[FieldTrip] One-sample t-test with cluster-based permutation test

[Seung Goo Kim, Ph.D.]

Dear Eelke,

Thank you for an informative response. I just wonder how the discussion around the one-sample T-test finally has ended and why it was decided not to be included.

I think you're right that it should be tested with the paired T-test. I thought the autocorrelation in MEG data should be minimal, I think it would make more sense to pair them, especially the signal was low-pass filtered.

I think if you create a flat timeseries with the mean values of each baseline period as a "baseline trial" and do a paired t-test, the result should be identical to flipping signs of baseline-corrected activation trials.

During the permutation, if the labels are not swapped, then the paired difference at one trial-pair would be y - x0 (y is a timeseries in an activation period; x0 is the mean of the baseline period). If the labels are swapped, then the difference would be x0 - y. Since baseline-corrected activation trial is also y - x0, this is equivalent to flipping signs of the corrected activation trials.

But I'm not sure if I really want to use only the mean of the baseline period because then it would discard underlying noise structure in the baseline period. But also because the baseline period is not evoked by any stimuli of interest, we don't assume any sample-by-sample correspondence between the baseline and activation periods. So it feels a bit strange to do so too.

Best,
-SG

On 2018-10-05, at 04:07, Eelke Spaak <e.spaak at donders.ru.nl<mailto:e.spaak at donders.ru.nl>> wrote:

Dear SG,

Rather than go into the whole one-sample T test business again, I will
respond to one aspect of your question which I think might be useful.

but the result was quite different from the two-sample t-test comparing baseline vs activation trials.

I don't know exactly how you were doing the baseline vs activation
test, but I'll note two things here. First, you will typically not
want to do a *two-sample* test, but a *paired-sample* test; i.e. each
unit of observation (trial or subject) has both a baseline and an
activation period, and it's the paired difference that matters.

Second, you will want to test the activation period against the *mean*
across the entire baseline period (since we typically assume a
stationary baseline). If you were to simply do (1) select baseline
window as condition A; (2) select activation window as condition B;
(3) compare the two using cluster stats; then the statistic would be
comparing each time point for the activation period against the
matching time point in the baseline.

So basically, one way of doing activation versus baseline cluster
stats is to average the baseline window across time (probably repmat()
the mean over time again) and then use paired statistics against the
activation window. This should work at first or second level.

Hope that helps!

Best,
Eelke



Is there really no way to use cluster-based permutation test for one-sample T-test at the second level?

Best,
--
Seung-Goo ("SG") Kim, PhD

Postdoctoral Research Associate
O-Lab, Department of Psychology & Neuroscience, Duke University
Postal: 308 Research Drive, Durham, NC 27708, USA
Email: solleo at gmail.com<mailto:solleo at gmail.com>

_______________________________________________
fieldtrip mailing list
https://urldefense.proofpoint.com/v2/url?u=https-3A__mailman.science.ru.nl_mailman_listinfo_fieldtrip&d=DwIGaQ&c=imBPVzF25OnBgGmVOlcsiEgHoG1i6YHLR0Sj_gZ4adc&r=Oo6kXvV9N4AHFS-wWXXAKKDl4dUe6z92y4XtIzLkZJY&m=XJZB_02aOg0SpO6JLAKP_HYe0fIywEx0eEE02bh4ztg&s=O3uZfr2pkofitVkCpUH6tjHpi5z80NBC5OwVFBKTQL8&e=
https://urldefense.proofpoint.com/v2/url?u=https-3A__doi.org_10.1371_journal.pcbi.1002202&d=DwIGaQ&c=imBPVzF25OnBgGmVOlcsiEgHoG1i6YHLR0Sj_gZ4adc&r=Oo6kXvV9N4AHFS-wWXXAKKDl4dUe6z92y4XtIzLkZJY&m=XJZB_02aOg0SpO6JLAKP_HYe0fIywEx0eEE02bh4ztg&s=KgYvBgZIFHTGoNd1roIMW-v3VAI91zJHE2TvYPHWLvM&e=
_______________________________________________
fieldtrip mailing list
https://urldefense.proofpoint.com/v2/url?u=https-3A__mailman.science.ru.nl_mailman_listinfo_fieldtrip&d=DwIGaQ&c=imBPVzF25OnBgGmVOlcsiEgHoG1i6YHLR0Sj_gZ4adc&r=Oo6kXvV9N4AHFS-wWXXAKKDl4dUe6z92y4XtIzLkZJY&m=XJZB_02aOg0SpO6JLAKP_HYe0fIywEx0eEE02bh4ztg&s=O3uZfr2pkofitVkCpUH6tjHpi5z80NBC5OwVFBKTQL8&e=
https://urldefense.proofpoint.com/v2/url?u=https-3A__doi.org_10.1371_journal.pcbi.1002202&d=DwIGaQ&c=imBPVzF25OnBgGmVOlcsiEgHoG1i6YHLR0Sj_gZ4adc&r=Oo6kXvV9N4AHFS-wWXXAKKDl4dUe6z92y4XtIzLkZJY&m=XJZB_02aOg0SpO6JLAKP_HYe0fIywEx0eEE02bh4ztg&s=KgYvBgZIFHTGoNd1roIMW-v3VAI91zJHE2TvYPHWLvM&e=

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20181005/cf6ae4a2/attachment-0001.html>