The 78% drop in Covid that was four times the water
Showing the misleading chart
A patient newsletter we drew from California’s wastewater surveillance data reports that the Covid virus in San Francisco’s sewage fell 78% in four days in December 2022. The sample was taken on a rainy Sunday, when the city’s combined sewers sent the Southeast plant four times its usual water. Count the virus instead of its strength per litre and it fell 11%, and three days later it was the highest of any sample in the six weeks.
01The claim
The Covid virus in San Francisco’s sewage fell 78% in four days, from 936,000 copies per litre on Wednesday 7 December 2022 to 207,000 on Sunday 11 December.
02The trick
Both concentrations are the state’s, and the drop is real for what it measures: copies of the virus per litre of sewage. But a litre of sewage is not a fixed amount of anything. San Francisco’s sewers are combined: the same pipes that carry sewage also collect the rain off its streets, and 1.29 inches fell downtown on 10 and 11 December. The Southeast plant’s flow that Sunday was 184.8 million gallons, against 45.7 million on the Wednesday and a dry-day median of 43 million across the six weeks. The same virus shed by the same people was spread through four times the water, so each litre held about a quarter as much. The concentration fell 78% because its denominator rose 4.04-fold, not because its numerator shrank. Nothing on the chart shows the water. The axis says “per litre”, starts at zero and runs to the top of the data, every sample is plotted, and the line still falls off a cliff that is really a rainstorm. The newsletter’s own window had already shown the pattern once: rain on 1 to 5 December doubled the flow to 93 and 86 million gallons, and the concentration fell 49% and 55% from 30 November while the amount of virus moved far less (284, 292 and 235 million copies per resident). (The newsletter is our own demonstration, dated as if written on Monday 12 December 2022. The clinic, its newsletter and its advice are invented, and every measurement is real.)
03The fix
The redraw puts the water back. Above the newsletter’s line it draws the day’s rainfall and the plant’s flow on their own axes, so both wet spells are visible, and below it the load: each concentration times the day’s flow, divided by the 750,000 people the plant serves. On 7 December that is 936,037 copies per litre times 173 million litres, 162 trillion copies, or 216 million per resident. On 11 December it is 206,679 times 700 million litres, 145 trillion copies, or 193 million per resident. That is a fall of 11%, smaller than several dry-day swings in the same six weeks, and it lands at the median of all 18 samples (194 million). Wet-day loads are noisier than dry ones: the next day’s sample, with the flow still at 95 million gallons, works out at 147 million per resident. So the safe reading is that the storm days tell you little either way, not that nothing changed. A second ruler that needs no flow meter agrees. Labs also measure PMMoV, a plant virus people eat in peppers and shed fairly steadily. It fell 71% in the same Sunday sample, so the ratio of Covid to PMMoV fell 25%, not 78%. And the samples after the newsletter show which way things were going. By 14 December the rain had drained away and the concentration was seven times Sunday’s, 1.45 million per litre. The load was 350 million per resident, the highest of the six weeks and more than three times 13 November’s 104 million, before settling back near the median by 22 December: no sign of the fall the newsletter announced. Wastewater dashboards report loads or marker ratios partly for this reason. When a series is a concentration, find out what the volume did on the day it moved, and read a single wet-day sample as weather, not news.