Griselda J Garrido wrote:
> Dear Will and SPMers,
>
>
>>I would recommend just analysing each subject separately and
>>reporting the results as a series of case studies (unless you
>>can get data on another 8 subjects).
>>
>>So 'at the first level' you plot maps of regions where
>>you can reject the null hypothesis: In subject X at location
>>y there is no difference in activation between the two conditions.
>>
>
> I have a similar study with SPECT (10 subjects, 2 conditions each).
> I tried to analyze each subject separately using either
> Basic models - Paired t-test
> PET/SPECT models - Population main effect (paired t test)
>
> But the sole error message unestimable condition effects appears.
> Should I use another approach?
> Could you comment on the degrees of freedom of this kind of analysis?
> I'm not statistician but could it be n-1 equal 0?
>
Assuming you only have one scan per condition then with 2 conditions
you could get 1 DF per subject - though SPM uses a mean and then
two condition variables per subject and so this is why
you get an error (ie. not enough DF). The standard approach is therefore
to combine data from 10 subjects to do a fixed effects analysis
(ie. analyse data from all subjects together).
In SPM, go to Basic models, paired t-test and then select
10 pairs and then enter the scans for each condition and subject in
order [Sub 1, Condition 1], [Sub 1, Condition 2] then [Sub 2, Condition 1]
etc. as prompted.
Best wishes,
Will.
> Best regards,
>
> Griselda
>
>
> -----------------------------------------
> Griselda J Garrido, PhD
> Division of Informatics - Heart Institute
> University of Sao Paulo - Brazil
> www.fisicamedica.com.br/griselda
> [log in to unmask]
> -----------------------------------------
>
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|