Bad jokes

Oh yeah, you know who’s been long overdue a challenge? The trans community. Oh they’ve had their guard down for too long if you ask me. — James Acaster

What’s this about?

First off, if you watch James Acaster’s 5 minutes on edgy comedians maybe you can skip the word salad below.

TL;DR: If you’re going to make jokes that involve a marginalised or oppressed group (where membership is not voluntary), do it thoughtfully. Try to understand how a member of that group will see your joke and try to understand whether your joke is going to make their day better or worse. If it’s going to make it worse, or you just couldn’t be bothered thinking about it, don’t complain about the consequences if you go ahead and tell the joke anyway.

At TCB we’ve had some incidents involving jokes that cross a line. I wanted to write down my own thinking on this, in the hope that it will help others to make their own decisions about jokes and to clarify how TCB makes decisions (although I’m not speaking for the club here).

First off, this is not a new problem. I recommend listening to Marc Maron’s interview with a historian of comedy on this topic. Acceptable speech and acceptable humour has always been changing and people have been complaining that “you can’t say anything anymore” for at least a century.

What is new is that it’s no longer TV-execs who get to decide. Now audiences through social media have huge influence, the “woke mob”, if you like. This has also made audiences in real life more vocal and more willing to complain.

Comedy clubs also need to get involved here. They put people on a stage in front of an audience. They have a responsibility to ensure that they are not making the world a shittier place in the process.

Offence is a red herring

I think talking about “offence” is counter-productive. People can be offended by anything. It’s almost a choice on their behalf. You can’t argue with them. You can’t tell them they’re not offended. If I offend you, I haven’t necessarily done anything bad. You might just have very different sensibilities.

I do material that would offend nationalists and xenophobes (of several countries). I am happy to offend them. I consider it a positive thing.

TCB’s policy about unacceptable jokes is offensive to some people!

This is why offence cannot be the basis of a policy for a club (it can factor into policies for specific shows). Anthony Jeselnik is a highly offensive comedian but people aren’t trying to cancel him like they’re trying to cancel Dave Chapelle.

Being offensive is not an ethical issue.

You can skip the rest of the “offence” stuff and go straight to “harm” if you want. I just want to say a few more things about offence while I’m at it. They’re not part of the bigger argument.

Context

We don’t want customers who’ve come for lighthearted fun getting an hour of dead-baby jokes. That is not a good way to build a customer base.

That said, people being offended is not a huge concern at TCB. For two reasons.

• we have shows that specialize in offensive material like Roast Battle and Your Hood’s a Joke. The audience want offensive material. Comics can get their fix by performing in these.
• at more general shows, comics read the room and choose how far to go without alienating the audience.

We don’t disallow offensive material but if you drop a massive load on the stage don’t be surprised if you receive some pointed advice. Not only have you done yourself no favours, the MC has to do the work to get the crowd back to a happy place before the bringing on the next act.

Doing it anyway

Some comics do material that is definitely offensive, myself included. I do material that might offend some people who I have no particular quarrel with, people who just have a lower tolerance for nastiness. I hope it’s mixed in with enough other good material that by the time I get there, they’ll let me do it or if it was too much, they can let it slide.

Offence can easily be forgiven if the joke is good enough. An offensive joke should be worth it. The perfect offensive joke is one where people laugh involuntarily and then cover their face in shame.

Causing offence is a practical matter. It has an impact

• your popularity as a comedian.
• who can be a happy audience member when you’re on stage.
• your bookability

but it’s rarely about crossing a line.

Harm is the problem

There are some jokes that cause harm to members of the audience or to groups in society. There are two key points that are frequently misunderstood.

• jokes can harm individuals and groups
• not all groups are equal

Causing harm is an ethical issue.

Why some groups are special

There are some groups you can’t leave. I was born in Ireland. So I’m a member of the “Irish” group. There’s nothing I can ever do to change that. Even if I gave up my citizenship, I would always be ethnically Irish. If someone made a joke about Irish people, it would be about me. The same goes for skin colour, gender, sexual orientation, disability, blood type (yes blood type, you should read a Japanese company’s code of conduct!) and many other factors. Some groups are not groups you are born into but once you join them you cannot leave, e.g. victims of assault.

There are some groups you can leave. It’s a choice to be a neo-Nazi or to stand on the road telling LGBTQ+ people that they’re going to hell. It’s also a choice to be a climate activist or a vegan. It might be hard to leave those groups but people do.

Religion is on the border line here. You can’t change what religion you were born or raised but you can change what religion you are now although this is easier said than done.

So when your joke involves a group of people, you should consider whether membership of that group is optional. If membership is not optional then you probably need to think about what impact your joke will have.

How do jokes harm?

Realistically, it’s very unlikely that any one of us will ever be in a position to do significant harm to any group of people. This is really about not being part of the problem, not adding fuel to the fire, not making things worse for people who are already oppressed, marginalized and struggling because of things they do not control.

It’s about the impact your words have on the people in the audience that night but it’s also about not giving oxygen to hate-speech. People are literally killed for being members of groups they did not choose to join and cannot leave.

If you are not a member of one of these groups you may not have thought much about this. I went through life as a straight, white male, in a 99%-white country, mostly not thinking about this stuff. It wasn’t until I joined a company with a very diverse and very vocal set of people from all kinds of marginalised groups that I started to understand it.

I am incredibly lucky that life never forced me to understand this stuff the hard way. I learned it by reading other people’s stories and their explanations of how seemingly harmless statements, policies or practices were actually causing harm in ways I had never thought of.

I wish I could recommend books on this topic. If anyone has suggestions, please let me know.

Just to align with James Acaster’s bit and because some of the incidents at TCB have been about this, I will talk about trans-phobic jokes.

Realistically, nobody is going to go beat up a trans person because you made a trans-phobic joke. The problem is that the trans person in the audience, who has a legitimate fear that they may be beaten up for just walking down the street but who came out anyway, now has deal with trans-phobia instead of the night of comedy they planned.

It’s not about whether you are trans-phobic or not. I don’t think I’m trans-phobic but I’m sure I could thoughtlessly make a trans-phobic joke. If I was told I had done so, I hope I would accept the feedback.

This would not be someone telling me what I can or can’t say. This would be someone telling me the impact that my joke had on them or others and me deciding to avoid that impact in future.

But they do jokes like that

Sometimes gay comedians make jokes about gay people. Sometimes (always!) Japanese comedians make small-dick jokes.

I think of this a similar to self-deprecating humour. There is a tolerance for people making fun of themselves that extends to their own groups. It’s fine for me to get on stage and make jokes about my bad fashion sense or my thinning hair. Unless it’s a roast battle, it’s not OK for you to just attack me from the stage. The same applies to groups.

This might be OK, it might not. Really though, that is for other gay people or other Japanese people to decide. It does not give anyone else a pass to make the same jokes.

Why is it OK to attack straight, white men?

Personally, I don’t think it’s a great idea. Comedy-wise it’s lazy to rely on broad stereotypes (no offence intended Roast Battle). I also think it’s counter-productive in that it gives ammunition to the “anti-woke” narrative where straight, white men are the “real victims”, the only people left who can be attacked.

There are plenty of straight, white men who are struggling. However they almost certainly are not struggling because of their straight-white-manness. That’s why it gets a pass.

Straight, white men can mostly just shrug it off and enjoy the rest of the show. They’re not going to spend the rest of their night mentally composing a detailed refutal of the stereotype with an attached bibliography.

Cumulative harm

While no one comedian is likely to cause harm directly, at a comedy club, the effect of multiple comedians accumulates. Having an LGBTQ+ night is nothing if the LGBTQ+ community can’t go to your club any other night. If the club does not want to be X-phobic that can’t be a part-time policy.

Enforcing this policy consistently is hard. I hope this document can help people to understand the ideas behind it. They’re not complicated but they’re also not obvious.

Censorship

When someone asks you not do a joke for the reasons above, are you being censored? Yes, you are! If you got onstage and said “those people are different and shouldn’t be able to exist in society without being humiliated every now and then” you would also be censored but I don’t think you would be surprised.

I think the key issue is that people don’t understand that some jokes are just a different version of “those people are different and…”. It’s absolutely fine to do that to Nazis. They can stop being Nazis any time they want.

Wrap up

When a show-runner asks you not to do a particular joke they’re really asking you not to hurt a group that they want to protect. They are prioritizing that group’s welfare over your joke.

The above is my understanding of this issue. It’s “woke” in as much as being woke means trying to understand how you might be doing harm and choosing not to do it.

It’s not comprehensive. I haven’t said anything about making comedy from traumatic subjects. I have nothing useful to say there except that you should be very wary of attempting this. You are quite likely to get no laughs AND to upset people in the audience.

I think there’s also something to be said about character comedy, where it may be OK to say horrible things when portraying a character that the audience should disagree with. I love that Neil Hamburger was not allowed to do this joke on Fox News because it was “homophobic”. I think it’s only homophobic if you don’t understand that Neil Hamburger is the joke. In fairness to Fox News, they probably made the right call for their viewership.

I don’t think the stuff above is obvious. It’s not difficult to understand but I didn’t understand a lot of it for a large part of my life simply because I had no reason to (luckily I was nowhere near a stage).

I’m happy to discuss any part of this, in person or virtually.

A simple and surprisingly good model of SARS2 growth in Japan

A friend of mine has been playing with simple models of the spread of SARS2 in Japan since the beginning. It was always interesting but nothing mind-blowing. Recently he added vaccination into it and things got very interesting.

This thread will cover implications of the model and speculation about Japan’s future epidemic the technical details of the model speculation about why such a simple model is so good (in Japan)

Implications: The model strongly implies that the recent massive drop is due to vax. This bottom chart shows the 2 components, mobility and vax. Vax has been a massive negative contribution in the last few months, outweighing mobility since late August.

The vaccination effect gave us a lot of headroom to increase mobility without getting back into exponential growth but it looks like we're already at the edge of growth (according to the model and the public test results).

As vaccination-induced immunity wanes, the numbers could rise sharply but vaccination is still ongoing (at a lower rate than before), so that might be a while. The uncertainty about waning makes predictions very hard.

In general, forecasting in Japan has not been great. E.g. the recent massive drop seems to have blind-sided everyone. This simple model doesn’t really help to forecast case numbers because we can’t predict the G-Mobility data.

What this simple model can do is tell us how much we can add to mobility before entering growth. This could give us some control over our fate by triggering changes in behaviour or vaccination schedules.

Depending on what interventions remain in place, it might be possible for Japan to have a very normal life and stay out of growth all or most of the year with well-timed vaccinations.

This article makes a very similar point. I cannot find any details of Prof Osawa’s model.

https://www.japantimes.co.jp/news/2021/10/09/national/japan-booster-shot-plan/

Meta-implications: G-Mobility and vax are enough to account for almost all variation in the rate of spread. That implies that, in Japan, G-Mobility is an excellent proxy for interpersonal infection risk. Any model without something like that is probably not a good model.

Forecasting vaccination rates is pretty easy. Forecasting Google Mobility is hard. It is likely influenced by

• weather
• govt policies
• headlines about people dying untreated in ambulances
• backlash against holding the Olympics

but some of these are somewhat predictable.

The model: It’s relatively simple. It assumes that the rate on a given day (R_t) is some combination of

• 6 G-Mobility indices
• number of people fully-vaxxed
• number of people who were fully-vaxxed over 168 days ago (to model waning effect)
• fraction of delta vs non-delta

With just these, it does a remarkable job. Here’s the modelled vs the observed R_t for Tokyo (middle chart). It captures the trends very well - the various waves, the massive drop and the recent reversal of this as people get out and about.

It assumes that R_t is a weighted linear sum of these 9 things. It chooses the weights using standard linear regression against the historical data. That’s all there is.

