You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Proposals to define, test and address confounding in fMRI
By Manjari Narayan, Stanford University
Theme: Past, Present and Future of Open Science
Format: Emergent session
Abstract
Advances in open source tools and pipelines like fmriprep and the BIDS ecosystem make it feasible to conduct more thorough investigations of putative nuisance variables as good proxies for confounding as defined in recent work in theoretical epidemiology.
The goal would be to have a discussion on theoretical definitions of confounding in the literature because doing so will clarify how to structure models that incorporate control for confounding in a more principled way and to design tests/benchmarks for deciding when sets of variables constitute good surrogate confounders.
Participants will discuss at least one of the following questions
What do we think of recent formal definitions of confounding (as opposed to non-confounding) bias in theoretical epidemiology? Can we use this to clarify colloquial usage of noise/nuisance/confound variables in neuroimaging and create a better consensus on terms used and types of bias?
How do we empirically test assumptions regarding whether there is residual confounding in data?
What are current approaches being used in the context of non-BOLD (e.g. motion), BOLD (e.g. vascular), demographic/sampling biases (which includes both confounding and selection bias due to epidemiological design) in fMRI studies? How can researchers who work on these problems benefit from recent theoretical clarify on confounding?
Can or should we organize a special issue or some kind of manylabs challenge to empirically test presence of residual confounding and validity of current approaches in accounting for them? This could involve planned/pre-registered analyses of different datasets with different epidemiological study designs with different strategies.
Proposals to define, test and address confounding in fMRI
By Manjari Narayan, Stanford University
Abstract
Advances in open source tools and pipelines like fmriprep and the BIDS ecosystem make it feasible to conduct more thorough investigations of putative nuisance variables as good proxies for confounding as defined in recent work in theoretical epidemiology.
The goal would be to have a discussion on theoretical definitions of confounding in the literature because doing so will clarify how to structure models that incorporate control for confounding in a more principled way and to design tests/benchmarks for deciding when sets of variables constitute good surrogate confounders.
Participants will discuss at least one of the following questions
What do we think of recent formal definitions of confounding (as opposed to non-confounding) bias in theoretical epidemiology? Can we use this to clarify colloquial usage of noise/nuisance/confound variables in neuroimaging and create a better consensus on terms used and types of bias?
How do we empirically test assumptions regarding whether there is residual confounding in data?
What are current approaches being used in the context of non-BOLD (e.g. motion), BOLD (e.g. vascular), demographic/sampling biases (which includes both confounding and selection bias due to epidemiological design) in fMRI studies? How can researchers who work on these problems benefit from recent theoretical clarify on confounding?
Can or should we organize a special issue or some kind of manylabs challenge to empirically test presence of residual confounding and validity of current approaches in accounting for them? This could involve planned/pre-registered analyses of different datasets with different epidemiological study designs with different strategies.
Useful Links
Tagging @mnarayan
The text was updated successfully, but these errors were encountered: