Date: Nov 30, 2012 4:06 PM
Author: Luis A. Afonso
Subject: A short comment on a Elena. Kulinskaya paper
The A. starts (Abstract) with the assertive justification ?two side statistical tests and p-values are well defined only when the test statistical in question have a symmetrical distribution.? And so invented a way to correct this ?anomaly? by means a ?conditional p-value, Pc? which is obtained by weighting the acceptance interval two sides, left and right, in order that it accounts with the density unique mode: The weights, sum 1, are chosen inverse proportional to the probability tails, and alpha= P(X <=A) + P(X >A), X denoting the test statistics, and A location parameter value like the mean, median or mode. . .
The only and obvious critic to the method is that probabilities we chose for the tails are completely independent from the density shape: alpha/2 for each represents the feature. The fact that p-values are preferentially near to one of the Confidence Interval bounds is completely irrelevant, I mean.
Because arXiv is unmoderated to dump whatever is allowed.
Luis A. Afonso