unfortunately there is no easy way .. the point you want to make, if I
understand, is that one type of fit (model) is better than the other ...
there are many techniques regarding model selection (actually entire
books dedicated to this topic)
the easiest/simplest would be to use the Mean Square Error (MSE) which
is stored in the ResMS image. For each model and each subject take the
sqrt(ResMS) in your ROI (ie extract data from the ResMS and compute the
square root) and then you can test if there is a significant difference
between the root mean square (RMSE) of model1 vs model 2. For RMSE, the
smaller, the better.
Alternatively you can compute the maximum likelihood estimation (MLE).
The log likelihood = apha*MSE^2+beta with alpha = -n/(2*var) and beta =
-n*log(sqrt(2pi)*std). In that case, the larger the MLE the better.To
compute the log likelihood you will need n the nb of scans and also the
variance (stored in SPM.VResMS) .. so it come down to how good you are
with Matlab to do all this. MLE thends to be better because it
accommodates different variances in the data (y)
Finally, I also want to point out that these two methods are not ideal
because your 2 models have different complexity (2 vs 4 parameters) and
the more complex, the more likely you will fit the data. Something along
the lines of Akaike Information Criterion
(http://en.wikipedia.org/wiki/Akaike_information_criterion) or Bayesian
Information Criterion
(http://en.wikipedia.org/wiki/Bayesian_information_criterion) are more
likely to be accurate but that will take you some hardcore coding to get
there ..
Cyril
> Dear Cyril,
>
> I have two models. Could I get the goodness of fit measures using SPM? Or do I have to do this with a statistical software. What are the values that I should in order to do this? Is the contrast estimates calculated by spm enough? Sorry about thesr basic questions but I am always confused about the Y axis when plotting the contrast estimate for parametric modulations using SPM . Values range from 0.04 to 0.06.
>
> Regards,
>
> AS
>
> On 3 Jul 2013, at 12:47 PM, cyril pernet <[log in to unmask]> wrote:
>
>> If you have fitted 2 different models then you could derive a goodness of fit measure - if the data come from the same model this would not work because parametric regressors are serially orthogonalized
>>
>> cyril
>>
>>> Dear all,
>>>
>>> If I have two scans. One with 2 points regressored as parametric modulation and one with 4 points. I want to compare the first order effects of them in terms of the fitting lines. Would it be correct and possible if I take the contrast estimate and T scores and do comparisons between them? If what is the best way of doing this? I got the contrast and T values from ROI's using marsbar.
>>>
>>> Regards,
>>>
>>> AS
>>
>> --
>> The University of Edinburgh is a charitable body, registered in
>> Scotland, with registration number SC005336.
>>
>
>
--
Dr Cyril Pernet,
Academic Fellow
Brain Research Imaging Center
http://www.bric.ed.ac.uk/
Division of Clinical Neurosciences
University of Edinburgh
Western General Hospital
Crewe Road
Edinburgh
EH4 2XU
Scotland, UK
[log in to unmask]
tel: +44(0)1315373661
http://www.sbirc.ed.ac.uk/LCL/
http://www.sbirc.ed.ac.uk/cyril
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|