Mitzi
summary
Claims that the ICAR as improper prior with a sum-to-zero constraint gives fast, usable areal smoothing; implements the pairwise-difference formulation in Stan and uses spdep/shapefiles to build adjacency. Applies to New York City data with ~2100 census tracts (Brooklyn subset 709) and shows neighbor borrowing reduces variance in low-population areas. Reports model fits in ~1 hour for large problems, notes potential weird geometry and fitting difficulty, and compares roughly equal results to INLA.
AI-written summary · about
chapters (6)
- 0:00Spatial models for aerial data
- 3:20Adjacency in spatial models
- 9:10Intrinsic CAR model explained
- 13:19Improper prior and sum-to-zero
- 18:32Implementing I-CAR in Stan with NYC
- 22:32Conclusion
related talks
05mExploratory Spatial Data Analysis in the tidyverse
06mInsights in 5-D! (Using magic small-multiples layouts)
28mAnthony Bak, Principal Data Scientist at Ayasdi @ MLconf SEA
22mMessy maps to make sense of a complex world
41mUsing the Shapely Package for the Care and Feeding of your 2-D Spatial Data- 05mExploratory Spatial Data Analysis in the tidyverse
34mParallel Efficient Clustering with Graph-Based Approximate Nearest Neighbor Search
13mGeometry for Social Good