This is from Nathan Sanders, who presented this work with Victor Lei at StanCon in 2017:
I wanted to share that our analysis of US mass shootings using a Gaussian process timeseries model has now been accepted by the journal Statistics and Public Policy. The analysis remains much the same as was presented at StanCon, but we have reformulated the discussion to focus on the role of informative priors. The title is “The Role of Prior Information in Inference on the Annualized Rates of Mass Shootings in the United States.”
We show that providing informative prior information (in this case, on the GP timescale parameter) is necessary to draw conclusions about the short timescale (i.e. a few years) variation in the mass shooting rate. With a weak prior on the timescale parameter, the conclusion is inevitably that there is no short timescale variation, which we argue is a recipe for inaction in a policymaking context. At the same time, performing full Bayesian inference with Stan allows us to deeply explore the implications of our prior choice and understand its relation to our conclusions.
A preprint of the accepted version and all code and data associated with the project has been posted here: https://github.com/nesanders/spp-massshootings
Thank you for all your help and feedback to this work, and of course for Stan! Any further feedback would be welcome.