Date: Feb 8, 2013 10:09 AM
Author: Luis A. Afonso
Subject: Re: Sample Distributions by Simulation

Follow-up
In what concerns natural data we rarely know what probability model we are dealing with. If so, one is restricted to use replications of the data we try, even so to estimate the parameters we are interested to know. The procedure is not completely safe; in fact the information brought is not as sound as we wanted to be. The reasons are; the sample could not be Population-representative, the replications definitively are not completely equivalent to fresh data, their items random and equally independent distributed, i.i.d., and consequently the parameters are shifted by a not computable amount. Restrictive features we must live with . . .
Conclusion
The use of models, (1st post) with synthesized data even the Population constitutive parameters are know with sufficient accuracy, which going without saying is very hard to achieve, have the drawbacks yet indicated: it can be always argue against that there is not, surely, a natural Population following exactly a statistical model even it was lead to high accuracy. In contrast when we prefer simulative methods we have to deal with inaccuracies here sketched. In every case a Principle, Occam´s Razor Blade, must be taken in consideration: among all possible models of an experimental or observational phenomenon we should choose the simpler . . . resulting that the introduction of deficiently estimated parameters is avoided.
The dilemma is like to cho0se between: models - perfect but lacking realism - or ? only pretty refined but more or less close to the true world.
_______
Another treatment option, relative to p-values determination, will follow, I hope.

Luis A. Afonso