Teste para igualdade de médias independentes com erro normal assimétrico

Abstract

In design studies, the normality of the error terms is the usual assumption w hen tests for eq uality of independent means are performed. H ow ev er, this assumption could b e unrealistic and the inferenc es w hich come from these tests can not b e reliab le. In this paper, the normality assumption w as encapsulated b y the sk ew normally distrib uted error assumption. By the w ay, a new test for mean eq uality w as proposed, b ased on q uadratic forms. It w ill b e used to test the ov erall eq uality hypothesis of independent means in the Bayesian approach. A practical application w ith a real data set w as dev eloped and discussed. O n the ev aluation of the performanc e of the proposed test, it w as presented a M onte Carlo simulation in sev eral ex perimental conditions. M oreov er, the simulation results w ere compared w ith the usual F test. The proposed test has presented good performanc e b y b eing ex act or conserv ativ e in all simulated designs and, at the same time, has presented close pow er rates to the usual F test.