Code and simulations accompanying:
Zhu, H., Beierholm, U., & Shams, L. (2026). Analyzing multisensory integration: do's and don'ts. .
This repository provides the model-based inference pipeline, simulation scripts, and figure-generation code used in the paper. The pipeline disentangles unisensory reliability, amodal priors, and binding tendency within a Bayesian Causal Inference (BCI) framework, and demonstrates why behavioral indices such as illusion magnitude and the temporal binding window cannot be interpreted as direct readouts of "integration strength."
The code in this repository reproduces:
-
Figure 1. Simulations of how unisensory reliability and binding tendency (
p_common) jointly shape crossmodal bias and the sound-induced flash illusion under BCI. -
Figure 2. Multiple parameters independently shape the temporal binding window in the sound-induced flash illusion.
-
The minimal, reproducible BCI model specification described in Section 3 of the paper.
-
The model-based inference pipeline (design → unisensory baselines → parameter estimation → interpretation).
All simulations are built on top of the BCI Python Toolbox (Zhu et al., 2024).