Just another problem with Bayesian statistical programs

 

As someone who uses the R coding statistical environment lots, I much prefer frequentist methods for analyzing datasets. Generalized linear models or just linear models and more complicated things like GAMs, run really fast in R. Literally as soon as you press enter on the keyboard in most cases. And sometimes even before you push enter on the keyboard because you accidentally brushed the mouse pad or pushed some random key combination that magically made it all work.

 For anyone who's trying to solve the problem of the likelihood of you having an illness because a test came back positive for that illness but with the background knowledge that there are false positives and false negatives for a test will be aware that probability is far from straightforward or intuitive. You have to subtract some things from each other and divide it by the other things and sometimes you swap the order of those things around. It is about as intuitive as trying to just eat the holes in Swiss cheese. Like most normal people I will fail to solve these probability problems unless the correct formula is written down next to me from some cheatsheet and I have a computer to plug the proper values into.

 Bayesian statistical programs in R are even worse. Bayesian statistical programs run on multiple iterations of complicated probability surfaces. Running even simple formula in them takes a rather long time. But the first challenge is to install one.  My recent experience at attempting to install code submitted in reproducible formats for a peer reviewed journal made me realize that installing Bayesian statistical libraries is a bit like throwing a dart at a dartboard, where things only work if you hit the bulls eye. Except it is based on the prior assumption that you are holding a dart. Then when you throw the dart it turns out you were actually holding a Bunny. And this leads to all sorts of philosophical conundrums and dead ends and circular thinking. Am I allowed to throw bunnies at dartboards? Should I now be throwing the board at the bunny? But I don’t want rabbit for dinner, all I wanted to do was play darts! Or was it install JAGs? Then I get stuck in 10000 iterations of tangential meanderings over probability surfaces that look like dartboards created either using LSD or a randomization function applied to RColorBrewer. 

I emerged from it all burnt out after a too long burn in period, and much reduced self-confidence intervals. While I have so far frequently solved my analytical problems, I fear that I can no longer run away from the Bayesian hounds closing in. I really hope that when they catch me, they turn out just to be baying bunnies. 

 

 

Full credit to Rasmus Baath for the basis for the visualisation: http://www.sumsar.net/blog/2018/12/my-introductory-course-on-bayesian-statistics/