulb, sometimes charged phones
    • Can't afford appliances to use with electricity
  • Education findings: kids studied 20 minutes more evening, 20 minutes less afternoon - net zero change in test scores
  • Contrast with US 1930s New Deal rural electrification: actively subsidized appliance purchases, documented benefits
  • Published null results despite potential disappointment to funders/government

Cash Transfers Study - Broader Impact Measurement

  • Collaboration with GiveDirectly NGO in Kenya (2013-2014)
  • Designed large-scale study to measure broader spillover effects
  • Randomized at higher aggregation levels (sublocation) - certain local economies got lots of cash, others didn't
  • Surveyed cash recipients, neighbors, conducted census of local enterprises, measured market prices
  • Results:
    • Replicated existing studies showing direct benefits to cash recipients
    • Measured multiplier effects: 2.5x multiplier (every $1 generates $2.50 economic activity)
    • Local businesses benefit: revenue increases, hire more workers including from non-recipient households
    • Minimal price inflation found
    • Long-term health tracking: infant mortality rates fell >1/3 in cash areas
    • Cash + health clinic access = much larger child survival benefits than cash alone

Machine Learning Applications

  • CEGA leader in bringing ML tools to development work
  • Togo pandemic response example (colleague Josh Blumenstock):
    • Government needed to target cash transfers to poor but lacked census/household data
    • Used ML to predict poverty using cell phone records, remote sensing data
    • Within ~1 month identified several hundred thousand poorest households
    • Government + GiveDirectly targeted cash transfers successfully
    • Approach now adopted in multiple other countries
  • Google.org + GiveDirectly hurricane response in Florida:
    • Satellite imagery + ML tools identified neediest areas
    • Cash transfers delivered in days rather than weeks

Intervention Effectiveness Comparisons

  • Micro finance: large-scale RCTs found minimal benefits on average
    • Most people aren't great entrepreneurs
    • Small subset start effective businesses, others don't utilize cash well
    • Near zero benefits for large-scale programs on living standards
  • One Laptop Per Child: well-studied in large-scale RCTs, found minimal benefits
  • Both had significant enthusiasm/hype but research showed limited impact
  • Research helps move policy and funding priorities based on evidence

CEGA Organization and Structure

  • Based at Berkeley, spans multiple West Coast campuses
  • 160+ faculty affiliates, hundreds of grad students
  • Partners: Stanford, UCSD, UCLA, University of Washington
  • Distinguishing features:
    • Interdisciplinary from start: technologists, computer scientists, climate experts
    • ~2/3 economists, 1/3 other disciplines bringing new technologies/tools
    • Methodological flexibility: early RCT adopters, now ML pioneers
    • Strong commitment to early-stage researcher support
    • Partnerships with institutions in East/West Africa, Bangladesh
    • Inclusive global vision for research impact

Scaling and External Validity Challenges

  • Government engagement: launch projects in partnership, include in early discussions
  • CEGA maintains relationships with governments, foundations, NGOs for research partnerships
  • External validity concerns: results may not transfer across settings
  • Deworming example: worked in area with >90% infection rates - wouldn't expect impact where no infections
  • Rural electrification: richest 10-15% of villages gained from electricity (bought appliances, started businesses), other 90% didn't
  • Need sufficient data and analysis to help governments plan over time
  • Equity considerations: some interventions may exacerbate inequality in short run

Research Surprises and Lessons

  • Rapid adoption of improved methods in development economics field
  • What was unusual 10-15 years ago now commonplace: RCTs, new data sources, ML tools, pre-analysis plans
  • Many programs fail - lots of null results in practice
  • Sierra Leone governance reform example: World Bank program to weaken traditional chiefs, empower women - found null results across outcomes
  • Traditional institutions very persistent
  • Need to temper enthusiasm in absence of evidence
  • Rigorous methods + commitment to share all results regardless of findings = powerful combination

Funding and Future Research Priorities

  • Balance between funding proven interventions vs. research for future discoveries
  • Billions in low/middle-income countries across scores of countries still unexplored
  • Research investment makes future giving more effective
  • Without past research wouldn't know about deworming, cash transfers, bed nets returns
  • Pace of learning and evidence generation not slowing down
  • Need long time horizons to keep identifying wins for tomorrow, next year, 10+ years from now

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome