anova contrasts problem

kk49

New Member
#1
Hi,
I have data from an experiment with 3 conditions (and a few covariates), and I am running ANOVA (or ancova).

The conventional approach is to run anova on "condition" (all three conditions) and then do contrasts to see which treament differs from the mean. But I am wondering if I could create two dummies for conditions 2 and 3 and see if those are significantly different from the baseline.

I am asking, in part, because when I use the dummies, I get results that seem inconsistent with the results i get from anova contrasts. Specifically, the dummies show that condition 2 is significantly diff from the baseline condition (while condition 3 is not), but anova contrast shows that condition 2 is no differnt from the baseline. I am using STATA's anovacontrast.

Is there anything I am doing wrong?
Thank you very much for your help
 
#2
Contrasts should be laid out a priori and aren't usually carried out as an after the fact analysis; i think you are referring to post hoc testing, which makes all possible pair-wise comparisons between conditions.

If you have a few simple contrasts of interest in mind, I'd say run a t-test on the groups, and replace the denominator with the error term from the overall ANOVA. You'll have to calculate this manually.

In terms of complex contrasts (i.e, conditions 1 & 2 VS 3 or 3 VS 1 & 2 etc), SPSS can handle them by clicking the contrast option box in the 1 way anova procedure window.

If you have more than 2 contrasts, be sure to correct for an inflated type 1 error rate.