help
help > RE: Sessions with different durations
Aug 12, 2017 12:08 AM | Alfonso Nieto-Castanon - Boston University
RE: Sessions with different durations
Dear Ella,
I would suggest defining covariates that encode the number of datapoints (or perhaps the square root of the number of datapoints) per condition, and then using those covariates as control variables (or covariates of interest, depending on your focus) in your second-level analyses.
In general second-level GLM analyses are typically considered robust against heterscedasticity violations (i.e. the presence of different standard errors between conditions/sessions/subjects, for example those associated with different number of timepoints), but there could well be also temporal effects that may introduce some biases in connectivity measures obtained at different temporal scales. If you are mainly concerned about potential heteroscedasticity violations then I would probably suggest using linear mixed model designs (e.g in FSL), while if you are mostly concerned about potential biases in connectivity introduced by the different temporal scales analyzed then using the above-mentioned control covariates should work perfectly fine.
Best
Alfonso
Originally posted by Ella Gabitov:
I would suggest defining covariates that encode the number of datapoints (or perhaps the square root of the number of datapoints) per condition, and then using those covariates as control variables (or covariates of interest, depending on your focus) in your second-level analyses.
In general second-level GLM analyses are typically considered robust against heterscedasticity violations (i.e. the presence of different standard errors between conditions/sessions/subjects, for example those associated with different number of timepoints), but there could well be also temporal effects that may introduce some biases in connectivity measures obtained at different temporal scales. If you are mainly concerned about potential heteroscedasticity violations then I would probably suggest using linear mixed model designs (e.g in FSL), while if you are mostly concerned about potential biases in connectivity introduced by the different temporal scales analyzed then using the above-mentioned control covariates should work perfectly fine.
Best
Alfonso
Originally posted by Ella Gabitov:
Dear all,
I want to compare functional connectivity between sessions with different durations. In my current data set, the number of data-points varies not only between sessions but also between subjects. I would prefer to use all data rather than reducing it to the minimum number of available data-points and would be glad to know what do you think about it. Instead of disregarding data-points that exceeds the minimum, is there any other way to control for potential differences due to different standard errors between the sessions and subjects?
Thank you in advance,
Ella
I want to compare functional connectivity between sessions with different durations. In my current data set, the number of data-points varies not only between sessions but also between subjects. I would prefer to use all data rather than reducing it to the minimum number of available data-points and would be glad to know what do you think about it. Instead of disregarding data-points that exceeds the minimum, is there any other way to control for potential differences due to different standard errors between the sessions and subjects?
Thank you in advance,
Ella
Threaded View
| Title | Author | Date |
|---|---|---|
| Ella Gabitov | Aug 11, 2017 | |
| Alfonso Nieto-Castanon | Aug 12, 2017 | |
| Ella Gabitov | Aug 14, 2017 | |