Mapping Childhood Asthma: Projection, Classification, and the Map That Misleads
A county health department must produce a one-page asthma map for a school board vote on port-adjacent truck routing. The county has 40 census tracts. Tract-level childhood asthma rates run from 2 to 31 cases per 1,000 residents, with a single high tract of about 900 residents next to the port terminal. The GIS analyst has two projections available (Web Mercator and Albers equal-area) and three classification schemes (equal interval, quantile with eight classes, natural breaks). The analyst's draft used five equal-interval classes on a Web Mercator basemap. A board member has already asked why the map makes the poor neighborhoods look so big. The department has one revision cycle before the vote and no budget for tract-level confidence intervals, though three-year averages are available on request.
1. Which projection should the analyst use, and why does the choice matter here?
Read the worked answer
Albers equal-area. A choropleth encodes data in the area of each polygon, so a projection that inflates area distorts the map twice: the tract looks larger and its rate appears to occupy more territory. Web Mercator is a navigation projection built for conformality at the expense of area; at county scale the distortion is modest, but it is the wrong tool and it feeds exactly the board member's suspicion about why some neighborhoods look big.
Albers is the standard for continental-scale thematic mapping and is familiar to planners, and it keeps relative areas comparable across the county. State the projection in the legend - Albers equal-area, NAD83 - because an unlabeled map invites readers to treat shapes as neutral facts.
The transferable teaching point: every map is an argument about area, and the analyst chooses the argument before the reader sees the data.
2. Which classification scheme should be used, and what does each choice do to the board's perception?
Read the worked answer
Equal interval over a 2–31 range in five classes yields bands about 5.8 wide. With most tracts clustered low, nearly all of them fall into the first class and the county reads as uniformly safe except one red dot - that is the draft's problem.
Quantile forces equal counts per class, guaranteeing visible variation, but it splits tracts with nearly identical rates into different colors and makes the middle look worse than it is. Natural breaks puts boundaries where the data actually gap, which is defensible here because the port tract is genuinely separate, but the breaks are data-driven and can look arbitrary to a skeptical board.
Best choice: natural breaks, or a custom scheme that isolates the port tract, with explicit cut points in the legend. Add a footnote that the high tract has about 900 residents, that a single-year rate is unstable at that denominator, and that a three-year average is available. A one-page map can carry that in two lines.
3. The board member asks why the map makes poor neighborhoods look big. How do you respond, and what does the question illustrate?
Read the worked answer
First check whether it is true. It may be a projection artifact, a classification artifact, or simply that those tracts are physically large because they are industrial or low-density. Answering before checking would confirm the suspicion you are trying to dispel.
Then explain the modifiable areal unit problem: tract boundaries are drawn for census administration, not health analysis, so the pattern shifts if you aggregate to ZIP codes or block groups, or if you split a tract. A rate calculated on 900 people carries wide uncertainty, and coloring it darkest red treats an estimate as a fact.
Practical move: put the same data side by side under two classifications for one slide, state the cut points, and add a caveat line. The civics lesson is direct - the map is not the territory, and choosing the unit of analysis is a policy decision as much as a technical one.