League of Legends: Testing the Effects of Suspense, Surprise,
and Friendship on Satisfaction and Entertainment
Brian Lance, Wenjie Zhang, Jose Santiago-Calderon, Sean Bjurstrom, Carlin Crisanti
SSSPE / Economics, IST / Data Analytics, SSSPE / Politics
Overview
What is League of Legends?
Model: Suspense and Surprise
Analysis Plan
This project seeks to explore two questions using Big Data
from the world’s most popular computer game today,
League of Legends:
League of Legends is a free online computer game developed by Riot Games and released in 2009. It is referred to
as a Multiplayer Online Battle Arena (MOBA).
Premise: Why do people consume useless information?
Why do people read mystery novels? Why is a close
tennis match more fun to watch than a land slide?
1. Receive Raw Data: Riot Games has informed us that
the raw data will be delivered as dozens (possibly hundreds) of separate SQL tables.
(1) Does playing with friends leads to more satisfying
game experiences? Ravaja et al (2005) argue that playing with friends may create a greater sense of personal
connection during the game, giving players more satisfaction regardless of whether they win or lose.
It is extremely popular, with an estimated 67 million monthly players, and peaks of 7.5 million concurrent players.
Theory: People have preferences for being surprised or
held in suspense. They often consume useless information
for entertainment value, preferring experiences which
cause a greater sense of suspense and/or surprise.
2. Assemble Working Datasets: Once we’ve received
the data, we will write server-side scripts to produce unified, working SQL tables that can be analyzed in statistical
analysis software.
Setup: An audience (agent) consumes non-instrumental
(useless) information inducing suspense and/or surprise
about the true state, which is gradually revealed by the information designer (principal). In our case, the agent is the
typical LoL player and the principal is Riot Games.
3. Exploratory Analysis: Starting out with hundreds of
variables, we will run model selection techniques to identify
the most important variable subsets: Lasso (Tibshirant,
1996; R package: glmnet), “post-double-selection” (Belloni, et al., 2014) and Principal Component Analysis (R
package: pls and prcomp).
Model Components
4. Testing the Model: The primary challenge will be to
develop suspense and surprise scores for each game.
This means we need a method for estimating player’s
beliefs throughout a game. Our first strategy will be to use
model-based predictions of victory as a substitute for subjective player beliefs.
Personal accounts track each player’s match history and
provide bonuses as they continue playing the game.
(2) Does the experience of suspense or surprise increase the entertainment value of playing a match?
In a recently introduced theoretical economic model of
suspense and surprise, Ely, Frankel and Kamenica (2015)
argue that one reason people consume non-instrumental
(i.e. useless) information is because they derive entertainment value from suspenseful and surprising experiences.
Current Status: After several conversations with the data
provider throughout the winter, we submitted our final data
request in early April and are waiting to receive raw data
(expected June 2015).
Gameplay screenshot
Objective: Destroy opposing team’s base (called the
“Nexus”), located in the opposite corner of the map.
True state
Belief at time t about
probability of true state
Belief path throughout the
game (ends at true state)
“Anticipated beliefs”
at time t
Preference for Suspense: Utility increases with increases
in the variance of anticipated beliefs about the true state.
Setup: Players summon/choose a champion to play with
and join or create a team. The champions are all sci-fi /
fantasy characters (see left).
Arena: Each match involves two teams of 5 battling on a
symmetric map called Summoner’s Rift (see below).
Preference for Surprise: Utility increases with increases in
the distance between current beliefs and previous beliefs.
5. Testing Playing with Friends: We anticipate that
social effects will be straightforward compared to suspense and surprise. We will simply observe the effect of
joining a game with friends on satisfaction ratings and the
likelihood/hazard that that they immediately play again.
Match loading screenshot, top team vs bottom team
Example of high suspense: Djokovic beats Federer
References
Bainbridge, William Sims. "The scientific research potential of virtual worlds." science
317.5837 (2007): 472-476.
Castronova, Edward. "On virtual economies." 752 (2002).
Castronova, Edward. Synthetic worlds: The business and culture of online games. University of Chicago press, 2008.
Castronova, Edward et al. "As real as real? Macroeconomic behavior in a large-scale virtual
world." New Media & Society 11.5 (2009): 685-707.
Ely, Jeffrey, Alexander Frankel, and Emir Kamenica. "Suspense and surprise." Forthcoming
in Journal of Political Economy (2013).
Ravaja, Niklas, et al. "Spatial presence and emotions during video game playing: Does it
matter with whom you play?." Presence: Teleoperators and Virtual Environments 15.4
(2006): 381-392.
Tibshirani, Robert. "Regression shrinkage and selection via the lasso." Journal of the Royal
Statistical Society. Series B (Methodological) (1996): 267-288.
To begin this process, we will run a series of predictive
models in order to identify the in-game variables that best
predict which team will ultimately win. Using this, we will
generate probability vectors of the outcome at each
moment in a game, and use this to generate scores for
overall suspense and surprise. Then we will test the model
by looking at whether the suspense and surprise scores
predict positive post-game player satisfaction ratings and
the likelihood/hazard that a player will immediately start another game.
Summoner’s Rift, bird’s eye view
Example of low suspense: Murray beats Nadal