Hi Josh, there is not a way that I know of to do the paired analysis if some are missing a scan. A "real" statistician might know of a way theoretically, and I might be able to implement that theory, but I don't know myself. So does that mean you want to do an unpaired test? doug
Roffman, Joshua L.,M.D. wrote:
Hi Doug, Thanks for weighing in. For this study, while most subjects have 2 sets of good scan data (pre and post treatment), a number have only one set (pre or post). We're hoping to use all of the data given that it's overall a pretty small study. Is it possible to run the paired analysis if we are missing pre or post data for some subjects? Best, Josh
-----Original Message----- From: Douglas N Greve [mailto:greve@nmr.mgh.harvard.edu] Sent: Thu 9/15/2011 4:54 PM To: Adam Nitenson Cc: freesurfer@nmr.mgh.harvard.edu; Roffman, Joshua L.,M.D. Subject: Re: [Freesurfer] FSGD Contrast Setup Question
Hi Adam, you have a "repeated measures" design. In your case, this is best handled using a paired analysis. You can search for "paired" on the wiki to get instructions. The easiest way to do this is to run mris_preproc with the --paired-diff option, then follow the instructions for the "standard" one or two group analysis (search for "fsgd").
doug
Adam Nitenson wrote:
Hi Freesurfers,
I am trying to set up both a 2 Group 1 Covariate and a 4 Group 1Covariate analysis but just want to make sure I'm using the correct contrast option.
For the first analyis, the two groups are Drug Pre and Drug Post,with folate-serum level as a covaraite. We want to keep the covariate under consideration in our analyses, but we really want to look at the difference between the Drug Pre and Drug Post activation (so folate-serum level is more of a nuisance variable). It seems that the first option under the 2G1V Page "Null Hypothesis: is there a difference between the group intercepts? Is there a difference between groups regressing out the effect of age? 1 -1 0 0" reflects this idea best.
For the second analysis, the 4 Groups are Drug Pre, Drug Post,Placebo Pre, and Placebo Post, with the same folate-serum level covariate. We are intersted in the double subtraction of (Drug Post
- Drug Pre) MINUS (Placebo Post - Placebo Pre), but again wish to
covary by folate level. For this it seems like contast 7 on the 4G1V page "is there an interaction between gender and handedness regressing out the effects of age? 0.5 -0.5 -0.5 0.5 0 0 0 0" is best.
Is this the best way to go about analyzing these two questions?
Thanks,
Adam Nitenson, B.S. Brain Genomics Lab Massachusetts General Hospital nitenson@nmr.mgh.harvard.edu Phone: 617-643-3215 _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Hi Josh,
With sophisticated mixed modeling procedures that use MLE estimation (e.g., SAS's PROC MIXED) it would be possible to use all your data while still modeling the necessary covariances. If you extracted ROI-based measures that would be a possible approach you could take. However, I don't know of any voxel-level analysis package that is built to do the equivalent sort of thing.
cheers, -MH
On Thu, 2011-09-15 at 18:18 -0400, Douglas N Greve wrote:
Hi Josh, there is not a way that I know of to do the paired analysis if some are missing a scan. A "real" statistician might know of a way theoretically, and I might be able to implement that theory, but I don't know myself. So does that mean you want to do an unpaired test? doug
Roffman, Joshua L.,M.D. wrote:
Hi Doug, Thanks for weighing in. For this study, while most subjects have 2 sets of good scan data (pre and post treatment), a number have only one set (pre or post). We're hoping to use all of the data given that it's overall a pretty small study. Is it possible to run the paired analysis if we are missing pre or post data for some subjects? Best, Josh
-----Original Message----- From: Douglas N Greve [mailto:greve@nmr.mgh.harvard.edu] Sent: Thu 9/15/2011 4:54 PM To: Adam Nitenson Cc: freesurfer@nmr.mgh.harvard.edu; Roffman, Joshua L.,M.D. Subject: Re: [Freesurfer] FSGD Contrast Setup Question
Hi Adam, you have a "repeated measures" design. In your case, this is best handled using a paired analysis. You can search for "paired" on the wiki to get instructions. The easiest way to do this is to run mris_preproc with the --paired-diff option, then follow the instructions for the "standard" one or two group analysis (search for "fsgd").
doug
Adam Nitenson wrote:
Hi Freesurfers,
I am trying to set up both a 2 Group 1 Covariate and a 4 Group 1Covariate analysis but just want to make sure I'm using the correct contrast option.
For the first analyis, the two groups are Drug Pre and Drug Post,with folate-serum level as a covaraite. We want to keep the covariate under consideration in our analyses, but we really want to look at the difference between the Drug Pre and Drug Post activation (so folate-serum level is more of a nuisance variable). It seems that the first option under the 2G1V Page "Null Hypothesis: is there a difference between the group intercepts? Is there a difference between groups regressing out the effect of age? 1 -1 0 0" reflects this idea best.
For the second analysis, the 4 Groups are Drug Pre, Drug Post,Placebo Pre, and Placebo Post, with the same folate-serum level covariate. We are intersted in the double subtraction of (Drug Post
- Drug Pre) MINUS (Placebo Post - Placebo Pre), but again wish to
covary by folate level. For this it seems like contast 7 on the 4G1V page "is there an interaction between gender and handedness regressing out the effects of age? 0.5 -0.5 -0.5 0.5 0 0 0 0" is best.
Is this the best way to go about analyzing these two questions?
Thanks,
Adam Nitenson, B.S. Brain Genomics Lab Massachusetts General Hospital nitenson@nmr.mgh.harvard.edu Phone: 617-643-3215 _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Michael Harms