Thank you for your excellent answers, they are very helpful. Please allow me to summarise and ask for additional clarification:
1) The main problem I'd like to solve is, as you spotted, whether the sequence will have required power to detect and effect (if there is one!). So, please disagree with me if this is incorrect, but one way to evaluate the potential of the sequence in
terms of detecting an effect would be to plug the condition files directly into a Fsl FEAT analysis design matrix (FEAT>Stats>Full Model Setup> Efficiency), which would come out with an efficiency/estimability % to detect a statistically significant effect
for a given comparison. The lower the effect required, the more efficient the design.
2) The second question is related to spacing of NULL trials. I would like to first run a univariate analysis and then run a multivariate analysis. You mentioned NULL trials would be 'helpful'- could you please clarify whether you mean having ANY null trials
or having null trials between EACH trial would be helpful?
My current design is allocating as much total null time as for any other condition (as recommended in freesurver list). This means that I do have some NULL trials, and they are randomly inserted in the sequence. However, most trials have no ITI between
them. If I did force ITIs between EACH trial I would have to decrease number of trials for each condition type, which could lower the design efficiency.
IF the problem of having carry-over effects of closely very spaced trials (i.e. majority of trials with no separation between trials) could be alleviated by rigorous first order counter-balancing within Optseq2 using the --focb option, is it necessary
at all to add a null trial between EACH trial into the design?
Thanks for your clarification.
Best,
Betina
Hi Betina, see answers below
doug
On 04/23/2012 12:04 PM, Betina Ip wrote:
> Dear Freesurfer list,
>
> I am using optseq2 for the first time and wanted to check with you whether my protocol makes sense, many thanks in advance for your insight.
>
> 1) I start by choosing the maximum scan duration equalling ~30 min:
You might want to break this up into shorter chunks. It's totally up to
you, you just don't need to have everything in a single run.
> 2) set the psd window so it can capture the entire waveform of a 4 second event
> 3) select equal amounts of trials for each of the 4 event types, I played around with it and 80 trials give me the greatest efficiency (2.72) and variance reduction factor average (54.37). Are these values acceptable? They are the best I could get tweaking the number of trials.
You can't tell whether a VRF of 54 is good enough. You are asking a
question about the power of your experiment, which optseq can't answer.
If you don't have an effect, then no number of trials or efficiency will
be good enough. Having said that, I usually shoot for a VRF of between
20 and 40, but that's just a rule of thumb.
>
> The command is here:
>
> ./optseq2 --ntp 440 --tr 4 --psdwin 0 20 \
> --ev evt1 4 80 \
> --ev evt2 4 80 \
> --ev evt3 4 80 \
> --ev evt4 4 80 \
> --nkeep 3 \
> --o gfatrial \
> --focb 10 \
> --nsearch 10000
>
> A sample trial list is here:
>
> 0.0000 2 4.000 1.0000 evt2
> 4.0000 4 4.000 1.0000 evt4
> 8.0000 0 4.000 1.0000 NULL
> 12.0000 3 4.000 1.0000 evt3
> 16.0000 3 4.000 1.0000 evt3
> 20.0000 2 4.000 1.0000 evt2
> …
>
> The other question I had concerns the position of null trials. Many trials are not separated by a NULL event, however studies of rapid event-related designs to report a minimum ITI, in this case the minimum seems to be 0? To assess the effect of forcing a minimum ITI on the sequence, I have run the search defining the --tnullmin to 4 sec (equal to 1 TR) however efficiency and VRFAvg drop to .79 and 16.07, obviously due to trade of between ntp and ntrials (ntp=440, ntrials=62).
>
> Are NULL trials between each event required at all if a standard FIR is used for the analysis? My aim is to first apply a univariate BOLD-analysis and subsequently a multivariate analysis to the dataset. Thanks for clarifying.
They are not strictly necessary, however, they may be helpful. If you
really don't want them as part of your design (eg, it would interfere
with the psychology), then don't use them but make sure to optimize the
FOCB more (I usually use 100 regardless). The disadvantage of not having
the nulls is that there will probably be some non-linearity in that the
second of two closely spaced trials will have a lower amplitude than if
it occurred in isolation. This will hurt your power and might cause a
false difference between conditions if they are not well counter-balanced.
doug
>
> Betina
> _______________________________________________
> Freesurfer mailing list
> Freesurfer at nmr.mgh.harvard.edu
> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
>
>
--
Douglas N. Greve, Ph.D.
MGH-NMR Center
greve at nmr.mgh.harvard.edu
Phone Number: 617-724-2358
Fax: 617-726-7422
Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting
FileDrop: www.nmr.mgh.harvard.edu/facility/filedrop/index.html