I keep reading that in order for a t-test to be valid the data need to be normal. I have distributions which are clearly skewed to the right (long right tails) so they are not normal. However I have a sample of of n>100 in both bases so the sampling distributions are essentially normal.
My question therefore is: does the t-test requirement for normality refer to the original distributions for the raw data, or for the sampling distributions for the sample means?
Jun 30th 2011, 02:16 PM
Re: t-test and normality
The t-test itself, tests the sample mean, which will be normally distributed.
If your data itself is skewed but n is large enough the t-test is considered a robust test.
I would recommend using a t-test and then using a different test, maybe a non-parametric test like the Kolmogorov-Smirnov test.
Are the data sets considered different in both tests?