On 10 Apr 97 21:32:03 EDT David Thiel <firstname.lastname@example.org> writes:
>OK, but without a "line in the sand" to begin with, 25 people could >have >25 different interpretations of its significance. > >What if a new drug has been tested to treat some malady. Are we >saying >that the P-value should be printed on the bottle and the patient left >to >interpret whether it does what it was intended to do? That may be an >extreme example, but without a benchmark to start with, it's almost as >if >we can fudge a little. Seems a bit unethical, really. > >I realize that P=.049 and P=.051 are not practically different, but if > >alpha=.05, it makes a world of difference. Or does it? It's tough >enough that statistics is not an exact science.
So who sets alpha=0.05? God? I'm not that existential, but it seems that we're stuck with the decision for an alpha value.
Also, in any particular case the relevant community will already have decided (or be in the process of deciding) what is acceptable and what is not. For instance, in education, one sometimes sees accepted p-values of 0.15 and even lower because instructional situations are not nicely analyzed quantitatively. (This is partly why a lot of math ed. research today qualitative and not quantitative.) Or in industry, decision-makers have to decide what level of risk they are willing to accept.