English

Student's $t$-test for scale mixture errors

Statistics Theory 2016-08-16 v1 Statistics Theory

Abstract

Generalized t-tests are constructed under weaker than normal conditions. In the first part of this paper we assume only the symmetry (around zero) of the error distribution (i). In the second part we assume that the error distribution is a Gaussian scale mixture (ii). The optimal (smallest) critical values can be computed from generalizations of Student's cumulative distribution function (cdf), tn(x)t_n(x). The cdf's of the generalized tt-test statistics are denoted by (i) tnS(x)t_n^S(x) and (ii) tnG(x)t_n^G(x), resp. As the sample size nn\to \infty we get the counterparts of the standard normal cdf Φ(x)\Phi(x): (i) ΦS(x):=limntnS(x)\Phi^S(x):=\operatorname {lim}_{n\to \infty}t_n^S(x), and (ii) ΦG(x):=limntnG(x)\Phi^G(x):=\operatorname {lim}_{n\to \infty}t_n^G(x). Explicit formulae are given for the underlying new cdf's. For example ΦG(x)=Φ(x)\Phi^G(x)=\Phi(x) iff x3|x|\ge \sqrt{3}. Thus the classical 95% confidence interval for the unknown expected value of Gaussian distributions covers the center of symmetry with at least 95% probability for Gaussian scale mixture distributions. On the other hand, the 90% quantile of ΦG\Phi^G is 43/5=1.385...>Φ1(0.9)=1.282...4\sqrt{3}/5=1.385... >\Phi^{-1}(0.9)=1.282....

Keywords

Cite

@article{arxiv.math/0610838,
  title  = {Student's $t$-test for scale mixture errors},
  author = {Gábor J. Székely},
  journal= {arXiv preprint arXiv:math/0610838},
  year   = {2016}
}

Comments

Published at http://dx.doi.org/10.1214/074921706000000365 in the IMS Lecture Notes--Monograph Series (http://www.imstat.org/publications/lecnotes.htm) by the Institute of Mathematical Statistics (http://www.imstat.org)