Let me start from the beginning...
You have found an interaction/post-hoc t-test of part of the
interaction that shows task1>task2 is greater at time 1 than time 2.
There are 6 possibilities that could lead to this -- all would have
significant interactions:
(1) task1>task2 is significant for both time1 and time2.
(2) task1>task2 is significant for only time 1. task2>task1 is not
significant at time 2 either.
(3) task1>task2 is not significant at either time 1 or time 2.
(4) task2>task1 is significant at time 2, but not time 1.
(5) task2>task1 is significant for both time points.
(6) task1>task2 is sig. at time 1 and task2>task1 is sig. at time 2.
Any of these 6 are possible in the interaction. However, these 6
occurs do not mean that you have an interaction. The interaction must
be tested as you have done.
Now, you stated that you wanted to see if possibility #2. You've got
the test for task1>task2 at time 1 correct. However, you haven't
properly tested that neither task1>task2 nor task2>task1 are not
significant. Both have to be non-significant to meet your criteria.
Thus, you need to test task2>task1 for time 2 as well as task1>task2.
In my opinion, this is overkill though.
I'd stop at the interaction since you already know that one is greater
than the other. All you need to know is the sign of the individual
components, not whether each component is significant.
Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Postdoctoral Research Fellow, GRECC, Bedford VA
Website: http://www.martinos.org/~mclaren
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 Wed, Jan 9, 2013 at 11:21 PM, Jun Wang <[log in to unmask]> wrote:
> sorry, I didn't get you. The inclusive mask I used is from post hoc t
> contrast interaction. Then I apply this inclusive mask to two simple effect
> t tests on time1 and time2 separately. These one-sided spm t test should
> able to tell me the direction whether task1 > task2 or task2 > task1 at
> time1 or time2. why do I need exclusive mask of task2>task1 at time2.
>
> thanks
> Jun
>
>
> On Wed, Jan 9, 2013 at 10:05 PM, MCLAREN, Donald <[log in to unmask]>
> wrote:
>>
>> Yes. You could do that, but that's a rather restrictive definition for
>> an interaction and you'd need another mask to truly get to your
>> definition, see below.
>>
>> An interaction could be: task1>task 2 at time 1 AND task1>task2 at
>> time2, but the task effect is much greater than that at task 1.
>> Alternatively, task2>task1 at time2 could be significant. The latter
>> would not fit your definition as you've said task1 and task2 need to
>> be not different at time 2. You'd need an exclusive mask of
>> task2>task1 at time 2 to meet your definition.
>>
>> Best Regards, Donald McLaren
>> =================
>> D.G. McLaren, Ph.D.
>> Research Fellow, Department of Neurology, Massachusetts General Hospital
>> and
>> Harvard Medical School
>> Postdoctoral Research Fellow, GRECC, Bedford VA
>> Website: http://www.martinos.org/~mclaren
>> 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 Wed, Jan 9, 2013 at 10:55 PM, Jun Wang <[log in to unmask]> wrote:
>> > Sorry, Donald. I have to ask another question. my purpose is just to
>> > tell
>> > the direction for post hoc t contrast test. for example, [0 0 0 0 0 1 -1
>> > 0
>> > -1 1 0]. I want to know the the significant cluster in this t test is
>> > due to
>> > task difference on time window 1,but no difference on time window 2,
>> > then I
>> > should use the inclusive mask for simple effects [0 0 0 1 -1 1 0 0 -1 0
>> > 0]
>> > and [0 0 0 1 -1 0 1 0 0 -1 0]. please correct me if I am wrong.
>> >
>> > really appreciated your help
>> > Jun
>> >
>> >
>> >
>> > On Wed, Jan 9, 2013 at 9:41 PM, MCLAREN, Donald
>> > <[log in to unmask]>
>> > wrote:
>> >>
>> >> For the simple effects (e.g. [0 0 0 1 -1 1 0 0 -1 0 0]), you want to
>> >> use an an exclusive mask of the interaction as you generally cannot
>> >> interpret the simple effects in the presence of an interaction.
>> >>
>> >> I'd create a mask from the significant effects of the interaction,
>> >> then apply the mask as an exclusive mask. If you need an inclusive
>> >> mask, then create an image of all 1s, then subtract the interaction
>> >> mask.
>> >>
>> >> Best Regards, Donald McLaren
>> >> =================
>> >> D.G. McLaren, Ph.D.
