01
RCT
Randomized trials
ATE = E[Y1] − E[Y0]
Random assignment eliminates selection bias
A FIELD GUIDE TO CAUSE & EFFECT
Mastering Causal Metrics
Better questions.
Fairer comparisons.
Does insurance improve health? Does education raise earnings? Learn how researchers move from a compelling question to credible evidence.
New to causal inference? You’re in the right place.
A learning project by Carlos Mendez
Based on Mastering ’Metrics
by Joshua Angrist & Jörn-Steffen Pischke.
THE FURIOUS FIVE01 / 05
Randomized trials
Does health insurance
improve health?
No clear general-health improvement.
In RAND, the free-care, deductible, and coinsurance effects on general health all have 95% intervals spanning zero. These comparisons are against catastrophic coverage, not no insurance.
Explore randomized trialsRecomputed RAND results · family-clustered intervals.
Colab source ↗ · Artwork is illustrative.
Regression
Does a pricier college
pay off?
The premium shrinks with controls.
Chapter 2 summarizes the private-college premium as about 14%, then 7%, then close to zero as controls improve. The study data are unavailable here; this is the notebook’s literature summary.
Explore regressionNotebook literature summary · approximate findings.
Colab source ↗ · Artwork is illustrative.
Instrumental variables
Can a school lottery
reveal a school’s impact?
A positive effect for lottery compliers.
Chapter 3 reports a +0.48 standard-deviation math effect of KIPP attendance for lottery compliers. The causal reading requires the lottery to affect scores only through attendance.
Explore instrumental variablesNotebook-reported KIPP estimates · SD = standard deviations.
Colab source ↗ · Artwork is illustrative.
Regression discontinuity
Does alcohol access
change health risks?
An upward jump at age 21.
These age-cell mortality rates and separate linear fits reproduce Chapter 4. Mortality rises at the cutoff; a causal reading requires other determinants to change smoothly there.
Explore regression discontinuityObserved age cells and fitted lines · Chapter 4.
Colab source ↗ · Artwork is illustrative.
Differences-in-differences
Can a policy keep
banks from failing?
Fewer losses, not a recovery.
Bank counts fall in both districts. In 1934, the 6th has 102 banks versus a counterfactual of 79. The 23-bank difference relies on the parallel-trends assumption.
Explore differences-in-differencesNotebook bank series · dashed line is counterfactual.
Colab source ↗ · Artwork is illustrative.
Changes every 10 seconds
6 chapters / 5 causal methods / Learn by doing
A question is just the beginningYOUR FIRST STEPS
From “what if”
to “how do we know?”
You don’t need all the answers to begin.
Just a question worth investigating.
- 01 / READ
Read the question.
Start with health insurance. Meet the people, the problem, and the comparison behind the evidence.
Begin the first chapter - 02 / EXPLORE
Explore the evidence.
Open a guided notebook, make a plot, and notice what the data can—and cannot—tell you.
Meet the RAND data - 03 / EXPLAIN
Make the idea yours.
Explain the comparison in your own words. Work through your questions with an AI tutor.
Think it through with a tutor
SIX CHAPTERS. QUESTIONS THAT MATTER.
Learn to see
cause differently.
Follow the chapters in order, or return to the question you’re working on. Read, code, watch, or listen—each chapter gives you a way in.
The Furious Five
A visual overview of the methodsFive methods to separate cause from correlation.
02
OLS
Regression
Y = α + βX + ε
All confounders are observed and controlled
03
IV
Instrumental variables
βIV = Cov(Y, Z)Cov(D, Z)
Instrument affects Y only through D
04
RD
Regression discontinuity
τ = limcutoff Y+ − limcutoff Y−
No manipulation of the running variable
05
DiD
Differences-in-differences
δ = (YT,post − YT,pre)− (YC,post − YC,pre)
Parallel trends in the absence of treatment
01Randomized TrialsStart here
Does health insurance make people healthier?
Begin with a question that matters. Discover how the RAND experiment and the Oregon lottery help researchers build a fair comparison.
AI podcast · 39 min
Chapter 1 · Randomized Trials
Now: Opening
00:0038:56
02Regression
Is an elite college worth the extra tuition?
Students arrive with different backgrounds and ambitions. Explore what regression can account for—and what it can still miss.
AI podcast · 40 min
Chapter 2 · Regression
Now: Opening
00:0039:31
03Instrumental Variables
What if people ignore their assigned treatment?
A lottery can change an opportunity without deciding what people do. Follow that variation to understand whose causal effect we can learn about.
AI podcast · 35 min
Chapter 3 · Instrumental Variables
Now: Opening
00:0035:22
04Regression Discontinuity
Can a birthday reveal the effect of alcohol access?
At age 21, the rules change. See how comparing people just either side of a cutoff can reveal a local causal effect.
AI podcast · 35 min
Chapter 4 · Regression Discontinuity
Now: Opening
00:0035:12
05Differences-in-Differences
Did emergency lending save banks?
Two Federal Reserve districts faced the same crisis with different policies. Compare their changes over time—and examine the assumptions.
AI podcast · 32 min
Chapter 5 · Differences-in-Differences
Now: Opening
00:0031:55
06The Wages of SchoolingBring it together
Does more education cause higher earnings?
Bring the methods together. Investigate schooling, ability, and credentials through twins, birth timing, and policy changes.
AI podcast · 46 min
Chapter 6 · The Wages of Schooling
Now: Opening
00:0045:33
Podcasts narrated with Gemini · AI videos generated with NotebookLM · Lecture videos by Marginal Revolution University
TURN UNDERSTANDING INTO PRACTICE
Less setup.
More discovery.
Open a Python notebook right in your browser. Follow the code, explore real data, and connect each estimate to the comparison behind it.
Google Colab’s AI tools can help explain code and errors. You bring the questions—and check the reasoning.
Open your first notebook Runs in Google Colab · no local installationVideo: Julianne DeMars-Smith and Key Jung Lee · Google Research
KEEP FOLLOWING YOUR CURIOSITY
There’s more to explore.
Books, courses, and practical guides for your next question.
No resources found in this category yet.