Remember in 2020 there was a map showing how well-prepared different countries were, which received widespread derision for its terrible accuracy? I analyzed the underlying data to see how poorly it predicted pandemic outcomes.
The
#economist
is publishing a regularly-updated model of excess deaths due to COVID, which is being used to tell just-so stories about countries the Economist doesn’t like. Let’s talk about the dangers of using machine learning for analysis when you don’t have data.
For those wondering at why the Australian government would abandon its Afghan “allies” and comrades of ADF soldiers to an uncertain fate after the Taliban have taken over
#Afghanistan
, a little history lesson in how Australia has historically treated Afghan refugees
Weird how there’s a storm of complaint online about Eugene Gu, but Naomi Osaka’s decision to represent Japan at the Tokyo Olympics got barely a mention. I wonder what the difference in these two cases might be?
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
This figure shows the change in birth rates in Japan from 1985 to 1990, with the municipalities where the change was >30% shown in red. Let’s talk about the problem with using demographic data to try and prove genocide.
@dakekang
and
@huizhong_wu
your reporting on prison rates in your latest article about Xinjiang is wrong and misleading. Assuming your linked list is true, the imprisonment rate is not “the highest anywhere in the world” and your numbers are just wrong.
And finally I’ll say this: if your government had all these systems in place but still threw you to the wolves the moment those policies looked like they might cost their rich buddies a few pennies, imagine what they’ll do when global warming bites!
@aspi_icpc
@Nrg8000
@jleibold
I reanalyzed the data from your “family de-planning” report. Contra your findings, “coercive birth control policies” in Uyghur-majority areas had less effect than in Uyghur-minority areas. How do you explain this?
The same week that every western media organization and anti-China grifter is writing 1000s of words on Peng Shuai, this happens. Will the IOC be required to call this woman to confirm her safety? Or is international outrage selective?
I found clear evidence that the more “ready” a country was on the score, the worse it performed. Every 1 point increase in the score meant a 5% increase in total infections. I plot this result here, with some key countries identified, and the regression results.
But what we actually saw in practice in 2020 was that although rich countries had the policies in place, most refused to enact them for fear of endangering the economy they depended on, and the libertarian policies that underpin it.
None of these preparedness measures matter at all if your country has no political will to protect the lives of its people, or any desire to make even the vaguest voluntary or enforced restrictions on social behavior.
This figure shows the number of births that were “lost” in Japan due to falling birthrates since 2010. Nearly 1 million over just 10 years! Shocking! But no outcry from western thinktanks. Let’s discuss attributing sinister motives to good policy in
@adrianzenz
's latest work
Of course it includes the usual Economist “intelligence” unit nonsense about governance. E.g. the USA was given the second best rank for “government effectiveness”, a fact that was released to the public about the time Trump was advising Americans to inject bleach.
Wow the NYT is so concerned about China’s Uyghur minority that they denied the Uyghur ethnicity of one of its Olympic representatives. Definite good faith reporting on minority rights in China here …
China chose two athletes — including one it said was of Uyghur heritage — to light the Olympic cauldron at the
#OpeningCeremony
in Beijing.
The country has conducted a mass detention campaign targeting Uyghur Muslims that the U.S. has called genocidal.
It doesn’t work because like many policy assessments it assumes that the documented policies will actually be implemented, and that there is some relationship between the policies on the books and their implementation.
The economist believes China has had as many excess deaths as the USA. This is simply not credible. It’s not possible that China would have an epidemic on that scale and nobody would know.
It’s not as bad over here. Western governments failed on COVID-19, and no amount of dodgy modeling and just-so stories by their mates in the dismal “science” will let them off the hook for their criminal response to this challenge.
@amyyqin
@chenweihua
@JeromeTaylor
There I was, trapped in my hotel room, drinking room-service 獺祭 on my employer's tab, while I waited half of a shift for the most effective COVID-19 prevention system in the world to protect my health. It was authoritarian hell, kids, you'll never understand what I suffered!
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. A model can’t tell that the low testing rate in China is because of no cases – it will assume under-testing.
Before we do, let’s just take a moment to appreciate the deep racism of this picture from a report using the Economist's numbers by the Centre for Strategic and International Studies. When you see a cover pic like this, you can guess what the contents are gonna be.
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. See also the IHME’s estimates of COVID deaths in Japan.
Also the single biggest determinant of a successful response is adherence to basic public health measures as recommended by the WHO – test, trace, isolation – and we know most rich countries refused to do this.
Last year I wrote a long thread about a model in the Economist that claimed China suffered 700,000 excess COVID-19 deaths, which I said was preposterous. This thread is a follow up to that, with new info.
The
#economist
is publishing a regularly-updated model of excess deaths due to COVID, which is being used to tell just-so stories about countries the Economist doesn’t like. Let’s talk about the dangers of using machine learning for analysis when you don’t have data.
To summarize: The Economist is doing a shoddy version of work that has already been done by experts in epidemiology, and conveniently confirming the suspicions of western ideologues that those shifty Asians are hiding the truth, and COVID-19 is really bad over here.
They seem to have included some index of political freedom from “Freedom House” but failed to include any variable indicating whether the country *actually has a COVID epidemic*. Hence, the model assumes China and Vietnam are under-diagnosing. This is a bad and simple error.
It’s so exhausting dealing with this torrent of bad-faith data “analysis” from the Economist. The latest is an analysis of satellite data on night time lights and gdp growth that suggests “autocratic” states fiddle their numbers on gdp growth.
Basically, this scale failed the one time it had to be used. I think the JHSPH people need to go back to the drawing board on their assessment of global preparedness, because this index doesn’t work. And once again, the Economist needs to stay far away from public health!
There are many other problems with the Economist’s work. First, they seem to apply a linear regression model to deaths, rather than over-dispersed Poisson regression. This allows the prediction of negative deaths in a country!
This table shows the rate of imprisonment of minorities in a couple of different countries. Let’s talk about genocide denialism among China “experts” who are laser-focused on proving genocide in Xinjiang while they ignore the legacy of genocide in their own countries.
This data is simply wrong. So how is it possible the economist got it so wrong? And why do they publish anyway? The answer to this question is also relevant to other projects that aim to estimate health data where that data is missing, like the
#globalburdendisease
studies.
They then build a massive model based on this data to estimate what the true excess mortality might be using a machine learning method called gradient boosting. They predict deaths in other countries based on the countries with data.
The economist 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.
Having seen who is using this data and hazarded a wild guess at why, let’s look at some results. First, let’s compare the UK and Vietnam. According to the Economist, Vietnam had nearly as many excess deaths due to COVID-19 as the UK, despite containing the virus for a year.
Since the 1st August 2021 when Vietnam’s outbreak began, it has never once had more daily cases than the UK. Yet this country with this epidemic experience is supposed to have had as many deaths as the UK.
Hong Kong has experienced a wave of
#covid19
cases and deaths, and some media are blaming this on Chinese COVID vaccines, saying they don’t work. Let’s talk about whether this is true, and the implications for global vaccine equity of vaccine misinformation.
And who managed the 2001 election campaign which slandered refugees from the Taliban with the “children overboard” lie and put anti-refugee rhetoric at its centre? Scott Morrison, the current prime minister of Australia who is being faulted for failing to save Afghan workers.
First, you report that US prison rates are 364 per 100000. This is not correct. The number is actually 537, but you didn’t include US Jails in your figure. Please correct it. We don’t need more articles understating the USA’s incarceration epidemic.
As part of that operation the navy fired into the water in front of the ship. The prime minister John Howard claimed the asylum seekers had thrown their children into the sea. This was a lie, intended to slander refugees during an election that Howard wasn’t sure he would win.
In case you have doubts about the quality of western coverage of Asia, this guy was Japan bureau chief when the Nagasaki mayor was shot by a right wing assassin, but he apparently didn’t cover it.
In my six years in Japan as a correspondent and
@nytimes
bureau chief the only shootings I covered involved yakuza, arguing over territory. Even those were rare. For political assassinations you have to go back to the ‘30’s. Prayers for Abe Shinzo.
Next, they treat weeks and months as “fixed effects”, a weird obsession within economics, rather than fitting a spline or some other model that is less arbitrary. This is not how we do this in epidemiology.
So how did it perform? I downloaded the 2019 data and linked it to WHO data on total COVID-19 cases, deaths and vaccinations, and UN population data. I then used Poisson regression to explore the relationship between the total readiness score and these outcomes.
This is ironic because they reference a BMJ paper of excess deaths in Hubei which tells them to do all of these things (except the prediction limits). There is a method for excess deaths, called the Farrington method, used by the CDC. But they ignored it!
The map is based on the Global Health Security Index, a numerical measure of pandemic preparedness compiled by Economist Impact in collaboration with Johns Hopkins and others. There is a published report, with a clear methodology.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Basically Zenz et al are weaponizing demography to repurpose family planning campaigns as genocide. This is very dangerous. If it is successful, in future any family planning campaign can be spun into an accusation of genocide.
For example, using data on covid tests per person, positivity rates, human development indices etc they can guess what deaths might be in countries without detailed death data. Gradient boosting selects the “best” variables and their effectiveness.
Please correct this incredibly inaccurate article so it is reflective of the reality of mass incarceration in the USA, in the interests of the 1.8 million people currently imprisoned in the USA, and the 10s of millions who have experienced the brutality of the US carceral system.
