Empirical evidence for resource-rational anchoring and adjustment

Falk Lieder, Thomas L. Griffiths, Quentin J. M. Huys, Noah D. GoodmanPsychonomic 2017 · DOI

Introduction

Anchoring bias—the tendency to rely too heavily on an initial piece of information when making decisions—has long been considered a hallmark of human irrationality. Since the pioneering work of Tversky and Kahneman in the 1970s, the "anchoring-and-adjustment" heuristic has been viewed as a flawed process where individuals fail to adjust sufficiently from a starting point, leading to systematic errors. However, recent research suggests that this "bias" might actually be a sophisticated, resource-rational strategy.

The Resource-Rational Hypothesis

The authors of this paper challenge the assumption that insufficient adjustment is a sign of a broken cognitive system. Instead, they propose that humans have finite time and limited cognitive energy. In a resource-rational framework, a rational agent should only continue adjusting their estimate as long as the expected improvement in accuracy is worth the "cost" of the time and mental effort spent.

Figure 1: Theoretical predictions of anchoring bias across different cost conditions. The model predicts that bias increases with time cost (TC) and decreases with error cost (EC).

As illustrated in the conceptual model above, the degree of anchoring bias is predicted to vary systematically based on environmental incentives. When time is expensive, the rational agent stops adjusting sooner to save resources, leading to higher bias. Conversely, when the penalty for error is high, the agent invests more cognitive "fuel" to reach a more accurate estimate, reducing the bias.

Experimental Methodology

To test this hypothesis, researchers developed a novel numerical estimation paradigm where participants predicted bus arrival times based on a provided timetable and previous examples. This task allowed for precise control over the participants' prior knowledge and the specific incentives they faced.

Figure 2: The experimental interface used in the study. Participants predicted bus arrival times while managing a budget of points that decreased based on time spent (Time Cost) and final prediction error (Error Cost).

The researchers independently manipulated two key variables:

  1. Time Cost: Points lost per second of deliberation.
  2. Error Cost: Points lost per unit of prediction error.

The study included two main experiments. Experiment 1 focused on self-generated anchors (internal starting points), while Experiment 2 used provided anchors (external suggestions) to determine if the same rational logic applied to both types of anchoring phenomena.

Key Results

The behavioral data strongly supported the resource-rational model across both experiments. Participants proved to be remarkably flexible, adjusting their cognitive effort in response to the cost-benefit structure of the task:

  • Time Cost Effect: When time was expensive (30 points/sec), participants were faster but less accurate. In Experiment 1, relative adjustment from the anchor dropped from 50.7% to 31.0% when time costs were introduced.
  • Error Cost Effect: When the penalty for inaccuracy was high, participants slowed down and increased their adjustment from 31.6% to 50.1%.
  • Linear Relationship: The magnitude of the anchoring bias grew linearly with the distance between the anchor and the true value, exactly as the computational model predicted.
  • Unified Logic: Experiment 2 confirmed that externally provided anchors are subject to the same cost-benefit tradeoffs as internal, self-generated ones.

Conclusion and Future Work

This research reframes anchoring bias as an optimal tradeoff between accuracy and effort rather than a fundamental flaw. By showing that people adaptively change their level of bias based on the specific costs of time and error, the authors provide compelling evidence that human judgment is more rational than previously thought—provided we account for the resource constraints under which the brain operates.

While the study successfully demonstrates how we adjust, future work is needed to understand the origins of the initial anchors themselves in complex real-world scenarios. Additionally, it remains to be seen whether the brain explicitly calculates these tradeoffs or employs simpler heuristics that have evolved to mimic rational behavior.

Summary updated

BibTeX
@inproceedings{lieder2017empirical,
  author = {Falk Lieder and Thomas L. Griffiths and Quentin J. M. Huys and Noah D. Goodman},
  title = {{Empirical evidence for resource-rational anchoring and adjustment}},
  booktitle = {Psychonomic},
  year = {2017},
  doi = {10.3758/s13423-017-1288-6},
}
Figure