•Multiplicity adjusted (“exact”) P values can be reported with Bonferroni, Tukey, Dunnett, Dunn, or Holm-Sidak multiple comparisons testing following one- or two-way ANOVA.  An adjusted P value is an “exact P value” reported for each comparison, but its value depends on the number of comparisons.

•Fisher’s unprotected Least Significant Difference (LSD) comparisons are a second approach to report "exact P values" following ANOVA. This method does not correct for multiple comparisons.

•New multiple comparisons test. Holm-Sidak has more power than the Tukey or Dunnett methods, but cannot provide confidence intervals.  

•Dunnett's test can now can be computed with 256 groups. Prism 5 was limited to 20 groups.

•For all multiple comparisons that create confidence intervals, check an option and Prism will create a new graph with those multiple comparisons confidence intervals.

•Multiple comparisons after one-way repeated measures ANOVA, without assuming sphericity. Prism 6 can compute confidence intervals from repeated measures one-way ANOVA using two different methods. The new method does not assume sphericity. Instead of using the pooled variation from all the groups (which depends on the sphericity assumption), the multiple comparisons are computed using only the data in the two treatments being compared. The calculations still account for the total number of comparisons, but don’t pool the variation in all groups.

•When performing multiple comparisons testing, it can make sense to use a  overall (familywise) significance threshold higher than 0.05. Prism 6 lets you set it to 0.1.  

• Main and simple effects for multiple comparisons after two-way ANOVA.