help
help > RE: contrast ANOVA
Nov 19, 2014 02:11 AM | Andrew Zalesky
RE: contrast ANOVA
Hi Andreas,
What you have described is normal and possible.
When you increased to F=5, it is possible that the extent of your observed effect became smaller or remained the same. However, increasing your threshold also means that it is less likely for larger subnetworks to emerge in the randomized data.
So, it is perfectly fine to observe a significant effect for F=5, but not for F=4.
Andrew
Originally posted by Andreas Hahn:
What you have described is normal and possible.
When you increased to F=5, it is possible that the extent of your observed effect became smaller or remained the same. However, increasing your threshold also means that it is less likely for larger subnetworks to emerge in the randomized data.
So, it is perfectly fine to observe a significant effect for F=5, but not for F=4.
Andrew
Originally posted by Andreas Hahn:
Dear
Andrew,
Thanks for the help!
Another question regarding thresholds: I tried several thresholds for the F-test, but interestingly F=5 gave a significant cluster but F=4 did not. How can this be? I thought a lower initial threshold would just make the cluster larger?
Best,
Andreas
Originally posted by Andrew Zalesky:
Thanks for the help!
Another question regarding thresholds: I tried several thresholds for the F-test, but interestingly F=5 gave a significant cluster but F=4 did not. How can this be? I thought a lower initial threshold would just make the cluster larger?
Best,
Andreas
Originally posted by Andrew Zalesky:
Dear Andreas,
NBS differs from SPM in this respect. However, both design matrix formats will give exactly the same result. Note that for a one-way ANOVA with four groups, you could also use a design matrix in the NBS of the following form:
1 1 0 0
1 1 0 0
1 -1 1 0
1 -1 1 0
1 0 -1 1
1 0 -1 1
1 0 0 -1
1 0 0 -1
The contrast is then: [ 0 1 1 1].
Looking at the above design matrix, it can be seen that columns 2, 3 and 4 correspond to the rows of your SPM contrast matrix.
I hope this provides you with some intuition above the equivalence between them.
The initial threshold refers to the F-statistic (if you select an F-test). So if you use an F-test, you would typically select a larger threshold than if you were using a t-test.
Originally posted by Andreas Hahn:
NBS differs from SPM in this respect. However, both design matrix formats will give exactly the same result. Note that for a one-way ANOVA with four groups, you could also use a design matrix in the NBS of the following form:
1 1 0 0
1 1 0 0
1 -1 1 0
1 -1 1 0
1 0 -1 1
1 0 -1 1
1 0 0 -1
1 0 0 -1
The contrast is then: [ 0 1 1 1].
Looking at the above design matrix, it can be seen that columns 2, 3 and 4 correspond to the rows of your SPM contrast matrix.
I hope this provides you with some intuition above the equivalence between them.
The initial threshold refers to the F-statistic (if you select an F-test). So if you use an F-test, you would typically select a larger threshold than if you were using a t-test.
Originally posted by Andreas Hahn:
Dear Andrew,
Thank you very much for providing the NBS toolbox!
I have a question regarding the contrast setting for a one-way ANOVA with four independent groups of subjects with the following design matrix:
1 0 0 0
1 0 0 0
0 1 0 0
0 1 0 0
0 0 1 0
0 0 1 0
0 0 0 1
0 0 0 1
(with more than two subjects in each group).
Can you please explain why the F-contrast according to the example in the help is [1 1 1 1] to test if any of the means is significantly different? Being used to SPM contrasts it appears to test if the sum of all 4 groups is different from zero and the following would be more intuitive for me to test the differences across groups [1 -1 0 0; 0 1 -1 0; 0 0 1 -1].
Another quick question: does the initial "threshold" refer to the F-value (or t-value respectively) for the initial mass univariate testing?
Thank you and best regards,
Andreas
Thank you very much for providing the NBS toolbox!
I have a question regarding the contrast setting for a one-way ANOVA with four independent groups of subjects with the following design matrix:
1 0 0 0
1 0 0 0
0 1 0 0
0 1 0 0
0 0 1 0
0 0 1 0
0 0 0 1
0 0 0 1
(with more than two subjects in each group).
Can you please explain why the F-contrast according to the example in the help is [1 1 1 1] to test if any of the means is significantly different? Being used to SPM contrasts it appears to test if the sum of all 4 groups is different from zero and the following would be more intuitive for me to test the differences across groups [1 -1 0 0; 0 1 -1 0; 0 0 1 -1].
Another quick question: does the initial "threshold" refer to the F-value (or t-value respectively) for the initial mass univariate testing?
Thank you and best regards,
Andreas
Threaded View
| Title | Author | Date |
|---|---|---|
| Andreas Hahn | Nov 14, 2014 | |
| Andrew Zalesky | Nov 16, 2014 | |
| Charanya Muralidharan | Nov 9, 2017 | |
| Andrew Zalesky | Nov 10, 2017 | |
| Charanya Muralidharan | Nov 13, 2017 | |
| Yuan-Fang Zhao | Oct 14, 2016 | |
| Andrew Zalesky | Oct 15, 2016 | |
| Andreas Hahn | Nov 18, 2014 | |
| Andrew Zalesky | Nov 19, 2014 | |