help
help > RE: Design matrix and contrast for 2 way ANOVA, and f-value
Jun 10, 2019 06:06 AM | Andrew Zalesky
RE: Design matrix and contrast for 2 way ANOVA, and f-value
Hi Woomi,
1) Yes - these contrasts are correct for both the first and second design matrix.
However, the results for the main effect will be different between the first and second design. This is because you have included an interaction term in the design matrix. Therefore, when you talk about the main effect of Sex, you need to ask yourself whether you are interested in the main effect of Sex in the Drug or Water group. Because there is an interaction, the main effect of Sex will be different in the Drug group compared to the Water group. Which main effect of Sex do you want to consider? Drug or Water? The first design matrix will tell you the "average" main effect of sex across both the drug and water group. The second design matrix will tell you the main effect of sex in the drug group only.
Both design matrices are correct. And both contrasts are correct. You just need to interpret the main effect carefully in the presence of an interaction. This is a general property of the GLM and not specific to the NBS.
2) Both models are equally good. One model is not better than the other.
3) Technically yes. But I think that it would be ok if you just used 0.05 as your p-value threshold.
4) The degrees of freedom are only based on the number of subjects. The number of nodes does not affect the degrees of freedom. The degrees of freedom is approximately the square root of the number of subjects.
Andrew
Originally posted by Woomi Ban:
1) Yes - these contrasts are correct for both the first and second design matrix.
However, the results for the main effect will be different between the first and second design. This is because you have included an interaction term in the design matrix. Therefore, when you talk about the main effect of Sex, you need to ask yourself whether you are interested in the main effect of Sex in the Drug or Water group. Because there is an interaction, the main effect of Sex will be different in the Drug group compared to the Water group. Which main effect of Sex do you want to consider? Drug or Water? The first design matrix will tell you the "average" main effect of sex across both the drug and water group. The second design matrix will tell you the main effect of sex in the drug group only.
Both design matrices are correct. And both contrasts are correct. You just need to interpret the main effect carefully in the presence of an interaction. This is a general property of the GLM and not specific to the NBS.
2) Both models are equally good. One model is not better than the other.
3) Technically yes. But I think that it would be ok if you just used 0.05 as your p-value threshold.
4) The degrees of freedom are only based on the number of subjects. The number of nodes does not affect the degrees of freedom. The degrees of freedom is approximately the square root of the number of subjects.
Andrew
Originally posted by Woomi Ban:
Hi Dr.Zalesky,
Thanks for the quick answer and detail explanation.
It is very helpful to understand your tool. Base on your answer, I tried to find reference to other comments in the forum, but still have several questions.
Sorry for too many questions.
1. You suggested two different design before. My understanding is ...
1) For the main effect of Sex or Exposure,
I can use 1st design matrix and contrast will be [0 1/-1 0 0] for Sex and [0 0 1/-1 0] exposure.
Can I get the both main effect from the 2nd design matrix, too?
If so, how can I put the contrast for main effect of Exposure using the 2nd design matrix?
2) If I modeled 'The connectivity will be increased in Drug and Female', then which design is better?
2. t-test will return a one-sided p-value.
1) I'm trying to get significant level of data(p<0.05) using t-test.
Then, do I need to put 0.025 for Significance column in NBS tool for each independt test(contrast [0 0 0 1] or [0 0 0 -1])?
3. Threshold
1) When the t-value is calculated which factors are used for degree of freedom?
Is it include nummer of subjects and nodes? Or only number of subjects?
Thanks again!
Woomi
Thanks for the quick answer and detail explanation.
It is very helpful to understand your tool. Base on your answer, I tried to find reference to other comments in the forum, but still have several questions.
Sorry for too many questions.
1. You suggested two different design before. My understanding is ...
1) For the main effect of Sex or Exposure,
I can use 1st design matrix and contrast will be [0 1/-1 0 0] for Sex and [0 0 1/-1 0] exposure.
Can I get the both main effect from the 2nd design matrix, too?
If so, how can I put the contrast for main effect of Exposure using the 2nd design matrix?
2) If I modeled 'The connectivity will be increased in Drug and Female', then which design is better?
2. t-test will return a one-sided p-value.
1) I'm trying to get significant level of data(p<0.05) using t-test.
Then, do I need to put 0.025 for Significance column in NBS tool for each independt test(contrast [0 0 0 1] or [0 0 0 -1])?
3. Threshold
1) When the t-value is calculated which factors are used for degree of freedom?
Is it include nummer of subjects and nodes? Or only number of subjects?
Thanks again!
Woomi
Threaded View
| Title | Author | Date |
|---|---|---|
| Woomi Ban | May 30, 2019 | |
| davideit7 | Dec 5, 2019 | |
| Andrew Zalesky | Dec 6, 2019 | |
| davideit7 | Dec 6, 2019 | |
| Andrew Zalesky | Dec 7, 2019 | |
| davideit7 | Jan 20, 2020 | |
| Andrew Zalesky | Jan 21, 2020 | |
| davideit7 | Jan 22, 2020 | |
| Andrew Zalesky | Jan 23, 2020 | |
| Woomi Ban | Jun 10, 2019 | |
| Andrew Zalesky | Jun 10, 2019 | |
| Andrew Zalesky | Jun 1, 2019 | |