-1

I am dealing with normaly testing of large samples. As stated here:

Is normality testing 'essentially useless'?

Normally testing is essentially useless, if the sample is too large. Even visual testing cannot give a clear statement about the distribution?

So, how to test a large sample if the distribution is normal distributed?

Community
  • 1
Le Max
  • 3,559
  • 9
  • 26
  • 26
  • What is the purpose of your normality testing? – Peter Flom Mar 17 '13 at 11:46
  • My purpose is, to test if the distribution is normaly distributed or not? – Le Max Mar 17 '13 at 12:01
  • 1
    Why do you want to know? What will you do if it is normal? What will you do if it is not normal? Is this a testing of residuals from a regression? – Peter Flom Mar 17 '13 at 12:23
  • Yes it is testing of residuals from a regression. However, I am interested how huge data sets are testen for normal distribution, and if its possible to get a meaningful result? – Le Max Mar 17 '13 at 12:59
  • See the points made [here](http://stats.stackexchange.com/questions/36212/what-tests-do-i-use-to-confirm-that-residuals-are-normally-distributed/36220#36220) – Glen_b Mar 18 '13 at 00:19

1 Answers1

1

Testing residuals is the classic case where formal normality testing goes astray, see the first answer in the question you linked to. This isn't unique to normality testing, it's a problem with p-values for any very large data set - that is, results can be statistically highly significant but of no practical import.

But I think this may be an exact duplicate question, unless you can give some reason why it is different than the question you linked to.

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
  • Ok so generally said, it depends on your goal how valid your answer about the normal distribution is? – Le Max Mar 17 '13 at 14:09
  • 3
    It depends on the consequences of the amount of non-normality for the technique that you wish to use. – Peter Flom Mar 17 '13 at 14:12