Based on my experience and knowledge of statistics (I'm not a trained
statistician), repeated measures designs should use the flexible
factorial model. The reason for this is two-fold: (1) the degrees of
freedom in the full factorial are higher than statistical text state
they should be for an ANOVA; (2) including the subject term constrains
the model.
The main effect of group in the flexible factorial uses the same
degrees of freedom and error term as the within-subject effects and
this does not match with the gold standard of the one-sample t-test.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.
On Fri, Sep 2, 2011 at 11:50 AM, Bedda Rosario <[log in to unmask]> wrote:
> Hello Donald,
>
> I ran a similar analysis (but for PET data) by specifying a full factorial
> model (2 groups (between) and 2 conditions (within)- pre and post). I
> defined contrasts to test main effects and interaction between group and
> condition. I thought the difference between full factorial and flexible
> factorial is that you use a flexible factorial when there is not sufficient
> data to test all possible effects. Do both models account for nonsphericity?
>
> So, if you have a repeated measures design, you should run a flexible
> factorial model in spm?
>
> Thank you,
> Bedda
>
> On Fri, Sep 2, 2011 at 10:30 AM, MCLAREN, Donald <[log in to unmask]>
> wrote:
>>
>> Alec,
>>
>> The full factorial is invalid for repeated-measure designs.
>>
>> For repeated-measure designs you need to do the following:
>> (1) Use the flexible factorial model with subject, group, and
>> condition factors. Then you can investigate the group*condition and
>> condition effects (within-subject effects).
>>
>> (2) For group effects (between-subject effects), you need to average
>> the two conditions and do a two-sample t-test. If you want one of the
>> conditions compared between groups, then you need to do a two-sample
>> t-test of that condition.
>>
>> (3) For group*condition or condition (within-subject effects), the
>> flexible factorial model is accurate.
>>
>> (4) For multiple within-subject factors, all standard models in SPM are
>> invalid.
>>
>> The reason for the above 4 statements is related to the
>> degrees-of-freedom and the error terms used in the statistical
>> computations.
>> In the future, there should be a toolbox to do this within one model.
>> For more details see: http://www.martinos.org/~mclaren/ftp/presentations
>> ===============
>>
>> Your specific questions:
>> 1. Independence should be no for within-subject factors (e.g.
>> condition), independence should be yes for between-subject factors.
>> The reason for this is that subjects are independent of each other,
>> but conditions are not as they are collected within the same subject.
>> Variance for subject and conditions should be equal as the conditions
>> and subjects both come from the same source; variance for group
>> should be unequal since the groups could have different means and
>> variances.
>>
>> 2. See above about the full factorial model.
>>
>> 3. If you use the steps above, then you should be fine.
>>
>>
>> Best Regards, Donald McLaren
>> =================
>> D.G. McLaren, Ph.D.
>> Postdoctoral Research Fellow, GRECC, Bedford VA
>> Research Fellow, Department of Neurology, Massachusetts General Hospital
>> and
>> Harvard Medical School
>> Office: (773) 406-2464
>> =====================
>> This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
>> HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
>> intended only for the use of the individual or entity named above. If the
>> reader of the e-mail is not the intended recipient or the employee or
>> agent
>> responsible for delivering it to the intended recipient, you are hereby
>> notified that you are in possession of confidential and privileged
>> information. Any unauthorized use, disclosure, copying or the taking of
>> any
>> action in reliance on the contents of this information is strictly
>> prohibited and may be unlawful. If you have received this e-mail
>> unintentionally, please immediately notify the sender via telephone at
>> (773)
>> 406-2464 or email.
>>
>>
>>
>>
>> On Fri, Sep 2, 2011 at 4:40 AM, Alec Sproten
>> <[log in to unmask]> wrote:
>> > Dear SPM experts,
>> >
>> > I have been reading in the archives topics which relate to my following
>> > problem but I ended up a bit confused with no confidence regarding what is
>> > best to do and why.
>> > I would be more than happy if you could help me to understand the issue
>> > of sphericity and solve my specific problem to ensure a good design for the
>> > analysis.
>> >
>> > My experiment has two subject groups (25 in group 1 and 21 subjects in
>> > group 2). Each subject perform 2 types of task (i.e. 2 conditions).
>> > I am interested in both: in-between subjects effects and within-subjects
>> > effects. So for the 2nd level analysis I set the model specification with
>> > Full-Factorial Design with 2 factors (f1: task type, f2: subject group). f1
>> > has 2 levels for the two types of task and f2 has also 2 levels for the two
>> > group of subjects.
>> > I understand that in this ANOVA design analysis I included both the
>> > between-subject analysis (the f2) and the within-subject analysis (the f1).
>> > My questions are the following:
>> > 1. What should I choose regarding the independence (yes/no) and the
>> > variance (equal/unequal) in my design? And most importantly why?
>> > 2. Does my design fit for my question or should I use other type of
>> > design such as flexible factorial design? And if so what shall be the
>> > parameters then?
>> > 3. If there is anything else that I should know or check in my
>> > design please let me know.
>> > I thank you very much in advance,
>> > Alec
>> >
>
>
|
