 |
 |
I've always been under the impression that: if you have a repeated measures design (e.g. 2 measures per individual) and don't model a subject term, then you violate the assumption that the observations (each row) are independent as the program won't know they are replicates. The only way to specify observations as replicates, that I know about, is to include a subject term in the model. Is there another way to deal with non-independence of observations? According to a recent document (conweights.pdf by Darren and Jan Glascher): When you specify subject, the contrast needs to be non-zero for those columns when contrasting group. For example, 2 groups with 10 subjects each where each subject has two measurements would be: 1 -1 ones(1,10)/10 -ones(1,10) 0 0 Darren, Is that what you meant when you said to "model differences at the subject level"? 2009/4/3 Laura Menenti <[log in to unmask]> > Dear Radek (and Thomas and Natasha), > > You asked whether you should specify the subject factor when testing for > between-group main effects in a flexible factorial design. As you correctly > noted, the between-subject variance is crucial for assessing the > significance of between-group effects. Specifying the subject factor > removes > just this variance from the error term. > > Therefore, you should *not* specify the subject factor when testing for > between-group effects. As we understand from your email, Radek, you were > testing for the main effect of group. > > Of course, specifying the subject factor is correct when testing for > interactions between within-subject and a between-subject factors. We think > this is the point Darren makes in his reply: > > "Assuming you have two groups, for > example, the main effect should be a difference, which should also > model differences at the subject level between groups and with respect > to any interactions." > > Stephen states the same: > > " So, while the "main effect" of subject is regressed out, the interaction > of subject and treatment still remains and is used to compute significance > of the condition effect." > > So, specify the subject factor when testing for interactions with group, > but > not when testing for main effect of group. Therefore, "yes and no" is > actually a correct answer to your question. > > As an aside, we note that this is a common error, as exemplified also by > the > email "2nd Level ANOVA 2x3x3xsubj" by Thomas. Here the parameterplot of his > F-contrast of the group effect nicely illustrates that most of the relevant > variance between groups has been (erroneously) explained away by adding the > subject factor (though we are not sure this actually causes the error > message when plotting). > > Natasha (Between group tests for task main effects & group*task > interactions > - one-way ANOVA?), we hope this answers (part of) your questions too. > > HTH, > Laura Menenti and Lennart Verhagen > > > > -----Original Message----- > From: Darren Gitelman [mailto:[log in to unmask]] > Sent: Wednesday, 01 April, 2009 16:50 > Subject: Re: flexible factorial - main effect of subject factor > > Radek > > If you want to do a repeated measures mixed effects design you should > enter the subject effects. Since you didn't specify the contrasts you > used I wonder if you may have setup an average effect of group rather > than the main effect of group. Assuming you have two groups, for > example, the main effect should be a difference, which should also > model differences at the subject level between groups and with respect > to any interactions. > > ----- > Darren Gitelman > > > > 2009/4/1 Radek Mareček <[log in to unmask]>: > > Dear SPMers, > > > > we have an experiment with 2 groups of subjects (controls/patients) and > > 3 conditions for each subject. We set the 2nd level analysis using > > flexible factorial design option in SPM5. > > > > The main effect of subject, group and condition and interaction of > > group/condition were included in design matrix. > > > > The contrast weights for main effect of group was set according to > > technical note of Jan Glascher and Darren Gitelman. > > > > The problem is when the main effect of subject is included in design > > matrix. This results in widespread activation on 0.05 FWE level with > > very high t-values. When the main effect of subject factor is not > > included in the design matrix the activation is less extensive (and > > probably more credible). > > This is in accord with the Darren Gitelman's note of improved > > sensitivity of a model where the main effect of subject is included in > > the design matrix. > > > > The question: > > When the subject effect is included the inter-subject variability > > doesn't end up in residuals (or at least not all) which induce higher > > t-values. But isn't the inter-subject variability crucial for assesing > > of significance of tested effect? > > Am I completely wrong? Which design is more resonable then? > > > > Thank you for any comments or notes on this issue. > > > > > > Radek Marecek > > Dep. of Neurology > > St. Anne's University Hospital > > Masaryk University > > Brno, Czech Republic > > > -- Best Regards, Donald McLaren ===================== D.G. McLaren University of Wisconsin - Madison Neuroscience Training Program Office: (608) 265-9672 Lab: (608) 256-1901 ext 12914 ===================== This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is intended only for the use of the individual or entity named above. If the reader of the e-mail is not the intended recipient or the employee or agent responsible for delivering it to the intended recipient, you are hereby notified that you are in possession of confidential and privileged information. Any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited and may be unlawful. If you have received this e-mail unintentionally, please immediately notify the sender via telephone at (608) 265-9672 or email.