Skip to content

← Thinking Clearly

Shortcuts

Chapter 31 · The one source you haven't checked

In this chapter 5 sections
  1. What comes to mind
  2. A number in the air
  3. Saved or lost
  4. The better story
  5. Knowing isn’t enough

How tall is Mount Everest? In Nepal plenty of people could tell you, and since a joint survey by Nepal and China in 2020 the answer has been exact: 8,848.86 meters, which is a little over 29,000 feet.

Now suppose you didn’t know. Before you guess, you’re told one thing: Mount Everest is taller than 2,000 feet. That’s true, and it’s no help at all, since thousands of ordinary hills are taller than that. Then you guess. Someone in the next room is told something different: Mount Everest is shorter than 45,500 feet. Also true, also no help. Then they guess.

That’s what thousands of people did in Many Labs, the large test of psychology’s findings from Chapter 23, run in 36 places, at universities and online, and published in 2014. Most of them were Americans, who think of heights in feet. The people told “taller than 2,000 feet” guessed, on average, about 11,000 feet. The people told “shorter than 45,500” guessed about 34,000. A number that carried no information had pulled each group’s guess toward itself, and the two groups ended up more than 20,000 feet apart, which is most of the height of Everest itself.

Nobody in that test was stupid. They were doing what everybody does when a question is hard and an answer is wanted now: they used a shortcut. Psychologists call such shortcuts heuristics. In 1974 Amos Tversky and Daniel Kahneman described three of them in the journal Science, and they were careful to say that the shortcuts are “highly economical and usually effective”. Without them nobody could get through a day. The trouble is that each one also goes wrong in its own particular way, “systematic and predictable”, so that you can say in advance which way a guess will lean.

They opened with an example from seeing. How far away is that mountain? Part of the answer your eyes give you comes from how sharp it looks, since far things are usually hazier than near ones. That rule is mostly right. But on a hazy day it makes distances seem longer than they are, and on a very clear day shorter. By the same rule, the snow peaks seen from Kathmandu on a clear morning after rain should seem nearer than the same peaks through the haze. Judging how likely something is, they argued, runs on rules of the same kind.

Most claims come from somewhere else: a post, a study, an expert, a chart. This Part turns to the one source nobody checks, which is your own judgment. It sticks to mistakes you can catch in your own reasoning, and to the ones that have been tested hard. Psychology has had a rough time of repeating its own famous results, and several well-known findings didn’t survive. So the shortcuts here are the ones that came through, each with how well it held up.

What comes to mind

In the autumn of 1974, in Eugene, Oregon, a team led by Sarah Lichtenstein and Paul Slovic paid college students to compare causes of death in pairs. Which kills more people in the United States each year, and by how many times? They later asked members of the town’s League of Women Voters, chosen as some of the best-informed citizens around, to do the same.

Stroke killed about 85% more Americans than all accidents put together, car crashes, falls, drownings, fires and everything else. Only about one in five of the students, and about the same share of the league members, said stroke was the bigger killer. On average the students thought accidents were 25 times more common than stroke. The students judged tornadoes to kill more people than asthma, which killed 20 times as many, and accidents about as common a cause of death as disease, when disease killed about 15 times as many.

The pattern ran right through the results. Causes of death that are dramatic, that make the news, that are easy to picture, came out too high. Quiet ones, the diseases that kill people one at a time at home or in hospital, came out too low. The team traced it to how much people had seen and heard of each, and to how easily each came to mind.

That is the first shortcut, and it’s called availability: we judge how common something is by how easily examples come to mind. Most of the time that works, because common things usually are easier to recall. It goes wrong when something is easy to recall for another reason, because it was frightening, or recent, or on every screen.

Think of the Yeti Airlines plane that came down near Pokhara on 15 January 2023. All 72 people on board died, in Nepal’s worst air crash since 1992, and video from inside the cabin was widely shared. Now think of Nepal’s roads. In the year to mid-July 2026, Nepal Police counted 2,707 people killed in road accidents, about seven a day. At that rate the roads kill as many people as that crash every ten days or so. But road deaths mostly reach the news as a daily count, under headlines like “Seven killed, 105 injured in road accidents across Nepal in past 24 hours”. Which of the two comes to mind more easily? We found no study of how Nepalis judge these risks, so nobody can say how far their guesses lean. The point is only which way the shortcut would pull.

Availability’s record is the thinnest of the three in this chapter. As far as we could find, it has had no big many-labs repeat like the other two. The Oregon study was a careful series of experiments, and a later test of how people choose between two risks found that recalling the cases they knew fitted their choices well, though a simpler explanation, that people squeeze very big and very small numbers toward the middle, fitted about as well. But one of the most famous early demonstrations, about how often letters of the alphabet come first in English words, didn’t hold up well when other researchers measured it directly. So availability is best held as a likely reason for a well-measured pattern, not as a law.

A number in the air

Everest’s anchors were not the first. In the original 1974 paper, people watched a wheel of fortune spun in front of them, numbered 0 to 100. Then they were asked whether the share of African countries in the United Nations was higher or lower than the number that came up, and then to guess it. People who had seen the wheel stop at 10 guessed 25%, as the middle answer. People who had seen it stop at 65 guessed 45%. They knew the wheel was chance. It pulled them anyway, and paying people for accurate answers didn’t stop it.

Another of their tests can be done on a friend. Give someone five seconds to estimate 8 × 7 × 6 × 5 × 4 × 3 × 2 × 1. Give someone else five seconds for 1 × 2 × 3 × 4 × 5 × 6 × 7 × 8. It’s the same sum. With five seconds, people work out the first few steps and guess from there, and the first few steps of the descending version are bigger. In the 1974 test the middle guess for the descending version was 2,250, and for the ascending one 512. The right answer is 40,320. Both groups started too low and never moved far enough.

That is anchoring: a number you’ve just seen pulls your estimate toward it, even when you know the number means nothing. You start from it and adjust, and you don’t adjust enough.

Anchoring has one of the strongest records in psychology. In the 2014 test across 36 samples, all four anchoring questions, Everest among them, gave a clear effect in every single sample, and the effect came out bigger than the original study had suggested, for Everest more than twice as big. The team noticed the size varied from place to place and wondered whether that was partly because people in some countries simply knew more about Everest, or about the other questions. That was their guess, not something they tested.

Anchors are easy to imagine in daily life: the first price named at a shop, the “was Rs 5,000, now Rs 2,500” on a label, a number in a headline that sticks while the article that corrects it doesn’t.

Saved or lost

In 1981 Tversky and Kahneman gave students at Stanford University and the University of British Columbia a problem. A country is preparing for an unusual disease that is expected to kill 600 people. There are two plans. With Plan A, 200 people will be saved. With Plan B, there is a one-in-three chance that all 600 will be saved and a two-in-three chance that nobody will be. Which do you choose?

Of 152 students, 72% chose Plan A, the sure thing.

A second group got the same story with the plans described differently. With Plan C, 400 people will die. With Plan D, there is a one-in-three chance that nobody will die and a two-in-three chance that all 600 will. Of 155 students, 78% chose Plan D, the gamble.

Plans A and C are the same plan. Out of 600 people, 200 saved means 400 dead. Plans B and D are the same plan too. The only thing that changed was the wording, lives saved or lives lost, and it was enough to turn a large majority one way and then the other. Put in terms of gains, people played it safe. Put in terms of losses, they gambled. The authors found the same reversal among university teachers and among doctors.

That’s framing: the same facts, described as a gain or as a loss, lead to different choices. A medicine that “saves 90 patients in 100” and one that “lets 10 in 100 die” are the same medicine, and they don’t sound it.

The 2014 test included this problem too, leaving out the word “Asian” the original had used and putting each country’s own name in place of “the U.S.” The effect came through, significant in 86% of the samples, though at about half the size of the original. Then in 2020 a team led by Kai Ruggeri went back to the 1979 study in which Kahneman and Tversky had first laid out their theory of how people weigh gains and losses, a set of choices between sure and risky amounts of money. They asked 4,098 people in 19 countries and 13 languages, with only the money changed to local currencies. The results came out as before for 94% of the questions, a little weaker. Their verdict was that the foundations of the theory replicate “beyond any reasonable thresholds”.

The better story

Here is someone who never existed:

Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.

Which is more likely? That Linda is a bank teller? Or that Linda is a bank teller and is active in the feminist movement?

