The question
In the early 1970s, Amos Tversky and Daniel Kahneman were cataloguing the mental shortcuts people use when they have to guess. One of them they called “adjustment and anchoring”: we often start from some number that is at hand and adjust from it to reach an answer. Their claim was that the adjustment usually stops too soon, so the final answer stays too close to where we started.
That would not matter much if the starting number were always sensible. The sharper question was: what if the starting number is obviously meaningless? Would it still drag the answer towards it?
How it was done
Tversky and Kahneman described two demonstrations in their 1974 article in Science.
The wheel of fortune. Participants were asked to estimate various quantities given as percentages, for example the percentage of African countries in the United Nations. Before each estimate, a wheel marked from 0 to 100 was spun in front of them. They first said whether the true value was higher or lower than the number on the wheel, and then gave their own estimate. Different groups got different numbers; the article reports the results for two of them, 10 and 65. In his 2011 book Thinking, Fast and Slow, Kahneman later explained that the wheel had been fixed so that it could only stop on those two numbers.
The multiplication task. Two groups of high-school students had five seconds to estimate a product. One group saw 8×7×6×5×4×3×2×1; the other saw the same numbers in the opposite order, 1×2×3×4×5×6×7×8. Five seconds is too short to finish, so people do a few steps and guess the rest. The prediction was that the first steps would act as a starting point.
The article does not say how many people took part or where the studies were run.
What came out
The wheel moved the estimates. The group that saw 10 gave a median estimate of 25%. The group that saw 65 gave 45%. A number everyone had just watched come from a wheel shifted the typical answer by 20 percentage points. Offering money for accurate answers did not make the effect go away.
The order of the numbers mattered. For the sequence starting with 8, the median estimate was 2,250. For the sequence starting with 1, it was 512. Both were far too low: the true answer is 40,320. The group that began with the bigger numbers guessed higher, and everyone underestimated, which is what you would expect if people stopped adjusting too early.
Tversky and Kahneman also argued that anchoring helps explain other errors, such as why people overestimate the chance that a chain of likely events will all happen, and why the ranges people give when they feel highly confident are usually too narrow.
What really happened (replications and critiques)
Why it works. Tversky and Kahneman’s account was insufficient adjustment. In 1997 Strack and Mussweiler proposed a different mechanism, which they called selective accessibility. When you ask yourself whether a value could be as high as the anchor, you search your memory for facts that would fit that value. Those facts are then easier to recall when you make your own estimate. In their first study, with 32 students at the University of Würzburg, people compared a landmark’s height or width with a high or low anchor. The anchor moved their later estimate much more when it was about the same dimension (height after a height comparison) than when it was about the other one. Simple adjustment from a number does not easily explain that difference. In another study, anchors that were plainly absurd, such as comparing Gandhi’s age at death with 140 or with 9, still pulled the estimates.
The classic effect replicates very strongly. In 2014, the first Many Labs project re-ran 13 classic findings in 36 samples with 6,344 participants, in the United States and nine other countries. Its anchoring task came from a 1995 study by Jacowitz and Kahneman: people compared a quantity, such as the height of Mount Everest or the population of Chicago, with a high or low number and then gave their own estimate. All four anchoring questions showed the effect in every one of the 36 samples. They were among the largest effects of the whole project, including the two largest, between about 1.2 and 2.4 standard deviations depending on the question, larger than the original study’s overall estimate of 0.93. The size varied between sites, possibly because people knew the facts to different degrees.
Prices are a harder case. In 2003, Ariely, Loewenstein and Prelec asked 55 MBA students at MIT whether they would buy products such as a cordless keyboard or a bottle of wine for a price equal to the last two digits of their social security number. Then they stated, with real money at stake, the most they would pay. Students in the top fifth of social security numbers offered roughly two to three and a half times as much as those in the bottom fifth: on average $55.64 for the keyboard versus $16.09. Yet almost everyone still valued the better wine above the cheaper one. The authors called this mix of arbitrary levels and consistent ordering “coherent arbitrariness”.
Later work did not settle the matter:
- In 2012, Drew Fudenberg, David Levine and Zacharias Maniadis found only very weak anchoring for everyday goods and none for simple lotteries.
- In 2014, Maniadis, Fabio Tufano and John List re-ran a different experiment from the 2003 paper, in which people named the price at which they would agree to hear an unpleasant sound. With 116 students in Bologna, the effect was about half the original and not statistically significant. They argued that a single striking result should not be trusted before independent replications.
- The Data Colada authors (Uri Simonsohn, Leif Nelson and Joseph Simmons) replied that this replication was too imprecise to rule out a sizable effect, partly because nearly half of the answers were 10 cents or less, so they could hardly fall any lower.
- In 2019, Yoon, Fong and Dimoka ran three new studies and pooled earlier ones. They found that anchoring of willingness to pay was robust, but smaller than in 2003.
- In 2020, Ioannidis, Offerman and Sloof found no anchoring at all when 316 participants rolled a die to set a random selling price for a bottle of wine.
The fair summary: random anchors probably do affect what people will pay, but the 2003 effect sizes look like the high end, and some studies find nothing.
Incidental vs. informative anchors. In all the studies above, people are asked to think about the anchor, usually by comparing it with the answer. A bolder claim is that numbers merely present in the surroundings also anchor us. In 2008, Critcher and Gilovich reported, among other things, that people predicted different sales for a new phone depending on whether it was called the “P97” or the “P17”, and were willing to pay more at a restaurant called “Studio 97” than at one called “Studio 17”. These results did not hold up. Many Labs 2 (2018) re-ran the phone study with more than 6,800 participants and found an effect close to zero. In 2020, Shanks and colleagues ran three studies with 1,291 participants, including a large preregistered one, and found no reliable incidental anchoring. In the same participants, classic comparison anchoring worked strongly. Their conclusion was that people probably have to actively think about a number for it to anchor them.
