Table 2 Comparative evaluation of model performances for dataset 2: likelihoods, frequencies, PEP (protected exceedance probabilities), and posterior group parameters

From: Expectation violations signal goals in novel human communication

Model

Log likelihood

Model frequency

PEP

τ

ε

λ

γ

α

Surprise model

−60.46

0.78

0.99

6.80

0.31

0.11

2.03

1.83

State model

−69.48

0.22

0.00

2.77

0.45

–

–

1.15

Movement model

−94.26

0.00

0.00

2.50

0.70

0.15

1.98

–

  1. Model: refers to the three tested computational models: Surprise, State, and Movement models. Log likelihood: this represents the log-transformed probability of the data given the model, where higher values indicate a better fit of the model to the observed data. Model frequency: reflects the proportion of subjects for which a given model is most likely the best fit, with values closer to 1 indicating higher generalizability across the sample population. PEP: is the probability that a model is more likely than any other model, while also considering the possibility that no single model is a good fit for the data population. Posterior group parameters (τ, ε, λ, γ, α): estimated group values for parameters: \(\varepsilon\) (discount rate for planning horizon), \(\lambda\) (backward movement),\(\,\gamma\) (forward and left/right movement), \(\alpha\) (slope of State prior), and \(\tau\) (choice stochasticity).
Back to article page