Before using Excel's t-test, how to know if the data is nomal distribution?

  • Thread starter Thread starter tony lincoln
  • Start date Start date
T

tony lincoln

Dear friends,

I have to test some groups of data using Excel's t-test. As I know,
the prerequisite of t-test is the distribution of the data sets should
be normal distribution. How can I evaluate my data set, to be sure
that they are nomal distribution?

Many thanks.
tony
 
Current views are that one really can't determine if a sample is normal or not from a small sample (i.e. less than 30). One can use the summary statistics output in Excel and look at the skewness value. If the skewness is substantially different from zero then the t test is probably in error. The t-test is fairly insensitive to large kurtosis values, but very sensitive to skewness.
 
There are many tests to check for normality, some of which are
specifically designed for the Normal.

One test uses the Jarque-Bera statistic:
(#observations/6) * (skew^2 + 0.25*(kurtosis-3)^2). It is chisq
distrivuted with 2 degrees of freedom. At 95% confidence level, that is
about 5.99. So if your JB stat > 5.99, reject the null hypothesis that
the data are Normal.

There are numerous other tests; Google for them, paying particular
attention to those that are designed just for the Normal, unlike more
general tests such as K-S. You can refine your search to tests such as
Shapiro-Wilks, Shapiro-Francia, D'Agostino's D, Spiegelhalter's T, etc.
No single test is uniformly most powerful; the best choice depends on
what you have (and what you know)

HTH
Dave Braden
 
Back
Top