How would you set this up? My understanding is that for a standard
multi run, multi subject experiment it would be a 3 level analysis
(run, subj, group). True? So if I want to do a paired t-test within
subject across runs (8 nov1 runs vs 8 nov2 runs per subj) how would I
set this up? I figured I'd have to do a second level paired t within
subject using the 8 vs 8 copes... Ah! here's where my brain fuzzed
before and added a non-existing level. The paired t-test yields the
subj map, so all it needs is a further level three group. I assume this
is what you were getting at.
On Jun 29, 2004, at 5:03 AM, Stephen Smith wrote:
> Hi - you don't need to worry about this at all - multiple comparison
> corrections are with respect to _questions_ (ie final thresholdings)
> not whatever (non-thresholding) stages led up to the final test. So
> having multiple levels doesn't contribute to a MC problem.
> p.s. - are you sure you need all the levels - for example, it may be
> possible to include the nov1&2&3 and cross-run model in a more complex
> second-level within-subject model.
> On Fri, 25 Jun 2004, Robin Goldman wrote:
>> Back with a question about analyzing the repeated novelty data (read
>> below for details). We've
>> decided to pair the runs that have the same stimuli (so, the Nov1 run
>> with it's matched Nov2 run).
>> This means that in addition to doing a straight forward first (run),
>> second (subject), and third
>> (group) level analysis I will also be doing a second level paired
>> t-test of the Nov1 copes vs the
>> Nov2 copes within subject, then a third (subject) and fourth (group)
>> level for these copes.
>> My question is this -- when you do the higher level analysis, do you
>> take the number of tests into
>> account for significance of the maps? If so, how do I adjust this for
>> the fact that I'm doing a set of
>> contrasts outside the normal analysis chain?
>> On Fri, 2 Apr 2004 16:46:53 -0500, Robin Goldman
>> <[log in to unmask]> wrote:
>>> Hi there,
>>> New day, new question. We're running an event related novelty oddball
>>> study. (Give people a task, surprise them every now and then with a
>>> "novelty" event (Nov1) that is always something new, after a while,
>>> another run, repeat some of these novelty events (Nov2) to see what
>>> happens). We scan the subjects over multiple runs (12). All of these
>>> contain Nov1 events, all but the first contain Nov2 events. (None of
>>> the novelty events are repeated in the same run, only in later ones).
>>> So here's the question. It seems to me there are a number of ways to
>>> analyze this data to see if there is a difference between first time
>>> and repeated novel events (Nov1 vs Nov2).
>>> 1. In all but the first run (where there are no Nov2), model an EV
>>> gives some relationship between the Nov1 and Nov2 events. But I don't
>>> want to a priori do this.
>>> 2. Have one EV for Nov1 and one EV for Nov2, do a contrast within
>>> run (all but run 1) of Nov1 vs Nov2. Then do a second level within
>>> subject of this cope, then a third level across subjects.
>>> 3. Have one EV for Nov1 and one EV for Nov2. Do a contrast for each
>>> Do a second level analysis within subject of the Nov1 contrast and
>>> the Nov2 contrast (so, 12 copes for Nov1 and 11 for Nov2). Do a third
>>> level within subject of the Nov1 cope vs the Nov2 cope. Then a fourth
>>> across subjects. (As I'm writing this, I'm sure this is not the way
>>> So perhaps I answered my own question... thoughts?
>>> thanks again.
>>> Robin Goldman, Ph.D.
>>> Hatch Center for MR Research
>>> Columbia University
>>> 710 W. 168th Street, NIB-1
>>> New York, NY 10032
>>> (212) 342-0867
> Stephen M. Smith DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
Robin Goldman, Ph.D.
Hatch Center for MR Research
710 W. 168th Street, NIB-1
New York, NY 10032