Previous week: Covid 7/9: Lies, Damn Lies and Death Rates

Two weeks ago I predicted a surge in Covid deaths starting on July 2 and picking up on July 7.

On July 6, the rolling 7-day average of Covid deaths reached a local low of 480.

Seven days later that average was 721, almost exactly 50% higher. It’s since pulled back slightly to 696.

No doubt holiday delays are a small portion of that story, and helps to explain why things didn’t pick up until after the holiday backlog was gone, which I failed to anticipate at the time. That is a small part of this story, but only a small part.

The good news is that, once again, things are getting worse slower than I would have anticipated two weeks ago. We might be remarkably close to turning the corner.

The other good news is that masks, especially cloth masks, might be a lot of the story of reduced death rates, by reducing initial viral loads. Despite have a full debate and writing a whole post about this, I still didn’t take this possibility seriously enough in the last few explorations. That wouldn’t be quite as good news as the virus becoming less deadly, but it’s probably the second best possible news.

Where do we go from here? When and where will deaths and infections peak? How long until we turn things around? How much locking down do we have left in the tank after all?

Let’s run the numbers.

Positive Test Counts By Region

May 7-May 13 22419 43256 37591 56892
May 14-May 20 22725 42762 40343 52982
May 21-May 27 23979 39418 42977 37029
May 28-June 3 32200 31504 50039 33370
June 4-June 10 35487 24674 55731 22693
June 11-June 17 41976 22510 75787 17891
June 18-June 24 66292 26792 107221 15446
June 25-July 1 85761 34974 163472 16303
July 2-July 8 103879 40139 202863 18226
July 9-July 15 108395 53229 250072 20276

Infections by Region 7-1


Deaths by Region

May 7-May 13 1082 2288 1597 5327
Apr 23-29 1090 2060 1442 4541
Apr 30-May 6 775 1723 1290 3008
May 28-June 3 875 1666 1387 2557
June 4-June 10 743 1297 1230 1936
June 11-June 17 778 1040 1207 1495
June 18-June 24 831 859 1204 1061
June 25-July 1 858 658 1285 818
July 2-July 8 894 559 1503 761
July 9-July 15 1380 539 2278 650

Deaths by Region 7-1


Positive Test Percentages and Test Counts

Date USA tests Positive % NY tests Positive %
May 14-May 20 2,618,882 6.1% 246,929 5.6%
May 21-May 27 2,679,365 5.5% 305,708 3.5%
May 28-June 3 3,041,311 5.0% 417,929 2.2%
June 4-June 10 3,171,227 4.4% 438,695 1.4%
June 11-June 17 3,447,528 4.6% 442,951 1.1%
June 18-June 24 3,636,525 6.0% 440,833 1.0%
June 25-July 1 4,326,304 7.0% 419,696 1.2%
July 2-July 8 4,463,797 8.2% 429,804 1.1%
July 9-July 15 5,125,361 8.5% 447,073 1.1%

It’s clear that things were worse this week than last week. There were more positive tests in all four regions. The West may have managed to reimpose enough controls slash adjust enough behaviors to be in a better spot, in particular with California and Arizona looking like they may have turned the corner. The Midwest and South don’t share that pattern. The Northeast has reason to worry, and Pennsylvania is attempting today to walk back some of its reopening.

But it’s also clear that the second derivative improved, and the trend line broke. Positive test rates are only slightly up, and didn’t seem to rise much day to day throughout the week. It’s possible that the second wave peak of infections will be in July. There was a large jump in test count this week, and wasn’t one last week, so the smooth lines on the infection count graphs are misleading. There’s good reasons to mostly be optimistic here.

New York has had a scary last two days, but mostly I see this as the state having its internal problems under control while importing cases from elsewhere on a continuous basis. One friend had a neighbor infected because they flew to Florida for a July 4 party, which qualifies for first rank Covidiot status.

New York and New Jersey continue to see declining death counts week over week, despite infection rates that clearly stabilized three or more weeks ago. Given how rapidly this is continuing, this is a sign that deaths substantially lag more than a mere few weeks. I can accept declining real death rates, but not this fast. As usual, see my spreadsheet for full details.

Mysterious Deaths Solved

Twice previously I’d talked about the apparent rise in ‘mysterious deaths,’ or those classified under R00-R99. The speculation I’d read was that this was being used to hide increasing numbers of Covid-19 deaths, which seemed plausible to me. Last week, I finally got linked to raw data, and posted an analysis.

I learned a few things along the way (see post for details), but the big thing is that no, this wasn’t a massive conspiracy or distributed cover-up or anything like that. It is our system for classifying deaths being slower than molasses, often taking months to figure out causes of death. So most of the increase in mysterious death in recent weeks will fade away over time, leaving a bump not big enough to impact the grand scheme of things.

Mysterious Tests Incoming

However, in the data becoming fraudulent front, we now have a new problem on its way.

According to reports I have seen, the Trump administration has awarded a no-bid contract to a private corporation, and instructed hospitals to send their test results for compilation to this company, rather than sending them to the C.D.C. 

I’ve been hard on the C.D.C. lately. I think it’s been deservedly so. I disagree with their policy recommendations, and what I presume is the motivation behind them making those recommendations. I’m also still rather frustrated with the botched initial testing kits and how that was handled.

This, however, is something else. As far as I can tell, Trump was displeased that the C.D.C. refused to neuter its school reopening guidelines when he requested they do so, and continues to engage in the heinous act of telling people how often Americans test positive for Covid-19. Frustrated for long enough, he has decided to do something about these scandals, and strike back.

I hope this is a false alarm, and either the reports are incorrect, or hospitals will ignore this new instruction and send the data to the C.D.C. (and/or make it public) regardless of whether they also send the data to the new provider. Alas, it seems the reports are probably true.

We must remain vigilant. Ensure the worst does not occur, if we can do that. Recognize the worst if it does occur. It’s less bad to slow the statistical compilation of test results than to actually slow the testing down, as Trump previously explicitly confessed to doing. It’s still very, very bad.

It’s an interesting question whether it’s better or worse to falsify the records outright, which likely is his preference here, than to actively slow down testing. My gut and aesthetics say falsifying the records would actually be worse, as hard as that might be to defend to many.

From here on in, we need to constantly ask the question, are the numbers real, or are they manipulated lies, and if so to what extent.

The good news is that the Covid Tracking Project does not anticipate that this will interfere with their methodology, since it is assembled at the state level. It would be better if the Federal Government compiled accurate data, for many reasons, but as long as we have the state-level data that was all , hopefully we should be fine.

Where Is The Line

This week, Robin Hanson noted that many media outlets and experts predicted a surge in Covid-19 deaths Real Soon Now, and asked if any of them were offering betting odds. In particular, he asked about odds on when we would reach an average of 1100 daily deaths, roughly midway on a log scale between deaths at the time and deaths at the previous peak.

Needless to say, no. No one was offering betting odds. With notably rare exceptions, such people never offer even probabilities, let alone probability distributions, let alone betting odds of any kind, let alone betting odds on people dying. We are lucky we get (actively and openly distorted to account for how regular people interpret and react to probabilities) odds on tomorrow’s weather. On its face it is a rather silly question.

As a request for such odds, from those such as myself who might ever offer them, rather than attempting to be a rhetorical argument against such ‘experts,’ it is far more reasonable.

Thus there were three questions.

What is my fair price?

What is the price at which I am confident I have substantially the best of it?

Is this a topic on which one can or should bet, given it is people dying?

To figure out how I expected short-term death counts to go, I looked at a few inputs.

The number of infections was going up rapidly, but how much? Positive test rates had bottomed out nationwide around 4.4%. At the time of the question, it looked like our new rate was roughly double that off 50% more tests. So it seemed like the default was that infections had somewhat more than doubled. It also seemed clear that death rates had stopped declining by 7/2, if you corrected for the holiday weekend, with the previous lowest true rate being at or around around 6/9. So we were looking at just over a three week delay there, with about enough increased cases to get us to around 1100.

There’s the argument that death rates were declining. That does seem to be the case, but how rapidly is that going to keep happening? Are conditions further changing? If anything, I’d expect death rates to rise again, because hospitals are becoming overwhelmed, and that should be the primary change from now to next month in terms of outcomes.

Thus, if we assume that reported infections hit the necessary levels around 7/9, something around 8/1 is when I would expect to see us back at 1100 deaths per day under baseline scenarios. It is already too late to impact this result much. That’s about why I replied that my fair was 8/1. To get a betting line, I asked how long it would take to take care of any plausible delay, and to account for things clearly looking sufficiently bad, and came up with 8/15. I offered a small wager at that point, and got no reply or counter-offer.

It is now close to a week later. I’d move up my timelines slightly, but only slightly, as the rises that we’ve seen were not unanticipated.

All that of course assumes the data is not likely to become (substantially more) fraudulent and hide deaths. Given the section above, I’m not sure we can assume that.

Predicting The End

If one were to extend prediction farther into the future, what is the best guess for what will happen over the next few months?

Here’s how I currently roughly think about projecting things into the future.

There are lots of variables, but we know a lot about what values they likely take. If I was going to go for a baseline scenario, I have clear best guess assumptions for them.

One big unknown is the policy response. This means both whether governments at various levels will be willing and able to impose restrictions, and to what extent citizens will choose to abide by those restrictions and/or do voluntary other action to slow the spread.

In my baseline scenario, we don’t see much difference in action from what we are seeing now, and thus the background rate of infection does not change much on this basis. I do expect to get modestly increased mask adherence over time, but I don’t want to count on too much gain from that.

Are there events coming up that will make the situation worse, and could create a third wave or sustain the current one? Yes. There’s one hell of an event coming up in September called school. It won’t come full blast, but it will happen to a large extent in many places, including many places it shouldn’t happen.

This also plays into my model that there are control systems at work. When things get worse or better, people adapt their behaviors.

The other question is how rapidly herd immunity sets in as a function of confirmed infections. That means we need to convert from confirmed infections to actual infections, decide how long we believe immunity lasts, and then guess how impactful those people being immune is on the new rate of infection.

As I have stated many times before, I will state again that I see no evidence that immunity is going to run out any time soon for substantial numbers of previously infected individuals. Thus, I’m going to assume that immune in context means immune period, for as long as is relevant here.

My best conservative guess for the current multiplier from identified to true cases is at least 5 times the number of confirmed infections, and it could still be as high as 10. It’s clear death rates have dropped. Previously a rate of 1% was my conservative estimate, there’s a large delay on deaths coming in, and the current overall CFR without NY/NJ is sitting just above 3%, albeit with a large number of cases that haven’t seen their final outcomes yet.

What does a given amount of immunity mean for the rate of infection? As I’ve long said, much more than one would naively expect. Power laws apply. My best guess is that the first 5% that we’ve already seen actually represents at least a 20% reduction in infection rates, and that a 10% immunity rate would be good for at least 30%. Again, I consider these very conservative guesses. It could be a lot more than that, especially if there is existing immunity slash variation in vulnerability, and also especially if many people continue taking extreme precautions while others are forced to or choose to accept massive exposure. Demographics shows that risks for different populations varies greatly, so there’s no good way to deny this effect.

Real rates of infection have roughly doubled over the last month. That’s about six serial intervals, meaning a best guess for general R0 (while noting that there is a ton of regional variation) of about 1.12. Thus, we could expect that a 10% reduction would get us back to roughly 1.0. If we have to do it all with immunity, going from roughly 5% to roughly 10% should definitely do it, barring compensating adjustments in behavior. If behaviors adapt to intentionally put immune people in positions of risk, things could improve further.

How long will all that take? Right now a reasonable guess is we have at least 300,000 true cases per day, so that would take at most three months if the case rate was already stable. It’s not stable under this scenario, so this should be closer to two. We then have to deal with schools reopening, and with changes in the weather, and with changes in which populations are currently infected, and so forth.

No matter how bad things are now, within the range of possibilities, we should be turning the corner, with the underlying infection counts starting to drop, by mid September, and definitely by the end of September. If things are worse now, then we’ll get more immunity faster, which will make up for it. That doesn’t mean things are even as good then as they are now, merely that they start improving.

Deaths likely peak somewhere between 1,200 and 2,500 per day. Less deaths than that would mean death rates have fallen a lot more than I expect, or herd immunity works the way my intuitive models think it does and therefore is already setting in to a large extent, either because power laws are more powerful or more people are immune than the conservative assumptions. I do think there’s a bunch of probability mass in ‘death rates have dropped and herd immunity is already turning the tide so we peak relatively soon at a number that isn’t that high.’ Note that this is compatible with usually hitting 1,100 deaths relatively soon, because in relatively good scenarios the peak comes relatively soon.

More deaths than about 2,500 would mean a health care system collapse.

Then we see if schools set things up for wave three. Given experiences in various places including Israel, as well as common sense and my belief in physics, I believe that schools will increase infection rates substantially wherever they are open, and doing ‘part time’ is going to not help much – see the discussion last week, since my analysis has not importantly changed.

There wouldn’t be a fourth wave unless slash until immunity fades, by which time I hope slash expect a vaccine. Vaccine progress tentatively looks quite good.

Remembering to Take Initial Viral Load Seriously

Suddenly this week it is out there in the zeitgeist that masks could reduce your own risk from Covid-19, not only by reducing your risk of infection somewhat, but also by reducing your initial viral load. Thus, if you are wearing a mask and get infected, your illness can be expected to be less severe.

Masks reducing initial viral load is obviously correct because physics. The question is only whether that load matters. Because our society is, in effect if not in principle, fundamentally opposed to science and learning things, we don’t know. Priors say that it is likely true.

Why was this message ignored for so long? Why is it still not being broadcast aggressively? This seems like a much more convincing argument to get people to mask up. Protecting others is great, but protecting the wearer will always be a contribution to one’s favorite charity. And the idea of ‘get less sick if you get sick’ is if anything more appealing than not getting sick as often for those not at much risk, because you get immunity on the cheap, as it were. For those at risk, the benefit is clear.

Shout it from the rooftops, I say. Any reasonable prior says we should expect it to be true. People should wear masks because it protects everyone involved by reducing the severity of any infections that do happen, in addition to preventing many infections from happening at all. Great pitch.

From what I can tell, the surge in discussion of this comes from this LA Times article. Someone wrote a snappy headline and framed this as ‘mounting evidence’ to fit it into the narrative. Kodus to the authors for doing so, as this seems to be an honest write-up. Highly above average journalism.

It also might come somewhat from this paper.

The evidence in it that could possibly have been news was the asymptomatic rate being high among some groups that were given masks. I worry these have a lot to do with the details of the clusters in the question other than masks, and also hope that the asymptomatic rate is actually very high in general, and also we shouldn’t need this information. But if it’s what gets people’s attention, great, let’s go with that.

The article also points to death rates in mask-wearing countries as highly suggestive, which could have been pointed out any number of times previously.

Why haven’t I been mentioning this recently, until now? Why didn’t this even make the list of hypotheses when discussing lowered death rates? What does this suggest about why everyone else doesn’t talk about it?

I pretty much have no excuse, given I did a bunch of work on the subject.

I have even less excuse given I take active measures to minimize my exposure in this way all the time. It’s central to how I’ve structured my choices of actions, worrying less about small stuff while keeping the small stuff small, and taking extreme measures to avoid big risks.

Despite all that, it faded from my mind in the context of writing, and I stopped thinking about it in that context until I was reminded. I need to think more about the details of how I allowed that to happen, so similar things don’t get forgotten in the future.

In the meantime, yes, this. Wear a mask to protect yourself and others, not only from catching Covid-19, but to ensure relatively mild cases for those who do catch Covid-19.

Sports Go Sports

It’s time.

Athletics are number one. Participants are heroes. Go team, yeah!

Except all completely straight. It makes it that much better to defeat you.

In case that wasn’t clear: We’re great, and also, you suck.

In an world of increasingly toxic and escalating wars between ingroup and outgroup, a never-ending parade of mutual cancellations and Aysmmetric Justice. In world of increasingly unfair play. In a world of increasingly falsified records. In a world where we’re afraid to go outside and entertainment production has come to a screeching halt.

We need sports. Real American sports. To teach us what it means to be the loyal opposition engaged in fair play. To raise up high those who accomplish great things, rather than looking for what we can tear down. To remind us to keep good statistics so as to understand the world. To have something to watch and then argue about all day without destroying Western civilization – the destruction of which, in an increasingly hot take, I think would be a bad thing.

Sports. Now, more than ever.

For those of you who don’t get the sports aesthetic, that’s the best I can do without writing a difficult post that’s been on my stack for a long time. Eventually, I swear.

Things are rather desperate. I tried watching NASCAR. If it hadn’t been for the ‘competition caution’ I would have kept watching.

I have missed being temporarily adjacently victorious, or at least temporarily adjacently agonizingly defeated, so much.

It’s time.

Baseball needs to happen on schedule next week, or I am going to be a very sad and angry panda. Looks like they’re going to at least try. Basketball is needed, too.

And remember: No football, no peace.

Make no mistake. There are going to be problems. Athletes are going to test positive, not only on arrival but during the season. The show must go on.

My instinctual first response is, take one for the team. Where that team is not only your team, but also America.

Sports as a whole bring joy to hundreds of millions of Americans. Individual events frequently bring joy to millions. They bring us together.

If a few athletes get sick as a result, that’s bad, and let’s do everything we can to prevent it that’s compatible with playing, but, well, is any of this meaningfully new?

How many weeks of playing football is worse for your long term health than definitely getting Covid-19? When a huge portion of young healthy people are getting it anyway? When a lot of athletes are getting it off of the field before play even starts? 

By participating in sports, such players get access to constant testing. They get far superior isolation procedures, in place of what would otherwise be frequent high-risk behavior for many of them.

It isn’t obvious to me that playing even raises the risk of infection for them at all. Let alone net increases the resulting health risks. It seems like it’s mainly Copenhagen Interpretation of Ethics, where these infections are things we’re interacting with and thus responsible for, even if there aren’t net more of them, whereas other infections we are not responsible for.

As for those who ask if it’s ‘responsible’ to ‘use all this testing capacity’ I remind everyone involved that there isn’t a constant amount of capacity over this kind of time frame. The leagues will pay good money for it so we’ll make more. The scandal is that the government is not doing that same thing. Back in March, I thought this argument was mostly nonsense but it made some sense. Now in July, it’s completely vacant.

We will do our best, and deal with problems when they arrive, but fundamentally, The Show Must Go On. If we occasionally need to hit pause to do a re-quarantine, well, so be it. But again. The. Show. Must. Go. On.

A seemingly troubling sign is that colleges are pushing sports into the spring, or limiting games only to conference play. I call upon us to embrace the upside of these actions.

By focusing on conference play, we make every game good. What was the point of cupcake games to begin with? There are 12 or more teams in each conference, so we can get a full schedule of good games. Every game counts. Every game matters. We should still do our best to do a rivalry week at the end, when there’s no risk of cross-contamination afterwards, to allow the most important out of conference game for each team. But that’s good enough. This is a better schedule. 

The college football playoff should by all rights expand, of course, because without cross-conference play it’s impossible to compare properly and choose who gets left out. Another opportunity! Let’s expand to six teams for this year. With round robins or close to it, each conference can send their champion, and the committee can choose one team at large to join them. Play two games prior to the bowls to get down to four, then play the Rose Bowl Pac-12 vs. Big 10, and Sugar Bowl as SEC vs. ACC/Big-12, as tradition mandates, or replace losing champions from the first round with whoever beat them. Championship game in the Orange Bowl a week later.

Even better, let’s intentionally postpone all of this until the Spring. 

Every year, sports fans face a cycle. Baseball provides the boys of summer. Then we have two forms of football plus the MLB playoffs, so the days are packed. There’s games all Saturday, all Sunday, Monday night, Thursday night and usually other nights too, and that’s without the basketball or baseball. One can’t keep up! Then we have NBA plus both footballs, and they’re mostly packed again. Then February rolls around and I hope you like shooting hoops, cause that’s all you’re going to be doing until April.

Or, alternatively, we could finally do the right thing. NFL plays Saturday and Sunday plus Monday night, gets rid of that dreadful Thursday night game. This allows full testing periods between games, so any flare-up gets caught. It gives us the chance to watch lots more NFL, and fight less with baseball and the World Series.

Then, once the NFL is down to four teams, we start playing NCAA Football in January. Conference play plus a final rivalry week for non-conference opponents. Like NFL, they get to play on both Saturday and Sunday, so now we get full Football weekends all the way through April. Then we play the bowl games in the north rather than the south.

Come on, everyone.

Let’s do it and be legends.

Dreams of a Vaccine

This week we get to end on legitimately great news. The vaccines look like they work!

Check out this Nature paper. All 36 patients responded. Excellent. Both Moderna and Oxford look like they’ve got strong candidate vaccines and are making rapid progress.

Meanwhile, China is confident enough in its candidate to use it for its military. That’s not something one does without high confidence in safety and a strong prior that it’s likely to work.

There’s also this call for challenge trials, from some people whose voices have some chance of being heard. Still a long shot, but Americans are known for doing the right thing after exhausting all alternatives. It might be time!

I Have Decided To Give Back to the Community

However, unlike Twitter, I have not been hacked, so I will do this via providing hopefully useful information, rather than having it take the form of providing a Bitcoin address. I thought about it, sure, cause who knows what might happen. But it seemed rather tacky.

See everyone next week.




New Comment
20 comments, sorted by Click to highlight new comments since: Today at 8:03 PM

Thought I would plug a really fascinating recent preprint. It is entirely a theory paper but it makes sense of several interesting things that have been seen about the fact that type A blood is a risk factor for infection, type O blood is protective, and only a fraction of infected people infect most new people. If you have the ability to read biomedical literature I recommend it wholeheartedly along with all its references.

"Modelling suggests blood group incompatibility may substantially reduce SARS-CoV-2 transmission"

Short version: it is possible that transmission of the virus from one person to another is substantially hampered by blood transfusion incompatibility. Type O blood individuals become 'universal viral donors' able to maximally infect most people, while being at substantially lower risk of infection from non-type-O individuals.

It has already been established that HIV, SARS-classic, and measles virions produced in tissue culture from cells expressing the type A antigen can be deactivated by serum containing anti-A antibodies, albeit at a lower rate than using antibodies specifically raised against the viral proteins. The ABO antigen is not a protein, but instead is a polysaccharide that is used to decorate many other proteins. If the viral spike protein incorporates an ABO antigen in its glycosylation it can be bound by antibodies, some of which will neutralize it. This would ONLY affect the incoming viral inoculum, not any virus produced in your own cells.

The data in the paper suggests a risk to type O inviduals of getting the virus from a type A individual of about 40% baseline.

This would also be complicated by 'secretor' status. About 80% of the European population puts the ABO antigen on proteins on all their cells rather than just blood cells, but 20% are 'nonsecretors' who only put it on blood cells. These people would fail to put the ABO antigen onto virions and transmit as if they were type O 'universal donors' but still get infected with heightened risk factors if they have non-O blood type status. This would also imply that the rate of getting infected by someone of an incompatible blood type who was a secretor would actually be only about 20% of baseline.

An implication of this work is that spread should be slower in populations with a greater *diversity* of blood types - a more even mix of A, B, and O - and that superspreaders should preferentially be nonsecretors and type O individuals.

Needs in vitro work and careful contact tracing epidemiology to back it up. But it makes way too much sense...

Note that this paper is just a mathematical model, without any actual data, so our default stance should be skepticism. The mechanism seems unlikely, but not impossible. Respiratory droplets apparently do contain whole cells (source), so a way this could happen would be if transmission involves not just virus particles in droplets, but infected whole host cells; in that case, the surface markers of those cells could make a big difference to the initial immune response.


No, I am talking about actual ABO antigen on actual virus particles, and specifically probably on actual spike protein.

The ABO antigen is not a protein, unlike the Rh antigen (the positive/negative factor - which is not empirically relevant for disease risk unlike ABO). It is a motif of a few sugars hooked together (three particular sugars in a chain for O, and two different branched 4-sugar motifs for A and B) that is used as a component of polysaccharide chains used to decorate proteins via glycosylation. It winds up attached to loads and loads of membrane proteins made by cells. In a nonsecretor it winds up on surface proteins of blood cells and endothelial cells only, in the other 80% of the population it winds up on surface proteins on all cells.

The spike protein is COVERED with glycosylation sites. They are places on the protein that polysaccharide chains defined by the particular enzyme milieu of the cell get added. Like many glycosylated proteins they're rather variable molecule to molecule and not precisely genetically predetermined by the protein sequence, though the protein certainly biases particular types of chains to wind up on particular pieces. In the case of the spike protein, these chains actually significantly help make sure that there's very few places on the protein that an antibody can bind without needing to bind to the sugars, since the sugars are the same sugars that you decorate your own cells with and you thus cannot generate an immune reaction against.

If you look at the references of the paper above, you will see it has been experimentally determined for SARS, measles, and HIV that the ABO antigen winds up decorating the spike/fusion protein and that anti-A antibodies can neutralize particles created in tissue culture that expresses the A antigen. It doesn't neutralize nearly as vigorously as antibodies raised against the protein proper, but a whole lot of the particles get neutralized in in vitro experiments. Presumably they bind to sugar chains near the receptor binding site and just physically block it from being able to touch its receptor.

So, virions made in your own body are effectively cloaked by glycosylation. But if an incoming inoculum includes a bit of glycosylation that you have antibodies against - an incompatible ABO blood type - antibodies might be able to bind to and neutralize a reasonable fraction of incoming viral particles before they are able to infect you and make virions that have your own ABO sugars.


I am quite aware that this is only modeling. The feature is that the risk factors of different blood types line up with what you would expect if there was this blood type incompatibility of spread in which getting the virus from someone with an incompatible blood type was only 40% as likely as from someone with a compatible blood type, and that this could explain a component of the extreme dispersion seen in which some people barely spread the disease and some spread it quite freely. Definitely requires in vitro and careful contact tracing work to follow up on and see if the blood type associations are somehow a fluke with a different mechanism.


What's the practical implication if true? Presumably it implies things like "You're type A so you go to the type O butcher's shop, or have an office meeting with a Type O colleague, and everyone is safe?" Does that mean that type A people become more valuable because there are less of them and they match easier? What about type B? Are ABs worse than As or are they now superstars? Etc.

Does this mean that blood relatives are in general more likely to infect within-household than friends or spouses, since their types match more often? Any implications? Etc.


Important to note that the author suggests that the best fit from observed blood type disease risk differences to a model like this would indicate that getting a virus from someone with an incompatible blood type would be about 40% as easy as getting infected from a compatible blood type.

One might be able to make arguments about intra-familial transmission versus extra-familial transmission rates with more math than I have time for right now.

In this model, if you ignore secretion status, type O people would be hardest to infect but most likely to spread. In the US about 45% of the population is type O, about 40% is type A, about 11% is type B, and about 4% is type AB according to the first source I found. You get the following fractions of the population being easy to transmit to and receive from for each blood type:

O - to 100%, from 45%

B - to 15%, from 56%

A - to 44%, from 85%

AB - to 4%, from 100%

If you assume that you are 40% as likely to spread to someone of a noncompatible blood type, you get a ease-of-infecting-others score for each type of: O = 100%, B = 49%, A = 66%, AB = 42%. Assuming you don't know their secretion/nonsecretion status.

If you take into account non-secretors, then this is all slightly wrong and you actually have most non-O people slightly less infectious than that and some people who are A, B, or AB with their usual higher risk of getting infected but who transmit like type O. This is me - I am a type A nonsecretor and thus would be more vulnerable to transmission from a full 85% of the population, while being good at transmitting to 100% of the population. Such individuals might be more likely to be important nodes in the network.

If you look at the paper they show that the biggest result of this kind of an effect on an epidemiological model is that the more diverse the population is in terms of blood types - a more even distribution of A, B, and O - the slower spread happens, and the more homogenous a population is in terms of blood type the faster it happens regardless of which blood type is dominant.

So we can test AB-blood populations for antibodies and compare that to the general population, and you'll know quickly if this theory is right or not?


No. It is already known that type O people are at lower risk of disease, with higher risk to type B then higher still to type A then highest to type AB, and that when you do a GWAS study for associations with disease the ABO locus is one of exactly two loci that fall out as very important. The question is, is that due to some intrinsic degree of resistance to disease brought on by the ABO locus or is it due to this sort of transmission incompatibility? To know you need to figure out actual pairs of people you know transmitted to each other and see if particular pairwise patterns of blood types are more likely than others (which has never been done), and test the virions themselves to see that they can be neutralized by anti ABO antibody levels that you find in mucous membranes (though this is highly likely given previous work on other viruses).

