Delta is now two thirds of sequenced samples from the past week, so we can be confident that it has taken over, and soon will be most cases in America, and soon after that most cases around the world. That’s bad news, but given we know the case numbers, it sort of is good news. It means we’ve already ‘taken the hit’ to the reproduction rate, mostly, and if we can stabilize things one final time, then unless there’s a new even worse variant then the last scare is over.
Things are not yet stable, but they continue to echo what happened this time last year, and things continue to be quite hot. If things cool down a bit (in temperature terms), several weeks pass and these trends continue, then they’ll be more clearly where we are at going forward, and we’ll be counting on further vaccinations and on past vaccinations kicking in. We should have enough momentum there to still get us over the finish line, but in many places it’s going to be close.
Let’s run the numbers.
I forgot that July 4th was coming, so my deaths prediction was dumb. Whoops.
Prediction from last week: Positivity rate of 2.7% (up 0.3%) and deaths decline by 5%.
Result: Positivity rate of 2.9% (up 0.5%) and deaths decline by 20% likely due to July 4th.
Prediction for next week: Positivity rate of 3.3% (up 0.4%) and deaths increase by 7%.
Predicting deaths this week is weird, and I’m predicting an increase because I think this week’s count is artificially low due to the holiday. So there will be some catch-up reporting and some reversion, even if the new rise in cases won’t have caught up to us yet. For the positivity rate, it jumped a lot the last day, but I have to assume it will keep rising for now. As it stops being constrained, the positivity rate will once again become a useful measure of our situation.
|May 27-June 2||527||838||1170||456||2991|
|June 3-June 9||720||817||915||431||2883|
|Jun 10-Jun 16||368||611||961||314||2254|
|Jun 17-Jun 23||529||443||831||263||2066|
|Jun 24-Jun 30||550||459||706||186||1901|
|Jul 1-Jul 7||459||329||612||128||1528|
I’d love for this to be real, but it’s probably not. It’s July 4th, and I keep forgetting to look ahead to the holidays in the next week before I make predictions. There’s no way this full decline is real, and it’s hard to know where the real number was.
My brain continues to not understand how it being the 4th of July causes three days of missing data across the country. That’s not a mindset I can process, even now, not fully. It definitely did happen, whether I can process it or not, and it’s very unclear how much of the difference was made up after the weekend was over.
|May 20-May 26||33,890||34,694||48,973||24,849||142,406|
|May 27-June 2||31,172||20,044||33,293||14,660||99,169|
|Jun 3-Jun 9||25,987||18,267||32,545||11,540||88,339|
|Jun 10-Jun 16||23,700||14,472||25,752||8,177||72,101|
|Jun 17-Jun 23||23,854||12,801||26,456||6,464||69,575|
|Jun 24-Jun 30||23,246||14,521||31,773||6,388||75,928|
|Jul 1-Jul 7||27,413||17,460||40,031||7,050||91,954|
That’s an unsettling jump especially given the holiday weekend with its disruptions to reporting, but given the jump I’m going to presume that there wasn’t much disruption to testing, only to deaths. That means we saw a 20% or so rise in cases, up from 10% last week, which was across the board and concentrated in the places one would expect. In terms of absolute levels things are still fine, and there isn’t much room for further acceleration because Delta is already dominant, but at 20% per week (R ~ 1.15) things will get ugly fast if trends don’t improve.
These changes are very similar to the regional changes one year ago this week, by region, in percentage terms.
Further discussion of likely future trends in the Delta section.
The decrease is unwelcome, but also clearly linked to the holiday, so it’s reasonable to expect some amount of rebound over the coming week. It could also be seen as the end of a previous temporary bump, in which case we wouldn’t expect much further decline right away but wouldn’t expect a bounce-back. My guess is the next week will hold steady from here, which is still not bad at all. Last week we were at 46.7% fully vaccinated and 54.4% with one does, and we picked up 0.9% and 0.7% on those respectively. That’s at least an effective 1.5% drop in R0, and I’m guessing more than that. As I say each week, that adds up fast.
Periodic reminder from MR that fractional dosing of vaccines would save many lives and end the pandemic faster with no downside. This links to a new paper in Nature Medicine arguing the same thing. And a preprint showing 1/4 doses of Moderna create immunity similar to natural infection. As a reminder, I continue to believe that Moderna’s day-after side effects are due to the dose being actively too large rather than simply not being strictly necessary. And Moderna is getting half-doses ready for children, so we know they can provide them, and also arguing feasibility was always silly. For children, the new half-doses should be overkill the same way the full-doses are overkill for adults.
Or simpler version:
If you have had one shot of the Johnson & Johnson vaccine, should you then get a shot of Pfizer or Moderna? If it is available, absolutely, yes you should. There’s every reason to think that mix and match will work here, one shot of J&J isn’t that differently effective than one shot of mRNA, and no reason beyond supply limitations to not take the second shot. Yet because it is not Officially Recognized all the Very Serious People are falling in line and finding ways to tie themselves up in knots and pretend that a second shot would not be useful. Here’s Nate Silver commenting on the bizarre NYTimes attempt at this. Here’s the Washington Post attempt. It’s all disgraceful bull**** designed to back up official policy that doesn’t make sense, with gems like ‘if you don’t feel safe enough without a booster, use additional measures like mask wearing.’ They’ve committed to one J&J shot counting as ‘vaccinated’ while one shot of Moderna doesn’t, and now they have to pretend that definition makes sense.
Remember to always be praising people when they finally do the right thing, regardless of whether they first exhausted all alternatives. The last thing you want to do is punish correct action by using it as justification for someone now being that much more blameworthy for not doing it sooner! Still, sometimes…
More vaccinations are the only viable way to deal with Delta. As Matt says here, there’s no going backwards, and we should have and will have very little tolerance for attempts to reimpose intrusive measures:
Within a few weeks Delta will be almost all cases, and there won’t be a need for a distinct Delta section, but for now it still makes sense.
Where are we on that right now? Mostly we’re there.
This is a seven day rolling average of data that’s lagged days behind that to begin with, and this matches where I previously presumed the trend-line was. If it was 65% or so in this 7-day rolling average of lagged data, current case counts likely reflect something more like 75%-80%, and the Alpha+Gamma is accounting for almost all of the remainder as per this graph. That means we have accounted for over 90% of the effect of the shift from pre-Alpha to Delta.
If we take this week’s numbers seriously in that context, and don’t make any adjustments, we’d get a final R0 for Delta under exact current conditions of R0 = 1.18 or so, once we take into account the lag in our case numbers.
Trevor Bedford thread attempts to ballpark the potential size of the wave coming from Delta. It’s useful to check one’s intuitions and estimate calculations against those of others. Trevor uses a basic SIR model, assumes that vaccinations are fixed and independent of infections (and counts only those currently fully vaccinated, but gives them credit for full immunity, and ignores varying distributions of vaccinations, much of which he notes is wrong but that he hopes such effects will cancel) and as always ignores all control systems in all directions and comes up with a further 11% of the population likely to be infected before this is all over.
In addition to all that, I’ve written several times about why SIR is not a great approximation in situations where different people are making varying adjustments in their behaviors and taking radically different levels of risk, which he’s also not considering. Nor does it take the children versus adults distinction seriously.
I think these issues mostly point in the direction of a smaller wave, and there’s one that towers over the rest of them, which is that vaccinations will continue until morale improves.
We are continuing to see vaccine shots given out at a good clip. At a bare minimum, I would expect the bulk of those who have only had their first shot to get their second shot. He’s starting with 46% fully vaccinated and an R0 = 1.18 (matching my estimate above). If we move that 46% to something like 55%, that’s enough to get R0 = 1, and all that would require is everyone who has had a first shot as of July 5 getting a second shot. It doesn’t work that way, because the first shot’s protection is already present and can’t be double counted, but other factors point the other way.
Another way to think about this is that when Trevor says 11% additional infections, that’s 11% additional infections or vaccinations, and can mostly be considered a hard upper bound. More vaccinations means less overshooting after R0 gets down to 1, combined with all the other issues, make me continue to think that the Delta wave is unlikely to get all that big.
There’s also the seasonality perspective. This week, we saw a 21% increase in cases. A year ago this week, we saw a 21% increase in cases, both with similar regional patterns. A reminder:
|Date in 2020||Cases Total|
|June 11-June 17||158164|
|June 18-June 24||215751|
|June 25-July 1||300510|
|July 2-July 8||365107|
|July 9-July 15||431972|
|July 16-July 22||461441|
|July 23-July 29||444797|
|July 30-Aug 5||392193|
|Aug 6-Aug 12||365028|
|Aug 13-Aug 19||322126|
Thus, it seems premature to conclude that we are in a permanent R=1.18 world going forward pending additional vaccinations. We also have this situation at very low overall Covid levels, so control systems via individual action have a long way to adjust if necessary. We also have vaccinations continuing, with this week’s disappointing crop still dropping R0 at least by 1.5% (and at least one vaccination week won’t be reflected in cases yet, likely 2-3, so we have a head start here).
My baseline scenario continues to be that cases rise for a bit, but things stabilize in most places before reaching levels that require behavioral adjustments, especially by the vaccinated. But I do continue to expect some regional/local issues in places with lower vaccination rates.
Periodically commenters will ask what the evidence is that Delta is more lethal than Alpha. I’ve seen such estimates from several sources, and no formal estimates doubting it, but such effects can’t hold a candle to the power of vaccination, and especially of vaccination of the most vulnerable populations.
Scott Alexander Analyzes Lockdowns
Scott Alexander writes Lockdown Effectiveness: Much More Than You Wanted To Know. This was not, in this case, more than I wanted to know. Instead it felt more like a ton of empty calories, with comparison after comparison and calculation after calculation that had so many caveats (that were explicitly mentioned – Scott plays fair on such matters) that I don’t feel much more enlightened than when I began reading, and the main update I made was that the evidence available to find was such that Scott was unable to provide an update.
Comparisons of this form stack the deck in favor of lockdowns, because they discount non-GDP effects (SBF’s #3), and also by considering the average countermeasure against the average gain, instead of comparing effects on the margin.
That second one is worth unpacking a bit if it isn’t obvious to you. The value of lockdown was being considered en masse rather than on the margin. Given that we end up controlling the situation either way and never end up in ‘everyone is infected’ mode, stricter long-term lockdowns have increasing marginal cost per case prevented or life saved. Thus, if it’s not clear whether lockdowns in general are good, or whether lockdowns above a given much lower level (like red state level versus blue state level) are good, that means that on the margin we did too much locking down. If we didn’t do enough, lockdowns up to that point would look very good.
Another note is that Scott is much too kind about the ‘maybe some of these provisions were not all that great’ aspect of the problem. Closing parks and beaches and playgrounds isn’t ‘we made lives worse for not much gain,’ it’s ‘we actively forced people into more dangerous situations and made the pandemic worse, while also making lived experience worse and burning out public tolerance and trust.’
What was the right rate of mandated lockdown, given our ability to prioritize measures? My belief is that private action reacts better and the control systems are very strong, and that the real reason to do lockdowns is the tail risk of complete disasters (and the big silent reason not to do them is to keep that ammo for when you really need it and/or to minimize risk of complete disaster via loss of civil order and people losing their minds).
That’s the argument that an analysis like Scott’s is missing the central point of decision making under uncertainty, rather than stacking (or not stacking) the deck in a particular way.
Once we got past the first few months, I’m firmly in the ‘we did too much locking down on the margin’ camp. I believe things would have gone better if we had let people make more of their own decisions. Even better would have been smarter restrictions, but I’ve learned not to make that the comparison point in such questions.
In Other News
This seems like an excellent idea. Businesses liking it makes sense too, but in general there is the worry that ‘the government tells you that you can’t open, then tells you that you can open if you do X’ will make you happy whether or not the prohibition or requirement makes any sense, as it’s better to pay a small additional cost than stay closed.
Poor ventilation is quite bad for many non-Covid reasons, and it’s very good to see this acknowledged. Fixing air quality is an underappreciated cause area, whether via filters or better designs, and if that caught on more it would be a large upside to everything that’s happened.
Great news, New York gets to keep its outdoor dining:
Which of course led to this as the featured comment, gotta love the user name:
To which I’d respond on the sidewalk because I have never seen an outdoor dining area block a sidewalk to pedestrians in a meaningful way, not once, what city does this person even live in. It’s amazing we have any nice things at all.
And not as great news, New York, insert your own joke here:
More surveys at the original post, including that worry is down to pre-pandemic levels, but daily enjoyment still lags behind. My top takeaway here is that tough times make people better appreciate life, and that we all went through some tough times and then things improved, and we now have comparison points that make me feel better. Daily life scores are still lagging, so the ‘life is actually better now’ hypothesis doesn’t fit the data. It also doesn’t fit my observations otherwise, as life in general is pretty great but it’s still not fully back to the old baseline.
MR links to this LW post on the recent mask wearing studies, titled ‘we still don’t know if masks work.’ I agree with its finding that the study in question didn’t prove anything in any direction, but that doesn’t mean we don’t know if masks work, because we are allowed to know things that didn’t come from a Properly Done and Formatted Scientific Study.
Evidence that some people’s defenses against Covid kick in without causing antibody tests to come back positive. For now I don’t think this has much practical impact, since proving that something happens and showing that it happens frequently enough to matter are very different things.
Lab leak: Confirmation that many scientists will affirm its plausibility privately but not publicly. I continue to skip over most lab-related items, and the situation hasn’t changed.
NYTimes so no link, but MR points us to a birthday paper. The key finding is that your risk of Covid goes up when you have a birthday, presumably because you have a birthday party, and the effect size cares not whether you’re blue tribe or red tribe, implying their choices of party were remarkably similar.
In not-ready-for-prime-time-players AI news, IBM Watson has decided to start making tennis predictions. They are, shall we say, not so great.
That’s not how this works. That’s not how any of this works. Djokovic absolutely does not lose to Garin 43% of the time, or anything of the sort. You don’t need to know much about tennis to know that the world #1 is going to dominate the world #8 most of the time, yet Watson can’t figure this out, despite the training data providing more than enough information to solve this puzzle.
That would be fine if quite odd, if this was a little experiment IBM had privately done and then said ‘whoops, I wonder why that didn’t work.’ Instead, it’s publishing the output as if it’s the next iteration of tennis analysis. I don’t blame Watson, it’s a computer program. I blame the humans, who took a problem that AI should be quite good at, clearly botched it at the most basic level, then didn’t notice they’d botched it somehow.
That’s the part I don’t understand. I’ve gotten similarly nonsensical analytical results plenty of times, but when I do I say the whoops thing and learn about how to make better predictions. To paraphrase Seth, if they think this is publishable tennis analysis, keep Watson the hell away from any and all medical care, or it will end quite badly for all concerned.
The other part I don’t understand is how they got this kind of answer, given what I know about machine learning, and I’d love to hear a plausible gears-level theory of how this happened, even ignoring that the humans should catch it within five seconds. How did the algorithm get one of the most normal basically-in-sample situations so wrong?
If one is curious, my friend Seth offers this tracking of how it would be doing gambling, which can best be summarized as ‘lighting money on fire.’