[EMAIL PROTECTED] (Herman Rubin) wrote in message news:<[EMAIL PROTECTED]>... > The difference of the mean and median in a typical > distribution, especially a symmetric unimodal distribution, > is asymptotically normal with a variance O(1/n) if the > variance exists. In such situations, the power of the > test will not improve with sample size.
Yes, for tests of symmetric null against symmetric alternatives it's going to be useless. For a test of a symmetric null (e.g. normal) against a skew (in the sense of Pearson's 2nd coefficient of skewness) alternative it will have some power. If the only alternatives one is interested in detecting are skew, this may not be such a terrible thing to do. Just don't use it for a general alternative. Glen . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
