sdm-help-list
89 Subscribers 
sdm-help-list > A question about the image result
Showing 1-2 of 2 posts
Apr 12, 2018 11:04 AM | Yangqian Shen - Psychology Institute, Chinese Academcy of Science
A question about the image result
Dear SDM fellows,
I have a question regarding of the result viewing. I got an image that showed significant brain regions. But there was a cluster with a big hole (looked like a circle) in the middle of it. The area of the hole was not significant but the surroundings. I was wondering is it for some reasons or just how it should be.
Thank you very much!
Jane Shen
I have a question regarding of the result viewing. I got an image that showed significant brain regions. But there was a cluster with a big hole (looked like a circle) in the middle of it. The area of the hole was not significant but the surroundings. I was wondering is it for some reasons or just how it should be.
Thank you very much!
Jane Shen
Apr 17, 2018 11:04 AM | Joaquim Radua
RE: A question about the image result
Dear Jane,
I think that this may be an issue related to the use of random-effects models. Specifically, in the center of the cluster the estimated heterogeneity might be larger, and thus the z-value might be occasionally lower.
This is not a problem of SDM, but of random-effects meta-analysis, as you can check in R:
However, this is a minor problem as compared to the multiple advantages of using random-effects. See e.g. https://doi.org/10.3389/fnins.2017.00745
With the best wishes,
Joaquim
I think that this may be an issue related to the use of random-effects models. Specifically, in the center of the cluster the estimated heterogeneity might be larger, and thus the z-value might be occasionally lower.
This is not a problem of SDM, but of random-effects meta-analysis, as you can check in R:
library(metafor)
rma(c(0,0,0,0,0,0,0,0,2,3),rep(1,10))
rma(c(0,0,0,0,0,0,0,0,3,3),rep(1,10)) # The arithmetic mean is larger but the estimate and z-value is lower
rma(c(0,0,0,0,0,0,0,0,2,3),rep(1,10))
rma(c(0,0,0,0,0,0,0,0,3,3),rep(1,10)) # The arithmetic mean is larger but the estimate and z-value is lower
However, this is a minor problem as compared to the multiple advantages of using random-effects. See e.g. https://doi.org/10.3389/fnins.2017.00745
With the best wishes,
Joaquim