Dear Isadora,

On 05/10/16 17:30, Isadora Olive Ferreira wrote:
> Great! But I really thought that when it comes to anovas spm worked with
> only one error term for all factors and interactions? There are for some
> years now loads of debate on the web about it, mostly carried by the glm
> flex people.

You can also read these documents:
  http://www.fil.ion.ucl.ac.uk/~wpenny/publications/rik_anova.pdf

http://www.mrc-cbu.cam.ac.uk/wp-content/uploads/2015/03/Henson_EN_15_ANOVA.pdf

> And my model is  between factor 'group', 3 levels: controls, patients1
> and patients2; within factor 'condition' , 2 levels, cond1 and cond2.

For the main effect of 'group', compute for each subject the average
effect of 'condition' and enter them in a one-way anova (with three
levels) and use F-contrast [1 -1 0;0 1 -1].

For the main effect of 'condition',  compute for each subject the
difference between the two levels of 'condition' and enter them in a big
one-sample t-test at the second level and use t/F-contrast [1].

For the group x condition interaction, compute for each subject the
difference between the two levels of 'condition' and enter them in a
one-way anova (with three levels) and use F-contrast [1 -1 0;0 1 -1].

> And I profit to ask you also about flexible factorial models, which I
> would prefer to work in order to model subject variance. As I am working
> with highly diverse individuals, mainly because of their condition, can
> I assume that their variance is unequal?

I guess that if you indeed don't know, assuming unequal variance is the
safe option.

Best regards,
Guillaume.


> ------------------------------------------------------------------------
> *From:* Guillaume Flandin <[log in to unmask]>
> *Sent:* October 5, 2016 11:19 AM
> *To:* Isadora Olive Ferreira
> *Cc:* [log in to unmask]
> *Subject:* Re: [SPM] glm flex fast4 posthocs t
>  
> Dear Isadora,
> 
> to my knowledge, there is nothing special about partitioned error models
> and you can implement them in (any version of) SPM. In your mixed
> design, you don't precise which factor has three levels: is it the
> within or the between subject factor? It will help to describe how to
> create the relevant partitioned error models.
> 
> Best regards,
> Guillaume.
> 
> 
> On 05/10/16 17:04, Isadora Olive Ferreira wrote:
>> Bonjour Guillaume!
>> 
>> 
>> thank you for your quick reply! Yes, I'm trying for about a week to have
>> an answer in glm flex fast-related issues, without success.
>> 
>> 
>> I am working with a mixed model design 3x2. I would prefer to work with
>> glm flex tools mainly because it works with partitioned error terms. Any
>> ideas of how I could eventually do the same with spm are most welcome!!
>> 
>> 
>> Au plaisir de recevoir votre reponse,
>> 
>> Isadora
>> 
>> 
>> 
>> ------------------------------------------------------------------------
>> *From:* Guillaume Flandin <[log in to unmask]>
>> *Sent:* October 5, 2016 10:59 AM
>> *To:* Isadora Olive Ferreira
>> *Cc:* [log in to unmask]
>> *Subject:* Re: [SPM] glm flex fast4 posthocs t
>>  
>> Dear Isadora,
>> 
>> I was about to suggest you to post your question on the dedicated forum
>> but I realise you already did:
>>   https://groups.google.com/forum/#!topic/fmri_matlab_tools/8JcJ32go9cU
>> I'm not familiar with "glm flex fast4" so I hope someone will be able to
>> help you here but I would be curious to know what your data and model
>> are, that you cannot specify with core SPM functionalities.
>> 
>> Best regards,
>> Guillaume.
>> 
>> 
>> On 05/10/16 16:32, Isadora Olivé wrote:
>>> Good morning!
>>> 
>>> as I'm trying to run glm flex fast4 for the first time, I'm stumping into an error specifically concerning my post-hoc t-tests:
>>> 
>>>>> model
>>> Error using model (line 40)
>>> Dimensions of matrices being concatenated are not consistent.
>>> 
>>> Whenever I take out all the t-tests and make the code run only with f-tests, the problem disappears and the code runs smoothly.
>>> 
>>> Can anyone help me solving this problem?
>>> 
>>> I'm working on matlab 2016b and spm 12.
>>> 
>>> thank you for all help!
>>> Isadora
>>> 
>> 
>> -- 
>> 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

-- 
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG