A VARIETY OF Internet online services are designed based
on contests. A canonical example is crowdsourcing
services, which solicit solutions to tasks by open calls to
online communities. Here the tasks can be of different
categories, such as art design, software development,
data-science problems, and various challenges such as
planetary-scale locating of objects. 12, 28 These services
operate under certain contest rules that include
specifying a prize allocation mechanism, for example,
awarding only a first-place prize or several position
prizes. The prizes can be monetary, or in-kind rewards
such as in terms of attention, status, or computing
resources, for example, CPU, bandwidth, and storage.
We refer to a contest as any situation in which agents
invest irreversible and costly efforts toward winning a
prize, which is allocated based on relative performance.
We use the term “contest theory” in a broad sense
to refer to a set of theories developed for the better
understanding and informed design of contests.
A central question in contest theory
is: How to allocate prizes to maximize a
desired objective? The objective may be
to maximize the utility of production
to the agent who solicits solutions to a
task, or to the whole society. The question of how to allocate prizes was studied as early as 1902 by Galton. 15 A study
of how to allocate prizes necessitates
to consider the incentives of contestants, who act strategically in investing
costly production efforts. 1, 11, 44 Game
theory models of contests have been
studied in auction theory, economic
theory, operations research, as well
as theoretical biology; for example,
Bishop and Smith. 3 The use of compensation schemes based on an individual’s ordinal rank rather than absolute performance in firms have been
studied by economists; for example,
Lazear and Rosen. 27 Game theory and
pertinent computational questions
have been studied by computer scientists. 31, 33, 36 Several new contributions
have been made on optimal allocation
of prizes in crowdsourcing contests,
equilibrium outcomes in games that
model simultaneous contests, and the
worst-case efficiency of production
in equilibrium outcomes of various
games that model contests.
The skill-rating methods that use
observations of relative performance
comparisons as input, such as ranking
outcomes in contests, have been studied extensively in the past. They are now
widely used in various applications,
such as sport competitions, online
gaming, and online labor platforms.
Exploring the basic game theory models
of contests found in online services.
BY MILAN VOJNOVIĆ
˽ The operation of various online platforms
relies on incentive mechanisms for
eliciting user contributions, which take
the form of a contest.
˽ Contest theory refers to a set of theories
for the better understanding and
informed design of contests.
˽ The theory provides insights into what
user behavior may arise in equilibrium,
guidelines of how to allocate prizes,
and algorithms for estimating skills
of individuals based on observed