Nonparametic tests ARE simply parametric tests applied after a (rank) 
transformation of the data.

You have a problem with heterogeneity of variance?  Just use the separate 
variances t test rather than the pooled variances t test.  It has been 
around since the 1940's, but most psychologists seem not to be aware of it. 
So sad.  Major stat packs provide it whenever you do a t test.   More than 
two groups?  Use a Welch test.  Your stat pack probably provides it as an 
option when you do ANOVA.

Significance testing variances or normality prior to selecting a test?  VERY 
POOR PRACTICE.  The parametric tests will be most sensitive to violations of 
their assumptions when sample sizes are small, but in that case the tests of 
assumptions will have so little power that you will not be able to detect 
serious problems.  With larger sample sizes the parametric tests will be 
more robust to such assumptions and the tests of such assumptions will have 
so much power that they will detect as significant deviations from equal 
variances or normal distributions that are so minor as to be of absolutely 
no consequence.

Cheers,

Karl W.
http://core.ecu.edu/psyc/wuenschk/KLW.htm
----- Original Message ----- 
From: "Daniel Leucuta" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Wednesday, April 12, 2006 3:23 PM
Subject: Re: Dilemma about testing normality, and equality of variance to 
check assumptions for tests before applying them


Thanks everyone for your precious help!!!

I really enjoyed, the answers and find them useful. They revealed to
me a part of the beauty of statistics :)

A side question relating to the one asked is: what about transforming
non-normal data and testing it with parametric tests. Is this a better
approach than using non-parametric tests? What are implication to the
interpretation of the results?

Kind regards!!!

Daniel

Subscription, posting, archives, etc. can be found at 
http://lists.psu.edu/archives/edstat-l.html



---
To make changes to your subscription go to:
http://acsun.frostburg.edu/cgi-bin/lyris.pl?enter=tips&text_mode=0&lang=english

Reply via email to