@dakekang
@NuryVittachi
Your "best estimates" are just an extrapolation from interviews with 8 arbitrarily selected people. If I did that in the USA I could get a "best estimate" of the number of people in prison in the 10s of millions. That's not a "best estimate", it's making up numbers
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Zenz and the ASPI need to 1) Retract these reports; 2) Inform media and NGOs who have broadcast them that they are built on bad data, bad statistics and bad demography; 3) commit never to report on these issues again
Managing the offshore processing camps costs about 800 million dollars per year, because Australia would rather burn money than allow a single Afghan refugee to land.
There are a few weird problems with the index, of course. The current version includes a measure of access to paid sick leave but the infamous 2019 one doesn’t. It doesn’t seem to have anything about union membership or workplace rights, marginalization or stigma.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
This cannot be allowed to happen. Family planning is essential to sustainable communities, crucial in women’s empowerment, and critical to poverty alleviation programs. This misuse of statistics needs to stop and needs to be stopped.
Second, you say that the Konasheher country rate (3789) is “the highest known imprisonment rate in the world”. This is false, because you compare a county with countries. There are *many* counties in the USA with higher rates. See e.g. Indiana.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Decisions are being made internationally partially on the basis of these erroneous reports; they’re being debated at the UN. These organizations need to withdraw these reports and commit never to do this kind of shoddy work again.
It’s no surprise that these Afghans have been abandoned, and any senior media or policy commentator in Australia who tells you they are surprised and disappointed is either a fool or a liar.
Also, as I have shown before, the rate of imprisonment of Canadian indigenous people is higher than this, and similar in regions – e.g. in Nunavut in 2019 it was 3400 per 100,000.
And don’t doubt that almost every senior journalist or defence policy think tank worker who is crying over the abandonment of Afghan workers in Kabul today, was either defending or ignoring this policy over the past 20 years.
In a month this same newspaper will unironically publish articles about how China has an unelected leader for life, dissent is impossible and the entire nation’s media is captive to the state.
In September 2021 an anti-vaccination meme started which claimed 70 members of Pfizer’s “investment board” are members of China’s Communist Party. Let’s look at how major news services and anti-China thinktanks drive anti-vaccination sentiments that kill Americans.
Next I used the same Poisson model to compare cumulative deaths with the score. The same result holds – about a 4% increase in deaths for every unit increase in the score. Here I plot this with the same countries identified, and the regression results.
The data is available from their website, giving 195 countries an overall score and also scoring them on six sub-domains which measure things like anti-microbial resistance (AMR) preparedness, adherence to international health regulations, and so on.
If you want to see how excess deaths are actually calculated (when data is available), check out papers by some of my colleagues and I that find limited changes in Japan.
A general problem with these big multi-country studies that try to fill in data for countries they can’t acces is that they use data-driven methods to try and infer missing information. It’s at the core of the global burden of disease (GBD) study and it’s dangerous.
This fit isn’t great – there isn’t a strong relationship between pandemic preparedness measured with the Global Health Security Index and total mortality from COVID, but you’d think the “better prepared” countries would not be in the top half of the mortality figures …
In August 2001 the MV Tampa rescued 433 refugees at sea. 244 of them were Afghans, fleeing the Taliban, and following the laws of the sea Tampa attempted to land them in Australia. The Australian government refused to take them.
They calculate excess deaths as a difference from the predicted deaths, even though deaths naturally change every year. They should generate uncertainty ranges around the predicted deaths for 2021, and then report excess/exiguous deaths *over or under those limits*.
@aspi_icpc
@Nrg8000
@jleibold
In fact we can build a better model that shows “coercive birth-control policies” were less effective in Uyghur areas. It explains more of the variation than the figures in your report. Why did you not present this model?
After the “Children overboard” incident, the government hired a dodgy company that claimed to be able to identify the “true” origins of asylum seekers from their accent and speech. This company redefined Afghan refugees as Pakistani tribesfolk so their claims could be rejected.
Fortunately this data is available for you to check, so you could see how large swathes of the USA imprison people at rates higher than those you say are “the highest known in the world.”
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Zenz found sterilization rates of 2.5 per 1000 adult XJ women. Scary stuff! Or is it? In 1990s UK 5 per 1000 adult women were sterilized every year, and recent surveys find 3-7% of UK women are sterilized. :
But anyway, regardless of specific flaws, the 2019 scale completely fails to predict any form of effective pandemic response, and it doesn’t look like the 2020 one is better. Although there is a single paper showing some sub-scales predict outcomes, the scale's mostly useless.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
This paper Zenz et al don’t reference also surveys 1000 Uyghur living in Hotan county, and finds 68% of them want less children. Why be surprised that with a properly managed and funded project birth rates drop by 50% in a few years?
[BTW as an aside, let’s compare Poisson regression of this data with the preference of many on
#econtwitter
, OLS regression of logged rates. Compare the fit – analyzing logged rates is never a good plan!]
