Baking The Perfect Cake

Trying to fathom Test & Trace data tables (rev 1.2)

I am a simple man with a simple mind. I like numbers to add up and I like to be able to understand what those numbers mean. This morning I took it upon myself to update my spreadsheet with the latest Test & Trace England figures and run a few checks just ensure all was well and groovy.

But all was not groovy for I once again noted more people being tested than tests processed which doesn’t make sense. In my book if 10 people are tested this means you’ve processed 10 tests at the very minimum, and considerably more if those 10 people re-test themselves a few times per week. Hence I am expecting to see virus test to persons ratios exceeding 1.0, but that’s not what we find!

If I take the pillar 1 scheme¹ processed PCR total and the pillar 1 scheme people tested figures I get this chart:

This starts out groovy but rapidly sinks into nonsense as the blue wiggle descends below the ratio of 1.0, and particularly so during the autumn of 2021. Either we are missing tests or have double-counted people; then again I may have got the wrong end of the Test & Trace stick!

De-duplicating People

Stuffed down in the notes section of the source spreadsheet called NHS_Test_and_Trace_Data_Tables_20220526.xlsx I find the following entry in cell A11 underneath the table headed Table 1. People tested for COVID-19 each week, including those people testing positive, Pillars 1 and 2, England, 28 May 2020 to 18 May 2022 [Note 1, 2, 3, 4]:

[1] Deduplicated for each reporting week. Deduplication began on 15 October 2020.

I’ve covered the issue of de-duplication before but it is worth mentioning again because it’s one of those tricks that the data authorities play that the public are unaware of.

De-duplication in this sense means removing subsequent data records for those folk who decided to test themselves again within the same reporting week. This sounds sensible until you realise that you’ve trashed (negative) test results that should be going toward estimating the infection rate in an unbiased manner. Repeat this across the nation of England for all those who are required to test themselves more than once per week and it amounts to a very large number of missing negative test results indeed. Whilst this may suit the political aim of the data authorities it doesn’t suit the aims of analysts wanting to unveil the truth and so somehow we have to make estimates and put all those missing results back. This requires a formidable tin of biscuits.

Just To Be Sure

Just to be sure I wasn’t having a senior moment I decided to pull down the daily pillar 1 test figures for England from the UK GOV dashboard and compare weekly totals with the Test & Trace counts. Herewith that little tray bake:

That’s not too shabby, though we might ask why the two sets of figures are not precisely the same! There are several likely reasons for these differences starting with the reporting cycle and ending with methods of de-duplication. No doubt some bright spark could give us a lecture but the fact that these two curves mirror each other means something is distinctly wacko with how tested folk and processed tests are handled such that we end up with more people than tests.

The Grass Is Not Greener

With pillar 1 giving me a headache I decided to switch to pillar 2² only to be thrown a googly. Pillar 2 ratios simply didn’t make sense at all until I started dialling in all the different sources for test results. In the slide below you’ll see the virus tests per person ratio zip well below a value of 1.0 for central lab PCR testing at the beginning of 2021. If we add in satellite laboratory processing of PCR tests things look a little better, and better still when we add in home kit PCR testing. However, it is not until we dial-in the lateral flow device (LFD) use that the virus test to person ratio reaches a state of reasoned logic and the pale green curve wiggles along above a ratio of 1.0.

Whilst this is a blessed relief it also raises a concern on what denominators the authorities are using when they calculate case rates. My preference is positive cases detected per 1,000 tests to avoid multiple test bias³ but then again I’m cranky. Once I can fathom how to best approach the pillar 1 missing tests issue I shall reveal some super groovy wiggles!

Meanwhile keep your eye on the small print and be wary of anything that has undergone de-duplication. In my book de-duplication is like surreptitiously slicing the burnt bits off your Victoria sponge and claiming you’ve baked the perfect cake. Like the test result pool your cake will also be smaller.

Here’s some not so small print, which may be found here.

1. Clinical need: testing of those in care and frontline workers.

2. Testing of the wider community.

3. Each additional test a person may undertake may return a false positive result but what’s worse is that if a person tests negative and positive within the same week then only the positive result is counted.