Friday, 9 April 2021

Seems plausible

The BBC's headline is: Covid: UK vaccine rollout 'breaking link' between infections and death.

Seems plausible to me. I took the stat's on new daily cases and new daily deaths from, overlapped them and shifted deaths back by about two weeks to get the best match. The brown line for 'deaths' is clearly dropping much faster than the blue line for 'cases':
The alternative explanation is that all the really susceptible people have already succumbed, anybody who's survived this far is resistant or immune.

Obviously, neither case numbers nor death numbers are accurate in absolute terms, but as long as the stat's are compiled consistently, the trend is reliable.

Interestingly, assuming the stat's to be reliable, the death rate appears to be about 2.5% (much higher than often assumed) i.e. if there are 1,000 new cases today, there will be 25 new deaths today in two weeks (or whatever the lag is between diagnosis and death).


DCBain said...

The statistics are being manufactured to suit someone's narrative. "Of wuflu" and "with wuflu" are being conflated every day and the test numbers are the result of a not-for-purpose test with user-defined parameters. Pardon my scepticism . . .

Mark Wadsworth said...

DCB, you've missed the point.
The 'deaths' are almost certainly overstated - but consistently so. The 'cases' are almost certainly understated - but consistently so.
Therefore, an apparent *relative* change reflects an underlying *real* change.

Piotr Wasik said...

DCB, also, I don't think this particular error (counting "with wuflu" as "of wuflu") changes stats a lot. they have a month window from positive covid test to death to count it as "covid death". Probability of dying within any given month is not very high, even for an octagenarian (except for terminally ill people). I think I heard it is about 1 in 6 per year for an octagenarian, so it would be ~ 1 in 70 in any particular month. Of course if zerohedge finds a single instance of counting a motorcycle accident victim as "covid death", they will ridicule it aloud (as they did once).

MW - "an apparent *relative* change reflects an underlying *real* change" unless they change the way they count tests but not deaths if they test random people a lot more, and still test almost every suspicious death, it may appear as death ratio is going down.

Mark Wadsworth said...

PW, your second observation is quite correct - you can see that during the first wave the apparent death-to-case ratio was much higher, because they simply weren't doing that much testing.

The official number of tests (from here) shows that the absolute number shot up in early March (when kids went back to school) and has started falling again.

Despite this, the number of 'cases' (positive tests) did not go up a lot in early March.

So I assume that the official number of 'cases' over the last few months is a reasonable indicator of the general trend in cases. By the same token, it would be daft to compare the official number of cases of March 2021 with the official number of cases in March 2020.

Piotr Wasik said...

MW - "Despite this, the number of 'cases' (positive tests) did not go up a lot in early March." - agree for aggregate numbers. if you break it down into age groups: - it seems that infection rate among oldies went down (vaccinations?), and among youngsters - went up (schools?). They offset each other. Not sure what to make of 2-3 year old children, they don't attend schools. Why all graphs go up and then down? The person who twitted it says it is a modelling artifact, I can't comment myself.

Piotr Wasik said...

and as of risk of dying for octagenarians, I remember now, you posted this :-)

it is indeed 1 in 6 (M) or 1 in 7 (F) but only after 85, so I still stand by my first remark as well. "with wuflu" is a good approximation of "of wuflu".

Mark Wadsworth said...

PW, I think you are reading too much into this.

The point of my post was, that for once, the BBC headline appears to be correct.

Will it turn out later that it was all bollocks? Possible but unlikely.