From your pics it looks like your doing GLM with one regressor and
the intercept.
y = intercept + level +error
y=B0 +B1X1 +E
The model estimated the beta representing the mean level of
activation in the data column 2
n column 2 you have the effect (linear trend) over and beyond the mean.
So your questions seem to be
1 0 to find regions showing a positive linear trend
-10 to find region showing a negative linear trend
I am not seeing how the 1-1 or -1 1 subtractions would make sense.
Does SPM really let you do this subtraction?
Thanks for letting me think about this. i hope others will join the
discussion.
Jen
On Dec 17, 2008, at 12:48 PM, Dorian P wrote:
> Hello all,
>
> I sent this message some time ago but didn't have an answer.
> I would really be interested to know what is the "mean" column created
> in multiple regression. Can I use the contrast [1 -1] to compare the
> effect of my regressor against the mean or I can use only [1 0] and
> never put the mean in a contrast. My effects are coming highly
> significant if I compare against the mean, but not if I compare the
> regressor alone.
>
> You can see the covariate matrix on my previous message:
> https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0812&L=SPM&P=R35067
>
>
> Thank you
> Dorian
>
>
>
> 2008/12/10 Dorian P <[log in to unmask]>:
>> Dear all,
>>
>> Hope the analysis are flowing for you :)
>>
>> We are running a multiple regression analysis in 2nd level to catch
>> the differences of 3 levels of memory (20%, 40% and 60%). After
>> inserting the con_*.img files they are weighted in a Covariate with
>> 20, 40 or 60 (the vector corresponds to the scans in input).
>>
>> Now we have a design matrix of 2 columns, one is the covariate and
>> the
>> other the mean. We are running various contrasts ([1 -1], [-1 1])
>> etc,
>> but we don't understand what are we contrasting exactly (even though
>> the results are nice for what we expect).
>>
>> Can somebody help on the topic? Attached you can find the Covariates
>> matrix and the example of the fitted response for one contrast.
>>
>> Best Regards.
>>
>> Dorian.
>>
|