Dunnett's Test and Degrees of Freedom

#1
Can someone clue me in as to how degrees of freedom are calculated for Dunnett's test for multiple comparisons? We have a data set with 5 groups (1 control and 4 treated). The n-numbers are 16, 16, 16, 15, and 16, respectively. The stat package we're using (TOXSTAT) gives output that shows the degrees of freedom as 24,4. The 4, I get (K-1, where K=5). But I can't for the life of me figure out the 24. I thought that DF would be the summation of n-1, but obviously that's not even close. What am I missing? Thanks.
 

JohnM

TS Contributor
#2
I don't know anything about TOXSTAT, but you're correct - the dof for Dunnet's test should be 74 in this case, which is N - k (total sample size minus the number of groups).
 
#3
Thanks, John. I've since discovered that there appear to be 2 versions of the program floating around. A recent one, that gives the correct degrees of freedom, and a previous version, that does not. Guess which one was used. Fortunately, it appears to be looking up the correct critical value for the assignment of significance, so it's kind of a typo thing with printing out the DF. Ya can't depend on anything these days!
 
#5
Hobgoblin:

I think that in this case, (and John, correct me if I'm wrong), the "k-1" refers to what is called "between groups" degrees of freedom, while the "n-k" refers to "within groups" degrees of freedom. The within groups d.f. can be calculated by summing the "n-1" for all the groups, where "n" = the number of subjects in the group. It can also be done, as John stated it, by "n-k", where "n"= the TOTAL number of subjects in all the groups, and "k" = the number of groups. More than that, I don't know.