Detecting and predicting changes.

Area of Science:

• Cognitive psychology
• Decision science
• Machine learning

Background:

• Human prediction of sequential events often involves flawed strategies, like assuming patterns where none exist.
• Dynamic environments offer opportunities to detect true temporal structure, challenging intuitive prediction methods.

Purpose of the Study:

• To investigate human ability to predict sequential events in changing environments.
• To compare statistical inference accuracy with prediction accuracy for future outcomes.
• To model human performance using computational approaches.

Main Methods:

• Two experiments were conducted with observers predicting outcomes in dynamic environments.
• Participants' statistical decisions and future outcome predictions were recorded.
• A particle filter model was employed to simulate and explain observed human performance.

Main Results:

• Participants accurately inferred hidden states of data-generating processes.
• Prediction accuracy for future outcomes was lower than statistical inference accuracy, despite identical normative correctness.
• Individual differences in performance were largely explained by model parameters.

Conclusions:

• Humans can perform correct statistical inference in dynamic environments but struggle with accurate future predictions.
• A particle filter model provides a plausible account of human sequential event prediction and individual variability.
• The findings highlight a dissociation between statistical inference and predictive accuracy in changing environments.