The Cutoff Trick That Proves Causation Without a Single A/B Test

A hands-on guide to Regression Discontinuity Design in Python — the causal inference method hiding in plain sight at every exam pass mark, age limit, and funding threshold.

In 2004, a kid in India scored 649 out of 700 on a scholarship exam. The cutoff was 650.

One point. That’s it. One point separated her from four years of paid tuition, mentorship, and a resume line that would follow her into every job interview for the next decade. The kid sitting next to her scored 650 and got all of it.

Nothing about these two students was meaningfully different. Same neighborhood, same prep classes, same nervous energy walking into that exam hall. If you tested them again tomorrow, their scores could easily flip. But one number — one arbitrary line drawn by an administrator — decided who got the scholarship and who didn’t.

Here’s the uncomfortable part: that line is a gift to anyone who wants to measure cause and effect.

Most people think you need a randomized experiment to prove causation — flip a coin, assign treatment, compare groups, done. It’s the gold standard, and for good reason. But real life rarely hands you a coin flip. Governments don’t randomize who gets a subsidy. Schools don’t randomize who gets held back a grade. Hospitals don’t randomize who gets flagged as high-risk.