How to handle mixed block/event design with very different durations (20s blocks + stick events) given single stimdur?

[sibo-lin]

Dear Dr. Kay,

Thank you for this excellent toolbox!

I have a question about handling a mixed event-related/block design where events have very different durations.

Design:
My experiment has two types of events per run:

• Image blocks : 20s
• Motor events: brief Cue and Go signals following Image block

Problem:
Since stimdur must be a single scalar, I cannot set 20s for Image and ~0.1s for Motor simultaneously. I tried upsampling to TR=0.1s — because event onsets do not align to the original 1.0s TR grid , rounding to the nearest 1s TR would introduce up to 0.5s timing error — and modeling Image blocks as 200 consecutive stick functions (stimdur=0.1s). However, this settings consistently crashes due to memory issues

Question:
Would it be appropriate to run two separate GLMsingle passes using extraregressors?

• Pass 1 (Motor): Estimate Cue/Go betas (stimdur=0.1s). Image blocks are pre-convolved with canonical HRF and passed as extraregressors.
• Pass 2 (Image): Estimate Image betas (stimdur=20s). Motor events are pre-convolved with canonical HRF and passed as extraregressors.

Is this a valid approach, or would you recommend a different strategy for this type of mixed design?

Thank you in advance!

[kendrickkay]

The approach you sketched out is clever. I think, it is in principle, valid. But ultimately it's sort of a lot of bookkeeping and a pain to keep track of? I would suggest upsampling to a TR of 0.5 s. And you can use a stimdur of 0.1 s. The rounding error from a TR of 0.5 s is probably negligible given how slow and sluggish HRFs are. Hopefully that will get around your memory limitations

…

On Mar 26, 2026, at 2:09 AM, sibo-lin ***@***.***> wrote: sibo-lin created an issue (cvnlab/GLMsingle#204) <#204> Dear Dr. Kay, Thank you for this excellent toolbox! I have a question about handling a mixed event-related/block design where events have very different durations. Design: My experiment has two types of events per run: Image blocks : 20s Motor events: brief Cue and Go signals following Image block Problem: Since stimdur must be a single scalar, I cannot set 20s for Image and ~0.1s for Motor simultaneously. I tried upsampling to TR=0.1s — because event onsets do not align to the original 1.0s TR grid , rounding to the nearest 1s TR would introduce up to 0.5s timing error — and modeling Image blocks as 200 consecutive stick functions (stimdur=0.1s). However, this settings consistently crashes due to memory issues Question: Would it be appropriate to run two separate GLMsingle passes using extraregressors? Pass 1 (Motor): Estimate Cue/Go betas (stimdur=0.1s). Image blocks are pre-convolved with canonical HRF and passed as extraregressors. Pass 2 (Image): Estimate Image betas (stimdur=20s). Motor events are pre-convolved with canonical HRF and passed as extraregressors. Is this a valid approach, or would you recommend a different strategy for this type of mixed design? Thank you in advance! — Reply to this email directly, view it on GitHub <#204?email_source=notifications&email_token=AAU2DETZVHSFHD5CNRKT3R34STJRZA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBRGQYTENZTGI3TLJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAU2DEVCTCBPNCYXM2P2FML4STJRZAVCNFSM6AAAAACW76TH7GVHI2DSMVQWIX3LMV43ASLTON2WKOZUGE2DCMRXGMZDONI>. You are receiving this because you are subscribed to this thread.