cent Hispanic. The results were similar to, but more muted than, those for the no-car variable by block group; a cluster of high-Hispanic voting-age population in the northwest section of the town of Islip stands out, but numerous tracts flare up to higher levels (and deeper shading on the map) in the upper bound map due to large margins of error. He displayed the results from a few other approaches:
- “Triangulation”/corroboration with related, auxiliary data: In this specific setting, he displayed a map that looked at the Hispanic voting-age (18 and older) population from the 2010 census summary data file. This is not the same population because it is not restricted to citizens, but it can serve as an intuitive check—or an effective, true upper bound—on the CVAP data. The high-Hispanic-concentration areas in Suffolk County that registered strongly in the CVAP maps show up on the 2010 summary file map—northwest Islip, Brentwood, Huntington. And one could construct a “corrected” map through this triangulation—for instance, an area in East Shoreham shows up in the 11–25 percent Hispanic shading level on the upper-bound CVAP map but that percentage is logically capped at the 6– 10 percent category because that is the 2010 census percentage Hispanic-and-voting-age for the tract (without regard to citizen status).
- Omission of unreliable-estimate areas: A variant of the masking approach, another take would be to visually omit tracts with highly unreliable estimates (coefficient of variation 25 percent or greater) rather than call attention to them. In practice, this took the form of leaving the unreliable-estimate tracts unshaded (white). Though he repeated that he has made no conclusions, he did suggest that this approach is undesirable because—in general—“you want to try to avoid losing information from the map,” and this feels like going out of the way to not present the whole picture.
- Overlay some visual marker on areas where margins of error could shift estimate to a different color category: In this sample map, Romalewski shaded the Suffolk County tracts in some tint of green based on Hispanics as a percentage of total CVAP population, but overlaid a dot-mesh layer on tracts where the interval estimate could place the estimate in one of the other green-tint ranges. He said that his basic reaction to this picture is that it is “okay to an extent,” but it seems to put a very high burden on the map’s readers to sort out what the map is really telling them.
He closed by noting that the Cornell Applied Demographics program is doing particular work on techniques to flag potentially unreliable areas in an interactive, online mapping application, focusing the viewer’s attention on the possible pitfalls and permitting a more detailed look at the reason for the flags by clicking on them.