Table 2.

Free parameters and parameter fits, for the best-fitting behavioral model for each classa

Behavioral modelFree parametersMean ± SERange
Bayesian_nolearningβ: softmax inverse temperature4.04 ± 2.22[0, ∞]
Additiveβ: softmax inverse temperature7.04 ± 3.46[0, ∞]
Mostleast_voter (voting for or against the sectors in which an animal was the most or least common)v_pos: size of positive vote; v_neg: size of negative vote1.69 ± 3.39; 0.754 ± 1.39[0, ∞]; [0, ∞]
Bayesian_feedbackRL (learning from all feedback, no scaling of learning rates, and αpos = αneg)α: learning rate; β: softmax inverse temperature0.0515 ± 0.161; 4.82 ± 2.52[0, 1]; [0, ∞]
  • aThe best-fitting behavioral models for all classes did not model recency or primacy biases, and used each participant's subjective estimates of the animal likelihoods rather than the actual likelihoods. For model classes that had additional variants, the best-fitting settings are described in parentheses.