Dear Carmen

I'm  unclear whether "emotion regulation success under high arousal " and "emotion regulation success under low arousal" are at the within-subjects or between-subjects level. I will assume the latter, i.e., you have one number per subject for each measure.



In which case, yes, I recommend you include both regressors in a single PEB model (after the constant regressor). If you're running an automatic search over reduced models, then there's nothing more to consider. However, if you're comparing pre-defined reduced models with particular connections on/off, then you'll need to run the analysis twice, switching the order of the two covariates (because the software only varies column 2 of the design matrix when comparing pre-defined models).



It's important not to think in terms of thresholds when performing Bayesian analysis. A connection with 80% probability of being non-zero might be just as interesting as one with 95% probability - albeit with lower confidence. The purpose of thresholding in the GUI is simply to focus one's attention on the most probable connections - but all of those that survive Bayesian model reduction and averaging could be important.



Let me know if anything remains unclear.



Best

Peter



On 04/12/2020, 13:51, "Morawetz, Carmen" <[log in to unmask]> wrote:



    Dear Peter,

    

    I hope this email finds you well. I have one more question regarding the spDCM. We set up two models with following covariates (emotion regulation success): 

    model 1: covariate emotion regulation success under high arousal and 

    model 2: covariate emotion regulation success under low arousal

    

    My first question relates to the evaluation of the model: For each model, we assessed the posterior probability of the connections using a threshold of <.95. Now the question arises, whether a connection between regions should only be reported if the posterior probability of the baseline connectivity/commonalities is <.95 AND this connection is modulated by the covariate (<.95). This means, is an existing baseline connectivity is the prerequsiste for reporting a modulation by the covariate. (I just know from DCM analysis that a n instrisic connectivity has to eixist before a modulatory effect can be reported). OR would you report the modulation by the covariate also if the baseline connection is not <.95. 

    

    My second question relates to covariates of no interest. Do you think we need to include covariate 2 in model 1 as covariate of no interest and vice versa? I.e. in model 1 with the covariate emotion regulation success under high arousal (covariate of interest) the covariate emotion regulation success under low arousal (covariate of no interest).

    

    Thanks a lot in advance. 

    Best wishes,

    Carmen

    

    

    

    ———————————————

    

    

    

    Universität Innsbruck   

    Institut für Psychologie     

    Prof. Dr. C. MORAWETZ 

     

    Innrain 9, Zi.Nr. , 6020 Innsbruck, Österreich 

    Telefon +43 507 507-56064 

    E-Mail  [log in to unmask] 

    Internet           https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.affectiveneuro.at%2F&amp;data=04%7C01%7C%7Cfc22c7e053cf4a1c20ea08d8985ba7a2%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637426866707466741%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=GKag%2BXOIdIRkCmP0%2BnLMXxHENFOUQ77h9zelXhWXXQI%3D&amp;reserved=0 

     

    

    

    

    -----Ursprüngliche Nachricht-----

    Von: Zeidman, Peter <[log in to unmask]> 

    Gesendet: Montag, 30. November 2020 10:51

    An: Morawetz, Carmen <[log in to unmask]>; [log in to unmask]

    Betreff: Re: [SPM] spDCM corss validation

    

    Dear Carmen

    Bayesian model comparison and leave-one-out cross validation have different purposes and the latter is not typically mandatory for experimental research.

    

    Model comparison is about testing hypotheses. So if you applied an automatic search using PEB, it will compare hundreds of models and provide you with the probability with/without each connection. The posterior probability of each connection can then be reported.

    

    Cross-validation is about asking whether the effect sizes are large enough to predict some covariate. E.g., whether the difference in a connection strength between groups is large enough to be clinically relevant, if your covariate were a diagnosis. For performing this analysis, we recommend including only the connections with the highest probability from the model comparison.

    

    Best

    Peter

    

    On 30/11/2020, 07:44, "SPM (Statistical Parametric Mapping) on behalf of Carmen" <[log in to unmask] on behalf of [log in to unmask]> wrote:

    

        Dear DCM experts,

        I hope this message finds you will. It is my first time to implement spDCM and I am not entirely sure whether at the end of the analysis the step of the leave-one-out cross validtion is demanding. For example, what if the cross validtion is not significant. Am I still able to report the results of the model or is it not interpretable becasue the cross validtion was not significant? What is the importance of this last step? Is it a mandatory analysis?

        Thanks a lot in advance.

        Best wishes,

        Carmen