Snow's Map, and What It Actually Did
The founding story: in 1854, John Snow mapped cholera deaths in Soho and found them clustered around the Broad Street pump. He had the handle removed. The outbreak ended.
The story is told as data visualization defeating superstition, and nearly every part of that telling is wrong in an instructive way. The outbreak was already declining when the handle came off. And the map did not generate Snow's hypothesis — he already believed cholera was waterborne, and drew the map to argue it against a medical establishment committed to miasma.
What the map genuinely did is more interesting than the legend: it made a spatial argument checkable. Its power was in the anomalies — the brewery workers who drank beer and did not die, the widow who had Broad Street water delivered across town and died of it. Those are the cases that no miasma theory could accommodate, and you can only see them once the deaths are on a map with the pumps.
That is exactly what GIS does. It does not answer questions; it makes spatial arguments visible, checkable, and falsifiable — and it is worth being clear that Snow's map was a rhetorical instrument, deployed in an argument he was already having.
Layers, and the Question They Enable
GIS is layers over shared coordinates: terrain, rivers, roads, settlements, soils, parishes, deaths, prices. The value is in the overlay — each layer is ordinary, and the relationships between them are where the history is.
| Operation | The question it answers |
|---|---|
| Overlay | What coincides with what? |
| Buffer | What is within X of this? |
| Least-cost path | Where would movement actually go, given terrain? |
| Viewshed | What can be seen from here? |
| Density / clustering | Is this pattern more concentrated than chance? |
Least-cost path deserves emphasis because it is the one that changes historical practice. Straight-line distance is nearly meaningless in real terrain: what matters is the cost surface — slope, rivers, marsh, season. Compute the cheapest route across it and Roman roads frequently fall out of the model, which is both a validation and a tool. Where the model and the road disagree, you have found something: a reason that was not about cost. Politics, ritual, an estate someone owned, a decision.
Viewshed answers questions that were previously pure assertion. Could that signal tower see the next one? Was this castle sited to be seen, or to see? Put the observer at the parapet and the model tells you, and a good deal of confident writing about "commanding positions" has not survived the test.
Georeferencing, and the Trap It Sets
Historical maps must be stretched onto modern coordinates to join the layers — which requires control points, and imports every one of the old map's errors into a system that displays them with perfect crispness.
This is the central danger of the whole enterprise. GIS output looks authoritative in a way that spreadsheets do not. A map renders a guess and a survey identically, and puts a boundary on a screen at sub-metre precision when the source was a sketch. The parish boundary you digitized was approximate, seasonal, and disputed; on your screen it is a crisp line, and three papers later it is a fact.
The related trap is the ecological fallacy: an aggregate pattern does not license a claim about individuals. Districts with more X have more Y does not mean the people with X have Y. Spatial data invites this error constantly, because the unit you can map is almost never the unit you want to talk about.
The rule is the one from the digital-humanities lesson, restated: the method changes the scale, not the epistemology. Ask the layer what you would ask a document — who made it, why, what is missing, at what precision. A crisp line is a claim about evidence, and GIS will render your assumptions beautifully.