help
help > RE: Second level random effects analysis
Jun 8, 2015 05:06 PM | Alfonso Nieto-Castanon - Boston University
RE: Second level random effects analysis
Hi Kaitlin,
"Import values" will store those connectivity values for each condition and for each subject as new second-level covariates into CONN. If you would prefer to have them stored into a .txt file you could:
a) when selecting 'display values' check the GUI element that says something like 'launch REX gui'
b) on the new REX gui select the option that reads 'save REX.mat file' and change that to 'save REX.mat and output files' and then select the 'data.txt' file option.
c) click on the 'Extract' button, and that will extract those connectivity values for each subject and each condition and place that info into a text file.
Regarding the confidence intervals, those are not being stored anywhere (sorry about that) but they could be easily computed from your subject-specific values. In particular, the 90% errorbar lengths for each condition can be computed as 2*spm_invTcdf(.95,N-1)*std(x)/sqrt(N) where x is your subject by condition matrix and N is the number of subjects.
Hope this helps
Alfonso
ps. As a side note in terms of the interpretation of these errorbars, these errorbars constructed with the formula above only capture the between-subjects variance separately for each condition (i.e. they are reflective of the significance of an associated one-sample t-test testing each condition value against a zero-mean null hypothesis), but not the repeated-measures between-conditions variance (i.e. they are NOT reflective of the significance of an associated paired t-test comparing each pair of conditions against each other). This means that if an errorbar does not overlap the x-axis then that means that you would get a p<.05 value when testing the average effect of that condition using a one-sample t-test, but you may have two conditions with overlapping errorbars/effect-sizes which still show a significant between-condition difference when testing the difference between those conditions using a paired t-test. I thought I wold mention because that is a typical source of confusion regarding the interpretation of error bars in the context of repeated-measure analyses.
Originally posted by Kaitlin Cassady:
"Import values" will store those connectivity values for each condition and for each subject as new second-level covariates into CONN. If you would prefer to have them stored into a .txt file you could:
a) when selecting 'display values' check the GUI element that says something like 'launch REX gui'
b) on the new REX gui select the option that reads 'save REX.mat file' and change that to 'save REX.mat and output files' and then select the 'data.txt' file option.
c) click on the 'Extract' button, and that will extract those connectivity values for each subject and each condition and place that info into a text file.
Regarding the confidence intervals, those are not being stored anywhere (sorry about that) but they could be easily computed from your subject-specific values. In particular, the 90% errorbar lengths for each condition can be computed as 2*spm_invTcdf(.95,N-1)*std(x)/sqrt(N) where x is your subject by condition matrix and N is the number of subjects.
Hope this helps
Alfonso
ps. As a side note in terms of the interpretation of these errorbars, these errorbars constructed with the formula above only capture the between-subjects variance separately for each condition (i.e. they are reflective of the significance of an associated one-sample t-test testing each condition value against a zero-mean null hypothesis), but not the repeated-measures between-conditions variance (i.e. they are NOT reflective of the significance of an associated paired t-test comparing each pair of conditions against each other). This means that if an errorbar does not overlap the x-axis then that means that you would get a p<.05 value when testing the average effect of that condition using a one-sample t-test, but you may have two conditions with overlapping errorbars/effect-sizes which still show a significant between-condition difference when testing the difference between those conditions using a paired t-test. I thought I wold mention because that is a typical source of confusion regarding the interpretation of error bars in the context of repeated-measure analyses.
Originally posted by Kaitlin Cassady:
Hi Alfonso,
This is exactly what I was after - thanks so much for your help! I noticed that I am able to explore these effect sizes further with "import values" in which I can look at the effect sizes for each subject. Is there a way to export these values for each subject? In addition, is there a way to export the standard errors for each of the 7 time points (the values that produce the error bars in the plot)?
Thanks again!
This is exactly what I was after - thanks so much for your help! I noticed that I am able to explore these effect sizes further with "import values" in which I can look at the effect sizes for each subject. Is there a way to export these values for each subject? In addition, is there a way to export the standard errors for each of the 7 time points (the values that produce the error bars in the plot)?
Thanks again!
Threaded View
| Title | Author | Date |
|---|---|---|
| Kaitlin Cassady | May 30, 2015 | |
| Alfonso Nieto-Castanon | May 30, 2015 | |
| Kaitlin Cassady | Jun 4, 2015 | |
| Alfonso Nieto-Castanon | Jun 4, 2015 | |
| Kaitlin Cassady | Jun 5, 2015 | |
| Alfonso Nieto-Castanon | Jun 8, 2015 | |
| Kevin Mann | Jun 16, 2016 | |
| Kaitlin Cassady | Jun 8, 2015 | |
| Alfonso Nieto-Castanon | Jun 9, 2015 | |
| Kaitlin Cassady | Jun 9, 2015 | |