You can read a bit more about it here

https://github.com/graphy-covidjp/graphy-covidjp.github.io/blob/master/blog/2021-11-01.md

You can find lots more graphs here

https://graphy-covidjp.github.io/#predictions

It’s updated daily.

Why it works so well? I think part of the reason is that the interventions in Japan fall into 2 categories masks, ventilation etc. These have been pretty constant from early on. govt mandates. These have all been of the type that show up clearly in the Google mobility data.

In other countries there are varying mandates for masks in public/school/etc. sometimes varying a lot by town/county. Tracking these and integrating them into a model adds much more complexity.

People in Japan rarely visit friends’ houses. In other countries, when closures are enforced, people go to their friends’ houses. This is invisible in G-Mobility. In Japan, “residential” almost certainly means “in your own home”.

I have concerns about the modelling of waning in that we don’t really have a lot of people in the 200+ days group, so its influence may have been underestimated and the 200 days threshold is somewhat arbitrary.

That paper about NPIs again

This is a follow up to my previous post about a paper that had some very surprising result about non-pharmaceutical interventions (NPI) and how strict lockdowns seemed to have less impact than light lockdowns.

The crux of this entire thing is that they feed delta(log(cumulative)) into a linear regression model for the log of the growth rate (g) but there is no linear relationship between delta(log(cumulative)) and g when g is moving around. There is if you use delta(log(daily)) but they didn’t. The result is that it is heavily biased towards NPIs which happen earlier and of course everywhere starts off by trying light NPIs and switches to heavier later. Hence the paper’s surprising results.

What’s new in this post is that I’ve fired up R and recreated the linear regression from the paper. The results are spectacular. I had no idea how biased this was.

The R notebook with several simulated epidemics is over here. The highlights are:

• 2 equally effective NPIs come out with an estimated impact on g of -0.2063621 and -0.0646044 respectively (with the later one losing out)
• An NPI that increases growth by 5% looks better than an NPI that immediately stops the epidemic!

That’s right, with this broken methodology, an NPI that makes things worse beats an NPI that ends the epidemic, simply because it happens earlier.

An algebraic approach to that lockdown paper

I just posted about this paper and I realised there is a simpler way to poke a hole in it.

The paper says

We define the dependent variable as the daily difference in the natural log of the number of confirmed cases, which approximates the daily growth rate of infections

and then defines a linear model for g. The details of the model are not important, let's just assume that nothing changes at all. What happens if we let the epidemic play out with a fixed g?

The important point is that they have used the cumulative confirmed case numbers. So let C_n be the cumulative daily totals of cases. Then
g = ln(C_n+1) - ln(C_n)
g = ln(C_n+1/C_n)
e^g = C_n+1/C_n
C_n+1 = e^gC_n

So their model for COVID19 with everything else held fixed, gives exonential growth directly in C_n. This is actually OK if the growth initial rate never changes but of course the whole point of this paper is to study changes in growth rate. C_n+1 is just C_n plus tomorrow's new cases. That depends only the reproductive rate of the virus and how many cases there were about a week ago. How many cases there were a few weeks or months ago has no place in the calculation. This is clearly not a valid model.

This is a much simpler way to see the flaw in the paper but unlike my earlier post, it doesn't give any insight into how this error skews the results in favour of earlier interventions.

Surprising methods lead to surprising results in lockdown paper

I ran across a paper by Eran Bendavid, Christopher Oh, Jay Bhattacharya and John P. A. Ioannidis. It looks at the impact of non-pharmacological interventions (NPIs) on the growth of COVID19 cases. It finds that less restrictive NPIs (lrNPIs) have good effects while more restrictive NPIs (mrNPIs) have no clear beneficial effect. This is a very surprising result.

I’m going to say up-front that I am neither an epidemiologist nor someone who works with this kind of data regularly. If I’ve missed something obvious or subtle, I would love to know about it.

TL;DR

This paper seems wrong. This paper takes an odd definition of case growth, one that will exaggerate the impact of NPIs that occur earlier and lessen the impact of later NPIs. Since almost every country starts with lrNPIs and then moves to mrNPIs later, the result of applying this methodology is that lrNPIs come out looking more effective than mrNPIs.

If you read the text of the paper the definition of growth seems standard. It’s the logarithm of the ratio of cases from one day to the next. However if you read the supplied source code, you find that it’s actually the ratio of cumulative cases. That is a very important difference and that is not made clear in the text of the paper. This statistic is not linear in the rate of transmission and running linear regression on this to measure the impact of NPIs does not seem superior to more usual ways of estimating growth, in fact it doesn’t seem valid at all.

Details

This paper has quite a remarkable finding. How could it be that forcing people to stay in their houses has no measurable impact on transmission while asking people to social distance does? It’s counter-intuitive but of course that’s what makes this worth publishing. And here it is published in a peer reviewed journal but I still couldn’t believe it.

So I looked for rebuttals of the paper. I found this, where the authors respond to several criticisms. Unfortunately the original letters are pay-walled. In the response to the letters, I found this surprising paragraph.

Fuchs worries about omitting the period of declining daily case numbers, but this is a misunderstanding. We measure growth of cumulative cases, which are monotonically increasing, and therefore never go below 0 (negative growth) in our study’s figure 1 (“Growth rate in cases for study countries”). The data that we include cover the period up to the elimination of rapid growth in the first wave (Figure 1).

Growth in cumulative cases? That’s an odd choice. Is this a error? The paper includes source code in a zip file. I uploaded it to a GitHub repo. You can see in the source for the Spain calculations

gen l_cum_confirmed_cases = log(cum_confirmed_cases)
lab var l_cum_confirmed_cases "log(cum_confirmed_cases)"

gen D_l_cum_confirmed_cases = D.l_cum_confirmed_cases
lab var D_l_cum_confirmed_cases "change in log(cum_confirmed_cases)"

...

reghdfe D_l_cum_confirmed_cases p_esp_*, absorb(i.adm1_id i.dow, savefe) cluster(t) resid

There’s no error. They used delta(log(cumulative)).

Now look at what they wrote in the paper.

We define the dependent variable as the daily difference in the natural log of the number of confirmed cases, which approximates the daily growth rate of infections (g). We then estimate the following linear models:

No mention of “cumulative”. The paper is not inaccurate but it is ambiguous. This ambiguity is resolved in the code in a way that is surprising and that also happens to favour finding lrNPIs vs mrNPIs.

The usual thing to do is to take either “ratio of cases day by day” or “logarithm of cases day by day”. Logarithms are useful because they turn exponential growth into straight lines and we have lots of tools for dealing with things that behave like straight lines - specifically linear regressions

So we have delta(log(daily)) (common) and delta(log(cumulative)) (this paper). They are similar and the paper was not explicit about which one was used, so I supposed the reviewers of the paper just assumed it was delta(log(daily)). The problem is that they are hugely different. This should have been made explicit in the text of the paper.

The Problem with delta(log(cumulative))

So what’s the difference between delta(log(daily)) and delta(log(cumulative))?

Let’s simulate an epidemic. I’m not going to use an SIR model. I’m just going to simulate exponential growth/decay. 10 days at 1.5x per day, another 10 at 1.1x and finally 10 at 0.6x This simulates steady growth, followed by a weak NPI reducing g from 1.5->1.1 and then a stronger NPI reducing g from 1.1 to 0.6. Whether you think additively or multiplicatively the second NPI is stronger

• additive -0.4 vs -0.5
• multiplicative 0.73x vs 0.55x

I’ve done this in a google sheet.

So let’s look at the graphs of what happens.

Here’s are the basic graphs showing daily new cases and cumulative cases. Nothing terribly interesting here although I would point out that the NPIs are far more clearly visible to the human eye on the daily graph than on the cumulative graph.

Now let’s look at log(delta(cumulative)) vs delta(log(cumulative)).

You can see that delta(log(daily)) gives us a flat line during periods of constant growth. This is exactly what we want and perfect for applying linear regression. delta(log(cumulative)) on the other hand gives us… I don’t know what. It’s dropping rapidly, even when cases are still rising! It’s not linear at all despite periods of constant growth. Applying linear regression to this seems unjustified, to put it mildly.

More importantly you can see that the later NPI is very clearly stronger in the delta(log(daily)) graph and applying linear regression will discover that. While in the delta(log(cumulative)) growth, it’s the weaker NPI which appears to have a larger impact!

Finally and most absurdly, if a perfect NPI appears in week 3 and stops transmission dead in its tracks, delta(log(cumulative)) would drop from from 0.2 to 0. So this perfect NPI would still appear less effective than the first 1.5x->1.1x NPI!

This definition of g is fundementally broken.

The authors themselves state that g is bounded by zero. So they know about this. If you look at the graphs in their paper you can see this effect quite clearly, they all head to 0 as time goes on.

Per-country growth from the paper

Contrast this with graphs of estimated R for the 10 countries in the paper.

At the end of the period (around 2020-04-10), UK, US, Sweden and Netherlands all still had positive growth rates but in the paper’s graphs, they are all hanging around close to 0.

This should have been a huge red-flag for the authors that something was up with their methodology.

Further oddness

This paper was published in 2021-01 but only looks at data until 2020-04. The zip file that comes with the paper contains data for cases and NPIs far beyond 2020-04. Why don’t they use more data? Wouldn’t more data give a clearer result? There is no justification given for the stopping point. The nearest is this statement in their reply to letters.

The data that we include cover the period up to the elimination of rapid growth in the first wave.

More data would certainly have made it very obvious that the delta(log(cumulative)) has some very odd behaviour. The graphs of “growth” would have stayed around 0 even as Spain, France and Netherlands headed into new waves.

For example here’s Spain’s daily cases and delta(log(cumulative)) (the data comes from the paper’s ZIP file it has some blips and squeaks that I didn’t bother to try clean up).

The massive winter wave is almost invisible with delta(log(cumulative)). As is pretty much everything else except the initial wave. This shows what a non-useful method it is to take delta(log(cumulative)).

Conclusion

When the methodology gets the wrong answer on simulated, perfect, noiseless data, something is wrong. When the methodology is surprising and the surprise is not even mentioned, let-alone justified in the paper or validate on well-understood cases, it’s alarming.

It’s possible I have missed some key point about delta(log(cumulative)) that makes it suitable for use or superior to log(daily). As far as I can see, the only “good” thing about it is that it gets the answer that the authors wanted.

I don’t have access to the software needed to run their code (Stata is expensive and I have no use for it day to day). If I did I would try 2 things

1. create some fake countries with simulated growth and simulated NPIs and apply the paper’s code to see if it gets the wrong answers
2. fix their code to measure growth correctly and run it all again to see how the results change. I suspect it would come out quite differently.

As it stands, the conclusions drawn by this paper don’t seem to be justified at all.

Super-spreader event on Japanese flight with strong evidence of airborne spread

Japan’s National Institute for Infectious Diseases published an epidemiological investigation showing long range airborne transmission on a plane. It occurred in Mar 2020. It was published in Oct 2020, in Japanese only. Google translate makes it pretty readable.

Main takeaways:

• started off following the “2 row rule” but after finding a bunch of infections near the index case they expanded and expanded and eventually tested 122 out of 141 passengers.
• found 14 PCR positive passengers.
• found several others with symptoms who they did not test.
• confirmed that all positives were an RNA match for the index case.
• it travelled far - furthest infection was 16 rows in front of index case with another 4 infections 9 rows in front and one 6 rows behind. Also, some of the symptomatic, untested people were far from the index case.
• index case had a severe cough but did not wear a mask

I have translated the seating diagram published in the report from Japanese to English. I don’t know why there are 2 “3rd tests”, that was in the original Japanese.

Seating diagram of flight with test results etc

In the discussion they mention droplet and “マイクロ飛沫感染” - micro-droplet - infection in-flight. They also say that they didn’t have aircondtioning and ventilation information or pre-boarding passenger interaction information.

I often hear that “Japan understood this was airborne from the start” and it’s half true - some scientists here knew and the “3 Cs” guidance has been good but a lot of the response has been focused on droplets and fomites, cleaning and perpex barriers.

All of the information in this report was available in March 2020. It’s really disappointing that this was not published sooner and more broadly as it seems like it would have been strong evidence for airborne spread and also strong evidence against the safety of air travel. Especially given the full RNA analysis and almost complete test coverage os passengers

Are variants the reason Tokyo is back into State of Emergency? I don't think so

I keep hearing that the spread of SARS2 in Japan recently has been exceptional and variants are a problem etc. That's kind-of unclear. The week-on-week growth rate for Tokyo has not gone above 1.3x [1]. The national growth rate is similar [2].

Last April we saw up to 3x growth week-on-week. Obviously they were different times but Tokyo also saw some 1.5x weekly growth back in Nov too.

It's possible that if we didn't have the SOE-lite (State of Emergency) for the last 2 weeks things would be worse but it's hard to know if that has had any impact at all.

What is new is that the last SOE ended long before hospitals emptied out. I don't have Tokyo's numbers but nationally ICU usage peaked at 1043 on 2021-01-27 and was only down to 631 on 2021-03-21 when the last SOE ended (cases were at 300/day in Tokyo).

I think we're entering SOE again a month later not because the variants are crazy but because ICU beds were still 60-70% full when we allowed the growth to restart and there was lots of virus around.

Maybe PM Suga really believed that after lifting the SOE in March cases would keep falling. That makes no sense to me. He's talking about lifting this SOE after 2 weeks. Good luck with that. It's a stronger SOE, so cases will drop faster but hospitals empty out quite slowly, there's not a lot he can do about that.

Stay safe everyone!

1 Tokyo Stats

2 National 7-day growth stats

How not to calculate excess mortality

I got in a twitter argument with someone about COVID19 and they threw a surprising stat at me. South Korea had over 20k excess deaths this year. This made no sense to me. SK is maniacal about testing, their official COVID19 death toll for 2020 was 917. Did they miss 20x that many? Was there some other big killer? Is it a statistical blip? The answer is "none of the above".

The source was this WSJ article. It's pay-walled but the key information is this info-graphic which purports to show that many countries have vast quantities of excess death, above their official COVID numbers.

So the world has massively under-counted COVID19 deaths? Probably but the other key information is how they calculated excess deaths

Methodology: To analyze the pandemic’s toll, the Journal compiled weekly or monthly death data for 2020 and for 2015-19, where available. Most of the data was collected from national statistical agencies, either directly or indirectly through inter-governmental or academic groups. In a handful of nations, data was collected by health data organizations or local analysts. Epidemiologists use several methods to calculate excess deaths, adjusting for age composition, incomplete data and other factors. The Journal used a straightforward method, summing deaths for the portion of 2020 available and subtracting from that total the average number of deaths that occurred in the same span of each year from 2015-19. When the result falls below zero—when the 2020 death total fell below the average—some countries adjust the result to zero, boosting excess death totals. The Journal did not adjust in those cases. All totals are based on actual counts and comparisons. For some nations, the average was based on three or four recent years, typically 2016-19.

I have bolded the important part.

Unfortunately this straight-forward method is a fundamentally flawed methodology (did they not talk to an epidemiologist before publishing?). It ignores the fact that most countries have underlying mortality trends due to their demographics. Using their methodology South Korea has +24k excess death in 2020 but guess what, it had +21k excess death in 2019! This is what SK's recent excess deaths look like with WSJ's methodology. Here is the sheet if you want to explore.

As you can see, SK's mortality is rising pretty rapidly, presumably due to a population explosion in the 50s and 60s. I believe most countries are similar. This makes all of the numbers in the article somewhere between questionable and meaningless.

Applying their methodology to the whole world, we get an excess of 1.7M in 2020 and 1.3M in 2019. They did a subset of the world and got 2.8M which is also interesting, I don't know where that discrepancy comes from.

So while the world has surely under-counted official COVID death, WSJ's figures could almost be anything, an over-count or an under-count. What's bizarre is that they said "Epidemiologists use several methods to calculate excess deaths, adjusting for age composition, incomplete data and other factors" and then proceeded to just do it wrong anyway.

Addendum: New Zealand is quite similar. Here's the data

Japan Choral Association declares itself safe while ignoring aerosols spread of COVID19

TL;DR The Japan Choral Association have given themselves the all-clear to continue singing by producing a scientific report that focuses on splashes and pretends that COVID19 is not airborne.  

CBS reports that the Japan Choral Association has been involved in research on the production of splashes/droplets during singing. They measured droplets singing in 3 languages, German, Italian and Japanese. German produced the most, Japanese the least (the song apparently is a fairly aggressive kids' song).

It's awful because they chose not to measure aerosols. Despite knowing form early on that this was airborne (the 3Cs are precautions for airborne spread), many people in Japan remain super-focused on droplets (I had a nurse friend explain to me how 1 patient infected the 3 others in his room because they all used the same toilet, also the recent Oedo-line dorm faucet event).

There's a bit at the end where it says that it's not yet clear what size droplets are carrying the infection but that most important is thought to be large droplets. If this was true then ventilation would not matter. Non-droplet airborne infection has been demonstrated in the labs between animals. Also in www.superspreadingdatabase.com there is just 1 outdoor event out of 2000+, droplets exist indoor and outdoor, aerosols only accumulate indoors. The many choral (and other) super-spreader events cannot be explained by droplets, with people being infected many meters from the index case.

The safeguards they have developed for singing, spacing and patterning entirely assume droplet transmission.

The original Japanese research findings are here.

Worst of all they demonstrate a mouth shield blocks all splashes (yes it is but that doesn't make it safe).

I posted this as a long thread on Twitter. Reposting here for completeness.

I've been notified of a close contact by the COVID19 app. On the the 20th, the only time I went out for long enough was dinner. We (wife+kids) sat at the only outdoor seat of a restaurant. I assume someone inside the restaurant or maybe the bar next door is the contact...

First off, the app only notified me because I opened it. It does not check in the background. Another friend opened it after hearing about this and was immediately notified of a contact. The app has been broken since the start and the developers have been told multiple times...

Anyway, it seems pretty unlikely that I am infected but I did go inside to use the toilet and pay the bill with mask on. I have a bit of a rough throat since last night but that could just be the changing weather...

I phoned the number, gave all the details and now I have an appointment to get a saliva PCR test in 48 hours!

All relatively smooth. Mostly in Japanese. I started with the English helpline last night but they just gave me the number for the Japanese one. |If my Japanese was worse, I don't know what would have happened.

Also, I have to pay for the test (3000Y)! My Japanese is not good enough to have a fight on the phone but I thought this should be free.

More problematic is the outcome...

I'm the only one being tested because I'm the only one with symptoms. Is the app meaningless? They should be testing all or none. Has Japan stopped believing in asymptomatic spread again?

If they actually think I'm at risk, this is day 6-7 which can be highly infectious. Why let me sit at home with my family for another 48 hours before testing? If I'm being tested, I should be tested ASAP, not in 48 hours.

@nishiurah

Also, I was given no instructions about isolating, no questions about my kids' school or my wife. From a test-trace-isolate perspective, this is useless.

Results takes a day or two, so maybe 96 hours from now, I will know.

If my kids are infected then they were probably spreading in school on Thu/Fri but with this policy, the earliest they would even be tested would be if my result comes back positive on Wednesday. By then their contacts from last week will already be spreading...

Urgency and aggressive testing is the difference between winning and losing this war. Japan is asleep at the wheel. If winter and economic opening brings a spike in cases, this attitude to testing will be completely ineffective and a big shutdown will be required.

#COVID19

Disney and cultural genocide in Xinjiang

Disney has reached a new low in scumminess. I wouldn't expect them to make a live-action Mulan without filming in China but there was really no need to film in an area that was then and is now conducting cultural genocide and then thank them in the credits. This seems to be Disney demonstrating its willingness to basically do anything to be friends with China. There are a bunch of articles reporting this but this one goes into a fair bit of details and interviews some observers. If you must see Mulan, please find a way to do it without giving money to Disney...

COVID 19 and aerosols FAQ

The idea that aerosol transmission of COVID is very important is gaining traction. Some scientists are even speculating that it’s the most important vector, especially for super-spreader events. The lady who infected 27 people in a Starbucks in Seoul by just sitting under the aircon for 2 hours seems to be a very clear case of massive aerosol spread. WHO and CDC are not on board yet, but they seem to be generally slow to adopt new ideas. Japan has been onboard from way back with the 3Cs but didn’t really do anything to convince the world about it.

The conventional wisdom on aerosol transmission seems to stem from an old book on the transmission of diseases but there is no basis for believing that only a highly infectious disease like measles can spread this way.

This FAQ has been put together by a bunch of researchers and provides advice and good justifications for all the advice. In particular the sections on general protection and on masks are well worth reading. They justify everything with references, they are clear on what’s still unknown and they debunk a bunch of misconceptions. Key points are summarized below.

They have also made a truly awful acronym to help you remember. It’s Avoid Crowding, Indoors, low Ventilation, Close proximity, long Duration, Unmasked, Talking/singing/Yelling/breathing hard (“A CIViC DUTY”).

Just remember that an infected person, breathing in an unventilated room can fill that room with floating virus, possibly for hours after they leave. And there are a bunch of fairly obvious factors that make you more or less likely to get infected:

• how many sources there are (more people => more chances of an infected person)
• filtering (their mask and your mask)
• how close you are to the source
• how much aerosol the source is producing (from normal breathing to screaming)
• ventilation (how often is the air replaced by clean air)
• how long you’re exposed

A tight fitting mask, even a cloth mask protects you and others (great news, if aerosols really are a big factor your mask really protects you now too, not just everyone else, maybe now even selfish pricks can wear a mask). You want multiple layers of cloth and don’t worry about “the virus is much smaller than the holes in your mask” that is not how aerosol filtering works. Your mask is not going to be perfect but it’s like sunburn, factor 50 is way better than factor 10 which is still better than nothing.

There’s lots more detail in the doc it’s worth reading those sections in full.

If you were being careful, a lot of the above is stuff you were doing anyway. However, people and places that take very visible precautions against droplets but do nothing for aerosols are potentially making things more dangerous by giving the impression of safety and encouraging people to expose themselves to more danger.

the best age to have a child

 I recommend having a baby at 36. Then there will be be 9 times when your age is an even multiple of their age. Every time, you get to say "your age is definitely a factor" which is a very important dad joke. So if your child is 0 when you are 36 then here's your age, their age and the multiple

37 = 1 x 37
38 = 2 x 19
39 = 3 x 13
40 = 4 x 10
42 = 6 x 7
45 = 9 x 5
48 = 12 x 4
54 = 18 x 3
2 = 36 x 2

48 gives you 10 times but only if you make to 96!

So the math problem here is is there a limit to or can you find higher and higher numbers N such that i divide N+i for more and more values of i? I think there's no limit.. 36 and 48 have lots of 2s and 3s in them, so if i has lots of 2s and 3s so will N+i, the 5s and 7s are a lucky bonus.

Here we go again

TL;DR: More testing means much highers numbers but slower growth. We are probably going to have to wait quite a while before any emergency is declared. In the end, the spread will probably be much greater than last time.

Things are not entirely the same as before but not entirely different either.

The good:

Testing is way up from before and testing policy has changed. You can get tested on demand privately, if you want to pay for it. It seems getting tested publicly is now easy too. I know of two recent stories where people with mild symptoms got tested and it was pretty easy. When doing contact tracing they now test all contacts, not just those with symptoms. Combined with the targeting of some high risk groups for mass testing and this certainly has a large impact on the number of cases discovered. The official goal of the govt now is to identify positive cases as soon as possible, including asymptomatic cases.

Even though the numbers are much higher than before, it does seem like the rate of growth is lower. This might seem odd but it makes sense. If you go from finding and isolating 10% of infections to 30% of infections your numbers triple on paper but you reduce the speed of the spread (10% and 30% are very made-up numbers).

Treatment is improving. The govt has approved a couple of medicines that help suppress the violent immune over-reaction. So we may see fewer serious cases and fewer deaths for the same number of infections.

The mixed:

This time round, it’s spreading in younger people. Some possible reasons:

• younger people get no symptoms or light symptoms and we refused to test them last time around. Maybe nothing has changed and it’s just that we are detecting it in young people.
• (not sure if this is actually true) older people are sheltering more than before
• (not sure if this is actually true) nursing and caring staff are being tested regularly so infections are not spreading into hospitals and care homes
• targeting of host clubs etc biases detection towards the young
• when you shut everything down, undetected COVID mostly survives in young people. It could take several weeks for the infection to spread from young social circles back to old social circles.

If they can keep the infection away from old people, it will keep the ICU wards empty and the deaths low.

Hospital capacity has been increased. Many more hospitals are now designated to take COVID cases. Obviously more preparation and available care is a good thing. The downside is that the govt clearly knows that it’s not going to win and is betting on higher capacity to let it drag this round out as long as possible.

The bad:

There’s no doubt that we are losing again. A few weeks ago I was hoping that maybe the spike was mostly due to the testing change. It wasn’t.

The “serious” (i.e. ICU/ECMO) number has stopped falling and has been consistently rising for a week. It’s still single digits per day but it’s growing. Those people going into ICU in the last week were infected 2-3 weeks ago.

Testing capacity is growing incredibly slowly. They started publishing the numbers a little over a month ago and it has grown from 28254 to 33030 in that time (column P). Tokyo’s all-time record is 4,507. Ireland tests about 4500/day and finds 20-40 new cases per day (Tokyo finds 250-300/dat currently). Ireland is 1/3 the size of Tokyo. The rest of Japan’s testing numbers are even worse.

The avg daily testing rate (Column Q) is about half of the official capacity. It seems likely that Tokyo is using much more than half of its capacity and will soon max out. Then they will have to get stingy on tests again, e.g. no more proactive testing of host clubs or maybe make it hard for mild cases to get tested.

The national govt really don’t seem to care or have any ability to help. In fact they seem actively harmful. No responsible govt in any country is running a domestic tourism campaign right now, they’re all watching the numbers as they carefully try to restore their economies. Japan’s govt seems to see everything as secondary to the economy and their campaign-funding lobby groups. I don’t think they put any value on quality of life or are paying attention to the fact that COVID has many non-fatal but extremely nasty and chronic outcomes.

Speculation:

Present

The current numbers are not good but the numbers don’t mean the same thing anymore. My thinking had been that the true infection rate is maybe 10x what was being detected. Japan’s Case Fatality Rate was about 5% while it seems like in reality, COVID-19 has a fatality rate of about .2%-.5% that implies that the real case numbers were 10x bigger. That assumes that the virus is just as fatal here as anywhere else and that Japan’s fatality numbers are accurate (Japan was extremely stingy with testing, the true number of deaths may be much higher).

The new testing policy changes that. We might be detecting 2x or even 5x as much as previously, especially in Tokyo where they proactively test some high risk groups.

So we’re closer to the bottom of the curve than we might seem. We’re also climbing more slowly.

Near Future

I expect the govt are going to try to squeeze as much economic value as they can out of this round until one of these happens

• ambulances cannot find hospitals for patients
• ICU/ECMO capacity runs out

What’s extremely dangerous here is that it takes about 2-3 weeks for someone to go from infection to ICU. This means that we have to shut down 3 weeks before capacity runs out. If we screw up, we may exceed capacity. Exceeding capacity for ICU means picking who we try to save and who we leave to die.

There are a couple of ways to screw up.

• Wait too long.
• Have a sudden spike. E.g. as Tokyo runs out of testing capacity the rate of spread may increase, causing a numbers to spike.
• The SOE (state of emergency) is not as effective as expected. People are tired and don’t want to stay home. Young people are already acting like there is no problem.

Also the extra capacity means that the consequences of a screw up could be much larger. Essentially we will be going much faster than before when we hit the brakes.

The SOE was declared on 2020-04-16. By that day 185 people had died (officially). By 2020-06-16 927 had died. Now it’s 996. That means that 75% of deaths occurred in the two months after hitting the brakes. If we keep going until we fill ICU capacity the death toll of the second round could be enormous.

We’ll probably see deaths start to move again soon but hopefully more slowly, given better treatment and younger patients.

In the end, I think the infection will spread much more widely than April. Given the reduced speed of spread, it might take a long time for the govt to declare an SOE. Maybe 4 weeks, as a crazy estimate that I will surely regret. I suspect Tokyo is soon going to start racing ahead of the rest of the country (moreso than it already is). So Tokyo might go into SOE earlier than that.

Until then we will have an extended period with a very large number of asymptomatic cases wandering around. That will make this round much more dangerous for people with preexisting conditions. This Bloomberg story about South Korea describes someone getting COVID-19 from a person in a neighbouring karaoke box! I was hoping I would be going out and doing fun stuff by now but I’m staying put for now.

Is corona spiking in Japan again?

TL;DR This is all me extracting unjustified conclusions from too little information. There has been a pretty big change in testing, including testing asymptomatic contacts when contact tracing. The spike in cases in Tokyo and nationally is hopefully due mostly to more aggressive testing which will bring things under control but... It could also just reflect actual massive spread due to reopening. We can't know for a few weeks.

Something is happening with the corona virus numbers, most notably in Tokyo but also nationally. There are big jumps in the numbers and, it seems, some long-overdue changes to testing. Unfortunately there's still not a lot of transparency so everything is deduced from random quotes and reports here and there and then trying to read meaning into that. If you want to know what real transparency looks like, check out the Korean CDC[1], they have an FAQ and their daily reports describe all of the active clusters and tell you which clusters are still producing cases. I don't believe there is a Japanese equivalent to either for Japan (in Japanese or in English).

Things that have changed recently:

• Testing is largely being done by private labs now (the beige), with the public health centres being a fairly small portion now (the reddish). Graph [8]
• For Tokyo, the percentage of infections from unknown sources has been over 40% and even over 50% for weeks, although the 7-day avg is dropping rapidly, presumably a result of a small number of clubs providing a large number of cases. Graph [5]
  
• After 2-3 weeks of flatness, PCR positive numbers have started to spike up. The 7 day avg [2] is currently 165, up from 75 a week ago and 62 a week before that.
• Testing numbers are up having almost doubled from 3 weeks ago, to 5000-6000/day and apparently a desire for many more.
• Governor Koike says the spike is due to testing[9].
• There are several reports [3] that the contact tracers are now testing all contacts whereas before they would only test and isolate people with symptoms (anecdote from months ago, 2 bar staff test positive, one lives with the bar owner, they didn't test the owner because they had no symptoms).
• Shinjuku-ku will give you 100,000JPY (~1000USD) if you are found to be positive in recognition that you will need to stop working etc.[10]
• In Tokyo they are aggressively testing host bars and finding a ton of young male hosts who are positive. It's unclear if/why they are not testing hostesses. I've heard their clubs are not cooperating. Apparently the customers of host clubs are often hostesses whereas the customers of hostess clubs are businessmen[10]. On some days, I think, more than half of the cases announced were from clubs.
• The age profile of the positive cases is way younger than before. Consistent with detection of mild and asymptomatic cases and also hosts.
• The number of people hospitalised [4] had been dropping rapidly since early May. The number in ICU was also dropping. They started dropping around the same time. Hospitalised has been rising for about 2 weeks and ICU was still dropping until today.
• The head of the LDP's corona virus panel laid out some good stuff on what needs to be done, a refreshing change from the self-congratulation and mission accomplished nonsense. Including saying that previously testing was rationed due to lack of resources.
• They have started issuing alerts through the contact tracing app [6]
• The expert panel has been disbanded and will be replaced by a new panel. This is a mixed bag. It seems they conflicted with the politicians. It's very hard to know who was to blame for the previous bad testing policies. The new panel retains some of the old panel. It's also got Shinya Yamanaka a Nobel laureate who has been vocally critical about the lack of testing. This is a great sign but might also just be a good way to shut him up. Who knows...

As well as the links to the spreadsheet of the national stats, the Tokyo site [5] has a lot of the same numbers with nice graphs but Tokyo-specific.

So what does this all add up to?

It does seem pretty clear that testing policy has changed and we are finally looking for the asymptomatic cases and Shinjuku-ku are even offering a bounty. That bounty should be national policy, not something we have to wait for 23 wards and 47 other kens to implement. Maybe 40% of cases have no symptoms but can still spread. Catching them with contact tracing could be the difference between growing and shrinking case numbers. Wakayama did exactly that back in March [7] but at the time it had to defy policy. Maybe finally policy has caught up to them.

I have read recent stories in forums from people who are properly sick and still cannot get tested but if you are willing to pay for it, I believe testing is quite available. It's extremely short-sighted to put any barriers to testing people who have symptoms or are backed up by a request from a doctor. Hopefully the enthusiasm for more testing will spread. It would be really nice to see a strong, clear national policy that allowed any doctor get someone tested without charge. I suspect that the free testing at the public health centres is going to remain slow, manual and tightly rationed.

This part is total speculation. The spike in positives seems likely to be a mix of aggressive testing and some real growth. I feel like the jump is so sudden and large that it is more about testing than growth but that I might be doing some wishful thinking. The continued unlockdowning might also be a significant factor.

If my hope is correct, then we have a few weeks of continued rising numbers as we find more and more of the clusters that are already out there. The number of cases out there will not be growing rapidly, just the number that we are confirming. Now, for every cluster, we will find twice or three times as many infected people as before. Keep an eye on the number of people in ICUs and deaths. That lags by a few weeks but is much less impacted by PCR testing policy. If it stays low then the spike in numbers is coming from better testing.

Even if the new testing policy leads to control instead of growth, it would be great to kick-start the clean-up by mass-testing a few other at-risk industries, hostesses, massage, nail & hair, health, retail.

The alternative is that not that much has changed, Koike is demonising the night-entertainment industry because it's convenient [10] and we are all due for a another cycle of growth and lockdown (maybe locking down over Obon in August but I doubt we'll make it that far if it's really happening again).

I think both scenarios fit with the information and it's too early to tell which one is real. It would be awesome if Tokyo followed through on the poop studies [8]. It seems like waste water is able to give us almost real-time information on the infection levels in a city but until then I'll keep reading the tea-leaves - apparently floaters are good luck :)

1 https://www.cdc.go.kr/board/board.es?mid=&bid=0030
2 https://docs.google.com/spreadsheets/d/1t-EDGaXP5ehDgvPLOyLlESONMzFTFlZ-y5vrOs9nvVk/edit#gid=162562862
3 https://twitter.com/fergal_whatever/status/1278979040213262336
4 https://docs.google.com/spreadsheets/d/1t-EDGaXP5ehDgvPLOyLlESONMzFTFlZ-y5vrOs9nvVk/edit#gid=38005457
5 https://stopcovid19.metro.tokyo.lg.jp/en/
6 https://www.japantimes.co.jp/news/2020/07/03/national/japan-infection-codes-coronavirus-app/#.XwAPWnUzZhE
7 https://www.washingtonpost.com/world/asia_pacific/japan-coronavirus-wakayama/2020/03/22/02da83bc-65f5-11ea-8a8e-5c5336b32760_story.html
8 https://www.mhlw.go.jp/stf/covid-19/kokunainohasseijoukyou.html
9 https://www.japantimes.co.jp/news/2020/06/15/national/tokyo-47-covid-19-cases/
10 https://asiatimes.com/2020/07/tokyo-host-bars-take-the-heat-for-covid-revival/