sdm-help-list > effect size estimation and CCMA
Showing 1-2 of 2 posts
Apr 27, 2015 02:04 PM | Nobody
effect size estimation and CCMA
Dear Joaquim,
I want to include a VBM study in a meta-analysis, which used non-parametric statistic (i.e. threshold free cluster enhancement). To adequately address the non-parametric approach, I did only include the coordinates of the significant voxels. The manual states that "the SDM software conducts a pre-analysis to provide an effect size for these peaks". Could you please elaborate or refer me to a manuscript, how this is actually achieved? I failed to find further information for this case and need to include this aspect in the methods section of our manuscript.
If conceivable, I would like to read your take on continuously cumulating meta-analyses in the field of neuroscience (Braver et al., 2014). We did previously publish a VBM-analysis (single study). Recent unpublished results of our own research group fail to replicate some of these findings. To avoid selective citations as well as the file-drawer problem, I want to combine SPMs of both studies as well as the above mentioned peaks of a third study in a meta-analysis to provide a more accurate estimation of gray matter abnormalities in this mental disorder. What's your opinion about such small meta-analyses?
All the best, (and thank you for AES-SDM),
Dirk
I want to include a VBM study in a meta-analysis, which used non-parametric statistic (i.e. threshold free cluster enhancement). To adequately address the non-parametric approach, I did only include the coordinates of the significant voxels. The manual states that "the SDM software conducts a pre-analysis to provide an effect size for these peaks". Could you please elaborate or refer me to a manuscript, how this is actually achieved? I failed to find further information for this case and need to include this aspect in the methods section of our manuscript.
If conceivable, I would like to read your take on continuously cumulating meta-analyses in the field of neuroscience (Braver et al., 2014). We did previously publish a VBM-analysis (single study). Recent unpublished results of our own research group fail to replicate some of these findings. To avoid selective citations as well as the file-drawer problem, I want to combine SPMs of both studies as well as the above mentioned peaks of a third study in a meta-analysis to provide a more accurate estimation of gray matter abnormalities in this mental disorder. What's your opinion about such small meta-analyses?
All the best, (and thank you for AES-SDM),
Dirk
May 4, 2015 09:05 AM | Joaquim Radua
RE: effect size estimation and CCMA
Dear Dirk,
I think that the best way to include a TFCE study is still the inclusion of its t-values map, or otherwise of the t-values of the peaks. However, as none of them are probably reported, you or SDM may have to "guess" the t-values of the peaks. In that case I would still suggest to derive them from the peak p-values, even if the latter were found using TFCE. Of course these t-values will be only an approximation, but they will probably better reflect the true peak t-values. I.e. a peak with a high t-value will probably have a small TFCE-found p-value, and thus the p-value-derived t-value will be high, whereas a peak with a low t-value will probably have a large TFCE-found p-value, and thus the p-value-derived t-value will be low. SDM, conversely, "would try its best" but would assign the same t-value to all peaks (you can read about the algorithm SDM uses here: http://dx.doi.org/10.1016/j.eurpsy.2011....).
My opinion about continuously cumulating meta-analyses is that they are probably complementary to standard meta-analyses. The latter, which involve careful human inspection of the studies, potential publication bias, modulators, and etcetera, are needed in order to provide the most reliable information. The former, conversely, may provide the most updated information.
Hope this helps,
Joaquim
I think that the best way to include a TFCE study is still the inclusion of its t-values map, or otherwise of the t-values of the peaks. However, as none of them are probably reported, you or SDM may have to "guess" the t-values of the peaks. In that case I would still suggest to derive them from the peak p-values, even if the latter were found using TFCE. Of course these t-values will be only an approximation, but they will probably better reflect the true peak t-values. I.e. a peak with a high t-value will probably have a small TFCE-found p-value, and thus the p-value-derived t-value will be high, whereas a peak with a low t-value will probably have a large TFCE-found p-value, and thus the p-value-derived t-value will be low. SDM, conversely, "would try its best" but would assign the same t-value to all peaks (you can read about the algorithm SDM uses here: http://dx.doi.org/10.1016/j.eurpsy.2011....).
My opinion about continuously cumulating meta-analyses is that they are probably complementary to standard meta-analyses. The latter, which involve careful human inspection of the studies, potential publication bias, modulators, and etcetera, are needed in order to provide the most reliable information. The former, conversely, may provide the most updated information.
Hope this helps,
Joaquim