need to mean center covariates in SnPM?

Kathleen Hupfeld

unread,

Sep 15, 2020, 10:12:48 PM9/15/20

to Statistical Nonparametric Mapping

Hello,

I'm running a two groups, two sample t-test in SnPM and want to include age and sex as covariates.

In SPM second level models, when you enter a new covariate, SPM defaults to overall mean centering each covariate.

In SnPM there is just a space to enter the covariate name & vector of values. Is it recommended that e.g., for age we demean the covariate values before entering into SnPM? I couldn't find mention of this in the batch or manual so I wasn't sure for SnPM models.

Any insight would be greatly appreciated. Thank you.

Best wishes,

Kathleen

Thomas Nichols

unread,

Sep 16, 2020, 3:17:14 AM9/16/20

to Kathleen Hupfeld, Statistical Nonparametric Mapping

Dear Kathleen,

The exact behviour SnPM depends on the particular plug in. For the Two Sample T-test plug in, covariates are always centered, so there's no need to do it, and, anyway, it won't matter: Centering or not-centering will give the same results for the group mean difference.

-Tom

--
You received this message because you are subscribed to the Google Groups "Statistical Nonparametric Mapping" group.
To unsubscribe from this group and stop receiving emails from it, send an email to snpm-support...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/snpm-support/f02e2297-3a62-40c2-a8a2-deef794696f4n%40googlegroups.com.

Kathleen Hupfeld

unread,

Sep 16, 2020, 9:32:15 AM9/16/20

to Statistical Nonparametric Mapping

Great, thanks so much! Would that be the same (i.e., I do not need to center before entering #'s) for the MultiSub: Simple Regression, 1 covariate of interest? i.e., if I'd like to enter age/sex as covariates of no interest, but a behavioral metric as a covariate of interest?

Best wishes,
Kathleen

Thomas Nichols

unread,

Sep 16, 2020, 10:26:35 AM9/16/20

to Kathleen Hupfeld, Statistical Nonparametric Mapping

Sorry this isn't better documented.

Yes, in MultiSub: Simple Regression, covariate of interest and all others supplied are centered.

Do you think this is a problem?

-Tom

To view this discussion on the web, visit https://groups.google.com/d/msgid/snpm-support/7c5d2225-fba0-4715-b445-546c748e090en%40googlegroups.com.

Hupfeld,Kathleen E

unread,

Sep 16, 2020, 10:46:00 AM9/16/20

to Thomas Nichols, Statistical Nonparametric Mapping

No worries. I do not think it's a problem! I think it's preferable to default to centering all covariates, particularly for folks accustomed to setting up SPM models (where the default is to mean center all covariates).

-K

From: Thomas Nichols <thomas....@bdi.ox.ac.uk>
Sent: Wednesday, September 16, 2020 10:25 AM
To: Hupfeld,Kathleen E <khup...@ufl.edu>
Cc: Statistical Nonparametric Mapping <snpm-s...@googlegroups.com>
Subject: Re: need to mean center covariates in SnPM?

[External Email]

Reply all

Reply to author

Forward