What is the anchoring effect, and how strong is the evidence?

The anchoring effect is the tendency for numerical estimates to move toward a number a person has just considered, even when that number is plainly irrelevant. People who first see a higher number give higher estimates than people who first see a lower one. Large meta-analyses find sizable average effects, though the effects vary a lot between studies, and a recent high-powered replication found a far smaller effect than the study it replicated. [1] [2] [3]

A review of the literature notes that the word has three meanings: an experimental procedure in which attention is directed to a number before a judgment, the empirical effect itself, which is treated as a cognitive bias, and the heuristic or mechanisms thought to explain it. The classic demonstration is by Tversky and Kahneman in 1974. They spun a rigged wheel of fortune that stopped on 10 or 65, then asked participants to estimate the number of African member states in the UN. Those shown 10 averaged an estimate of 25, and those shown 65 averaged 45. [1]

A 2026 meta-analysis in Management Science covered 2,601 effect sizes and found a large average effect (Hedges' g of 0.825) that stayed large after accounting for extensive publication bias. An earlier meta-analysis of 393 effects from 17,708 participants found g = 0.689, and described anchoring effects as among the largest and easiest to replicate in social psychology. [2] [4]

Both analyses also report heavy heterogeneity, and both find that some conditions shrink or remove the effect. Anchors communicated as randomly determined showed no effect in the earlier analysis. The newer one reports reduced or null effects for incidental anchoring, anchors from a different dimension, random numbers, incentives, debiasing interventions, and anchors lacking directional information. The two analyses differ on incentives: the earlier found they did not change the effect's size, while the newer lists them among conditions with reduced or null effects. [2] [4]

A 2025 paper in Economic Inquiry argues that anchoring studies, including replications, routinely have statistical power below 30%, which tends to exaggerate significant results. The authors replicated a field experiment on willingness to pay for gardening products that had found a 31% rise in bids after a high anchor. With power raised from 46% to 96%, they found an increase of 3.4%, with a confidence interval from about -3.4% to 10%, which excludes the original estimate. They conclude that effects are more likely to be modest, though still possibly meaningful given how cheap anchors are to use. [3] [5]

The positions therefore differ in emphasis. The meta-analysts hold that the overall literature shows a robust, large effect even after bias corrections. The replication authors hold that small-sample studies overstate magnitudes, at least in applied field settings. The sources do not show these views to be strictly incompatible: one concerns the average across many mostly laboratory studies, the other a single well-powered field replication. The results do not settle which view specialists as a whole favor. [2] [3]

Reviews say anchoring results from several cognitive processes rather than one. The earlier meta-analysis found that theories stressing rationality and nonlinearity were best supported overall, but it also noted that it was unclear which theory is best and that many findings on moderators were contradictory. [1] [4]