help > NBS- no significant result interpretation and design matrix cross-validation
Showing 1-5 of 5 posts
Apr 20, 2018 03:04 PM | hex_dex - AUTH
NBS- no significant result interpretation and design matrix cross-validation
Hello! I am new to NBS and though I find it an amazing tool I still
am not sure I completely do understand how to set it. I have the
eeg data from 2 groups (A & B) which represent different activities
and for each group I have 2 time points (pre & post). For each
group I have the eeg of 20 people so in total, I have 40 sets of 2
time points. My connectivity matrix is 22*22*80 and my design
matrix is 80*42. Specifically, my design matrix is
Time Time*Group Regression#1 Regression#2 ... Regression #39 Regression#40
line1 0 0 1 0 0 0
line2 1 1 1 0 0 0
line3 0 0 0 1 0 0
line 4 1 1 0 1 0 0
. .
. .
. .
line 37 0 1 0 0 1 0
line 38 1 0 0 0 1 0
line 39 0 1 0 0 0 1
line 40 1 0 0 0 0 1
I want a time x group analysis and my contrast matrix is [ 0 1 0 0 ...0] which consists of 42 columns. I run an F-test. However I take no significant results and I think I am doing something wrong. Is my design matrix ok? Do you see anything wrong?
Best regards,
Katerina
Time Time*Group Regression#1 Regression#2 ... Regression #39 Regression#40
line1 0 0 1 0 0 0
line2 1 1 1 0 0 0
line3 0 0 0 1 0 0
line 4 1 1 0 1 0 0
. .
. .
. .
line 37 0 1 0 0 1 0
line 38 1 0 0 0 1 0
line 39 0 1 0 0 0 1
line 40 1 0 0 0 0 1
I want a time x group analysis and my contrast matrix is [ 0 1 0 0 ...0] which consists of 42 columns. I run an F-test. However I take no significant results and I think I am doing something wrong. Is my design matrix ok? Do you see anything wrong?
Best regards,
Katerina
Apr 22, 2018 12:04 AM | Andrew Zalesky
RE: NBS- no significant result interpretation and design matrix cross-validation
Hi Katerina,
I don't think that your Time*Group column is correct. The Time* Group column should be the multiplication of the Group and Time columns. Let's assume line 1-4 is labelled group -1 and line 37-40 is labelled group 1. In this case,
GROUP TIME GROUP*TIME
line1 -1 0 0
line2 -1 1 -1
line3 -1 0 0
line4 -1 1 -1
line37 1 0 0
line38 1 1 1
line39 1 0 0
line40 1 1 1
Remember to remove the Group column from the design matrix. I have only shown it here so you can see how the Group and Time columns are multiplied to get the interaction effect.
Andrew
Originally posted by hex_dex:
I don't think that your Time*Group column is correct. The Time* Group column should be the multiplication of the Group and Time columns. Let's assume line 1-4 is labelled group -1 and line 37-40 is labelled group 1. In this case,
GROUP TIME GROUP*TIME
line1 -1 0 0
line2 -1 1 -1
line3 -1 0 0
line4 -1 1 -1
line37 1 0 0
line38 1 1 1
line39 1 0 0
line40 1 1 1
Remember to remove the Group column from the design matrix. I have only shown it here so you can see how the Group and Time columns are multiplied to get the interaction effect.
Andrew
Originally posted by hex_dex:
Hello! I am new to NBS and though I find it an
amazing tool I still am not sure I completely do understand how to
set it. I have the eeg data from 2 groups (A & B) which represent
different activities and for each group I have 2 time points (pre &
post). For each group I have the eeg of 20 people so in total, I
have 40 sets of 2 time points. My connectivity matrix is 22*22*80
and my design matrix is 80*42. Specifically, my design matrix
is
Time Time*Group Regression#1 Regression#2 ... Regression #39 Regression#40
line1 0 0 1 0 0 0
line2 1 1 1 0 0 0
line3 0 0 0 1 0 0
line 4 1 1 0 1 0 0
. .
. .
. .
line 37 0 1 0 0 1 0
line 38 1 0 0 0 1 0
line 39 0 1 0 0 0 1
line 40 1 0 0 0 0 1
I want a time x group analysis and my contrast matrix is [ 0 1 0 0 ...0] which consists of 42 columns. I run an F-test. However I take no significant results and I think I am doing something wrong. Is my design matrix ok? Do you see anything wrong?
Best regards,
Katerina
Time Time*Group Regression#1 Regression#2 ... Regression #39 Regression#40
line1 0 0 1 0 0 0
line2 1 1 1 0 0 0
line3 0 0 0 1 0 0
line 4 1 1 0 1 0 0
. .
. .
. .
line 37 0 1 0 0 1 0
line 38 1 0 0 0 1 0
line 39 0 1 0 0 0 1
line 40 1 0 0 0 0 1
I want a time x group analysis and my contrast matrix is [ 0 1 0 0 ...0] which consists of 42 columns. I run an F-test. However I take no significant results and I think I am doing something wrong. Is my design matrix ok? Do you see anything wrong?
Best regards,
Katerina
Apr 22, 2018 08:04 PM | hex_dex - AUTH
RE: NBS- no significant result interpretation and design matrix cross-validation
Dear Andrew,
thank you very much for your response. I prepared the design matrix under your instructions and although I still don't get any significant results, the p-values appear improved now.:-)
Best regards,
Katerina
thank you very much for your response. I prepared the design matrix under your instructions and although I still don't get any significant results, the p-values appear improved now.:-)
Best regards,
Katerina
Apr 22, 2018 11:04 PM | Andrew Zalesky
RE: NBS- no significant result interpretation and design matrix cross-validation
Hi Katerina,
this is possible and a valid result. You might want to experiment with different thresholds.
Andrew
Originally posted by hex_dex:
this is possible and a valid result. You might want to experiment with different thresholds.
Andrew
Originally posted by hex_dex:
Dear Andrew,
thank you very much for your response. I prepared the design matrix under your instructions and although I still don't get any significant results, the p-values appear improved now.:-)
Best regards,
Katerina
thank you very much for your response. I prepared the design matrix under your instructions and although I still don't get any significant results, the p-values appear improved now.:-)
Best regards,
Katerina
Apr 23, 2018 10:04 AM | hex_dex - AUTH
RE: NBS- no significant result interpretation and design matrix cross-validation
Hi Andrew,
once again thank you. I experimented with loads of thresholds and I seem to get a graph for a threshold of 1.18. Does this look a valid threshold or is it too low?
Katerina
Originally posted by Andrew Zalesky:
once again thank you. I experimented with loads of thresholds and I seem to get a graph for a threshold of 1.18. Does this look a valid threshold or is it too low?
Katerina
Originally posted by Andrew Zalesky:
Hi
Katerina,
this is possible and a valid result. You might want to experiment with different thresholds.
Andrew
this is possible and a valid result. You might want to experiment with different thresholds.
Andrew