>> >> Research Fellow, Department of Neurology, Massachusetts General
>> >> Hospital
>> >> and
>> >> Harvard Medical School
>> >> Postdoctoral Research Fellow, GRECC, Bedford VA
>> >> Website: http://www.martinos.org/~mclaren
>> >> 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 Wed, Jan 9, 2013 at 10:39 PM, Jun Wang <[log in to unmask]> wrote:
>> >> > thanks, Donald. This is really helpful. Just one last question, I
>> >> > searched
>> >> > across the list and saw some of your posts stating for simple effect
>> >> > tests
>> >> > [0 0 0 1 -1 1 0 0 -1 0 0], we use mask with interaction contrast or
>> >> > use
>> >> > significant cluster image from interaction[0 0 0 0 0 1 -1 0 -1 1
>> >> > 0](Maybe I
>> >> > am wrong about that). Right?
>> >> >
>> >> > thanks again for your help
>> >> > Jun
>> >> >
>> >> >
>> >> > On Wed, Jan 9, 2013 at 9:25 PM, MCLAREN, Donald
>> >> > <[log in to unmask]>
>> >> > wrote:
>> >> >>
>> >> >> For the 3x2 ANOVA, the F-contrast should be: [0 0 0 0 0 1 -1 0 -1 1
>> >> >> 0;
>> >> >> 0 0 0 0 0 0 1 -1 0 -1 1].
>> >> >>
>> >> >> The post-hoc tests:
>> >> >> [0 0 0 0 0 1 -1 0 -1 1 0]
>> >> >> [0 0 0 0 0 0 1 -1 0 -1 1]
>> >> >> [0 0 0 0 0 1 0 -1 -1 0 1]
>> >> >>
>> >> >> For task differences at a specific time window, I'd really recommend
>> >> >> GLM Flex (partition the variance by each within-subject factor), but
>> >> >> in the case of SPMs GLM:
>> >> >> [0 0 0 1 -1 1 0 0 -1 0 0]
>> >> >>
>> >> >> I've written several post on creating contrasts and would suggest
>> >> >> that
>> >> >> you search the list archives. Use the term S1G1C1.
>> >> >>
>> >> >> ** This assumes that the first 3 columns are time windows, the next
>> >> >> 2
>> >> >> are task, the next three of time windows of task 1 and the last 3
>> >> >> are
>> >> >> time windows for task 2.
>> >> >> ** You should be using the flexible factorial
>> >> >> ** You should also have a subject factor in the model, if its not
>> >> >> already
>> >> >> there.
>> >> >>
>> >> >> For viewing purposes, you might also want to create the negative of
>> >> >> the above contrasts, although peak_nii and OrthoView can both pull
>> >> >> out
>> >> >> both positive and negative values.
>> >> >>
>> >> >> Best Regards, Donald McLaren
>> >> >> =================
>> >> >> D.G. McLaren, Ph.D.
>> >> >> Research Fellow, Department of Neurology, Massachusetts General
>> >> >> Hospital
>> >> >> and
>> >> >> Harvard Medical School
>> >> >> Postdoctoral Research Fellow, GRECC, Bedford VA
>> >> >> Website: http://www.martinos.org/~mclaren
>> >> >> 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 Wed, Jan 9, 2013 at 10:11 PM, Jun Wang <[log in to unmask]>
>> >> >> wrote:
>> >> >> > sorry, could you make it more clear. I am pretty new for this
>> >> >> > contrast
>> >> >> > stuff. In this 2X3 repeated design, the F contrast should be set
>> >> >> > as
>> >> >> > [0 0
>> >> >> > 0 0
>> >> >> > 0 1 0 -1 -1 0 1] or something else. and what the contrast set if I
>> >> >> > want
>> >> >> > to
>> >> >> > know the simple effect about task difference on different time
>> >> >> > window.
>> >> >> >
>> >> >> > thanks
>> >> >> > Jun
>> >> >> >
>> >> >> >
>> >> >> > On Wed, Jan 9, 2013 at 9:05 PM, MCLAREN, Donald
>> >> >> > <[log in to unmask]>
>> >> >> > wrote:
>> >> >> >>
>> >> >> >> You'd still need the F-contrast even in that case. Post-hoc tests
>> >> >> >> would be the same as I stated.
>> >> >> >> There are some things in the Glascher document that should be
>> >> >> >> updated.
>> >> >> >> It's been on my list for quite some time, but never at the top of
>> >> >> >> the
>> >> >> >> list.
>> >> >> >>
>> >> >> >> Best Regards, Donald McLaren
>> >> >> >> =================
>> >> >> >> D.G. McLaren, Ph.D.
