help > How to find a reasonable threshold when performing NBS analyses?
Showing 1-3 of 3 posts
Sep 27, 2021 05:09 PM | Ting Qiu
How to find a reasonable threshold when performing NBS analyses?
Hi Andrew,
NBS is really great! Thank you for making it to the public.
Now I am using NBS to examine structural connection differences among four groups (which means that i need to do 6 pairs of comparisons).
I checked the paper [1] and found the relationship between effect size d and t values: d=t *sqrt(1/n1+1/n2),which means that I could determine a threshold for NBS; but for the effect size, there seems that no criteria but I knew the small effect size (0.2), the medium effect size (0.5) and large effect size (0.8)... my question is: could you give me some suggestions about how to determine the threshold? and are there any papers talking about this that you would recommend?
Another question is: I have four groups, do you think comparing each pairs of the four groups (ttest) in terms of connection differences is a good way?
Thank you very much and looking forward to your reply.
All the best,
Ting
NBS is really great! Thank you for making it to the public.
Now I am using NBS to examine structural connection differences among four groups (which means that i need to do 6 pairs of comparisons).
I checked the paper [1] and found the relationship between effect size d and t values: d=t *sqrt(1/n1+1/n2),which means that I could determine a threshold for NBS; but for the effect size, there seems that no criteria but I knew the small effect size (0.2), the medium effect size (0.5) and large effect size (0.8)... my question is: could you give me some suggestions about how to determine the threshold? and are there any papers talking about this that you would recommend?
Another question is: I have four groups, do you think comparing each pairs of the four groups (ttest) in terms of connection differences is a good way?
Thank you very much and looking forward to your reply.
All the best,
Ting
Sep 27, 2021 11:09 PM | Andrew Zalesky
RE: How to find a reasonable threshold when performing NBS analyses?
Hi Ting,
Comparing each pair of groups separately is ok. Alternatively, you may want to consider an ANOVA (F-test) to test whether any of the groups differ from each other. This would reduce the number of multiple comparisons.
It is difficult to provide a concrete recommendation about the threshold without further knowledge of your project. Setting the threshold based on a minimum effect size of interest sounds reasonable. There is no right or wrong threshold.
Andrew
Originally posted by Ting Qiu:
Comparing each pair of groups separately is ok. Alternatively, you may want to consider an ANOVA (F-test) to test whether any of the groups differ from each other. This would reduce the number of multiple comparisons.
It is difficult to provide a concrete recommendation about the threshold without further knowledge of your project. Setting the threshold based on a minimum effect size of interest sounds reasonable. There is no right or wrong threshold.
Andrew
Originally posted by Ting Qiu:
Hi Andrew,
NBS is really great! Thank you for making it to the public.
Now I am using NBS to examine structural connection differences among four groups (which means that i need to do 6 pairs of comparisons).
I checked the paper [1] and found the relationship between effect size d and t values: d=t *sqrt(1/n1+1/n2),which means that I could determine a threshold for NBS; but for the effect size, there seems that no criteria but I knew the small effect size (0.2), the medium effect size (0.5) and large effect size (0.8)... my question is: could you give me some suggestions about how to determine the threshold? and are there any papers talking about this that you would recommend?
Another question is: I have four groups, do you think comparing each pairs of the four groups (ttest) in terms of connection differences is a good way?
Thank you very much and looking forward to your reply.
All the best,
Ting
NBS is really great! Thank you for making it to the public.
Now I am using NBS to examine structural connection differences among four groups (which means that i need to do 6 pairs of comparisons).
I checked the paper [1] and found the relationship between effect size d and t values: d=t *sqrt(1/n1+1/n2),which means that I could determine a threshold for NBS; but for the effect size, there seems that no criteria but I knew the small effect size (0.2), the medium effect size (0.5) and large effect size (0.8)... my question is: could you give me some suggestions about how to determine the threshold? and are there any papers talking about this that you would recommend?
Another question is: I have four groups, do you think comparing each pairs of the four groups (ttest) in terms of connection differences is a good way?
Thank you very much and looking forward to your reply.
All the best,
Ting
Sep 28, 2021 01:09 PM | Ting Qiu
RE: How to find a reasonable threshold when performing NBS analyses?
Hi Andrew,
Thank you so much for your reply.
Have a great one!
Ting
Thank you so much for your reply.
Have a great one!
Ting