Dear Lisbeth,
Yes, the bug fix had to do with the partition of the design matrix in
condition effects and block effects. As a consequence, it changes the
definition of the 'effects of interest' F-contrast that SPM uses to
select voxels entering non-sphericity estimation. The ensuing correction
(ie whitening) is then different, which is what you observe by looking
at the design matrix.
Best regards,
Guillaume.
On 07/03/14 16:47, Van Ruijssevelt Lisbeth wrote:
> -----Original Message-----
> From: Van Ruijssevelt Lisbeth
> Sent: vrijdag 7 maart 2014 17:47
> To: 'Guillaume Flandin'
> Subject: RE: [SPM] results flexible factorial design after SPM8 update
>
> Dear Guillaume,
>
> Thank you for your quick reply. Was the bug which was fixed in r4290 related to the recognition of the number of conditions and blocks in the design matrix. I noticed that in the older version I always get 22 conditions when I specify factor 1=subjects with 16 levels and within-factor 2 with 6 levels and in the updated version I get 6 conditions (related to the 6 levels of my factor 2?) and 16 blocks (related to my number of subjects). What is the importance of the separation in blocks and conditions if your total number of parameters remains correct? Is this the cause of the difference in my results? My apologies for the basic question.
>
> Kind regards,
> Lisbeth
>
>
> Lisbeth VAN RUIJSSEVELT
> PhD student
> Department of Biomedical Sciences
> Bio-Imaging Lab - Prof. Dr. A. Van der Linden
> Universiteitsplein 1 - Uc 1.18
> BE-2610 Antwerpen
> BELGIUM
>
> Tel +32 (0)3 2652796
> Email [log in to unmask]
> Lab website: http://www.uantwerp.be/bio-imaging-lab
>
>
> -----Original Message-----
> From: Guillaume Flandin [mailto:[log in to unmask]]
> Sent: vrijdag 7 maart 2014 13:36
> To: Van Ruijssevelt Lisbeth
> Cc: [log in to unmask]
> Subject: Re: [SPM] results flexible factorial design after SPM8 update
>
> Dear Lisbeth,
>
> the differences you observe on the design matrices are due to the consequences of a bug fix that took place in r4290 and mentioned in the release notes of that version:
> http://www.fil.ion.ucl.ac.uk/spm/software/spm8/SPM8_Release_Notes_r4290.pdf
> This bug fix does not affect your stats at the first level though.
>
> Best regards,
> Guillaume.
>
>
> On 07/03/14 11:01, Van Ruijssevelt Lisbeth wrote:
>> Dear SPM experts,
>>
>>
>>
>> We started using SPM8 within a specific project, update r4010. To
>> increase comparability between the different studies we stopped
>> updating the program at that moment for all future studies within that
>> same project. To verify the results of my current experiment, I
>> processed my data in SPM8 (r4010) as well as in the last updated
>> version of the software (r5235), both starting from the raw data. I
>> expected some difference in the resulting stats which I indeed noticed
>> in first level statistics: same coordinates for global max but subtle
>> differences in T values. However at second level I get different
>> clusters resulting from my contrast of interest. I noticed that
>> although the batches are identical, the matrix of my factorial design
>> shows some differences in the parameters: # conditions, blocks etc. I
>> attached both design matrices to this email. Is this just due to a
>> different visualization of the matrix after the updates or could this
>> be the cause of the difference between both versions? If this is the
>> case, which one should I trust?
>>
>>
>>
>> SPM8 versions I use:
>>
>> 1. SPM8 (r4010) on PC running Windows 8.1 + Matlab R2011a (7.12.0)
>>
>> 2. SPM8 (r5235) on PC running Windows 8.1 + Matlab R2013a
>>
>>
>>
>> Many thanks in advance for your advice.
>>
>>
>>
>> Best regards,
>>
>> Lisbeth
>>
>>
>>
>>
>>
>>
>>
>> Description: Description: cid:image001.jpg@01C940EE.DF567870
>>
>> Lisbeth VAN RUIJSSEVELT
>>
>> PhD student
>>
>> Department of Biomedical Sciences
>>
>> Bio-Imaging Lab- Prof. Dr. A. Van der Linden
>>
>> Universiteitsplein 1 -*Uc 1.18 *
>>
>> BE-2610 Antwerpen
>>
>> BELGIUM
>>
>>
>>
>> Tel +32 (0)3 265*2796 *
>>
>> Email [log in to unmask]
>> <mailto:[log in to unmask]>
>>
>> Lab website: http://www.uantwerp.be/bio-imaging-lab
>>
>>
>>
>
> --
> Guillaume Flandin, PhD
> Wellcome Trust Centre for Neuroimaging
> University College London
> 12 Queen Square
> London WC1N 3BG
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|