JavaScript is not available.

ZOE Covid Study - public data files

Aug 25, 2022

Replying to

@AlastairGrant4

and

@timspector

I have some feelings about that last part, heh. To be constructive, if people are looking at Zoe, I hope more people will play with the "covid-public-data" files. Let a thousand graphs bloom! They're not at all well-known, so I wrote it up here.

sourcejedi.github.io

A blog?

Paul Mainwood

@PaulMainwood

14h

Update. I can also quote retweet them. Given that's a common way of harrassing, that seems really, really bad.

Trannery Sultanas

@Chican3ry

Just carving out bizarrely specific terms for which medical conditions allow you to still register as a woman based on the vibes of the athletics body with no real scientific basis. Frankly it's hard to believe this is legally justifiable.

164

Trannery Sultanas

@Chican3ry

Important correction: because so many women with PCOS have higher levels of testosterone, the way to be allowed to bypass the testosterone test is to be one of the 30% of women who have pcos Which is even more silly as it shows how the limit is arbitrary

Quote Tweet

Natalie Washington

@Transsomething

Replying to @TotallyTills and @Chican3ry

Just checked on this one btw - the policy document excludes PCOS in one of the footnotes, so PCOS causing a T level above 2.5nmol\L alone wouldn't cause a women to be banned

189

Ian Dunt

@IanDunt

22h

Terrible approach, with a seeming insinuation that - what? - Labour would try to make cannabis Class A? That would be utterly insane.

Quote Tweet

Matt Dathan

@matt_dathan

Keir Starmer insists a Labour government will not treat recreational use of cannabis as a "low level" crime. He says "cannabis smoke creeping in from the street downstairs" is "ruining children's lives". Quite a contrast to Sadiq Khan's approach to cannabis...

773

Paul Mainwood

@PaulMainwood

I'm blocked by four people (I think - may be more). That's fine; their choice. What's weird is that their recent tweets are being pushed to my "For you" tab. Not sure about how blocking is meant to work, but surely that shouldn't happen?

John Roberts

@john_actuary

19h

And that's it. Sincere thanks to all those involved at the ONS, its partner organisations, Oxford and Manchester, Universities, Wellcome Trust and UKHSA, not to mention all those who collected data and all those who participated in the surveys over the last 3 years. The End. 5/5

SpeakOutSister

@speakoutsister

Mar 24

The permitted levels of testosterone have now been lowered to a very narrow "normal" range. This will probably exclude even more cis women.

150

1,380

SpeakOutSister

@speakoutsister

Mar 24

Trans women have been able to compete in international athletics for 12 years. In this period no trans women have competed but cis women have been barred because of their naturally.occuring testosterone levels.

1,442

6,781

Your personalized program is ready

Trannery Sultanas

@Chican3ry

Mar 24

It looks essentially like they will end up banning about half of women with PCOS and a handful of women without PCOS in order to exclude trans women none of whom were competitive at an elite level anyway.

393

I feel people look at "ZOE Health Study" and see pandemic stats. But most pandemic stats were, like - questioned by press - regulated by ONS or UK statistics authority, etc - sharing their current methods for review by scientific peers - accountable to voters - not funded by VC

I don't like talking about this, but I think a lot of progressive, covid-conscious people would be - or are - very smart about ZOE's business. Venture capital funded diet app. How scientifically rigorous it is. And their prices.

archive.is

archived 23 Mar 2023 22:33:43 UTC

I has a coping strategy: I don't think too many people are paying attention to ZOE COVID estimates (and communications). At this point, it can't do society-level damage.

Quote Tweet

Unfortunately I fear that the result will be a misinformation gap, where the usual suspects can whip up hype about some new "killer variant" and scatter a few local ZOE graphs, and cooler heads won't be able to point at the variant percentages and growth rates to defuse that.

Show this thread

Quote Tweet

Replying to @AlastairGrant4 and @timspector

Quote Tweet

Replying to @sourcejedi @peter_boring and 3 others

5/n. By their actions, ZOE agreed that the jump is unacceptable. On both Friday and Monday: ZOE published files with this jump in the morning, and then revised it away in the afternoon. It appeared on Sunday and they didn't revise it away. But Sunday is the weekend.

Quote Tweet

Replying to @sourcejedi @peter_boring and 3 others

4/n. When ZOE first recalibrated against ONS, they produced sharp discontinuities. Similar to our anomalous jump. Except the discontinuities did not affect the last ~70 days of the time series. twitter.com/sourcejedi/sta

Quote Tweet

Replying to @sourcejedi @peter_boring and 3 others

3/n. Therefore, the jump is due to a transformation applied on top of the 14-day average. @/JoePajak has also been blaming "recalibration". ZOE recently announced they would now recalibrate against ONS. So there is potential for teething problems. health-study.joinzoe.com/blog/covid-est

If anyone indeed thinks they are "watching closely", to see if the distance from the red line to the blue line will "mark the peak of the wave", they are very confused. And whatever they may be "experts" in, it is not the ZOE COVID estimates. inews.co.uk/news/science/c

ZOE England estimate of "newly symptomatic infections".
Comparing file from today with yesterday

- published 22/03 (red)
- published 23/03 (blue)

As previously described, the red line jumps up by 30% on March 15.
The blue line does not, and therefore ends up at a lower level.

read image description

ALT

Although it’s not advised to read too much into daily changes, experts will be watching the figures closely over the next few days to see if the drop on Tuesday (the latest day for which figures are available) does in fact mark the peak of the wave – or whether, as expected, cases will continue to rise for the next week or so.

read image description

ALT

If you want to claim this 14-day average dropped an impossible 20% per day? Don't say the drop happened on Tuesday 21st! It makes it sound like you and your sources can't read a graph. storage.googleapis.com/covid-public-d

Daily incidence rates across nations.

This graph lets us zoom in on a shorter time period than the UK graph.

The last two days for England look relatively flat.

England is the largest UK nation. England COVID trends usually dominate the UK trends.

read image description

ALT

read image description

ALT

Or, we can look at the last two dates on today's graphs. "By nominal date". Monday 20th (UK): 110,032 Tuesday 21st (UK): 111,241 Rising 1% per day is much more believable. (FWIW, it is a moderate rise). Numbers from this ZOE spreadsheet file: storage.googleapis.com/covid-public-d

read image description

ALT

The numbers above were the last two UK estimates "by publish date". Wednesday 22nd: 136,722 Thursday 23rd: 111,241 It does not tell us COVID dropped by 20% per day. It tells us ZOE removed a jump of 30% per day. Both swerves are impossible and not "real".

To be clear about dates. Today is Thursday, March 23. The estimate published today is > Based on PCR and LFT test data as of two days ago. > 111,241 daily new cases [...] in the UK on average over the two weeks up to [Tuesday] 21 March 2023. web.archive.org/web/2023031218

111,241

Daily new COVID cases in the UK

Figures updated daily at 9am and are based on PCR and LFT test data as of two days ago.

read image description

ALT

What daily data would produce this average? It would need to jump up about 400% in one day (March 15). It would then have to jump *down* on the next day (March 16). Enough to keep the average from rising much. But not so much it let the average fall. And so on. Nonsense.

Daily incidence rates across nations

This ZOE graph lets us look closer at the most recent incidence rates.

The incidence rates for all nations jump up ~30% on a single day, March 15.

The data with the narrowest confidence interval - least random error - is for England. Following the jump, the England estimate appears relatively flat.

read image description

ALT

The 30% jump on March 15 made no sense. Because ZOE is based on a 14-day average. storage.googleapis.com/covid-public-d

Daily new cases of symptomatic COVID

We estimate there have been 111241 daily new cases of symptomatic COVID in the UK on average over the two weeks up to 21 March 2023. This is based on the number of newly symptomatic app users per day, and the proportion of these who give positive swab tests.

read image description

ALT

Today's ZOE graph did not jump. Therefore the final estimate was lower. ~20% lower. storage.googleapis.com/covid-public-d

Daily new cases in the UK

111241 new cases in UK on 21 March 2023

read image description

ALT

Yesterday's ZOE graph jumped up ~30%, on March 15. storage.googleapis.com/covid-public-d

Daily new cases in the UK

136722 new cases in UK on 20 March 2023

read image description

ALT

Me: "of course nobody will report the increase in ZOE incidence as actual news, because it's clearly a statistical artefact" INews: hold my beer inews.co.uk/news/science/c (And of course ISAGE are all over it)

Quote Tweet

Alan Jenkins (@sourcejedi_ben@mastodonapp.uk)

Today, 111241. My current hypothesis is that there's one doomer and one minimiser who work for ZOE, and whoever gets to the office first gets to put the numbers on the webpage that day. twitter.com/BristOliver/st…

Content warning: ZOE #COVID estimates are not working as they usually do

Mar 22

The part I'm confused about is when people keep telling us the parties were NBD and everyone knew, when Allegra Stratton had to resign just for joking about one of them.

Luke Weston

@lukeweston

Mar 22

Full moon calls thee - Shai-hulud shall thou see; Red the night, dusky sky…

Quote Tweet

Unwanted Blog

@UnwantedBlog

Mar 21

2/2 Basically a tube with a number of external ring segments each having a flattened foot. By undulating the rings (“traveling wave bellows”), the vehicle could trundle along over any lunar terrain. Tops speed would be 5 mph. Range was to be 100 mi, 1000 lb payload.

Replying to

136,722 on file for this morning, and high figures restored all the way back to the 15th...

Replying to

Nope - we were wrong. ZOE at 7:20 am says 136,722. So they went for the "high" algorithm. Thus I am picking 110k-120k for tomorrow!🤣 ZOE says 1,548,734 for total cases, btw. Whether they update the update remains to be seen.

Mar 21

Thread.

Quote Tweet

Mar 21

Replying to @peter_boring @PJeffcock and 2 others

1/n. Right. These are good questions. The incidence time series (graph) which jumps 28% in one day is not "real". twitter.com/BristOliver/st

Replying to

Spot the difference

read image description

ALT

read image description

ALT

Doug Parr

@doug_parr

Mar 16

Such a good graphic from

@FT

on the difference between a climate warming of 1.5C and 2C On many of the metrics like crop yields, species loss, heatwaves, the impacts don't just go up a little, they double ft.com/content/450a59

287

408

Replying to

and

I appreciate you stopped showing incidence graphs just *before* they showed this impossible jump. For good reasons! But right now, if you only show the prevalence graph, you end up hiding the anomaly and smoothing it into a steady rise. See:

Today's incidence by age group graph, which continues to show the impossible jump on graph date March 15, and comparatively dead flat for three days afterwards.

read image description

ALT

Quote Tweet

Mar 19

Replying to @sourcejedi @ProfJSDavies and 2 others

(3/4) Pajak's graph is prevalence, which shows 3 days sloping up. Spector's graph above is incidence, which shows 1 jump + two flat days. This is because ZOE calculate prevalence from incidence. An easy way to think of it is as a weighted rolling average. github.com/sourcejedi/nov

scream is on mastodon: @botsin.space/@scream

@infinite_scream

Mar 14

AAAAAAAAAAAAAAAAAAAAAAHHHHHHHHHHHHHHH

Replying to

and

... and this is the plot issued on Friday - also has a problem on the extreme right data points. The bug is also affecting the other plots and the headline rate.

Replying to

It looks like there's a problem with the data for the last three days in the orange and green lines in your tweet. Here is yesterday's plot - perfectly normal looking data.