A SAS Macro for Automated Stopping of Markov Chain Monte Carlo Estimation in Bayesian Modeling with PROC MCMC
Round 1
Reviewer 1 Report (Previous Reviewer 1)
No further comments
Author Response
We are very thankful for your positive evaluation of our manuscript!
Reviewer 2 Report (New Reviewer)
Very interesting paper and certainly having a more objective set of tools for deciding when to stop MCMC in SAS would be helpful. For this reason, I do think that the current paper brings something useful to the literature. I have some specific comments that I hope will help the authors as they continue with their work.
1. P. 2, line 64 Please provide the formal definition for PSR.
2. P. 2, line 77 What constitutes a large number of draws from the posterior?
3. P. 2, line 80 Which of PSR and ESS are preferable? How would they be used together?
4. P. 3, line 96 Please provide a reference for using 1.01 as the convergence criterion for PSR.
5. P. 3, line 124 How many chains would the authors recommend that researchers use in practice?
Author Response
We are very thankful for the positive evaluation of our manuscript and your feedback! Please find our responses in the attached PDF.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
This article presents a description of how to use a MCMC module of SAS, which has some practical guidance value. However, the example given are too simple and lack comparison with existing softwares. It is recommended that the authors further show the readers why this module should be used, rather than trying to use free Stan, JAGS or BUGS softwares.
I don't have much to say about the article itself, because it is clearly written and too simple.
Author Response
We are very thankful for pointing out that guidance regarding the choice of specific Bayesian software packages might be helpful. Such software comparisons, however, were clearly beyond the scope of our manuscript. With the manuscript, we aimed at providing a SAS macro for users who are willing to apply PROC MCMC (e.g., because they are familiar with SAS or PROC MCMC more specifically) that automatically monitors stopping criteria. A software comparison regarding flexibility of modeling, computational efficiency, and user-friendliness between different software programs would be helpful as well but also a different scope that would also require to focus on different model classes to show (dis-)advantages of each software. With regard to Bayesian multilevel models, for instance, such a software comparison can be found in Hecht, Weirich, & Zitzmann (2021) in this special issue and also in Mai and Zhang (2018). We added the following paragraph in the Discussion section:
„It should be noted that a comparison of different Bayesian general-purpose software packages was beyond the scope of this article. It would nevertheless be interesting to provide an overview about the pros and cons of available packages with regard to flexibility of modeling, computational efficiency, and user-friendliness for different classes of Bayesian models, which is a topic of future research. With a focus on Bayesian multilevel modeling, such an overview—including general-purpose as well as more specialized Bayesian software for multilevel modeling—can be found in Mai and Zhang [11] (see also [28]).“
Reviewer 2 Report
This manuscript does not provide original research on Bayesian techniques of data analysis. It merely suggests a program tool on the use of stopping criteria in an automated process for the evaluation of MCMC procedures for approximating posterior distributions. It is a well-written tutorial for less trained users of Bayesian techniques (if you will).
Author Response
We thank you for your feedback! That was exactly the aim of our manuscript: To provide a tool for monitoring stopping criteria for applied users who are willing to estimate Bayesian models with PROC MCMC.
Round 2
Reviewer 1 Report
I am not satisfied with the author's response. I don't think a package manual needs to be submitted and published as a paper. Just put it directly on the website for readers to read for free.
Reviewer 2 Report
In my view, this manuscript does not really provide a progress in respect of Bayesian psychometric analysis but it is a well-written tutorial for applied researchers. I am not against the publication of this manuscript if it (as a tutorial paper ) is within the scope of the special issue.