Fig. 5: Model comparison for Study 1 conditions with a first phase.

Results from the ratings condition (n = 51 participants) are shown in the left column and from the full condition (n = 50) on the right. Human participant sampling rates in the first row are reproduced from Fig. 2. In the first row, horizontal solid lines link samples data for humans (black points) and Ideal Observer (grey points) when BF₀₁ > 3 (at least moderate evidence for equal means) while dotted lines indicate when BF₁₀ > 3 (at least moderate evidence for different means). Human and Ideal Observer sampling never showed BF₀₁ > 3 (at least moderate evidence for equal means). Statistical details for these pairwise tests can be found in Supplementary Table 1. There is clearer evidence for participant undersampling in the ratings than in the full condition. The second row shows BIC values where lower values indicate better model fit. Horizontal lines are shown in the colour corresponding to the point spread of the better-fitting model when BF₁₀ > 3 or in black when BF₀₁ > 3. Statistical details for these pairwise tests can be found in Table 5. The third row shows number of participants that each model best fitted. In both conditions, some version of the Biased Prior model (blue) fits better than any other model, though the ratings condition is more ambiguous. Model point spread data colours (See also legend in lower right panel): Cost to Sample (green), Cut Off (orange), Biased Prior (blue). Boxplots reflect first, second (median) and third quartiles, while whiskers reflect 1.5 interquartile range. Point spreads reflect individual participant mean values. Abbreviations: Subj = Models that make choices about subjective values; Obj = Models that makes choices about objective values.