Crowdsourcing with Endogenous Entry: Why "Pay-to-Play" Might Yield Better Quality

TL;DR

In most crowdsourcing platforms, showing up is half the battle—but it costs effort. This paper by Ghosh and McAfee reveals a counterintuitive truth: if you want high-quality results from a crowd, you shouldn't subsidize entry. Instead, a "Winner-Take-All" approach, or even charging a small participation fee, forces a strategic "quality filter" that ultimately yields a better top-tier outcome.

The "Endogenous" Missing Link

Most economic models of contests assume that N people will definitely show up. In reality, contributors on platforms like TopCoder or Quora make a two-step decision:

1. Should I even participate? (Given the cost of understanding the task and the risk of winning nothing).
2. How much effort (quality) should I produce?

Prior work often failed because it didn't account for the fact that participation itself is a strategic choice. If rewards are spread too thin to encourage more people to join, the "serious" players lose the incentive to push for excellence.

Methodology: The Strategic Balancing Act

The researchers modeled this as a rank-order mechanism. Agents follow a symmetric mixed-strategy equilibrium $(p,G(q))$, where $p$ is the probability of entering and $G(q)$ is the distribution of their quality.

Breaking the Symmetry

The authors prove that there is no "pure strategy" equilibrium—everyone can't just pick one quality level. If they did, a tiny bit of extra effort would guarantee a win, leading to a perpetual "arms race." Thus, agents must randomize their effort levels.

Figure 1: The derivative of the benefit function $U$ with respect to rank-order rewards.

Two Worlds of Rewards

1. Attention Rewards (Social Computing)

In Q&A forums, "attention" is the currency. The site decides how many answers to display.

• The Insight: If you want the best possible answer, display all contributions except perhaps the very worst one. Curating too aggressively reduces the overall reward pool, which kills participation more than it boosts effort.

2. Redistributable Budgets (Contests)

When you have a fixed $1,000 prize, how do you split it?

• The Findings: Spread the money across top 5 places? No.
• The Winner-Take-All Principle: Giving everything to the winner ($a_1$) is usually optimal.
• The "Taxation" Twist: Surprisingly, making participants pay a small fee to enter (which is then added back to the winner's pot) can actually increase the maximum expected quality. This "tax" discourages low-effort "spammers" and increases the stakes for high-effort contributors.

Figure 2: The paper provides a mathematical framework for unique equilibrium construction.

Experiments & Mathematical Proofs

The paper uses rigorous calculus to show that for any cost function where the marginal cost relative to total cost is non-increasing ($c^{\prime }(q)/c(q)$), a "Winner-Take-All" contest dominates.

Specifically, when p < 1 (participation is not guaranteed), the authors demonstrate that shifting reward from the winner to a lower rank $s$ always decreases the expected maximum quality: ${\frac{dE{q}_{max}}{d{a}_s}|}_{B,a_1}\le 0$

Critical Insight: The Quality vs. Quantity Tradeoff

The industry takeaway is profound: More participants do not necessarily mean better results.

In a world of endogenous entry, the platform's goal is to manage the "crowding out" effect. By concentrating rewards at the top, you ensure that those who do participate are the ones willing to exert significant effort. While "entry subsidies" might increase the total number of submissions (average quality), they dilute the incentive to be the best.

Conclusion & Future Work

Ghosh and McAfee have provided a foundational proof that "gentle" incentives often fail in crowdsourcing. To get the best from the crowd, the mechanism must be "sharp"—rewarding the peak, even if it means fewer people show up.

Limitations: The model assumes "homogeneous" agents (everyone has similar potential). Future research should integrate "ability-heavy" scenarios where some agents are naturally more talented than others, which might justify different prize distributions.