Hi,

I have a repeated measure experiment (pre, post intervention), and I would like to test for changes induced by the intervention. When I test, even at near 0 treshold, I don't get any results. Hence I would like to validate my design matrix and exchange blocks (which I don't fully understand). Any suggestions welcome!

I have set up my files in alphanumerical order : 

aXXX01-pre.txt
aXXX02-pre.txt
aXXX03-pre.txt
bXXX01-post.txt
bXXX02-post.txt
bXXX03-post.txt
...

My design matrix is comprised of one column per subject, and a contrast column :

1 0 0 1
0 1 0 1
0 0 1 1
1 0 0 -1
0 1 0 -1
0 0 1 -1

Contrast = [ 0, 0, 0, 1] / or [1, zeros(1, nsub)]; (tried both ways)

Exchange bloacks = [ 1 2 3 4 1 2 3 4 ]
This is the part where I am not sure what it does!
I have tried, like in some exmples, to specify it as [1;2;3;4;1;2;3;4], and as reshape(repmat(1:nsub, 1, 2), 1, []);
In the last two cases, I get 'exchange blocks : no' in the output.

In all cases, even at very low T, I get no results. Thanks for letting me know if I did it the right way (and that there is no results in these data).

Great thanks,

Dylan

Hy Dylan, 

the design matrix and exchange blocks look fine to me. Simply using  [ 1 2 3 4 1 2 3 4 ]  for exchange blocks should be fine. 

You may also want to test the opposite contrast of [ 0 0 0 -1]. Currently it seems that you are only testing one direction of difference. 

Not being able to reject the null hypothesis is a result. 

Andrew

Originally posted by dylan sutterlin:

Hi,

I have a repeated measure experiment (pre, post intervention), and I would like to test for changes induced by the intervention. When I test, even at near 0 treshold, I don't get any results. Hence I would like to validate my design matrix and exchange blocks (which I don't fully understand). Any suggestions welcome!

I have set up my files in alphanumerical order : 

aXXX01-pre.txt
aXXX02-pre.txt
aXXX03-pre.txt
bXXX01-post.txt
bXXX02-post.txt
bXXX03-post.txt
...

My design matrix is comprised of one column per subject, and a contrast column :

1 0 0 1
0 1 0 1
0 0 1 1
1 0 0 -1
0 1 0 -1
0 0 1 -1

Contrast = [ 0, 0, 0, 1] / or [1, zeros(1, nsub)]; (tried both ways)

Exchange bloacks = [ 1 2 3 4 1 2 3 4 ]
This is the part where I am not sure what it does!
I have tried, like in some exmples, to specify it as [1;2;3;4;1;2;3;4], and as reshape(repmat(1:nsub, 1, 2), 1, []);
In the last two cases, I get 'exchange blocks : no' in the output.

In all cases, even at very low T, I get no results. Thanks for letting me know if I did it the right way (and that there is no results in these data).

Great thanks,

Dylan

 

Hi ,

Thank you for this quick and helpful response/validation (and for a reminder that indeed not rejecting null hyp. is informative at some point)!

Can you explain what the exchange blocks do? I just realize that I used the example [1 2 3 4], but with only 3 participants in my example, so I am a bit confused about its usage. Let's say that my final design matrix has 60 rows (cond a x sub00-sub30 and cond. b x sub00-sub30; like the example above), and 31 columns (1/sub + the contrast one).

Then would the exchange block look like : [1 2 3 4 5 ...30 1 2 3 4 5 ... 30] ? 

Thank you very much,

Dylan

Exchange blocks enable you to constrain permutations to occur within a group of observations, such as repeat measurments within the same individual.

There is an explanation of their use in the NBS manual. 

Originally posted by dylan sutterlin:

Hi ,

Thank you for this quick and helpful response/validation (and for a reminder that indeed not rejecting null hyp. is informative at some point)!

Can you explain what the exchange blocks do? I just realize that I used the example [1 2 3 4], but with only 3 participants in my example, so I am a bit confused about its usage. Let's say that my final design matrix has 60 rows (cond a x sub00-sub30 and cond. b x sub00-sub30; like the example above), and 31 columns (1/sub + the contrast one).

Then would the exchange block look like : [1 2 3 4 5 ...30 1 2 3 4 5 ... 30] ? 

Thank you very much,

Dylan

Thank you for the precision!

Best,

Dylan