Scope of inference

Handle the two randomizations separately, because they answer different questions. Random assignment of treatments makes the groups comparable, so a significant difference can be credited to the treatment; random selection makes the sample representative, so the result extends to the population sampled. For example, randomly assigning treatments to 60 volunteers supports cause and effect for people like those volunteers only, while a random survey of 60 town residents describes the town but cannot establish cause. A study with both random selection and random assignment is the only kind that supports a causal claim about the whole population.