If I want to compare the means of two groups but the group sizes are uneven (230 and 25) what is the best test to use?

I am basically trying to see if one group performs significantly better than the other.

Thanks!

Hi,

just had the same problem.
Kramer developed modifications for post-hoc tests for unequal n. For example for Tukey's HSD > Tukey-Kramer test. I don't know if this modification fits for all tests, or special modifications are needed for each test.
P.S. In SPSS this modification is automatically used in unequal sample sizes.

hobglobin

Hi,

just had the same problem.
Kramer developed modifications for post-hoc tests for unequal n. For example for Tukey's HSD > Tukey-Kramer test. I don't know if this modification fits for all tests, or special modifications are needed for each test.
P.S. In SPSS this modification is automatically used in unequal sample sizes.
hobglobin

Just read further about this:
Availble tests are: GT2; T' method; Games-Howell; T3; Tukey-Kramer (list is not exhaustive). The paper recommends the Tukey-Kramer test. Most of them are implemented in SPSS, SAS, R. Or automatically used in beginner software ;-).

best wishes
hobglobin

Thanks, Hobgoblin.

So if I run an ANOVA in SPSS, does that mean I am covered because the software accounts for the different group sizes for me? Or should I be doing something in addition to the ANOVA?

Would it help if I did a Brown-Forsythe test to ensure equal variances of the two groups?

Thanks again.

Hi,
before ANOVA it should (or better must) always be controlled that the data fit the ANOVA assumptions (especially variance homogeneity, normally distributed data). The post-hoc tests also rely on this, each test has a little bit different needs. Games-Howell and T3 (and others) for example are relatively robust against unequal variances (if they are not to strong).
To check the data I use the Levene test and box-plots.
In SPSS you can choose several tests, in my version e.g. GT2, Games-Howell are available, and if Tukey is chosen, automatically Tukey-Kramer is used for unequal sample sizes (at least it is written is the help).
Hope this helps

Peter