There is a further distinction. In real life, many anchors are not meaningless: a list price or an expert’s first offer usually does carry some information, so being influenced by it is not always a mistake. What made the 1974 studies striking was that the anchor was openly uninformative.
Ethics
The wheel was presented as random, while, by Kahneman’s later account, it could only land on two numbers. That is a mild deception. Otherwise the tasks were harmless: guessing percentages and multiplying numbers. No distress was reported, and the 1974 article does not describe a debriefing. A study like this would easily pass an ethics review today, with consent and a short explanation afterwards.
What it means today
Anchoring on estimates is one of the most reliable findings in psychology. Anchoring by numbers you only glimpsed is not. That makes the practical lesson narrower, and more useful, than it is often presented.
The following habits go beyond what these experiments tested, but they follow from them:
- Before you look at someone else’s figure (a price, a salary offer, a forecast), write down your own estimate first.
- When a number has been put in front of you, ask whether it carries real information or simply happened to come first.
- In a group, collect independent estimates before anyone says a number out loud.
Would you have passed?
Coming soon. Spin the wheel, say whether the answer is higher or lower, then make your estimate. At the end you can compare your answers with those of people who got the other number, and with the 25% and 45% medians from 1974. A caveat: once you know what the experiment is about, you may correct for it, so your result will not show how strong the effect is on people who don’t know.
Questions people ask
What is the anchoring effect?
It is the tendency for an estimate to stay close to a number you considered just before, even when that number is arbitrary. In Tversky and Kahneman's 1974 example, people who first saw 10 on a wheel of fortune gave a median estimate of 25% for the share of African countries in the UN; people who saw 65 gave 45%.
Has the anchoring effect been replicated?
The classic version, where people first compare a quantity with a high or low number and then estimate it, has been replicated many times. In Many Labs 1 (2014), it appeared in all 36 samples and produced some of the largest effects of the 13 findings tested.
Did the social security number experiment replicate?
Partly. Several studies found anchoring of willingness to pay, but usually much weaker than in Ariely, Loewenstein and Prelec (2003). Others found very weak or no effects. Researchers still disagree on how large the real effect is.
Does a random number in the background, like a price on a nearby sign, anchor people?
The evidence says probably not, or only very weakly. Large replications in 2018 and 2020 found no reliable effect of incidental numbers that people were not asked to think about, while the classic comparison-based anchoring worked well in the same participants.
Sources
- Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. doi:10.1126/science.185.4157.1124
- Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.
- Strack, F., & Mussweiler, T. (1997). Explaining the enigmatic anchoring effect: Mechanisms of selective accessibility. Journal of Personality and Social Psychology, 73(3), 437–446. doi:10.1037/0022-3514.73.3.437
- Ariely, D., Loewenstein, G., & Prelec, D. (2003). "Coherent arbitrariness": Stable demand curves without stable preferences. The Quarterly Journal of Economics, 118(1), 73–106. doi:10.1162/00335530360535153
- Fudenberg, D., Levine, D. K., & Maniadis, Z. (2012). On the robustness of anchoring effects in WTP and WTA experiments. American Economic Journal: Microeconomics, 4(2), 131–145. doi:10.1257/mic.4.2.131
- Maniadis, Z., Tufano, F., & List, J. A. (2014). One swallow doesn't make a summer: New evidence on anchoring effects. American Economic Review, 104(1), 277–290. doi:10.1257/aer.104.1.277
- Simonsohn, U., Simmons, J., & Nelson, L. (2013, November 11). [7] Forthcoming in the American Economic Review: A misdiagnosed failure-to-replicate. Data Colada. datacolada.org/7
- Yoon, S., Fong, N. M., & Dimoka, A. (2019). The robustness of anchoring effects on preferential judgments. Judgment and Decision Making, 14(4), 470–487. dlab.sauder.ubc.ca/sjdm/journal/19/190426/jdm190426.html
- Ioannidis, K., Offerman, T., & Sloof, R. (2020). On the effect of anchoring on valuations when the anchor is transparently uninformative. Journal of the Economic Science Association, 6(1), 77–94. doi:10.1007/s40881-020-00094-1
- Jacowitz, K. E., & Kahneman, D. (1995). Measures of anchoring in estimation tasks. Personality and Social Psychology Bulletin, 21(11), 1161–1166. doi:10.1177/01461672952111004
- Klein, R. A., Ratliff, K. A., Vianello, M., et al. (2014). Investigating variation in replicability: A "many labs" replication project. Social Psychology, 45(3), 142–152. doi:10.1027/1864-9335/a000178
- Critcher, C. R., & Gilovich, T. (2008). Incidental environmental anchors. Journal of Behavioral Decision Making, 21(3), 241–251. doi:10.1002/bdm.586
- Klein, R. A., Vianello, M., Hasselman, F., et al. (2018). Many Labs 2: Investigating variation in replicability across samples and settings. Advances in Methods and Practices in Psychological Science, 1(4), 443–490. doi:10.1177/2515245918810225
- Shanks, D. R., Barbieri-Hermitte, P., & Vadillo, M. A. (2020). Do incidental environmental anchors bias consumers' price estimations? Collabra: Psychology, 6(1), 19. doi:10.1525/collabra.310