What should we believe and not believe about R?

This is from my email, highly recommended, and I will not apply further indentation:

“Although there’s a lot of pre-peer-reviewed and strongly-incorrect work out there, I’ll single out Kevin Systrom’s rt.live as being deeply problematic. Estimating R from noisy real-world data when you don’t know the underlying model is fundamentally difficult, but a minimal baseline capability is to get sign(R-1) right (at least when |R-1| isn’t small), and rt.live is going to often be badly (and confidently) wrong about that because it fails to account for how the confirmed count data it’s based on is noisy enough to be mostly garbage. (Many serious modelers have given up on case counts and just model death counts.) For an obvious example, consider their graph for WA: it’s deeply implausible on its face that WA had R=.24 on 10 April and R=1.4 on 17 April. (In an epidemiological model with fixed waiting times, the implication would be that infectious people started interacting with non-infectious people five times as often over the course of a week with no policy changes.) Digging into the data and the math, you can see that a few days of falling case counts will make the system confident of a very low R, and a few days of rising counts will make it confident of a very high one, but we know from other sources that both can and do happen due to changes in test and test processing availability. (There are additional serious methodological problems with rt.live, but trying to nowcast R from observed case counts is already garbage-in so will be garbage-out.)

However, folks are (understandably, given the difficulty and the rush) missing a lot of harder stuff too. You linked a study and wrote “Good and extensive west coast Kaiser data set, and further evidence that R doesn’t fall nearly as much as you might wish for.” We read the study tonight, and the data set seems great and important, but we don’t buy the claims about R at all — we think there are major statistical issues. (I could go into it if you want, although it’s fairly subtle, and of course there’s some chance that *we’re* wrong…)

Ultimately, the models and statistics in the field aren’t designed to handle rapidly changing R, and everything is made much worse by the massive inconsistencies in the observed data. R itself is a surprisingly subtle concept (especially in changing systems): for instance, rt.live uses a simple relationship between R and the observed rate of growth, but their claimed relationship only holds for the simplest SIR model (not epidemiologically plausible at all for COVID-19), and it has as an input the median serial interval, which is also substantially uncertain for COVID-19 (they treat it as a known constant). These things make it easy to badly missestimate R. Usually these errors pull or push R away from 1 — rt.live would at least get sign(R – 1) right if their data weren’t garbage and they fixed other statistical problems — but of course getting sign(R – 1) right is a low bar, it’s just figuring out whether what you’re observing is growing or shrinking. Many folks would actually be better off not trying to forecast R and just looking carefully at whether they believe the thing they’re observing is growing or shrinking and how quickly.

All that said, the growing (not total, but mostly shared) consensus among both folks I’ve talked to inside Google and with academic epidemiologists who are thinking hard about this is:

  • Lockdowns, including Western-style lockdowns, very likely drive R substantially below 1 (say .7 or lower), even without perfect compliance. Best evidence is the daily death graphs from Italy, Spain, and probably France (their data’s a mess): those were some non-perfect lockdowns (compared to China), and you see a clear peak followed by a clear decline after basically one time constant (people who died at peak were getting infected right around the lockdown). If R was > 1 you’d see exponential growth up to herd immunity, if R was 0.9 you’d see a much bigger and later peak (there’s a lot of momentum in these systems). This is good news if true (and we think it’s probably true), since it means there’s at least some room to relax policy while keeping things under control. Another implication is the “first wave” is going to end over the next month-ish, as IHME and UTexas (my preferred public deaths forecaster; they don’t do R) predict.
  • Cases are of course massively undercounted, but the weight of evidence is that they’re *probably* not *so* massively undercounted that we’re anywhere near herd immunity (though this would of course be great news). Looking at Iceland, Diamond Princess, the other studies, the flaws in the Stanford study, we’re very likely still at < ~2-3% infected in the US. (25% in large parts of NYC wouldn’t be a shock though).

Anyways, I guess my single biggest point is that if you see a result that says something about R, there’s a very good chance it’s just mathematically broken or observationally broken and isn’t actually saying that thing at all.”

That is all from Rif A. Saurous, Research Director at Google, currently working on COVID-19 modeling.

Currently it seems to me that those are the smartest and best informed views “out there,” so at least for now they are my views too.


Rif A. Saurous

Rif - your riff on R is just what we need in this time of sad car commercials. We’re all in this together.

I am seeing a change in local news coverage. Now it’s all how things are getting better. There is also coverage of demonstrations in the state capital. Watch this thing turn sharply.

What's up all, here every person is sharing such know-how, therefore it's good to read this blog, and I used to
go to see this webpage everyday.

I much prefer Python to R.

I was at my computer trying to teach myself R when I saw this post. What a coincidence.

American capitalism can never be shutdown! Reopen America Now!


Live free or die indeed. Who gave these sh*theads the power to take our jobs and put is under house arrest?

The schmucks at Google are not the experts on modeling virus pandemics. Peter Theil got Google right - it is not a technology company.

Here's one of the real experts - Dr. John Ioaniddis at Stanford.

Watch, listen, and learn.


Great email. Shoutout to Rif.

1. Hmm -
"(25% in large parts of NYC wouldn’t be a shock though)."

Back of the envelope - that puts IFR >= 0.4% in NYC ...

2. Who's doing the best synthetic work on how all of this plays out? MR has been great. But I wonder: Is Ed Yong's piece in the Atlantic ("our pandemic summer") the best "high-level, intended for laymen" view about how all this plays out?

Or - what's in the running for that - who is doing the most credible (and thus sufficiently uncertain) mainstream work on what the next 3-18 months look like?

538 has been strong; parts of the Atlantic (particularly Yong, Khazan, Hamblin) seem strong; Robin Hanson's twitter feed has been great; Trevor Bedford, Scott Gottlieb, etc.

Who's keeping it all together?

So the full picture seems to be emerging.

COVID-19 is essentially a 5x deadlier flu. New/unknown and more lethal therefore terrifying, but at its root a seasonal respiratory disease.

If we had let it rip deaths would have been about 250,000 but we went ham and locked down the economy so COVID-19 will instead be a 2x deadlier flu.

If we had let it rip and called it a flu variant the headline would’ve been:
“Deadliest Flu Season in 61 Years”

COVID-19 is 5x-10x deadlier seasonal flu that spreads pretty well before/without symptoms, and there is no existing immunity or vaccine.
2% IFR is reasonable if you don't count asymptomatic cases.
Asymptomatic cases are around 50%, judging from various ships.
So let 'er rip = 1% of 70% of 330 million dead = 20 million dead.
Those are the sorts of figures that make the head spin and got Boris Johnson to quickly backpedal on his plans to "let er rip."
"Average" flu season = 30,000 dead, by the way.
Now of course after the first few hundred thousand died, if the government did nothing, half the population would hide indoors and the economy would crash anyhow.
Soo ... "let 'er rip" is not a solution.

