Back to PLS Helpmixed contrast
Posted on 02/02/07 12:36:08
Number of posts: 100
Author: Zara (158.223.64.---)
Date: 01-15-07 10:55
Hi there,
I'm trying to run a Non-Rotated Task PLS on ERP data. I have a mixed design with 2 groups and 3 repeated measures conditions. I enter a set of contrasts with a Helmert type breakdown like this below:
contrast 1: -1 -1 -1 1 1 1 (main effect group)
contrast 2: -.5 1 -.5 -.5 1 -.5 (cond 2 vs mean of cond 1+3)
contrast 3: -1 0 1 -1 0 1 (cond 1 vs 3)
contrast 4: .5 -1 .5 -.5 1 -.5 (group * contrast 2 interaction)
contrast 5: 1 0 -1 -1 0 1 (group * contrast 3 interaction)
However, when I save this contrast file I get error messages saying that the contrasts are rank deficient and correlated, even though they are orthogonal? Can you please tell me what I'm doing wrong and if there is a way to run the above analysis?
Thanks very much!
Zara
Date: 01-15-07 10:55
Hi there,
I'm trying to run a Non-Rotated Task PLS on ERP data. I have a mixed design with 2 groups and 3 repeated measures conditions. I enter a set of contrasts with a Helmert type breakdown like this below:
contrast 1: -1 -1 -1 1 1 1 (main effect group)
contrast 2: -.5 1 -.5 -.5 1 -.5 (cond 2 vs mean of cond 1+3)
contrast 3: -1 0 1 -1 0 1 (cond 1 vs 3)
contrast 4: .5 -1 .5 -.5 1 -.5 (group * contrast 2 interaction)
contrast 5: 1 0 -1 -1 0 1 (group * contrast 3 interaction)
However, when I save this contrast file I get error messages saying that the contrasts are rank deficient and correlated, even though they are orthogonal? Can you please tell me what I'm doing wrong and if there is a way to run the above analysis?
Thanks very much!
Zara
Replies:
mixed contrast addendum
Posted on 02/02/07 12:36:27
Number of posts: 100
Author: Zara (158.223.64.---)
Date: 01-15-07 11:28
I also tried it with the orthonormal contrasts below but I still get the same error messages.
-0.4082 -0.4082 -0.4082 0.4082 0.4082 0.4082
-0.2887 0.5774 -0.2887 -0.2887 0.5774 -0.2887
-0.5000 0 0.5000 -0.5000 0 0.5000
0.2887 -0.5774 0.2887 -0.2887 0.5774 -0.2887
0.5000 0 -0.5000 -0.5000 0 0.5000
Thanks!
Date: 01-15-07 11:28
I also tried it with the orthonormal contrasts below but I still get the same error messages.
-0.4082 -0.4082 -0.4082 0.4082 0.4082 0.4082
-0.2887 0.5774 -0.2887 -0.2887 0.5774 -0.2887
-0.5000 0 0.5000 -0.5000 0 0.5000
0.2887 -0.5774 0.2887 -0.2887 0.5774 -0.2887
0.5000 0 -0.5000 -0.5000 0 0.5000
Thanks!
Re: mixed contrast addendum
Posted on 02/02/07 12:36:45
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 11:32
I looked into both of your contrast matrices, and both of them should be fine.
I also looked into the program, but did not find any problem.
Would you please upload your data (*session.mat, *datamat.mat and *data.mat) onto your ftp site, so I can download and look further?
Date: 01-15-07 11:32
I looked into both of your contrast matrices, and both of them should be fine.
I also looked into the program, but did not find any problem.
Would you please upload your data (*session.mat, *datamat.mat and *data.mat) onto your ftp site, so I can download and look further?
Re: mixed contrast addendum
Posted on 02/02/07 12:37:00
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 11:37
i also need to know how do you input contrast matrix for the 2 groups. are the contrasts that you input the same? like:
-0.4082 -0.4082 -0.4082 0.4082 0.4082 0.4082
-0.2887 0.5774 -0.2887 -0.2887 0.5774 -0.2887
-0.5000 0 0.5000 -0.5000 0 0.5000
0.2887 -0.5774 0.2887 -0.2887 0.5774 -0.2887
0.5000 0 -0.5000 -0.5000 0 0.5000
Date: 01-15-07 11:37
i also need to know how do you input contrast matrix for the 2 groups. are the contrasts that you input the same? like:
-0.4082 -0.4082 -0.4082 0.4082 0.4082 0.4082
-0.2887 0.5774 -0.2887 -0.2887 0.5774 -0.2887
-0.5000 0 0.5000 -0.5000 0 0.5000
0.2887 -0.5774 0.2887 -0.2887 0.5774 -0.2887
0.5000 0 -0.5000 -0.5000 0 0.5000
Re: mixed contrast addendum
Posted on 02/02/07 12:37:19
Number of posts: 100
Author: Zara (158.223.64.---)
Date: 01-15-07 11:55
Hi Jimmy,
sorry I don't understand what you mean by "are the contrasts that you input the same"?
I'm using the gui to input the contrasts, so to enter contrast 1, I click on group 1 in the 'group' box, and enter e.g. -0.4082 -0.4082 -0.4082 and add as the first contrast, then I click on group 2 and enter 0.4082 0.4082 0.4082, etc. I get an error message if the number of values entered doesn't match the number of within ss conditions.
The contrasts and the output all looks fine in the results file, except that the lvintercorrs are quite high between some of them (up to .4).
I'm afraid I don't have access to an ftp site...
Thanks!
(I'm using version 4.0608291)
Date: 01-15-07 11:55
Hi Jimmy,
sorry I don't understand what you mean by "are the contrasts that you input the same"?
