Introduction
The "anchoring bias"—our tendency to stay too close to an initial starting value—is often cited as a classic example of human irrationality. However, recent research suggests this bias may actually be a "resource-rational" strategy. By modeling the mind as a bounded agent that trades off the cost of thinking against the cost of error, we can see that insufficient adjustment is often the mathematically optimal way to use limited cognitive resources.
The Problem: Heuristics vs. Rationality
For decades, the "Heuristics and Biases" program has suggested that human judgment systematically violates logic and probability. The anchoring-and-adjustment heuristic is a prime example: when estimating a value, people start from an anchor and adjust away, but they almost always stop too soon. This paper addresses the tension between human intelligence in complex tasks and apparent irrationality in simple experiments, proposing that standard definitions of rationality ignore the finite nature of brainpower.
Methods: Resource-Rational Analysis
The researchers applied Resource-Rational Analysis to evaluate whether cognitive processes are optimal given hardware constraints.
- The Model: The "adjustment" phase is modeled as a Markov Chain Monte Carlo (MCMC) process. The mind "searches" for the correct value by taking stochastic steps away from the anchor.
- The Trade-off: A cost function is defined: $Total Cost = Expected Error + (Time Cost \times Number of Steps)$.
- Optimization: Solving for the optimal number of adjustment steps ($t^*$) reveals that because MCMC algorithms exhibit geometric convergence, a rational agent should stop adjusting long before reaching the true mean of their internal belief distribution.
Key Insights: The Logic of "Close Enough"
A central finding is the diminishing returns of thinking. While the cost of computation is linear, the benefit is sub-linear. If an estimate is already "close enough," further effort to move the needle by a small percentage is a waste of resources.
Furthermore, the model explains why cognitive load or intoxication increases anchoring: these states increase the "cost per adjustment," making it rational to stop even sooner. It also resolves the "Uncertainty Paradox," showing why financial incentives help for self-generated anchors but not provided ones; high uncertainty creates a flat internal "landscape" where adjustments are less effective.
Key Results
The resource-rational model provides a unifying explanation for ten different anchoring phenomena:
- Quantitative Fit: It matches the empirical "Anchoring Index" of approximately 50% found in literature.
- Linearity: It predicts that bias increases linearly with the anchor's distance from the true value.
- Motivation Effects: It accurately simulates how incentives and forewarnings interact with different anchor types.
Limitations and Future Work
The current model focuses on the adjustment process and assumes the anchor is either provided or easily retrieved. Future research should explore how the mind selects anchors and whether it uses parallel "population-based" sampling. Additionally, integrating social cues—where anchors may be cooperative hints—is a promising next step for ecological validity.
Reference: Lieder, F., Griffiths, T. L., Huys, Q. J. M., & Goodman, N. D. (2018). The anchoring bias reflects rational use of cognitive resources. Psychonomic Bulletin & Review, 25, 322–349.
Summary updated