How to Tell When Statistics Are Bullshit
We asked Tim Harford, host of statistics podcast More Or Less, how we can find out when newspapers and politicians are being liberal with the truth.
It's basically impossible to illustrate a story about statistics with anything other than weird 90s business stock–imagery. This one's called
In this world of competing and confusing numbers, More or Less, a statistics show hosted by Tim Harford, is a small oasis of mathematical rigor in a sea of endless numerical bullshit.
Here's a good example of what they do well. Recently they looked at the issue of shared parental leave. A year ago, the law changed on maternity and paternity leave in the UK to allow eligible couples to share parental leave between partners. This was a big policy for the government, and a year on a survey was conducted to find out how many fathers had taken up the option. Here's how the Telegraph reported it:
"On the first anniversary of a revolutionary policy that gives men the opportunity to care full time for their new baby, a new study reveals that just 1% of men have so far taken up the opportunity to do so... reasons for the reluctance include it being 'financially unworkable ', 'a lack of awareness,' and 'women (55%) refusing to share their maternity leave.'"
The Guardian, BBC News, and others reported the story in much the same way. But when the staff at More or Less looked into the story, they found the survey was conducted in a way that is close to insanity. It turned out that it wasn't 1 percent of new fathers who were eligible for shared parental leave who took it, but 1 percent of all men in the workforce, whether they had fathered a child in the past year or not. They found that even if every man who was eligible for shared parental leave had taken it, that overall figure would only have been 5 percent.
All of this is more relevant now than ever before, because most news operations are 24-hour rolling cycles, where press releases, particularly those from official-sounding bodies, may be posted unchecked. The shared parental-leave story ended up being analyzed by politicians and comment writers, and everything they were saying was based on utter, utter bullshit. And that's just traditional news sources. Increasingly, we are getting our information from Facebook memes and social media, and those statistics are even less verified.
More or Less, as well as doing some more whimsical stuff like "How long would it take to ride to the moon on a number 8 bus?", delves into the statistics that are bandied about by newspapers and politicians and tries to find out where they come from. I spoke to its host, statistics bad boy and Financial Times writer Tim Harford, to find out what we should be looking for when we see stats in the news.
VICE: The word "statistics" has a bad reputation. They are often seen as the most boring part of a news story. Why should we care about them?
Tim: Statistics are a very important way of understanding the world. There are sixty to sixty-five million people in the UK. There are seven billion people in the world. You can't get a picture of what's happening purely by anecdote. You need to see what the numbers actually say—whether you're trying to understand the doctors strike, inequality, health stories, the financial crisis, environmental change, anything really. Statistics will tell you things you can't learn any other way. One of the things that we do on More or Less is to ask quite simple questions that other people could ask but don't.
OK, so when you see a news story with a lot of statistics in, what are the sort of things that might raise a red flag?
I think there's a basic reality check you can do on most statistical claims. You can just ask: Does that sound reasonable? For example, somebody was tweeting me saying, "Apparently we throw away millions of disposable coffee cups every day, can this be true?" And I figured, well, I don't go out and count them personally, but even if it were six million, that would mean that ten percent of the population throws away a disposable coffee cup every day. Is it likely that ten percent of the population goes to Starbucks, buys a coffee, and then throws it away? Yeah, that sounds about right.
When I replied, she said, "Well that's a shocking waste." I thought, But is it? A lot of things times sixty million are going to seem like big numbers. It's not that difficult.
So are there certain big numbers, like the number of adults in the UK or the total UK public spending, that are worth keeping in our heads, so when we see stories that have big numbers in, we can provide some context?
The thing is, it's not very hard to check. A lot of this stuff you can find really quickly. If you've got Google, you can often find a proper fact-check of a claim that you doubt quite quickly from organizations that devote themselves to independent fact-checking, such as Full Fact. Or Snopes, who look at urban myths, the sort of statistics you might see on Facebook.
What does a statistical urban myth look like?
Well, for example, with the EU referendum, there's one going around saying that the EU has a twenty-seven thousand word memo on the regulation of cabbages, as a way of showing that it's a big unnecessary bureaucracy. And Snopes very simply lays it out and explains that story started in America in the 1940s and has been doing the rounds ever since. It's never been true.
If all this information is widely available, why do you think we are still misled by statistics in news stories?
I think we instinctively reject the statistics that were made by people we perceive to be political opponents, and accept the ones that chime with our view of the world. I saw a graphic on Instagram, I think it was from the Washington Post, which should know better, and it was about support for some aspect of gay rights. I can't remember the details, but it was showing that support for gay marriage is increasing in the US. And for me, that's wonderful news, I'm very pro-gay rights, so I just retweeted that. Then someone replied and said, "Have you seen the axis on that graph?" I looked, and the first two dots were twenty years apart. Then the next dot was two years further on. It was just a very dodgy graph. I hadn't checked. It would have taken about five seconds for me to have a proper look at that graph, but I just retweeted it, because it said something that I emotionally responded to, and that's just what we do.
Bullshit circulates at a higher velocity than it's ever circulated before.
Is the amount of statistics available on the internet a problem? Even when you have very good sources, it sometimes feels as if you get competing facts. That wealth of information can sometimes feel as though it's having a negative impact on trying to find the truth.
I think as with anything, the internet makes it very easy to drown in low quality noise if you're not careful, but it also makes it easier than ever to find high quality analysis. It used to be quite hard to find out things like development trends—what's happening to child mortality, for example—but now you can just click and download the data. You can see academics in the area blogging and analyzing. It's much easier to get really good, solid analysis than it ever was, but obviously bullshit circulates at a higher velocity than it's ever circulated before.
In terms of the years you've been doing the show, are there particular cases where statistics have been incredibly misleading that stick out to you?
In health and nutrition journalism, you see single studies given a lot of prominence, when in most cases there will have been hundreds of studies, and some new study doesn't overturn all the studies that have already been done. So often you'll see the new study reported as a sudden new truth.
In the field of politics, I think the issue is that often politicians are very good at making statements that are true but not helpful, or are a bit misleading. There's a situation where one side of the debate is yelling this not very useful number, people on the other side of the debate are rejecting it out of hand, and the fact-checker in the middle is in a tricky situation because the details are a bit complicated and weigh people down.
What if we want to be that fact-checker? What should we ask?
OK, so first, if you've been given a number, just ask: compared to what? What was it last year, what was it five years ago, what was it ten years ago? What was it under a previous government? What is it in other countries? Is it going up or down? Does it go up and down all the time? So just a bit of context.
The second thing is—what does the statistic actually refer to? If you see a claim that inequality is up, there are lots of things you could be unequal about, notably: wealth, income, consumption. It can be measured in different ways. You can measure the share of the top 1 percent, or you can measure something called the GE coefficient. There are lots of different ways of doing it, so just ask yourself when this claim is being made, what is actually being said.
A couple of weeks ago you looked at the Oxfam report on wealth, which said just sixty-two people have an equal level of wealth to the poorest half of the world. But you said wealth might not be a very useful figure here, because someone on a $1 a day actually has more wealth than a banker in millions of dollars of debt.
Wealth tells you a lot about the people with a lot of wealth, but not about people with not much wealth. Levels of income can tell you more, and there's an argument that says levels of consumption are even more informative. Let's say you're a banker, you have a good income, you decide to switch firms, and you're put on garden leave for a year, so you've got no income. How do you spot that, statistically? Well actually it's very easy: look at consumption. If someone's spending lots, then he or she is comfortably off, even if his or her actual income is low temporarily. So for that reason, the IFS, in a lot of its work, will look at consumption rather than income. There's no single right way to measure any of this, but there's always a gap between, for example, an academic trying to understand the world, and a campaigning organization that is not trying to understand the world but tell a message or sell their brand.
I feel like we should talk a little bit about the EU referendum, because it feels so much like a statistics referendum, particularly for the Remain campaign. Can you actually work out whether or not Britain will be worse off out of the EU?
I wouldn't take any particular number too seriously. For example, the Treasury says each household will be £4,300 [$6,300] better off if we stay in, and I question figures like that. The Leave campaign points to the fact that trade with China is growing by half, and it says the future is China. But hang on, context. How big is trade with China? It turns out that we still don't trade that much with China compared with the EU. It is increasing very rapidly, but our trade relationship with the EU is vastly bigger than with China. Which is not surprising, but just understanding that sort of thing helps you understand the decision involved.
Or EU membership fees—how much do we pay? How much is that per person? I lose track because I've debunked so many false claims. The answer is basically that it's not a huge number, but it's not small either. But you need to understand that there are a lot of situations where if we left the EU, we would still have to pay some of that money: Switzerland has to pay some of that money, and Norway has to pay some of that money. They're outside the EU, but they have to pay these fees, and that's for access to a single market. This sort of thing is not going to tell anybody which way to vote, but it provides some context, and I think context is very important.
Tim's latest book, The Undercover Economist Strikes Back, is out now.
"Statistics, Transparency, Company" via.