Snow's Cholera Map Was Drawn After the Pump Handle Came Off
On 31 August 1854, cholera broke out in Soho. About 500 people living within a few streets of each other died in the next ten days. On the evening of 7 September, a physician named John Snow met the Board of Guardians of St James parish and argued that the water pump on Broad Street was the source. The handle came off the next morning.
The version of this story that gets told in data circles goes: Snow plotted the deaths on a map, saw them cluster around the pump, took the handle off, and the outbreak ended. The map is held up as the moment a picture of data changed the world.
Almost every clause in that version is out of order.
What the record says
Snow did not have the map when he went to the Board of Guardians. He had a list of deaths from the General Register Office and the results of walking the district asking households where they drew water. The map was first shown on 4 December 1854, at a meeting of the Epidemiological Society of London, three months after the handle was removed. A revised version with a dotted line marking walking distance to the nearest rival pump went to the parish Cholera Inquiry Committee with a report Snow dated 12 December. A man named Edmund Cooper had already drawn a death map of the same streets for the Metropolitan Commission of Sewers before Snow started his.
The handle coming off did not end the outbreak either. Snow said so himself. He wrote that the attacks "had so far diminished before the use of the water was stopped" that it was impossible to tell whether the pump still contained the contamination. The epidemic was already in steep decline, and a large share of the residents who could leave Soho had already fled.
And the authorities were not convinced. The Board of Health investigated and attributed the epidemic to miasma, bad air, the prevailing theory of the day. When the outbreak subsided, the parish put the pump handle back on.
So the famous chart was made after the decision, did not cause the outcome it is credited with, and failed to persuade the people it was aimed at. That sounds like a story about a chart that did not work. It is the opposite. It is a story about what charts are for.
What did the work
Two things, neither of them a picture.
The first was a controlled comparison Snow ran across south London during the same epidemic, which he called an experiment "on the grandest scale." Two water companies served the same neighborhoods, sometimes the same street, with their pipes intermingled house by house. Southwark and Vauxhall drew from the Thames downstream of the sewers. Lambeth had moved its intake upstream to cleaner water. The households were alike in everything except which company's pipe came through the wall. In the 1854 epidemic, the Southwark and Vauxhall houses died of cholera at fourteen times the rate of the Lambeth houses. That is the evidence. It has a treatment group, a control group, and a mechanism that explains the gap.
The second was legwork by Henry Whitehead, the assistant curate at St Luke's in Soho, who knew the streets and the families in them. Whitehead went door to door in the streets he served and asked who had drunk from the pump and who had not, including the people who had stayed healthy. His interviews turned up the index case: a five-month-old girl named Frances Lewis, who died on 2 September, and whose mother had rinsed the soiled nappies into a cesspit that sat about three feet from the pump's well. That is the causal chain, found by a man walking and asking, not by a man plotting.
The map, when it finally appeared, was a way of showing a committee what the walking and the comparison had already established. It was the argument's illustration, not its proof.
The mechanism, and why we run it backwards
Here is the part that changes what you do Monday.
A chart can show that two things sit near each other. It cannot show why, and it cannot rule out the other explanations. In Soho, the deaths clustered around the pump, but they also clustered around the busiest streets, the poorest housing, and the worst drains, all of which the miasma theory could claim just as well. The cluster was consistent with Snow's theory. It was also consistent with the theory he was arguing against. A picture that is consistent with both sides persuades neither.
What breaks the tie is the comparison that holds everything else constant. The south London households were a natural experiment because the pipes were intermingled: same air, same streets, same poverty, different water. Once that comparison exists, the map becomes readable, because the reader already knows which explanation survived.
Most reporting runs this in the opposite order. A number moves. Someone builds a chart. The chart goes in front of a decision-maker with the implicit claim that the picture is the evidence. The decision-maker, if they are any good, asks the miasma question: what else could produce this shape? And the meeting ends with the chart being argued about, because nobody built the comparison that would have answered the question before it was asked.
Snow's sequence was: hypothesis, comparison that isolates the mechanism, legwork that finds the causal chain, and then, last, the picture for the committee. The modern sequence is often: picture, then argue, then maybe go find out.
The order to run it in
When a number needs to change a decision, not just decorate one, this is the order that works.
Start with the rival explanations. Write down the two or three stories that could produce the pattern you see, including the boring one. The exciting hypothesis usually has a boring cousin, and the boring cousin is cheaper to check.
Then find your Lambeth. Look for the comparison that already exists in your data where the thing you suspect varies and the other things do not: two regions on different payment processors, two cohorts that signed up under different pricing, two stores where only one changed the menu. That comparison is worth more than any amount of trend-line, because it answers the question the trend-line only raises.
Then walk the street. Pick five rows from the group that moved and read them end to end, the way Whitehead read households. Find one case where you can trace the whole chain from input to number. If you cannot find one, the pattern is still a hypothesis, however good the chart looks.
Then, and only then, draw the picture. Make it for the committee, in their language, with the comparison in it. It will land, because by that point it is showing them something that is already true.
Snow's map is still a very good map. It just was not the evidence. The evidence was the fourteen-to-one ratio and a curate's notebook, and the picture came after. That is the right order, and it is the one most teams would have to reverse to match.