VI
Mark học AIAnchoring · Amos Tversky and Daniel Kahneman, Science, 1974
When we estimate a number, does the number we just saw drag us along, even when we know it means nothing? The answer after 50 years: yes, quite reliably. And at the negotiating table, the first number often wins.
A wheel numbered 0 to 100, spun in front of participants, then two questions.
The two explanations do not exclude each other. Both pull the answer toward the anchor.
At the negotiating table, numbers always look relevant. That is why anchoring there stays strong.
10 common claims about anchoring, checked against current research. Click a row to see why.
The original experiment (Tversky and Kahneman, 1974) and hundreds of studies since. Many Labs (Klein et al., 2014) redid four anchoring questions in 36 samples: all four replicated, among the clearest effects of the project.
Many Labs had 6,344 people. The authors found that setting (lab or online, US or elsewhere) mostly did not change effect size. Note: Many Labs used fixed anchors in comparison questions, not a wheel.
Tversky and Kahneman (1974) report that paying for accuracy did not reduce the effect. Wilson et al. (1996) found that neither an incentive nor detailed warnings about anchoring made the effect smaller.
Estate agents' valuations followed the listing price they were given (Northcraft and Neale, 1987). Experienced judges were pulled by a number from dice when proposing a sentence (Englich, Mussweiler and Strack, 2006). Individual samples are not large, but many studies point the same way.
Galinsky and Mussweiler (2001), 3 experiments. A meta-analysis of 90 studies, 16,334 people (Petrowsky et al., 2025): there is a first-mover advantage, and first-offer level correlates 0.62 with the final price. Two preregistered experiments with 2,121 people replicated this.
Ariely et al. (2003) results were much weaker on replication: very weak anchoring for goods, none for lotteries (Fudenberg, Levine and Maniadis, 2012); the unpleasant-sounds experiment gave a weak, non-significant effect (Maniadis, Tufano and List, 2014). A meta-analysis of 53 price studies (Li, Maniadis and Sedikides, 2021): non-random anchors pull harder.
Petrowsky et al. (2025): ambitious first offers give better final prices, but mean fewer deals and less satisfied recipients. The advantage depends on the deal happening at all.
In the same meta-analysis, first-offer effects shrink when many issues are negotiated at once. If you lack market information, going first can anchor yourself low; this is a rule of thumb from the Harvard Program on Negotiation, not a measured number.
Mason et al. (2013): recipients of precise offers counter closer to them. Leib et al. (2022) still found the effect but confirmed only part of the mechanism. A small effect, use it as a minor tip.
Wilson et al. (1996): warnings, whatever their form or timing, did not remove the anchor. What works is a concrete step: think of reasons for the opposite (Mussweiler, Strack and Pfeiffer, 2000), or focus on your own target, the other side's reservation price and their alternatives (Galinsky and Mussweiler, 2001).
Robustness 8/10. One of the most robust effects in psychology. Be careful only with "random numbers pull the value of everything" stories.
In psychology's "replication crisis", many famous effects shrank or vanished when redone with large samples. Anchoring is different: in Many Labs, the anchoring questions gave clear results at almost every site.
One likely reason is that the effect is large and easy to measure: all you need is a number and a question. The more surprising versions, like a social security number pulling the price of a bottle of wine, depend a lot on method and are smaller on replication.
The lesson for readers: trust the basic effect, be wary of too-good stories built on it.
Work skill: salary talks, quotes, budget sign-off. Who names the number first, and how to handle someone else's anchor.
Every number in this page, the Facebook carousel and the captions comes from the studies below.