To everyone that responded to my query, thank you for your replies and
helpful suggestions.
Very much appreciated.
I am still a little stuck however.
With the design I mentioned in my initial email my end goal was to get a
contrast image for each of my participants so that I could do a
correlation with a behavioural variable at the second level. To let me
simplify the design, I am interested in a design such as (A < B) > (C <
D).
Due to the suggestions from this list, I created a new t-contrast which
looks like -1 1 1 -1
If I understand correctly (B - A) - (D - C) would be the same as B - A -
D + C.
I hope this is correct.
The other step required that was mentioned was to inclusively mask the
output of the above contrast with B > A
I could do this by running the analyses and pushing save in the SPM5 GUI
for both contrasts then putting both into imcalc and using the
expression (i1>0)&(i2>0).
I'm not sure this gives me the type of file as an output that I want to
run my correlation with however.
An alternative would be to make the saved image from the analysis of B >
A binary using imcalc (e.g. i1>0) and then multiply the outputted image
with the other saved image (A < B)>(C < D). I think this would preserve
the values in the image that I am interested in as I would be
multiplying only with 1's.
But again I'm not sure that this final image is appropriate for a
correlation?
Has anyone any suggestions please?
My goal is to get a usable image for each participant which I can then
enter into a 2nd level model and run a correlation with a behavioural
variable.
I have only ever used contrast images taken from the 1st level of
analysis for this purpose before.
Thanks,
Will
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Benjamin, Christopher
Sent: 05 July 2010 14:31
To: [log in to unmask]
Subject: Re: [SPM] contrast query
Yes, you can run the contrast (B>C)>(E>F), but you then need to
inclusively mask the output (e.g. using imcalc) w the contrast of
interest (i.e., B>C).
See Don McLaren's note:
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind1001&L=SPM&D=0&I=-3&d=No
+Match%3BMatch%3BMatches&P=123461
Hello All,
I would do as Chris suggests but inclusively mask with B>C.
Geoff
On 5 July 2010 02:07, Paloyelis, Yannis <[log in to unmask]>
wrote:
Hi,
I wonder if [0 1 -1 0 -1 1] would be appropriate, or whether B>C - E>F
should be calculated separately and then use imcalc to get their
difference. -If the purpose is to get voxels responding to B>C uniquely
...
Any comments-I'm also interested in this contrast.
Thanks s lot,
Yannis
On 2 Jul 2010, at 18:48, Chris Watson
<[log in to unmask]> wrote:
I think you would want [0 1 -1 0 -1 1]
because (B - C) - (E - F) = B - C - E + F.
Will McGeown wrote:
Dear experts,
I was wondering how I might model the following design in a first level
analysis. I would like to receive one contrast image per participant
(in order to run a second level correlation).
The design is across 4 sessions and looks something like this:
A B C A B C D E F D E F
I would like to know what areas are significant for (B > C)>(E>F)
The directions of the analysis are very important.
If it is possible, how should I define the contrast to model this?
Thanks.
Will
*****************************************************************************************
To view the terms under which this email is distributed, please go to http://www.hull.ac.uk/legal/email_disclaimer.html
*****************************************************************************************
|