Planet Rugby Forum

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Just out of a staff update meeting; & management were asked if the company was investigating producing ventilators; (we're a medical device company). Response was that there was consideration; but that it wasn't practical with all the requirements.

Instead the company has offered Medtronic, (one of our rivals), any other support they could; materials, labs, logistics, people, even cleanroom space.

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Scientists questioning other scientists work. Who'd of thunk?

JM2K6 wrote:The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Can you link me to these please.

The ones I've seen so far were the one stating that it could be wiped out, and there wouldn't be bouncing resurgences as restrictions were lifted.

The other about the behavioural science is a tough one to blame on imperial, since you need behavioural predictions, and that's not going to be their speciality. And quite simply, we don't know if the data they were given was right or wrong, since that's only going to be known after the restrictions are in place. They predicted 75% compliance, probably as a pretty bad overall picture, worst case scenario type thing.

message #2527204 wrote:

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Scientists questioning other scientists work. Who'd of thunk?

Taleb isn't strictly speaking, a scientist

message #2527204 wrote:

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Scientists questioning other scientists work. Who'd of thunk?

That's how science is supposed to work. Good science holds up under criticism and analysis.

Yer Man wrote:.

edit: Saint had already pointed out the f*cking obvious

I think people are missing the point. They had to abandon one model a do a U-Turn. As I said I don't agree. I don't know exactly how they made these models, but the idea that they use assumptions and speculation, even the overall approach, abet of a scientific basis, means there is likely areas for flaws in these models.

Raggs wrote:

JM2K6 wrote:The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Can you link me to these please.

The ones I've seen so far were the one stating that it could be wiped out, and there wouldn't be bouncing resurgences as restrictions were lifted.

The other about the behavioural science is a tough one to blame on imperial, since you need behavioural predictions, and that's not going to be their speciality. And quite simply, we don't know if the data they were given was right or wrong, since that's only going to be known after the restrictions are in place. They predicted 75% compliance, probably as a pretty bad overall picture, worst case scenario type thing.

Pretty sure you've already seen Taleb & co's review of the imperial model, no? https://twitter.com/nntaleb/status/1239 ... 32/photo/1 - they're far from the only ones.

As for the behavioural science, I wasn't blaming imperial, I was saying the nudge unit appears to be pretty dodgy - but someone somewhere made the decision to use it.

All this stuff absolutely should be more rigorous. The fact that lives are at stake and this is a crisis means we should be more certain about the science, not less.

JM2K6 wrote:

message #2527204 wrote:

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Scientists questioning other scientists work. Who'd of thunk?

That's how science is supposed to work. Good science holds up under criticism and analysis.

Indeed. And it's not a matter of the model being 'wrong', more how accurate or inaccurate it is. Anyone could say washing hands like manics and staying away from people, especially the old will prevent disease spreading. But are the assumptions that produce the outcomes, which help guide when and where actions should be implemented so their effects will be at their optimum for example are important they they are accurate.

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

“As Taleb made clear”

Taleb isn’t a virologist, epidemiologist or Doctor of medicine.

You now quote him as being infallible because of what quite frankly is a “gut feel” you’re so guilty of what you regularly accuse others of and you can’t even see it.

JM2K6 wrote:

Raggs wrote:

JM2K6 wrote:The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Can you link me to these please.

The ones I've seen so far were the one stating that it could be wiped out, and there wouldn't be bouncing resurgences as restrictions were lifted.

The other about the behavioural science is a tough one to blame on imperial, since you need behavioural predictions, and that's not going to be their speciality. And quite simply, we don't know if the data they were given was right or wrong, since that's only going to be known after the restrictions are in place. They predicted 75% compliance, probably as a pretty bad overall picture, worst case scenario type thing.

Pretty sure you've already seen Taleb & co's review of the imperial model, no? https://twitter.com/nntaleb/status/1239 ... 32/photo/1 - they're far from the only ones.

As for the behavioural science, I wasn't blaming imperial, I was saying the nudge unit appears to be pretty dodgy - but someone somewhere made the decision to use it.

All this stuff absolutely should be more rigorous. The fact that lives are at stake and this is a crisis means we should be more certain about the science, not less.

The Taleb one is where they say that it can be wiped out with a lock down, and tracing. Criticising the model for not assuming the virus disappears is not a valid criticism for me. Criticising it for not including a process that the country isn't as yet capable of, isn't a valid criticism. As the imperial guys pointed out, they didn't include tracing since the capability quite simply wasn't there (and unless they get upto hundreds of thousands of tests a week, still isn't). Meaning being able to test door to door is even further into the lands of fiction. I'd rather the modellers use the world as it is now, rather than a hoped for one.

Taleb makes no accounting for people not obeying the lockdown, not a luxury the Imperial modellers can take, yes China can do it, the UK? Not so easy.

Localised outbreaks is a fair point. But again, even with government advice, it's tough to enforce people not to go to their second homes (and look at that, it's happened).

They've complained the model doesn't take into account real world factors, whilst completely ignoring quite a few (testing being a major one).

Lobby wrote:

Homer wrote:What will be interesting longer term is to assess whether this virus has any significant impact on total deaths. To compare England and Wales registered deaths for 1st 10 weeks of the year:

2018 129,983
2019 114,794
2020 116,017

So far no spikes noticed in the weekly total deaths, but that could quickly change. Will total deaths for the year go up? Will it be fairly level at the year end, maybe with some people dieing six months earlier than they otherwise would?

What will be also interesting is when they get a significant number of antibody test kits available. If they find a large portion of the population have already had it, they may assess transmissibility is far higher than originally thought and morbidity is far lower (is this why China has already re-opened)?

I think over 90,000 people have been tested in the UK so far with only about 10% testing positive. The idea that lots of people have already had it really isn’t credible, no matter how many times it is repeated.

The current tests show whether somebody actively has covid-19. They do not show if somebody has had it and recovered.

bimboman wrote:

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

“As Taleb made clear”

Taleb isn’t a virologist, epidemiologist or Doctor of medicine.

You now quote him as being infallible because of what quite frankly is a “gut feel” you’re so guilty of what you regularly accuse others of and you can’t even see it.

cool, I'll let the data scientists I work with know that they need to hold PhDs in the stuff that the data refers to in order to do their jobs

wait

no

turns out you don't need to be a virologist to criticise statistical models of complex systems - but Taleb is an expert on statistical models of complex systems

JM2K6 wrote:

message #2527204 wrote:

JM2K6 wrote:

Raggs wrote:

eldanielfire wrote:It's not quite like that is it? No one is criticising the strategies put in place.

The government had to change tack because even following the initial modelling of what to do, the real life data contradicted their own model and they drastically changed plans based off new models wit the new data trends. A good first model would have evolved with new data, with certain points coming a bit earlier or a bit later. That is a bit different to changing things altogether. The best thing is that the Government have been willing to have a dynamic approach to this and follow the Science at all times.

The model did evolve as new data (in that case, higher hospitalisation rate) was entered. What changed was the fact that with that new data, is the model showed that mitigation was no longer possible, as even minimising the spread as much as possible, whilst still allowing it to spread, led to the health system getting overwhelmed, only squashing it completely worked, hence the change of strategy.

Same model, different data, different strategy.

The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Scientists questioning other scientists work. Who'd of thunk?

That's how science is supposed to work. Good science holds up under criticism and analysis.

Exactly.
The worry it's that tabloids and Twitter ( and Piers bloody Morgan) will just seed panic because they don't understand that.

Raggs wrote:

JM2K6 wrote:

Raggs wrote:

JM2K6 wrote:The criticism of the model extends beyond simple "your data is out of date" stuff, as Taleb made very clear and as the criticism of the nudge unit - which also feeds into the model given it relies on some behavioural science stuff - also makes clear.

The argument isn't that the model just needs the right data, it's that the model itself is quite flawed and they're making a lot of bad assumptions and ignoring a lot of other factors and presenting this untested work as 'science'.

Can you link me to these please.

The ones I've seen so far were the one stating that it could be wiped out, and there wouldn't be bouncing resurgences as restrictions were lifted.

The other about the behavioural science is a tough one to blame on imperial, since you need behavioural predictions, and that's not going to be their speciality. And quite simply, we don't know if the data they were given was right or wrong, since that's only going to be known after the restrictions are in place. They predicted 75% compliance, probably as a pretty bad overall picture, worst case scenario type thing.

Pretty sure you've already seen Taleb & co's review of the imperial model, no? https://twitter.com/nntaleb/status/1239 ... 32/photo/1 - they're far from the only ones.

As for the behavioural science, I wasn't blaming imperial, I was saying the nudge unit appears to be pretty dodgy - but someone somewhere made the decision to use it.

All this stuff absolutely should be more rigorous. The fact that lives are at stake and this is a crisis means we should be more certain about the science, not less.

The Taleb one is where they say that it can be wiped out with a lock down, and tracing. Criticising the model for not assuming the virus disappears is not a valid criticism for me. Criticising it for not including a process that the country isn't as yet capable of, isn't a valid criticism. As the imperial guys pointed out, they didn't include tracing since the capability quite simply wasn't there (and unless they get upto hundreds of thousands of tests a week, still isn't). Meaning being able to test door to door is even further into the lands of fiction. I'd rather the modellers use the world as it is now, rather than a hoped for one.

Taleb makes no accounting for people not obeying the lockdown, not a luxury the Imperial modellers can take, yes China can do it, the UK? Not so easy.

Localised outbreaks is a fair point. But again, even with government advice, it's tough to enforce people not to go to their second homes (and look at that, it's happened).

They've complained the model doesn't take into account real world factors, whilst completely ignoring quite a few (testing being a major one).

The first model was based on years of academic work - it's all well and good saying we don't have the capability right now, but we easily could have done and that should have been a factor earlier in the process. It's a failure of preparation, and you have to wonder whether Imperial's inability to consider it led to the government's inability to consider it, leaving us caught cold and unable to repeat what SK did.

"Government advice" is part of the problem, I suppose. It shouldn't be just advice...

Anyway, this terrifies me: https://twitter.com/nntaleb/status/1242235586578190338 - the model is "thousands of lines of undocumented C". Sweet Jesus. This is exactly the sort of modeling that Bimbo fears wrt climate change.

Fuckety balls how is it 4:30pm I need to do some actual work

JM2K6 wrote:

Raggs wrote:The Taleb one is where they say that it can be wiped out with a lock down, and tracing. Criticising the model for not assuming the virus disappears is not a valid criticism for me. Criticising it for not including a process that the country isn't as yet capable of, isn't a valid criticism. As the imperial guys pointed out, they didn't include tracing since the capability quite simply wasn't there (and unless they get upto hundreds of thousands of tests a week, still isn't). Meaning being able to test door to door is even further into the lands of fiction. I'd rather the modellers use the world as it is now, rather than a hoped for one.

Taleb makes no accounting for people not obeying the lockdown, not a luxury the Imperial modellers can take, yes China can do it, the UK? Not so easy.

Localised outbreaks is a fair point. But again, even with government advice, it's tough to enforce people not to go to their second homes (and look at that, it's happened).

They've complained the model doesn't take into account real world factors, whilst completely ignoring quite a few (testing being a major one).

The first model was based on years of academic work - it's all well and good saying we don't have the capability right now, but we easily could have done and that should have been a factor earlier in the process. It's a failure of preparation, and you have to wonder whether Imperial's inability to consider it led to the government's inability to consider it, leaving us caught cold and unable to repeat what SK did.

"Government advice" is part of the problem, I suppose. It shouldn't be just advice...

Anyway, this terrifies me: https://twitter.com/nntaleb/status/1242235586578190338 - the model is "thousands of lines of undocumented C". Sweet Jesus. This is exactly the sort of modeling that Bimbo fears wrt climate change.

What difference would it make? Honestly? I'm positive they'd have had advised that testing is extremely important, but in terms of the model? Does it change the need/effect of the lockdown? No. All it does is perhaps avoid the bounces later on, but as of here and now, you could start entering all sorts of possibilities. What if it mutates to something more/less agressive? etc etc.

Tracing is not possible now. Door to Door is absolutely dreamworld right now. It's not hard to know that both those things were going to be the case for the coming months. How long for? Who knows. Antibody tests, vaccines, treatments, testing, all these things can have a huge effect, but they're all irrelevant at this point in time.

And now, you're terrified about the complexity of the model, yet you wanted them to include the possibilites of contact tracing, and door to door testing etc etc. Making the model more complex.

