When you are deciding on the software you would prefer to use for Bayesian meta-analysis, there are several aspects to bear in mind. We thought it would be useful to put them into one table. This appears at the back of the book as an appendix, but we can extend and update it here.
| Software | Language interfaces | GUI | Open source | Algorithm | Variance parameter |
|---|---|---|---|---|---|
| BUGS | stand-alone (Windows, not maintained), R, Python, Stata (not maintained), Julia | yes (Windows only) but not maintained | OpenBUGS but not WinBUGS (programmed in C) | Gibbs | precision |
| JAGS | R, Python, Stata (experimental), Julia | no | yes (C++) | Gibbs | precision |
| Stan | command line, R, Python, Stata, Julia, etc. | no | yes (C++) | Hamiltonian, variational, Laplace | standard deviation |
| bayesmeta | R | no | yes (R) | no sampling, analytic distributions instead | standard deviation |
| brms | R | no | yes (R and C++) | front end for Stan | standard deviation |
| Stata | stand-alone, Python | yes | mostly (ado or Mata) | RWMH, Gibbs | variance |
| JASP | stand-alone | yes | yes, various R packages | front end for various R packages | standard deviation |
| MLwiN | stand-alone | yes | no | RWMH, Gibbs | any |
| SAS | stand-alone | no | no | RWMH | standard deviation |
| PyMC | Python | no | yes (python) | RWMH, Gibbs, Hamiltonian, variational | standard deviation |
| multiNMA | R (beta testing at present) | no | yes (R) | front end for Stan | standard deviation |
These options are mostly freeware. The exceptions are:
- MLwiN is free upon application from UK academics, otherwise costs between 400 and 800 GBP
- Stata costs about 1000 USD for a one-year license for one user, significant discounts are available for students
- SAS would typically be installed organisation- or team-wide, rather than for a single user, but starts from about 10,000 USD per year
These costs are generally reduced for multiple users/years, and are correct at the time of writing.