Dear Jessica,
You do not necessarily need to include these regressors with dummy
onset. But if you do, they must get a weight of 0 in the contrast
vector. Then make sure that your contrast will compute the average
effect for each condition across sessions as described in slide 12 of:
https://www.fil.ion.ucl.ac.uk/spm/course/slides19-may/03_Contrasts_and_Inference.pptx
Best regards,
Guillaume.
On 03/03/2020 16:55, Jessica Emily wrote:
> Dear SPM users,
>
> I have a question regarding treating missing events. Let's say I have
> four events A, B, C, and D, in which it is designed to be balanced in
> numbers for the whole experiment and divided across three sessions. This
> leads me to encounter a situation in which some of the sessions, they do
> not have specific condition (only A, C, and D). However, to keep the
> design matrix balanced, I introduced a dummy onset, which is taken from
> the last scan of that session (referring to this previous
> thread: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;f8bcba42.1206
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.jiscmail.ac.uk%2Fcgi-bin%2Fwebadmin%3FA2%3Dspm%3Bf8bcba42.1206&data=02%7C01%7C%7Cae285f25adc94996139408d7bf93c155%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637188513633088411&sdata=WfQivEdOeGkml1X6v6c31g5ho7IaJ5d0EhXWwZWyQGA%3D&reserved=0>).
>
> This strategy is problematic for me during model specification in the
> 1st-level analysis, in which the dummy onsets are considered to be not
> uniquely specified. I cannot move forward to create a contrast due to
> this. Regarding this issue, do you think this is a correct strategy to
> approach the problem of having missing events? If yes, I would be glad
> of some guidance about how to get the dummy regressor to be uniquely
> defined.
>
> I also found an alternative to tackle this missing event by
> concatenating the sessions
> (https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=spm;fd0d0c9b.1010
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.jiscmail.ac.uk%2Fcgi-bin%2Fwa.exe%3FA2%3Dspm%3Bfd0d0c9b.1010&data=02%7C01%7C%7Cae285f25adc94996139408d7bf93c155%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C1%7C637188513633088411&sdata=jDJCW%2BvbqKXBqG6W6%2F%2BJJ5m3rVo7RLUEV2HFnaSSojY%3D&reserved=0>),
> which would be my second part of the question. The way I define my
> design matrix for each session is by selecting new session/subject in
> the first level model specification, so that each of the session has its
> own constant, and the regressors of interest multiplies by 3 for the
> contrast vector (see illustration below). Is there a way in SPM to
> concatenace all regressors of interest across the sessions?
>
> image.png
>
>
> Thank you in advance!
>
> Best,
> Jessica Emily Antono
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|