Taleb just finished criticising, saying the model doesn't take into account enough things (tracing, door to door, super spreaders, travel etc etc), and is also now saying it's too complex.

Sorry, you don't get to complain that something is too simple, and too complex.

Homer wrote:

Lobby wrote:

Homer wrote:What will be interesting longer term is to assess whether this virus has any significant impact on total deaths. To compare England and Wales registered deaths for 1st 10 weeks of the year:

2018 129,983
2019 114,794
2020 116,017

So far no spikes noticed in the weekly total deaths, but that could quickly change. Will total deaths for the year go up? Will it be fairly level at the year end, maybe with some people dieing six months earlier than they otherwise would?

What will be also interesting is when they get a significant number of antibody test kits available. If they find a large portion of the population have already had it, they may assess transmissibility is far higher than originally thought and morbidity is far lower (is this why China has already re-opened)?

I think over 90,000 people have been tested in the UK so far with only about 10% testing positive. The idea that lots of people have already had it really isn’t credible, no matter how many times it is repeated.

The current tests show whether somebody actively has covid-19. They do not show if somebody has had it and recovered.

But if Covid19 was much more prevalent in the populations than previously thought, you would expect more positive results than just 10%. And you would also have to wonder why it didn't result in a surge in cases requiring hospitalisation and intensive care until now.

Taleb has a pretty shotgun approach to criticism, so I'd always consider his pronouncements carefully.

He has been very consistent on one thing though - when you have less information you need to be more cautious with regard to risk
There were (and are) fundamental assumptions baked into the Imperial models (which include pieces of code written over many years) - and many of those are about the likelihood of extreme values. E.g. if the average person gives the disease to 2.5 people, how much does that vary from person to person? What is the probability that one particular person gives it to 20 or 50? In traditional models these probabilities are next to zero, because the statistical distributions they use are shaped that way - they fit the data very well - but the data doesn't usually have many examples of these extreme values. But if a more appropriate model has 'fat tails' then it gives significant chances of some extreme values. We know lots of natural events - floods, diseases, stock market fluctuations are actually better described by fat-tailed distributions. But they've tended not to be popular in the stats literature and they make the modelling more complex. In many cases that complexity doesn't bring much extra value - the model will give similar enough answers. But if the nature of the problem is exponential, then it is massively different. So a small number of super-spreaders can change the outcome dramatically one way or the other. That is a big risk that is ignored in the Imperial model.

Also another major flaw is that assumes that casualty rates are constant - when in fact they area a function of how much pressure the health system is under - as we see in Italy.

He has been very consistent on one thing though - when you have less information you need to be more cautious with regard to risk

That cautiousness was built in with them changing advice as it went along and saying that’s what they’d do as information became available.

Much of that information was human behaviour and the unknown of how the population would react and where the reactions would be strongest.

JM2K6 wrote: Anyway, this terrifies me: https://twitter.com/nntaleb/status/1242235586578190338 - the model is "thousands of lines of undocumented C". Sweet Jesus. This is exactly the sort of modeling that Bimbo fears wrt climate change.

Any function longer than 5 lines, written in a language without a rigorous type checker and a strong bias towards pure functions and immutability, and without tests, has a high likelihood that it doesn't do what the author thinks it does. Or at any rate, doesn't after it's been being edited by multiple people for more than a few weeks.

Raggs wrote:

JM2K6 wrote:

Raggs wrote:The Taleb one is where they say that it can be wiped out with a lock down, and tracing. Criticising the model for not assuming the virus disappears is not a valid criticism for me. Criticising it for not including a process that the country isn't as yet capable of, isn't a valid criticism. As the imperial guys pointed out, they didn't include tracing since the capability quite simply wasn't there (and unless they get upto hundreds of thousands of tests a week, still isn't). Meaning being able to test door to door is even further into the lands of fiction. I'd rather the modellers use the world as it is now, rather than a hoped for one.

Taleb makes no accounting for people not obeying the lockdown, not a luxury the Imperial modellers can take, yes China can do it, the UK? Not so easy.

Localised outbreaks is a fair point. But again, even with government advice, it's tough to enforce people not to go to their second homes (and look at that, it's happened).

They've complained the model doesn't take into account real world factors, whilst completely ignoring quite a few (testing being a major one).