Most people who see this pick the second. It fits. The description sounds like someone who would be active in a cause, and the plain “bank teller” sounds like a stranger.

But it can’t be more likely. Every bank teller who is a feminist is a bank teller. However many feminist bank tellers there are, there must be at least as many bank tellers. Adding a detail can never make a story more probable. It can only keep it the same or make it less. Tversky and Kahneman called the mistake the conjunction fallacy, a conjunction being two things joined by “and”.

The shortcut behind it is the one their 1974 paper called representativeness: we judge how likely something is by how well it matches a picture in our heads. A detailed story that matches the picture feels likelier than a thin one that doesn’t, even when the thin one includes it. That has a cost well beyond Linda. A rumor with names, a place and a time feels truer than a vague one. A plan that needs ten things to go right sounds solid because each step does. In the 1974 paper they pointed out that when each step of a project is very likely, the chance of all of them happening can still be low, which is one reason so many plans run late.

Linda’s record is a story of its own. Some researchers argued that people weren’t making a logic mistake at all: they were reading “and” in its everyday sense, the way “friends and colleagues” can mean either. Ralph Hertwig, who held that view, and Kahneman, who didn’t, agreed to settle it together. They designed the experiments jointly, with a third psychologist, Barbara Mellers, as referee, and published the result in 2001, still disagreeing about what it meant. Neither side got everything. Asking “of 100 people like Linda, how many are…?” didn’t make the mistake vanish on its own, and reading “and” as “or” turned out to be at most one part of it. But with some wordings and set-ups the mistake disappeared. In 2021 a team ran a registered repeat of that work with 1,032 people online. For Linda, the mistake came back clearly, and the “and” explanation again found no support. For a second description used in the same study, about a man called James, it didn’t appear at all. So the effect is real, and it’s stronger for some stories than others.

Knowing isn’t enough

The Optical Illusions pages on this site have the Müller-Lyer lines: lines of the same length, some with fins pointing out and some with arrowheads. Measure them and you know they’re equal. Look again and one still looks longer. Knowing the answer doesn’t change what you see.

The shortcuts behave like that. In 1981 Tversky and Kahneman compared framing to walking in the mountains, where which of two peaks looks taller changes as you move. The Oregon team tried the obvious cure. They gave one group of 30 students a note before they started, explaining that people misjudge these risks, that what comes easily to mind and what makes the news can lead them astray, and asking them not to let it. That group did no better than a group of 22 who got no note. A warning is a fact, and the shortcut isn’t listening to facts.

What does help is a habit, something you do with the question in front of you, not something you know. The evidence here is thinner than for the shortcuts themselves, but it points one way.

For availability, the habit is to ask where the number is. “It feels common” is a report on your memory. The count of stroke deaths or road deaths is a report on the world, and it’s often a search away.

For anchors, the habit is to argue against the number. In a 2000 study with experts as the people tested, Thomas Mussweiler and his colleagues found that listing reasons why a given number was wrong made it pull less. It reduced the effect rather than removing it.

For framing, the habit is to turn the sentence around. If it says “saves 200”, say “400 die” to yourself, and see whether you still feel the same. If the two versions pull you in different directions, neither feeling was about the facts.

For the better story, the habit is to put the plain version beside the detailed one and ask which includes the other. In one of Kahneman and Tversky’s tests, people estimated how many of a group of women like Linda were bank tellers, and how many were feminist bank tellers. Most of those asked both at once kept to the rule. Those asked only one of the two, in separate groups, broke it: on average they put the feminist bank tellers higher.

Can this be trained? In 2015 Carey Morewedge and his colleagues gave people a single session, either a computer game with feedback or a video, about anchoring, representativeness and another bias. Mistakes fell by roughly a fifth to a third right away, and most of the gain was still there two months later. It’s one team’s work, but it suggests the habits can be learned, and that practice with feedback works better than being told.

That mountain comparison of 1981 ends with a thought worth keeping. A traveler who notices that which peak looks taller changes as he walks will conclude that some of his impressions must be wrong, even with no map to tell him which. Your judgment is like that. You can’t step outside it to see the true height of things. But you can notice when the same question, asked two ways, gets two answers from you, and take that as a sign to stop and count.

Found a mistake? Please tell us.