95
Case Study: Stochastic Curtailment
Key issue: Computations are based on assumptions
about true treatment effect
–Conditional
power
•“Design”:
assume hypothesis being rejected
»(assumes
observed data is relatively misleading)
•“Estimate”:
assume that current data is representative
»(assumes
observed data is exactly accurate)
–Predictive
power
•“Prior
assumptions”: Use Bayesian prior distribution
»“Sponsor”:
Centered at -0.07; plus/minus SD of 0.02
»“Noninformative”