Friday, July 03, 2020

Part of the Covid Mortality Puzzle

We would like to know  the mortality rate, the percentage of people who get the disease who die as a result. That requires two numbers: number infected and number dead as a result.

The best estimates of number infected come from seroprevalence studies, testing a random sample of people for evidence that they have had the disease. There has been a good deal of discussion of the results, which generally indicate a much larger number, up to ten times as large, as the number of recorded cases.

I have not seen a careful discussion of problems with the other number. Under current circumstances, I think it unlikely that many deaths in developed countries will be missed, since the symptoms that suggest Covid are obvious and tests available. But we might err in the other direction, by counting deaths due to other causes but occurring to people infected with Covid. As a number of people have pointed out hospitals have a financial incentive to classify a death whose cause is ambiguous as due to Covid.

How large could that effect be?

I start with a high value of number infected in the U.S. — ten times the number of known cases, or 28 million. Assume, for simplicity, that those infections were spread out evenly over a period of four months and that each would test positive for a month. Then the number who would test positive at any given time was 7 million, or about 2% of the population.

The  U.S. mortality rate from all causes is 2,813,503 per year. I want to know how many of those would have died from causes other than Covid, over a four month period, while infected, so I multiply by .02 for number infected, divide by 3 for a third of a year of mortality. The result comes to about 19,000.

That should be an upper bound on by how much we are overcounting covid mortality by classifying deaths of people infected with Covid but due to other causes as due to Covid. The current count for total deaths in the U.S. due to Covid is 131,000. Assuming I haven't made any mistakes in my very rough calculation, that might be too high, but not by very much.

Readers are welcome to point out errors in either the logic or the numbers I am using, remembering that what I am looking for is an upper bound on the overcount.

Tuesday, June 30, 2020

Ayn Rand's _Fountainhead_ and the New York Times

One of the subplots in Rand's novel involves a conflict between the wealthy owner of a major newspaper and his staff. The owner wants the paper to support the novel's protagonist, an architect who, having provided the design for a housing project on condition that nothing be changed, destroyed the project when that condition was violated. The staff, left wingers, probably young, under the intellectual influence of the novel's villain, want the paper to take the opposite position.

The staff wins.

I was reminded of that by news stories about the conflict on the New York Times over the decision to publish an op-ed by a Republican senator arguing for the use of troops in response to rioting. That position is apparently supported by about half the population, but was viewed by the more woke members of the Time's staff as not merely worth disputing but so far wrong that it should not have been permitted to pollute paper's pages, even as a signed op-ed. They won, the paper apologized for having published the op-ed, and the editor responsible resigned, presumably under pressure.

It sounds from news stories as though the conflict was between older staff members with conventional liberal views, including the view that the appropriate response to positions you disagreed with was to argue with them, and younger staff members who preferred that such positions never be seen at all. We don't know what the view of the paper's owner was or to what degree the decision was due to pressure by staff members, to what degree by the belief that it would gain more subscribers than it lost.

But it was still similar enough to remind me of a story I must have read some fifty years ago.

Friday, June 26, 2020

Slate Star Codex and The New York Times

Slate Star Codex is a blog run by a young psychiatrist with a very wide range of interests and an extraordinary amount of intellectual energy, posting under the name of Scott Alexander. In recent years, more than half of my online time has been spent reading and posting on it — one reason I have neglected this blog.

Scott posts thoughtful, intelligent, entertaining, often original essays. Some are book reviews, including reviews of two of my books. Some consist of carefully reading a collection of scientific articles on some topic and summarizing the results. One old one looked at articles on Alcoholics Anonymous and concluded that both the claim that research showed it worked and the claim that research showed it did not work were false, none of the studies adequately measuring its effects. A recent one, published before official sources switched from telling people not to wear masks to telling people to wear masks, analyzed the existing literature on the subject and concluded that, while the size of the effect was uncertain, wearing a mask substantially reduced the chance of spreading the disease if you had it and probably reduced by a little the chance of getting the disease if you didn't.

One post, some years ago, started by explaining that he had noticed some intelligent people online with odd ideas who called themselves neoreactionaries. He looked for a summary and defense of their views, was unable to find it, so wrote one. If you ignored the introduction, you would assume the piece was written by a supporter of the position.

Then he wrote another post explaining why they were wrong. 

Some of his posts are humor, of a somewhat special sort. A recent one described several fictional political systems, I think in part inspired by my recent Legal Systems book. The first was acausal democracy. 

Electing someone president, as we do, before the beginning of his term makes no sense, since we don't know how good a job he will do. So instead they hold the election at the end of his term, to see how many people approve of the job he has done. This leaves them with the problem of propagating the information backwards in time, so that the candidate who will get the highest approval is appointed.

