Thursday, June 23, 2016

Agency promotes risk-seeking

People and other animals are generally risk-averse.  We tend to prefer sure things, despite their low payoff value, to risky things that may have a high payoff value.  However, a recent study demonstrates that people can change their behavior dramatically, becoming risk-seeking, if they participate in games in which there is an illusion of control.  That is, if you believe that your actions are responsible for the outcome of the gamble, then you are more likely to be risk-seeking.

In a paper that was recently published in the Proceedings of the National Academy of Sciences, Sam McDougle, Jordan Taylor and colleagues at Princeton and Berkeley designed a game where there were two targets.  The participant chose one of the targets, and depending on the hit probability of that target, and the payoff value of the target (both manipulated by the experimenters), the participant received points, which would later be exchanged for money.  The hit probability and payoff were randomly varied but always in proportion to each other.  This way, the expected value of each target (hit probability times payoff value) was constant in time, as well as equal among the targets.  This made it so that there was no “optimum” behavior, that is, there was no target that they could pick which would maximize their winnings. 

If a participant was risk-averse, then she should usually pick the target that had the higher hit probability, ignoring the fact that the other target, though less likely to hit, had a higher payoff value.  On the other hand, if the participant was risk-seeking, then she should usually pick the target that had the low hit probability, gambling to get the higher payoff value.

The authors divided the participants into a few groups.  In the Standard group, the volunteers pressed one of two keys, selecting a right or left target.  These participants tended to usually pick the high hit probability target.  They exhibited risk-averse behavior. 

Does risk-aversion change if people believe that they can control the outcome of the gamble through their behavior?  To test for this, the authors changed the task: now instead of pushing a button to indicate the chosen target, the participants reached toward the target while holding a joystick.  As they moved the joystick, a cursor indicated whether they landed on target or not.  This was manipulated through the hit probability.  In reality, it actually did not matter how they moved the joystick.  If the hit probability was low, the cursor would not land on target (but close to it), giving the illusion that they missed the target because of a poor movement.  Remarkably, these participants behaved in a risk-seeking manor.  They usually picked the target that was less likely to hit, gambling that they could capture the high payoff by improving their movement.

The key difference between the two groups was that in the key-press group, the action had very low variability: you press a button, giving the impression that there is little relationship between what you do and the hit probability of the target.  However, in the reach group, the action had high variability: you see that the cursor that appears to be connected to your actions did not hit the target.  Perhaps by reaching slightly differently, you will hit it the next time.  In reality, in both cases the hit probability was predetermined, having nothing to do with your actions.

The work suggests that greater risk-seeking arises in circumstances where there is a likelihood that the actions themselves were responsible for the outcome, and not some external agent.  If, on the other hand, one believes that the outcomes were predetermined, due to an external agent, then they are likely to be more risk-averse, avoiding actions that have low probability of success (but potentially high reward).

Let me take these results and make some generalizations.  Suppose you wanted to build a casino.  The results of this study would say that you should buy slot machines that required some large motion of the arm to pull down a lever, rather than a button-press.  That would make your customers believe that the way they pull the lever has something to do with their winnings, and this belief would make them more risk-seeking.  (The down side is that pulling a lever may get tiring.).

Cultures and religions differ in the relative importance of agency and concept of destiny.  Perhaps a belief in pre-determination of events and an “ultimate cause” would reduce the sense of agency, which in turn would promote risk-averse behavior.


Samuel McDougle, Matthew Boggess, Mattew Crossley, Darius Parvin, Richard Ivry, and Jordan Taylor (2016) Credit assignment in movement-dependent reinforcement learning.  Proceedings of the National Academy of Science (USA) 113:6797-6802.