I noticed that you didn't discuss a winter second wave triggered by a rise in the R0 (caused by easier transmission in cold) or the R_t (caused by different behaviour patterns), because your predictions seem to indicate that it will all be over by then in the US, with partial herd immunity in many places. For parts of the US that aren't on that trajectory, or for Europe, this UK government report may be useful.

They go over a bunch of factors that might increase transmission and say that a 'reasonable worst case' scenario is R_t increasing to 1.7 in September and remaining constant, assuming effectively zero government action - total second wave deaths are about double the first, with a similar peak of currently infected individuals and the peak in January (meaning a lot of time to course-correct and reimpose measures). Honestly, this is a fair bit better than I would have guessed for the worst case scenario - a far cry from the sorts of things we discussed here in March.

They don't say how plausible they think this scenario is or give explicit motivation for R_t=1.7, just model the consequences of that change. This is because they claim that the degree of seasonality of Covid-19 is highly uncertain. Is this true? I've heard some sources say it's probably not that seasonal and others say it definitely is.

Another example of (Simulacra level B?) disinformation on Herd Immunity (

Prof Jonathan Heeney, a virologist at the University of Cambridge, said the findings had put “another nail in the coffin of the dangerous concept of herd immunity”.

There was no sign of any reluctance to make inferences about what the antibody levels imply even though that is highly uncertain, even from the domain experts they interviewed.

Meanwhile, China is confident enough in its candidate to use it for its military. That’s not something one does without high confidence in safety and a strong prior that it’s likely to work.

That makes sense for most countries, but my model of China is that they would be pretty willing to do it even if on low-ish confidence.


One big downside that I think you have to factor in is that a return to sports would signal a return to normalcy, which would make people less careful about wearing masks and stuff, and also probably influence policy-makers to be less careful.

That may not be a huge deal if you're correct about where this is headed though. It seems that the endgame is to have herd immunity save us, and that we're not too far away from it. I guess the big thing would be keeping the curve flat enough so that hospitals don't get overwhelmed in the meantime, so the question is to what extent sports would move us away from that.

I assume China is at least doing a cost-benefit - it's risky and expensive to do the vaccinations, and it's not a great source of information on whether it works, so the benefits at that scale are mostly because you think it will likely work. I agree that it's much less evidence than if e.g. Germany were doing it!

I dunno if seeing sports played to empty stadiums signals much of a return to normality. I agree that sports shutting down was a big wake-up call, and helped a lot, but I think every game will have a definite air of 'things are not normal' when they pan to empty fan sections, see the sidelines full of masks and distancing, and talk about which teams have had testing concerns. It might even model good behavior.

it's risky and expensive to do the vaccinations

Ah ok, I wasn't thinking about that part. I assumed that it would be trivial in comparison to the prestige and money they'd get if it worked. I was thinking more about how much they care about harming people.

I dunno if seeing sports played to empty stadiums signals much of a return to normality.

My guess is that even with the empty stadiums it would be at least a moderate push for people away from being careful, but I don't feel particularly confident in that.

My experience is that people seem to have a general single setting 'risk-dial' and are not taking lumpiness of risk into account much. Since they aren't going to form a gears level model, it's important for the list of 'things that are actually dangerous' to be small enough that they can remember them. I've been giving advice accordingly, most do not ask for details.

Agreed that this is how most people seem to effectively act, so you need to find a rule to tell them that 80/20s things as best one can. It's unfortunate that this is *wildly* worse than having a gears model, but dunno what can be done about that.

, but as long as we have the state-level data that was all ,





(the post still includes "[TODO] [RUN THE NUMBERS]".)

Fixed it for Zvi. Given that it was no longer present on the version on his blog.


I've noticed there seems to be a policy of not deleting typo threads after they've completed their missions, not sure why that's the case - when I ask for deletion after the fix it doesn't happen.

Huh, I usually delete them when someone asks us. Do you have an example of a case where we didn't delete them? 

New to LessWrong?