Affective forecasting as an adaptive learning process.

Area of Science:

• Cognitive Psychology
• Affective Science
• Computational Neuroscience

Background:

• Human affective forecasting is theorized as an adaptive learning process driven by prediction errors.
• While a Kalman filter can model this process, human forecasts are often suboptimal and biased.

Purpose of the Study:

• To compare the accuracy of human affective forecasts with statistical forecasts generated by a Kalman filter.
• To explore the discrepancies between human and statistical affective forecasting processes.

Main Methods:

• Participants (N=240) from the general population made repeated forecasts of their affect.
• Forecasts were made for future time points ranging from 1 minute to 2-3 hours.
• Data collected in daily life and experimental settings across three studies.

Main Results:

• Human forecasts exhibited larger absolute errors than Kalman filter forecasts for hour-long predictions (dz = 0.42, 0.30).
• No significant difference in absolute errors was found for minute-long forecasts (dz = 0.17).
• Participants showed a consistent bias of projecting current affective states onto future forecasts.

Conclusions:

• Human affective forecasting accuracy is lower than statistical models for longer durations.
• The tendency to project current affect may underlie human-specific forecasting errors.
• Findings highlight potential deviations from optimal prediction error-driven learning in human emotion forecasting.