vasishth / bayeslmmtutorial Goto Github PK

Are we going to get into trouble for citing Nicenboim and Vasishth 2016? There we also introduce Bayesian methods, but don't use Stan proper (rstanarm). I see our paper as very different in aim from NV16. Do you guys agree? If there is too much overlap in your opinion I can drop all citations to it. I added some citations to it because we show how to compute things like Bayes Factors and LOO and WAIC for model comparison there.

Dear Prof. Vasishth,

Is there any shareable R package or script for Bayesian hierarchical logistic regression?

Thanks.

Shicheng

One criticism I got from one of my students that we introduce chains suddenly without any explanation. It's very disconcerting to the beginner.

I wonder how I can vectorize

for (i in 1:N)
    rt[i] ~ lognormal(X[i] * beta + Z_u[i] * u[subj[i]] + Z_w[i] * w[item[i]], sigma_e);
}

and make it suitable for a multivariate lmm ? I am trying to modify your varying-slope/varying-intercept/no-correlation for a multivariate regression.

Thanks.

This Table was supposed to have a correlation parameter (a reviewer pointed this out), but we didn't put it in. This should be added.

Since we don't show the lmer fit, the reader cannot know that the correlations are +1 and -1. Yet we say:

"However, the classical LMM is overparameterized due to an insufficient number of data points. Hence, correlations between random effects could not be estimated, as indicated by perfect correlations of $-1$ and $1$."

I would show the lmer fit and the Stan fit for all variance components side by side (e.g., in Table 5). In Nicenboim and Vasishth 2016 I think we did this.

Why is the posterior prob. in our paper 95% and in Nicenboim and Vasishth 89%? I'm surprised by this huge discrepancy for the same data. Need to look into this.