The R in Ireland

I was playing about with different ways of presenting the Covid-19 data I’ve been collecting here to make the trends clearer. This is what the daily confirmed cases and reported deaths look like if smoothed with a simple 7-day moving average and plotted on a log-linear scale:

This confirms something I’ve suspected over the last couple of weeks: that the number of confirmed cases has been edging upwards. This is not so clear in the raw counts, but is suggested: the smoothing makes this easier to see by reducing the noise and removing any weekend reporting artefacts:

This recent upward trend is consistent with the latest estimates of the basic reproduction number R that suggest it has crept up to around unity.

The number of cases per day remains low and confined to particular clusters. Hopefully contact tracing and isolation will prevent the increase getting out of hand.

It seems about two thirds (15 out of 23) of the new cases are associated with travel, though, so any loosening of restrictions on overseas travel would be very unwise.

The maximum age of any of the new cases reported yesterday is 44 and 77% are under 25. Perhaps its younger people who are less likely to observe social distancing.

I worry a bit that Ireland may be unlocking too quickly and people may be getting a bit complacent about the situation.

This is not over.

36 Responses to “The R in Ireland”

  1. Anton Garrett Says:

    No indeed. Only two strategies are worth anything, I reckon: herd immunity or total lockdown for 3 weeks, prior to which the government issues everybody with 3 weeks of canned food and institutes all-hours curfew by martial law, allowing only farmers, electricity, gas, water and health workers and one or two others to move, meanwhile closing borders and airports to all people and allowing in only goods. Then reopening borders only to other ‘clean’ countries. Any strategy in between these two is worse than useless as it fails to prevent the virus while extending economic damage. But neither is likely to be viable in a democracy; not that I wish to live anywhere else. Which of the two to choose depends on the virulence of the virus in question, and SARS-CoV-2 seems to be on every borderline as far as decisions go.

    This virus is likely to get less when the weather warms again. But what of the autumn? And it may or may not go away for no identified reason, as SARS-CoV-1 did. Also we may or may not find a vaccine using smarter strategies than were available two decades ago. Having had it, I’m particularly keen to learn (the easy way) what the immunity period is.

    • telescoper Says:

      The UK has over 40,000 dead with only a few percent of the population infected. `Herd immunity’ would involve an even more catastrophic loss of life.

      • Anton Garrett Says:

        These figures need to be set in the context of the normal number of deaths per month and – more importantly – the expected loss of life in years/months across a cohort grouped by age and prior health, after which a mean can be taken. If herd immunity is the strategy (and I did not advocate it, just said that it was one of only two decent strategies) then I agree that it ought to be slowed to the point that hospitals do not overflow.

        The Victorians used to talk freely about death, whereas sex was taboo. Today it is the other way round. But we all have to die.

      • telescoper Says:

        The figures also need to be seen in the context of the fact excess mortality in the UK over the last four months is about 70,000 (far in excess of the official figures).

      • Anton Garrett Says:

        I’d be surprised if the UK government is actually diddling stats of how many people die each month. The iffy stuff comes in about what they die of, which is a difficult enough debate even before political interests obtrude (as they surely do). And as you rightly pointed out here, some of the increase is because the NHS has reduced doing other things.

      • telescoper Says:

        I don’t think they’re necessarily fiddling. I just think the reporting is a mess.

    • Phillip Helbig Says:

      “Only two strategies are worth anything, I reckon: herd immunity or total lockdown for 3 weeks, prior to which the government issues everybody with 3 weeks of canned food and institutes all-hours curfew by martial law, allowing only farmers, electricity, gas, water and health workers and one or two others to move, meanwhile closing borders and airports to all people and allowing in only goods.”

      What about restricting meetings between people who don’t live together, 2 m distance in public, masks, restricting the number of people in buildings, and so on? Countries which have adopted this strategy have done reasonably well, without the deaths from herd-immunity-through-infection nor a complete stillstand of the economy.

    • Anton Garrett Says:

      Articles in the Guardian and Telegraph today suggest that the immunity period is a few months. Not great news.

  2. Dave Carter Says:

    Herd immunity is something your get with a vaccine. Suggesting that we go for herd immunity by infection, is in my view totally immoral, as we are pretty sure from Ferguson’s work that it would involve around half a million deaths. Its fashionable in certain far right circles to try to discredit Ferguson, but I see nothing in his record or his methodology to lead be to disbelieve his numbers.

    Back to the original post, Peter do you think that the increase in the detected infection rate is due to an increase in the real infection rate, or to an increase in testing amongst age groups who weren’t being tested before?

    • Anton Garrett Says:

      I have no confidence in Ferguson’s work and not much in anybody else’s.

    • telescoper Says:

      The testing strategy here is very focused and informed by contact tracing. What we see is a small number of clusters (largely initiated by people who have been abroad) where most of the members of a cluster are being identified and tested. That often involves younger people, hence the higher rates among younger (often asymptomatic) people. So I think correct answer is a bit of both, though you have to remember that the actual number of new cases per day is around 15-20 on average, which is still quite low.

    • Anton Garrett Says:

      Re the morality of strategies, recessions lead to increased rates of death among the poor, and this needs to be taken into account too.

  3. Anton Garrett Says:

    The latest idea, for those suffering dangerous lung complications due to SARS-CoV-2 (which is how it kills those who die directly of it), is to inhale appropriate steroid drugs, notably budesonide:

    The Far Eastern countries where the death rate is conspicuously low are doing this, apparently, and doing it as a routine treatment – not waiting until people get severely ill.

    All of the google hits for budesonide + covid are of very recent date, so this appears to be recent information.

    • So you don’t accept published models of epidemics, but you do promote as yet untested ‘treatments’? It might be better to apply scientific methodologies for testing models and ideas, rather than using gut feelings.

      • Anton Garrett Says:

        I put up that clip because I was impressed by the science, not by any gut feeling. Perhaps it would be more constructive if you spoke about the science rather than about me?

        I am floating steroid inhalation as a layman, and I described it above as an “idea” rather than something proven. In a situation in which the medical profession are learning in real time just as much as the general public I make no apology for that. (The difference is that experts have a far greater context into which to fit – and by which to judge – new information.) Laymen can also move more quickly than governments and experts who dare not say anything wrong (although they manage it anyway).

        If you can make a scientific critique of the YouTube clip above then I’d welcome it.

        Did not Ferguson’s software, even after polishing by Microsoft (?!), give different answers on different machines? In any case I do not think any epidemiological models take into account variations in contagiousness due to climate, ie sunlight and air temperature. The lipid coating of SARS-CoV-2 melts at not much above body temperature, so the R-number is liable to vary with both time (the season) and place (latitude). That is just one variable that is important but not taken into account.

      • Steroid drugs are currently not recommended for SARS by either the WHO or other medical organisations, to the best of my knowledge. They were tried for SARS-1 in China and Hong Kong and discontinued when the side effects became clear. There was no evidence of improvement in mortality rates, and a significant fraction of recovered patients (reports mention 1/3-1/2) suffered from avascular necrosis. To put such a contentious drug back into the picture requires extensive testing with control samples. But it is hard to see people volunteering for this, seeing the published research from SARS-1.

        I am not aware of any studies that invalidate the Ferguson models. It has assumptions and limitations, but to say you have ‘no confidence’ is not based on evidence that I am aware of.

        I apologize that my comment came across as a personal attack. That was not intended.

      • Anton Garrett Says:

        Thanks. We’re all in this together.

  4. Dave Carter Says:

    Anton, please just point to one peer reviewed paper which supports your critiques on Ferguson’s software? The only criticisms I have seen seem to be based upon the fact that it is written in Fortran not Python. But I haven’t seen one peer reviewed or even academic source which disputes his results.

    • telescoper Says:

      This is the only thing I’ve read on the matter… https://www.nature.com/articles/d41586-020-01685-y

    • Anton Garrett Says:

      Come off it, stuff saying only that a model is unreliable isn’t even going to get submitted to a peer reviewed journal. I can give you stuff that makes cogent sense but isn’t peer reviewed if you wish; please say. We all know that nothing wrong has ever passed peer review…

    • Dave Carter Says:

      Right so no evidence that Ferguson’s models are not robust, and plenty of evidence that they are. I think this case is closed, and you should know better than to promulgate scurrilous rumours in the sewer press such there is any kind of problem here.

      • Anton Garrett Says:

        O there’s plenty of evidence if you care to look but you haven’t asked me for it (I’ll give it if you do) and I don’t appreciate being categorised with Nazis and other racists on the far right because I use scientific criteria to criticise a scientific model.

  5. Dave Carter Says:

    Right, so you can cite a peer reviewed paper then, or at the very least a paper submitted to a peer reviewed journal. Otherwise you have no evidence. The article in Peter’s link is not as far as I see peer reviewed, but it is at least in Nature, a respectable scientific journal.

    • Anton Garrett Says:

      No evidence that you’d accept, but that is not the same. If you are willing to stoop to ask me, I can and will provide. Not that this blog is peer-reviewed…

      • Dave Carter Says:

        Ok, go ahead then, I would be interested to know what kind of sources you have.

      • Anton Garrett Says:

        Will do. I’ve already given one critique – of *all* models of this epidemic, consequently including Ferguson’s – about the dubiousness of a universal value of R independent of climate, and of exporting an R-value estimated from data at one time and place to another. That is because of the fragility to UV and high temperature of SARS-CoV-2’s lipid coating. I had that idea myself, although I expect that others better qualified will have said so in the public domain. I’ve no idea whether it’s been stated in anything peer-reviewed, but peer review moves more slowly than blogs and this is a fast-moving situation, and in any case if an argument is poorer for the absence of peer review then it should be easier to knock down.

        I’ll document the variability of Ferguson’s model across machines and also cite other critiques within 24 hours.

      • Dave Carter Says:

        Anton, you haven’t yet posted one link in this thread, only a YouTube video about an untested drug. There is nothing, peer-reviewed or otherwise.

      • Anton Garrett Says:

        Perhaps you were drafting your comment as I uploaded mine, but (as I stated) I shall post something not later than 1.19pm blog time tomorrow (ie, Tuesday)

      • Anton Garrett Says:

        Dave, FYI I uploaded it last night but it contains multiple links and requires Peter to clear it.

  6. Anton Garrett Says:

    I said that I had no confidence in Ferguson’s computer model and not much in any computer model. I’ll start with some generic criticisms – which will obviously take in Ferguson’s model – and then move on to matters specific to Ferguson’s. I’ve already given one critique of all models of this epidemic, about the dubiousness of a universal value of R independent of climate, and of exporting an R-value estimated from data at one time and place to another. That is because of the fragility to UV and high temperature of SARS-CoV-2’s lipid coating. Models used in the UK early on – which is when the big political decisions were taken – were not based wholly on UK data, for the virus was further on elsewhere. There are also different protocols across jurisdictions as to whether covid-19 was the cause of death, which could impair estimates of R in one place from another. There may also be genetic factors. If genetic factors are correlated with persons living in different parts of the world, the virus would be better adapted to the genotype it ‘passaged’ through many times early on; also, Vitamin D levels appear to be relevant in conferring protection, and these differ with differing diet, exposure to sunlight and skin tone. I do not know how relevant these factors are, but I doubt that a Bayesian model comparison was performed in order to determine the optimal choice of ‘floating’ parameters in the model (too few or too many obviously won’t work).

    Here is a general critique of the science of modelling epidemics:

    https://forecasters.org/blog/2020/06/14/forecasting-for-covid-19-has-failed/

    As for Ferguson’s model,

    Many have claimed that it is almost impossible to reproduce the same results from the same data, using the same code. Scientists from the University of Edinburgh reported such an issue, saying they got different results when they used different machines, and even in some cases, when they used the same machines. “There appears to be a bug in either the creation or re-use of the network file. If we attempt two completely identical runs, only varying in that the second should use the network file produced by the first, the results are quite different,” the Edinburgh researchers wrote on the Github file. After a discussion with one of the Github developers, a fix was later provided. This is said to be one of a number of bugs discovered within the system. The Github developers explained this by saying that the model is “stochastic”, and that “multiple runs with different seeds should be undertaken to see average behaviour”. However, it has prompted questions from specialists, who say “models must be capable of passing the basic scientific test of producing the same results given the same initial set of parameters…otherwise, there is simply no way of knowing whether they will be reliable.”

    https://www.telegraph.co.uk/technology/2020/05/16/coding-led-lockdown-totally-unreliable-buggy-mess-say-experts/

    There appears to be some muddying of the water here. If the model is stochastic then you want multiple runs with *identical* seeds to see average behaviour! If the code is not meant to contain a random number generator then the variability it gives for identical inputs means that it is untrustworthy; would you fly in an aircraft governed by such software? If it is meant to contain one, what is its purpose – elsewhere one reads that runs are done with slightly differing data in order to test stability, which is good practice; but in those cases you vary the data, not the model.

    Ferguson never released the code on which his advice to government was based. It was cleaned up for 6 weeks by Microsoft engineers working for GitHub before being publicly inspectable. I can well understand that Ferguson was working flat out in March and was too busy to respond to FOI requests, but failure to release the actual code that was acted on is deeply shocking. What was there to hide? See this detailed critique of what *was* released, by a software developer:

    https://lockdownsceptics.org/code-review-of-fergusons-model/

    Presumably what was used to inform the government was worse still. Here is an astronomer commenting on lousy academic software writing practices, including a mea culpa, and saying that he recognises the same sort of problem with Ferguson’s code, and arguing – as I think Peter has done – that code *must* be made public along with the results it provides:

    https://thecritic.co.uk/a-series-of-tubes/

    • Dave Carter Says:

      Ok, as you promised none meet any acceptable standards of peer review, but I will go through them one by one anyway:

      Link 1: Comes from the International Institute of Forecasters, which I have never heard of, and I have never heard of any of the people associated with it. It starts out with this:

      “Experienced modelers drew early on parallels between COVID-19 and the Spanish flu [2] that caused >50 million deaths with mean age at death being 28. We all lament the current loss of life. However, as of June 8, total fatalities are ~410,000 with median age ~80 and typically multiple comorbidities.”

      The question of whether the COVID-19 death toll will eventually exceed the 1919 H1N1 (I refuse to call it Spanish flu) death toll is far from settled. Antonio Guterres, Secretary General of the UN wrote an article published in the Observer and elsewhere where he cautioned against the “despicable meme” that the lives of the elderly were more expendable than those of the young. Here is that despicable meme, right there.

      When we get to the only bit which is specifically about the Imperial models, we get this:

      “conversely many models have assumed 25-fold reduction in deaths (e.g. from 510,000 deaths to 20,000 deaths in the Imperial College model) with adopted measures”

      No, the models forecast 25-fold reduction in deaths, they didn’t assume it. Sadly that reduction has not been so large, more like a factor 8, but still a factor of 8 well worth having. The most likely reason is that they were not aware of, and didn’t predict, the appalling situation in care homes.

      The rest of this is really just tittle-tattle from various US local newspapers are right wing politicians (Andrew Cuomo). The section about California is looking pretty silly right now.

      Link 2: Is behind the Telegraph paywall, so I can’t read it. But the people it is written by are not scientists of any kind.

      Link 3: Comes from lockdownsceptics.org, which appears to be a vehicle for Toby Young. It is written by someone who claims to be a software engineer, but won’t give their name. It contains a lot of misconceptions, including that the Edinburgh team could not reproduce the Imperial results, when according to them they did. And whoever wrote it doesn’t seem to understand the meaning of “stochastic”, it does not mean that you get different results on different computers. The author doesn’t really seem to understand that scientific programming is in some ways different from software engineering, and this is a scientific programming problem.

      Link 4: From The Critic (funded by Jeremy Hosking, vote Leave and Tory donor, edited by Michael Mosbacher, founder of the far right Standpoint magazine, and Christopher Montgomery, ex-DUP and ERG strategist). Its regular contributors include David Starkey and Toby Young (yes, him again). But back to the article. Written by one Ben Lewis, who used to be a PhD student in magnetohydrodynamics and published in MNRAS, and makes the absolutely scandalous accusation that scientists might make errors in their code on purpose, and then write a correction so they can claim two papers instead of one. Regarding the Imperial code this article is pretty full on in its criticism, throwing around descriptions like “exceptionally abysmal”, and “worst production code I have ever seen”, without any indication as to in what way these descriptions might apply. I could quite easily write a critique like this about his papers in MNRAS, but it wouldn’t mean anything unless I had actually done some analysis of his methodology and results. This link disappoints me most of all, because it is written by someone with a PhD in astrophysics, and who seems to have held a couple of postdoc positions since. So someone who is capable of a proper analysis, but hasn’t done one.

      • Anton Garrett Says:

        So a fair amount of playing the man rather than the ball, an appeal to peer review which we all know is infallible, and total silence about the fact that Ferguson has never released the code on the basis of which he advised the government, only a version that took multiple software experts 6 weeks to rejig. Why?

        I agree that there is confusion over whether the code contains a random number generator or not and that this might be the cause of differing results. If Yes, why were people ever complaining that it gave differing results? If No, so not those differing results show an instability bug in the code?

      • telescoper Says:

        The usual practice with (pseudo)random number generators is to use a standard one from a library rather than include it in the code. That’s what I do anyway. It is true that if you compile the code on a different machine the chances are that it will produce different numbers from the same seed as you may well not be exactly the same bit of code being included. In effect you may not running the same code when you compile it on a different machine. This is a known issue with pseudorandom number generators.

        Of course I don’t know that this is the case for the Ferguson code, which is yet another reason to maintain my view that it should be released publicly. I think that should be a general principle for scientific programming, not just in this case.

        One of the main reasons scientists are reluctant to do this is that though their code may well work, it may be very messy and inefficiently written. Another benefit of having to make codes public would be that it would make coders take more care over the design of their software.

      • Dave Carter Says:

        When I last did coding I am sure I used a NAG one. Now that dates me….

      • telescoper Says:

        I have used NAG routines in the past, but not really since I switched to Python.

        I think the Ferguson code is written in Fortran rather than Python, but the Python pseudorandom number generations definitely produce different sequences on different installations. These Black Box routines are very useful but if anything goes wrong it’s very hard to figure out what.

      • But the bottom line is whether it invalidates the code. I do not see anything here that points at that.

Leave a Reply

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

WordPress.com Logo

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

Google photo

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

Twitter picture

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

Facebook photo

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

Connecting to %s

%d bloggers like this: