Hello,
I've got a question about reaction times in a cognitive task being modelled as a regressor of interest or regressed out from the estimate.
In my experiment, subjects performed a (variant of a) mental rotation task in the scanner. Stimuli were presented in an event-related manner: they disappeared when the subject gave a response.
Therefore, I have the onsets and reaction times. My research question is: which brain regions are involved in this type of mental rotation task? In order to model and estimate that, I thought about:
- Regress out reaction times as parametric modulators (RT are confounders). Since the BOLD signal is correlated with the reaction time (doi.org/10.1371/journal.pone.0004257) and I'm not exactly interested in brain-mapping the reaction time (but a cognitive aspect), it makes sense to do it. However, mental rotation effects are mainly seen in RTs and I'm afraid I'd be regressing out the cognitive aspect together with the RT.
- Look at the parametric modulators (RTs are a proxy of the mental rotation effect). This parametric analysis would inform about brain regions correlated with the task load (rotation), but entangled with a measurement that is correlated with the duration of the trial.
- Report both above and discuss this entanglement (doi.org/10.1016/j.neuroimage.2014.05.073).
Does anybody have an opinion about this?
Thanks in advance!
Gustavo
I am not sure whether I understood correctly what you would like to do, but maybe you could do the following:
- add the demeaned reaction times as an explanatory variable (EVs).
- set different contrasts to analyse:
* 1 for your time series and 0 for the demeaned reaction times: this would remove the effect of reaction time.
* 1 for your time series and -1 for the demeaned reaction times: to find regions that the activation for your time series is greater than due to the reaction times.
* 1 for you time series and 1 for the demeaned reaction times: find where you have activation due to both time-series and reaction time.
- than you can combine the contrasts according to you interest.
Hi Andre, many thanks for your answer! It's a nice technical summary of how to look at the effects separately.
My question was rather in terms of disentangling the inconvenient correlation between BOLD signal and reaction time: reaction time alone modulates the estimation based on the BOLD signal (time on task effect). Then, one can't really tell whether the the estimation comes from a cognitive aspect or simply from a measure that took longer to be made.
I came across this paper: Article Inconvenient correlation – RT–BOLD relationship for homogene...
. Regressing out the RT might be desired to remove the time on task effect, but the authors pointed out that this effect is closely related to sustained attention during the task (that's why it's common to see the dorsal attention network in RT-dependent estimates).I'm still reading about the topic, but I'm prone to conclude that one should report both effects: the estimate that is independent on the RT (neural engagement) and the one that is dependent on the RT (neural effort) (see: Article Interpreting response time effects in functional imaging studies
)Cheers,
Gustavo
I suspected your question wasnt as simple as I had understood 😅.
I have just a guess. Maybe you could combine with other fMRI acquisition with a different task? Just to see if you have a clue on the reaction time?
After some research, I would like to share some insights that may help other researchers who have the same question.
The estimation of the BOLD response is inherently coupled to the reaction time (i.e. the duration of a neural cognitive response to a task). When one measures the neural response associated to a given condition which is proportional to the reaction time, one ends up also measuring neural activity associated to the time-on-task. Depending on the researcher's intentions, it is not a simple task to dissociate informative from non-informative estimations.
When the experimental design is already defined and/or it is cumbersome to separate cognitive processes during the experiment, one could try to address this entanglement in the analysis. One idea is to estimate the BOLD signal regressing out the reaction time which is related to neural engagement, while covariations with RT would be related to neural effort. I recommend this paper for further details: Article Interpreting response time effects in functional imaging studies
In SPM, the order of the parametric regressors matters and one can control the influence of the reaction time in a condition by defining it as the last parametric regressor. Details of how to correctly define it can be found here: Article Orthogonalization of Regressors in fMRI Models
The fact that several psychometric tasks recruit the same network can also be that it shares a common cognitive aspect, such as sustained attention. See the discussion in Article Inconvenient correlation – RT–BOLD relationship for homogene...
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression