Dear all,
I have a problem (at least partially due to lack of knowledge I guess) concerning custom contrasts.
I'm in the following situation: I did an experiment with 3 groups (A, B, and C). A oneway ANOVA shows significant differences in means.
anova Y X
Using pwcompre I can see that the differences between all three groups are significant.
Now my hypothesis requires me to show that A is different from B and C.
At first, one could assume this is easy to show by
contrast h.X or -using custom contrasts- contrast {X 1 -.5 -.5} This would say the mean of A is different from the average mean of B and C.
But: The problem is that (consistent with theory) group B shows a somewhat negative deviation and group C somewhat positive. So this sums up and I can't see a significant effect.
So I am thinking about trying something like contrast {X 1 -2 3}. What would that tell me as it is significant? Or is there another possibility to 'put different weights to the group means when comparing them'? Maybe you also know a paper that uses these kinds of 'asymmetric contrasts', i.e., not simply comparing one mean with the average of various other means.
I realize this sounds a bit fuzzy while writig it down. Still, I hope someone can help.
Thanks in advance for your helpful suggestions,
Kai
I have a problem (at least partially due to lack of knowledge I guess) concerning custom contrasts.
I'm in the following situation: I did an experiment with 3 groups (A, B, and C). A oneway ANOVA shows significant differences in means.
anova Y X
Using pwcompre I can see that the differences between all three groups are significant.
Now my hypothesis requires me to show that A is different from B and C.
At first, one could assume this is easy to show by
contrast h.X or -using custom contrasts- contrast {X 1 -.5 -.5} This would say the mean of A is different from the average mean of B and C.
But: The problem is that (consistent with theory) group B shows a somewhat negative deviation and group C somewhat positive. So this sums up and I can't see a significant effect.
So I am thinking about trying something like contrast {X 1 -2 3}. What would that tell me as it is significant? Or is there another possibility to 'put different weights to the group means when comparing them'? Maybe you also know a paper that uses these kinds of 'asymmetric contrasts', i.e., not simply comparing one mean with the average of various other means.
I realize this sounds a bit fuzzy while writig it down. Still, I hope someone can help.
Thanks in advance for your helpful suggestions,
Kai
