Modeling mean estimation tasks in within-trial and across-trial contexts.

Area of Science:

• Cognitive psychology
• Computational neuroscience
• Visual perception

Background:

• Mean estimation is crucial for ensemble coding and cue integration.
• Understanding information summarization in perception is key.

Purpose of the Study:

• To formalize information summarization in mean estimation using computational models.
• To compare the predictive power of the Fidelity-based Integration Model (FIM) against other models.
• To investigate within-trial weight distribution, across-trial integration, and set-size effects.

Main Methods:

• Development and comparison of computational models, including FIM.
• Experimental investigation of mean estimation tasks (sequential and simultaneous).
• Analysis of observer behavior regarding trial weighting and estimation biases.

Main Results:

• Observed non-equal weighting within trials and biases in over/underestimation of means.
• Demonstrated declining and stabilizing mean estimation accuracy with increasing set sizes.
• FIM successfully simulated all experimental patterns, while other models failed.

Conclusions:

• FIM provides a robust framework for understanding information processing in mean estimation.
• FIM's structure offers insights into visual working memory capacity and sub-sampling.
• FIM facilitates task-dependent modeling for ensemble coding research and synthesis.