Nov-09
: In this lesson, we will plan reanalysis or replication of COVID-19 spatial accessibility.
Prepare for lecture
Brainstorm ways in which the research compendium for this notebook could be improved, including revisions to:
fix errors
reduce uncertainty
improve reproducibility
improve visualization of results
We will collaborate in an in-person class discussion to more deeply understand the Python script for this analysis and identify opportunities for improving and extending its functionality, especially to more completely implement the analysis in the published research paper.
Results from spring 2021
The Spring 2021 lecture concluded with a challenge: what is the largest source of error in the current version of the python Jupyter notebook?
We determined the answer to be boundary effects introduced to the analysis when the geographic extent was limited to Chicago.
The road network was limited to boundaries for Chicago that exclude Midway Airport, while population data was limited to boundaries for Chicago including the airport, and hospitals were included within a buffer distance of Chicago.
The inconsistent spatial extents caused the Midway Airport area to seem inaccessible to health care, and caused hospitals in the buffer outside of Chicago to snap to the nearest node of the road network without accounting for the distance between the hospital and the boundary of Chicago.
Finally, services outside of Chicago were included but population demand on those services was not.
Spring 2021 class questions and ideas for reanalysis
Fix boundary / edge effects stemming from limited extent of road network and need to expand analysis beyond city of Chicago for accurate results within the city. Done by including a buffered road network and population/households outside the buffer.
Add time benchmark to key sections of code with %%time as the first line of the code block to assess the main contribution of the paper: computational speed on cyberinfrastructure. Done.
How important or efficient is the network simplification function? Does this impact the locations/connectivity of any hospitals? Done by counting edges before and after, and by buffering the road network and excluding hospitals beyond the road network.
Clarify use of packages and functions throughout the code. Improved to a great extent, but it would be far easier for novices to understand the methods without all of the extra code for parallel processing or for dynamically choosing the population and the health resource type.
How are weights justified for the enhanced distance bands? Would different weights significantly change outcomes? Done by some previous student reanalyses.
For reproducibility, we should filter hospital types in the code, rather than beforehand. Done, but we could not find original data on ventilators.
Pay attention to CRSs: are they always transformed when they need to be? Can the final maps be projected? Done
Could area-weighted reaggregation be used for the overlap analysis? This could happen in the overlap_calc function.
Why are hexagons used for the analysis as opposed to other spatial data structures, e.g. a raster-like tessellation of squares? In other words, is there a modifiable areal unit problem? Does the use of different sizes and shapes for the unit of analysis matter?
Should the default network speed be set to 35mph? Hint: could we use the new osmnx.speed module to fix speed limits? If so, we need to be careful about the measurement units for distance, speed, and time.
Spring 2021 class ideas beyond the scope of a reanalysis
There are other models besides two-step catchment analysis for spatial interaction in health geography, e.g. gravity model.
Could other forms of transit be included? Why is the analysis dependent upon personal vehicle use?
Is there a better indicator for hospital capacity other than ventilators or ICU Beds?
Fall 2023 class questions and ideas for reanalysis
Area-weighted reaggregation of population to hospital catchments
Area-weighted reaggregation of hospital catchments to hexagons
Disaggregate population data into finer geographic units
Mask areas without population from final maps / visualize areas with low accessibility and large populations in final results
Visualize social vulnerability index
Calculate correlation between accessibility and social vulnerability
Fine-tune hospital catchments based on patient behavior, including distances, distance weights
Improve transportation network modeling with factors such as multimodal transportation, traffic congestion, and transportation access/choices for different populations (e.g. people with no cars)