Somebody has to count¶

A park that lays off its rangers loses two things: the people who stop poachers, and the people who write poaching down. After covid took away tourism fees, IPBES reported more poaching, in one place with a citation and in another as anecdote. Much of the record of poaching comes from patrols, so after layoffs a fall in recorded poaching could mean less poaching or less looking. The count moves with the thing it counts, and with the counter.

Emergence records hide the same trap. A new disease in Maryland gets written up and the same disease in Kinshasa may not, so Allen’s group corrected for how hard anyone was looking.

Some counts cannot be checked against the world at all. In 2023 Marani and colleagues corrected their record of epidemics. Their code had counted events in two successive years as though they fell in one. After the fix, an epidemic as intense as the 1918 influenza pandemic recurred on average once in 879 years instead of about 400, so the error had made such epidemics look more than twice as common. An audit of the code found it. Future epidemics could not have settled between 400 and 879 years within any lifetime.

Marani’s record counts epidemics by how intense they were; the chain counts outbreaks that spread far enough to be called pandemics. The record can check the chain’s output but does not supply it, and a tally of how often an outbreak crosses that line is still missing. A credit rule, a ranger’s wage or a misinformation campaign can be tested sooner on what it changes directly: hectares cleared, poaching, doses given.

Neither one did it¶

A wave that fills the wards may also be a wave of older or sicker patients, so fuller wards and a higher fatality ratio can rise together without one raising the other. Comparing patients of the same age, in the same week, in hospitals at different occupancy narrows that down. It still leaves whatever sends sicker patients to fuller hospitals.

Imported cases and travel restrictions can share a cause too. Gwee’s group found imported cases reported before the restrictions, but the outbreaks in Europe could have driven both.

Easier said than done¶

Loomba’s trial moved stated intent by about six percentage points among people who had said they would definitely take a vaccine. Coverage is counted separately, from doses given. Joining the two needs the same people’s intent and vaccination records.

Dates that lie¶

Dates can show which came first, if they record when the thing happened. Brockmann’s arrival dates are reporting dates, not arrivals. Whether trust falls after deaths rise, or restrictions follow reported cases, needs the date of each event, not of its report. Coming first still does not make a cause, since a shared cause can put either one first.

The uncounted¶

Nobody counts the four things that set how often a pathogen crosses into people: how many reservoir animals are infected, how often people meet them, how likely a meeting is to infect, and how many pathogens the animals carry. Lloyd-Smith’s group found the contact rate and the chance of infection “typically arise from multiple and largely unobserved ecological, behavioural and physiological factors”.

Clearing, wildlife trade and warming all bring people and infected animals together, and laying off rangers adds to the trade. Brazil’s credit rule is measured as far as the clearing it prevented, about 60% below the counterfactual, and nobody measured what it did to contact.

Antibodies can find infections nobody saw. Ballmann’s group found them in 34 of 72 lactating cows on a Dutch farm where the virus had first turned up in a cat. Surveys in people living beside cleared land, and beside forest still standing, would find infections that never reached a clinic.

A difference between the two places would show where infections are commoner, not that clearing caused them, because people beside cleared land may also hunt more or keep more livestock. Like Ballmann’s, the surveys would not say when or how each infection happened, or which animal it came from.

Once a pathogen is in people, nobody counts how easily it spreads or how far it has adapted. Antia’s chains are countable once primary cases can be told from secondary ones. Caserta’s sequences showed that a farm cat’s virus came from the cattle. Along a chain of human cases they can show whether the virus changed; telling who infected whom often needs contact tracing as well.

Only two by design¶

Two studies in this series were built to tell a cause from a coincidence. Loomba’s group decided at random who saw the misinformation. Assuncao’s group compared municipalities before and after the credit rule, against what clearing would have been without it. Bollyky’s group call their own study ecological and “not designed to provide information about the causes”, and the series leans on its associations between age, body weight, trust and a wave’s infections and deaths.

Whether fuller wards raise the share of patients who die, whether deaths make people cut contacts or lose trust, and whether more doses mean fewer infections, no study in this series has tested. Each question sits between numbers counted every week: cases, beds occupied, staff off sick, deaths, doses given. The numbers exist.

First counts¶

Travel and farm movement can wait, because Chinazzi’s and Pulliam’s groups have already modelled those routes from outbreak data.

• Antibodies in people near reservoir animals, where land is cleared and where it is not.

• Occupancy, staff absence and outcome, patient by patient, across hospitals in the same week, to see whether patients die more often when the ward is full.

• Vaccination intent and doses given for the same people, and antibodies across the population, to see whether misinformation that changes what people say also changes who gets vaccinated, and how many end up immune.

• Trust surveyed before, during and after a wave, against deaths per week, to see whether trust falls after deaths rise, and how long after.

• Park budgets, ranger numbers and a poaching count whose coverage does not fall when the patrols do, as visitors leave and return.

• Sequences along chains of human cases, with contact tracing, for chain length and for any change in the virus.

The series began with a slogan, pandemic risk as emergence plus propagation plus fragility. Multiplied instead of added, it became a chain whose output no record supplies. Its parts can be counted. Blood samples from villages beside new pasture, a ward’s admissions set against its staff rota, and a park’s accounts set against a poaching count that does not depend on its rangers could each be read within a few years. Each would measure a link the chain now only names. None would say how many pandemics start next year.