Introduction
Sometimes people think I know what i’m doing.
Note for this to be visible, you need to install the quarto extension inside the paper directory with
quarto install extension sellorm/quarto-social-embedsLiterature Review
You should be using open-source. The practices here are a good way to do ‘full-stack’ open and reproducible science.
Methods
We often say that spatial is special because of two phenomena known as spatial dependence and spatial heterogeneity (Anselin 1989, 1988). Two classic spatial econometric models designed to handle these effects include the Spatial Lag Model, defined as
\[y = \rho Wy + \beta X + \epsilon\]
and the Spatial Error Model defined as
\[ \begin{split} y & = \beta X + u\\ u & = \lambda W u + \epsilon \end{split} \]
The real point being that you can dropdown to real latex any time you need to. If necessary, add the library in the header-includes section of the quarto yaml. For example if you needed aligned equations, you can split with ampersands where you want to set the alignment
Results
This section uses quarto’s conditional formatting to swap out the correct table depending out output. The problem here is pandas can write nice latex tables, but those don’t convert to html. Instead you can just write both formats out to file and select the correct one on-demand.
| n-total-pop | median-household-income |
|---|---|
| 1577.0 | 150688.0 |
| 1673.0 | 127292.0 |
| 1915.0 | 90673.0 |
| 1271.0 | 65219.0 |
| 695.0 | NaN |
| 2617.0 | 81250.0 |
| 500.0 | 64631.0 |
| 808.0 | 64787.0 |
| 1682.0 | 59010.0 |
| 1151.0 | 79725.0 |
You can also do the same thing with figures, e.g. to swap in an interactive map in the html output and use a static map in the pdf.
Everyone from the R world will recognize Figure 1 as coming from ggplot. It shows up either way. But the blockgroups in San Diego show up differently depending on the output.
Here is a map showing the spatial graph of blockgroups in San Diego using a Rook contiguity rule. But Figure 2 shows up differently depending on the output, thanks to quarto’s conditional include syntax (awesome).
This is kinda hacky because it relies on an iFrame that requires the embedded map to be available the relative URL set above (you cant download this html file and expect it to work). The embed-resources option in quarto wont work for an iframe either.
In the end, we seamlessly weave together geosnap (Knaap and Rey 2024), PySAL (Rey et al. 2021), and tidycensus (Walker and Herman 2024). Use the best tool for the job, then let quarto sew things together.
Discussion
Note that since pixi also installs binary dependencies, the install is complete with tricky deps like GDAL and the correct GCC compiler (hello windows friends!). That lets you R people use install.packages for anything that’s not (yet) on conda and your build system is guaranteed to work (though you wont have a lockfile for those packages).
Thus, you get a sandboxed install for this project, with all deps on all platforms, and dependencies right inside the repo. Just like npm (if that’s your kinda thing). This is especially nice for making sure your analysis environment isn’t broken by the evolving ecosystem while your paper is under review.
Conclusion
This is the way.