Hi all,
I wanted to share our new preprint:
Statistical models of repeated categorical ratings: the R package {rater} (https://arxiv.org/abs/2010.09335)
which presents an R package rater designed to fit Bayesian versions of the Dawid-Skene model.
The paper:
- Describes variations of the Dawid-Skene model and how they relate to other latent class models
- Describes how we can fit the Dawid-Skene model in Stan using marginalization even though it has latent discrete parameters
- Discusses the theory of doing this marginalization and how it relates to Rao-Blackwellization
- Present all the many features of the R package
Of particular interest to people here are:
-
The package itself, which is another example of an R package using RStan on CRAN
-
The description of the ‘theory’ of marginalization.
We wrote this section because we found interpreting the marginalized output from Stan confusing and because we wanted to understand how marginalization related to ‘Rao-Blackwellization’. If other people here are confused regarding these points I would recommend having a look at that section of the paper!
All the best,
Jeffrey