Fascinating, thanks for sharing. The points made in the article seem obvious, but doesn't offer much in terms of a suggested solution (conscientiously redraw zip codes). Do you know of any effort to have a more accurate "health boundary" region designation?
A major problem in any demographics research is data coding. For health-related demographics, HIPPA starts playing in.
Because mailing address is almost always provided in a healthcare setting, ZIP Code data are present in records. Similarly for coronor or death reports. Other alternatives, say, census block number, exist, but would have to be coded into the data. If you're already working with ZIP rollup data, you simply cannot recode for finer-grained analysis.
The most obvious data target would be the ZIP+4 encoding, though here in a provider context you'd likely run into HIPPA concerns (I've not worked in the field for some years). For death reports, that's obviated, so access to county death records at household detail would be an option, but you'd need to acquire data across all (or some sampling of) the 3,144 counties in the US.
Some of that data may be available in electronic form, but I'd suspect some level of fieldwork might still be required.
The optimal solution is likely a huge set of overlapping boundaries each accounting for different things.
Like if easy access to fresh groceries matters, the boundaries for that will not line up with the water system boundary (there could be some interesting correlations between the areas, but a good analysis would have to split out the boundaries to account for each effect).
