Extended Data Fig. 2: Task behaviour.

(a) Proportion of abandonment choices as a function of trial number (mean values +/− SD, n = 30). (b) Results of a logistic regression analysis predicting goal switching as a function of simple task parameters (mean beta weights associated with the regressors are plotted, +/− SEM, n = 30, dots show individual participant beta weights). The plot demonstrates participants were sensitive to the key elements of the task: they were more likely to switch when the offers associated with alternative goods were high (two-sided t-test against zero; t(29) = 14.74, p = 5.29 × 10⁻¹⁵), and less likely to switch when the offers for their current good were high (two-sided t-test; t(29) = −16.84, p = 1.63 × 10⁻¹⁶), or after having accumulated many goods in their net (two-sided t-test; t(29) = −27.67, p = 2.12 × 10⁻²²). We also found that people were more likely to abandon a goal when the size of the target net was larger (two-sided t-test; t(29) = 5.35, p = 9.51 × 10⁻⁶). We found no effect of the second-best alternative on abandonment decisions (t(29) = 0.11, p = 0.912). (c) Reaction times plotted against goal progress (mean values +/−SEM, n = 30). (d) Relationship between reaction times and trial type (mean values +/− SEM, n = 30, dots show individual participant means). Participants are much slower to respond on abandonment trials compared to persist trials (mean RT on persist trial = 1.52 seconds, SD = 0.28; mean RT on abandonment trial = 2.59 seconds, SD = 0.62; two-sided paired t-test for difference in RTs: t(29) = 11.05, p = 6.42 × 10⁻¹²). Participants were also faster on persist trials than the first trial of a block (mean RT on the first trial = 2.17 seconds, SD = 0.65; two-sided paired t-test for difference between first trial and persist trial: t(29) = 6.05, p = 1.38 × 10⁻⁶), but slower on abandonment trials than the first trial (two-sided paired t-test for difference between first trial and abandon trial t(29) = 2.99, p = 0.006).