Oh by the way IFR would probably get well above 1-2% if hospitals are asked to treat 10x the number of patients that they have resources for. That's where the 5% IFR in early days in Wuhan came from.

Correction: estimate should be 2 million dead, not 20 million. Point still stands.

Yes, basically as deadly as a flu without a vaccine.

Hits older, fatter people much harder than the flu. Conversely, hits younger people a lot less.

Biggest thing is very long time the virus seems to stick around -- 8 to 21 days. Nobody is trying culture viruses at 21 days but would be worthwhile.

All this was known during Wuhan. About the only big difference we've learned is cigarette smoking doesn't seem to be a risk factor in the west, and obesity is.

Too many people watched Contagion and learned the wrong lesson.

This is very false. COVID-19 is much more dangerous to young, healthy people than the flu is.

In fact, the % increase for young people is actually larger. COVID is more dangerous to both young and old people than the flu, but the increase in fatality rate is actually slightly more pronounced among young people. You only see many more old people dying because old people are more vulnerable to both.

As of 2:30 today, NYC is counting 13,683 dead from COVID-19 (9,101confirmed, 4,582 probable).

If we go with the full 25% infected (instead of 25% in 'large parts of' NYC), that's an IFR of 0.67%.

That's in under two months, and with a lockdown - not 'letting it rip'. (NYC was at about 50 deaths when the lockdown occurred.)

CDC estimates that Seasonal flu has taken 24-62K in the US in '19-20. That's with flu shots and built-up immunity. There is neither for COVID-19.

tl/dr: numbers look like they would be far north of 250k, depending on your definition of 'let it rip'.

You say IFR of .6%
Latest seroprevalence study from LA county says .2%
Diamond Princess IFR .18%
Iceland .17%

Flu IFR .1%

“5x deadlier flu” may be conservative.

Letting it rip would be horrifying and tragic but by the numbers more of a really nasty flu season with deaths on the order of 250K, maybe far less.

Diamond princess had 712 infected, 13 died. That's 1.8% IFR. (See https://tinyurl.com/s78aj7f).

Iceland is also miscalculated. As of today, 10 deaths in 1,789 cases (which given their extremely high testing rate, is assumed to be close to actual infections) = 0.56 IFR.

The LA (and all other) serology numbers have significant questions lingering about false positives pushing up the calculated community spread.

Except for, you know, all those refrigerator trucks parked outside NYC hospitals you'd have to explain away.

How's the weather in Moscow?

"...at its root a seasonal respiratory disease."
Seasonal? No evidence for that, just wishful thinking.

If IFR is proven to be at 15-25% in NYC, wouldn't this show that herd immunity is indeed achievable over 3-4 months without breaking down your medical system?

When you have significant deaths in your frontline medical teams due to caseloads and lack of equipment, you have a medical system breakdown. This privileged 'not affecting me' attitude towards how well things are going is severely wrong.

No. The question is whether anyone needing care can get it.

No. That's another measure. PMedical staff and their families are not exendables.

Everyone is expendable. Isn't that the point? Cost-benefit analysis rules the day

Get used to it. We are all expendable in these situations.

A few die and save thousands. A few don't die and thousands die.

It isn't nice, and doesn't mean that the lives lost are not a tragedy, or by better preparation could be prevented.

It simply is reality. I'm saying this because this isn't done yet. The easy decisions have been made, they only get harder and harder from here on. The precipice of the supply chain collapse is behind us now, look at the oil price. When it becomes evident it is too late.

And remember that shutdowns have serious consequences, worldwide. Entire nations could go starving if this lasts long enough, triggering civil wars and general unrest.

We make decisions that favor economic growth and convenience over lives everyday. Lowering the speed limit to 15mph nationwide would save an enormous number of lives. Is that anymore ludicrous of an idea than a national quarantine? The economic damage is not sustainable, nor is the Fed and Treasury bankrolling it for months to come. We're on the verge of destroying the hospitality and energy business, which will subsequently destroy the credit markets. The consequence will be felt by people. The narrative is that "corporations" will take the hit, but the reality is people will lose their lives, businesses, access to further education, homes, and well being.

"enormous number" = ~40k. Still, it's a fair point and many are asking. Why is COVID-19 treated more severely than other deaths where the government could intervene and save lives, but they don't.

Nearly all the whataboutisms I hear used as comparisons (car accidents, for example) aren't contagious diseases. If I go out with my Car Accident in public, I don't cause 245 other people to have Car Accidents within a month (R=2.5, 5 day infection delay).

Given that NYC is probably going to see an increase of 2x more infected before the first wave really dies out and given that a second wave is more or less inevitable, we may know in 3-4 months whether herd immunity is going to work or not for covid-19.

I don't think that many locations could deal with an outbreak like NY's without the medical system breaking down though.

"If IFR is proven to be at 15-25% in NYC, wouldn't this show that herd immunity is indeed achievable over 3-4 months without breaking down your medical system?"

It does tend to look that way. Granted, NYC got a lot of extra supplies brought in, the US couldn't have provided that much simultaneously for the entire country. On the other hand, it looks like the combination of high density, mass transit, low to non-existent use of mask by the general public, a lot of international traffic and a lack of preparedness by the city and state contributed to a much worse than normal case.

+1, good summary.

New York was always vulnerable but they did a fair amount of this to themselves. Good to see things are getting under control.

You're welcome,

The rest of America

I think you meant prevalence of infection. An "IFR" of 0.6%-0.7% would produce a total death toll of around 13,000 in NYC, if one assumes that around a quarter of the population is already infected, The IFR would be higher if it's only 15% of the population infected. More like 1.2%. Perhaps herd immunity in NYC will be reached with an additional 15,000-35,000 deaths in the City. That's still a lot of dead New Yorkers.

And we have theories, oh boy do we have theories.

Several epidemiologists have said that forecasting is somewhere between hard and impossible because there are too many parameters and not enough data. So we should expect a bunch of math-smart people from outside the field coming in to fix it and failing. See xkcd.com/1831 for the time frame on that.

Maybe. But if you read what he is saying, he is confirming what the epidemiologists are suggesting is working. The social distancing helps change the spread dynamics, which is pretty obvious even to a layman. Perfection is impossible, again obvious, but in a dynamic situation large effects are possible without 100% compliance, again obvious. And the death rates are less noisy as an indicator, and the trends on that front are noteworthy.

From there on, it is a matter of adjusting on the fly. I think most people have figured that out, and are doing those adjustments as well.

The UTexas models have no consideration of second wave dynamics, openly stating it. And their stuff is probably among the best, but severely limited.

Where did the this information come from? " (2) The model estimates the extent of social distancing using geolocation data from mobile phones and assumes that the extent of social distancing does not change during the period of forecasting." Copy paste from the UTexas site.

"boy do we have theories"

I think there are a lot of agnostics as well.

I for one don't know how this is going to look in a week, a month, or a year.

But I think we can address current problems, which may be identified without too much modeling.

The Guardian online had a good piece today connecting death rates to pollution levels. One of my hunches is that our focus on climate change has taken our eyes off the pollution ball. A reduction in particulates would do more for public health than many other measures.

That would be interesting. I've heard that "you've got to use" GW as your hook to get any attention, but obviously that can produce a counter-reaction.

(I can't think right off of a particulate source that isn't also a co2 emission ..)


Ah right. California's central valley suffers natural and agricultural dust.

There are other countries out there and lots of them are poor.

"3.8 million people a year die prematurely from illness attributable to the household air pollution caused by the inefficient use of solid fuels and kerosene for cooking." WHO

People who take an interest in environmentalism but aren't aware of stuff like this amaze me.

Oops, I misread your comment. Burning wood and kerosene produces CO2, but that's just a footnote.

How much of Ray Lopez's carbon footprint is attributable to his burning garbage in the Philippines?

Hi Etalon d'Silomar. We will build an incinerator using forced air to burn our garbage in PH. It's quite a problem, getting rid of trash in the PH countryside where they don't have trash pickup. Imagine burning plastic in an open pit, black carcinogenic smoke, kids around, not fun. Incinerator should help. Ideally they could build a trash pickup service and some World Bank / UN initiative was doing just that but they either ran out of money (mismanagement) or it never got to our neck of the woods. Bye.

The West region (California) has worse PM2.5 than the NorthEast. Indeed the West region is the worst in the country. And all of the US is far better than China.

Reference: https://www.epa.gov/air-trends/particulate-matter-pm25-trends#pmreg

Here is a detailed map:


The LA Basin ain't great, but the red zones are hot interior valleys, especially those under irrigation and cultivation.

It’s a hell of a lot better than it used to be.

I also wonder how much the smog moves. Pollution doesn’t seem like it’s something that stays local.

It is interesting that the Owens Valley is a red zone. Most of the land up there is natural, but it is dry and windy.

Some environmentalists do blame all of that on water theft, but I think water theft is only part of it.

It's probably correct that there'd be dust in the Owens Valley anyway. But LA's appropriation/theft of the water from the Owens River (see the movie "Chinatown") turned the valley into a dust bowl.

Several decades and lawsuits later, Owens Valley got some of its water back including agreements to monitor dust and pollution levels (the dried-up bed of the Owens Lake has a ton of minerals that are unhealthy to breathe) and re-allocate water to wet down the lake bed when necessary.

I'm making my assumption that the Owens Valley was always somewhat dry and dusty by reference to its neighbors, the Panamint Valley and Death Valley. But I don't really know; maybe the ground was grassland that could withstand seasonal waterflows and dry seasons, or maybe irrigation had made the valley green.

I remember driving over the Owens Mountains towards the Owens Valley a couple of decades ago; when you're in tall mountains you often find yourself looking down onto the clouds. I thought that's what was happening as I reached the top of the ridgeline; I was in clear air with a blue sky but I could see that the Owens Valley was filled with a big white cloud.

Which I thought was clouds or fog. But as I drove down into the valley I discovered that the white cloud was dust kicked up by big winds.

This was before the latter lawsuits were settled, so I'm guessing those dust storms are less frequent now.

I was in Independence once when it went from nice and sunny to full on sandstorm in 10 minutes. Impressive. I was glad I was in town and could get my car behind a building. It was a sandblasting.

I now recall that the source of the dust was not so much the Owens Valley, but instead the dried-up lakebed of Owens Lake. And that was what one of the lawsuits against LA Water & Power was about -- re-allocating enough water to re-wet the lake bed to keep the dust down.

The rest of the valley might very well produce dust too, but not at the apocalyptic levels of the lakebed.

[Posting this a second time because it didn't show up the first time]

it's probably correct that there'd be some dust in the Owens Valley anyway. But LA's appropriation/theft of the water from the Owens River (see the movie "Chinatown") turned the valley into a dust bowl.

Several decades and lawsuits later, Owens Valley got some of its water back including agreements to monitor dust and pollution levels (the dried-up bed of the Owens Lake has a ton of minerals that are unhealthy to breathe) and re-allocate water to wet down the lake bed when necessary.

I'm making my assumption that the Owens Valley was always somewhat dry and dusty by reference to its neighbors, the Panamint Valley and Death Valley. But I don't really know; maybe the ground was grassland that could withstand seasonal waterflows and dry seasons, or maybe irrigation had made the valley green.

I remember driving over the Owens Mountains towards the Owens Valley a couple of decades ago; when you're in tall mountains you often find yourself looking down onto the clouds. I thought that's what was happening as I reached the top of the ridgeline; I was in clear air with a blue sky but I could see that the Owens Valley was filled with a big white cloud, thousands of feet tall.

Which I thought was clouds or fog. But as I drove down into the valley I discovered that the white cloud was dust kicked up by big winds.

This was before the latter lawsuits were settled, so I'm guessing those dust storms are less frequent now.

NYC has air that's cleaner than a lot of places, like Dehli, Seoul, Shanghai, Beijing, Hong Kong, London, Berlin, Toronto .... If that's what a mild COVID outbreak looks like, then it looks like we've got problems.

Judging by the tax rate on net CO2 emissions, we never had our eye on the climate change ball, even without taking account of the tax on fossil fuels would also have had salutary effects on particulate emissions.

Speaking of measuring in the moment, a LA USC study says "about 2.8% to 5.6% of the county's adult population has antibody to the virus"


"These results indicate that many persons may have been unknowingly infected and at risk of transmitting the virus to others," said Dr. Barbara Ferrer, director of the L.A. County Department of Public Health. "These findings underscore the importance of expanded polymerase chain reaction (PCR) testing to diagnose those with infection so they can be isolated and quarantined, while also maintaining the broad social distancing interventions."


Yes, that's me.

Wow, I thought you were extinct. Nice to know that you are still above the sod.

Google is a specialized niche where unusual laws of selection apply.

You got that right; sure got that right.

Rich, go put on some camo. The "liberate Colorado" protest needs more COVIDiots to take on those counter-protesting health workers.

Get off my lawn!

" but we know from other sources that both can and do happen due to changes in test and test processing availability."

Seems like an easily fixable problem, testing data is available.

"but of course getting sign(R – 1) right is a low bar, it’s just figuring out whether what you’re observing is growing or shrinking"

Still seems useful, no?

In the mean timezzz....

“Like it or not”

Ruin has come whether we like it or not
Sit at home or labor with fraught

The men of means all sit in doors
Pointing at peasants mopping their floors

Go about or settle down in your spot
Ruin has come whether we like it or not.

There are those that say we must go or burst they sit at home while others bear the worst

Whether sitting or going the germ is wrought
Ruin has come whether we like it or not

How long can this go from towers they say
Until our mistakes the others shalt pay

Open or close it won’t change our lot
Ruin has come whether we like it or not

The time is come to eat me must trot
Ruin has come whether we like it or not

I Love a Rainy Night, by Eddie Rabbitt, Tom Jones

Nice poem Student. I am designing here in GR a gravity fed watering system, using suction and a watering tank with drip irrigation system. Should work fine, I am guessing I need 12 metric tons of water per week during the summer months until the trees get good roots. Gravity is free, no need for a generator to pump water into the field! Take care.

"Lockdowns, including Western-style lockdowns, very likely drive R substantially below 1 (say .7 or lower), even without perfect compliance. "

False. This is a Deep State lie they use to steal our freedoms.

"Cases are of course massively undercounted"

That's right. Keep people in fear to grab more power. We The People know your games. You can't stop freedom.

Actually, the statement argues strongly against any sort of *government-mandated* lockdowns. The case for a government-mandated lockdowns is an extremely narrow one. It requires that voluntary measures --- like washing hands, maintaining 6-ft separation, workplaces voluntarily closing, and workers voluntarily working from home --- aren't quite enough to push R below 1 yet are close enough that the incremental gain from government mandates are enough to push R below 1. If one can get all the way to 0.7 without perfect compliance, it says that government mandates are unnecessary. With all the ambiguities around models, data, etc., even if we did happen to be in that very narrow channel that would justify government mandates, how would we possibly know?

Also, saying that cases are massively undercounted is just another way of saying that the virus is not as lethal as originally thought, another strike against continuing mandatory lockdown measures.

I am willing to admit when someone has made a good case. To those that argue for ending government-ordered lockdowns, I admit it: you're right.

What were your GRE scores?

You must be an economist.

Are they offering curbside pickup?
Whole body condoms?
Are they disinfecting their customers?

The sex workers I follow on Twitter have been taking time off.

Nation's Sex Workers Call For Stimulus: "Give it to us! Oh god, give it to us right now! We need it so bad!"

If we aren’t essential, I don’t know who is.

How much money does he make? What were his GRE scores? Did he vote for Democrats or Republicans? Don't mean to be rude but is this guy a smart guy because he agrees with your views?

He also supports and is supported big business, thinks the goal of the universe is maximizing GDP, and socialism for the rich is like very awesome, because we need to care for them, they create wealth and do ALL of the labor.

Mostly agree, but the upper estimates of total infection in the second bullet would be my mid-lower estimates. Notice that these Oxford Centre for Evidence Based Medicine pages (i) have revised their CFR estimates down even further recently, and (ii) use some pretty blunt language about policy.

i) https://www.cebm.net/covid-19/global-covid-19-case-fatality-rates/
ii) https://www.cebm.net/covid-19/covid-19-the-tipping-point/

A question. The number of infections seems to be a guess so far, the data is pretty sketchy.

How much of the estimates are derived backwards, ie. a model of spread leading to this measured result (deaths) indicates an infection rate in the population of x?

The first link I posted shows their workings pretty well. Yes it is extremely guessy and sketchy. Even deaths could be badly under or overcounted (though I think under is more likely). It is a combination of models that fit back to deaths or positive swabs, patchy flawed antibody tests, and whole small communities like cruise ship or repatriation flights.

I just worry that we are confusing "we aren't sure so we should act as though the worst case scenario were true" (fine) with "we aren't sure so we should believe that the worst case scenario is true" (illogical).

Indeed. That is well stated.

This is a very smart take.

As he says, short-term dynamics in the number of official cases detected in a specific location are incredibly prone to error. These are subject to a lot of volatility due to availability of testing, so trying to estimate R off these is rather crazy.

For deaths we have good reason to think the degree of undercounting is much smaller, and probably fairly stable across time. Perhaps deaths are undercounted by 50%, perhaps it's 10%, but the trends in the data should be much better at capturing reality. Alternately, where hospitals are not at capacity (which is to say, throughout the US), the number of people hospitalised by day in a location is again much easier to measure accurately across time. Under the fairly reasonable assumption that the hospitalisation and mortality rates (as a share of total infections, observed or unobserved) are roughly constant over time, these provide much more useful information for ascertaining R (or at least, what R was several weeks ago).

Take NY for example. The number of deaths reported daily is declining steadily and fairly rapidly. The number of people visiting NYC emergency departments daily with respiratory/ILI complaints has fallen ~65% since the peak approx 4 weeks ago. And the proportion of test results in NYC that are coming back positive (which is rather informative since it abstracts from the number of tests performed, and tells you something about the marginal test taker) has fallen fairly dramatically, from ~50% to the low 30s% in recent days.

These numbers are not consistent with R ~= 1.

Yet rt.live claims NY has spent much of the past month with R > 1 (and averaging over 1). Enough said.

One complicating factor everywhere is that rather than specifically COVID-19 deaths, we want to look at all-cause mortality to catch people that die without being confirmed as infected (dying at home causes a lot of this). A lot of countries and states are late in publishing the all cause data, but it eventually gets out there. (Yes, it's true there may be people who die while positive but of something else, death is complicated, but the spikes in all cause mortality are larger and show that unconfirmed cases are much higher than that, especially considering that some causes, like non-COVID-19 infectious diseases and traffic deaths, are down a lot.)

Luckily, NY is one of the most rapid states at publishing reliably all cause mortality, and indeed their data does show that they're getting past the enormous peak, which given incubation time means that things are working to get R below 1.

> Luckily, NY is one of the most rapid states at publishing reliably all cause mortality

Yes, but they and other states are in the process of sandbagging so that the feds pick up the tab.

That is, 6 months on O2 + dialysis with failing kidney's and a bad liver...While in the hospital you get pneumonia and die a week later. Presto: Call it covid and the feds will pay the bill. Not sure if they feds pay the entire bill. But it's causing everyone to be fast and loose on the coding of deaths, that's for sure.

