Where does multiple testing arise

If an honest author describes the analysis of half a dozen different end-points or 20 sub-group analyses, one of which is significant, then the multiplicity is overt and an educated reader can be suitably cautious. However, the unscrupulous may conduct endless analyses and only report the one that hit the 

Break the data down into larqe numbers of sub-groups and publish the one that produces a significant result 

Wicked The effectiveness of this one depends only on how unscrupulous you are prepared to be. If you are prepared simply to publish the sub-group that best suits your purposes and keep quiet about the fact that it was part of a larger experiment, there is no limit to what you can prove. Those with some residual scruples may prefer to own up to the full data set, but invent some spurious reason why there is justification for focusing on the particular sub-group where a significant result was obtained. 

---