Dear SPM experts,

I have completed the first-level analysis of subjective post-scan ratings of stimuli from two conditions (A and B). At this level I created con images using
0 1

and
0 0 0 1

My goal is to create 2nd level results that will indicate which neural activations increase as ratings increase. I would like to have separate results for each of two groups, for each gender separately.

For example, I would like to report that for Group1 females, neural activation occurred in X as Condition A ratings increased.

I have tried flexible factorial (unsuccessfully), and therefore, I am wondering whether it would be appropriate to conduct 8 separate one-sample t-tests. If so, can I rely on the significant FWE/FDR p values, as given, for each result?

Or is flexible factorial more appropriate, and in that case, would someone be able to advise  me to sort out my confusion.

I realize that in my first message (copied below)  I was confusing matrix and contrasts.

I updated my I matrix to show only four columns:

repl  subject   GroupGender   Condition

I have one row per scan.

repl always = 1

subject:  the number for that participant

GroupGender: values from 1 to 4 (Group1female, Group1male, Group2female, Group2male)

Condition (1, 2):  not independent

I indicated two main effects (for 3, and for 4)
and one interaction (3 4)

The design matrix could not be created. error message "factorial design specification   attempted to access I(:,5): index out of bounds because size (I) = [#subjects*2, 4]

I read through spm_run_factorial_design.m
and as a result I thought perhaps I should not be creating factors for repl and subject

so I amended the GUI to have 2 factors (GroupGender, and Condition), 2 main effects (1), (2), and one interaction (1 2)

The design matrix was created successfully but it is inappropriate.

I looked through SPM.mat, and noticed that (for example) it created 4 conditions, not 2, so it was reading off the GroupGender column as though that indicated conditions. 

I was then thinking that perhaps I should try the "non-power-user" method:

Select Subjects.
However, I do not understand the example in the manual.

For the 4th subject (that you describe in the example) why do you need
1
2
1
2

Why not simply
1
2

to capture the one task with its two levels (interacting with the subject)

??

or if there are in fact two task factors in the example, why repeat them with identical numbers
for the same subject ??

Thank you in advance for your advice.

J.M.

Date:    Fri, 22 Mar 2013 22:28:49 +0000
From:    Julie Morgan <[log in to unmask]>
Subject: <No subject given>

Dear SPM experts,

I hope that you will not mind helping me with what is probably a very basic
question.

Participants provided subjective post-scan ratings, each on a 7-point
scale, for two types of stimuli, A and B.
For the parametric analysis, I conducted first-level analyses, using
spm8, as:
ConditionA  with one parametric modulator     ConditionB  with one
parametric modulator     and entered a textfile with the 6 motion regressors

I chose hrf (with no derivatives).

To create the con files, my contrasts were:
0 1
and
0 0 0 1

I have tried to figure out how to conduct the second-level analysis.
I have read through the SPM8 manual, four publications that have reported
the use of parametric modulation, and have scoured the SPM helplist since
2008.  Unfortunately, I am still confused.

I would like to generate results showing how neural activation increases as
ratings increase.
I am hoping to have 8 separate results:
A and B ratings for each of two groups, for each gender.

I tried entering 8 factors into a flexible factorial, as main effects. I
used the "specify all" option.
For the factor matrix, I used the following (the first line is for the
"replication" as advised in the manual):


1 1 1 1 1 1 1 1
1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0
0 0 0 0 0 1 0 0
0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 1

When selecting the scans, I chose the Group1 females (conditionA con
images), Group1 females (conditionB con images),
Group1 males (conditionA con images), Group1 males (conditionB con images),
Group2 females (conditionA con images), and so on.

Unfortunately, this design did not run. The error message indicated "factor
matrix must have four columns". However, I am not interested in comparing
ratings A and B, or comparing either between-subjects factor; I am hoping
to simply generate 8 results.

Is it acceptable to run 8 separate one-sample t-tests?  If so, for multiple
comparisons, do I simply accept the FWE/FDR cluster p-values as given, for
each analysis?

Thank you in advance for your help.

J.M.