PakMediNet Discussion Forum : Biostatistics : NORMAL DISTRIBUTION

PLEASE GUIDE ME:

WHAT IS THE DIFFERENCE BETWEEN NORMALLY DISTRIBUTED DATA AND NON-NORMAL DATA?

HOW TO DETERMINE THAT WHETHER THE DATA IS NORMALLY DISTRIBUTED OR NOT?

[Edited by tariqasl on 20-03-2011 at 01:14 PM GMT]

Posted by: ** tariqasl Posts: 42** :: 20-03-2011 :: | Reply to this Message

I know that I am replying very late, but still I think, following may be helpful;

You have to understand that Normal (Gausian) distribution is the one of the most important thing/formula in Statistics.

Now, there is an important theorem in Statistics (you may call staisticians' pet) known as Central limit theorem (CLT). Keeping mathematics aside, CLT (not a disease name for you doctors!)says in simple english that if a sample has 30 [a magic number (dont ask me why)] or more observations, then the sample follows normal distribution. That's what we statisticians want to hear as most of statistical methods (correlation, regression, ANOVA to name few) are based on the assumption that sample follows normal distribution.

The prerequiste of most of the statistical methods is the assumption of normality (i.e. data follows normal distribution) which users and experts alike usually do not perform and hence have wrong conclusions. This is because the specific method is based on the assumption of normality (there may be other as well), which data is not following. In order to have reliable results, my humble advice is, before applying any statistical test first look the assumptions and then decide whether one can apply the spefic test or not.

Next the million dollar question is how to check the normality? Answer is very simple. There are sevreral methods (you will find in literature, tens if not hundreds with every method has it pros and cons)I will suggest to just log on to any software (SPSS, R which Dr. Rehan Qayyum likes very much) and it will tell you that data is following noraml distribution or not? If

yes, then proceed accordingly

and not then what? There are several alternatives: simplest is to trandform the data in such a way that eventually data follows normal distribution.

In recent times (say last 40 to 50 years) methods have been developed where apparently data is not following noramlity assumption. That's altogether different story and we statisticians call it nonparametric or distribution methods.

I hope above will clarify some of your worries.

Posted by: ** anwer_khur Posts: 30** :: 20-10-2011 :: | Reply to this Message

Dear sir,

Thanks a lot.

Would you please provide your email and cell?

May i contact you, in case of some problem?

Posted by: ** tariqasl Posts: 42** :: 14-11-2011 :: | Reply to this Message

Presently I am out of Pakistan, but you may contact me at anwer_khurshid@yahoo.com

Posted by: ** anwer_khur Posts: 30** :: 15-11-2011 :: | Reply to this Message