[FieldTrip] Regression of behavioral data against EEG data

[Schoffelen, J.M. (Jan Mathijs)]

Hi Burcu,

Conventially, the parameter estimates (i.e. the beta weights) are taken to the second level for an inferential statistical test. Averaging p-values, or r-squared values is usually not done, and also does not make much sense.

Best wishes,
Jan-Mathijs


> On 10 Dec 2022, at 17:07, Burcu Bayram via fieldtrip <fieldtrip at science.ru.nl> wrote:
> 
> Dear FieldTrip community,
> 
> I'm new to regression analysis of EEG data and unsure which regression outputs to use. Beta-coefficients give a very different pattern of results compared to r squared values or p-values, although all of them (to my understanding) should express a form of relation between the two datasets. We are looking for help regarding the interpretation of those data and which one to select for our analysis. In our model, behaviour is the predictor and EEG activity is the outcome. The datapoints for each are single experimental trials (~2000 per subject).
> So far, we just used simple linear regression, but the plan is to use multiple linear regression at a later stage.
> The idea is to plot and interpret the regression results as if they were EEG amplitudes. So we get a time course and a topography of regression results, that help us to determine where and when in the brain behavior predicts neural activity. Our main questions are:
> 
> 1. Which value makes most sense to use as an indication of brain/ behavior relationship? The betas should provide the quality/ direction of the relationship, but don't say anything about how large or important that relationship is. The r squared or also the t or p values for each coefficient tell something about the strength of the relationship. The issue is, that they give really different activity patterns. You can see the topographies of beta values, r squared values and p-values in the attached images.
> 
> 2. The second question is which of the single-subject regression outputs actually can be used for group level plots and statistics: Is it possible to average over e.g. betas or p-vales across subjects, and also do group level statistics with (e.g. compare group-level p-values or betas between two conditions)?
> 
> Thank you so much in advance!
> 
> Best regards,
> Burcu
> <beta_av_postmean.png><p_values_av_postmean.png><r_squared_av_postmean.png>_______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://urldefense.com/v3/__https://doi.org/10.1371/journal.pcbi.1002202__;!!HJOPV4FYYWzcc1jazlU!-BVAi-vLCyuI-z25XshuL5Ot-zKj9ciiKIvw7cD8r0qhP2CQfYup6W8y10f8RVGleO0IZjn6YH1s39xwbLKu4dPeUFzFOfIwfYy5Qg$