Dear Alexa and Christian,
First of all, thank you so much for your thoughtful replies!
I now understand from your emails that contrasts such as [0 1 -1 0 0 0 -1
1 0 0] are obtained by subtracting [0 -1 1 0 0 0 0 0 0 0] (ie. condition 3
group 1) from [0 0 0 0 0 0 -1 1 0 0] (ie. condition 3 group 2). And the fact it
is an interactions of group x condition plus some more enlightments. Heaps more
information pertaining to stats will make me to work on it to understand for a
week, haha
-_-;;
I further discussed with my lab group and decided tentatively to use;
condition 3 - condition 2 for group 2 > group 1: [0 1 -1 0 0 0 -1 1 0 0]
condition 4 - condition 2 for group 2 > group 1: [0 1 0 -1 0 0 -1 0 1 0]
condition 5 - condition 2 for group 2 > group 1: [0 1 0 0 -1 0 -1 0 0 1]
and in addition to these contrasts also employ:
condition 3 - condition 2 for group 1 > group 2: [0 -1 1 0 0 0 1 -1 0 0]
condition 4 - condition 2 for group 1 > group 2: [0 -1 0 1 0 0 1 0 -1 0]
condition 5 - condition 2 for group 1 > group 2: [0 -1 0 0 1 0 1 0 0 -1]
We think that by using these additional contrasts, 'two-tail test-like' analysis
can be done. I hope these are ok... Please correct me if I'm wrong. Thank you
again!!!!
^^ ~~
Best wishes,
Yerina Ji
Hi Christian, Yerina
I would say there¡¯s more than one kind of freak and two of these would work
just fine! In more technical language, Yerina¡¯s original suggestion is a
pooled-error RFX model and Christian¡¯s final option with 2-group t-tests use
partitioned errors. See numerous recent postings in the archive. The
intermediate-looking option however is not quite corect:
You compute the following contrasts:
condition 3 group 1: [0 -1 1 0 0 0 0 0 0 0]
condition 4 group 1: [0 -1 0 1 0 0 0 0 0 0]
condition 5 group 1: [0 -1 0 0 1 0 0 0 0 0]
condition 3 group 2: [0 0 0 0 0 0 -1 1 0 0]
condition 4 group 2: [0 0 0 0 0 0 -1 0 1 0]
condition 5 group 2: [0 0 0 0 0 0 -1 0 0 1]
Now you are able to compare the effect of condition 3 between the two groups
with the two-tailed t-test...
What it looks like you are suggesting here it taking 2nd level contrasts on to
yet another level of analysis. This is best avoided as it¡¯s confusing. In fact
I think the contrasts themselves (being just linear combinations of the first
level contrasts) would be fine and an inference on them valid but I think (I¡¯m
not certain) that the prewhitening done for the 2nd level model might mean that
the ¡®3rd level¡¯ inference whilst statistically valid might not actually be
about quite the right thing. Someone please correct me if I am rambling!
잹ut in any case it¡¯s best to stick to one of the 2 standard
alternatives.
In Yerina¡¯s 2nd level contrasts:
?font size=1 face="Times New Roman"> How shall I specify my contrasts? -
eg. if I were to compare 'group 2 > group 1'
> for 'condition 3 > condition 2', can I assign contrasts of
> [0 -1 1 0 0 0 1 -1 0 0] and [0 1 -1 0 0 0 -1 1 0 0] for two-tailed t-test?
The contrast you could enter are...
condition 3 - condition 2 for group 2 > group 1: [0 1 -1 0 0 0 -1 1 0 0]
condition 3 - condition 2 for group 2 > group 1: [0 1 0 -1 0 0 -1 0 1 0]
condition 3 - condition 2 for group 2 > group 1: [0 1 0 0 -1 0 -1 0 0 1] ?o:p>
These look like interactions because they are – they are interactions of
group x condition, which from Yerina¡¯s description sounds just right.
HTH, Alexa
--------------------------------------------------------------------------------
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf
Of Mohr, Christian
Sent: 05 October 2006 08:08
To: [log in to unmask]
Subject: Re: [SPM] Contrasts for Between groups analysis
Hi Yerina!
I am clearly no expert but I belive in my ability to help you with your
confusion!
Since I am no expert to all:
Please correct me if I post misleading information.
Yerina Ji schrieb:
> Dear experts,
>
> I've specified a factorial design for between groups analysis (SPM5). The
design
> matrix has 5 conditions each for 2 groups; so it goes like ;
>
> (condition 1, group 1) (condition 2, group 1) (condtion 3, group 1) (condition
> 4, group 1 ) (condition 5, group 1) (condition 1, group 2) (condition 2, group
> 2) (condition 3, group 2) (condtion 4, group 2) (condition 5, group 2)
Ok, I got that!
>
> It has 10 cells all together. Now I'm trying to set up some contrasts in order
> to compare group 2 over group 1 (i.e., group2 > group1) for different
> conditions. What makes it difficult is that the first two conditions are
> 'baseline' conditions employed for subtraction purpose for each of the latter
> three 'active' conditions. Hence, I need to subtract condition 2 (ignore
> condition 1) from either condition 3 or 4 or 5 for each groups, before
> comparing between groups.
You compute the following contrasts:
condition 3 group 1: [0 -1 1 0 0 0 0 0 0 0]
condition 4 group 1: [0 -1 0 1 0 0 0 0 0 0]
condition 5 group 1: [0 -1 0 0 1 0 0 0 0 0]
condition 3 group 2: [0 0 0 0 0 0 -1 1 0 0]
condition 4 group 2: [0 0 0 0 0 0 -1 0 1 0]
condition 5 group 2: [0 0 0 0 0 0 -1 0 0 1]
Now you are able to compare the effect of condition 3 between the two groups
with the two-tailed t-test...
But I would do it in a different way, because if I got that right you would have
not the degrees of freedom for the intersubject-effect in this test...
More further down...
> How shall I specify my contrasts? - eg. if I were to compare 'group 2 > group
1'
> for 'condition 3 > condition 2', can I assign contrasts of
> [0 -1 1 0 0 0 1 -1 0 0] and [0 1 -1 0 0 0 -1 1 0 0] for two-tailed t-test?
The contrast you could enter are...
condition 3 - condition 2 for group 2 > group 1: [0 1 -1 0 0 0 -1 1 0 0]
condition 3 - condition 2 for group 2 > group 1: [0 1 0 -1 0 0 -1 0 1 0]
condition 3 - condition 2 for group 2 > group 1: [0 1 0 0 -1 0 -1 0 0 1]
But is it really what you want??? I do not really know if this way is the
correct one. You want do a categorical comparison an these terms look like
interactions...
My solution would be:
Using the first-level GLM of each single subject computing the following
contrasts:
condition 3 subject 1: [0 -1 1 0 0]
condition 4 subject 1: [0 -1 0 1 0]
condition 5 subject 1: [0 -1 0 0 1]
condition 3 subject 2: [0 -1 1 0 0]
condition 4 subject 2: [0 -1 0 1 0]
condition 5 subject 2: [0 -1 0 0 1]
..
Then you sort them into two groups. Subject 1 - n = group 1 and subject n+1 - m
= group 2.
Now you do the two tailed t-test for condition 3 group 1 > group 2, condition 4
group 1 > group 2, condition 5 group 1 > group 2 seperately.
And you got a real rfx-analysis (two-tailed t-test) with the intersubject
effect..
I hope you want to do this kind of statistics... If not I produced the next
freaky posting...
Best,
Chris
> I know it's a freaky email and thank you so much if anyone can please reply
with
> something !!!!!!!!!!!!
>
>
> Yerina
Christian Mohr
Klinik f? Neurologie
Universit?sklinikum Schleswig-Holstein
Campus L?eck
Ratzeburger Allee 160
23538 L?eck
Telefon: +49 451-500-3709
FAX: + 49 451-500-2489
email1: [log in to unmask]
email2: [log in to unmask]
|