Tinker A/B testing

by Kondasamy Jayaraman · drawn with Colorfigures by April Zhu

The same checkout on both sides.

Chapter 2 of 7

Luck

The new checkout won by about two points. Then someone asks: would two identical checkouts differ too? The team pits today’s checkout against itself.

This chapter: How big a gap can luck alone make?

About this chapter

You will learn

  • An A/A test shows the noise between identical groups
  • What significant means
  • Why a few A/A tests still call a winner

Words you’ll meet

A/A test
A test of a version against an identical copy of itself: any gap is luck.
Noise
Gaps that come from which shoppers happened to buy, not from any change.
Significant
A gap bigger than luck alone makes 19 times in 20 (a 5% level).
False win
A significant result when nothing truly differs.

Step 1 of 5

An splits shoppers between two identical checkouts. Any gap is : luck in who happened to buy.

Run the A/A test.