Explaining Away, Augmentation, and the Assumption of Independence

Area of Science:

• Cognitive Psychology
• Decision Science
• Artificial Intelligence (Bayesian Networks)

Background:

• The 'explaining away' effect is a key inference in common-effect causal structures.
• This phenomenon involves updating beliefs about causes when an effect and one cause are known.
• Existing research shows people exhibit this effect but often to a lesser degree than normative models predict.

Purpose of the Study:

• To investigate the conditions under which the 'explaining away' effect is observed in human probability judgments.
• To compare human responses to predictions from Bayesian network models across various information conditions.
• To identify factors influencing deviations from normative 'explaining away' predictions.

Main Methods:

• Participants estimated cause probabilities under different scenarios of effect presence/absence and evidence certainty.
• Bayesian network modeling was used to generate normative probability predictions.
• Sensitivity analysis assessed the impact of information conditions on normative probability changes.

Main Results:

• Explaining away occurred less frequently when causes were perceived as interacting rather than independent.
• The magnitude of normative probability change influenced the subjective observation of the effect.
• Participants showed difficulties with negated evidence, indicating a potential 'double negation' effect.

Conclusions:

• Human 'explaining away' is modulated by perceived causal interaction and the subjective relevance of probability shifts.
• Certainty about the absence of an effect serves as a diagnostic condition for reasoning about common-effect structures.
• Further research is needed to understand reasoning with complex evidential structures, including double negations.