Why the unequal variance (Welch) t test can report a different P value in Prism 6 than from prior versions.

The P value is computed from the value of the t ratio and the number of degrees of freedom (df). The unequal-variance t test computes the number of df using a complicated equation that can end up with a fractional df. Prism 5 rounds this value down to the nearest integer. Prism 6 can compute the P value using a df that is fractional. Since Prism 6 doesn't need to round the df, it gives a more accurate result. Since df will always be a bit higher (or have the same value) in Prism 6 (i.e. 7.6 rather than 7), the P will be a bit lower (or the same).

This issue only applies with the unequal-variance t test. The usual t test, which assumes equal variances, always uses a df value that is an integer so all versions of Prism report the same P value.