Scrutable

Bird on a Fire wrote: ↑

Mon Mar 01, 2021 11:11 pm

Ah, I think he's not plotting 7- vs 14- (etc-)day rolling averages, but the average of the most recent 7 vs 14 etc days - ie, averaging over the period shown by the lines.

Good spot. I think my point stands, that indicating a recent change in a 7-day average by comparing to a 14-day (or 21 or 28 or ...) average is inferior to comparing to the 7-day average from yesterday, or the day before or ...., i.e. looking at the rolling mean.

Bird on a Fire wrote: ↑

Mon Mar 01, 2021 11:11 pm

All his stuff is quite unconventional, though. He models the pandemic trajectory as a random walk with no underlying statistical model whatsoever, just fitting to the data.

His plots do seem to do a good job of fitting, but I can't recall seeing a comparison of say, last month's forecast with the actual observed figures, because the plots he posts each day are based on an updated version of the model.

As I've said, I couldn't easily find a description of how or why he made the decisions implicit in his charts - but I'm not skilled in Twitter . But charts like this... ... explicitly refer to "Modelled cases" and the "plumes" look like some kind of uncertainty interval. Any fit or forecast requires a model, whether acknowledged or not, and whether it's called "statistical" or not.

Even descriptive statistics such as means entail important assumptions. For example, the 7-day arithmetic mean of daily cases has a clear and simple interpretation, the total cases in a week (divided by 7). The geometric mean (or mean of logs) of cases is much more debatable - what is the meaning of the product of daily cases.
On the other hand, the 7 day geometric mean (or mean of logs) of daily ratios is easily interpretable, the ratio over a complete week (7th rooted). Here the arithmetic mean is more debatable - what is the meaning of the arithmetic sum of ratios?

The model details are somewhere on his site, rather than on twitter (which is not the easiest medium to navigate!) - here's his latest post https://bskiesresearch.wordpress.com/20 ... -velocity/ but I think the stuff about his models are way back in time. Might go digging later if nobody else finds them.

Try https://www.medrxiv.org/content/10.1101 ... 20065227v2

Bird on a Fire wrote: ↑

Tue Mar 02, 2021 6:06 pm

The model details are somewhere on his site, rather than on twitter (which is not the easiest medium to navigate!) - here's his latest post https://bskiesresearch.wordpress.com/20 ... -velocity/ but I think the stuff about his models are way back in time. Might go digging later if nobody else finds them.

shpalman wrote: ↑

Tue Mar 02, 2021 6:14 pm

Try https://www.medrxiv.org/content/10.1101 ... 20065227v2

Thanks to both, looks like some interesting reading.

Just been looking at the government figures, the last time that there were less than 100 deaths reported within 28 days of a covid test was in October last year. Over 4 months ago. I have become numb to that fact.

I can cope with a small number of deaths remaining as a background, there are people dying of preventable diseases and health issues all the time, but the horror of all this has been somewhat lost in the relentlessness of it.

A live blog copies a press release posted by an agency in which there's a thing.

"This is a good thing" said some guy. "I've seen it and you haven't so f.ck you".

The thing is similar to other things which we post about but don't even give you enough information to google to find it for yourself let alone link to where you can read it. Because f.ck you, we're journalists and we know what's best for you.

Some other guy says Look! there's a press release! All European countries must now change their medical policy! It's the only sensible thing to do.

“Well, here are the data.”

WHERE

WHERE ARE THE f.cking DATA

WHERE ARE THEY

WHERE

Grumble wrote: ↑

Tue Mar 02, 2021 11:38 pm

Just been looking at the government figures, the last time that there were less than 100 deaths reported within 28 days of a covid test was in October last year. Over 4 months ago. I have become numb to that fact.

I can cope with a small number of deaths remaining as a background, there are people dying of preventable diseases and health issues all the time, but the horror of all this has been somewhat lost in the relentlessness of it.

Same. It seems like there was generally a lot less coverage of the number of people dying a day at the peak of this wave compared to the first too.

I feel the same about case numbers. Numbers that I'm sure used to feel big now feel small compared to 80,000 a day. 6000-8000 feels tiny until you realise that that's about where we were towards the end of September and not over the summer.

Yes, I posted something similar on a footie forum where folk were going on about how low cases are and we should open everything. Cases aren't low at all, they just look low compared to when it was scores of thousands a day. Problem is exponential growth, we are still only two or three doublings away from being back in the shite.

wilsontown wrote: ↑

Wed Mar 03, 2021 2:28 pm

Yes, I posted something similar on a footie forum where folk were going on about how low cases are and we should open everything. Cases aren't low at all, they just look low compared to when it was scores of thousands a day. Problem is exponential growth, we are still only two or three doublings away from being back in the shite.

Yes. We should be locking down even harder from now until we are fully vaccinated. Schools should not be opening.

wilsontown wrote: ↑

Wed Mar 03, 2021 2:28 pm

Yes, I posted something similar on a footie forum where folk were going on about how low cases are and we should open everything. Cases aren't low at all, they just look low compared to when it was scores of thousands a day. Problem is exponential growth, we are still only two or three doublings away from being back in the shite.