I'm using the gui to input the contrasts, so to enter contrast 1, I click on group 1 in the 'group' box, and enter e.g. -0.4082 -0.4082 -0.4082 and add as the first contrast, then I click on group 2 and enter 0.4082 0.4082 0.4082, etc. I get an error message if the number of values entered doesn't match the number of within ss conditions.
The contrasts and the output all looks fine in the results file, except that the lvintercorrs are quite high between some of them (up to .4).
I'm afraid I don't have access to an ftp site...
Thanks!
(I'm using version 4.0608291)
Re: mixed contrast addendum
Posted on 02/02/07 12:37:38
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 12:03
Oh yes, you actually already told me that you had 3 repeated measures conditions.
For 3 conditions, you can only have a maximum of 2 degree of freedom, i.e. a maximum of 2 contrasts.
Increase the number of group will not increase the degree of freedom. On the other hand, all the conditions must be the same for all the groups.
Date: 01-15-07 12:03
Oh yes, you actually already told me that you had 3 repeated measures conditions.
For 3 conditions, you can only have a maximum of 2 degree of freedom, i.e. a maximum of 2 contrasts.
Increase the number of group will not increase the degree of freedom. On the other hand, all the conditions must be the same for all the groups.
Re: mixed contrast addendum
Posted on 02/02/07 12:37:55
Number of posts: 100
Author: Randy (---.dsl.bell.ca)
Date: 01-15-07 12:49
Jimmy wrote:
> Increase the number of group will not increase the degree of
> freedom. On the other hand, all the conditions must be the same
> for all the groups.
this is technically incorrect:
with k tasks within N groups there are:
N-1 degrees of freedom for group main effects
k-1 degrees of freedom for task main effects
(N-1)*(k-1) degrees of freedom for interactions
There is a bug/feature in the program that does not consider "group" in checking the contrast matrix.
Date: 01-15-07 12:49
Jimmy wrote:
> Increase the number of group will not increase the degree of
> freedom. On the other hand, all the conditions must be the same
> for all the groups.
this is technically incorrect:
with k tasks within N groups there are:
N-1 degrees of freedom for group main effects
k-1 degrees of freedom for task main effects
(N-1)*(k-1) degrees of freedom for interactions
There is a bug/feature in the program that does not consider "group" in checking the contrast matrix.
Re: mixed contrast addendum
Posted on 02/02/07 12:38:13
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 12:54
I agree, and the bug will be fixed soon.
At the mean time, please ignore the warning message, and Zara should be able to save the contrast data and run PLS analysis.
Date: 01-15-07 12:54
I agree, and the bug will be fixed soon.
At the mean time, please ignore the warning message, and Zara should be able to save the contrast data and run PLS analysis.
Re: mixed contrast addendum
Posted on 02/02/07 12:38:36
Number of posts: 100
Author: Zara (158.223.64.---)
Date: 01-15-07 13:01
Thanks very much! Now I can finally sleep at night:)
Date: 01-15-07 13:01
Thanks very much! Now I can finally sleep at night:)
Re: mixed contrast addendum
Posted on 02/02/07 12:38:51
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 16:11
This bug is fixed.
> I agree, and the bug will be fixed soon.
>
> At the mean time, please ignore the warning message, and Zara
> should be able to save the contrast data and run PLS analysis.
Date: 01-15-07 16:11
This bug is fixed.
> I agree, and the bug will be fixed soon.
>
> At the mean time, please ignore the warning message, and Zara
> should be able to save the contrast data and run PLS analysis.
Re: mixed contrast
Posted on 02/02/07 12:39:09
Number of posts: 100
Author: Randy (---.dsl.bell.ca)
Date: 01-15-07 12:12
You are correct, Zara, that these contrasts are orthogonal and I am puzzled that you get the error.
Jimmy, can you please look into this? The correlations amongst the contrasts are zero (identity matrix) so there should be no error message.
Regarding the interlvcorrs being high, this is likely because the effects in your data are not exactly the same as the contrast you input. For example, lets assume the strongest effect in your data is the difference between the first condition and the other three. The two task contrasts you use capture parts of such an effect, but you need both to perfectly characterize it. In this case, the correlations between the singular images will be >0 since each captures part of the same effect.
Date: 01-15-07 12:12
You are correct, Zara, that these contrasts are orthogonal and I am puzzled that you get the error.
Jimmy, can you please look into this? The correlations amongst the contrasts are zero (identity matrix) so there should be no error message.
Regarding the interlvcorrs being high, this is likely because the effects in your data are not exactly the same as the contrast you input. For example, lets assume the strongest effect in your data is the difference between the first condition and the other three. The two task contrasts you use capture parts of such an effect, but you need both to perfectly characterize it. In this case, the correlations between the singular images will be >0 since each captures part of the same effect.
Re: mixed contrast
Posted on 02/02/07 12:39:26
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 12:27
Because Zara put 5 contrasts for 3 conditions, the program give warning that the contrasts are rank deficient; however, as we discussed, the program will not stop Zara from running Non-Rotated Task PLS, no matter the contrasts are rank deficient or even not orthogonal, if he ignore the warning message.
Date: 01-15-07 12:27
Because Zara put 5 contrasts for 3 conditions, the program give warning that the contrasts are rank deficient; however, as we discussed, the program will not stop Zara from running Non-Rotated Task PLS, no matter the contrasts are rank deficient or even not orthogonal, if he ignore the warning message.
Re: mixed contrast
Posted on 02/02/07 12:39:45
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-15-07 16:12
> Jimmy, can you please look into this? The correlations amongst
> the contrasts are zero (identity matrix) so there should be no
> error message.
This bug is fixed.
Date: 01-15-07 16:12
> Jimmy, can you please look into this? The correlations amongst
> the contrasts are zero (identity matrix) so there should be no
> error message.
This bug is fixed.
Login to reply to this topic.
