[FieldTrip] fieldtrip Digest, Vol 93, Issue 18

[Maris, E.G.G. (Eric)]

Dear Eelke and others,

"Would it not be fair to say that the null hypothesis here is that the expected value of our data equals zero?”

This question can be answered by “Yes” if one can produce a proof of the false alarm rate control of this “one-sample randomisation test” under the null hypothesis that you mention. I welcome the results of all successful attempts to this challenge. You do not have to produce the proof yourself, a reference to a descent peer-reviewed publication (e.g., a statistics journal) is also OK.

best,
Eric



From: Eelke Spaak <e.spaak at donders.ru.nl<mailto:e.spaak at donders.ru.nl>>
Subject: Re: [FieldTrip] One-sample cluster based permutation t-test ERP data
Date: 21 August 2018 at 09:53:14 CEST
To: FieldTrip discussion list <fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>>


Dear Eric and all,

Following up on your point (1), I've wondered in the past about the
one-sample randomization test (let's call it that to avoid the name
"permutation"). Would it not be fair to say that the null hypothesis
here is that the expected value of our data equals zero?

Cheers,
Eelke

On 19 August 2018 at 23:57, Maris, E.G.G. (Eric) <e.maris at donders.ru.nl<mailto:e.maris at donders.ru.nl>> wrote:
Dear FT-ers,

The one-sample statistical test continues to be an issue that raises
confusion. Although it is impossible to explain here the statistical
background in all detail, the following points are relevant for empirical
neuroscientists that apply cluster-based permutation tests to their data:

1. A permutation test can only be used for comparing two or more
experimental conditions. Thus, Stephen’s proposal is not a permutation test.
However, it does produce a p-value, and by comparing it with some nominal
value (e.g., 0.05) it can be used to take a decision. The problem with this
procedure is that it is unclear what is the null hypothesis to which this
decision pertains. Here lies the important difference with a permutation
test for the difference between two experimental conditions: tthe null
hypothesis involves that the biological data in the two conditions are
generated by the same probability distribution.

2. Comparing the activation (post-stimulus) with the baseline (pre-stimulus)
period can be performed using a permutation test, regardless whether the raw
data were transformed into a time-frequency representation or not. The
reason why some people think it cannot be used on the raw data (i.e., for
testing effects on the time-locked average) is that the baseline period is
typically used to normalize the activation period (by removing the DC
component). With this normalisation, the null hypothesis pertains to two
sets of biological data of which one (the activation data) is already a
function of the other (as a result of the normalisation procedure). If you
want to use the actvsblT statfun, It makes more sense to perform a high-pass
filter on the raw data rather than normalising the activation data using the
baseline data.

best,
Eric Maris





 1. Re: One-sample cluster based permutation t-test ERP data
    (Stephen Politzer-Ahles)

From: Stephen Politzer-Ahles <politzerahless at gmail.com<mailto:politzerahless at gmail.com>>
Subject: Re: [FieldTrip] One-sample cluster based permutation t-test ERP
data
Date: 19 August 2018 at 10:36:09 CEST
To: <fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>>


I'm not sure if there is or not, but in the past I have managed this by
creating an ERP dataset that is all zeroes, and then comparing my real
dataset to it with a cluster test. Since a one-sample test is the same as a
paired samples test between something and zero (e.g., comparing X and Y via
t-test is the same as comparing {X-Y} to zero), this should give the same
result.


---
Stephen Politzer-Ahles
The Hong Kong Polytechnic University
Department of Chinese and Bilingual Studies
http://www.mypolyuweb.hk/~sjpolit/


----------------------------------------------------------------------

Message: 1
Date: Fri, 17 Aug 2018 16:01:29 +0000
From: Alex Sel <alex.sel at psy.ox.ac.uk<mailto:alex.sel at psy.ox.ac.uk>>
To: "fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>" <fieldtrip at science.ru.nl<mailto:fieldtrip at science.ru.nl>>
Subject: [FieldTrip] One-sample cluster based permutation t-test ERP
       data
Message-ID: <47637187-A1A5-4040-8917-F04C501E65CF at psy.ox.ac.uk<mailto:47637187-A1A5-4040-8917-F04C501E65CF at psy.ox.ac.uk>>
Content-Type: text/plain; charset="utf-8"

Dear list,

I wonder if there has been a function developed to do a one-sample cluster
based permutation t-test ERP data. I am aware this is possible to do with
actvsblT for time-frequency data.

There is a forum thread from 2012 saying that this wasn’t implemented. But
I wonder if there is anyone there who might have resolve this issue and
wouldn’t mind sharing the solution.

Any help would be appreciated.

Best wishes,

Alex Sel, PhD
Associate Fellow of the Higher Education Academy
Wellcome Centre for Integrative Neuroimaging
Department of Experimental Psychology
University of Oxford
The Tinsley Building
OX1 3SR







Eric Maris | Donders Institute for Brain, Cognition, and Behaviour & Faculty of Social Sciences | Radboud University | PO Box 9104, 6500 HE Nijmegen | (024) 3612651 | www.ru.nl<http://www.ru.nl>

This message (and any attachments) is intended solely for the addressee(s) and may contain confidential information. If you are not the addressee, do not copy this message (and any attachments), forward or share this message with third parties. You are requested to notify the sender immediately and delete this message.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20180822/81f96ea4/attachment.html>