Simulating the MDE

• Set the sample size, n
• Set the control rate, r
• Set the significance level, a
• Set the power, b
• Set the simulation count, nsim=10000
• Set the possible effect sizes d1, …, dk For each effect size d_i:
  • set detected = 0
  • for _ in range(nsim):
    • simulate a control and treatment data set
    • compute the treatment effect CI at sig level alpha
    • check if the CI spans zero. if not, detected += 1
  • the estimate power for d_i is detected / nsim
• Return the smallest d_i such that the observed detection rate >= p