reason for symmetric priors

[vasishth]

This was necessary because we added trial id as a predictor as was suggested by a reviewer.

i don't understand this explanation.

[vasishth]

is this true? pls cite the page no.

It is not possible to include symmetrical and directional priors within a single brms model (Bu ̈rkner, 2021), thus all priors were changed to symmetrical ones.

if i were a reviewer, for me this would be a strange reason. i would have just fit it in stan then.

wasn't the reason that we are using savage dickey to figure out the van dyke bfs and using a truncated prior there would not make sense as that was the first study on this design? our truncated priors were based on van dyke claims. so for consistency we used symmetric priors everywhere?

[vasishth]

we also switched to savage dickey to compute van dyke bfs no?

[PSchoknecht]

In the first response to reviewers, we have given this explanation:

"It is not possible to use symmetrical priors (for trial id) and directional priors (which we had used originally for the effects of interest to reflect longer reading times in high interference conditions) within the same brms model \citep{brms}. So in the revised analyses, we now use symmetrical priors on all slopes which are a priori agnostic regarding the sign of the effect. To ensure consistency in our analyses and comparability between the analyses, we changed all our analyses (SPR and ERP data) to use symmetrical priors. These revised analyses have produced the same result pattern as our previous analyses (no syntactic interference or interaction, but a semantic interference effect)."

There were no objections to this reasoning. Too simplify the current response, I can say that the priors were changed to be consistent. That is also true.

[vasishth]

yes ok

[PSchoknecht]

complete revised response:

Yes, as we had explained in the response to reviewers on August 29, 2024, all analyses were changed in the first round of revisions. For a fair comparison of our results to the results of the previous studies, especially Van Dyke (2007), it was necessary to compute Bayes factors using the Savage-Dickey ratio (in the original version of the manuscript we had used bridge sampling). Furthermore, for the Van Dyke (2007) data, the first on this design, it only made sense to use symmetrical priors which are agnostic regarding the sign of the effects. For consistency, we then used symmetrical priors for all our analyses (compared to the directional priors which we had used originally). These changes did not change the results qualitatively, but led to some numerical changes which probably piqued the reviewer’s interest.

[PSchoknecht]

Please close if ok