new technique: mass-univariate general linear model

[lzehl]

@skoehnen could you provide a nice definition for this one?

[UlrikeS91]

Not sure what to do with this. @skoehnen, @lzehl what should we do with this one? Discuss with @tgbugs?

[lzehl]

@mdenker how would you best define this analysis technique? I'm struggling a bit to create a clear one sentence definition for these kind of analysis techniques in general. It would be great if you could support us?

[lzehl]

To start a discussion:

Definition for GLM (from wikipedia): A 'general linear model' is a generalization of multiple linear regressions to the case of more than one dependent variable.

Extending to mass-univariate GLM: A 'mass-univariate general linear model' is a general linear model that is performed individually on a large number of elements of the measured data (e.g., voxels of images).

[skoehnen]

I was not familiar with this model before, but it would say the Definition that @lzehl posted sound good:

Definition for GLM (from wikipedia): A 'general linear model' is a generalization of multiple linear regressions to the case of more than one dependent variable.

Extending to mass-univariate GLM: A 'mass-univariate general linear model' is a general linear model that is performed individually on a large number of elements of the measured data (e.g., voxels of images).

The last sentence in the Applications section on Wikipedia worries me a bit:

It is usually tested in a univariate way (usually referred to a mass-univariate in this setting) and is often referred to as statistical parametric mapping

Would we create the potential for confusion if introduce this, do we already have a technique for SPM? Then this should be added as synonym.

[skoehnen]

do we already have a technique for SPM?

Does not look like it, could not find it with a search.

But I would argue that SPM is a quite common name in neuroscience circles (largely due to the software), should we rather create that.

[lzehl]

@skoehnen I would argue that SPM is used specifically in the imaging community while the applied mathematical technique is actually not imaging specific, correct? In principle the mass-univariate GLM can be used on all kinds of data, not only imaging data.

[lzehl]

@skoehnen & @tgbugs question is then: should we stick to a modality specific name / variant of a mathematical / analysis technique or not / or both ?

[lzehl]

@mdenker what is your insight here?

[skoehnen]

@skoehnen & @tgbugs question is then: should we stick to a modality specific name / variant of a mathematical / analysis technique or not / or both ?

I would say the name "mass-univariate general linear model" is the more "correct" name and should be used. SPM would be a specific application and should then be covered with a synonym.

[mdenker]

I would tend to agree that mass Univariate GL M is the more generic name, and also it is more telling in terms of the underlying idea. It's also the name I saw pop up when doing some searching. I would go with that. The definition seems good to me, a faik this is indeed used to analyze images pixel by pixel (or voxel) using a G L M, but really I an no expert here,

[lzehl]

@skoehnen and @mdenker thank you so much for your insight on this term. We will prepare then a respective PR for this (TODO: add definition to GLM term, add mass-univariate GLM as new term).