Exactly. And our ICU occupancy has fallen to " high"

Herainestold wrote: ↑

Wed Mar 03, 2021 2:36 pm

wilsontown wrote: ↑

Wed Mar 03, 2021 2:28 pm

Yes, I posted something similar on a footie forum where folk were going on about how low cases are and we should open everything. Cases aren't low at all, they just look low compared to when it was scores of thousands a day. Problem is exponential growth, we are still only two or three doublings away from being back in the shite.

Yes. We should be locking down even harder from now until we are fully vaccinated. Schools should not be opening.

I don't know about this. Having schools closed is an economic and social disaster, so there's a clear case for opening them if at all possible. And right now they aren't actually completely closed with children of key workers etc being accommodated. Last summer there didn't really seem to be a massive impact on case numbers when schools reopened in early September. The disastrous rise in cases seemed to start a few weeks later when universities reopened for teaching.

It's all a bit "what I reckon", but certainly in York cases were initially concentrated in the student age population and were particularly high in areas of the city that have large student populations.

shpalman wrote: ↑

Wed Mar 03, 2021 12:30 pm

Please note, this analysis has not yet been peer-reviewed or published, so the data is in preliminary form which only journalists are allowed to see, not scientists

I'm guessing it wasn't there when you looked earlier, but I followed the link to the SMC comments and at the bottom of that page there was actually a link to the preprint: https://papers.ssrn.com/sol3/papers.cfm ... id=3796835 (full text as PDF: long url)

jimbob wrote: ↑

Wed Mar 03, 2021 2:42 pm

That chart shows just how fast things can change from getting better to getting much, much worse.

jdc wrote: ↑

Wed Mar 03, 2021 7:03 pm

shpalman wrote: ↑

Wed Mar 03, 2021 12:30 pm

Please note, this analysis has not yet been peer-reviewed or published, so the data is in preliminary form which only journalists are allowed to see, not scientists

I'm guessing it wasn't there when you looked earlier, but I followed the link to the SMC comments and at the bottom of that page there was actually a link to the preprint: https://papers.ssrn.com/sol3/papers.cfm ... id=3796835 (full text as PDF: long url)

Oh right, thanks, I think that's how I found the preprint about the Scottish data too.

Sciolus wrote: ↑

Wed Mar 03, 2021 7:16 pm

jimbob wrote: ↑

Wed Mar 03, 2021 2:42 pm

That chart shows just how fast things can change from getting better to getting much, much worse.

Yup. Almost as though exponential growth is bad when it comes to infections.

I do wonder if some of the Covid deniers have actually started rooting for the virus, so are trying to push the opening up before the vaccine can protect too many

shpalman wrote: ↑

Wed Mar 03, 2021 7:17 pm

jdc wrote: ↑

Wed Mar 03, 2021 7:03 pm

shpalman wrote: ↑

Wed Mar 03, 2021 12:30 pm

Please note, this analysis has not yet been peer-reviewed or published, so the data is in preliminary form which only journalists are allowed to see, not scientists

I'm guessing it wasn't there when you looked earlier, but I followed the link to the SMC comments and at the bottom of that page there was actually a link to the preprint: https://papers.ssrn.com/sol3/papers.cfm ... id=3796835 (full text as PDF: long url)

Oh right, thanks, I think that's how I found the preprint about the Scottish data too.

At times, getting hold of preprints does seem to be a bit bottom of a locked filing cabinet stuck in a disused lavatory with a sign on the door saying Beware of the Leopard.

The frustrating thing is when the press and the experts they email for comment have all seen it but won't tell us poor f.ckers where to find it.

COVID-19: a new emerging respiratory disease from the neurological perspective

Coronavirus disease 2019 (COVID-19) has become a challenging public health catastrophe worldwide. The newly emerged disease spread in almost all countries and infected 100 million persons worldwide. The infection is not limited to the respiratory system but involves various body systems and may lead to multiple organ failure. Tissue degenerative changes result from direct viral invasion, indirect consequences, or through an uncontrolled immune response. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spreads to the brain via hematogenous and neural routes accompanied with dysfunction of the blood–brain barrier. The involvement of the central nervous system is now suspected to be among the main causes of death. The present review discusses the historical background of coronaviruses, their role in previous and ongoing pandemics, the way they escape the immune system, why they are able to spread despite all undertaken measures, in addition to the neurological manifestations, long-term consequences of the disease, and various routes of viral introduction to the CNS.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7884096/

Allegations of bad practice regarding a NEJM article on school closures in Sweden: https://www.sciencemag.org/news/2021/03 ... ks-swedish

In a review about children’s role in the pandemic, published in Acta Paediatrica in May 2020, he reported there had been “no major school outbreaks in Sweden,” which he attributed to “personal communication” from Tegnell. But as critics noted, Swedish media had reported several school outbreaks by then, including one in which at least 18 of 76 staff were infected and one teacher died. (Children were not tested.)

His NEJM letter sounded another reassuring note. It reported that in all of Sweden, only 15 children, 10 preschool teachers, and 20 school teachers were admitted to intensive care units for COVID-19 complications between March and June 2020. The authors noted that 69 children ages 1 to 16 died of any cause in Sweden during that same period, compared with 65 between November 2019 and February 2020, suggesting the pandemic had not led to an increase in child deaths.

But the emails obtained by Malmberg show that in July 2020, Ludvigsson wrote to Tegnell that “unfortunately we see a clear indication of excess mortality among children ages 7-16 old, the ages where ‘kids went to school.’” For the years 2015 through 2019, an average of 30.4 children in that age group died in the four spring months; in 2020, 51 children in that age group died, “= excess mortality +68%,” Ludvigsson wrote. The increase could be a fluke, he wrote, especially because the numbers are small. Deaths in 1- to 6-year-olds were below average during the same period, so combining the age groups helped even out the increase, he noted.

KAJ wrote: ↑

Tue Mar 02, 2021 6:45 pm

Bird on a Fire wrote:

Tue Mar 02, 2021 6:06 pm

The model details are somewhere on his site, rather than on twitter (which is not the easiest medium to navigate!) - here's his latest post https://bskiesresearch.wordpress.com/20 ... -velocity/ but I think the stuff about his models are way back in time. Might go digging later if nobody else finds them.

shpalman wrote: ↑

Tue Mar 02, 2021 6:14 pm

Try https://www.medrxiv.org/content/10.1101 ... 20065227v2

Thanks to both, looks like some interesting reading.

Well...
I've had a [too] quick look at those links and explored [as well as I can] James Annan's Twitter feed (apologies if that's the wrong terminology).

I think it's very clear that his now/forecasts (such as in my post yesterday) are firmly based on statistical/epidemiological models. I don't think it's correct to say ...

Bird on a Fire wrote: ↑

Mon Mar 01, 2021 11:11 pm

<snip> He models the pandemic trajectory as a random walk with no underlying statistical model whatsoever, just fitting to the data. <snip>

I'm not equipped to assess his models and I'm carefully avoiding inferring mechanisms or forecasts, so I haven't studied this aspect in any detail.

Turning to the case numbers charts which sparked this conversation (e.g. link) I couldn't find any real description, explanation, or motivation. I think the first appearance of the cases plot in his Twitter feed is on 16 Jul 2020 (failed to make a link to that ) when he said: "Also: new plot showing weekly change in daily case numbers, the last week is averaging small growth. The dots are differences from value 7 days previous". That charts Y-axis was "Difference in log(Cases)" on a scale apparently linear in logs. I think 9 Sep 2020 is the first chart with a Y-axis "Ratio of cases" on a scale apparently linear in ratio. I surmise ( ) that the earlier means were of logs and the later of ratios.

As I suggested above I mildly prefer:

• moving average to latest multiple latest averages
• arithmetic means for cases
• geometric means for ratios

I've also averaged before differencing, which allows me to use a 1-day difference. On today's data that gives me this: ... which makes clear that the feature attracting lpm's attention is dominated by 22 Feb. (In hindsight this was also clear in the chart I posted on Monday which more closely follow Annan's protocol.)

Here's my usual fit to PubCases (R-sq = 97%): Now we know where to look it does seem as if 22 Feb is a bit above the fit. It becomes really clear if we look at the residuals from that fit: While the points near 22 Feb are perhaps, maybe, a little, higher than usual - without the 22 Feb point they wouldn't really be noticeable. I see something similar but much less marked looking at cases by specimen date (not shown 'cos can't seem to include more than 3 pictures).

OK. Enough for now. It's been interesting but I'm parking this subject for now.

Interesting. The 22nd clearly pulls it upwards.

But my first expression of concern was on 19th. Probably with data up to 18th. That's the smaller but distinct upwards bit of the chart prior - the worst upward tick since the 5th Jan lockdown bedded in.

The 22nd outlier then reinforced that concern. There was a string of 10 positive residuals from 16th which is unique in the sequence.

I'd say things genuinely looked worrying for over a week - but was probably noise.

Brazil Breaks A New Record with 1,840 Covid-19 Deaths in 24 Hours
The country also recorded, for the fifth consecutive day, a new record for the moving average of deaths, 1,332; it's 42 days in a row with an average of over 1,000

https://www1.folha.uol.com.br/internaci ... ours.shtml

shpalman wrote: ↑

Mon Mar 01, 2021 10:45 am

Pfizer vaccine may be less effective in people with obesity

Italian researchers have discovered that healthcare workers with obesity produced only about half the amount of antibodies in response to a second dose of the jab compared with healthy people. Although it is too soon to know what this means for the efficacy of the vaccine, it might imply that people with obesity need an additional booster dose to ensure they are adequately protected against coronavirus.

Previous research has suggested that obesity – which is defined as having a body mass index (BMI) over 30 – increases the risk of dying of Covid-19 by nearly 50%, as well as increasing the risk of ending up in hospital by 113%.

Non-peer-reviewed preprint at https://www.medrxiv.org/content/10.1101 ... 21251664v1

Covid deaths high in countries with more overweight people, says report

I think the report they mean is this pdf: https://www.worldobesityday.org/assets/ ... -Atlas.pdf