topic Re: T-TEST in SAS Procedures

T-TEST

robertrao — Wed, 04 Sep 2013 20:27:49 GMT

Hi,

In interpreting T-test results which of the following should we pick for the p-value?????

Method variances DF tvalue pr>|t|

Pooled equal 42 3.68 0.0007

Satterthwaite unequal 9.094 2.53 0.0320

Equality of Variences

Method NumDF DenDF Fvalue Pr>f

Folded F 8 34 3.85 0.0051

Thanks

Re: T-TEST

ballardw — Wed, 04 Sep 2013 22:17:52 GMT

The second part of the display on equality of variances tests if the two goups of data have equal variance. If you have large F then the likelihood of equal variances is very low (Pr>f) and reject equality. Which would mean to use the row in the first section with unequal variance and the Satterwaite adjustments to the basic T-test.

If the F is low then the likelihood is the variance are equal, or close enough, and use the equal variance.

Re: T-TEST

robertrao — Thu, 05 Sep 2013 13:29:57 GMT

Hi,

Thanks for the reply.

But how would we know if the F is low or high??? What do we need to compare the F to check to see if F is high or low??

Could you explain???

thanks

Re: T-TEST

SteveDenham — Thu, 05 Sep 2013 13:46:58 GMT

There is a probability associated with that F test. In the output, Pr>F is 0.0051. That is pretty strong evidence that the differences in the variances are not due solely to chance, and that you would be advised to use the Satterthwaite approximation value.

Note well, though, that any test of variance equality or homogeneity is much more sensitive to the assumption of normality than is the actual t-test itself. In this case, since the Satterthwaite degrees of freedom is so much different than the pooled method, I would be inclined to use the Satterthwaite t-test no matter what the result of the variance homogeneity was. I assume that the sample sizes differed greatly by group, yes?

Steve Denham