[color=#000000]Dear Pedro,[/color]
[color=#000000]Sorry that was confusing. You are right that there might not be obvious scenarios when you would want to use the 'copy task-related conditions to covariate list' option, since, as you state, those hrf-convolved condition regressors will already appear by default simply encoded as 'effect of [condition]' during Denoising and future steps, so copying there as a first-level covariate does not seem to add much if anything at all.[/color]
[color=#000000]In general, anything that is defined as a [b]condition[/b] will be hrf-convolved (assuming you have specified a continuous acquisition) and the resulting timeseries will be: a) shown in the default list of potential confounding effects in the [i]Denoising[/i] tab; and b) be used as weights in the [i]first-level[/i] tab for weighted-GLM or gPPI analyses for the estimation of condition-specific connectivity measures. In contrast, anything that is defined as a [b]first-level covariate[/b] will be: a) shown (as is, no hrf-convolution) in the default list of potential confounding effects in the [i]Denoising [/i]tab; b) appear as potential seed timeseries in the [i]first-level analysis [/i]tab (only those covariates not selected as confounding effects during the denoising step); and c) appear as potential interaction terms, both in the [i]Setup.Conditions [/i]tab as well as in the [i]first-level analysis [/i]tab for the "other temporal-modulation effects" analysis type.[/color]
[color=#000000]There are, of course, some somewhat-convoluted scenarios when you would want to genuinely use the 'copy task-related conditions...' option such as: a) for display purposes (for example, having the hrf-convolved conditions included in the list of first-level covariates allows you to include them in some plots such as QA which otherwise would need to be created manually; similarly if you want to use the 'covariate tools' gui to compute some summary measure of your conditions); or b) for more complex interaction analyses (e.g. CONN allows you define condition*covariate interactions, so sometimes it is useful to copy some subset of conditions into first-level covariates just to be able then to use the resulting timeseries as interaction terms). [/color]
[color=#000000]In practice, though, the most common use of this '[b]copy[/b] task-relate conditions...' option is a soft way to perform the '[b]move[/b] task-related conditions to covariate list' option in two steps (i.e. first use the 'copy ...' option, then, if everything looks fine, simply delete the original conditions). The 'move task-related conditions...' option is useful, as stated in the manual, when you want to perform Fair et al. -style analyses, where you still want to regress out anything that correlates with your conditions from the BOLD signal but you do not want to obtain condition-specific connectivity measures. In that case, moving a condition into a first-level covariate does exactly that, it still shows you the appropriate timeseries during the [i]Denoising [/i]step so the appropriate timeseries can still be included as a confounding effect, but it is no longer treated as a condition so CONN does not estimate condition-specific connectivity measures. [/color]
[color=#000000]Let me know if that clarifies[/color]
[color=#000000]Alfonso[/color]
[i]Originally posted by Pedro Valdes-Hernandez:[/i][quote]Dear CONN experts,
I'd like to know why one would want to copy task-related conditions to first level covariates.
Aren't these regressed out during the temporal preprocessing (denoising) anyway?
The original CONN paper (2012) suggests these effects are indeed removed in the Denoising step. It appears so since the conditions are imported as confounds with the name 'Effect of...'. I guess this is done to obtain "resting state" task-independent FC measures, as in Fair et al (2007).
However in the CONN User Manual states that, in order to achieve this, the conditions must be copied to the 1st level covariate list. Is this correct?
This is confusing. In a nutshell, what is the purpose of this 1st level covariate list, other than to provide regressors not to be HRF-convolved (like in SPM)?
On the other hand, is the HRF-free regression used to remove task effects or just HRF convolved conditions?
Looking forward any comment on this.
Pedro[/quote]
[color=#000000]Sorry that was confusing. You are right that there might not be obvious scenarios when you would want to use the 'copy task-related conditions to covariate list' option, since, as you state, those hrf-convolved condition regressors will already appear by default simply encoded as 'effect of [condition]' during Denoising and future steps, so copying there as a first-level covariate does not seem to add much if anything at all.[/color]
[color=#000000]In general, anything that is defined as a [b]condition[/b] will be hrf-convolved (assuming you have specified a continuous acquisition) and the resulting timeseries will be: a) shown in the default list of potential confounding effects in the [i]Denoising[/i] tab; and b) be used as weights in the [i]first-level[/i] tab for weighted-GLM or gPPI analyses for the estimation of condition-specific connectivity measures. In contrast, anything that is defined as a [b]first-level covariate[/b] will be: a) shown (as is, no hrf-convolution) in the default list of potential confounding effects in the [i]Denoising [/i]tab; b) appear as potential seed timeseries in the [i]first-level analysis [/i]tab (only those covariates not selected as confounding effects during the denoising step); and c) appear as potential interaction terms, both in the [i]Setup.Conditions [/i]tab as well as in the [i]first-level analysis [/i]tab for the "other temporal-modulation effects" analysis type.[/color]
[color=#000000]There are, of course, some somewhat-convoluted scenarios when you would want to genuinely use the 'copy task-related conditions...' option such as: a) for display purposes (for example, having the hrf-convolved conditions included in the list of first-level covariates allows you to include them in some plots such as QA which otherwise would need to be created manually; similarly if you want to use the 'covariate tools' gui to compute some summary measure of your conditions); or b) for more complex interaction analyses (e.g. CONN allows you define condition*covariate interactions, so sometimes it is useful to copy some subset of conditions into first-level covariates just to be able then to use the resulting timeseries as interaction terms). [/color]
[color=#000000]In practice, though, the most common use of this '[b]copy[/b] task-relate conditions...' option is a soft way to perform the '[b]move[/b] task-related conditions to covariate list' option in two steps (i.e. first use the 'copy ...' option, then, if everything looks fine, simply delete the original conditions). The 'move task-related conditions...' option is useful, as stated in the manual, when you want to perform Fair et al. -style analyses, where you still want to regress out anything that correlates with your conditions from the BOLD signal but you do not want to obtain condition-specific connectivity measures. In that case, moving a condition into a first-level covariate does exactly that, it still shows you the appropriate timeseries during the [i]Denoising [/i]step so the appropriate timeseries can still be included as a confounding effect, but it is no longer treated as a condition so CONN does not estimate condition-specific connectivity measures. [/color]
[color=#000000]Let me know if that clarifies[/color]
[color=#000000]Alfonso[/color]
[i]Originally posted by Pedro Valdes-Hernandez:[/i][quote]Dear CONN experts,
I'd like to know why one would want to copy task-related conditions to first level covariates.
Aren't these regressed out during the temporal preprocessing (denoising) anyway?
The original CONN paper (2012) suggests these effects are indeed removed in the Denoising step. It appears so since the conditions are imported as confounds with the name 'Effect of...'. I guess this is done to obtain "resting state" task-independent FC measures, as in Fair et al (2007).
However in the CONN User Manual states that, in order to achieve this, the conditions must be copied to the 1st level covariate list. Is this correct?
This is confusing. In a nutshell, what is the purpose of this 1st level covariate list, other than to provide regressors not to be HRF-convolved (like in SPM)?
On the other hand, is the HRF-free regression used to remove task effects or just HRF convolved conditions?
Looking forward any comment on this.
Pedro[/quote]