In many everyday situations, decision-making comes with feedback on the consequences of one's choices, linked to their outcome: financial gain or loss, success or failure, information about what could have happened if a different choice had been made... Dominant theories in economics, psychology, and neuroscience assume that this feedback allows for gradual adjustment of one's beliefs and learning to make better choices, especially within a framework of repeated decision-making.
To test this hypothesis, a research team from the Laboratoire de neurosciences cognitives computationnelles (Inserm/ENS) led by Stefano Palminteri, Inserm Research Director, in collaboration with the Paris School of Economics, conducted behavioral experiments involving more than 500 participants.
Illustration image Pixabay At each stage of these experiments, volunteers faced a binary choice situation between a safe option (for example, winning 20 points with certainty), and a risky option (for example, having a 50% chance of winning 40 points and a 50% chance of winning nothing at all). To rule out certain methodological biases, the scientists made the risky option more or less advantageous compared to the safe option, by varying the probability of winning (10%, 50%, 90%) and its value (40 or 60 points). The quality of the decisions was measured by the volunteers' ability to choose the option offering the highest average gain [1].
The feedback, when given, was either partial (only the result of the chosen option was given) or complete (the results of both possible options were given). Its effect was then assessed in two complementary ways: first, after making a choice, the person received feedback whose effect was examined on the following choice and allowed for observation of how behavior evolved. Secondly - and this is one of the most innovative aspects of the study - in some experiments, volunteers were not informed in advance that feedback would be provided, while in others, they explicitly knew that feedback would occur.
The results of the experiments show that the presence of feedback systematically increases risk-taking by 35% to 45%, but without improving the quality of decisions.
Depending on the situations, two distinct psychological mechanisms highlighted by the analysis of the experimental data could explain these results. When only the result of the chosen option is revealed (partial feedback), the increase in risk-taking would be linked to curiosity: choosing the risky option allows for obtaining more information (since the result of the safe option is, by default, always known). On the other hand, when volunteers also see what they could have won with the unchosen option (complete feedback), it is the anticipation of regret that would favor risky decisions in subsequent choices.
Another surprising result: among people who knew, before making their choice, that they would receive feedback, the increase in risk-taking appeared even before the volunteers had received any feedback.
"
These results suggest that the mere anticipation of receiving feedback would alter the attitude towards risk, even before any experience," explains Stefano Palminteri.
Finally, the team highlighted one last particularly counter-intuitive result: immediately after receiving positive feedback (a confirmation that the choice made resulted in the maximum possible gain), the probability that the person would choose a risky option again decreased.
"
According to dominant theories, one would expect positive feedback to lead to repeating the associated choice,
which would be compatible with a learning-by-experience effect, analyzes Stefano Palminteri,
yet, we observe the opposite. This can be explained by a cognitive bias called the 'gambler's fallacy': after winning, people estimate they have a lower chance of winning again immediately, and therefore avoid playing again."
These results therefore expose both paradoxical effects related to the anticipation of feedback and others related to its immediate consequences. According to the research team, they considerably limit the learning effect of feedback in a risky decision-making context.
"This research contributes to a finer understanding of the cognitive mechanisms of decision-making under risk, comments Antonios Nasioulas, PhD in economics at the Paris School of Economics, first author and corresponding co-author of this work
. Feedback is often presented as a 'debiasing' tool to improve decisions in applied contexts, such as financial management or medical choices.
This study shows, on the contrary, that feedback can introduce new biases, by altering the attitude towards risk rather than promoting rational learning."
This work thus opens new avenues for studying certain excessively risky behaviors observed in daily life or in behavioral studies. The authors also emphasize the importance of taking these results into account for the design of decision-support tools, whether intended for the general public, professionals, or integrated into digital systems.
Note:
[1] This approach is based on "expected value" (EV); for each option, the team calculated the average of possible gains, weighted by their probabilities: in the specific case of this study
EV = (probability of winning × possible gain). At each choice stage, it is therefore the option with the highest expected value that is considered "optimal." This method makes it possible to determine whether the choices are optimal from a mathematical point of view, independent of personal risk preferences.
Example: If the safe option yields 20 points, then EV
safe = 1 x 20 = 20; if the risky option offers a 50% chance of winning 90 points, then EV
risky = 0.5 x 90 = 45. Here EV
risky is therefore the most optimal choice.