The Anchoring Effect: How the First Number You Hear Hijacks Every Decision After.

In 1974, Amos Tversky and Daniel Kahneman spun a wheel of fortune in front of their research participants. The wheel was rigged to stop on either 10 or 65. Then they asked a simple question: is the percentage of African countries in the United Nations higher or lower than the number on the wheel? And what is your best estimate?^[1] The groups that saw 10 guessed a median of 25 per cent. The groups that saw 65 guessed 45. The correct answer was irrelevant to the demonstration. What mattered was that a random number, visibly generated by a carnival prop, had moved estimates by 20 percentage points. Nobody believed the wheel contained useful information. Nobody could explain why their answer had changed. But the number got in anyway.

That experiment launched a field. In the five decades since, anchoring bias (the tendency for an initial numerical reference to systematically distort subsequent judgment) has been replicated more consistently than almost any other finding in cognitive science.^[21][35] The Open Anchoring Quest dataset, aggregating 96 independent studies and 88,914 individual trials across Europe, Asia, North America, and South America, confirmed that the effect is not a Western laboratory curiosity but a species-level pattern.^[34] A comprehensive literature review by Furnham and Boo identified it as one of the most reliable cognitive biases in the psychological canon.^[21] The meta-analytic effect size across the entire literature is Hedges' g = 0.825, surviving extensive publication-bias correction, though the heterogeneity index (I² = 93.73%) warns that the "typical" effect varies enormously by condition, paradigm, and stakes.^[35]

The anchoring bias science reveals something uncomfortable about how numerical reasoning actually works. The mind does not evaluate numbers from a blank slate. It evaluates them relative to whatever number arrived first, then adjusts insufficiently, if it adjusts at all.^[3][9]

What makes anchoring remarkable is not merely its size but its resistance to the things that should eliminate it. Intelligence does not reliably protect against it: the correlation between cognitive ability and anchoring susceptibility is small and inconsistent.^[19] Forewarning helps only when the anchor is self-generated. For externally provided anchors (the kind encountered in negotiations, price tags, and courtroom arguments) knowing about the bias does not appreciably reduce it.^[16]

This is not a quirk that educated people can simply think their way out of. Anchoring is a structural property of how the brain constructs numerical estimates under uncertainty. Domain knowledge constrains the range of plausible responses but does not eliminate the effect. Experts show smaller but still significant anchoring.^[26] The question is not whether it affects you. The question is how much, and what can be done about it.

Kahneman framed this in terms of the mind's broader architecture. The anchor activates what he called WYSIATI (What You See Is All There Is): the principle that System 1 thinking builds its story from whatever information is immediately available, then treats that story as the whole story.^[13] When the first number you encounter is high, the information that becomes available in working memory is anchor-consistent. When it's low, so is the retrieval. The anchor doesn't persuade you. It curates your evidence.

The anchoring bias science has resolved a debate that ran for nearly three decades. Tversky and Kahneman originally proposed that anchoring worked through a simple mechanism: people start at the anchor, adjust away from it, and stop too soon.^[1] The adjustment was real but insufficient: a lazy search that terminated at the first plausible value rather than the correct one.^[9] That was one pathway. It turned out to be only half the story.

In 1997, Fritz Strack and Thomas Mussweiler proposed a second mechanism that would prove even more powerful. When you are asked whether a car is worth more or less than $50,000, your mind does not simply start at $50,000 and adjust. It tests the hypothesis that the car is worth $50,000, and in doing so it preferentially retrieves knowledge consistent with that hypothesis.^[2] Features that support the anchor become cognitively accessible. Features that contradict it stay dormant.

This is not a metaphor. Strack and Mussweiler demonstrated experimentally that anchor-consistent information became chronically accessible even after the comparison question was over. Wilson and colleagues confirmed that anchors primed numeric accessibility even without an explicit comparison task, though knowledge reduced but did not eliminate the effect.^[5] Chapman and Johnson showed that anchors increase the accessibility of features that anchor and target share. Prompting consideration of different features reduced anchoring, but only partially.^[4] The anchor had changed the contents of working memory.

The two mechanisms, anchoring-and-adjustment and selective accessibility, are not competing explanations. They operate under different conditions. Epley and Gilovich showed in 2001 that adjustment-based anchoring occurs specifically when the anchor is self-generated: you know the freezing point of water is 0°C, so when asked the freezing point of vodka, you start at 0 and adjust downward.^[3] The adjustment is effortful, serial, and terminates when it reaches the first plausible value. For externally provided anchors (the kind that arrive from price tags, prosecutor requests, and first offers) adjustment is largely absent. Instead, selective accessibility dominates: the external number changes what information your brain retrieves, and you build your estimate from that contaminated retrieval.^[3]

This distinction has practical consequences. Head-nodding (a motor acceptance cue) reduced adjustment from self-generated anchors, while head-shaking increased it, suggesting that the adjustment process is partially embodied and responsive to physical cues of acceptance and rejection.^[3] For externally provided anchors, no such motor override was found. The selective accessibility mechanism is harder to interrupt because it operates before the conscious estimation even begins. Chapman and Johnson framed this as the distinction between anchoring in belief judgments (factual estimates) and value judgments (preferences). Anchors affect both, but through different accessibility channels.^[10]

Mussweiler extended this in 2003 with a comprehensive model of comparison-based judgment. The key insight: when the comparison triggers similarity testing, the target assimilates toward the anchor. When it triggers dissimilarity testing, the target contrasts away from it.^[17] The default is similarity testing, which is why anchoring effects are predominantly assimilative.