32
Implementation
Define “rare data” for each hypothesis
•
Choose test statistic
–
Often based on an estimate of treatment effect
–
–
–
Reject low treatment effect when estimate is so
high as to only occur, say, with 5% probability