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Why don’t Zenz et al study the TFR, which measures actual fertility, rather than birth rates? Because they have no connection with Chinese scholars (and don't even reference them!)
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
In future the US far right will no doubt start to recast effective family planning campaigns in low-income countries this way, just as they have tried to do to Planned Parenthood in the USA.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
As you can see from that figure, birth rates in some municipalities (市区町村) in Japan dropped by more than 50%. Here are some example trajectories from 1970 – 2010.
Finally I modeled the relationship between vaccines (% of pop with >=1 shot) and the GHSI, using OLS regression of logit-transformed proportions. Here there *is* a relationship between improved “preparedness” and improved vax coverage.
GBD estimates of depression prevalence in Africa, for example, can be based on a small number of studies in a handful of countries to generate evidence on the whole continent. We should be very careful about this process.
Also, wherever possible these multi-country studies should incorporate data and collaborators from every country they include. The GBD studies do this; the Economist clearly have not. Their conclusions are not reliable as a result.
Between 2001 and now, except for a brief period under Labour, the Australian government has incarcerated all refugees coming by boat in concentration camps on Nauru and Manus Island. A total of 786 Afghan refugees fleeing the Taliban were imprisoned on Nauru in 2001-2002.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
At this time in Japan the oral contraceptive pill (OCP) or other forms of long-acting contraception were not widely available. If drops like this are possible in Japan without any intervention, why are they unusual in China?
Operation Slipper, Australia’s special operations activities in Afghanistan, is the subject of the Brereton report because 19 members have been implicated in the murder of 39 Afghan civilians and various other war crimes.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Neither the birth rate trend in Xinjiang nor the incidence of sterilization are unusual internationally. Chinese studies find high intention to have less children in communities with very high TFR. So what’s going on?
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Do Zenz and the ASPI know anything about fertility intentions in Xinjiang? They make big assumptions about what Uyghur families want, but research by Chinese people living in those areas doesn’t back them up. Why not?
The Tampa’s captain refused to turn around so Australia sent in the SAS. The commander of the SAS force that raided that ship, Vance Khan, was ultimately in charge of a squad that killed tribesmen as part of Operation Slipper in Afghanistan.
Australia is holding 1489 people, for an *average* period of 689 days, none guilty of any crime, or even charged. But not a peep from
@aspi_org
or
@Nrg8000
, despite their concerns about detention of muslims elsewhere. Why silent about Australia's crimes?
So why does the GHSI fail to adequately predict pandemic outcomes? You would think that a scale measuring pandemic preparedness would predict the response to the first actual pandemic that happened, wouldn’t you?
@dgurdasani1
Yes, every time a zero Covid country gets infected it’s presented as a failure of that country’s efforts rather than the fault of the countries that are exporting the virus to reinfect the world because they won’t control it
The scale is not badly designed – it’s a careful, thoughtful compilation of detailed info about 195 countries. For example, here is detail on just one of the scale elements (1.3.4, transport) for one country (Belgium). This is a detailed assessment!
@aspi_icpc
@Nrg8000
@jleibold
This figure shows different birth rate declines between prefectures. Higher baseline birth rates have larger drops – but this effect is smaller in the southern prefectures. How does this support your claim about coercive birth-control policies?
In October 2001, another boat carrying asylum seekers entered Australian waters. Nobody seems to have bothered reporting who these asylum seekers were, or we don’t know, because the navy was sent to drive them out of Australian waters.
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Let’s look at Prefectures, which have more stable populations, plotted by median drop: 50% of the prefectures saw a drop bigger than 17%. Prefecture populations in Japan range from 600,000 to 11,000,000, the same range as in Xinjiang.
@aspi_icpc
@Nrg8000
@jleibold
In this model, we see birth rates drop more when the baseline is higher. In Uyghur-minority areas, every 1% increase in baseline birth rate gives a 4.4%(point) drop in birth rates. In the southern prefectures, this drop is only 1.0%(points). Why?
@adrianzenz
@nrg8000
@aspi_org
@guardianAus
Next, let’s explore what population control means in Xinjiang. This publication gives a total fertility rate for Hotan county, Xinjiang of 3.3 – 3.5: on average women in Hotan have 3.3 babies in their life.
Australia’s “Pacific solution” dumps vulnerable and desperate people in cruel concentration camps far from Australia, and abandons them there with no hope of a future. The residents of these camps have constantly protested but the government just increases the cruelty.
@aspi_icpc
@Nrg8000
@jleibold
Third: Why did you use birth rates instead of total fertility rate (TFR)? Birth rates are vulnerable to changes in elderly, male and child populations and can be affected by migration flows, while TFR is not.
Since 2012 about 4000 people have been incarcerated on Pacific islands. Of the 217 people still incarcerated, 165 (72%) have been there over a year, and 3 have been there more than 4 years. They will never be resettled in Australia.