See below.


On Thu, May 15, 2014 at 6:34 AM, Ce Mo <[log in to unmask]> wrote:
Dear All,

 

      I got really weird results when I tried to analyze my data using one-way within-subject ANOVA.

 

As can be seen from the design matrix (see attached), my design involves a within-subject factor (4 levels, denoted as A1-A4 respectively) with 15 subjects.

 

My objective is to locate the brain regions that has a significant main effect. To do so, I first calculated the con images pertaining to the 4 conditions for each subject and feed these images to the within-subject ANOVA model (Independence = No; Variance = Unequal). I then setup the following F-contrast based on Dr.Mclaren's many posts on contrast definition and Glascher&Gitelman's tutorial. (The null hypothesis is A1=A2=A3=A4)

 

[ 1 -1 0 0 ones(1,15)

  1 0 -1 0 ones(1,15)

  1 0 0 -1 ones(1,15)

  0 1 -1 0 ones(1,15)

  0 1 0 -1 ones(1,15)

  0 0 1 -1 ones(1,15) ]


I am not sure what contrast this is testing. As you are comparing levels, the subject columns should have a weight of 0. 

 

However, the entire cerebral volume was found activated with spurious F-values as much as 2000! This is obviously wrong. Moreover, the df. of this F-contrast is 4, which is also incorrect. (See Attached Figure 1)

 

I then defined another contrast which excluded the weights for subject factor columns. The df. seems correct (df.=3) and the results were not as good as I expected, yet I do not get whole brain activations. (See Attached Figure 2)

 

[ 1 -1 0 0 

  1 0 -1 0 

  1 0 0 -1 

  0 1 -1 0

  0 1 0 -1 

  0 0 1 -1 ]


The correct contrast would be:
[1 -1 0 0; 0 1 -1 0; 0 0 1 -1]

I believe that you will get the same results as your contrast.
 

 

However, this contrast seems inconsistent with the preivous posts.  

 

Could anyone please point out where my problem lies?


I'm not sure why you say there is a problem. It is possible that the conditions are not different from each other in this small sample. I would set the variance of the within-subject factor as Equal. This may help.

 

 

Many thanks and Best regards,

                    Ce