I don't think it would come as a huge surprise if the UK is ringing around every country who makes ventilators asking if they have spare units. It was one of the strands of the multi-faceted approach.
But no way, Boris has been ringing up breathlessly begging for them between coughing fits.
Can you simply buy and use a US ventilator? Does it need approval, and is it technically compatible with UK medical equipment?
A US ventilator would have been approved by one of their regulatory bodies e.g. FDA or OSHA so no problem from a regulatory standpoint, the unit would need a simple electrical adaptor due to different current.
It confused the hell out of me for a while, because I'm used to seeing the tables the other way up.
No matter. This is what we non-life insurance actuaries call a claims numbers triangle. This is easy stuff now.
Based on the data provided, and using a simple "basic chain ladder" approach, I have projected the following "ultimate" deaths for the most recent days:
1 Apr: 495 (assuming no further development; we can test this against tomorrow's data) 2 Apr: 495 3 Apr: 499 4 Apr: 602 5 Apr: 640 6 Apr: 507 (but considerable uncertainty around this, could well be a lot higher. Typically, I'd overwrite or supplement this with some kind of prior estimate, but I don't really have one)
Based on that and assuming no weird effects (eg by day of week), I'd say we haven't hit peak yet, but we might be flattening out at the top. Need at least a few more days of reporting to give some real credibility to this.
But assuming they continue to publish the data in the same format from now on, I can keep updating the model. The reporting patterns look pretty stable, so (if this continues) every further day will add more and more credibility to the projections.
It confused the hell out of me for a while, because I'm used to seeing the tables the other way up.
No matter. This is what we non-life insurance actuaries call a claims numbers triangle. This is easy stuff now.
Based on the data provided, and using a simple "basic chain ladder" approach, I have projected the following "ultimate" deaths for the most recent days:
1 Apr: 495 (assuming no further development; we can test this against tomorrow's data) 2 Apr: 495 3 Apr: 499 4 Apr: 602 5 Apr: 640 6 Apr: 507 (but considerable uncertainty around this, could well be a lot higher. Typically, I'd overwrite or supplement this with some kind of prior estimate, but I don't really have one)
Based on that and assuming no weird effects (eg by day of week), I'd say we haven't hit peak yet, but we might be flattening out at the top. Need at least a few more days of reporting to give some real credibility to this.
But assuming they continue to publish the data in the same format from now on, I can keep updating the model. The reporting patterns look pretty stable, so (if this continues) every further day will add more and more credibility to the projections.
Lots of lovely spreadsheets there with various tabs to explore, but can't see the really useful one in the tweet! Can anyone help?
Have the horrible feeling it's just been compiled by the tweeter from a whole bunch of the spreadsheets but that's not nearly so convenient ...
Those numbers look far less noisy than the daily announcements. Oddly a lot of the professional infectious disease modellers are just using those daily announced totals (or worse, using versions of the numbers John Hopkins Uni and others have web scraped, often from unofficial sources like Worldometer, with the result that some days have been missed or only partial totals have been recorded). I suspect they'd gain something by delving deeper into these numbers, since they are readily available. I haven't done it but I reckon if you fit a Poisson/negative binomial time series regression to them, you'd get a much lower overdispersion than you do from the daily announced total, and probably less messy serial correlation patterns / weekend effects.
It confused the hell out of me for a while, because I'm used to seeing the tables the other way up.
No matter. This is what we non-life insurance actuaries call a claims numbers triangle. This is easy stuff now.
Based on the data provided, and using a simple "basic chain ladder" approach, I have projected the following "ultimate" deaths for the most recent days:
1 Apr: 495 (assuming no further development; we can test this against tomorrow's data) 2 Apr: 495 3 Apr: 499 4 Apr: 602 5 Apr: 640 6 Apr: 507 (but considerable uncertainty around this, could well be a lot higher. Typically, I'd overwrite or supplement this with some kind of prior estimate, but I don't really have one)
Based on that and assuming no weird effects (eg by day of week), I'd say we haven't hit peak yet, but we might be flattening out at the top. Need at least a few more days of reporting to give some real credibility to this.
But assuming they continue to publish the data in the same format from now on, I can keep updating the model. The reporting patterns look pretty stable, so (if this continues) every further day will add more and more credibility to the projections.
Lots of lovely spreadsheets there with various tabs to explore, but can't see the really useful one in the tweet! Can anyone help?
Have the horrible feeling it's just been compiled by the tweeter from a whole bunch of the spreadsheets but that's not nearly so convenient ...
Those numbers look far less noisy than the daily announcements. Oddly a lot of the professional infectious disease modellers are just using those daily announced totals (or worse, using versions of the numbers John Hopkins Uni and others have web scraped, often from unofficial sources like Worldometer, with the result that some days have been missed or only partial totals have been recorded). I suspect they'd gain something by delving deeper into these numbers, since they are readily available. I haven't done it but I reckon if you fit a Poisson/negative binomial time series regression to them, you'd get a much lower overdispersion than you do from the daily announced total, and probably less messy serial correlation patterns / weekend effects.
Yes, the table in the tweet isn't actually there. It's the totals from each day, so not hard to rebuild. Took me about 5 minutes plus a bit more to understand the layout.
I've long ago concluded the data on worldometer was worse than useless. This is the first dataset I've seen on the crisis that actually makes sense to me. If I have time next week I might start trying to calculate confidence intervals around my projections, possibly via bootstrapping residuals. Or Mack's method. I think I have a version somewhere that isn't my employer's IP.
The thing about this bug is this: it produces surreal dystopian images - of forsaken corpses, temporary morgues and mass graves, which everyone - because of normalcy bias - dismisses as fake, or Russian propaganda, or as wrongly interpreted.
Then it turns out they were real videos. This is what a plague looks like.
My favourite pb example is that video of filled body bags apparently left in the street in Wuhan. I posted this on here in early feb, I believe. At the time it was roundly dismissed. Charles, for example, suggested it was a video showing exhausted workers who had come home and gone to sleep in body bags.
It was no such thing. It was uncollected corpses.
That wasn’t me. The only thing I ever questioned was your predictions of 1.0-2.0m dead. I didn’t expect this lock down. But I did think that coronavirus would be episodic and endemic.
As soon as someone calls it The Plaque I stop reading their post.
Jesus, just listening to an ER doctor in NY. It appears they haven't even sectioned off ER admissions into suspected CV and others (as is standard here), so everybody is sitting there all together. Go with a broken leg and you have to sit there next to somebody coughing their lungs up.
Their healthcare system is atrocious unless you're a millionaire.
They confuse me. I thought they would maintain their values and ideas are already supported by the mainstream member and public, but they talk of the mistake of retreating from the challenge of making themselves mainstream.
How fascinating too that fighting between themselves is seen as a risk of centrists taking over the party - apparently that's the key fight, and fighting against the centrists and letting the Tories take over is no big deal. Focus on the real enemy I guess.
They confuse me. I thought they would maintain their values and ideas are already supported by the mainstream member and public, but they talk of the mistake of retreating from the challenge of making themselves mainstream.
How fascinating too that fighting between themselves is seen as a risk of centrists taking over the party - apparently that's the key fight, and fighting against the centrists and letting the Tories take over is no big deal. Focus on the real enemy I guess.
Isn't this just the old socialist nutter entryist philosophy in action?
The success of socialism is more important than the success of the Labour Party?
(Personally I would not mourn if they sunk together without trace like Tarka and Deadlock.).
It confused the hell out of me for a while, because I'm used to seeing the tables the other way up.
No matter. This is what we non-life insurance actuaries call a claims numbers triangle. This is easy stuff now.
Based on the data provided, and using a simple "basic chain ladder" approach, I have projected the following "ultimate" deaths for the most recent days:
1 Apr: 495 (assuming no further development; we can test this against tomorrow's data) 2 Apr: 495 3 Apr: 499 4 Apr: 602 5 Apr: 640 6 Apr: 507 (but considerable uncertainty around this, could well be a lot higher. Typically, I'd overwrite or supplement this with some kind of prior estimate, but I don't really have one)
Based on that and assuming no weird effects (eg by day of week), I'd say we haven't hit peak yet, but we might be flattening out at the top. Need at least a few more days of reporting to give some real credibility to this.
But assuming they continue to publish the data in the same format from now on, I can keep updating the model. The reporting patterns look pretty stable, so (if this continues) every further day will add more and more credibility to the projections.
Lots of lovely spreadsheets there with various tabs to explore, but can't see the really useful one in the tweet! Can anyone help?
Have the horrible feeling it's just been compiled by the tweeter from a whole bunch of the spreadsheets but that's not nearly so convenient ...
Those numbers look far less noisy than the daily announcements. Oddly a lot of the professional infectious disease modellers are just using those daily announced totals (or worse, using versions of the numbers John Hopkins Uni and others have web scraped, often from unofficial sources like Worldometer, with the result that some days have been missed or only partial totals have been recorded). I suspect they'd gain something by delving deeper into these numbers, since they are readily available. I haven't done it but I reckon if you fit a Poisson/negative binomial time series regression to them, you'd get a much lower overdispersion than you do from the daily announced total, and probably less messy serial correlation patterns / weekend effects.
Yes, the table in the tweet isn't actually there. It's the totals from each day, so not hard to rebuild. Took me about 5 minutes plus a bit more to understand the layout.
I've long ago concluded the data on worldometer was worse than useless. This is the first dataset I've seen on the crisis that actually makes sense to me. If I have time next week I might start trying to calculate confidence intervals around my projections, possibly via bootstrapping residuals. Or Mack's method. I think I have a version somewhere that isn't my employer's IP.
(This is more a note-to-self, and perhaps the handful of other R users on here, than a note-to-you!)
I can see rebuilding the table isn't hard (though it gets worse as the days roll on) but it is slightly irritating not to just be able to go to the NHS stats site and view it!
OT I've run out of Shoestring episodes. Just before HMG closed the shops I stockpiled some box sets including Life On Mars, which the BBC has just put back on iplayer to annoy me.
I have On Demand sets of: Morse, Endeavour, Lewis, Cranford, Yes Minister, Lark Rise to Candleford, BBT, and many more plus about a thousand films - should keep me going till the W/E
I have On Demand sets of: Morse, Endeavour, Lewis, Cranford, Yes Minister, Lark Rise to Candleford, BBT, and many more plus about a thousand films - should keep me going till the W/E
I've subbed to Netflix on the £6 per month option.
The thing about this bug is this: it produces surreal dystopian images - of forsaken corpses, temporary morgues and mass graves, which everyone - because of normalcy bias - dismisses as fake, or Russian propaganda, or as wrongly interpreted.
Then it turns out they were real videos. This is what a plague looks like.
My favourite pb example is that video of filled body bags apparently left in the street in Wuhan. I posted this on here in early feb, I believe. At the time it was roundly dismissed. Charles, for example, suggested it was a video showing exhausted workers who had come home and gone to sleep in body bags.
It was no such thing. It was uncollected corpses.
Is this the video where if you look you can see one of the 'bodies' move?
OT I've run out of Shoestring episodes. Just before HMG closed the shops I stockpiled some box sets including Life On Mars, which the BBC has just put back on iplayer to annoy me.
Great series especially Eddie Shoestring ..the relationship with his "landlady" eas particularly well played out
The thing about this bug is this: it produces surreal dystopian images - of forsaken corpses, temporary morgues and mass graves, which everyone - because of normalcy bias - dismisses as fake, or Russian propaganda, or as wrongly interpreted.
Then it turns out they were real videos. This is what a plague looks like.
My favourite pb example is that video of filled body bags apparently left in the street in Wuhan. I posted this on here in early feb, I believe. At the time it was roundly dismissed. Charles, for example, suggested it was a video showing exhausted workers who had come home and gone to sleep in body bags.
It was no such thing. It was uncollected corpses.
That wasn’t me. The only thing I ever questioned was your predictions of 1.0-2.0m dead. I didn’t expect this lock down. But I did think that coronavirus would be episodic and endemic.
I’m pretty sure it was you. Sorry. You were one of the worst denialists.
As for my predictions, yawn. I never predicted 2m dead. This is so tedious. If you can find a comment where I said that, knock yourself out. Hint: don’t bother. I didn’t say it.
What I did was extrapolate the extreme worst case scenario - by its nature very unlikely - from an infection rate of 80% and a Wuhan style CFR of 3-4%. That gives you millions of dead IN AN EXTREME WORST CASE SCENARIO
A reasonable worst case scenario was 250,000-500,000 British dead in the first year of disease, and indeed it is these figures which shifted HMG from herd immunity to suppression.
No - I just always questioned your numbers. Like @foxy I was early with zinc (in my case Sambucol) and social distancing. I also passed on some of the intelligence I was getting from Mike Pence’s office on travel etc.
You regularly used the reasonable worst case as the base case, which is what I criticised. (The 2m number was in one post about a week or so ago so a little unfair of me to highlight it)
He’s trying to rewrite history again. There are plenty of posts where he makes the casual assumption that 1% of the population (which would be over 650,000) will die, and the debate was around his own calculations based on higher death rates of up to 4%.
For Eadric now to describe 250-500,000 as his reasonable worst case scenario is simply dishonest, there is no other word for it.
The post he has clearly forgotten is the one where he promised to welcome the ordure he would deserve if his even worse predictions didn’t come to pass.
Comments
A US ventilator would have been approved by one of their regulatory bodies e.g. FDA or OSHA so no problem from a regulatory standpoint, the unit would need a simple electrical adaptor due to different current.
No matter. This is what we non-life insurance actuaries call a claims numbers triangle. This is easy stuff now.
Based on the data provided, and using a simple "basic chain ladder" approach, I have projected the following "ultimate" deaths for the most recent days:
1 Apr: 495 (assuming no further development; we can test this against tomorrow's data)
2 Apr: 495
3 Apr: 499
4 Apr: 602
5 Apr: 640
6 Apr: 507 (but considerable uncertainty around this, could well be a lot higher. Typically, I'd overwrite or supplement this with some kind of prior estimate, but I don't really have one)
Based on that and assuming no weird effects (eg by day of week), I'd say we haven't hit peak yet, but we might be flattening out at the top. Need at least a few more days of reporting to give some real credibility to this.
But assuming they continue to publish the data in the same format from now on, I can keep updating the model. The reporting patterns look pretty stable, so (if this continues) every further day will add more and more credibility to the projections.
Here's the link ...
Lots of lovely spreadsheets there with various tabs to explore, but can't see the really useful one in the tweet! Can anyone help?
Have the horrible feeling it's just been compiled by the tweeter from a whole bunch of the spreadsheets but that's not nearly so convenient ...
Those numbers look far less noisy than the daily announcements. Oddly a lot of the professional infectious disease modellers are just using those daily announced totals (or worse, using versions of the numbers John Hopkins Uni and others have web scraped, often from unofficial sources like Worldometer, with the result that some days have been missed or only partial totals have been recorded). I suspect they'd gain something by delving deeper into these numbers, since they are readily available. I haven't done it but I reckon if you fit a Poisson/negative binomial time series regression to them, you'd get a much lower overdispersion than you do from the daily announced total, and probably less messy serial correlation patterns / weekend effects.
I've long ago concluded the data on worldometer was worse than useless. This is the first dataset I've seen on the crisis that actually makes sense to me. If I have time next week I might start trying to calculate confidence intervals around my projections, possibly via bootstrapping residuals. Or Mack's method. I think I have a version somewhere that isn't my employer's IP.
How fascinating too that fighting between themselves is seen as a risk of centrists taking over the party - apparently that's the key fight, and fighting against the centrists and letting the Tories take over is no big deal. Focus on the real enemy I guess.
Stirring stuff, we'll get through this.
The success of socialism is more important than the success of the Labour Party?
(Personally I would not mourn if they sunk together without trace like Tarka and Deadlock.).
(This is more a note-to-self, and perhaps the handful of other R users on here, than a note-to-you!)
I can see rebuilding the table isn't hard (though it gets worse as the days roll on) but it is slightly irritating not to just be able to go to the NHS stats site and view it!
This thread has gone moistly into the dawn....
For Eadric now to describe 250-500,000 as his reasonable worst case scenario is simply dishonest, there is no other word for it.
The post he has clearly forgotten is the one where he promised to welcome the ordure he would deserve if his even worse predictions didn’t come to pass.