Created:
| Last Modified:
I teach stats in my graduate psychology program, and my approach is to use orthogonal contrasts to analyze ANOVA designs. (See this book among others.)
The "LSMeans Contrasts" option in the right-most section of the output is a fantastic way to analyze the data this way without actually having to built contrast codes in JMP columns. I start with the main effect contrasts and then figure out how to construct the interactions by assuming I'm multiplying the main effects, thus getting the signs right.
This works almost perfectly. All the inferential stats are just right. The one thing that is off–for interactions only–is the actual estimates of the interaction effects. I'm attaching a pdf of one of my slides that shows, on the left, the direct way with contrasts codes analyzed via multiple regression. On the right, it shows the output using "LSMeans Contrasts." The estimated effect on the right is half what it should be.
In another example I just did (with a 2x3 design), this same error happens.
In the unlikely event that anyone is following this with understanding and concern, any ideas?
