16.6 Design analysis using fake-data simulation

• Most general and often clearest method to understand a future study: Simulate it!

• Example using a randomized experiment on 100 students designed to test an intervention for improving scores.

  • Assumes intervention effect is a 5 point addition to the students score.
  • Final exam score are assumed to be random with mean 60 and standard deviation 20

See example “FakeMidTermFinal” in Tidy Ros

• Demonstrates effectiveness of using a pre-treatment predictor:

  • Reduce error on effect estimate
  • Control for selection bias