TL;DR
This paper introduces a "safety level" analysis for competing contests, showing that a simple "exclusive proportional policy" (EPP) contest can attract near-optimal effort from contestants. This policy provides a robust guarantee for contest designers, ensuring a minimum level of effort regardless of their competitor's strategy, especially as the number of contestants grows.
What Problem Does This Paper Address?
The paper tackles the challenge of designing effective contests in environments where multiple contests compete for the attention and effort of contestants. Think of platforms like Amazon Mechanical Turk or design marketplaces like 99designs, where individuals can choose to allocate their limited effort across various tasks or competitions. Traditional contest design literature often focuses on a single contest. However, in reality, contestants must decide which contest to invest in. This paper aims to understand how a contest designer can guarantee a certain level of effort for their contest, even when facing competition from another designer who can employ any strategy. This is crucial for designers who want to maximize the total effort invested in their contest, for example, to gather more survey responses or get more design submissions.
Methods
The authors model a two-stage game: first, contest designers choose their "contest policy" (how prizes are awarded based on effort), and second, contestants decide how much effort to invest in each contest, given their total effort budget. They introduce the concept of a "safety level" for a contest policy, which is the minimum total effort a contest can guarantee across all possible strategies of the competing contest designer and all possible Nash equilibria.
A key contribution is the analysis of a specific policy called the "Exclusive Proportional Policy" (EPP). Under EPP, a contest designer awards the prize equally only to those contestants who dedicate all their effort to that contest. This simplifies contestant behavior, as they will either invest all their effort or none. The analysis uses mathematical tools to derive inequalities that must hold in any Nash equilibrium and then uses these to establish lower bounds on the guaranteed effort. They also analyze the behavior of the safety level as the number of contestants ($n$) and risk aversion ($\beta$) change, and specifically examine the case of risk-neutral contestants ($\beta=1$).
Datasets
Not discussed in the paper.
Key Insights and Ideas
- The Power of Simplicity: The EPP policy, which rewards only "exclusive" participants, is surprisingly effective. It doesn't require complex prize structures or sophisticated contestant modeling.
- Robustness through "Safety Level": The safety level concept is powerful because it doesn't assume the competitor plays optimally or rationally. It provides a guaranteed minimum outcome, akin to a max-min strategy.
- Convergence to Relative Prize Power: As the number of contestants ($n$) grows large, the safety level of the EPP contest converges to its "relative prize power" ($t$). This means a contest designer can expect to capture a share of the total effort proportional to their contest's prize share, regardless of the competitor's policy.
- Risk Aversion Matters (for convergence): Higher risk aversion ($\beta < 1$) leads to faster convergence of the safety level to its asymptotic value as $n$ increases.
- Risk Neutrality is Different: When contestants are risk-neutral ($\beta = 1$), the safety level is generally lower and more sensitive to the "substitution effect" parameter ($\alpha$, which relates prize value to effort cost). The relationship between safety level and prize power becomes sub-linear and convex.
Key Results
- For a large number of contestants ($n \to \infty$) and risk-averse contestants ($\beta < 1$), the safety level of the EPP contest approaches the relative prize power ($t$) of the contest designer.
- The safety level function is convex with respect to the relative prize power ($t$). This implies that a contest designer with a larger share of the prize power gains more from an increase in that power.
- For risk-neutral contestants ($\beta = 1$), the safety level is bounded by a more complex formula that depends on $n$, $\alpha$, and $t$. In a specific case ($\alpha=1, t=0.5$), the safety level can be significantly lower than $t$, highlighting a real difference between risk-averse and risk-neutral behavior.
Limitations and Future Work
- Observable Effort: The model assumes contestant effort is observable, which is necessary for the EPP policy to function. The authors suggest future work could explore contest policies that don't require this assumption.
- Fixed Prize Size: The current model assumes contest designers fix their prize sizes. Future research could explore multi-round games where prize sizes might be adjusted based on previous rounds' outcomes.
- Characterizing Equilibria: While the safety level provides a robust guarantee, the paper notes that precisely characterizing all equilibrium outcomes remains an open area.
- Alternative Policies: The paper focuses on EPP. Investigating whether other contest policies can offer even higher safety levels is a natural extension.
Summary updated