Unequal group size, group representation in high scores?

#1
I'm trying to finish my thesis today, and I have run into the following two problems. I have tried searching on the net and in my old stats textbook and I can't seem to find the answers. I hope this is the right forum for these questions--if not, I apologize. I would be extremely grateful if someone could help me with these sometime today so that I can write the results section of my paper. :>

1) I'm running ANCOVAs in SPSS on a sample with three groups of unequal size (N=260, N=210, N=42). How does this affect my results--does SPSS correct for the size differences, or should I randomly select 42 members from the larger groups to equalize them? I really don't want to decrease my overall sample size.

2) My dependent variable is a test with scores from 0-30. I want to select only the high scorers (20-30) and then analyze them to see if one of the aforementioned three groups is disproportionately represented. I vaguely remember from stats class and from reading journal articles that there is a way to do this, but I don't remember how. I know how to select for the scores in SPSS--I'm just not sure what test to run on them. As you can see, my stats knowledge is very fuzzy. Any help would be much appreciated.
 

Mean Joe

TS Contributor
#2
2) My dependent variable is a test with scores from 0-30. I want to select only the high scorers (20-30) and then analyze them to see if one of the aforementioned three groups is disproportionately represented. I vaguely remember from stats class and from reading journal articles that there is a way to do this, but I don't remember how. I know how to select for the scores in SPSS--I'm just not sure what test to run on them. As you can see, my stats knowledge is very fuzzy. Any help would be much appreciated.
Sounds like you want to compare 3 proportions (the proportion of high scorers in group 1 vs proportion of high scorers in group 2 vs proportion of high scorers in group 3). I think you can do a chi-square test, if I'm understanding this correctly. Hopefully someone can confirm.

I'm sure SPSS has some way to do this test, but I'll attach an ugly-looking Excel worksheet that shows an example.

One note: if you do 3 tests (proportion 1 vs 2, 1 vs 3, and 2 vs 3), then instead of using the usual .05 "p-value hurdle", you should adjust it to .05/3 = .0167. This is done to account for the increased likelihood of making an error due to doing more tests.

Again, this is just my non-expert opinion. Hopefully someone can confirm/correct.

Good luck on your thesis, hope this helps.