Tom Chivers won the 'Explaining the facts' category in the 2020 RSS Statistical Excellence in Journalism Awards for his article, Do we really have a ‘suicidal generation’?
We asked Tom about his approach to using statistics in his writing.
How do statistics help you tell a good story?
So this might not be exactly the answer you're looking for, but I don't use statistics to tell a good story. My concern is that statistics, used unwisely, can tell stories that aren't true. For instance, I wrote this piece about a Sunday Times story that reported on a big jump in teen suicides; but they did it by cherry-picking their starting point and looking at a small subset of data, which allowed them to tell a really good story - but one which may well have been misleading. At the risk of sounding a bit pious, I try to use statistics to find out things that are true, to the best of my limited ability, rather than to tell good stories.
What do you think is the most important thing to bear in mind when reporting on statistics?
That the numbers you choose and the way you represent them can hugely change the story they tell. They are not some unvarnished window on the truth. For instance, you can say, 'eating Tic-Tacs raises the risk of heart attack by 38%', but if it goes up from 0.1% to 0.138% you might not care very much. Or you can say, 'the government will spend £50,000 on free corsets for the under-fives', which might sound a lot until you know how much the government spends in total. And, as we saw with the suicides thing, you can cherry-pick data, or you can do all sorts of other tricks, to find interesting-looking numbers which are actually misleading. This isn't to say statistics are all lies – they're the best way of finding out true things that we have – but they need to be used with enormous caution and humility. 'It's easy to lie with statistics,' goes the saying, 'but it's even easier to lie without them.'
What can statisticians do to help journalists in putting numbers into context for their readers?
Saying what a statistic doesn't mean is often helpful. If we see that two numbers A and B are correlated, it's worth explicitly pointing out that that doesn't mean that A causes B (or vice versa). And it's important to present risk in absolute, not relative, terms - 'this will mean your risk goes up from six out of 100 to eight out of 100, meaning two out of every 100 people will be affected', rather than 'your risk will go up by 25%'. And being clear that point estimates come with uncertainty intervals, that would be another one.
Has the Covid-19 pandemic made readers more stats-savvy?
Well, I'd need to see the numbers. (Seriously, I would.) I think it's made people more au fait with certain terms, like "R", and perhaps with things like prevalence. But I wouldn't necessarily say that it's made people better at understanding the uncertainty behind numbers. I hope it has, and I think it's perfectly plausible that it has, but I don't know.
How do you try to make statistics accessible when using them to explain a news story?
I'm from a non-mathematical background - I did philosophy at university. So I find mathematical notation, all those sigma signs and so on, really offputting and hard to navigate. But I find that if I write the steps of an equation out in words, I can follow them pretty easily. I kind of assume most people are similar: that if they see a load of conditional probability notation they'll kind of turn off, but that if you talk them through how the equation works, most people will be able to follow it. Sometimes that means whole paragraphs of writing dedicated to some mathematical concept that a statistician or economist would represent in a few characters, but I see it as helping others (and myself) follow the process.
How do you decide which numbers to use and which to leave out – do you have a limit to the amount of stats you use in any given story?
I don't really have any conscious method for deciding which numbers to use or not use, or any limit, beyond the ones that are necessary to explain the concept at hand. A few times I've thought 'OK, this piece is already pretty stats-heavy, I'd better stop throwing in new ones,' but there's no red line or anything.
The Economic and Social Research Council funds research on the social and economic questions facing us today. If you could choose a question to research, what would it be?
You know what, now you've mentioned it, I'd be really interested to know whether the Covid-19 pandemic has increased statistical literacy. If we could find some sort of baseline to measure against, I'd love to know if British people are more comfortable with statistics in the news than they used to be.
Read more about our 2020 Statistical Excellence in Journalism Awards in our announcement