help > Very high effect sizes
Showing 1-4 of 4 posts
Display:
Results per page:
Mar 26, 2025  01:03 PM | Erfan Naghavi - Tehran University of Medical Sciences
Very high effect sizes

Dear NBS Team,


Thank you for creating such an excellent tool. I am facing an issue that I believe you can help me resolve. I have computed a 22 × 22 functional connectivity matrix for a group of 850 participants (168 vs. 682). However, after applying NBS with a contrast array of [1, -1, 0, 0, 0, 0], I am observing exceptionally large effect sizes, reaching values as high as 11, 12, and even 14.


According to the NBS reference manual, the primary threshold is typically set between 2 and 4. However, using such a low threshold in my case results in an overly dense connectivity network, leading to unrealistic findings.


Given my situation, what would be an appropriate primary threshold, and how can I determine it? Would setting it to 10, 8, or 12 be reasonable? Additionally, what might be causing these unusually high effect sizes in my analysis?


Looking forward to your insights.


Best regards,
Erfan

Mar 27, 2025  10:03 AM | Andrew Zalesky
RE: Very high effect sizes

Hi Erfan, 


These effect sizes seem unusually high for Cohen's d. 


Of course, it is possibe that you may have a very strong effect in your comparison and it is difficult to assess this without know the specifics of your experiment. 


You can asjust the primary threshold as you wish. The range of 2-4 is simply a recommendation. 


 


Originally posted by Erfan Naghavi:



Dear NBS Team,


Thank you for creating such an excellent tool. I am facing an issue that I believe you can help me resolve. I have computed a 22 × 22 functional connectivity matrix for a group of 850 participants (168 vs. 682). However, after applying NBS with a contrast array of [1, -1, 0, 0, 0, 0], I am observing exceptionally large effect sizes, reaching values as high as 11, 12, and even 14.


According to the NBS reference manual, the primary threshold is typically set between 2 and 4. However, using such a low threshold in my case results in an overly dense connectivity network, leading to unrealistic findings.


Given my situation, what would be an appropriate primary threshold, and how can I determine it? Would setting it to 10, 8, or 12 be reasonable? Additionally, what might be causing these unusually high effect sizes in my analysis?


Looking forward to your insights.


Best regards,
Erfan



 

Mar 27, 2025  10:03 AM | Erfan Naghavi - Tehran University of Medical Sciences
RE: Very high effect sizes

Thank you for your reply! I recently found an equation that defines the primary threshold of an F-test in terms of Cohen's d: F = d^2 * (n₁*n₂/(n₁+n₂). 


I was wondering if I could use a desired effect size to determine a primary threshold with this equation and then apply this threshold in NBS as an F-test. Is it OK?


Best regards,
Erfan

Mar 27, 2025  11:03 PM | Andrew Zalesky
RE: Very high effect sizes

Yes - this sounds reasonable.


Originally posted by Erfan Naghavi:



Thank you for your reply! I recently found an equation that defines the primary threshold of an F-test in terms of Cohen's d: F = d^2 * (n₁*n₂/(n₁+n₂). 


I was wondering if I could use a desired effect size to determine a primary threshold with this equation and then apply this threshold in NBS as an F-test. Is it OK?


Best regards,
Erfan