Cognitive Biases in Human Causal Learning

Maldonado, Antonio; Catena, Andrés; Perales, José César; Cándido, Antonio · 2007 · Crossref

DOI: 10.1017/s1138741600006508

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper investigates the cognitive biases and psychological mechanisms that modulate human causal learning, challenging models that rely solely on either prior beliefs or statistical covariation. The authors argue that causal inference is a complex process requiring the integration of previous causal knowledge and direct empirical evidence. To explain this complexity, the paper proposes a three-level cognitive architecture: a lower level responsible for coding event frequencies, a middle level for computing statistical associations, and a higher level for integrating this evidence with prior beliefs. The authors review experimental evidence supporting this architecture, beginning with the "judgment frequency" effect. Studies show that when participants make causal judgments after every trial rather than after blocks of trials, their accuracy decreases, and they become overly sensitive to recent information. This supports a belief revision model where new evidence is computed via a weighted delta rule and integrated with prior judgments through an anchoring and adjustment mechanism. The paper further examines cue competition scenarios, such as overshadowing, blocking, and super-conditioning. Results indicate that prior beliefs significantly influence causal judgments in ambiguous situations (overshadowing), attributing more causal power to cues with higher prior propensity. However, in blocking and super-conditioning scenarios, where covariation data is sufficient to distinguish causes, empirical evidence dominates, and prior beliefs have less impact. Additionally, the research highlights the role of attention and reliability in causal detection. Experiments on "inattentional blindness" reveal that participants fail to detect negative or preventative causal relationships for incidental cues, not due to computational failure, but because they fail to encode or retrieve the necessary trial frequencies. This bias does not occur for positive or generative relationships. The study also demonstrates that the relative weight of beliefs versus covariation is modulated by the perceived reliability of the information. When empirical evidence is deemed highly reliable (e.g., through pre-training with neutral cues), it overrides prior beliefs, whereas low-reliability evidence allows prior beliefs to dominate, explaining confirmation biases. The significance of this work lies in its demonstration that neither beliefs nor covariation alone can explain human causal learning. Instead, the authors conclude that a hierarchical architecture is necessary, where attentional resources determine what is coded, statistical mechanisms compute contingency, and an integration mechanism weighs this evidence against prior knowledge based on reliability. This framework provides a more comprehensive explanation for the flexibility and biases observed in human causal inference and decision-making.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• inattentional change blindness