Learn

Designing Experiments

Design experiments that give answers you can trust: controls, randomisation, blinding, fair comparisons, ablations, A/B tests, natural experiments and a written plan before you start.

18 video lessons · about 32 min · free to study

Studying is free. The certificate is for members.

Join APJ-RAISE to do the project, take the quiz and get your certificate. Free, about 3 minutes.

Join now

Your progress is saved in this browser.

  1. Watch · module 1Why design matters3 lessons
  2. Watch · module 2Variables and bias3 lessons
  3. Watch · module 3Fair comparisons in ML3 lessons
  4. Watch · module 4Controls and checks3 lessons
  5. Watch · module 5Online and field experiments3 lessons
  6. Watch · module 6Planning3 lessons
  7. Graded · step 1Hands-on project

    Your experiment plan. Checked by a reviewer.

  8. Graded · step 2Graded test

    20 questions · 30 minutes · 70% to pass · sign in to take it

  9. FinishYour certificate

    For members. Joining is free and takes about 3 minutes.

What you will learn

  • Explain why controls and randomisation make results trustworthy
  • Spot confounders, and use blinding to remove hidden bias
  • Compare ML methods fairly: equal tuning, many seeds, no leakage
  • Design ablations, positive and negative controls, and sanity checks
  • Run A/B tests without peeking, and find natural experiments when you can't randomise
  • Write a pre-registered test plan and run a pilot first

Who it is for: Students about to run a study, a user test or an ML comparison. Statistics for Researchers helps but is not required.