>> >> >> >> Research Fellow, Department of Neurology, Massachusetts General
>> >> >> >> Hospital
>> >> >> >> and
>> >> >> >> Harvard Medical School
>> >> >> >> Postdoctoral Research Fellow, GRECC, Bedford VA
>> >> >> >> Website: http://www.martinos.org/~mclaren
>> >> >> >> 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 Wed, Jan 9, 2013 at 9:46 PM, Jun Wang <[log in to unmask]>
>> >> >> >> wrote:
>> >> >> >> > Thanks,Donald. If my two factors are all within subject
>> >> >> >> > factor(3
>> >> >> >> > time
>> >> >> >> > window
>> >> >> >> > and 2 task condition), then I can use the contrast [0 0 0 0 0
>> >> >> >> > 1 0
>> >> >> >> > -1
>> >> >> >> > -1
>> >> >> >> > 0
>> >> >> >> > 1]. right ( I just followed the example from Glascher's paper).
>> >> >> >> > in
>> >> >> >> > this
>> >> >> >> > case, how should I set the contrast for posthoc t test right.
>> >> >> >> > basically,
>> >> >> >> > I
>> >> >> >> > want to know the task difference on different time window
>> >> >> >> >
>> >> >> >> > thanks
>> >> >> >> > Jun
>> >> >> >> >
>> >> >> >> >
>> >> >> >> > On Wed, Jan 9, 2013 at 8:14 PM, MCLAREN, Donald
>> >> >> >> > <[log in to unmask]>
>> >> >> >> > wrote:
>> >> >> >> >>
>> >> >> >> >> On Wed, Jan 9, 2013 at 9:08 PM, Jun Wang <[log in to unmask]>
>> >> >> >> >> wrote:
>> >> >> >> >> > Dear spm experts,
>> >> >> >> >> > I have a question regarding to post hoc t test in spm. I
>> >> >> >> >> > have
>> >> >> >> >> > a
>> >> >> >> >> > 2X3
>> >> >> >> >> > (group
>> >> >> >> >> > by task ) repeated design and got significant interaction
>> >> >> >> >> > when
>> >> >> >> >> > I
>> >> >> >> >> > used
>> >> >> >> >> > contrast [0 0 0 0 0 1 0 -1 -1 0 1].
>> >> >> >> >>
>> >> >> >> >> >>>> The correct interaction contrast is an F-test. [0 0 0 0 0
>> >> >> >> >> >>>> 1
>> >> >> >> >> >>>> -1
>> >> >> >> >> >>>> 0
>> >> >> >> >> >>>> -1 0
>> >> >> >> >> >>>> 0; 0 0 0 0 0 0 1 -1 0 -1 1]. This will tell you if there
>> >> >> >> >> >>>> is
>> >> >> >> >> >>>> an
>> >> >> >> >> >>>> interaction
>> >> >> >> >> >>>> between the task effects and group. From this, you could
>> >> >> >> >> >>>> test
>> >> >> >> >> >>>> for
>> >> >> >> >> >>>> pairwise
>> >> >> >> >> >>>> task difference interacting with your groups. With the
>> >> >> >> >> >>>> post-hoc
>> >> >> >> >> >>>> t-test of [0
>> >> >> >> >> >>>> 0 0 0 0 1 0 -1 -1 0 1], you know the direction of the
>> >> >> >> >> >>>> effect
>> >> >> >> >> >>>> and
>> >> >> >> >> >>>> don't need
>> >> >> >> >> >>>> any further tests.
>> >> >> >> >>
>> >> >> >> >>
>> >> >> >> >> Then I want to do post hoc t test to
>> >> >> >> >> > tell the direction of interaction( e.g. group difference on
>> >> >> >> >> > task1).
>> >> >> >> >> > Should I
>> >> >> >> >> > extract ROI activity from significant cluster found in the
>> >> >> >> >> > interaction
>> >> >> >> >> > effect and do the simple t test outside spm or should I set
>> >> >> >> >> > different
>> >> >> >> >> > contrast in same SPM design. if it is the latter case, how
>> >> >> >> >> > should
>> >> >> >> >> > I
>> >> >> >> >> > set
>> >> >> >> >> > the
>> >> >> >> >> > contrast.
>> >> >> >> >>
>> >> >> >> >> >>>> Testing the group effect of a single task in a
>> >> >> >> >> >>>> repeated-measures
>> >> >> >> >> >>>> design is not valid with the standard GLM. You'll need a
>> >> >> >> >> >>>> 2-sample
>> >> >> >> >> >>>> t-test. If
>> >> >> >> >> >>>> you use GLM Flex, then you can test the group effect of
>> >> >> >> >> >>>> individual
>> >> >> >> >> >>>> tasks.
>> >> >> >> >>
>> >> >> >> >> >
>> >> >> >> >> > thanks for your input
>> >> >> >> >> >
>> >> >> >> >> > Jun
>> >> >> >> >
>> >> >> >> >
>> >> >> >
>> >> >> >
>> >> >
>> >> >
>> >
>> >
>
>
|
