关于两样本t检验统计的AlphaSim校正问题

Submitted by NCU_Zhou on Thu, 04/10/2014 - 17:10

 关于两样本t检验统计的AlphaSim校正问题，想向各位老师请教：

Reviewer给我的Comments是：

- The AlphaSim package within AFNI allows for both family-wise error correction or false discovery rate tests in multiple comparisons.  Which statistical method was used and did this limit the number of false positive voxels detected?

 
我的原统计描述是：
a two-sample t-test was performed to elucidate the ALFF differences between the two groups. The statistical significance was determined with Monte Carlo simulation (AlphaSim corrected, single voxel P = 0.05, FWHM = 6 mm , 10,000 simulations,) combined with cluster size ≥ 165 voxels. 

您看我这样回是否可行：
This correction was conducted using the AlphaSim program embedded into the REST Software (http://www.restfmri.net/forum/REST_V1.8), which applied Monte Carlo simulation to calculate the probability of false positive detection by considering both the individual voxel probability threshold and cluster size.
It can be the case that the results required cluster sizes (≥ 165 voxels) maybe miss some smaller volume in the statistical analysis with AlphaSim corrected. In fact, only three minor clusters with lower sizes (< 34 voxels) were discarded in this study.


我想AlphaSim和FWE校正还是有不一样的，AlphaSim是考虑了“ false positive voxels detected”吧。不知道我理解是否正确？

 
谢谢！
 
顺祝春祺！

Submitted by admin on Tue, 05/27/2014 - 01:30

 I guess the reviewer may want you to report a smaller alpha, since you have a very big cluster (165 voxels), whose false positive voxels maybe more.
In AFNI, there is a FDR q value, which can also be helped to control false positive voxels. 

Only my personal view, I don't know if smaller alpha is a proper choice. We can discuss about it in this forum.