Hi Marris,
A comment by Tom Nichols in the link below might help you understand
about no centering on dummy variables.
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;53e0444e.0804
<quote>
First, you have entered dummy variables manually, but then selected
'centering'. Dummy variables should not generally be centered, as
their 0's are special and not just a relative value. Further, if all
of your variables are centered and you drop the intercept, there's no
way to model the mean, and you'll get a horrible model fit. Hence, if
you're going to use these hand-entered dummy variables w/out an
intercept, turn off centering.
</quote>
Hope this helps,
Kiyotaka
2015-05-05 20:06 GMT+09:00 Marris Atwood <[log in to unmask]>:
> Hi Kiyotaka,
>
> Thank you so much for looking into my model so carefully. I sincerely
> appreciate that!
>
> I changed "Implicit Mask" to "Yes", but the error remains. So I guess I will
> need to look into my data, as suggested by you. In fact, I have run the
> analysis once with no mask at all. It runs OK, but I am not sure what are
> the potential pitfalls for not using the mask.
>
> As for the centering issue, indeed, I overlooked the suggestions in the
> tutorial. So I should NOT center age, gender, and group (indicating
> Alzheimer's, MCI, or normal). May I check why? To me, it is definitely not
> necessary, but it doesn' hurt, either.
>
> Thank you so much again for your kindness and advice!
>
> Warmest gratitude,
> Marris
>
> On Tue, May 5, 2015 at 1:11 PM, Kiyotaka Nemoto <[log in to unmask]>
> wrote:
>>
>> Hi Marris,
>>
>> I saw your settings and noticed that Implicit Mask is set to 'No'.
>> I think this should be set to 'Yes.'
>>
>> If you still see errors, something must be wrong in one or more files
>> in your dataset.
>>
>> I will go through each one of them.
>>
>> Guillaume and Donald gave a good advice on checking data a few days ago.
>>
>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;501709e8.1504
>>
>> <quote>
>> >> load SPM.mat
>> >> spm_check_registration(SPM.xY.P)
>>
>> and scroll through your entire dataset. Plotting the globals can also
>> give you a hint about suspicious scans:
>>
>> >> figure, plot(SPM.xGX.rg)
>> </quote>
>>
>> note that spm_check_registration can load up to 24 files.
>> Since your dataset is quite large, I think
>>
>> >> load SPM.mat
>> >> figure, plot(SPM.xGX.rg)
>>
>> will give you some hint.
>>
>> If you prefer GUI, I recommend using "Check sample homogeneity using
>> covariance" function implemented in VBM8 toolbox. It's easy to use and
>> quite useful to detect the poor quality data.
>>
>>
>>
>> Another concern: I see you set centering on Gender. Basically you
>> don't want to center dummy variables such as gender. It's also
>> described in VBM tutorial. I would set "No centering" on gender. (and
>> probably group if the order of group doesn't mean anything.)
>>
>> Best regards,
>>
>> Kiyotaka
>>
>>
>>
>> 2015-05-05 13:30 GMT+09:00 Marris Atwood <[log in to unmask]>:
>> > Hi Kiyotaka and Indrajeet,
>> >
>> > Sincere thanks to both of you! Now I understand what I edit should be a
>> > setting file instead of SPM.mat.
>> >
>> > Your pointers on global normalization are very enlightening. Yet, the
>> > strange thing is, I am not using "proportional" for global normalization.
>> > Instead, I am treating it as a covariate by using "ANCOVA".
>> >
>> > I have attached my saved settings here. Any guess is very much welcomed
>> > and appreciated!
>> >
>> > Thanks a lot,
>> > MA
>> >
>> >
>>
>>
>>
>> --
>> Kiyotaka Nemoto, M.D., Ph.D.
>> Assistant Professor
>> Department of Psychiatry
>> Division of Clinical Medicine, Faculty of Medicine
>> University of Tsukuba
>> 1-1-1Tennodai Tsukuba, Ibaraki 305-8575, Japan
>> E-mail: [log in to unmask]
|