 |
 |
>My problem here is knowing what value of p to put in step (iii). Victor and >Thom seem to be suggesting a test of p = 0.5, but personally I can't see why >that hypothesis would be of any interest. Under what model of people's >behaviour would there be anything special about people coming back with >probability 0.5? If there is some theory that makes this value interesting, In fact I broadly agree. The equiprobably test merely indicates evidence that they return less often than if the decision were random with no preference for returning or not returning. It isn't interesting given that we wouldnt expect this decision to be random with no preference. (One might argue it is a conservative test under the assumption that most patients _do_ return for a second dose for most treatments.) >Jim Fowler raised the question of whether there is any sampling at all. >Unless the question is made clearer, I think we can't say whether there is >or not. If the question is just about these 20 people, then Jim is right, >the proportion who came back is exactly 5 per cent, and that's all there is >to say. I suspect the question is not just about these people, though, but >about some (possibly ill-defined) population from which they were drawn. To >that extent it's like most medical research, where we have data from a >sample of patients, usually those who turned up to the right hospital at the >right time, satisfied the trial entry criteria, and agreed to participate, >and we make an inference to a wider population of 'all such patients' which >isn't too clearly defined. I agree entirely here. A significance test might still be appropriate if we wish to generalize to other possible observations on these same people. Thom