Estimating Daily People Tested (part 6)
Estimation of the number of people undergoing virus tests in England prior to de-duplication of data records (rev 1.0)
In my previous post in this series I ran subscribers through two more linear models that successfully predicted the unique number of people tested each day from the number of tests and number of cases detected. I say ‘cases’ because that is the sloppy jargon the experts are using these days to denote a positive test result (which may or may not be a truthful). I promised to stitch all three models together and here is the result:
Tasty innit? So what we are looking at here is the output of some linear modelling that accurately tracks counts of unique people tested generated by the England Test & Trace mob. In case some inquisitive types are wondering I can report the Pearson correlation coefficient between both series fetches up at r = 0.975 (p<0.001, n=860); that is to say the model accounts for 95% of the variation we see in the source data. That ain’t bad going and deserves a celebratory cuppa!
Yes, But Why Bother?
We are getting closer to the reason why I’ve been bovvered and am bothering to fart around for hours on end trying to fathom who got tested each day. What I can do now is convert the modelled daily series to a rolling 7-day count and compare this to the official UK GOV rolling 7-day count for England that may be found here. Here’s a rather telling slide:
In the blue corner we have the official rolling 7-day sum as provided by the UK GOV coronavirus dashboard and in the green corner we have a decently-modelled estimate based on what the England Test & Trace crew are claiming. Somebody can’t count and I don’t think it is me!
So here’s the thing…
The UK GOV coronavirus dashboard team are using their pretty blue curve to calculate the case positivity rate for the nation. Because they are missing rather a large number of people it means their estimates are seriously over-inflated. If we take the peaks of 3.5 million for the UK GOV series and 8 million for the modelled Test & Trace series we can see that over-inflation comes at a whopping great factor of 2.3x (230% inflated in old money). Halve the claims made by UK GOV for case positivity and you might just get closer to the truth!
It is interesting that this wasn’t always so for we see an initial period when the two curves were near-identical, this being prior to the test-everyone-like-crazy LFD farce that must have generated more false positives than I’ve had hot dinners. Once this nonsense kicked in it is clear that the authorities have struggled to count heads in a consistent manner. And on top of all this we are still only limiting counts to one test result per person per accounting week regardless of how many times they tested themselves!
Doh! I need a doughnut, or is it donut?
In my next newsletter we are going to do some cogitation because we have a tricky problem that I shall pose in the form of a question…
Q: how can we possibly put back all those (negative) test results that were junked through the process of de-duplication such that we get back to counting the actual number of times folk were subject to testing each week?
Kettle On!
Doughnut 😁