How ZIP codes nearly masked the lead problem in Flint (2016)

For the folks who in this thread say "just use a point map of occurrences and you would have seen the problem" you are right in this case, but it isn't really enough in most cases.

For one, the boundary of the city limit needs to be shown to see the correlation; perhaps one should map the point occurrences a la Snow, and then overlay all the boundaries you have (mosquito district, city boundary, water district boundary, county, etc...).

Additionally, a point map doesn't tell you rates of occurence, for that you need a denominator, or you can't calculate things like number of occurences divided by total population. One poster mentioned Census Tracts, and I would agree those should be the default boundary, subject to more investigation -- tracts correpsond well to "natural" boundaries, they are mostly consistent over census decades, they are mostly consistent with population (~5k), they nest perfectly with other census boundaries, and they are the smallest geography for which you can get mostly the full suite of population characteristics like education from the ACS.

For some reason tracts feel just wrong to people. I have managers who reject zip codes, but then fixate on weird systems like pixels or equal area hexagons or whatever for which you can't get denominator data and which don't correspond to anything real. I don't understand, though I don't think in public policy / social work education anybody bothers to teach you shit about GIS or Census data unless you know to ask specifically.

A related phenomenon is that different agencies have different boundaries, despite often sharing the same geographic name. Cities, water districts, sanitary districts, school districts, community college districts, counties, congressional districts, etc each have their own boundaries, and they are usually not aligned.

Lesson, if you want to correlate something spatially, use spatial coordinates.

[1] https://en.wikipedia.org/wiki/John_Snow

There are so many, many ways to lie with numbers, and once money and power depend on the outcome of the algorithms, you can never trust that people won't design them to produce the outcomes they want. I suspect the problem here was less 'didn't understand geography' and more 'didn't give a shit about the truth'

If not for that, there never would have been a problem, Flint would have eventually had its long term new supply without ever needing an intermediate option.

When I sat down a finally looked over the details of this story it's the one glaring omission that I noticed from all of the news coverage.

Last year I worked on a study where the geographic data had been fuzzed prior to passing it along to the study team, a process that unfortunately resulted in the patient addresses being simplified to just the ZIP code provided by the patient.

There were two major issues beyond the general unsuitability of ZIP codes as geography: first, ZIP codes don't seem to be verified during the course of patient care. They are recorded (incorrectly at a minimum rate of ~0.2%) as a part of a patient address during the creation of the initial record. Even if the address is used for mailings, a bad ZIP does not result in mis-delivered mail in the vast majority of cases due to the availability of city, state, and street address data in the full address. The issue is just never discovered until you are put into the position of relying on only the ZIP code for geography, at which point the mitigating data (the full address) is unavailable.

Second, even when ZIP codes do match to Census ZCTAs (the closest thing to 'ZIP code geographies' available and commonly utilized) there are some ZIP codes that do not have exactly matching ZCTAs. For example, 03911 and 04565 are real ZIP codes without (exact) matching ZCTAs.

It would be preferable if address data, before being obfuscated for data anonymity reasons, was properly geocoded. After geocoding, the resulting geography could then be fuzzed for anonymity. Of course, this would just lead to arguments about what constitutes proper diligence in geocoding :)

He needed geographic search boundaries that matched where the suspect water was being delivered. It so happens in Flint's case that it matches city boundaries. I suspect that's not the case everywhere...that is water delivery in some places might extend beyond city boundaries. Or perhaps some cities are split, and have more than one water source.

So the overall lesson here doesn't have much to do with zip codes, it has to do with searching the right boundaries. City boundaries wouldn't have been appropriate to search for mosquito borne issues. So that's not a good default either.

[1] https://www.hhs.gov/hipaa/for-professionals/privacy/special-...

The City of Flint's water system is isolated from the rest of the county, which falls under the Genesee County Drain Commission's jurisdiction and is connected to the Detroit water system. It is still difficult to get some people to understand that they are different things (have had to explain to family many times that while we live in "Flint", we are not on Flint water). It would be interesting to see a study done on the effect this had on businesses both within the city and in the greater Flint area.

My own ZIP code is a perfect example of this. We're in the ZIP code of Town A, but Town B is actually the closest incorporated town to us. The Town A ZIP code extends between the state line and a river, up the eastern edge of our county. That covers our large development, which has people from very poor to very wealthy, four smallish unincorporated communities that are mostly poor-to-median, an incorporated town of about 1,000 bordering Town A, and the actual Town A which is a National Historical Park and consequently is mostly shops and museums catering to tourists and actually has very few residents (about 300 out of about 13000 that live in the ZIP code).

Stop making up words for which there already are words. "estimate" works just fine in that context.