Roundup of worthwhile reading for Friday

3 April 2020

Here’s the best of what I’ve been reading recently.

Bill Gates: How to make up for lost time

Highly recommended op-ed in the Washington Post. The key points:

  1. Adopt a consistent approach to control measures nationwide; otherwise those areas with relatively lax measures put everyone at risk.
  2. Ramp testing faster; prioritise those on the frontlines, symptomatic people in high-risk groups, and those who may have been exposed.
  3. Stop having states compete for scare masks and ventilators.
  4. For therapeutics and vaccines, “stick with the process that works: Run rapid trials involving various candidates and inform the public when the results are in. Once we have a safe and effective treatment, we’ll need to ensure that the first doses go to the people who need them most.” Start ramping manufacturing capabilities now to be ready.

Might the death toll to date be higher?

We have strong reason to believe that the data we have on reported cases understates actual cases significantly, as I discuss here. What about the death toll? These are taken by most epidemiologists and modellers as a more reliable source, but have the disadvantage of being a lagging indicator.

The Economist presents data that suggest that the actual death toll so far could be significantly higher.

There are good reasons to be concerned about under-attribution of deaths:

“Official death tolls for covid-19 may exclude people who died before they could be tested. They also ignore people who succumbed to other causes, perhaps because hospitals had no room to treat them. The latter group has been large in other disasters.”

In a few regions, journalists have been trying to estimate the total number of deaths, from any cause, and to compare those figures to the number of deaths we would ordinarily expect in the absence of COVID-19. Here are the results:

There are two important caveats to this approach.

First, estimating the counterfactual of what deaths would have happened without COVID-19 is just that, an estimate. We can’t know for sure what it would have been. That said, if you look at the supplemental charts in the article, you’ll see that the weekly number of deaths in the examples given have been consistently within one range, and then jumped far outside of that range. So estimates of the baseline level of death seem reasonable.

Second, even if we’re confident in our estimate of the number of excess deaths, above that baseline, we can’t know that all of them are directly as a result of COVID-19 infection. There could be other reasons, even reasons related to our response to COVID-19. Perhaps it was harder for people with other illnesses to get medical help. Perhaps they were less likely to seek help. Perhaps social isolation has put more people at risk.

Still, the overall point is very clear and compelling: in the five regions the Economist article looked at, if all excess deaths were due to COVID-19, we could be underreporting COVID-19 deaths by something like 100%. And that could be consistent either with a higher Infection Fatality Rate than we think, or a higher attack rate in the population.

What are models good for?

Zeynep Tufekc’s measured and intelligent piece in The Atlantic, Don’t Believe the COVID-19 Models (That’s not what they’re for) is exactly how I think about modelling. The key points;

So if epidemiological models don’t give us certainty […] what good are they? Epidemiology gives us something more important: agency to identify and calibrate our actions with the goal of shaping our future. We can do this by pruning catastrophic branches of a tree of possibilities that lies before us.

The most important function of epidemiological models is as a simulation, a way to see our potential futures ahead of time, and how that interacts with the choices we make today. With COVID-19 models, we have one simple, urgent goal: to ignore all the optimistic branches and that thick trunk in the middle representing the most likely outcomes. Instead, we need to focus on the branches representing the worst outcomes, and prune them with all our might.

If you’ve been reading these updates for a while, you’ll know that I’m a committed Bayesian, and like to think about a range of potential outcomes and the probability of each (what in the statistics world are called probability density functions). Modelling, or more simply, asking “what if” questions and trying to estimate the chances that each scenario comes true under certain circumstances, is an indispensable tool for this.

But there are some subtle concepts that get completely lost in the reporting of modelling efforts like the influential Imperial College work.

The first is exactly the point made in the article above. The goal of modelling is not to be correct; it’s to inform policy decisions. At the extreme, as in the case of the Imperial College paper, a model can change policy response in such a way as to ensure that it is incorrect.

The second is that when using models in this way, there is an uncomfortable fact that our simplistic notion of what it means for a model to be correct or accurate isn’t that helpful. Ordinarily, we say a prediction is correct if it accurately predicts the future. But sophisticated models like those epidemiologists are using, particularly those that use Monte Carlo (random) simulations, or consider a range of possible inputs and the corresponding range of outputs, don’t make a single prediction; they make a range of predictions and assign probabilities to them. So how can we say if they are right or wrong after the fact? I’m not going to try to answer that question in detail here; it would require pages of statistics and philosophy. But to say that a given model didn’t correctly predict the actual outcome in our actual universe doesn’t mean the model wasn’t “right”.

Here’s perhaps a simpler way of thinking about this. When you flip a fair coin, you know that there’s a 1 in 2 chance of it coming up heads. Say you flip a coin twice and get two tails. Should we say your model, which forecast a 50% chance of heads, was wrong? What if you flip it 100 times and get 52 tails? What if you flip it 1 million times and heads comes up 50.532% of the time? There’s some complex math we can do that lets us estimate the chance that the coin is in fact biased based on the number of trials and the distribution of outcomes; but in general, in ordinary language, we wouldn’t say that our forecast of a 50% probability of heads was wrong at least in the the fits two of these three cases.

Bringing these points close to home, I made a prediction back in early March that we would see 1 million confirmed cases outside of Mainland China on the 3rd of April, based on an extremely simple model. It looks like my forecast will be right to within one day (will report back on 4 April). Does that mean my model was right?

I don’t think so. I knew, and advertised, that my model was overly simplistic in many ways. By design, it will be useless at predicting when the epidemic will turn. It was trying to forecast confirmed cases, not actual cases, and so to the extent to which it got the growth rate roughly right, there is some unearned “luck” (bad luck for humans) that we have been so bad at detecting and reporting cases.

But the model did do the one thing it was designed to do. It answered the what-if question, “what if confirmed cases around the world continue to grow exponentially at the same doubling time we saw in late February?” The fact that this came true is not a reflection of the quality of the model, but our collective failure to act more quickly and decisively.

Are viruses alive? And how did SARS-CoV-2 evolve?

Two recommendations via the excellent blog Pandemic Pondering (by a professor of immunology).

Dr. Baker linked to this classic Scientific American article, Are Viruses Alive?. It’s a quick and worthwhile read.

I was also fascinated by his summary of a recent paper in Nature on the evolution of SARS-CoV-2 (the virus that causes COVID-19). Key points:

  • “COVID-19 did not just appear but has aggressively evolved to infect humans. This has been a progressive process that has led to a virus very efficient in infecting human cells.”
  • The evolutionary history may explain why the virus infects and replicates in the nose and the throat, which may encourage asymptomatic transmission.
  • “Another piece of bad news is that the high affinity binding of the virus to target cells may make it harder to develop a vaccine. Very strong antibodies would be necessary to block the virus binding to the cell, so the vaccine would need to be very potent to induce a protective immune response.”
  • “The somewhat good news is that the virus has evolved so well it may have reached its zenith! Any further mutations in the spike binding protein would likely make the virus less virulent (potent). This may protect us from the development of other, more severe strains of COVID-19 and explain why more virulent strains have not developed thus far.”


I’ve previously linked to this well-written Atlantic article, How Will the Coronavirus End? But I should have re-linked to it in yesterday’s post on the same topic, as it’s so good. If you haven’t read it, do.

Ken Rogoff on the economic crisis

I’m a huge fan of Kenneth Rogoff’s work on the history of financial crises. So it was with some trepidation that I watched his brief interview on PBS (full transcript available here). Highlights; or lowlights, in this case:

It is really hard to think of a historical parallel.

We are going to see a recession, at least in the short term, the likes of which we have not seen at least going back to World War II.

But I think what we do know is, the short-term drop in output and quite possibly in employment could be worse than 2008.

If the virus can be conquered, and we can reach a stable situation, I’m very optimistic that we will be able to do that.

But it will be different for different countries. Italy’s in trouble. I mean, they have big pension problems, big debt problems, growth problems going into this. Emerging markets are in trouble. We’re actually already seeing emerging markets start to fold under the weight of this.

(Q: You are someone who’s worried about governments taking on too much debt. That’s now not a consideration anymore?)

No, absolutely not. I mean, there’s never been a concern about our government’s defaulting. The concern is being able to borrow massively when you need to. That’s the whole point of saving for a rainy day. When it rains, you want to really open up the floodgates.

And, here, I just — there’s no limit. We’re in a war. You have to win the war. I would have no problem with the government debt magically going up $5 trillion in the blink of an eye, if we could get out of this in two or three months healthily.

Per-capita infection and death rates in the US

A useful, interactive map showing infection rates and deaths per 10K residents, based ont the New York Times dataset.

A quibble: areas with a single death but very low populations look misleadingly bad.

Takeaways: the New York City area and greater New Orleans have clear clusters.

Adding to the Resources section of the blog.

Few patients requiring mechanical ventilation recover

NPR reports on several small studies. This 5 March 2020 Lancet paper gives details based on 52 patients given assisted ventilation (out of 710 confirmed COVID-19 patients admitted), of whom 79% died.

Clearly, the claim isn’t that mechanical ventilation causes fatalities; patients who require mechanical ventilation are the sickest.

But if ventilation is consistently this ineffective, it raises questions about the relative priority (given finite resources) that should be put into increasing ventilator capacity, vs other interventions that may be more impactful. (Of course, more ventilators are good; but at the limit there may ways to use the same resources that would save more lives.)

The studies are very small so we should be cautious about drawing conclusions at this point.

Prospective vaccines and therapeutics

I haven’t written much about the many prospective vaccines and therapeutics in the pipeline, mostly because I feel completely unequipped to say anything intelligent about them.

A friend shared this resource, which is apparently trying to track everything in the pipeline and has some interesting metrics.

Niskanen’s Live Video Briefing Series

I’m a great fan of the Niskanen Center, a centrist think-tank in Washington DC. If you’re not familiar with them, take a few minutes to read their conspectus (not COVID-19-related).

They have just announced a free, live video briefing series via Zoom, which looks very interesting. Below are the announced events. I’ll be joining several of them, particularly those featuring my friend Steve Teles and his co-author Brink Lindsey, co-authors of the outstanding book The Captured Economy.

Hosted by Niskanen’s Sam Hammond
Friday, April 3, 2020, 10 AM EST []The $2 trillion coronavirus aid package offers relief to several groups impacted by COVID-19, including state and local governments, businesses, public health, and education. But individuals will (arguably) be the biggest benefactors of the bill, receiving approximately $560 billion in aid. Niskanen’s work on child benefits crystalized this month when Republican Senators Romney, Cotton, and Hawley all voiced early support for emergency cash payments to every adult and child in the country. Niskanen’s Director of Poverty and Welfare Sam Hammond will discuss details of the package, and the important changes Niskanen advocated for to reduce the paperwork required to apply for recovery rebates, enhance fiscal stabilization for states amid falling sales revenue, increase SNAP benefits, and expand health coverage for COVID-19. Hammond will discuss the development of the CARES Act, details about the relief package, and how Niskanen plans to impact multiple aspects of the next phase of recovery. 
Hosted by Niskanen’s Geoff Kabaservice, Kodiak Hill-Davis, and Kristie De Peña
Thursday, April 9, 2020, 2 PM EST []COVID-19 has propelled the discussion on election security to the forefront of efforts to prepare for 2020 general elections. In the $2 trillion coronavirus aid package, $400 million was dedicated to “Election Security Grants” to prepare for, and respond to, coronavirus domestically or internationally. The CARES Act makes progress in ensuring the states are able to effectuate fair and reliable elections, but the changes are just the first step. Niskanen’s Vice President of Policy Kristie De Peña will review the provisions of the CARES Act and broadly highlight what improvements should be a part of a “Phase IV” package; Niskanen’s Director of Political Studies, Geoff Kabaservice will highlight specifics about proposed solutions, like mail-in voting, and additional voting modalities; and Niskanen’s Director of Government Affairs, Kodiak Hill-Davis, will review how Capitol Hill is responding to changes in policy and how politics will—as always—play a critical role in reform. 
Hosted by Niskanen’s Kristie De Peña and Matthew La Corte
Wednesday, April 15, 2020, 11 AM EST []COVID-19 affects everyone, regardless of their immigration status. It is more critical now than ever that all individuals have access to testing and treatment that will help America overcome this crisis, and are not subject to treatment and conditions that will increase their likelihood of exposure or needlessly risk their health and safety. Niskanen’s Vice President of Policy and Director of Immigration Kristie De Peña will review the multitude of impacts COVID-19 is having on immigrants’ ability to seek treatment and care, and where changes must be made in order to keep them—and the American public—safe. Niskanen’s Governmental Affairs Manager for Immigration, Matthew La Corte, will detail changes that would help protect the health and safety of key immigrant employees in the United States, and how we can capitalize on the role of immigrants in critical healthcare and essential jobs. 
Hosted by Niskanen’s Joseph Majkut and Nader Sobhani
Monday, April 20, 2020, 3 PM EST []The global pandemic and resulting economic shock will have implications for every issue in public policy, including climate change. In the next few months, government investment in clean energy and infrastructure will be positioned as a necessary economic stimulus, and bailouts for high-carbon industries could come with conditions for future environmental performance. In the medium term, climate advocates need to develop saleable plans for long-term recovery. Niskanen’s Director of Climate, Dr. Joseph Majkut, and climate policy associate Nader Sobhani, will discuss why COVID-19 reveals how poorly our society has prepared itself for predictable risks and how poorly designed government regulation can impede an effective response. It is essential that both climate advocates and skeptics learn from this experience and commit to not repeating them in our own domain.
Hosted by Niskanen’s Brink Lindsey and Daniel Takash
Tuesday, April 28, 2020, 9 AM EST []COVID-19 and our country’s faltering early response exposed a number of dysfunctions and rigidities in the American regulatory state, and thereby created openings for larger structural reforms that go beyond the current short-term waivers. First, the debacle over shortages of the delayed rollout of testing and shortages of PPE and ventilators highlights larger problems with the FDA’s highly risk-averse and slow-walk-everything institutional culture. In addition, various problems with the state licensing of healthcare workers have cropped up. State-based licensing has created obstacles to both telemedicine and moving healthcare workers across state lines to where the outbreak is most severe. Moreover, the scope of practice restrictions on nurse practitioners and physician assistants reduce the availability of primary care options at a time when all resources are under heavy strain. Niskanen’s Vice President Brink Lindsey and regulatory policy fellow Daniel Takash will discuss these and other issues and the prospects for constructive policy change in the wake of the crisis.
Hosted by Niskanen’s Steve Teles, Brink Lindsey, and Rachel Bitecofer
Tuesday, May 5, 2020, 11 AM EST []Niskanen’s Senior Fellow Rachel Bitecofer will lead off with an in-depth review of her election forecast, the methodology and assumptions that underlie it, and the various factors—including the pandemic—that could influence turnout and results in November general elections. Assuming a victory for the presumptive Democrat nominee, Joe Biden (for purposes of this discussion), Niskanen’s Vice President Brink Lindsey and Director of Political Studies, Geoff Kabaservice will discuss various scenarios for the correlation of political forces in 2021, and how that may influence the direction of the Biden administration. In particular, Lindsey and Kabaservice will consider factional rivalries between moderates and the progressive left in the Democratic party, and the possibility of some Republicans joining with moderate Democrats in a centrist coalition.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: