The Two-Tier Map: Why Half of America's Deer Data Isn't County Data

A twelve-state deer map looks like one dataset. It is two. In seven states, a hunter must check or register every animal, and the county of kill is a recorded field: real, countable, mappable. In five states, harvest never resolves below a management unit (four estimate it from a post-season hunter survey, and Minnesota registers every deer but reports only by permit area), so a county number does not exist in the source and cannot be recovered from it. Painting all twelve as a single county choropleth models half of the map. This is a structural fact about how the data is collected, not a hole someone forgot to fill.

Every deer harvest number in America is produced one of two ways. In the first regime, the state requires the hunter to register the animal: a phone call, a website, a check station, a tag scanned at a locker. Registration captures the county of kill as a field on the record. Illinois, Iowa, Missouri, Wisconsin, Indiana, Ohio, and Michigan all work this way, which is why a per-county harvest table for those states is a census of reports, not a model.

In the second regime the harvest never resolves to a county. Four of the five states (Kansas, North Dakota, Nebraska, and South Dakota) mail a post-season questionnaire to a sample of license holders, ask what they killed and roughly where, and project a statewide estimate by weighting the responses. Minnesota is the near-miss: it registers every deer, so it counts rather than estimates, but it publishes those totals only by deer permit area. In every case the geography is a management unit — permit area, data analysis unit, hunting district — drawn for biology, not to nest inside county lines. Their harvest figures are perfectly legitimate; they simply do not have a county inside them.

When a registration state reports 4,318 deer in a county, that is a count of registered kills whose county field reads that county — an administrative census with the ordinary caveats of any self-reported total. When a survey state reports a unit estimate, that is a modeled projection with a confidence interval, sliced to a geography the state chose. The two are not interchangeable, and the difference matters most at the boundary: a management unit routinely straddles four or five counties, so cutting a unit estimate to county lines requires an assumption (usually land area or habitat share) that the data itself never supplied. That assumption is where a survey-state 'county number' is born, and it is why we refuse to make one.

You cannot un-survey a survey. If a state never wrote the county down, no analyst can recover it. Only guess it, and call the guess data.

The temptation is obvious: twelve states, one legend, a smooth gradient of deer from Ohio to the Dakotas. It renders beautifully and it is wrong in a specific way. Over the seven registration states the color encodes a measured count. Over the five survey states the same color can only encode a number the mapmaker manufactured by chopping a unit estimate into county-shaped pieces. The viewer sees one continuous surface and reads one continuous meaning (deer density, harvest pressure, opportunity) with no signal that half the surface is measured and half is modeled. The map's own consistency is the deception.

Wisconsin is a useful reminder that grain can move even inside the honest half. The state's live county series covers 61 counties after a 2014 rebuild; an archival backfill recovers 72, including eleven Northern-Forest counties the agency later collapsed. We keep those vintages separate rather than blend them, because mixing a 2014-era compilation with the current series would break the one thing county grain is supposed to guarantee — that two counties in the same year are measured the same way. If we will not silently merge two vintages of the same state's real counts, we certainly will not invent counties for a state that never kept them.

The fix is not clever, only disciplined. Shade the seven registration states by county. Shade Minnesota, Kansas, Nebraska, and South Dakota by management unit, and North Dakota at the statewide level, with a legend that says so. The result is less uniform and more true: it shows the reader exactly where the resolution ends. A hunter comparing two Iowa counties is standing on measured ground; a hunter asking which North Dakota county kills the most deer is asking a question the state's own data cannot answer, and the map should say that out loud rather than paper over it with a plausible color.

Data literacy here reduces to one habit: before trusting a county number, ask whether the state registers deer or surveys hunters. If it registers, the county is real. If it surveys, the county is either a management unit wearing a county's clothes or an outright fabrication. The more polished the map that shows it to you, the more carefully you should check.