Experiment 1 design and subjective ratings.

(A) Relationships between actions and shapes with the objective probability of a particular shape following a given action ranging from 71% to 48%, 25% and 2%. At the end of the experiment, participants rated the probability for each of these 16 action-shape combinations. (B) Trial structure. On each trial, participants were cued to perform an action and then presented with an outcome – one of the four shapes according to the relationships outlined in (A). Participants’ task was to indicate as quickly and accurately as possible which shape was shown. (C) Participants’ subjective probability ratings for the four levels of objective probability. As objective probability increases, so does rated probability. Coloured dots represent each participant’s mean rating at each level of objective probability. Boxes denote lower, middle and upper quartiles, and whiskers extend to 1.5 times the interquartile range. Half-violin plots display density estimates.

Experiment 1 results.

(A) RTs as a function of objective probability, showing faster responses to outcomes that were objectively more probable. (B) Participant-specific slopes for the effect of objective probability on RTs, with most participants showing a negative slope. (C) RTs as a function of subjective probability ratings, indicating faster responses to outcomes rated as more probable. (D) Participant-specific slopes for the effect of subjective probability on RTs, with most participants exhibiting a negative relationship. (E) Accuracy as a function of objective probability, showing fewer errors for outcomes that were objectively more probable. (F) Participant-specific slopes for the effect of objective probability on accuracy, with the majority showing a positive association. (G) Accuracy as a function of subjective probability, showing fewer errors for outcomes rated as more probable. (H) Participant-specific slopes for the effect of subjective probability on accuracy, with the majority exhibiting a positive relationship. In (A), (C), (E) and (G), data points represent each participant’s mean performance within the four levels of objective probability or four equal-width bins of subjective probability ratings, created for visualization purposes only. Boxes show quartiles and whiskers extend to 1.5 times the interquartile range. Red lines represent the fixed effect estimates from the full model, with shaded ribbons showing 95% confidence intervals (CI). In (B), (D), (F) and (H), participant-specific slopes are displayed ordered by magnitude. Red points and horizontal bars indicate the fixed effect slopes and their 95% CI. Dashed vertical lines at zero indicate no effect.

Experiments 2 and 3 design and subjective ratings.

(A) Example mapping between tones and orientations. Each tone predicted one orientation with 75% and the other with 25% probability. (B) Trial structure. On each trial, participants were presented with a tone followed by a Gabor. They had to discriminate the Gabor orientation and also rated how expected (Experiment 2) or surprising (Experiment 3) they found the presented orientation. (C) Participants’ subjective expectation (Experiment 2) and surprise (Experiment 3) ratings over the course of the experiment for the 75% and 25% trials.

Experiment 2 results.

(A) RTs as a function of Rescorla-Wagner expectedness, showing faster responses when there was a higher objective associative strength between outcome and cue. (B) Participant-specific slopes for the effect of RW expectedness on RTs, with most participants showing a negative slope. (C) RTs as a function of subjective expectedness ratings, indicating faster responses to outcomes rated as more expected. (D) Participant-specific slopes for the effect of subjective expectedness on RTs, with most participants exhibiting a negative relationship. (E) Accuracy as a function of RW expectedness. When included alongside subjective expectedness ratings, objective RW expectedness had no significant effect on accuracy. (F) Participant-specific slopes for the effect of objective RW expectedness on accuracy. (G) Accuracy as a function of subjective expectedness, showing increasing accuracy as outcomes were rated as more expected. (H) Participant-specific slopes for the effect of subjective expectedness on accuracy, with the majority exhibiting a positive relationship. Plotting conventions are the same as for Fig. 2. For visualization purposes, data was grouped into eleven equal width bins of either RW expectedness value (A, E) or subjective expectedness rating (C, G) and participants’ mean performance was calculated. Bins span the minimum to maximum observed values (RW expectedness can theoretically range from –1 to 1).

Experiment 3 results.

(A) RTs as a function of Rescorla-Wagner prediction error, showing slower responses when the stimulus elicited a larger objective prediction error. (B) Participant-specific slopes for the effect of RW prediction error on RTs, with most participants showing a positive slope. (C) RTs as a function of subjective surprise ratings, indicating slower responses to outcomes rated as more surprising. (D) Participant-specific slopes for the effect of subjective surprise on RTs, with most participants exhibiting a positive relationship. (E) Accuracy as a function of RW prediction error. When included alongside subjective surprise ratings, objective RW prediction errors had no significant effect on accuracy. (F) Participant-specific slopes for the effect of objective RW prediction errors on accuracy. (G) Accuracy as a function of subjective surprise, showing decreasing accuracy as outcomes were rated as more surprising. (H) Participant-specific slopes for the effect of subjective surprise on accuracy, with the majority exhibiting a negative relationship. Plotting conventions are the same as for Fig. 4.