In this example the underlying structures are mechanical. In cases of interest for health and social policy they will be a mix of institutional, psychological and physical. The basic lesson is the same. Different underlying structures yield different causal and probabilistic relations. The problem is we often do not understand these underlying structures nor how they work to give rise to the causal relations an intervention might use. So we don't know when (i) is satisfied. For some causal relations it may be good to assume, as one economist recently claimed, that people are much the same wheresoever they are; for others that assumption can be disastrous. So the demands for exporting effect size from study to target population are generally far too great.