1. matching should be used for stratified sam- with a potential effect modifier many con
2. pling with respect to an exposure of trols may be required to achieve a specified
3. interest or a proxy. level of efficiency. Matching and counter-Reversal of these roles is not advanta- matching provide alternative means to this
4. geous. Matching for a variable closely asso- end by using information collected for the
5. ciated with exposure is known to lead to entire cohort to stratify the sampling in