Expect to see deaths spike even though they occurred weeks ago.

One positive sign is that all-cause flu cases as measured by the Kinsa health meter shows that estimates of US flu cases (which will mix both flu and unmeasured coronavirus) are at unusual lows (from 2% of the population typically down to the 0.10-0.15% levels). https://healthweather.us/?mode=Atypical

Or it just means herd immunity is being approached.

Where's the evidence for Covid-19 herd immunity and what's its credibility?

I wonder about the incentives of hospitals regarding the documentation of Rona deaths.

I understand that the Federal Govt is reimbursing hospitals treating Rona patients. I also understand hospitals are suffering considerable financial stress as many of their services have been suspended.

I'm not suggesting a conspiracy, or that anyone is deliberately or knowingly lying. But I expect, given the various incentives and stresses, US hospitals will likely to over-document deaths involving influenza-like illness symptoms as Rona-caused, whether or not there was a test.

Interestingly, the CDC shows that less than half of "coronavirus" deaths in the USA also had pneumonia.

Things that make you go hmmmmmmm.

Can’t we break out some good old time series econometrics here? Has anyone testing the time series of R to determine if it had a unit root? If it does, it has no variance and all these discussions about measuring it at points in time are meaningless?

I need a one armed economist to tell me what R is.

(Many serious modelers have given up on case counts and just model death counts.)

This was always obvious to me. I never fell for the 100K-200K US death nonsense, at least not in the first wave.

Worldwide deaths increased at 10% per day in the second half of March (meaning total deaths doubled weekly for 2-3 weeks, scary exponent.)

Since them the % increase in daily deaths has moved steadily downward and was 3.1% yesterday (that's a doubling every three weeks, much less scary.)

That number will keep coming down. Daily deaths have fallen in the past couple days, and no country with significant deaths is running anywhere near 10% daily increase anymore (maybe Canada, but their death total is still so low that a few more days of 10%+ growth are no big deal).

Anyway, we are now in the boring end game of wave one. Uncertainty has been wrung out of the system.

What's the game plan?

"Anyway, we are now in the boring end game of wave one. Uncertainty has been wrung out of the system.

What's the game plan?'

Start returning to normal with a strong emphasis on social distancing. Perhaps keep the worst areas locked down for an additional 6 weeks. (Though that's going to be rough on a lot of families in the impacted areas.) Then watch for mutations, expedite a vaccine and keep the testing capacity at high levels for Fall.

Moreover those are pretty much the policies that everyone agrees on, from the "open up/liberate" demonstrators to wannabe authoritarians who like ordering people around.

Obviously there's disagreement on the time schedule of returning to normal, especially over how much we should be concerned about a second wave of infections. But with so much uncertainty, well yeah people are gonna disagree.

Re: "Anyways, I guess my single biggest point is that if you see a result that says something about R, there’s a very good chance it’s just mathematically broken or observationally broken and isn’t actually saying that thing at all."

If R is varying, that tells you an awful about the success of your approach. And, in fact, that statement above is contradicted by the content of his post, where he points out that R is changing because of lockdowns without perfect compliance.

Maybe he didn't realize the internal contradiction of his ending statement and his supporting evidence.

Of course we want R to vary, and we do things to make it happen. That's not a problem, but a good result.

Thanks Bill. Let me try to clarify a bit.

I'm not saying it's *impossible* to say anything about R: as you point out, it sure looks to me (and others) like R is changing based on lockdowns with imperfect compliance. I think saying *precise* things about R is difficult: I think Italy's post-lockdown R is very likely less than .7, but I have no ability to tell if it's .7 or .4, and to pin it down much more precisely would involve knowing more about the underlying dynamics than I have much hope for. (Every country's outbreaks are of course the sum of multiple interlocking outbreaks with differing dynamics that we don't know enough to tease apart.) But yeah, the layman's claims about R are mostly true: it looks like introducing large amounts of social distancing at time t leads to a drop in new deaths t + 3 1/2 weeks later, which almost certainly means there'd be a drop in new infections at t + 1 week if you could observe it (which you can't), and that means R is driven "noticeably less than 1."

But my point was that many of the results we're seeing (rt.live, the study Tyler linked) aren't actually saying anything about R because they're fitting noise in data or using invalid statistical approaches.


In terms of modeling, you might want to look at my comment regarding defining "geographic markets" for covid using antitrust techniques, including Elzinga Hogarty. It was in the comment section in the previous post. Also, the more you are able to define a market closely, your data may not be as noisy.

In fact, since Hal Varian is your chief economist at Google, I would talk to Hal about using Elzinga Hogarty and more defined geographic spaces, rather than just states or even some SMSA.

Hal knows a lot about this.

Think in terms of not only where people travel (a vector) but also who comes into the preliminary market (another vector) and what that person connects to (eg, an airport, convention site, etc. And, then think about this as a hub for retransmission. So, for example, you might define a geographic area above as safe, based on these measurements, making them more open, but, at the same time watching the connections from the outside coming in.

Again, Go see Hal Varian. You can get a more refined market and then you might get less noise in your model and also be able to test hypotheses.

You might also want to talk to Prof Ghose at NYU as well. Ghose wrote a book on locationally aware advertising and apps: https://www.amazon.com/dp/B071CWDH8H/ref=dp-kindle-redirect?_encoding=UTF8&btkr=1

Also, Rif, if you look up Elzinga-Hogarty models, and talk to Hal, you should be able to do some pretty good zip code analysis to create "markets ". Hal Varian would be an excellent person to talk to on this.

If you want to look at spatial markets using a different technique, here is a good one: https://dl.acm.org/doi/10.1145/3058499

Also, the article I cited to above would also explain why communities of color have a higher transmission rate and a high prevalence of covid. Think of it as a disease that breaks out in the ghetto and there are few grocery stores and the grocery store is infected.

Rif, The comment I referred to was in the post on which relocations can be opened first.


Also, if you use a vector model, you can weight the vector in a way the represents the amount of time at the end point or origination point, so you are not just looking at where people go or come from but also how long they stay there.

Also, since you are at Google, you can look at where advertisers limit the geography of the ad, what are search patterns which would disclose how far searchers limit their search geographically, etc.

But R changes with everything. It depends on age of population, it depends on population density, it depends on weather, it depends on the number and duration of interactions you have per day, it depends on time since start of infection. Trying to determine it is as fraught as trying to determine climate sensitivity. Climate sensitivity estimates are a barn door--a factor of 3X generally. And at the lower bound, nothing drastic needs to be done. And at the upper end, it's an all out emergency. Like R, both are really useless measures except to illustrate a concept in the simplest of models.

If the climate community hasn't arrived a single climate climate sensitivity number with a tight 95% confidence over 20 years, what makes you think in 3 months a useful version of R can be determined on a target that has many more unknowns and is moving drastically week by week?

Yeah, but why do you listen to the weather forecast in the morning.

All models should start with the basic question: What is this model being used for.

If it is for a yes or no answer, is it for measuring big changes, is it for whatever.

How or whether you want something to be precise or not depends on the question you are asking.

Today's weather was sunny as predicted, but the afternoon temperature was off by 5 degrees.

Also, Phinton, the age of the population, its density, and other variables. do not change during the period of the study. You could treat those as constants. So. what you want to focus on is geography and then look at how changes in R correspond to changes in policy.

I'm not sure that 'geography'. as you seem to be defining it, is enough. For instance, take the meat packing plants in SD and western MN that suddenly became hotspots. Not a lot of airport connection, not much travel in and out, but once one case appears the immediate environment is so fraught that the transmission rate explodes.


This is interesting. According to newspaper reports, the two plants, although in different states, are actually close to each other and in some families, one person works in one plant, and the spouse works in the other. https://www.mprnews.org/story/2020/04/18/officials-say-at-least-20-workers-at-sw-minnesota-pork-processing-plant-have-covid19 http://tcbmag.com/news/articles/2020/april/worthington-pork-plant-closed-indefinitely

That's why geography, vectors, traffic connection patterns, shopping linkages, etc. ARE important.

Here is an article that mentions employee family overlaps: https://www.startribune.com/first-covid-19-case-confirmed-at-pork-plant-in-worthington-minn/569731152/

Wally, This will also be an interesting natural experiment one the effect of social distancing. SD did not; Mn did. What will be interesting to look at is whether Sioux Falls spread more from the plant than from the corresponding MN plant.

You can also look at spread between adjoining states with border cities which overlap state boundaries.

In antitrust, we always look for natural experiments to define boundaries and product and geographic markets.

> If it is for a yes or no answer, is it for measuring big changes, is it for whatever.

I would guess the model is primarily useful for forecasting ICU and other medical needs. But trying to get incredible accuracy from the model will end in tears, just as it had for warming where 99% of the predictions way over shot. Beyond a few degrees of freedom, it's nudging towards 100% modeler bias.

> Also, Phinton, the age of the population, its density, and other variables. do not change during the period of the study

They most certainly do. NYC boroughs are wildly different in density. Guttenberg, Union City, are above 50K/mi^2, Hoboken is 39K, Yonkers is 10K, Elizabeth (NJ) is 10K. They are all part of the NYC metro area. Manhattan is 70K, Brooklyn is 35, Queens is 20K.

This don't all "burn" at the same rate. And R will change as it burns through neighborhoods. Along with subway usage. If you zoom out enough, then the entire world has a single R, but that's not useful for modeling. And if you zoom in enough, the different neighborhoods have different R too. But that makes the complexity very difficult to manage.

But I don't think whatever R you come with for Manhattan will work for the Bronx (half the density).

And the R at the beginning of the infection (lots of old, frail people) isn't the same at the 3rd wave of the infection (all the frail people are gone).

Nursing homes, incredibly, are empty in many some because most of the residents died. Those are all old people that cannot be infected again in a next wave. Thus the R will be different for wave 2, 3, etc.

+1 Excellent.

PH, density doesn't change in a geographic area, nor do demographics. But, different geographic areas do have different compositions. But, if R is examined within a geographic area defined by interaction of persons, those features--density, demographic features--are constant over the period.

> But, if R is examined within a geographic area defined by interaction of persons, those features--density, demographic features--are constant over the period.

Yes, but you DEFINE it that way to simplify the problem--not because you want to, but because you have to. The real R isn't constant across NYC even if you define it to be.

Yes, from an economist's perspective I think R is a lot like many of the most important variables in macroeconomics, which are slippery to pin down for a couple of reasons: like R, they are highly endogenous and can and will change depending on how individuals behave and how policies change. And like macroecon variables, R is a single variable trying to summarize or aggregate a whole lot of sub-groups.

That doesn't mean R is irrelevant, any more than say the interest rate is irrelevant or investor confidence. But it needs to be used and interpreted with its very high endogeneity in mind.

Since R is a socially determined factor that itself is the product of the number of contacts per day times the probability of that contact transferring the virus and infecting the other person. That probability of transfer depends upon the details of the interaction and whether proper biosecurity practices protocols are being practiced. For example, if both people are wearing personal protection equipment and are using it correctly and sanitizing after contact like what occurs in medical settings, the probability of transfer is close enough to zero that with even many contacts per day the effective R reproductive rate is much less than 1.

A shut-down decreases the interactions per day and drives down R at the cost of the economy. Requiring proper biosecurity practices like using masks, external garments, hand and head covering and sanitizing those coverings and masks with heat for the general population would also drive down R by decreasing the probability of transfer. For civilian applications, heat treatment inactivates this virus at 140+ºF for 30 minutes in an oven, dryer or sauna at the correct temperature allowing for citizens PPE and full reuse. Hospitals also have other pathogens that aren't as easily killed so they need either disposal or more drastic sanitation methods.

The concept of R is good and focuses on the transfer probability, not just the physical distance that is irrelevant with good PPE.


+1 I would also add to it that R has to be considered in the context of a location defined by the degree of interactions among others in that geography, and also the number of folks that enter that geography from outside. See my comments to Rif above and also comment in the previous post.

"If R was > 1 you’d see exponential growth up to herd immunity, if R was 0.9 you’d see a much bigger and later peak (there’s a lot of momentum in these systems)."

Actually, herd immunity is probably close to being achieved in places like Lombardy and New York. For instance, in New York city has about 15,000 fatalities, that's .17% fatality rate over the total pop, which implicitly suggests about 50% of the population has been infected.

If R0 is driven down form 5-6 to 2 by New York lockdown, then that's the fraction that becomes infected before herd imunity is achieved.

"probably" "close" "implicitly" "suggests"

Is this just some more Google navel-gazing, or are they actually working with the US gov't or states on applying this research to policy?

Look at the UTexas link, and read at the bottom. Data from Google and/or Apple is used. Presumably some people are looking.

But why is it important to make R less than 1 ? This seems to be considered as obvious by TC and Rif A Saurus, and many commenters, but again, why ?

Obviously, the largest R, let us call it R_max, such that the health care system do not get overwhelmed is better that any lower R. For at R=R_max, we are going to have eventually the same number of infected and of deaths (because everyone will receive proper medical care), but we. will have them faster, which means we will approach herd immunity faster, and whatever costly measure we need to keep the R where we want it will be shorter. (And moreover these costly measures will be less costly if we aim at R=R_max than if we aim at R>R_max.)

