Causal Markov violations and hidden mechanisms.

Area of Science:

• Cognitive Psychology
• Causal Inference
• Decision Making

Background:

• Causal Bayes nets are widely used to model causal inferences.
• However, psychological experiments reveal persistent violations of the causal Markov condition.
• Previous explanations proposed divergent causal representations by participants.

Purpose of the Study:

• To present a novel method for determining participants' causal representations.
• To investigate how participants update causal representations with mechanistic information.
• To examine if hidden mechanisms explain Markov violations.

Main Methods:

• Systematic experiments investigating participants' representation of causal structures.
• Integration of mechanistic information to assess updates in causal representations.
• Analysis of Markov violations in relation to hidden variables and mechanistic knowledge.

Main Results:

• Participants' Markov violations are not explained by hidden variables or missing mechanistic knowledge.
• A persistent reasoning error was identified as the cause of these violations.
• This error is robust to task format variations but can be mitigated.

Conclusions:

• The causal Markov condition violations in psychological experiments are attributed to a consistent reasoning error.
• This error is independent of hidden variables or incomplete knowledge of causal mechanisms.
• Interventions can reduce these reasoning errors, offering avenues for cognitive training.