Matthew,
You can calculate the detection efficiency of different block orders. Years ago
I did this for 20 second blocks and the order didn't have much effect
on efficiency.
Efficiency is defined by (c'*inv(X'X)*c)^-1, where c is the contrast and
X the design matrix.
For a block experiment with two conditions and rest, see matlab
statements below to calculate efficiency for two possible
contrasts.
Jim Lee
theones = ones(125,1);
rest=SPM.xX.X(:,1);
fov=SPM.xX.X(:,2);
per=SPM.xX.X(:,3);
X=[rest fov per theones];
c = [-1 1 0 0]; (for fov > rest)
disp( 1 / (c*inv(X'*X)*c' ) ); = 21.14
c = [-1 0 1 0]; (for per > rest)
disp( 1 / (c*inv(X'*X)*c' ) ); = 21.03
On 4/25/16, Matthew B <[log in to unmask]> wrote:
> Thank you for your detailed answer, this is very helpful.
>
> Matthew
>
> On Thu, Apr 21, 2016 at 11:03 PM, <[log in to unmask]> wrote:
>
>> This isn’t the easiest question to give a perfect answer to, I think. I
>> am
>> assuming you are saying you have 35 second blocks of events, and not
>> necessarily a long gap between blocks. And you have several repetitions
>> of
>> each block.
>>
>>
>>
>> The order or presentation will have some impact on your statistical
>> power.
>> I think the most powerful design will have the maximum combinations of
>> different blocks following each other. Consider the end of a block to be
>> the critical point here. The best design will have the most and equal
>> conditions of Type 1 followed by type2, or type3, or type4, and the most
>> possible combinations of type2 followed by 1 or 3 or 4.
>>
>>
>>
>> For for example a design of blocks (indicated a T1 T2 T3 or T4) is weak
>> if
>> repeated
>>
>>
>>
>> e.g. T1 T2 T3 T4 T1 T2 T3 T4 T1 T2 T3 T4
>>
>>
>>
>> because T2 always follows T1.
>>
>>
>>
>> Whereas this for example order will be better:
>>
>> T1 T2 T3 T4 T1 T3 T2 T4 T2 T1 T4 T3
>>
>>
>>
>> Because it has more instances of different event types following each
>> other, which better allows the GLM to deconvolve the magnitude of effects
>> for each block.
>>
>>
>>
>> In theory it is also a good idea to have some instances of a repeating
>> block (e.g. T1 T1 T2 T3 T4 T4), but this may or may not be feasible for
>> your design.
>>
>>
>>
>> For what it is worth, I personally feel it is always simplest to present
>> the same order for all participants, especially if you are comparing
>> groups
>> (in case one group randomly tends to get a ‘more efficient’ design, on
>> average, and to avoid possible errors down the line).
>>
>>
>>
>> I hope that is helpful.
>>
>>
>>
>> Colin Hawco, PhD
>>
>> Neuranalysis Consulting
>>
>> Neuroimaging analysis and consultation
>>
>> www.neuranalysis.com
>>
>> [log in to unmask]
>>
>>
>>
>>
>>
>>
>>
>> *From:* SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
>> *On
>> Behalf Of *Matthew B
>> *Sent:* April-21-16 4:21 PM
>> *To:* [log in to unmask]
>> *Subject:* [SPM] block order
>>
>>
>>
>> Dear all,
>>
>>
>>
>> I'm resending this question hoping that someone could give me any
>> pointers
>> on the possible effects of using either a fixed, a completely randomized
>> or
>> a pseudorandomized order of blocks. Or is this choice not expected to
>> have
>> any impact on an experiment's detection power? Any thoughts on this topic
>> would be appreciated.
>>
>>
>>
>> Matthew
>>
>> ---------- Forwarded message ----------
>> From: *Matthew B* <[log in to unmask]>
>> Date: Monday, April 18, 2016
>> Subject: block order
>> To: [log in to unmask]
>>
>> Dear all,
>>
>> I have a question about the optimal block order in a block design. In a
>> block design with 4 conditions each lasting around 35 seconds, where most
>> first level contrasts would include only 2 of 4 conditions, is it more
>> optimal to randomize the order of the blocks, to use a fixed block order,
>> or to pseudorandomize the order (e.g. using cycles of the 4 conditions,
>> each with random order)?
>>
>> The HRF would be (35*4)*2 in each case?
>>
>> Thanks in advance for any help!
>>
>> Matthew
>>
>>
>>
>
|