Now what is R_max? I don't know but is certainly > 1 for now, since the health care system is not saturated, and is rising fast (e.g. with the production of many new ventilators, etc.).

Now R_max is your optimal R only if you attribute an infinite disutility to any death of a person infected by coronavirus and not receiving sufficient medical care. If you attribute a finite value to those lifes, then the optimal R, R_optimum, will be even greater than R_max. For if we aim at R > R_max, we will lose some infected people due to insufficient care, but we will save on the measures taken to lower R, lower the economic damage and the number of deaths it is bound to cause. So there is a trade-off here.

The above was mathematical, but let me do some unjustified estimate.
I think we would be good at R=1.5 for example. The number of infected will grow relatively quickly, but probably not too fast as to overwhelm the medical system. Them the R will begin to get lower naturally, because of the number of infected/immunized rising, and perhaps also because of the summer. There will be a second wave in the Fall, but much smaller than what it would have been with a R<1, because more people would have already been infected. And an R =1.5 can probably be achieved with voluntary social distancing, without the economic damage, the deaths, and the destruction of individual rights caused by a forced lock-down.

> Obviously, the largest R, let us call it R_max, such that the health care system do not get overwhelmed is better that any lower R.

Yes, bullseye. The vaccine isn't a certainty. Ideally you'd like to let the infection burn such that ICU utilization was managed near it's limit for those under, say, 55, All the while, you are putting people back on the street recovered (and presumably immune).

What is awesome about this is people can self select. If you are scared witless of the virus, stay inside and collect unemployment and eat a cheese wheel.

If you are under 40, then you can take comfort in the fact that your chances of dying from covid are MUCH better than dying from a car crash. And nobody is cowering indoors worried they might be killed in a car crash tomorrow.

WA state has had 0 people under the age of 40 die. If the ICU is empty, then twenty-somethings should be at the bars doing what they normally do. For the good of society.

Any R that is > 1 is going to cause exponential growth. However, what you could do is let the number of infections rise to a sustainable pace (if you are trying to pursue herd immunity), then drop R to hover around 1 until everyone has had it.

The current strategy is to drop R as low as possible to try to wipe out the virus to the point that every case can be isolated and traced. Personally I don't see it happening. There's too many people out there who have it to identify and trace them all.

"Personally I don't see it happening. "

+1, that's never going to happen with it this widespread.

Yes, agree with Hazel and J. Contact tracing I'd guess works best in the beginning, with efficacy falling fast as you have more cases. I just re-watched Contagion. In that movie Kate Winslet is a tracer, and she drives to see each person, the person is sobbing, scared witless, she asked them mundane questions, they asks questions back, they ask about treatment. if she's a typical contact tracer, you'd be lucky to talk to 4 people a day with travel.

And then imagine the follow-up you'd have to deal with "After lunch I went to the Piggley Wiggley on Broadway"..."do you know the address?" ..."No"..."Looks like there are 5 Piggly Wiggly on broadway...which one?"...."Oh, you know, the one by Chuck-E-Cheeze"..."actually I dont' know, I'm not from around here"

I mean, the process seems incredibly tedious and laborious.

The chinese managed to do it with 70,000 cases in Wuhan, but I think they started earlier on and a lot of the subsequent cases were people who were close contacts of earlier cases and already self isolated.
We already have 10x that many cases. Spread all over the country. In China it was much more concentrated in just Wuhan.

Agree that I don't understand why R<1 is the appropriate target. I've written here about how we could set a target and let the public use forecasts of whether we are going to meet that target to adjust their behaviour. I think the idea is neat in theory, but, if we can't forecast accurately (amongst other reasons), it won't work.


Great post Tyler. And great comment Joel. Also a great point by Hazel on test and trace being impractical.

It’s all about costs and benefits. Clearly lockdowns drove R below 1. Clearly we can ease lockdowns and keep R below 1. As Joel pointed out, we don’t need R below 1, we just need it low enough that hospitals are not overwhelmed.

Most people say we need test and trace on a massive scale. But it’s horribly expensive and unnecessary. In practice, no one gets tested unless they have symptoms. But people with symptoms should stay home anyway.

Tracing is a much bigger mess. It’s easy with nice people in nice neighborhoods who answer their phone. But how would you like to knock on the door of Pedro from the meat packing plant who lives with 20 other illegal immigrants?

Forget test and trace, and just do lots of temperature measurements. Employers should measure employees and send sick ones home.

Keep some other modest social distancing like face masks, 6 ft separation most of the time, and more isolation for the most vulnerable.

We can probably open our economy completely right now because behavior has already changed enough to get R to the optimal number that Joel wants.

Bars, restaurants, and planes will remain empty like they are in South Korea which did not force them to close.

It’s time to open our economy.

I would like to submit that R itself is not a terribly useful concept, give it's dependence on all sorts of external factors like weather and culture, We need some objective measure of infectiousness that is not dependent on social context. Especially if we are going to base policy that affects social context on a number that is itself dependent on social context.

Although the underling count of deaths is slightly different (mine come from the daily WaPo report) I get an almost identical estimate of daily deaths using log(D) = aT^2+bT_ c, without using any cellphone data. UT has 40.4 thousand dead; mine has 38.4 thousand.

Why read hard models when you can just adopt the "consensus inside Google".

Do you think the "consensus ... inside Google" is going to differ substantially from what their managers think will maximise Google's private profits?

so what you're telling me is, it's subtle.

The Icelandic data on April 4 yielded 0.6% positive and 2 weeks later on April 20 Iceland has 9 corresponding cumulative deaths . This yields an IFR of 0.41%. 95% CI = [0.2 %, 0.6%].
See NEJM paper on the results. Note that Heinsberg/Gangelt estimated 0.37%
From data in the Bronx we have a naive death rate of 175 per 100k population or 0.175% IFR. (assuming the unrealistic number of 100% infected 2 weeks ago) . This puts a lower bound on the IFR in the Bronx at 0.175%.
We can make some estimates of the population truly infected.
• Asymptomatic ~ 50% from spread of various studies
• symptomatic being tested.:-> 90% of people requesting a test by phone in NYC are told to stay home.
• Positivity of the untested symptomatic : we can use 2 numbers. Average positivity in the US = 19% or positivity in New York = 39%
• We get 2 numbers for the undetected/to confirmed cases ratio the high number : 20x or the lower number: 11x.
• The positive cases in the Bronx are 2000 per 100k today but were half this amount 2 weeks ago or 1.0% We get an estimate of the true number infected in the Bronx today of 22 % to 40 % , but 2 weeks ago of 11 % to 20%. This would give an Ro of 1.45 in that period. ( serial interval mean =3.96 days)
This then estimates the IFR for the Bronx in the range [ 0.88% - 1.58%].
This estimate is certainly higher than the Icelandic estimate keeping in mind that the Icelandic population is healthier and the Icelandic hospitals less busy.



Currently it seems to me that those are the smartest and best informed views “out there,” so at least for now they are my views too.

But they're wrong. Now that we have actual antibody testing results, it turns out that Covid is more contagious than the flu but no more than twice as deadly. The lockdowns are not only unnecessary, they're also ineffective. The reason NYC new infections are dropping so quickly is that most people there have already been infected and recovered. The rest of the country is only three or four weeks behind.

:Put the sourcing for your statement that covid is no more than twice as deadly than the flu.

Here is what I found: Anthony Fauci, MD, director of the National Institute of Allergy and Infectious Disease, put it plainly: "The seasonal flu that we deal with every year has a mortality of 0.1%,” he told the congressional panel, whereas coronavirus is "10 times more lethal than the seasonal flu," per STAT news.

Post your support below.

That was in response to Jeff above

Did you look at the link? LA County estimates 220K to 440K infections from their antibody testing, while there are 600 deaths. Since it takes about the same amount of time to develop antibodies as it does to die from the virus, you can just divide 600 by 220,000 to get an IFR of 0.27 percent, or divide 600 by 440,000 to get an IFR of 0.136 percent. Midpoint of that range is 0.204 percent, roughly twice Fauci's flu number.

I think we have to be careful They would have to publish more details. I couldn’t find the sensitivity of the BOSCH Biotech test online. They claim a very high sensitivity rate. In a published study of 9 commercially available tests in Denmark ( it did not include Premier Biotech), they found a sensitivity of 80 to 93% and a specificity of 80 to 100%. In the article you linked to they claim claim a 99.5% sensitivity which would be remarkable. The specificity is not given.
for the Cellex test , another aB test used in the US, the sensitivity is 93.8% and the specificity is 95.6%
Using Bayes Rule, with a 4.2% real background incidence if the LA test had the ( more reasonabale) Cellex specificity/sensitivity numbers , a positive would only be correct ~ 48% of the time.
I am not saying that the results are necessarily wrong but that caution is certainly needed when doing aB testing in a low incidence population

Yup. So it gets interesting looking at New York State. Worldometers has NY state deaths at over 18,000. Covidtracking.com has them at over 14,000. The discrepancy is that Covidtracking doesn't count the deaths reported by NYC as "probable" Covid deaths (hospital funding?) because NY State officially doesn't report them. Worldometers does. Again, shows how hard it is to really quantify what we're dealing with when governments can't even agree with each other.

Assuming that the LA rates hold true for New York, and we'd expect them to be HIGHER, not lower, because of increased ease of transmission due to containing the best mass spreading tool in the world, the NYC subway, at 41.5x confirmed cases, you're looking at 10,271,748 people in NY State that have been infected. Sure does seem to make sense, when you look at the dramatic declines in hospitalizations, ICU and deaths. 671 deaths reported on 4/13, 478 deaths on 4/20, nearly 30% decline in one week (per covidtracking). New York numbers look so bad because of almost instantaneous mass spreading; we saw how quickly it spread on cruise ships, well now add poor to zero ventilation, one of the most unsanitary mass environments in the US, and overcrowding on reduced trains (classic NYC). It's already been confirmed by Trevor Bedford the virus was in the US in January/February, so you had 1-1.5 MONTHS of millions of New Yorkers being exposed and exposing others, is it really so unlikely that 10+ million people in the state have had it in that time frame?

The estimated IFR in New York at those numbers results in 20,400 deaths, which we're either going to reach in a few days (worldometers) or may be extremely close to reaching when all is said and done (covidtracking).

Even if the number is 8M or 9M or whatever it is, that makes a lot more sense to me than NY being some sort of outlier of death. We're seeing incredibly consistent estimates of IFR all over the world, Iceland, the cruise ships (with their overly elderly population), Germany, etc.

Only thing that seems to make sense.

Wait a second, something Fauci said turned out to be wrong?

On the Today show, Feb. 29:


Why is there this obsession of picking "experts" and taking everything they say as gospel? The experts have been wrong about essentially everything before and during this pandemic. And wrong by 70-90%. Like, wrong by orders of magnitude outside of their confidence bars wrong. And for some reason we're still looking to the exact same experts? Ok, now they have more data, but that doesn't mean that their interpretations of that data are better now than they were a month ago when their interpretations contained approximately zero logic or reasoning ability whatsoever.

Bill - please follow your own advice and post a link for that Fauci quote. (Jeff did include a link, but Bill did not.)

I can also find an editorial co-written by Fauci - admittedly back in late February - stating that "If one assumes that the number of asymptomatic or minimally symptomatic cases is several times as high as the number of reported cases, the case fatality rate may be considerably less than 1%. This suggests that the overall clinical consequences of Covid-19 may ultimately be more akin to those of a severe seasonal influenza (which has a case fatality rate of approximately 0.1%) or a pandemic influenza (similar to those in 1957 and 1968)" ( https://www.nejm.org/doi/full/10.1056/NEJMe2002387 ).

More endless debates about meaningless numbers. This is why libertarianism never went anywhere. Lost in its own inanities.

There are a lot of good thoughts in that post, but there's also this: " (Many serious modelers have given up on case counts and just model death counts.)"

Somehow smart people still have the idea that death counts are accurate. They aren't. They're likely closer to reality than case counts, but in most countries only tested cases are being counted as COVID-19 deaths (and sometimes even then it's not counted), and the overwhelming majority of testing is hospital-only. No two countries are testing or counting quite the same way, and each country changes what they're doing over time. I've seen reports that in Germany people with CoV-SARS-2 who die, but have underlying medical conditions, aren't counted as COVID-19 deaths. In Belgium, they're counting everyone who dies in a nursing home. And deaths sometimes happen much much later than the initial infection, so any study that shows a cross-sectional snapshot of positive tests/deaths will significantly underplay fatalities.

Where death counts are useful is when they're done consistently and looked at within a country, which does indeed show that lockdowns are remarkably effective. The Google folks seem to understand all this, but it would be useful if they would point out that the death data is also problematic, and thwarts attempts at easy answers (as we see every day in the dozens of inane comments here).

Is there not a category of "Very Good Names?"

Comments for this post are closed