Lies, damned lies, and statistics about China’s Covid-19 death toll
- Beijing has been accused of lying or manipulating Covid death statistics; the same can be said about some of its fiercest critics, however impressive their academic or scientific credentials
In recent weeks, several mainstream British publications have resurrected the old lab leak theory about the origin of the Covid-19 pandemic. But what I find most interesting is another allegation that periodically resurfaces about China having vastly under-reported its Covid death toll.
The current interest starts with The Economist news magazine, which has released an intriguing machine learning program using big data and artificial intelligence to estimate excess deaths on top of the official death tolls from more than 110 countries and territories. The data, at least as it is currently presented on the publication’s webpage, doesn’t include China and Vietnam any more, two communist countries that do not release weekly and monthly statistics, but claim successes in containing the pandemic since the early stages of the outbreak in contrast to many Western countries.
Using the magazine’s estimates, researchers at the Washington-based Center for Strategic and International Studies (CSIS) claimed in an October article that the actual death toll in China is in excess of 600,000. Beijing currently claims a figure of 4,600-plus deaths.
Writing a three-part series in Forbes this month, George Calhoun, a professor specialising in quantitative finance at the Stevens Institute of Technology in New York, does even better by claiming the actual death toll is 1.7 million.
“The 800-to-1 ratio of US-to-Chinese mortality rates is a statistical, medical, biological, political and economic impossibility,” he wrote in the article, titled “Beijing Is Intentionally under-reporting China’s Covid Death Rate”, in which he equates under-reporting with lying.
But why was it impossible, when Americans had Donald Trump as president who denied the outbreak was a threat and pronounced it would disappear quickly on its own in the early months of its wildfire spread across the United States in 2020?
Calhoun continued: “China is another story. Its official statistics understate the Chinese Covid death rate by 17,000 per cent (according to The Economist model).
“In fact, based on excess mortality calculations, The Economist estimates that the true number of Covid deaths in China is not 4,636 – but something like 1.7 million. That is, China’s cumulative death toll is likely at least double that of the United States.”
I love the words “in fact”; it must be a slip of the pen. After all, we are dealing with estimates here.
In my line of work as a reporter, it’s usually think tank fellows and university professors who develop computer models to make new, preferably interesting findings, which are then reported in the media. In this case, it’s the academics who are borrowing from the computer model of a news publication. Interesting, to say the least!
In fact, a number of statistics specialists have since pointed out the dangers of extrapolating death toll estimates for countries whose data are not included in The Economist model. And that is exactly what Calhoun and the CSIS researchers have done. The CSIS piece, “Is China succeeding in shaping global narratives about Covid-19?” was updated last month, and references to the modelling of excess deaths in China from the magazine have been removed. Calhoun also moderated his claim somewhat in the last of this three-part series this week.
The Economist model uses a machine learning technique called gradient boosting, which assembles weaker prediction models to try to boost predictive outperformance using massive data regressively. In a series of online posts, Stuart Gilmour, a professor of biostatistics at St Luke’s International University in Tokyo, has cautioned against extending the magazine’s AI model to countries such as China and Vietnam.
“Let’s talk about the dangers of using machine learning for analysis when you don’t have data,” he wrote.
“The Economist [models] have limited data sources – they use complete mortality data from the human mortality database and the world mortality database, but these only cover about 110 countries. They don’t have anything for China or Vietnam, for example. They predict deaths in other countries based on the countries with data.”
Then, Gilmour goes in for the kill.
“This has a huge and simple problem: you can’t apply relations that exist in countries with full pandemics to countries where Covid is contained,” he wrote. “A model can’t tell that the low testing rate in China is because of no cases – it will assume under-testing.
“China does not have half a million excess Covid deaths, Vietnam doesn’t have 100k, lots of Asian countries are doing better than the USA, and fancy models that serve to reinforce Western wishful thinking don’t help.”
The latest offering from Calhoun is titled: “Anomalies In The Chinese Covid Data – Evidence Of Manipulation?” Note the question mark. Gone is that old certainty about “in fact, there has been X million of excess deaths”.
He no longer relies on The Economist model, which he says “has attracted much attention”. Instead, he now cites data on “crude death rates” from the World Bank and the United Nations between 2018 and 2021. But even he admits the UN data for 2020 and thereafter are projections that do not include any impacts from the Covid-19 pandemic. That adds another layer of uncertainty to Calhoun’s previous claim about a million or two of excess deaths in China.
Now his mistrust of official Chinese statistics is not entirely unjustified. He wrote: “It would be nice if we had similar data from China, where the virus originated. Beijing refuses to provide it.”
Of course, it would be nice. But it’s doubtful any level of disclosure and transparency from China would satisfy critics out to pick a fight. Be that as it may, Calhoun wrote: [The] effectiveness of China’s extreme public health countermeasures in slowing the spread of the disease … cannot account for the absence of mortality outcomes in people who do contract Covid. Once a person is infected, lockdowns and masks have no further effects on the medical prognosis.
“So how to account for this anomaly? The most likely explanation – really the only one that can explain this claim of perfect success in the midst of a chaotic public health crisis – the only explanation is the political explanation: Beijing ordered the mortality number to become 0.0 per cent.”
For the sake of argument, let’s accept Calhoun’s point about 0.0 per cent mortality rate being impossible after a certain official cut-off date. We still don’t know how much China has been “under-reporting”; The Economist model and UN data are no help. At most, they raise some questions about China’s data anomalies.
Well, I can go with that, but that doesn’t take us very far, if not back to square one.