18.3 Randomized experiments

• Unbiased estimator -> mean is equal to the estimand

• Efficient estimator -> small variance in the estimation of the estimand.

• Completely randomized experiment.

  • Treatment is independent of potential outcome -> unbiased estimator
  • High variance with small samples.

• Randomized Block experiment

  • Increase efficiency by assigning subject to different blocks with common characteristics.
  • Calculate separate treatment effects for each block, and then combine in a weighted average (for the sample or population as required)
  • Alternative: use a regression with indicators for each block. Example in section 19.2-19.4