June 12, 2020 - 7:00am

We bought some tasteful face-masks the other day. Iā€™d been wearing a crappy snood thing when I went into shops for a week or two, but it was a single layer of nylon and probably almost entirely ineffective.

The turnaround on mask policy has caught attention ā€” the UK government began by telling people they didnā€™t work, then ‘advising people to consider‘ using them in enclosed spaces, now mandating them on public transport; other national governments and the WHO have been on similar journeys.

To understand what happened there, itā€™s worth reading this post. It quotes an old BMJ paper, a joke one issued at Christmas, which pointed out that there are no randomised controlled trials showing that parachutes save lives when you jump out of a plane.

Itā€™s pretty obvious why that is; you canā€™t ethically make 50 people jump out of a plane with a placebo parachute. So you have to make do with observational evidence ā€” people who jump out of planes without parachutes tend to die more ā€” even though that canā€™t, strictly, prove causality.

The same happened with masks. You canā€™t have a bunch of healthcare workers walking around hospitals with no masks on, because it would put patients at risk. So the studies never got done, so the evidence remained really bad. And medical regulatory bodies demand really high standards of evidence to say something works, so if something has bad evidence, they say it doesnā€™t work.

Normally thatā€™s a good thing ā€” it means they donā€™t end up licensing a bunch of crappy quack medicines ā€” but in this case, it led to people confidently saying ā€œmasks donā€™t helpā€ when they meant ā€œwe canā€™t prove masks help, but thereā€™s a decent chance that they doā€.

Now though more evidence is coming in. On Twitter, the statistical epidemiologist Adam Kucharski points to two studies, one finding that masks reduce daily growth rate of infections by about 40%, and the other finding no evidence that people start behaving more riskily if they wear them.

Both studies were carried out in Germany, because mask policies were introduced by different German states at different times, so that there was a natural control group (equivalent to the people being thrown out of aeroplanes with placebo parachutes): you could see if Covid cases dropped in states with mask policies and not in those without. Itā€™s not a perfect method, because other things may be different, but itā€™s suggestive.

There are two lessons from this I think. One is obvious: people need to be better at talking about weak evidence and uncertainty, rather than ā€œit worksā€ or ā€œit doesnā€™t workā€. And the other is that if the evidence for some policy is weak, itā€™s a really good idea to allow local authorities some autonomy on whether to introduce it, in order to create these sorts of natural experiments which can provide better evidence.

I think itā€™s fair to say that now the evidence in favour of masks is pretty strong. And now I have my ones that actually work, thatā€™s even better.


Tom Chivers is a science writer. His second book, How to Read Numbers, is out now.

TomChivers