Protecting the overall rate of Type I errors for pairwise comparisons with an omnibus test statistic.