Part III
Quantifying Effects: Experiments, Causality, Regression, and Pricing
From what happened to what to do
This part is about acting on data by asking what would have happened otherwise. Every chapter chases the same object — the counterfactual — and the arc moves from naming it (Chapter 5 reframes any metric as a missing comparison) to earning the word "causal" for a regression (Chapter 6 opens with a from-scratch refresher on Southwest Airlines fares before making the identification argument precise), to recovering the counterfactual from field data no one randomized (Chapter 7's difference-in-differences, synthetic control, and heterogeneous effects), to spending it on the firm's highest-leverage lever (Chapter 8 turns an elasticity into a price, and settles whether Progresso's real habit of raising soup prices in the off-season is smart pricing or a mistake). A single thread of worked evidence — Southwest route fares, Progresso soup scanner data, a 1,700-store milk experiment, a Zillow-and-cannabis synthetic control — runs through all four, so the same number grows more trustworthy as the design tightens. The discipline it leaves behind is refusing to read a coefficient until you know which counterfactual produced it.
4 chapters · 16 articles
What you’ll learn
- Translate any business metric into a precise causal question — naming the lever, unit, horizon, comparison, and decision threshold before fitting a model
- Refresh simple and multiple regression on the Southwest Airlines fare data, then distinguish identification from estimation and demand the identification memo and diagnostics that separate a causal coefficient from a precisely-wrong one
- Recover treatment effects from unrandomized field data using difference-in-differences, synthetic control, and panel fixed effects, and audit each with balance and placebo checks
- Surface heterogeneous effects so a single average lift no longer hides which segments actually pay
- Convert an own-price elasticity into an optimal markup via the Lerner rule, apply it separately by season, and judge in dollars whether Progresso's real countercyclical pricing habit is justified by the data
Chapters in this part
Every metric worth acting on hides a counterfactual you must construct, not assume.
A regression number is only as trustworthy as the comparison it secretly makes.
Build the missing counterfactual — then trust the effect only as far as the design that produced it.
Turning an identified elasticity into a defensible price — and using it to judge whether Progresso's own countercyclical habit is smart pricing or a mistake.
Interactive studios
Hands-on studios paired with this part’s chapters — each opens in a new tab.