Greetings from Al-Ain the oasis city of the Middle East. A very interesting
question and I am not sure whether this is the right answer. But I will give
it a try. I found a few web sites discussing this issue.
Several assumptions about the process of collecting data and the shape of
the population distribution are usually made when using statistical methods.
When you reject the null hypothesis in a test, then a reasonable conclusion
is that the null hypothesis is false, provided all the distributional
assumptions made by the test are satisfied. If the assumptions are not
satisfied then this might be the cause of rejecting H0 . Hence, you should
always check assumptions to the best of your abilities.
According to the above (1) I think that if a we use a parametric test where
the normality assumptions is not fulfilled then we are more likely to commit
Type I error. We conclude that there is a difference when there is actually
no difference. To what extent this will happen – I hope our stats colleagues
will enlighten us on this
A chapter from Martin Bland’s book on the web. (His home page is worth a
visit as it contains a wealth of information on stats)
An excellent web article on non-parametric tests.
An excellent article on parametric test.
I hope this is of some help. (I stand to be corrected if I am wrong - Stats
is not one of my strong areas!)
Cheers & regards,
Dr.P.Badrinath M.D.,M.Phil.,(Epid) PhD(Cantab)
Clinical Assistant Professor and Epidemiologist,
Department of Community Medicine,
UAE University, PO Box 17666, Al Ain,
United Arab Emirates.
Tel: 00 971 3 7039 652
Fax: 00 971 3 7672022.
"For an excellent review of the current medical literature, go to Journals
Scan www.uaeu.ac.ae/jscan/index.htm" - BMJ 3rd June 2000,Reviews(Netlines)
Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp