Transfer of confidence in a novel observational learning task

Methods for measuring the confidence with which people make decisions, have traditionally relied on post-decision self-reports. We wanted to: 1) develop a new experimental paradigm that aims at measuring the confidence in a decision without participants having to explicitly rate their confidence, 2) characterize their decision using fMRI, and 3) show transfer of confidence through observation using the said paradigm. In our experiment, the participants have to move a cursor around a circle to catch particles moving from the centre to the edge of the circle. The direction of the particles is block-wise determined by a mean and its standard deviation from that mean. The participants can change the size of the cursor, and the amount of points rewarded for each catch is inversely proportional to the size of the cursor. The paradigm was tested in several deceit-free behavioural psychophysics experiments and analysed using computational modelling. To test that the task actually measures confidence, we added a condition in which blocks of trials were followed by a confidence rating scale. The model estimated trial-by-trial particle variance correlated strongly with the normalized ratings given by the participants. Similarly, the computational modelling found each participant’s trial-by-trial estimation of the particle variance to be correlated with the width of the cursor on the given trial. fMRI results show strong correlations between expected decision making and learning areas adding validity to the task. Finally, to investigate if it is possible to transfer confidence between participants, we ran an experiment where participants could observe the cursor width chosen by previous participants. Results show that participants that observed a player with higher base confidence chose a significantly smaller catcher than the group observing a player with lower base confidence, an effect that persisted after the other player’s choices were no longer displayed. In conclusion, we have developed a novel task that allows for measuring choice confidence implicitly, and used this task to show that participants adapt an observed level of confidence to their own choices, a level that persists even after observations are no longer available.