Table 1 Model comparison metrics for the BCI, FF, FS and BCI-NP models show superiority of the postdictive Bayesian causal inference (BCI) compared to a forced-fusion (FF), forced-segregation (FS), and non-postdictive BCI model.

From: Causal inference shapes crossmodal postdiction in multisensory integration

  1. Bayesian information criterion (BIC) and Akaike information criterion (AIC) quantify the trade-off between model fit (i.e., good explanation of the data) and model complexity (i.e., a sparse number of parameters). At group level, participant-specific relBICGroup and relAICGroup are summed over all participants (n.b. smaller relBICGroup and relAICGroup values indicate that a model provides a better explanation of our data). Nagelkerke’s coefficient of determination (R2) measured the proportion of explained variance against a null model of random guesses across 6 response options (i.e., across-participant mean reported). In a random-effects Bayesian model comparison, the protected exceedance probability (PEP) quantifies the probability that one model is more prevalent than competing models at group level, beyond which model frequencies could arise by random variations.
Back to article page