
Welcome - this site hosts my personal blog, software, publications, and bio. I'll be posting here about my work in geography and GIS, and whatever else interests me.
Are all models wrong? Some mathematical lessons on scientific realism
For some, it seems to be common sense that ‘all models are wrong’. The phrase ‘all models are wrong, but some are useful’ is attributed to George Box in reference to statistical modeling (where the phrase is arguably more innocuous), though the underlying attitude is prevalent in many fields. Humility is a virtue, but the statement that all scientific models—i.e., its theories and claims about the structure of our reality—are wrong amounts to something else entirely.
Are spatial regressions 'significantly biased'?
A raft of recent articles in journals of statistics assert that spatial regression models are ‘biased’. Some are now referring to this supposed bias as ‘spatial confounding’. Red warning flags abound in this literature, and I’ve written elsewhere why I’m not convinced by these claims. This post walks through some of the relevant concepts for spatial analysis and examines ‘spatial+’, which is among the more recent proposals for fixing the supposed bias of spatial regression models.
Space-time modeling in Stan: charting the evolution of U.S. mortality rates
This post is a tutorial on modeling spatial-temporal data using the Stan modeling language, with a focus on areal data. When we take the right approach, Stan can provide a great platform for spatial statistics (presumably, so would other Hamiltonian Monte Carlo samplers). I’ll illustrate by modeling mortality rates for U.S. states and D.C., covering the years 1999 through 2020.
A minimalist basemap for urban cartography
For a recent project, I a wanted a basemap that would display the local street network with some basic infrastructure and hydrology, but without any labels. It is now easy to add web maps to a GIS project by connecting to OpenStreetMap (OSM), CartoDB, or another web tile server. However, if you want to change the styling or control what type of objects are shown (and at what scales), then a web tile server won’t get you very far.
Modeling time trends in disease incidence with the 'surveil' R package
This post introduces the ‘surveil’ R package for modelling time trends in disease incidence or mortality, including age-standardization, percent change analyses, and other quantities of interest.
Spatial data analysis with geostan
Re-posting from r-spatial.org to introduce the geostan R package for Bayesian spatial analysis, but with a few updates (geostan now supports spatial econometric models).