Dear SPMers,

With Guillaume's help and reminder and getting several rejection mails of not being recognized. I hope I am finally subscribed to the mailing list. So I would like to attach my questions to see if it can really work successfully this time.

Thank you.

Best,

Naomi

---------- Forwarded message ----------
From: naomi chuang <[log in to unmask]>
Date: Mon, May 27, 2013 at 11:38 AM
Subject: Re: [fil.spm] 2nd Level Paired t-test with covariate
To: Guillaume Flandin <[log in to unmask]>

Dear Guillaume,

Thanks so much for your prompt reply. I clicked the subscribe link, filled out some info and is still waiting for the subscription confirmation letter.

I've also tried what you've suggested. However, I still have one question regarding to this One way ANOVA model. Because Factor 1 is obtained during MC phase of Control subjects, and Factor 2 is obtained during OV phase of the same Control subjects. So they sure are not independent. And Factor 3 and Factor 4 (Patients_MC phase, patients_OV phase, respectively) are not independent either. However, Factor 1 and factor 3, or Factor 2 and factor 4 should be independent to each other. Does this model representative enough to this problem?

Additionally, looking back to the 2-by-6 model I originally specified. If I contrast them with t-test: [0 0 0 0 1 -1] or [0 0 0 0 -1 1] what are we really testing?

Hope I will be able to send my question to the mailing list soon!

Thank you so much.

Best,

Naomi

On Fri, May 24, 2013 at 10:27 PM, Guillaume Flandin <[log in to unmask]> wrote:

Dear Naomi,

your email was not sent to the SPM mailing list because you need to
first subscribe to the list and then send your question to
<[log in to unmask]>:
http://www.fil.ion.ucl.ac.uk/spm/support/#JISCmail

I think that an important aspect of your design is to have your
covariate interact with group and conditions. One way to do so with the
current interface in SPM is to use a one-way anova with 4 levels (ie
modelling your 2x2 as a 1x4) and add your covariate having interaction
with factor 1. This will create a design matrix with 8 columns and you
can test for your two questions of interest with F-contrasts:
[0 0 0 0 1 -1 1 -1]
and
[0 0 0 0 1 -1 -1 1]

Best regards,
Guillaume.

On 24/05/13 10:53, naomi chuang wrote:
> Dear SPMers,
>
> I encountered a question about assigning contrasts to the covariate in
> the paired t-test. There are two groups of subjects in our study
> (Control vs. Patients). Each subject has two conditions (Different
> phases of menstrual cycle: Phase 1 and phase 2). I am interested in
> whether the effect of the estrogen is different between these two
> conditions. In addition, is this difference differs between the two groups?
>
> I already assigned three factors: Subject, Group, and Condition using a
> flexible design and one covariate (estrogen) having interaction with
> group was assigned, which gives me a 2-by-6 matrix (attached figure).
>
> row column
> 1 1: control_MC
> 1 2: control_OV
> 2 3: patient_MC
> 2 4: patient_OV
> 1 5: Estrogen of control subjects
> 1 6: Estrogen of patients
>
> I am quite confused about how to take this covariate into account.
>
> Thank you, indeed.
> Naomi Chuang
>
>
>
>
>

--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG