A study tries to count how many people in one region fell ill from dirty air over five years. It cannot count exactly. So it gives a range: somewhere between 8,000 and 34,000, most likely about 20,000. These numbers are made up. The shape is what every careful study looks like.
5 of 5
words that say how to read the number, still attached
The range is the honest part. Scientists draw it as a short line either side of the number and call it error bars. Then the number leaves the study. A press release keeps the top figure. A headline drops 'up to'. A politician rounds it. Each person who passes it on drops one word.
Low end8,000
Most likely20,000
High end34,000
The study says 8,000 to 34,000. The headline says 'up to 34,000'. True?
True. That is the finding.
Not yet. The finding is the range and the middle. A range reported as its top is a different claim with the same digits.
False. The study did not say 34,000.
Not yet. It did, as the top of a range. The headline did not invent the number. It dropped the lower numbers that came with it.
True and misleading: the top of a range, with the range removed.
Right. Every word in the headline is in the study. The words that are not in the headline are the ones that said how unsure the study was.
Put the hardening in order.
Tap them in the order they happen, first to last. Each tap gets a number.
A range→The top figure→A number→A fact→A citation
Right. Five honest steps, each dropping one word, and a range becomes a fact with a citation. A citation is a note saying where a number came from. Nobody in the chain saw more than one step of it.
'Up to 34,000'easy to say
vsvs
'8,000 to 34,000, most likely 20,000'a mouthful
Nobody in the chain lies. Each person keeps the part that is easiest to say. A top figure is easy. A range is a mouthful. 'Up to' sounds unsure, and unsure is dull. So the words that said how sure the number was go first. The number itself is never dropped.
The study's figure, retold. Change how many times it has been passed on.
5 of 5
words that say how to read the number, still attached
Retellings: The study
8,000 to 34,000, most likely 20,000, in one region, over five years.
4 of 5
words that say how to read the number, still attached
Retellings: The press release
Up to 34,000. The most likely figure is gone.
3 of 5
words that say how to read the number, still attached
Retellings: The headline
34,000. 'Up to' is gone.
1 of 5
words that say how to read the number, still attached
Retellings: A politician
30,000, rounded, and now said as a fact.
0 of 5
words that say how to read the number, still attached
Retellings: A second study cites it
30,000, with a citation. Nobody can get the range back from here.
What goes first?
The source.
Not yet. The source goes last, and often stays, which is the problem. A hardened number with a citation looks like a checked one.
The number itself.
Not yet. The number is the one thing that survives. The digits travel. The words that said how to read them are what get dropped.
The low end of the range and the 'most likely' figure.
Right. The bottom goes first, every time. A top figure makes a headline; a bottom figure does not. Then 'up to' goes, then the source.
Move the control.
Cases in 1,000 people without the drug2 cases
Cases in 1,000 people with the drug1 case
A different word can go missing: the 'in a thousand'. A drug is tested. Without it, two people in every thousand get a disease. With it, one in every thousand. The headline turns that change into a per cent.
The drug cuts the disease from two cases in every thousand people to one. The headline says risk fell by some per cent. Slide to the number the headline uses.
Slide the marker to where you think the answer is, then press Lock. Nothing is scored.
your guess · 0%
the answer · 50%
0%50%100%
Your guess: —
Halved, and the arithmetic is true: two in a thousand down to one in a thousand. The number that went missing is the thousand. For any one person the change is a tenth of one per cent. Most guesses land down there. That small number is the honest one for a single person.
Cases in 1,000 people without the drug2 cases
Cases in 1,000 people with the drug1 case
Four headlines for the same finding. Which one tells a reader what changed for them?
Biggest breakthrough in a decade
Not yet. A claim about the decade, not about the disease. There is no number in it at all.
One fewer case in every 1,000 people
Right. The only version with the thousand in it. A reader can stand among the thousand and see what changed. For 999 of them, nothing.
Risk halved
Not yet. True, and it does not say what was halved. Two in a thousand halved is one in a thousand. Two in ten halved would be a different story with the same word.
Risk down 50%
Not yet. The same as 'halved' with a number on it, and the same missing thousand.
Cases in 1,000 people without the drug2 cases
Cases in 1,000 people with the drug1 case
From two in a thousand to one in a thousand. How many people must take the drug for ONE of them to avoid the disease?
people
Right. A thousand. Both numbers are true: the risk halved, and a thousand people take the drug for one avoided case. The headline picked the number that sounds bigger.
The studya range
↓→
The last person to repeat ita number
↓→
A citationpoints here, not at the study
By the time a number carries a citation, it looks checked. But the citation points at the last person who repeated it, not at the study. So a hardened number can travel for years. Nobody along the way lied. Each of them dropped one word.
Cameras in the 2002 estimate, repeated for years4,200,000
Cameras in a 2011 police count1,850,000
For years, news stories said Britain had 4.2 million security cameras, one for every fourteen people. The figure came from a 2002 study that counted the cameras on one shopping street and multiplied up. What kind of number is 4.2 million?
A made-up number. The study lied.
Not yet. The study did not lie. It said what it did and how rough the result was. The people who repeated the number dropped the 'roughly'. Each step honest, the end result wrong.
Still the best figure there is.
Not yet. In 2011 a police force made a careful count and got about 1.85 million. The 4.2 million kept being printed for years afterwards, because a figure with a citation looks checked.
A rough estimate from one street, repeated until it read like a count.
Right. One street, multiplied up, then printed until it had a citation. When a police force made a careful count in 2011, it came to about 1.85 million. The 4.2 million kept being printed anyway.
A count. Someone counted 4.2 million cameras.
Not yet. Nobody counted 4.2 million cameras. One street was counted and the rest was multiplication. A number repeated for ten years starts to read like a count, and that is the hardening.
Lesson complete
A number hardens as it travels. Every retelling drops a word that said how to read it.