first level statistics of ISPC over trials

[Kirstin Heise]

Dear Mike,

I have calculated ISPC over trials between muscle and EEG channels and am now interested in testing on single participant level (1.) whether there is a significant difference between the ISPC during my time of interest and during baseline** and (2.) whether there is a difference between different task conditions.

I have been following your scripts related to chapter 19 and 26 for the computation of the ISPC and am trying to implement the permutation approach suggested in chapter 34 and in the respective script of your youtube lecture but seem to have manoeuvred myself into some severe confusion now…

What do I actually use to compute my null distribution?
I saw your suggestions to some related questions and am wondering how to implement the permutation in the case of a single difference matrix (time of interest - baseline or alternatively ISPC change in % of baseline ISPC) and testing this against 0. How would I implement this?

** Regarding the baseline, I am left with a shorter time period [-500:cue] than the time of interest [cue :1000]. I was wondering whether is sound to 'repmat' this time to the length of the time of interest or if using the mean of this time would be valid?

Any hint and prompt is highly welcome. Thanks a lot already.

Kind regards,
Kirstin

[Mike X Cohen]

Hi Kirsten. The baseline and trial windows don't need to be the same size. In general, it's good to try to match (roughly and to the extent possible) the data quality and characteristics between the conditions/windows; having similar N's is a good way to work towards that goal. 

In your case, test #2 is straight forward, because you can randomize the trial labels. 

Test #1 can be done by randomly moving around the baseline time window. For example, in one iteration, the "baseline" might end up being +1000 to +1500 ms.

Hope that helps!

Mike

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/analyzingneuraltimeseriesdata/9E3C6202-68B7-4FFD-BA6A-ECE50DA287A8%40kuleuven.be.

--

Mike X Cohen, PhD
Fresh look: mikexcohen.com

[Kirstin Heise]

Thank you for you speedy reply, Mike. 

I am going to try this for test #1 now. Just a quick question to be sure: Would you recommend doing the test on the baseline “normalised” (ISPCtoi-ISPCbl or ISPCtoi/ISPCbl*100 ) data or on the “raw” ISPC? 

Thanks again. 

kirstin

To view this discussion on the web visit https://groups.google.com/d/msgid/analyzingneuraltimeseriesdata/CAABvs%2Bu91ZtZ4wkjmP931iBd%2BA9K4sFOxC_Bg2gjFvNQJ-Zkhg%40mail.gmail.com.

[Mike X Cohen]

For ISPC (or any other phase-based measure), a baseline subtraction is fine, but you should avoid baseline divisions. ISPC during the baseline period can be really small, which can mean really large (possibly explosive) normalized values.

Mike

To view this discussion on the web visit https://groups.google.com/d/msgid/analyzingneuraltimeseriesdata/96F35F3D-A87E-4E64-B34A-1AA5A1D2CC84%40kuleuven.be.

[Kirstin Heise]

OK, got it and it is working now. 

Thanks a lot. 

And then my follow-up question (last before the weekend, I promise) would be regarding the best way how to proceed on group level. 

Following an advice from you in response to a related question in this group, I wonder whether exporting the individual z-maps to the group level would be reasonable in my case. And then to test whether the baseline change is significantly different from zero across the group. 

How would I approach this. I have not found this in your resources so far - maybe I looked in the wrong place?

kirstin

To view this discussion on the web visit https://groups.google.com/d/msgid/analyzingneuraltimeseriesdata/CAABvs%2BvNi_xQ1XOUMpuoXc8UY9irshND4t6NN8iMphyYyZQcaw%40mail.gmail.com.

[Mike X Cohen]

Hi Kirstin. Apologies for the delay in responding. 

Yes, once you have the data z-transformed for each subject, then you can test whether those z-transformed maps are consistently different from zero at the group level.

Mike

To view this discussion on the web visit https://groups.google.com/d/msgid/analyzingneuraltimeseriesdata/1ECFC06D-BFC4-47FE-A5E1-B7D7B33C53D9%40kuleuven.be.