Contrasts: Possible in Factorial Anovas or only On-way Anovas?

#1
Contrasts: Possible in Factorial Anovas or only One-way Anovas?

I would like to set up some contrasts in my factorial anova so that I can look for possible interactions between factor A and the contrasts of factor B. From my reading, it doesn't seem to be possible. All of the guides seem to refer to one-way anovas exclusively (e.g., http://www.statisticshell.com/contrasts.pdf). Any ideas? Thank you!
 
Last edited:

Dason

Ambassador to the humans
#2
I don't what you mean by "interactions between factor A and the contrasts of factor B" but contrasts are definitely possible in a factorial setting.
 
#3
I don't what you mean by "interactions between factor A and the contrasts of factor B" but contrasts are definitely possible in a factorial setting.

I hope this clarifies my question. Factor A has 3 levels, and Factor B has 2 levels. I would like to set up a contrast with Factor A (e.g., 0, -1, 1) and determine if this contrast interacts with Factor B (i.e., does the contrast look different at the two different levels of factor B).

How would I go about setting this up in a single factorial anova? Thanks!
 

Dason

Ambassador to the humans
#4
So Let's let A1B1 refer to the cell mean corresponding to the first level of A and the first level of B. You're interested in whether (A3B1 - A2B1) - (A3B2 - A2B2) = 0

Yes this is very possible to do and depending on what software you use isn't hard to code up.
 
#5
Dason, that is great news. I am using SPSS. Do you have any examples of the required syntax, or a reference to a guide or article that describes this process? Thank you!
 

Dason

Ambassador to the humans
#6
I know pretty much next to nothing about SPSS but there might be somebody else that can help you out.

As a last resort note that there is nothing different from running an ANOVA on some data treating it as a factorial design (with an interaction) than there is running it as a one-way ANOVA. You get a couple of preplanned contrasts spit out back at you (contrasts testing the main effects and one testing whether the interaction is significant) but any predictions you make will be the same. So if you know how to do contrasts for a one-way setting you could transfer that to the factorial setting by running it as a one-way anova and doing the contrasts for the relevant groups.