From: Kosha Ruparel <[log in to unmask]>

Date: 8 October 2009 18:23:31 BST

To: [log in to unmask]

Subject: [FSL] 2x2x5 Anova Design Matrix

Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>

Dear Fslers

Hello. I am trying to analyze a dataset having the following variables of interest:

2 Groups: G1 and G2
2 Sessions (in time): S1 and S2
5 Conditions: C1, C2, C3, C4, C5

i have two approaches that i have questions for.

1)non hierarchical model:

  i constructed a matrix for this 2x2x5 anova (http://www.med.upenn.edu/bbl/downloads/az_anova_model1.jpg) where
  i have two subjects from G1 and two subjects from G2 each having
  total 10 first level copes (5 copes per session).
  F1: Main effect of Group
  F2: Main effect of Session
  F3: Main Effect of Condition
  F4: Group x Session Interaction
  F5: Session x Condition Interaction
  F6: Group x Condition Interaction
  F7: Group x Session x Condition Interaction

  This is as you see a rank defincient matrix since the linear combination of the last 4 columns (subject effects) gives the first column of the group difference. To make this matrix not rank deficient, could i remove the first column
  and leave the model otherwise the same. in other words, Is it possible to remove the Main Effect of Group but model
  all other 2 way interactions including group by sess and groupbycondition with the 3 way interaction? does not interaction require
  that the main effect be present in the model?


2)To reduce the dimensionality, i ran second level analyses to get S1-S2 and S1+S2 for each condition for both groups.
  is it valid to run 2 anovas testing groupxdrug and groupxemotion seperately rather than in a one big model as above?


thanks in advance
Kosha