I am looking to determine whether patient's total number of prescriptions (RxTotal) differs by membership within 3 possible groups (ie run an ANOVA), and if so, to determine which groups differ (ie run multiple comparisons). My data is non-normal (Shapiro Wilk p<.0001) and the groups have unequal variances (Levene p<.0001). My group Ns are 113, 202, and 90. I understand that ANOVA is fairly robust against violations of normality and HOV as long as each group has equal Ns. I do not know if the Ns of my 3 groups could possibly be considered "equal" so I have chosen to go with using Welch's F statistic (p<.0001), although I am open to feedback to take other routes. I have read that the Games-Howell multiple comparisons test works well with HOV violations, and that it is the Welch's equivalent of the Tukey post hoc test. I have searched at length and cannot figure out what the syntax would be for running Games-Howell multiple comparisons tests in SAS. Is this perhaps because it requires a macro, or is there an option built into a standard SAS procedure that I have not found?

This is my syntax for running the Welch ANOVA, and the Tukey post hoc tests (which I am inclined not to use):

proc glm data=cutoffs1;
class group;
model rxtotal = group;
means group / welch tukey;

Thank you!

  • $\begingroup$ Unless the counts of total prescriptions are quite high, I'd probably use a count regression, either Poisson or negative binomial. $\endgroup$
    – Peter Flom
    Sep 6, 2023 at 18:54
  • $\begingroup$ Thank you so much. I would like to run multiple comparisons between all 3 groups on the count outcome, not just two groups compared to a reference group. Would there be an accepted way of using a Poisson regression approach that would allow for that? Thank you $\endgroup$
    – L.S.
    Sep 13, 2023 at 15:22
  • $\begingroup$ The same methods as with PROC GLM should work, I believe. $\endgroup$
    – Peter Flom
    Sep 13, 2023 at 16:03


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.