Hi MH,
this is not simple at all!
So why in the example of Cornelius below 'demeaning within groups' was
right? Which is the difference with my example?
Thanks!
Angela
> Hi Angela,
>
> Just to keep it simple, as general advice, one should not ever demean
> separately within group UNLESS you know exactly what implications this
> has on the analysis. In the case of your specific example of testing
> for differences between groups in the slopes of their continuous
> covariate, that is NOT a reason for demeaning separately within group.
>
> cheers,
> -MH
>
> On Fri, 2011-04-01 at 13:46 +0200, Cornelius Werner wrote:
>> On Fri, Apr 1, 2011 at 11:28 AM, Angela Favaro <[log in to unmask]>
>> wrote:
>> > Hi Cornelius,
>> > thank you for your reply.
>> > However, you did not completely answer to my doubts.
>> > Let me make an example:
>> >
>> > I am studying a brain area whose connectivity (resting state) seems to
>> > show a positive correlation with a cognitive variable in the patient
>> group
>> > and a negative correlation with the same variable in the control
>> group.
>> > This cognitive variable also differ between groups.
>> > If I want to test the differences between groups in SLOPES, I will
>> demean
>> > within groups. Is it right?
>>
>> If the intercept is of no interest and you are sure not to affect your
>> contrasts of the mean regressors in an invalid way, I'd say yes.
>>
>> > If I want to test the differences between groups in connectivity
>> removing
>> > the effects of the cognitive variable (suppose this has some sense), I
>> > will demean across groups or (according to type of variable) I can not
>> > demean at all. Right?
>>
>> Again, I'd say yes.
>>
>> > For the second question: in a design with two groups and a covariate
>> of
>> > interest demeaned across groups. Do I need to use the -D option? And
>> what
>> > if the demeaning is within groups?
>>
>> If you have regressors modelling the mean (say, one regressor with 1's
>> for each patient and 0's for controls, and a second regressor with the
>> other way around), you should not need the -D option. If there is no
>> mean regressor but only one regressor demeaned across groups, you
>> probably need the -D flag, as far as I got it. If you demeaned within
>> groups in this setting, you probably also need the -D flag, but you
>> will be testing for slopes only - any difference of the mean will not
>> appear in your test statistics. If the lines were parallel in that
>> setting, but miles apart, you wouldn't see it.
>> Before taking this all for true, I'd advise also checking on all the
>> previous posts in other threads - might be I missed something myself.
>>
>> Hope that helps,
>> Cornelius
>>
>> > Thanks!
>> > Angela
>> >
>> >
>> >> Hi Angela,
>> >>
>> >> let's see if I got it right.
>> >>
>> >> 1) Besides testing for slopes, I am also interested in average group
>> >> differences. Thus, if ages weren't matched, I would be introducing a
>> >> confound, i.e., any effect introduced by progressing age (e.g. task
>> >> speed) would also influence the group mean. As long as I demean
>> ACROSS
>> >> groups, this will not influence the *group means* and their contrasts
>> >> - in the GLM, any *shared* variability simply disappears (and will
>> >> lower "sensitivity" of either contrast, and rightly so. Teaches me to
>> >> match groups the next time, as Jesper put it two days ago).
>> >> If, on the other hand, I demeaned only within groups, I would not
>> >> correct for the fact that there was a significant contribution of the
>> >> factor "age" to either group. All variability due to the difference
>> of
>> >> age means would be soaked up by the group means and their contrasts.
>> >> Therefore, if these group contrasts showed something significant, it
>> >> might have been just due to the age difference (group a is slower
>> than
>> >> b, but also happens to be the older one!), but not due to treatment
>> or
>> >> diagnosis, or whatever I was actually interested in.
>> >>
>> >> 2) As far as I got it, if you are only interested in correlations
>> with
>> >> a (demeaned) covariate and did not model any group mean, you also
>> >> should demean the data before "randomise"ing. As an example: running
>> >> randomise on VBM data of a depressed patient cohort, looking for GM
>> >> changes correlating with a suicidal ideation score ranging from -5 to
>> >> +5, mean 0. In this case, randomise -D will do the demeaning of the
>> >> DATA (not the covariates) for you, saving you the effort of running
>> >> fslmaths on the data.
>> >>
>> >> If anything of this is wrong, I am sure one of the other contributors
>> >> will point it out rather quickly and I'll have lost posting rights
>> for
>> >> 4 weeks or so :-)
>> >>
>> >> Cheers,
>> >> Cornelius
>> >>
>> >> On Thu, Mar 31, 2011 at 11:36 PM, Angela Favaro
>> <[log in to unmask]>
>> >> wrote:
>> >>> Dear FSL Masters,
>> >>> this discussion has been very helpful for me.
>> >>> But I still have two doubts:
>> >>> 1. Demeaning within groups is more an exception than a rule, but it
>> is
>> >>> the
>> >>> correct thing when I want to test differences between slopes (and
>> not
>> >>> differences between groups). Is it correct?
>> >>> In the example below the two groups have a similar age. What happens
>> if
>> >>> the covariate differs in the two groups?
>> >>>
>> >>> 2. What continues to be unclear to me is the use of the -D option in
>> >>> randomise. When is it necessary/advisable to use it? Only in one
>> group
>> >>> covariate analysis?
>> >>>
>> >>> Thank you
>> >>>
>> >>> Angela
>> >>>
>> >>>
>> >>>
>> >>>> Yeah, that's what I thought. And basically that's why I asked in
>> the
>> >>>> first place :-)
>> >>>> But thanks for all the contributions to this topic. I believe I
>> have
>> >>>> an idea on how to go about it, now.
>> >>>> Best regards,
>> >>>> Cornelius
>> >>>>
>> >>>> On Thu, Mar 31, 2011 at 5:06 PM, Michael Harms
>> <[log in to unmask]>
>> >>>> wrote:
>> >>>>> Just wanted to chime in that demeaning the performance EV
>> separately
>> >>>>> within group is a rather unique case that is specific to this
>> >>>>> particular
>> >>>>> post.
>> >>>>>
>> >>>>> Recent posts by Jesper (just yesterday), Jeannette, Tom, and
>> myself
>> >>>>> have
>> >>>>> all advised that, in general, one should demean across all
>> subjects
>> >>>>> (NOT
>> >>>>> within group separately).
>> >>>>>
>> >>>>> Given the recent posts on this, I thought it was worth making
>> explicit
>> >>>>> that demeaning within groups is not a "typical" situation.
>> >>>>>
>> >>>>> And as a matter of good reporting practice, any time that
>> demeaning is
>> >>>>> performed separately within group, rather than across all
>> subjects,
>> >>>>> that
>> >>>>> should be noted (and justified) very explicitly in any
>> presentation of
>> >>>>> the ensuing results.
>> >>>>>
>> >>>>> cheers,
>> >>>>> -MH
>> >>>>>
>> >>>>> On Thu, 2011-03-31 at 08:42 +0100, Stephen Smith wrote:
>> >>>>>> Hi
>> >>>>>>
>> >>>>>> On 30 Mar 2011, at 11:30, Cornelius Werner wrote:
>> >>>>>>
>> >>>>>> > Hi,
>> >>>>>> >
>> >>>>>> > sorry to revive such a well-worn topic. But there is something
>> I
>> >>>>>> did
>> >>>>>> > not quite get so far.
>> >>>>>> > As an example, I am examining a patient cohort and a control
>> cohort
>> >>>>>> > in
>> >>>>>> > a Dual Regression setup (resting state data). Patients and
>> controls
>> >>>>>> > are matched for age and gender. They obviously differ in
>> diagnosis,
>> >>>>>> > but also in one performance score. I am interested in basic
>> group
>> >>>>>> > differences and the differential correlation of connectivity
>> >>>>>> > strength
>> >>>>>> > of several RSNs with performance. For the final randomise-step,
>> my
>> >>>>>> > design matrix has a column for group mean "patient" and one for
>> >>>>>> > "controls" (consisting of 1, padded with zeroes where
>> applicable),
>> >>>>>> > and
>> >>>>>> > two separate columns for age (as a confounder) - one for each
>> >>>>>> group,
>> >>>>>> > respectively, because an age*group interaction on
>> connectivities
>> >>>>>> > could
>> >>>>>> > not be excluded a priori. As I was modelling the group mean
>> >>>>>> > separately, only the slopes associated with age were tested. Is
>> >>>>>> that
>> >>>>>> > correct so far?
>> >>>>>>
>> >>>>>>
>> >>>>>> I think so - sounds fine.
>> >>>>>>
>> >>>>>> > As the age means did not differ (tested beforehand),
>> >>>>>> > does it matter if I demeaned within group or across groups?
>> >>>>>> > Shouldn't
>> >>>>>> > the intercept be modelled by the group mean regressor, in any
>> case?
>> >>>>>> > Following Tom's last post, I'd probably demean across groups.
>> >>>>>> >
>> >>>>>> > The next thing is even more unclear to me:
>> >>>>>> > Due to an expected group*performance interaction (i.e. steeper
>> >>>>>> slope
>> >>>>>> > of increases in connectivity along with better performance in
>> >>>>>> > contrast
>> >>>>>> > to the other group), also the performance scores are split.
>> BUT:
>> >>>>>> > should I demean?
>> >>>>>>
>> >>>>>>
>> >>>>>> Yes - if you want to compare the *slopes* between the two groups,
>> >>>>>> demean the performance scores within group before padding with
>> zeros,
>> >>>>>> for each group's performance EV.
>> >>>>>>
>> >>>>>> > And if so, within groups, or across groups? In this
>> >>>>>> > case, mean differences in performance are believed to be *due
>> to*
>> >>>>>> > diagnosis - therefore, variability associated with the mean
>> should
>> >>>>>> > go
>> >>>>>> > to the group regressor, shouldn't it? In this case, I'd be
>> inclined
>> >>>>>> > to
>> >>>>>> > demean in order not to affect the group mean regressor
>> negatively,
>> >>>>>> > and
>> >>>>>> > to demean within groups, because of the (clearly) attributable
>> mean
>> >>>>>> > variability...?!
>> >>>>>> >
>> >>>>>> > Example:
>> >>>>>> >
>> >>>>>> > EV1: Patient mean
>> >>>>>> > EV2: Control mean
>> >>>>>> > EV3: Patient age (demeaned across groups - EV of no interest)
>> >>>>>> >
>> >>>>>>
>> >>>>>>
>> >>>>>> I presume you mean demeaned within group, then padded with zeros.
>> >>>>>>
>> >>>>>>
>> >>>>>> Cheers.
>> >>>>>>
>> >>>>>> > EV4: Control age ( " )
>> >>>>>> > EV5: Patient performance score (demeaned within patients)
>> >>>>>> > EV6: Control performance score (demeaned within controls)
>> >>>>>> >
>> >>>>>> > Patients>controls: 1 -1 0 0 0 0
>> >>>>>> > Controls>patients: -1 1 0 0 0 0
>> >>>>>> > Slope(performance score) patients > Slope(performance score)
>> >>>>>> > controls:
>> >>>>>> > 0 0 0 0 1 -1
>> >>>>>> > Slope(performance score) controls > Slope(performance score)
>> >>>>>> > patients:
>> >>>>>> > 0 0 0 0 -1 1
>> >>>>>> >
>> >>>>>> > Please don't hit me - I'm having a hard time getting my head
>> around
>> >>>>>> > this :-)
>> >>>>>> > Cheers,
>> >>>>>> > Cornelius
>> >>>>>> >
>> >>>>>> >
>> >>>>>>
>> >>>>>> ---------------------------------------------------------------------------
>> >>>>>> Stephen M. Smith, Professor of Biomedical Engineering
>> >>>>>> Associate Director, Oxford University FMRIB Centre
>> >>>>>>
>> >>>>>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>> >>>>>> +44 (0) 1865 222726 (fax 222717)
>> >>>>>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>> >>>>>> ---------------------------------------------------------------------------
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>
>> >>>>
>> >>>>
>> >>>>
>> >>>> --
>> >>>> Dr. med. Cornelius J. Werner
>> >>>> Department of Neurology
>> >>>> RWTH Aachen University
>> >>>> Pauwelsstr. 30
>> >>>> 52074 Aachen
>> >>>> Germany
>> >>>>
>> >>>>
>> >>>
>> >>
>> >>
>> >>
>> >> --
>> >> Dr. med. Cornelius J. Werner
>> >> Department of Neurology
>> >> RWTH Aachen University
>> >> Pauwelsstr. 30
>> >> 52074 Aachen
>> >> Germany
>> >>
>> >>
>> >
>>
>>
>>
>
>
|