Testing Paired t-test Normality Assumptions

[Nebulus]

I have a quick question about testing the assumption for paired t-tests that " The distribution of the differences in the dependent variable between the two related groups should be approximately normally distributed".

Say I have two groups one and two. The way I am supposed to do this is to create a new variable "diff=one-two"

then do..

Proc Univariate data=x normal;

qqplot  / Normal (mu=est sigma=est);

var diff;

run;

Is this correct?

Or do i use this instead to test normality?

var one two;

I interpreted it as I am supposed to test the normality of the differences between groups not each group.

Thanks!

[ballardw]

Actually I'd probably use Proc Ttest. With ODS Graphics on the defaults will create some plots related to the distributions of differences. If I didn't see any problems then we're done.

Use the VAR Diff version. If you use two variables you will get separate results for each variable in Univariate.

[Ksharp]

I think both are equality. If both one and two are normal distribution, then dif=one-two  must be normal distribution. 
because dif is linear combination of normal distribution.