Causal Inference for Business: Difference-in-Differences, Matching & Natural Experiments
Tell cause from correlation in Excel: confounding and Simpson's paradox, randomised tests, difference-in-differences with LINEST, parallel-trend checks, nearest-neighbour matching, propensity-score weighting, regression discontinuity and instrumental variables, in 8 graded labs on one retailer's data.
The first lesson is free — no account needed.
About this course
A causal inference course for analysts who need to tell cause from correlation, built on one fictional retailer's data in a spreadsheet. You show how a coupon that looks harmful in the totals helps in every segment (Simpson's paradox), estimate the average treatment effect of a randomised store layout test and check balance, measure the effect of Sunday opening with a 2×2 difference-in-differences table and confirm it with a LINEST regression with an interaction term, test parallel trends with pre-period slopes, a placebo test and a chart, pair trained stores with their nearest untrained neighbour for the ATT, weight customers by inverse propensity scores, measure the jump at a loyalty-status cutoff with regression discontinuity and use a randomised reminder e-mail as an instrument in the Wald estimator. The first lab is free.
Lessons
- Correlation vs Causation: Confounding & Simpson's ParadoxFree
- The Gold Standard: Randomised Experiment & Average Treatment Effect Pro
- Difference-in-Differences: The 2×2 Table and the Regression Pro
- Checking Parallel Trends: Pre-Period Slopes, a Placebo Test & a Trend Chart Pro
- Matching: Nearest-Neighbour Controls with MIN, MATCH & INDEX Pro
- Propensity Scores & Inverse Probability Weighting (IPW) Pro
- Regression Discontinuity: The Jump at the Gold-Status Cutoff Pro
- Instrumental Variables: The Wald Estimator from a Randomised Nudge Pro