A researcher has decoded Beijing’s own propaganda to track a likely disconnect between COVID case counts and the official Party line.
A researcher has decoded Beijing’s own propaganda to track a likely disconnect between COVID case counts and the official Party line.
A t the outset of the pandemic, as Chinese Communist Party officials at all levels of government failed to provide trustworthy information, the rest of the world underestimated the disease.
But those who had witnessed or otherwise studied the Party’s cagey response to the 2003 SARS epidemic knew immediately what they were dealing with. The Taiwanese government sprang into action, crafting its famously successful pandemic response. And in the U.S., one Trump official — a former journalist who while based in China challenged the Chinese government’s SARS-era deceptions — started warning about the virus well before many realized that an outbreak could occur in the United States. People like him understood the lengths to which the Party-state would go to cover up the emergence of a deadly disease, a tendency that persists to this day even though its aggressive measures have brought the virus down from its early 2020 peak.
New research led by Weifeng Zhong, a Mercatus Center senior research fellow, sheds light on Beijing’s public-health deceptions. His team is using an innovative technology that can also examine changes in the Chinese government’s policies.
“We all know that the official number of diagnosed cases coming up from the Chinese authorities was not reliable,” Zhong told National Review in an interview following the release of his latest research paper in December. “But a quantitative question is, How unreliable are those numbers?”
Many have attempted to answer the same question. Some turned to reports about increased cremations in Wuhan, which suggested that official case numbers in early 2020 were being deliberately suppressed by the authorities. On-the-ground reports from Xinjiang shed light on an outbreak there that took place later in the year and that Chinese officials had downplayed.
But Zhong’s research — based on machine-learning analysis of articles that appear in the People’s Daily, the flagship Party propaganda outlet — offers an inventive way “to analyze the Chinese government’s own words.” It’s called the Policy Change Index (PCI).
“The reason why that actually would work is because words are very indicative of intentions in terms of the Chinese government’s policymaking process,” Zhong said. The idea behind the PCI is that, if the tone used by major state newspapers during the COVID pandemic sounds as urgent as the tone used by the same publications during the peaks of a previous disease outbreak, that could call into question the government’s claim of low case counts.
To construct this model, Zhong relied on a historical example — in this case, SARS — that resembles the rhetoric that the Party is using in COVID times. The algorithm analyzes recent People’s Daily articles to compare the tone expressed with articles published during the SARS outbreak nearly two decades ago.
Effectively, this allows the researchers to map a 2020 article onto a day in 2003 — and in the process, to match a COVID-era article with a point on the SARS epidemic trajectory. Articles similar in tone to those published at a particular time during the SARS era suggest that COVID would have been at a similar point at the time of publication. Zhong writes in his paper, for instance, that an article from COVID’s February peak in China resembles the urgent tone the People’s Daily used in March 2003, when SARS was cresting.
Even in a country transparently reporting the incidence of disease, the number of diagnosed cases might not align with the true extent of infection. Serological studies in several countries have shown that true infection rates are higher than what diagnostic nasal-swab and spit tests suggested (a recent Chinese CDC report out of Wuhan suggests that actual case counts were ten times higher than what was initially reported). But U.S. intelligence assessments have also shown that the Chinese authorities deliberately undercounted cases.
Zhong’s research suggests a similar conclusion. Whereas officially reported Chinese COVID numbers reached their peak in mid-February and declined steadily from there, Zhong’s outbreak index declines at a much slower rate, showing that officials were still conveying their sense of urgency through the state-run press. As the officially reported number of cases plummeted, the People’s Daily used language that’s associated with higher levels of infection.
Although some amount of the underreporting can be attributed to officials acting at the local level, state-media articles also suggest that underreporting by higher-ups was most pronounced during regional spikes that followed the one centered in Wuhan. Beijing was placed under a lockdown order in June, well after the initial outbreak had been suppressed. Although official case counts remained low in the capital city, Zhong notes that the People’s Daily sharply emphasized the importance of the lockdown measures “in marked contrast to the numbers, which indicate only about a dozen new cases per day in a city with a population of over 20 million.”
A July outbreak in Xinjiang — the far western region where Party officials have constructed a brutal techno-authoritarian ethnostate — barely registered in reporting of cases. But conditions in that region reflected a different reality, including strict lockdown measures much like the ones implemented in Wuhan and an attendant rhetoric of “wartime mode.” Zhong’s index picked that up, too.
To be sure, this method has its limits. It cannot offer a precise estimate of the true number of cases that China has experienced. And much can change over the course of 17 years, so it’s conceivable that official responses to SARS and COVID might not use exactly the same language.
The index, though, does offer a window into how officials have thought about COVID, how it compares with a previous disease outbreak, and, therefore, what they’re hiding from the world through distorted case reporting.
It has other applications, too, particularly in scrutinizing propaganda for insights about what the Party will do in the future. Zhong has constructed a similar index, called the PCI Crackdown, that uses the timeline of the Tiananmen Square massacre to analyze Beijing’s crackdown in Hong Kong.
That comparison might sound tenuous. But his team found some key parallels in how the government’s language in describing the demonstrators — first “people who have a good heart,” then, eventually, “traitors” — evolved. “We were surprised ourselves, too,” Zhong said. And although the researchers did not specifically predict the National Security Law that since May has been wielded to kill the pro-democracy movement, they detected key changes in the Party’s use of language in April.
So what’s on the horizon? Zhong told National Review that another one of his models — which analyzes the front page of the People’s Daily for accelerating changes in the way that Chinese officials talk about a range of issues — has picked up an alarming trend. “In the second quarter of this year, the Chinese government is showing an unusual emphasis on military aggression on the front page of the People’s Daily.”
Of course, the index notes changes only in what Party officials are communicating to the public, so this is just one data point among many others; it says little about the nature of potential military actions or the location of their targets. But Zhong says that he has observed increased Chinese social-media chatter about invading Taiwan to absorb its semiconductor production.
“So that’s something I think people should pay more attention to, especially in light of the fact that China is thinking about military power in a more salient way,” Zhong said. He’s far from the first to warn about the precarity of Taiwan’s security, and his Policy Change Index is an approximate measure. Still, the insights afforded by his method don’t seem far from the truth — in fact, they cut straight through the CCP’s lies.