Discover more from bad cattitude
the number of single vaxxed people strongly predicts covid deaths
more data from palestine and a metric from which we can make forward predictions
two weeks ago, i looked at palestine and the uptick in covid deaths that rapidly followed a rise in covid vaccination.
we saw this and it seemed quite provocative as the uptick in vaccination closely preceded the uptick in deaths (as in so many other places we have seen) and the drop in vaccine dosing looked to be leading to a possible flattening in deaths which is what we would predict if vaccines were causing a short term spike in deaths.
this is the “two week worry window” thesis that we have seen significant clinical and societal validation for. in the 2 weeks post vaccination, people seem immuno-suppressed and highly vulnerable. (walk through and data HERE)
this model seems to definitively address the “maybe it’s people rushing to vaccinate when deaths rise” hypothesis for correlation. it’s pretty clear which is leading which here.
so let’s update for the next 2 weeks and see what we have. (i have zoomed in on 7/15 to present to make this easier to see. all data from OWID. the vaccine data seems to lag a little and is therefore truncated vs deaths)
we can see that vaccines tick up on 8/15. on 8/20 or so (one could argue this a bit) deaths tick up implying a 5-8 day lag.
deaths then flatten 5 days after the peak of vaccinations though they do keep rising until peaking 9/21. it sort of fits, but is not as good as previous inflection.
of interest, on 9/19 vaccines then begin to spike again in terms of dosing. one might be tempted to presume that this will once more drive deaths to new highs.
i do not think it will.
here’s why: the surge from 8/15 was mostly first doses. it looks like the surge now is mostly second doses, and second doses do not seem to exhibit this effect.
we can get some intuition on this here.
but this is “stats with cats” so let’s see if we can go beyond intuition and quantify this. i used the rough expedient of taking the % of the population with one or more doses and subtracting from it the % of the population with 2 doses. this should get us pretty close to a measure of the size of the 1 dose group, a group shown to be the big danger category in the UK all cause deaths data.
then we can plot it against deaths. the results are provocative.
5 days after the slope of the growth of the single vaccinated (SV) cohort abruptly shifts on 9/4, deaths flatten.
they move to a lower slope from 9/8 until 9/21 peak, which is, tantalizingly, 5 days after the peak of the size of the SV cohort on 9/16.
we’re seeing a remarkably tight follow here and it really pops if you shift the data 5 days so that cause and effect are stacked temporally. cohort size aligns with truly eerie precision.
this level of correlation (below) validated by curve fit is OULANDISHLY high in a system this complex. (honestly, it’s so high i’m wondering if i’ve stumbled into some kind of autocorrelation i missed, but the alignment of the raw curves is also so tight and their inflections so similar that this appears real)
i think we’re onto something here. this looks like a strongly predictive variable for deaths, it’s clear which variable is leading which, and we have strong independent clinical basis to presume such a relationship.
and it lets us start to formulate a hypothesis with good backtested validation and the ability to make simple, testable predictions about the future.
if the size of the single vaccinated cohort keeps dropping, so too should deaths and they should follow at about a 5 day interval.
“number of vaccine doses” can rise and not drive a further rise in deaths.
it’s not about “overall doses,” it’s about “first doses.” (because that’s where the worry window is)
further, booster doses will act like first doses (as they have been in israel) and will cause rises in covid deaths like initial doses did.
so there’s the the model and the forward prediction.
and now, we wait and see if it proves out, because THAT and only that is the real test of a model.