To solve that problem they use the market, issuing bonds for each candidate which will be cancelled if he does not become president, pay an amount proportional to his vote at the end of his term if he does. The candidate whose bonds sell for the highest price, hence the one investors expect to get the highest approval if he is president, is elected.

Scott then described how the system is crashed by an unscrupulous candidate who figures out a way to game it — and goes on to his next fictional system.

All of this — I could go on for pages describing past articles — is only part, and not the largest part, of the reason I read the blog.

The typical article produced a comment thread of five hundred to a thousand comments. There were also open topic threads, where commenters were free to start conversations on anything they were interested in, again with a similar result.

The commenters ranged, politically, from socialists (and at least one self-identified communist) to anarcho-capitalists, religiously from believing Catholics to Atheists, professionally from a (literal) rocket scientist to a (literal) plumber. The conversation was intelligent and almost always civil. It was the only place I knew of online where I could have civil and interesting conversations across a wide range of views.

A little over a week ago I got an email from a New York Times journalist who wanted to interview me for an article on the blog. I spent a little over half an hour talking with him on the phone. It sounded as though it would be a friendly article, treating the blog as an interesting part of the intellectual life of the Silicon Valley community, which it is. 

He interviewed a number of other people and, via email, Scott. It is not hard, with a little research, to figure out what Scott's real name is, and the journalist had done so. Scott asked him not to include it in the article, explaining why doing so would have large negative effects on him. The journalist refused, claiming that NYT policy required him to publish the name of someone he wrote about. Scott responded by closing down the blog, which is why I have been using the past tense in referring to it. The article has not yet been published, and a fair number of people, myself among them, have been trying to persuade the NYT that if it is published it should not reveal Scott's real name.

Scott is basically a left winger with libertarian sympathies, intelligently critical of many parts of the left — he reminds me in that respect of George Orwell. The comments are open to a wide variety of people, including Trump supporters, libertarians,  pretty nearly anyone who is willing to engage in civil conversation. The result has been a good deal of hostility from people on the left, and I am worried that people on the Times will see it as a left vs right fight instead of an issue of internet privacy.

For one example of both why some on the left don't like Scott and why they are wrong, he wrote an essay arguing that Trump is not in fact particularly racist or hostile to gays, offering evidence both from Trump's speeches and from the election returns — apparently Trump did less well with whites and better with minorities than Romney. He did not publish the essay until after Trump had won, because he didn't want to encourage people to vote for him. But a lot of people, right and left, think the important question is not whether what you say is true but what side it supports.

Scott's explanation of why he shut down the blog. (Read this first)
A fairly detailed news article about the controversy. Like several others, it points out that the journalist's claim about NYT policy is inconsistent with past stories they have published, including a recent one on a podcast host who was identified only by his online pseudonym.
An essay about it by Larry Lessig
A comment on it by Scott Aaronson 
New Statesman article
How to find past posts by Scott, now that the blog has been deleted

Wednesday, May 20, 2020

Virtual Meetup

Back before the pandemic, we hosted meetups every month or two for people who read Slate Star Codex, the blog I spend a good deal of time on. People came over to our house on Saturday afternoon for food and conversation — typically twenty to forty people — starting at 2:00, ending at 10:00. I'm trying to revive the meetups online, using Mozilla Hubs. People who read this blog and don't read Slate Star Codex are welcome to come. Feel free to invite friends — from anywhere in the world.

Zoom and its competitors are designed for a one to many interaction, such as a lecture. Hubs is a VR program, usable from your browser or with a VR headset. It puts you in a world, in this case a house of four rooms that I have constructed online. The closer you are to someone, the louder his voice, so it is possible to have multiple conversations going and for an individual to wander around listening for an interesting one to join. That's how our meetups worked, and I am trying to construct the equivalent online.

For those interested in the general idea, there is another program also worth looking at — Online Town. Hubs gives you a virtual world rather like World of Warcraft, with your character moving around in it. Town is a top down view of a two dimensional world, with each person represented by a tiny icon. The nice thing about it is that when you get close enough to other people to talk with them you get to see their faces, a webcam view like what you get on Zoom. So you are having a conversation with people whose faces you can see, unlike Hubs, but there can be multiple such conversations going on at once and you can move among them, unlike Zoom. I may eventually try that as an alternative.

The link for the meetup, along with a link to information about Hubs, is a page on my site. It starts at one, will presumably go until everyone leaves.

Sunday, May 03, 2020

How to Test a Vaccine

I have been making some calculations on the alternative ways of testing a vaccine, and unless I misunderstand something, the current procedure not only takes longer, it probably kills more people. Here are my calculations:

Method 1: Give the vaccine to N1 people. Wait a month. If none of them get the disease, conclude that the vaccine works. 

Method 2: Give the vaccine to N2 people. Deliberately expose all of them to the disease. If none of them get the disease, conclude that the vaccine works.

The following calculations assume:

A: We select N1 and N2 to reduce the chance of a false positive to no more than .05 .

B: Someone not already immune who is deliberately exposed has a .5 chance of catching the disease.

C: The probability that the vaccine works is .1, but if it works it works perfectly — probability of catching the disease zero.

D: The probability that the vaccine not only does not work but gives the recipient the disease is .01 .

In the U.S. at present, about one person in a thousand gets the disease each month, so with method 1, in the U.S., if the vaccine does not work each test subject has a .001 probability of getting the disease. So if it does not work, the probability that none of them get the disease is .999^N1. If we set N1=3000, that comes to about .05.

With method 2, if the vaccine does not work, the probability that nobody gets the disease is .5^N2. We set N2=5, giving us a probability of about .03.

With method 1, the expected number of people who get the disease because of the vaccination is .01xN1=30. The number who get it because because they are in the test and the vaccination doesn’t work is zero, since their exposure is the same as if they were not in the test. The number who avoid getting the disease as a result of being in the test and the vaccine working is .3 . Net increase in disease due to Method 1 is 29.7 .

With method 2, the expected number of people who get the disease because of the vaccination is .01xN2=.05. The number who get it because of the exposure (and the vaccine doesn’t work) is .9x.5xN2= 2.25 . The number who don’t get the disease as a result of being in the test and the vaccine working is .0005. So the net increase in disease due to Method 2 is 2.3.

For simplicity, I am calculating the number of people in the test who don’t get the disease as a result of the vaccine over a month in both cases. It’s small with Method 1, trivially small with Method 2. 

Adding all of this up, Method 1 results in 29.7 people getting the disease as a result of the vaccine trial, Method 2 results in 2.3 people getting the disease as a result of the vaccine trial. Method 2 also gives a somewhat lower chance of a false positive and produces a result about a month faster. 

This is obviously a simplified analysis — a vaccine doesn’t have to work perfectly to be worth using, and my particular numbers were invented. But given how much larger the first figure is than the second, the argument that we must use the first because the second is too dangerous looks implausible unless one believes that the chance the vaccine gives people the disease is lower than the chance that it prevents the disease by substantially more than an order of magnitude. 

Also, even if there is no chance that the vaccine causes the disease, the downside of Method 2 is tiny. A small number of people, two or three with my numbers, get the disease as a result of the test. Since you will be using healthy young adult volunteers, the chance of death for each is about one in a thousand. Getting a vaccine out a month sooner, on the other hand, saves about 20,000 lives in the U.S. alone. 

Am I missing anything? Is there any plausible set of assumptions under which Method 1 is better than Method 2? Alternatively, have I misunderstood what the methods are?

Friday, April 24, 2020

Looking for a few more volunteer voices

I now have almost all of the recordings of quotes I need for the audiobook of my Future Imperfect. The ones I do not have satisfactory versions of are:

“I disagree with your principles, so will require you to die for mine.”
(It's a takeoff on a quote attributed to Voltaire, so a mild French accent would be appropriate but not essential)

“Abu Hurairah (may Allah be pleased with him) reported that the Prophet (peace and blessings of Allah be upon him) said: “Allah, may He be exalted, says: ‘Who does more wrong than the one who tries to create something like My creation? Let him create a grain of wheat or a kernel of corn.’” 
(Ideally in a mild Arabic accent, mild enough to be easily understandable).

“I would prefer my child take anabolic steroids and growth hormone than play rugby. Growth hormone is safer than rugby. At least I don’t know of any cases of quadriplegia caused by growth hormone.” 
(The speaker is an Australian academic, so an Australian accent would be ideal, but since the readers don't know he is Australian, not essential. An English accent would be fine.)

“The specialness of humanity is found only between our ears; if you go looking for it anywhere else, you’ll be disappointed.” 
(This is by Lee Silver. I have a good version of it, but it's spoken by a woman and Lee Silver is a man, so I would prefer a male version if possible).

That's it for ones I still need. If you feel like it, you are welcome to do any of the others I have listed in previous posts — I can always use a better version. 

Tuesday, March 31, 2020

Looking for More Volunteer Voices

A number of people have kindly recorded for me quotes for my audiobook of Future Imperfect, but there are still a fair number to be done. I could do all of them myself but would rather  have voices that better fit the speakers. 

 A list of the quotes that still need to be done is webbed.

The book is also webbed.
Voices I could use are:

William F. Buckley
H.L. Mencken
Dorothy from the Wizard of Oz
The King James Bible
A sexy female voice
A mild French accent
A mild Arabic accent
A mild Chinese accent
An Australian accent
Some random male and female voices.

Anyone more want to help? Ideally quotes should be recorded in .wav, but mp3 or anything else that most sound programs can read — I use Sound Studio, Audacity, and Switch — is fine. You can email them to me at

Thanks to everyone who responded to my previous post on this. Special thanks to Tim Worstall, who is the reason I no longer need any more English accents.