help
help > RE: low suprathreshold t-value
Sep 14, 2021 10:09 AM | Andrew Zalesky
RE: low suprathreshold t-value
Hi YaeJi,
Here are some approaches to consider:
- Set the threshold based on you minimum meaningful effects size. I agree that a Cohen's d of 0.14 might be considered quite low, but this really depends on the context of your specific study. It may be that d=0.1 (or lower) is meaningful in your context. Hence, I don't think that t=2.3 is necessarily too low.
- Consider trying a range of thresholds. This will give you a sense of the spatial extent and strength of the correlation as a function of effect size. Of course the danger here is that trying many different thresholds could be construed as data dredging, particularly if you only report one of the those thresholds (i.e. cherry picking).
- Consider using the TFCE version of the NBS. The TFCE version does not require selection of a threshold.
- While your sample size is not super large, it may be enough to divide into test and validation cohorts. This would allow you to test whether any relationships can be replication in the validation cohort.
best wishes,
Andrew
Originally posted by YaeJi Kim:
Here are some approaches to consider:
- Set the threshold based on you minimum meaningful effects size. I agree that a Cohen's d of 0.14 might be considered quite low, but this really depends on the context of your specific study. It may be that d=0.1 (or lower) is meaningful in your context. Hence, I don't think that t=2.3 is necessarily too low.
- Consider trying a range of thresholds. This will give you a sense of the spatial extent and strength of the correlation as a function of effect size. Of course the danger here is that trying many different thresholds could be construed as data dredging, particularly if you only report one of the those thresholds (i.e. cherry picking).
- Consider using the TFCE version of the NBS. The TFCE version does not require selection of a threshold.
- While your sample size is not super large, it may be enough to divide into test and validation cohorts. This would allow you to test whether any relationships can be replication in the validation cohort.
best wishes,
Andrew
Originally posted by YaeJi Kim:
Dear NBS experts,
Hello, I am currently studying about DTI measures and its correlational relationship with behavioral score.
I constructed network for each subject and ran the NBS correlational analysis.
I have total of 240 subjects in this study, and want to use a result which was calculated with t=2.3, NBS FWE p < 0.05.
However, I am afraid that t = 2.3 in 240 subjects is somewhat less significant to report, and probably reviewers would suggest me to increase the initial suprathreshold.
I read few threads talking about presenting effect size (cohen's d), and mine is very low (0.14).
Should I not report the result or is there any other great reasoning for the low t-value?
Much appreciated if anyone gives me a any kind of feedback!
Bests,
YaeJi
Hello, I am currently studying about DTI measures and its correlational relationship with behavioral score.
I constructed network for each subject and ran the NBS correlational analysis.
I have total of 240 subjects in this study, and want to use a result which was calculated with t=2.3, NBS FWE p < 0.05.
However, I am afraid that t = 2.3 in 240 subjects is somewhat less significant to report, and probably reviewers would suggest me to increase the initial suprathreshold.
I read few threads talking about presenting effect size (cohen's d), and mine is very low (0.14).
Should I not report the result or is there any other great reasoning for the low t-value?
Much appreciated if anyone gives me a any kind of feedback!
Bests,
YaeJi
Threaded View
| Title | Author | Date |
|---|---|---|
| YaeJi Kim | Sep 14, 2021 | |
| Andrew Zalesky | Sep 14, 2021 | |
| YaeJi Kim | Sep 16, 2021 | |
| Andrew Zalesky | Sep 16, 2021 | |