The first model was based on years of academic work - it's all well and good saying we don't have the capability right now, but we easily could have done and that should have been a factor earlier in the process. It's a failure of preparation, and you have to wonder whether Imperial's inability to consider it led to the government's inability to consider it, leaving us caught cold and unable to repeat what SK did.

"Government advice" is part of the problem, I suppose. It shouldn't be just advice...

Anyway, this terrifies me: https://twitter.com/nntaleb/status/1242235586578190338 - the model is "thousands of lines of undocumented C". Sweet Jesus. This is exactly the sort of modeling that Bimbo fears wrt climate change.

What difference would it make? Honestly? I'm positive they'd have had advised that testing is extremely important, but in terms of the model? Does it change the need/effect of the lockdown? No. All it does is perhaps avoid the bounces later on, but as of here and now, you could start entering all sorts of possibilities. What if it mutates to something more/less agressive? etc etc.

Tracing is not possible now. Door to Door is absolutely dreamworld right now. It's not hard to know that both those things were going to be the case for the coming months. How long for? Who knows. Antibody tests, vaccines, treatments, testing, all these things can have a huge effect, but they're all irrelevant at this point in time.

And now, you're terrified about the complexity of the model, yet you wanted them to include the possibilites of contact tracing, and door to door testing etc etc. Making the model more complex.

Taleb just finished criticising, saying the model doesn't take into account enough things (tracing, door to door, super spreaders, travel etc etc), and is also now saying it's too complex.

Sorry, you don't get to complain that something is too simple, and too complex.

Sure you do. Taleb's point in that tweet is that this model is _fragile_ (my point on that is a little different - that 'thousands of lines of undocumented C' is in itself a massive, massive danger sign, holy shit - there's no way that works as intended). There's just too many moving parts.

For a bit more detail on what his critique in the earlier link meant: https://twitter.com/nntaleb/status/1239933769982840832 + he said recently "Making parameters deterministic is what was incompetent in the UK government model.": https://twitter.com/nntaleb/status/1241738274495967234

He also likes to point out that given enough variables you can make a model say anything. It's fair to both criticise the model for being a huge fragile beast in terms of size and variables, and for ignoring some pretty basic stuff at the same time. Modeling complex systems is by all accounts bloody hard, and very bloody hard to do properly.

I think you'd like to read this paper - I've not had a chance and oh god I'm still meant to be working: https://twitter.com/nntaleb/status/1242498537562259458


You've not got my point re: tracing and that, but I don't have time to get into that right now, but you have to understand that this wasn't all magicked into existence the day we heard about it.

I really must get some work done so can't get any more involved in this today but I would strongly suggest following up on this as I know you're v interested in this shit

Dickhead learns lesson:



https://metro.co.uk/2020/03/25/attentio ... -12454901/

Raggs wrote: Sorry, you don't get to complain that something is too simple, and too complex.

That's a fair point. Also is a few thousand lines of code even necessarily that complex?

eldanielfire wrote:Dickhead learns lesson:



https://metro.co.uk/2020/03/25/attentio ... -12454901/

I'd spend a long time laughing if I knew it was only him that got it and he didn't pass it on but he sounds like someone who probably super spreaded it.

DeDoc wrote:Taleb has a pretty shotgun approach to criticism, so I'd always consider his pronouncements carefully.

He has been very consistent on one thing though - when you have less information you need to be more cautious with regard to risk
There were (and are) fundamental assumptions baked into the Imperial models (which include pieces of code written over many years) - and many of those are about the likelihood of extreme values. E.g. if the average person gives the disease to 2.5 people, how much does that vary from person to person? What is the probability that one particular person gives it to 20 or 50? In traditional models these probabilities are next to zero, because the statistical distributions they use are shaped that way - they fit the data very well - but the data doesn't usually have many examples of these extreme values. But if a more appropriate model has 'fat tails' then it gives significant chances of some extreme values. We know lots of natural events - floods, diseases, stock market fluctuations are actually better described by fat-tailed distributions. But they've tended not to be popular in the stats literature and they make the modelling more complex. In many cases that complexity doesn't bring much extra value - the model will give similar enough answers. But if the nature of the problem is exponential, then it is massively different. So a small number of super-spreaders can change the outcome dramatically one way or the other. That is a big risk that is ignored in the Imperial model.

Also another major flaw is that assumes that casualty rates are constant - when in fact they area a function of how much pressure the health system is under - as we see in Italy.

A good quote he retweeted

"Statisticians are not usually very adventurous people, perhaps because they are more interested in means than extremes."

Rishi announcing self-employed measures.

"You have not been forgotten..."

Taxable grant for 3 months, 80% of last 3 years up to £2,500

DeDoc wrote:Taleb has a pretty shotgun approach to criticism, so I'd always consider his pronouncements carefully.

He has been very consistent on one thing though - when you have less information you need to be more cautious with regard to risk
There were (and are) fundamental assumptions baked into the Imperial models (which include pieces of code written over many years) - and many of those are about the likelihood of extreme values. E.g. if the average person gives the disease to 2.5 people, how much does that vary from person to person? What is the probability that one particular person gives it to 20 or 50? In traditional models these probabilities are next to zero, because the statistical distributions they use are shaped that way - they fit the data very well - but the data doesn't usually have many examples of these extreme values. But if a more appropriate model has 'fat tails' then it gives significant chances of some extreme values. We know lots of natural events - floods, diseases, stock market fluctuations are actually better described by fat-tailed distributions. But they've tended not to be popular in the stats literature and they make the modelling more complex. In many cases that complexity doesn't bring much extra value - the model will give similar enough answers. But if the nature of the problem is exponential, then it is massively different. So a small number of super-spreaders can change the outcome dramatically one way or the other. That is a big risk that is ignored in the Imperial model.

Also another major flaw is that assumes that casualty rates are constant - when in fact they area a function of how much pressure the health system is under - as we see in Italy.

Thanks for that.

Another variable that might not be taken into account, but could be informed with more empirical data, is the virulence of the virus. Does it change over time and how is that accounted in modeling?

Trading profits up to £50k. Only if majority of income is from self-employment.

Must have a tax return for 2019.

Sounds very generous. Hopefully catches out all the under-reporters.

Gonna be a slew of late submissions on the way...up to 4 weeks to submit late 2019 returns.

JM2K6 wrote:Sure you do. Taleb's point in that tweet is that this model is _fragile_ (my point on that is a little different - that 'thousands of lines of undocumented C' is in itself a massive, massive danger sign, holy shit - there's no way that works as intended). There's just too many moving parts.

For a bit more detail on what his critique in the earlier link meant: https://twitter.com/nntaleb/status/1239933769982840832 + he said recently "Making parameters deterministic is what was incompetent in the UK government model.": https://twitter.com/nntaleb/status/1241738274495967234

He also likes to point out that given enough variables you can make a model say anything. It's fair to both criticise the model for being a huge fragile beast in terms of size and variables, and for ignoring some pretty basic stuff at the same time. Modeling complex systems is by all accounts bloody hard, and very bloody hard to do properly.

I think you'd like to read this paper - I've not had a chance and oh god I'm still meant to be working: https://twitter.com/nntaleb/status/1242498537562259458


You've not got my point re: tracing and that, but I don't have time to get into that right now, but you have to understand that this wasn't all magicked into existence the day we heard about it.

I really must get some work done so can't get any more involved in this today but I would strongly suggest following up on this as I know you're v interested in this shit

Not familiar enough with this stuff to do it without more learning, and I've been fasting today so my brains definitely not upto it. I will try and dig into it tomorrow though.

The model they have is fairly old (in terms of years), so I suspect it's been vigorously bug checked etc.

eldanielfire wrote:Dickhead learns lesson:



https://metro.co.uk/2020/03/25/attentio ... -12454901/

And they say the Chinese are unhygienic.

Ooooh, but they're gonna put your tax up - probably 12% NICs

Possibly no more dividend tax benefits? IR35 to go in harder?

RodneyRegis wrote:Taxable grant for 3 months, 80% of last 3 years up to £2,500

I'll bet a few who are heavily cash based regret cooking the books now

Raggs wrote:

JM2K6 wrote:Sure you do. Taleb's point in that tweet is that this model is _fragile_ (my point on that is a little different - that 'thousands of lines of undocumented C' is in itself a massive, massive danger sign, holy shit - there's no way that works as intended). There's just too many moving parts.

For a bit more detail on what his critique in the earlier link meant: https://twitter.com/nntaleb/status/1239933769982840832 + he said recently "Making parameters deterministic is what was incompetent in the UK government model.": https://twitter.com/nntaleb/status/1241738274495967234

He also likes to point out that given enough variables you can make a model say anything. It's fair to both criticise the model for being a huge fragile beast in terms of size and variables, and for ignoring some pretty basic stuff at the same time. Modeling complex systems is by all accounts bloody hard, and very bloody hard to do properly.

I think you'd like to read this paper - I've not had a chance and oh god I'm still meant to be working: https://twitter.com/nntaleb/status/1242498537562259458


You've not got my point re: tracing and that, but I don't have time to get into that right now, but you have to understand that this wasn't all magicked into existence the day we heard about it.

I really must get some work done so can't get any more involved in this today but I would strongly suggest following up on this as I know you're v interested in this shit

Not familiar enough with this stuff to do it without more learning, and I've been fasting today so my brains definitely not upto it. I will try and dig into it tomorrow though.

The model they have is fairly old (in terms of years), so I suspect it's been vigorously bug checked etc.

In my experience the older the codebase, the less likely it is to have been bug checked and the fewer people there are who can do that work. It's very, very, very (very) difficult to code review decade-old C [or any language, really, but C is one of the worst] with no documentation, especially as it seems to have been written by Ferguson himself, who is many things but a professional C programmer is unlikely to be one of them. There'll be a million silent bugs in there: without a thorough code review, it'll be near impossible to tell when it's not working properly because incorrect outputs just look like outputs and you've no way of knowing the difference.

Italy refusing to decline. Still in the 600s for deaths and an extra 1,000 cases over yesterday's total.

I suppose they are very much on a decline from an exponential perspective so that's something, I guess.

Less than a thousand off China's 'total' now

just on yesterday's lower UK death rates:

@KateEMcCann

For those asking/concerned about PHE figures on covid-19 deaths and why the number yesterday was low/why the way it is reported has changed, here is some more information:

NHS has been providing figures to PHE on covid deaths by 9am every day and PHE then publishing stats at 2pm. That was the intention. But as deaths rise it has become difficult to collate and verify the numbers in time.

That has led to figures being published after the 2pm deadline. To get on top of this, PHE and NHS have decided to change the time these numbers are reported to allow more time to check they're all correct.

Importantly, the way the deaths are collated has not changed, only the time. [I am still waiting to hear what the new cut off time is, and the new publishing time].

As a result, yesterday's death stats looked much lower than expected but unfortunately that was a crossover day when the times were being altered. So today the stats are likely to look far higher.

To give a more accurate picture (although I appreciate this is not ideal, stats experts) the suggestion is we add up the figures from yesterday and today and then divide by two to better understand what is happening.

There is also likely to be a note published later from officials to explain this formally, so that people can rest assured there is nothing funny going on with the numbers here.

(fingers crossed that won't be the case in bold)

Winnie wrote:

RodneyRegis wrote:Taxable grant for 3 months, 80% of last 3 years up to £2,500

I'll bet a few who are heavily cash based regret cooking the books now

Fudge ‘em

Seneca of the Night wrote:Sherelle provocative as ever:

(I'll post all of this as it's behind paywall but will spoiler it - there's a helluva lot going on in this)

Spoiler: show

[quote]The PM was panicked into abandoning a sensible Covid-19 strategy, and has plunged society into crisis
SHERELLE JACOBS
DAILY TELEGRAPH COLUMNIST
Follow 26 MARCH 2020 • 7:00AM

Millions will suffer because world leaders have chosen to follow the herd rather than back herd immunity

Boris Johnson risks being flattened by an obsolete ideology crumbling under the weight of its own contradictions

Beyond the slam of lockdown, does one detect the gentle quivering of a Prime Minister who has lost his nerve? What irony that Boris Johnson’s opponents have failed to pick up on this weakness. In their desperation to whip up hysteria against No 10’s “insufficient” coronavirus response, the liberal media has missed what could prove the century’s biggest scoop.

Namely that, faced with the protestations of the London bubble, the PM has jettisoned the only sensible strategy for dealing with the biggest global crisis since the Second World War. To put lockdown in the most cynical terms, the Government has decided to trash the economy rather than expose itself to political criticism. Unless Mr Johnson U-turns, the fallout could be cataclysmic.

Just a couple of weeks ago, the signs that our new Government would not resort to the same clunky damage control as other countries were reasonable. While Italy shooed people into their homes to stem all infections, the UK’s approach seemed more nuanced – getting the most vulnerable to self-isolate, while allowing lower risk people to get infected on a scale that wouldn’t overwhelm the NHS.

Through this “herd immunity” strategy, a resurgence of the virus after it had seemingly peaked would be avoided. Championed by Dominic Cummings, the approach was creepy, clinical and completely correct.

For a flicker, the Government seemed willing to withstand the paroxysms of its opponents and the shivers of its sympathists to take this long-termist course of action. Yes, it gambled on strong assumptions. But with leadership and clever use of numerical probability scale methods – which incidentally helped a clutch of obscure US superforecasters to actually predict Covid-19 – they may have pulled it off.

Instead, No 10 blinked, ditching herd immunity for an Imperial College research paper, which warned that hundreds of thousands could die without immediate, draconian action. It preposterously argued that lockdown may have to continue for as long as 18 months, until a vaccine is found. This despite the fact there is no scientific consensus (a rival paper claims a few weeks of lockdown may be sufficient).

Its recommendations also entail just as many risks and assumptions as the herd immunity strategy. In its assessment that 500,000 could die if the Government did nothing, the paper did not adequately address the question of how many of these victims would die anyway within a short period of something else.

Its modelling may also have underestimated the NHS’s ability to improve its intensive care capacity (the UK has just okayed medical ventilators that could equip the health service with 30,000 machines). Nor does it factor in the non-coronavirus deaths resulting from lockdown, like suicides.

So why has the PM traded in one controversial strategy for another that is, at the very least, equally vulnerable to deep criticism? Because the same old managerial elite dysfunction that got the world into this mess lingers beneath the surface of virtually all governments, like an undiagnosed cancer; this makes it impossible for them to defeat a simple virus, much like a Covid-19 victim with an “underlying illness”.

Thus “doing the right thing at the right time” has proved no match for wails about the need to be seen to be “doing whatever it takes”. And thus Mr Johnson, and other leaders, have ignored the unquantifiable damage of their actions (from the sinking of the world economy to the sacrifice of the global middle class) in order to meet spurious quantifiable targets.

There is, as usual with the Boris-Cummings duumvirate, a twist, though this time it’s of limited comfort. Rumours are aswirl that they are orchestrating herd immunity by stealth. The story goes that everything from low enforcement of lockdown to the dispersal of asymptomatic school children into family homes, is part of the plan. Critics will call this saving face. But, if true, it hits on the curious blind spot of a so-called populist: Mr Johnson’s insecure reluctance to square with the public.

He should pay heed to Trump, who is raring to get America up and running by Easter lest the cure be worst than the disease. Premature perhaps, but at least he is forcing Americans to frankly debate the trade-offs: millions of livelihoods versus thousands of lives.

One can’t help but wonder whether coronavirus is the West’s Berlin Wall moment. Liberal managerialism is collapsing under the weight of its own contradictions much in the same way communism did 30 years ago.

In puffing about climate change while ignoring threats like bio-engineered pandemics and nuclear war, UN junketeers, EU sycophants and Westminster charlatans and all the other globalist risk managers have shown themselves to be incapable of prioritising risks.

In blowing up the world economy, they have also shown themselves to be incapable of managing risks without exposing the planet to even greater dangers.

Most chilling of all perhaps, as this pandemic demonstrates, when managerial elites fail, they fall back on soft totalitarianism and the surveillance state to crawl their countries out of the messes they themselves have, through their sheer incompetence, created.

In the long term, total systems change in Britain now looks more inevitable than ever; we may look back on coronavirus as even more of a catalyst than Brexit in time. But for now, Mr Johnson’s short-term choice in coming weeks is clear: back herd immunity or be prepared to fall with the infirm herd of global elites, who will not survive this disgraceful fiasco.

https://www.telegraph.co.uk/politics/20 ... collapses/[/quote]
Johnson's Govt. wouldn't have survived deliberate inaction while all neighbouring countries shut down with the UK death count rising.