help > Reporting Results T-Value
Showing 1-5 of 5 posts
Mar 28, 2017 09:03 PM | Jennifer Siegel - University of Edinburgh
Reporting Results T-Value
Hello,
I just want to clarify something in regards to reporting my results. How do I obtain the individual T values for each significant cluster?
I notice there is a T-value listed at the top of the results page, but this is the same for all the clusters. I assume that this isn't what I want to report.
When I go into the "display values option", I see a cluster T-value, but the cluster p-FDR result is different to the results cluster p-FDR (and the order of my clusters changes slightly). Which one do I report and why are these different?
Thanks,
Jennifer
Disregard my attachment, it wouldn't let me delete it.
I just want to clarify something in regards to reporting my results. How do I obtain the individual T values for each significant cluster?
I notice there is a T-value listed at the top of the results page, but this is the same for all the clusters. I assume that this isn't what I want to report.
When I go into the "display values option", I see a cluster T-value, but the cluster p-FDR result is different to the results cluster p-FDR (and the order of my clusters changes slightly). Which one do I report and why are these different?
Thanks,
Jennifer
Disregard my attachment, it wouldn't let me delete it.
Mar 29, 2017 02:03 AM | Alfonso Nieto-Castanon - Boston University
RE: Reporting Results T-Value
Hi Jennifer,
Typically you would just report the combination of voxel-level "height" threshold (some times also referred to as "cluster-defining" or "cluster-forming" threshold; voxel-level uncorrected p<.001) and the cluster-level "extent" threshold (e.g. cluster-level p-FDR<.05) used to identify significant clusters. In the results explorer window these are the two p-value thresholds specified at the top of this figure.
It is not a bad idea to also report for each individual cluster a measure of its location (e.g. the MNI coordinates of the peak voxel, listed in the mid-figure list under the "(x,y,z)" column) as well as measures of each cluster size (listed under "size") and associated cluster-size p-values (e.g. under "size p-FDR" or "size p-FWE" in the same list).
Last, it is best if you can also report measures of effect-size within each cluster. These depend on the nature of your analyses, for example if you are looking at differences in connectivity between two groups it is a good idea to report within each cluster the average connectivity values within each group and/or the differences in connectivity between the two groups. For these you typically would go to "plot effects" and report the values listed under "beta" in the resulting plots (but this depends a bit on your specific analyses)
Hope this helps
Alfonso
Originally posted by Jennifer Siegel :
Typically you would just report the combination of voxel-level "height" threshold (some times also referred to as "cluster-defining" or "cluster-forming" threshold; voxel-level uncorrected p<.001) and the cluster-level "extent" threshold (e.g. cluster-level p-FDR<.05) used to identify significant clusters. In the results explorer window these are the two p-value thresholds specified at the top of this figure.
It is not a bad idea to also report for each individual cluster a measure of its location (e.g. the MNI coordinates of the peak voxel, listed in the mid-figure list under the "(x,y,z)" column) as well as measures of each cluster size (listed under "size") and associated cluster-size p-values (e.g. under "size p-FDR" or "size p-FWE" in the same list).
Last, it is best if you can also report measures of effect-size within each cluster. These depend on the nature of your analyses, for example if you are looking at differences in connectivity between two groups it is a good idea to report within each cluster the average connectivity values within each group and/or the differences in connectivity between the two groups. For these you typically would go to "plot effects" and report the values listed under "beta" in the resulting plots (but this depends a bit on your specific analyses)
Hope this helps
Alfonso
Originally posted by Jennifer Siegel :
Hello,
I just want to clarify something in regards to reporting my results. How do I obtain the individual T values for each significant cluster?
I notice there is a T-value listed at the top of the results page, but this is the same for all the clusters. I assume that this isn't what I want to report.
When I go into the "display values option", I see a cluster T-value, but the cluster p-FDR result is different to the results cluster p-FDR (and the order of my clusters changes slightly). Which one do I report and why are these different?
Thanks,
Jennifer
Disregard my attachment, it wouldn't let me delete it.
I just want to clarify something in regards to reporting my results. How do I obtain the individual T values for each significant cluster?
I notice there is a T-value listed at the top of the results page, but this is the same for all the clusters. I assume that this isn't what I want to report.
When I go into the "display values option", I see a cluster T-value, but the cluster p-FDR result is different to the results cluster p-FDR (and the order of my clusters changes slightly). Which one do I report and why are these different?
Thanks,
Jennifer
Disregard my attachment, it wouldn't let me delete it.
Mar 31, 2017 03:03 PM | Jennifer Siegel - University of Edinburgh
RE: Reporting Results T-Value
Hi,
Thanks for getting back about this and all of it was completely clear. So, you wouldn't suggest including the T-values? When I just quickly looked, I noticed that there were several papers which had also included T-values in their results.
Also, what is the plot effects table reporting? I notice that the p values are different from the results window.
Also, regarding reported the effect size. Would I still be reporting beta when doing a one-sample T test (for the full group and seed region)? Just wondering if anyone has any references on reporting effect size or papers that have done this?
Thanks,
Jennifer
Thanks for getting back about this and all of it was completely clear. So, you wouldn't suggest including the T-values? When I just quickly looked, I noticed that there were several papers which had also included T-values in their results.
Also, what is the plot effects table reporting? I notice that the p values are different from the results window.
Also, regarding reported the effect size. Would I still be reporting beta when doing a one-sample T test (for the full group and seed region)? Just wondering if anyone has any references on reporting effect size or papers that have done this?
Thanks,
Jennifer
Mar 31, 2017 06:03 PM | Alfonso Nieto-Castanon - Boston University
RE: Reporting Results T-Value
Hi Jennifer,
Yes, often times people also report peak-voxel statistics (e.g. T-values) and that is of course perfectly fine. The "plot effects' display is not showing you that, though, but rather it is performing ROI-level analyses on each cluster and displaying the associated statistics for those aggregated values. These ROI-level analyses and results are often useful for interpreting your results, displaying effect-sizes, and/or for post-hoc analyses. Note nevertheless that these post-hoc ROI-level results should never be used in lieu of the original cluster-level stats for hypothesis testing or inferences regarding your original analysis/test (if used without care post-hoc ROI-level stats are likely to contain non-trivial biases due to non-independence / double-dipping).
In your case, for a one-sample t-test, I would probably report:
a) the cluster-level statistics (e.g. position, size, and size p-FDR within each cluster) reported in the main "results explorer" window
and b) the beta values within each cluster reported in the "plot effects" graph/results (representing the average Fisher-transformed correlations with the seed within each cluster)
The T-values reported in the "plot effects" graph/results in your case represent the statistics associated with the average connectivity within each cluster. These are very informative for meta-analysis purposes but they contain expected biases so neither these T-values nor the associated p-values in these "plot effects" graph/results should be used for inferences regarding your individual cluster (in other words, it is fine to report them but it is not ok to use these p-values as basis to determine whether each cluster is "significantly" connected with your seed).
In case this helps, a simple/general and very good reference for best-practice reporting guidelines is Poldrack et al 2008 "Guidelines for reporting an fMRI study"
Best
Alfonso
Originally posted by Jennifer Siegel :
Yes, often times people also report peak-voxel statistics (e.g. T-values) and that is of course perfectly fine. The "plot effects' display is not showing you that, though, but rather it is performing ROI-level analyses on each cluster and displaying the associated statistics for those aggregated values. These ROI-level analyses and results are often useful for interpreting your results, displaying effect-sizes, and/or for post-hoc analyses. Note nevertheless that these post-hoc ROI-level results should never be used in lieu of the original cluster-level stats for hypothesis testing or inferences regarding your original analysis/test (if used without care post-hoc ROI-level stats are likely to contain non-trivial biases due to non-independence / double-dipping).
In your case, for a one-sample t-test, I would probably report:
a) the cluster-level statistics (e.g. position, size, and size p-FDR within each cluster) reported in the main "results explorer" window
and b) the beta values within each cluster reported in the "plot effects" graph/results (representing the average Fisher-transformed correlations with the seed within each cluster)
The T-values reported in the "plot effects" graph/results in your case represent the statistics associated with the average connectivity within each cluster. These are very informative for meta-analysis purposes but they contain expected biases so neither these T-values nor the associated p-values in these "plot effects" graph/results should be used for inferences regarding your individual cluster (in other words, it is fine to report them but it is not ok to use these p-values as basis to determine whether each cluster is "significantly" connected with your seed).
In case this helps, a simple/general and very good reference for best-practice reporting guidelines is Poldrack et al 2008 "Guidelines for reporting an fMRI study"
Best
Alfonso
Originally posted by Jennifer Siegel :
Hi,
Thanks for getting back about this and all of it was completely clear. So, you wouldn't suggest including the T-values? When I just quickly looked, I noticed that there were several papers which had also included T-values in their results.
Also, what is the plot effects table reporting? I notice that the p values are different from the results window.
Also, regarding reported the effect size. Would I still be reporting beta when doing a one-sample T test (for the full group and seed region)? Just wondering if anyone has any references on reporting effect size or papers that have done this?
Thanks,
Jennifer
Thanks for getting back about this and all of it was completely clear. So, you wouldn't suggest including the T-values? When I just quickly looked, I noticed that there were several papers which had also included T-values in their results.
Also, what is the plot effects table reporting? I notice that the p values are different from the results window.
Also, regarding reported the effect size. Would I still be reporting beta when doing a one-sample T test (for the full group and seed region)? Just wondering if anyone has any references on reporting effect size or papers that have done this?
Thanks,
Jennifer
Apr 14, 2017 03:04 PM | Jennifer Siegel - University of Edinburgh
RE: Reporting Results T-Value
Thanks again, that is really helpful.
One final and probably obvious question (sorry). Where do I get the peak voxel T value for each significant? I would still like to include this in my data (I am doing some group comparisons). I see a T-value at the top of the results explorer window, but I don't see a voxel specific T max. How would I obtain this? Sorry, there is probably a very obvious answer to this question.
Jennifer
One final and probably obvious question (sorry). Where do I get the peak voxel T value for each significant? I would still like to include this in my data (I am doing some group comparisons). I see a T-value at the top of the results explorer window, but I don't see a voxel specific T max. How would I obtain this? Sorry, there is probably a very obvious answer to this question.
Jennifer