Given variation in numerous covariates,
epidemiologists generally pursue a strategy of
attempting to isolate the effect of a single
measured quantity by conditioning in some way
(stratification, modeling) on covariates
We want to estimate the effect of E on D,
separate from the effects of covariates Z1….Zn
We usually want to answer the (causal) question: what would happen to D if I fiddled with E, while
letting the other variables run their natural course?
Compare Pr(D=1|Set[E=1]) with Pr(D=1|Set[E=0])
| Previous slide | Next slide | Back to first slide | View graphic version |