Why doesn't the UK test passengers arriving at airports? If you've been following this debate you'll have heard a lot about one particular number

Ministers from the PM on down have insisted that testing will only pick up 7% of infections

Is that right?

THREAD

I wanted to know more, so I dug up the source for this figure. It comes from a piece of modelling commissioned by Public Health England early in the crisis

This modelling doesn't rely on real world data. It simplifies the problem by making a set of assumptions, then running calculations to test different scenarios

The 7% figure is correct if the assumptions are correct

But the assumptions are, as one epidemiologist put it, "weird"

To see this, let's look at one assumption the paper makes. It assumes that no one flying to the UK is ill with COVID-19

Yes, that's right

In this model, no one gets on a plane with full-blown coronavirus - that is, coronavirus that has developed past the incubation period

It sounds unbelievable, but it's true

In the world of the model, anyone with coronavirus that can be detected by a test is presumed to be too ill to fly or is stopped from getting on the flight by screening systems in foreign countries

As you might imagine, this has a significant impact on the paper's conclusions

Just over 60% of infected passengers are counted as "non-fliers" who'll never reach the UK border

(Here's how they describe it in the paper )

I asked Luca Ferretti of the Oxford Big Data Institute what impact this assumption had on the model's results

He did a back-of-the-envelope calculation which showed that if *just half* of *asymptomatic* people weren't screened out, you'd catch 1 in 3 infected travellers

To repeat: the model assumes that every asymptomatic person is picked up by infallible screening systems

Without even getting into the question of whether ill people will fly (spoiler: they will), you can change its results significantly just by altering that one assumption

And the assumptions don't end there

The paper assumes that
- testing can't pick up pre-symptomatic people
- all countries have the same prevelance
- travellers move around in the same way as they do when they're at home

Statisticians don't like giving rough figures without having time to consider, but Dr Feretti very kindly gave me a rough idea of what airport testing could capture, using the figures from this model as a starting point

It's not 7%

It's 40%

None of this is meant as criticism of the modellers. To my non-expert eye it looks like very respectable work (and in any case they weren't actually considering the question of airport testing vs 14-day quarantine) https://www.gov.uk/government/publications/phe-investigation-into-the-effectiveness-of-double-testing-travellers-incoming-to-the-uk-for-signs-of-covid-19-17-june-2020

But I do think we should be careful with models are the truths they claim to represent - and I think we should expect government ministers to be careful as well

I asked Dr Ferretti about this. Here's his response

I asked DHSC about this, who told me the modelling had been approved by SAGE - and that work was ongoing on the benefits of airport testing

But I didn't get to find out why modelling was being used, when we now have an abundance of real world data

That to me is the big question

My piece, with lots more detail about the model

Models are great, but they're just a tool. The ideas that they can produce truth from complex, uncertain systems is an assumption we need to question https://news.sky.com/story/coronavirus-why-doesnt-the-uk-test-passengers-arriving-at-airports-12109065

(Oh, and thanks to Dr Ferretti and @Kit_Yates_Maths and colleagues for their help. As anyone who's ever seen me do even basic maths can attest, it was sorely needed)