Skip to content
 

Mitzi’s talk on spatial models in Ann Arbor, Thursday 5 April 2018

Mitzi returns to her alma mater to give a talk at joint meeting of the Ann Arbor useR and ASA Meetups:

Abstract

This case study shows how to efficiently encode and compute an intrinsic conditional autoregressive (ICAR) model in Stan. When data has a neighborhood structure, ICAR models provide spatial smoothing by averaging measurements of directly adjoining regions. The Besag, York, and MolliƩ (BYM) model is a Poisson generalized linear model (GLM) which includes both an ICAR component and an ordinary random-effects component for non-spatial heterogeneity. We compare two variants of the BYM model and fit two datasets taken from epidemiological studies of Scottish lip cancer (56 regions) and New York city pedestrian traffic deaths (700 regions).

It’s based on her Stan case study on ICAR models.

Registration

The event is open to the public. Here are the Meetup registration details.

6 Comments

  1. Louis Raes says:

    Is there any work on using spatial smoothing when implementing Mr. P ? Or is that just a silly idea?

    • Louis Raes says:

      I probably should clarify. As I understand it, mr.p. approaches have an indirect way of providing spatial smoothing in the sense that adjacent regions may have correlated observables or a similar composition, but I wondered whether there is work where there is a direct dependence in the model e.g. via an icar component (or something similar). I could not find it, but I am not very familiar with the literature.

      • Dan Simpson says:

        So Cici Chen, Jon Wakefield and Thomas Lumley wrote a lovely paper that smoothed post-stratified estimates with a CAR model, which is different to what you’re asking.

        As for MRP, I don’t think there’s anything weird about doing it. In the end, it doesn’t really matter how the regularised estimates are produced as long as they are sensible (and the extension from independent random effects to structured random effects isn’t a particularly large one, especially from a Bayesian point of view).

        • MJT says:

          that CWL paper is really good.

          i’m actually putting together an intro summary of how various regularization techniques fit into small area estimation.

          one thing i always thought about is, MrP shows that you can always post-stratify ANY model based estimate.

          whether whether your assisting model is appropriate or if you even need / want to post stratify seems to be what the end user needs to deliberate over

  2. Louis says:

    Great, thanks for the reference.

Leave a Reply