Anchoring Effect

A useful analogy might be drawn between the anchoring effect and the Stroop effect. In the classic Stroop effect, subjects are asked to name the ink color in which a word is printed. Reaction times are longer for color words that do not match the ink color (e.g., the word red printed in green ink) than for unrelated words. The meaning of the word is irrelevant information, yet it influences performance. In a similar fashion, an irrelevant anchor influences judgment, even when decision makers are instructed to ignore it. Like the Stroop effect, anchoring appears to be both prevalent and robust. The contaminating effects of irrelevant anchors can be observed in numerous real-world contexts: purchase offers are swayed by listing prices, estimates of uncertain quantities are pulled toward arbitrary values, and judgments of risk or self-efficacy are shaped by numbers that happen to be salient. Understanding the causes of anchoring and what they tell us about the efficacy of various potential debiasing techniques is thus of considerable practical importance.

Putting Adjustment Back in the Anchoring and Adjustment Heuristic

The Standard Paradigm and Its Robustness

In what year was George Washington elected President? What is the freezing point of vodka? Few people know the answers to these questions, but most can arrive at a reasonable estimate by tinkering with a value they know is wrong. Most know that the United States declared its independence in 1776, so Washington must have been elected sometime after that. And most people also know that alcohol freezes at a lower temperature than water, so vodka must freeze at something colder than 32°F. To answer questions like these, in other words, people may spontaneously anchor on information that readily comes to mind and adjust their responses in a direction that seems appropriate, using what Tversky and Kahneman (1974) called the anchoring and adjustment heuristic. Although this heuristic is often helpful, adjustments tend to be insufficient, leaving people’s final estimates biased toward the initial anchor value.

To examine this heuristic, Tversky and Kahneman developed a paradigm in which participants are given an irrelevant number and asked if the answer to a question is greater or less than that value. After this comparative assessment, participants provide an absolute answer. Countless experiments have shown that people’s absolute answers are influenced by the initial comparison with the irrelevant anchor. In one famous demonstration, a rigged wheel of fortune that stopped only at 10 or 65 was spun before participants were asked to estimate the percentage of African nations in the UN. Even though the wheel was obviously random and uninformative, the average estimates were 25% and 45% for those who saw 10 and 65, respectively. Similarly, people estimate that Gandhi lived to be roughly 67 years old if they first decided whether he died before or after the age of 140, but only 50 years old if they first decided whether he died before or after the age of 9.

Anchoring effects have been demonstrated in numerous contexts—the evaluation of gambles, estimates of risk and uncertainty, perceptions of self-efficacy, anticipations of future performance, and answers to general knowledge questions. Anchoring and adjustment has also figured prominently as an explanatory mechanism underlying such diverse phenomena as preference reversals, probability estimates, trait inference, language production and comprehension, and various egocentric biases such as the “spotlight effect” and the “illusion of transparency.”

Beyond the Standard Paradigm: Two Mechanisms for Two Systems

Anchoring effects have traditionally been interpreted as a result of insufficient adjustment from an irrelevant value, but subsequent research uncovered a more nuanced picture. In the standard paradigm, the attempt to answer the comparative question—say, whether Gandhi lived to be 140—leads an individual to test the hypothesis that the anchor value is correct. Because people typically evaluate hypotheses by attempting to confirm them, this search disproportionately activates anchor-consistent information. The absolute judgment is then biased by the evidence recruited in this confirmatory search. This alternative account, often called the selective accessibility model, emphasizes biased retrieval of target features rather than insufficient adjustment from an initial anchor.

This account distinguishes two mechanisms associated with the two broad modes of thinking: a deliberate process of adjustment (an operation of System 2) and an automatic priming effect (an operation of System 1). One important variable is the source of the anchor. In the standard paradigm, anchors are provided by an experimenter or other external source and must be taken seriously, if only for a moment, to answer the comparative question. This gives rise to selective accessibility and priming, where the mere consideration of a number automatically activates compatible information in memory.

Outside the laboratory, however, anchors are often self-generated rather than externally provided. Most people know George Washington was elected after 1776, but how long after? Or that vodka freezes at less than 32°F, but how much less? Self-generated anchors are known from the beginning to be wrong. Because the person already knows these anchors are wrong, testing their correctness is not the proposed explanation here. Instead, people start from the self-generated value and deliberately adjust away from it, but they typically stop too soon, leaving the final estimate biased toward the initial anchor. Thus, insufficient adjustment is the proposed explanation for these self-generated anchoring effects.

Conditions That Moderate Anchoring

Anchoring effects are remarkably robust across a wide range of conditions. They occur even when the anchor is extreme and implausible, when respondents are unaware of the effect, when they have been explicitly warned to avoid it, and when they are motivated to be accurate. The effect is most powerful when the anchor has received sufficient attention and when it is relevant to the target judgment—that is, when it is expressed on the same response scale and represents the same underlying dimension, thereby comprising a potential answer to the target question.