The Map That Stayed

The 1860 column predicts the 2020 column. For every ten-point increase in a county's enslaved share in 1860, Black mortality today runs 13.6 deaths per 100,000 higher.

And here is the number doing the real work: White mortality in the same counties runs 8.5 per 100,000 lower.

That is the detail worth stopping on, because it rules out the comfortable version. Poverty moves both lines the same way — a hard county is hard on everybody in it. This one splays. The 1860 map does not identify a poor place. It identifies a sorted place, one where the distance between two groups' odds of dying early is wider, and wider from both ends.

There is a second reading of that lower White figure worth saying out loud. A hundred and sixty-five years of people moving into and out of these counties is itself a sorting process, and age-adjustment tells you nothing about who left and who stayed; some part of that downward line may be a fact about which White population is there now rather than about what the place does to it. That doesn't dissolve the divergence. It does mean the divergence is partly a statement about the county's composition — which, as it happens, is what the rest of this is about.

Two responses to this are already loaded, and they are looking for the same thing. The first: proof that racism is ongoing and structural. The second: nobody now alive owned anybody, so this is history rather than accountability. They disagree about the verdict and agree entirely about the format. Both are arguments about perpetrators — present or absent, culpable or exonerated.

The study is not about perpetrators. It is about plumbing.

The authors tested what carries the signal forward: residential segregation, economic inequality, racial disparity in poverty, and racial disparity in college attainment, measured across the 1990s and 2000s. Those four accounted for the link. With them in the model, nothing the data can still see is traveling by any other route.

Full mediation reads, at first, as deflation — see, it isn't slavery, it's just segregation and money and schooling. That gets it backwards, and it is worth being exact about why.

If some fraction of the 1860 signal were still moving by a route nobody had named, something would have to be actively pushing it along — and a thing actively pushing can in principle be located, named, and stopped. What the study finds instead is that four ordinary institutional arrangements do the carrying, and all four are visible, measurable, and already named. School district boundaries. Tax base. Who holds title to what. Where the hospital is and where it isn't.

Take that at its statistical weakest and the point holds. Failing to detect a residual direct effect is not the same as showing there is none; with 901 counties, a real remaining path could sit under the threshold and produce exactly this result. But the argument does not need the null. It needs the four named channels to carry most of the weight, which they plainly do. A harm that requires a perpetrator needs continuous supply — someone has to keep doing it, and can be caught, shamed, or stopped. A harm converted into infrastructure needs nothing. It requires only that nobody rebuild it. Neglect is sufficient. Malice is redundant.

There is a literal map. In September 1861 the U.S. Coast Survey published a shaded map of the distribution of the enslaved population of the Southern states, drawn from the 1860 census — among the first American attempts to turn census figures into cartography. It was a war document, and its function was targeting: it showed where the Confederate economy's labor was concentrated, and therefore where a strike would do the most damage. Lincoln consulted it heavily enough that Francis Bicknell Carpenter, painting the first reading of the Emancipation Proclamation, noticed the habit and put the map in the lower right corner of the canvas.

That map is 165 years old and still serviceable at the county level, for a purpose other than the one it was drawn for. The easy reading is that it still works because the terrain never moved: Reconstruction, the Great Migration, the Civil Rights Acts, six decades of federal programs, sentiment shifting enormously, the shading holding.

But there is a more specific reason it still resolves, and it concerns the map rather than the ground. The 1860 census did not simply record where enslaved people were. It recorded it as a value attached to a county — a unit the federal government had already drawn and has largely left alone since. Everything built afterward inherited that grid. School districts were cut inside counties. Property tax bases are counties. So are health departments, hospital catchments, sheriffs' jurisdictions, the census tracts that nest within them, and most of what the New Deal and the Great Society administered.

So the correlation survives partly because the container survived. The 1860 map and the 2020 mortality file are legible against each other because both are indexed to the same object; genuinely redraw those boundaries and much of this signal would decay, not because anything had been repaired but because the instrument would have stopped resolving it.

That is the same logic running one layer down. Neglect was sufficient for the outcome; it was also sufficient for the category. Nobody decided that the unit built to count enslaved people should become the unit deciding where the money and the schools go. It only had to be the unit already lying there when somebody needed one.

The study cannot do everything, and it is better to say which things. It is observational — no experiment, no way to exclude some unmeasured factor that tracked slavery density in 1860 and tracks mortality now, including one sitting behind the mediators and making them look more load-bearing than they are. The mediators are snapshots rather than measures followed over time, and education enters only as college attainment, which misses everything that happens to a person before eighteen. Most of all: an association carried across 165 years is overwhelming evidence that something persists and nearly no evidence about what would move it. You cannot re-run 1860. The study names the channels; it does not tell you what happens if you go at one.

What it does establish is where the question lives. If the signal has been converted into segregation, tax base, and schooling, then those are not the excuse that explains the finding away. They are the finding, in its current form, still running.

Someone will do this study again in twenty years with a fresh mortality window. The 1860 column will not need updating. Neither will the county lines.