Frequentist conclusion depends on investigator’s intentions

• Inv. 1 planned several interim analyses

• Inv. 2 planned none (or says none!)

• They get identical data

• Inv. 2 claims statistical significance

• Inv. 1 cannot—only because looked at data with possibility of early stopping

• To avoid penalties, frequentists suggest not looking. Seems unscientific.

Previous slide

Next slide

Back to first slide

Slide Index

PNNL Bayesian homepage

View graphic version