Balls risk measurement task summary and demo.
Source code is here. Game demo directly below text. I wrote all of the code for the project, while Grigory Chernov (PhD student at Higher School of Economics), Alexis Belianin (Assistant Professor of Economics at Higher School of Economics), and me jointly discussed the ideas.
People have different risk attitudes. Most are risk averse, but the degree of risk aversion varies widely. Economists are interested in empirical estimation of risk preferences of people, since they are an important part of the decision-making processes, and being able to estimate them allows to better predict economic behavior.
The two most popular risk elicitation methods are the Balloon Analogue Risk Task (BART) and the Bomb Risk Elicitation Task (BRET), with BART being the older one, and being gradually supplanted by BRET. However, in our view, both of them are inadequate: in both games it's initially unclear to the player how exactly is risk related to the expected profit, thus we cannot distinguish whether the person is seeking risk, trying to avoid it, or is simply trying to maximize their profit (risk-neutral).
The end goal of this project is two-fold. First, it was to design a game, which would avoid this pitfall and would allow to estimate risk preferences in a "clean" way. Second, both BART and BRET measure risk-preferences in a setting, where the choice of the strategy is "a priori" and not based on the skill level. In the real world, the situation where we adjust the risks we take, based on our competencies, are extremely widespread, yet they're impossible to measure today. Examples of such cases are: choice of driving speed in a car, choice of the level of course load in a university, choice of the computer game difficulty, etc.
The design was guided by the desire to force people to quickly decide on one of *n* strategies that are available, which differ by risk and reward, and to allow to pick the level of desired cognitive load (because it seemed to better mimic the way risky decisions are made in real life).
So far we have the following problems to overcome (my personal summary):
- When trying the game out in real world, it became apparent that people treat it much more as a simple motor game, rather than a process in which they want to make decisions, which results in moderate, but not high enough, strategy variability, and high noise.
- Just like in BART and BRET we are still unable to convince ourselves that what we are really measuring is risk preferences, rather than exploration of profit maximization strategies or anything else.
- The fake balls were intended to make the task of picking the real balls harder, as the number of fake balls increased (cognitive load). This doesn't seem to be the case for any design we tried. People adapt to ignoring fake balls very quickly, not really bearing any cognitive load.
- Lastly, we underestimated the technical difficulties that arise from the game being dynamic. In particular, it's not clear if this design can be adapted so that the game could be played on a phone (or even simply without a computer mouse) and herewith save its core mechanic.
The problem of disentangling of risk-preferences from profit maximization can probably be solved by calculating the posterior probability of success in a chosen strategy during the game for each person, thus figuring out the optimal strategy and seeing how the actual strategy deviates from it. The remaining problems are likely innate to the fundamental design of the task and would require a significant redesign to solve (likely by removing the motor component and adding a different skill-based task with cognitive load or under time pressure).
Number of fake balls: