Fantasy Sports: A Game of Skill or

Chance
A Project report by:

U DINESH KUMAR*, SHAILAJA GROVER*, SHARADA

SRINGESWARA , ALOLIKA PATTANAYAK , ABHINAV SINHA**,
* **

JIGAR PANCHAL** , JITHIN PAUL**, MOUMITA SARKER**, RAMSU

SUNDARARAJAN**, TAPAN KHOPKAR**

* Indian Institute of Management Bangalore

** Cartesian Consulting Pvt. Ltd.
Executive Summary

Introduction to Online Fantasy Sports (OFS) Industry in India

Fantasy Sports is a form of skill-based online sports game where sports fans can create their own
team made up of real-life players from upcoming matches. These virtual teams garner points
based on the actual statistical performance of players during the real-life match and winners are
determined accordingly. OFS Industry, though nascent in India, has already become immensely
popular amongst sports fans with exponential growth from 20 lakh Indians playing in 2016 to over
7 Crore Indians playing in 2019. By 2020, as per industry estimates, there will be 30 crore Indian
sports fans watching sports online, so we are sure that at least 33% of them (10 crore) will play
fantasy sports, compared to the US where 65% of sports fans play fantasy sports. India has
become the largest OFS market in the world, surpassing USA’s 5 Crore Fantasy Sports user
base. Dream11, an Indian unicorn is now the largest Online Fantasy Sports company in the world
with over 5 Crore user base. Even the number of OFS operators has exponentially risen from 10
to 140+ in less than 3 years and expected to double by 2020.

Industry Drivers:

1. Digital Infrastructure: India has made exponential growth in terms of its digital
infrastructure. In recent times, sports & technology have converged beautifully to provide
Indian sports fans with an immersive and engaging experience, successfully bridging the
offline-online sporting divide. Growing affordability of smartphones, expansion of the
internet user base and plummeting data prices are fueling the growth. The internet
subscribers increased from 368 million in September 2016 to 560 million in September
2018 in India, which has helped fantasy sports to flourish in the digital gaming space.

2. User Demographics: OFS is being consumed alike by fans across metro and non-metro
cities. OFS users are mostly in the age range of 18-35 years and are tech-savvy
smartphone users. Users get to play the role of the all-important manager/selector who
decides the members of a team based on the statistical analysis of performance of players.

3. Sports Leagues: With the number of sports leagues increasing significantly, franchises
across various disciplines are looking to connect with their fans in a more meaningful way.
The advent of T20 leagues has further fueled more interest in the minds of sports fans
with a shortened attention span and the need for instant gratification, which is why single-
match freemium fantasy cricket has worked so well in India.
Indian Regulatory Landscape:
The ruling by the High Court of Punjab and Haryana and subsequent dismissal of SLP filed
challenging the High Court order on Dream11’s fantasy sports format as a game of skill has
helped spread awareness about the legality of fantasy sports and has encouraged the
participation of more users. Recently, the division bench of the Honorable Bombay High Court
has also ruled that fantasy sports Dream11 format is a game of skill and does not amount to either
gambling or betting.

Skills dominance - Learning effect, Consistency, Strategies and

payoffs
The dominant factor test methodology, as used by the United States Government, has also been
adopted by Indian courts to analyze the skill vs chance debate. While every game has some
element of chance, where skill is dominant in determining the result of the game are considered
skill based. These are games where the user’s knowledge, judgment and analytical application
are the more dominant than that of mere chance.

Skill dominance is established by statistically testing for learning effect, consistency effect and
impact of strategic choice on pay-offs. Learning effect states that playing more regularly or
practicing improves the chances of winning (improves skill) in a skill dominant game but has no
effect in games dominant by chance. Consistency effect indicates that scores or payoffs of users
in a skill dominant game will be consistent and not random (like in a game of chance). And lastly,
skill-based games necessitate the use of strategies or a series of decisions that lead to greater
payoff. In a chance dominant game, payoffs are independent of the strategies employed by the
players.

This study, a first of its kind in India, employs statistical hypotheses techniques used on actual,
anonymized fantasy sports data to test the above skill dominance tests. Dream11, the biggest
Fantasy Sports company in India, agreed to share historical, anonymized data for the purpose of
this study.

Hypothesis test as a methodology and results from Dream11 live

data
The question of “Skill vs Chance” dominance was converted into an analytics & statistics problem
on which hypotheses testing techniques like 2 sample t-test, Chi-square test and ANOVA were
applied.
A snapshot of business hypotheses, techniques used and result
Skill
Technique
Problem construct dominance Hypothesis Result
used
effect tested
Ho: No difference in
average scores of paid Since average scores of free and
and free contestants paid contestants are statistically
Performance of users different, we can infer that paid
of fantasy sports is Consistency (u1=u2) 2 sample t- contestants are using some
consistent (either Effect test strategy (such as selecting high
good or bad) performers) and hence are
Ha: Average scores of consistently scoring more than
paid and free contestants free contestants.

are different (u1 ≠ u2)

Number of rounds played has a
Scores of the users Ho: There is no
Linear significant effect on the chances of
improve as they play Learning relationship between
regression a user to be in the top percentile of
more number of Effect winning a round and the
model scores, hence proving a learning
games. number of rounds played.
effect
Ho: Scores of strategic
selection and random
Player performance Knowledge and skill have a
Selection bias selection are the same
has an impact on 2 sample t- statistically significant effect on
based on
his/her selection in a test scores compared to a random
performance Ha: Scores of the strategic
team selection of players to form a team.
and random selection are
not the same
In a game of chance,
The average scores are not equal.
then the average Ho: No difference in
Specifically, there is large variance
percentile score will Non-random average scores of users
ANOVA observed when one of the top 4
converge towards a payoffs Ha: Not all the user scores
players is either selected or not
coin-toss outcome are the same
selected as captain/vice-captain.
(50-50) for all users
Ho: Selection of players
and winning (high scores) Top scorers are playing
Selection of a player Chi-square
Impact of are independent of each strategically and choosing
as a captain or vice- test of
strategy on other. players as captain or vice-captain
captain increases the independenc
the payoff Ha: Selection of players with skill rather than randomly to
scores of users e
and winning (high scores) achieve high scores
are not independent.
The tests show that users learn and fare better as they play more. They use their understanding
and knowledge of the game to select players, captains, and vice-captains. With more stakes, paid
users to perform better than free users indicating the usage of a strategy.

Conclusion

With various hypotheses tests, there is enough evidence to conclude that performance in fantasy
sports is skill dominant. There is strong analysis that provides evidence of learning, consistency
and strategy-based usage all of which affect payoffs. There is little to no evidence of randomness
or chance. This case has already been published in Harvard Business Publication and is available
across more than 410 universities across 75 countries. The teaching note and case study is also
being used by Prof U Dinesh Kumar (protagonist author) as part of his student curriculum in IIM
Bangalore.
Project Report

INTRODUCTION
Fantasy sports is the fastest growing online gaming industry worth several billion dollars. An
important debate associated with fantasy sports across the world is whether it is a game of skill
or chance. In India, online gaming laws are not very well defined. While the game of skill is
permitted by Indian laws, a game of chance involving money is strictly prohibited. Indian courts
have recognized that no game is a game of pure skill or game of pure chance.1 When there is a
chance involved, the Indian courts decide based on the dominant factor test2 that requires
determining whether the skill is dominant, or chance is the dominant factor for a given instance.
Can this debate around whether the dominant factor is a skill or chance be settled using data?
With this objective, Dream11 which is one of the largest fantasy sports companies in India was
approached to understand the game as well as to collect data to test some of the hypotheses
which can help differentiate if in fantasy sports the dominant factor is skill or chance. Dream11
has created fantasy games for sports such as cricket, football, hockey, kabaddi, and basketball.

The main concern for lawmakers across the world was whether fantasy sports was a game of skill
or game of chance. Edelman in 2016 claimed that fantasy sports were a nerds’ game3. Similarly,
Jennifer Chu4 claimed that fantasy sports involve real skill. Authors such as Edelman and Jennifer
Chu believed that the players of fantasy sports would need a better understanding of the sports,
how the players perform under different playing conditions, injury to players, and the impact of
weather and other playing conditions. For example, till June 2019 Rafael Nadal won 12 of his 18
grand slam titles in the French Open, whereas Roger Federer won just one French Open out of
his 20 grand slam titles. This is an example to show that fantasy sports enthusiasts should be
aware of the strengths and weaknesses of players before selecting them. Through our study we
want to see if data could be used to check whether fantasy sports was a game of chance or skill,
especially whether skill is a dominant factor in winning fantasy sports competition as claimed by
Edelman.

1
Source: Supreme Court of India, Dr K R Lakshmanan Vs State of Tamil Nadu and ANR on January 12, 1996, available at
https://indiankanoon.org/doc/1248365/
2
G Gokhale et al. (2018), “The laws related to Fantasy Sports Games in India”, The Sports Law & Policy Centre,
3
M Edelman, “A Sure Bet? The Legal Status of Daily Fantasy Sports’, Pace Intellectual Property, Sports Entertainment
Law Forum”, 6(1), 1-21, 2016.
4
Jennifer Chu, “There is real skill in fantasy sports”, Science Daily, November 7, 2018. Accessible at
https://www.sciencedaily.com/releases/2018/11/181107093816.htm.
ABOUT THE FANTASY SPORTS INDUSTRY

Fantasy sports is was a competition game among fans that involves building an imaginary team
comprising players for upcoming real match game. Based on the combined l performance of the
players in a match, the fans are awarded points and they can win a prize5. History of fantasy
sports can be traced to the 1960s when Bill Gamson, former Professor of Psychology at Harvard
and Michigan University constructed a game called “The Baseball Seminar”6. Daniel Okrent,
student at the Michigan University learned about “Baseball Seminar” and created his own version
of the fantasy sports called “rotisserie baseball,7” whose idea had been crystallized in a Manhattan
restaurant named La Rotisserie Francaise8.

In 2018, the fantasy sports market worth was USD 13.9 billion and expected to grow at a CAGR
of 13.7% reaching USD 26.4 billion in 2024, according to a study carried out by Market Study
Report LLC9. A study conducted by Fantasy Sports and Trade Association, claimed that the
fantasy sports users grew at 25% annually10. With such high anticipated growth, channels such
as ESPN broadcast hour-long shows devoted to fantasy sports and players prior to major sports
seasons.

In the past 5-6 years, fantasy sports has become one of the most popular activities for sports
lovers in India. The major reason for this surge in fantasy sports users could be the rise of league-
based sports, and the growth of online viewership combined with increased smartphone
penetration. Technological advancement has also contributed to the growth in popularity of

5
Source: David O Klein et al., “Fantasy Sports – Rapidly Developing Legal Framework”, Law 360, September 20, 2017,
accessible at https://www.law360.com/articles/704275.
6
Justin Fielkow, “From Fantasy Sports to Reality: The evolution and legality of fantasy sports”, The Sports Esquire,
May 18, 2015 available at https://charlesfranklinlaw.com/wp-content/uploads/2016/12/2015.05.18-From-Fantasy-to-
Reality_-The-Evolution-and-Legality-of-Fantasy-Sports-Fielkow-The-Sports-Esquires-Revised.pdf
7
Ibid
8
Source: History and Evolution of Fantasy Sports available at http://futureoffantasy.com/the-history-and-evolution-of-
fantasy-sports
9
Source: https://www.marketwatch.com/press-release/at-137-cagr-fantasy-sports-market-size-is-anticipated-to-reach-
us-26400-million-by-2025-2019-03-29
10
Source: D Heitner, “The Hyper Growth of Daily Fantasy Sports is Going to Change Our Culture and Our Laws”,
Forbes, September 16, 2015. Available at https://www.forbes.com/sites/darrenheitner/2015/09/16/the-hyper-growth-of-
daily-fantasy-sports-is-going-to-change-our-culture-and-our-laws/#3331b7615aca
fantasy sports in India. While cricket has dominated the sports market in India, the
new generation is exploring and getting interested in other sports such as Kabaddi, football,
basketball and hockey.

The Indian fantasy sports market has shown a double-digit growth. A report by AC Nielsen,
commissioned by the Indian Federation of Sports Gaming (IFSG) across 12 cities in India claimed
that 67% of the survey respondents were aware of fantasy sports and the retention rate of users
was 89%11. The same report claimed an increase in number of users by 200% in 2017. While the
US and Canada had 59 million users in 2018, the Indian market saw an increase in users from 2
million users in 2016 to 20 million in 2018, more than the number of users in the UK12. An
interesting fact is that fantasy sports was witnessing 40% growth in female user participation
during the International Cricket Council (ICC) World Cup13. In 2019, India had more than 80 million
fantasy sports users across 130+ different platforms, which was expected to grow to 100 million
users by 202014.

Fantasy sports and the gaming industry was set to witness a massive growth in India, growing
from Rs. 43.8 billion to reach Rs 118.8 billion by FY23 at CAGR 22.1%, according to a report by
Indian Federation of Sports Gaming (IFSG), India’s first and only self-regulatory industry body for
the sports gaming sector and KPMG India Private Limited (KPMG). The report titled ‘The Evolving
Landscape of Sports Gaming in India’ provides an overview of the online gaming industry with a
focus on fantasy sports and eSports15.

FANTASY SPORTS COMPANIES IN INDIA

There were around 150+ online fantasy sports gaming companies in India such as Dream11,
MyTeam11, HalaPlay, BalleBaazi, Fanmojo, and 11Wickets. Dream11 is the biggest fantasy
sports company in India. Dream11 was co-founded by Harsh Jain and Bhavit Seth in 2008; and
in 2012, they introduced the freemium format of fantasy sports to the Indian cricket fans. In 2014,
the company reported 1 million registered users, which grew to 2 million in 2016 and to 60 million
in 201816 17.

11
Source: “Scoring Big with Sports Gaming”, A report on emergence, consumption, patterns, meteoric growth and
future of fantasy sports in India”, available at https://www.ifsg.in/wp-content/uploads/2018/03/IFSG-Report_33-low-
res.pdf
12
Ibid
13
Ibid
14
Source: Sporting Revolution in India available at https://www.thehindu.com/brandhub/a-sporting-revolution-in-india-
tracking-the-growth-of-fantasy-sports/article24026469.ece
15
Source: The evolving landscape of Sports Gaming in India available at https://www.ifsg.in/publications/
16
Source : https://www.outlookindia.com/newsscroll/fih-announces-partnership-with-dream11/1430388
17
Source : https://scroll.in/article/919582/big-game-how-a-fantasy-sports-startup-is-making-money-from-
indias-ipl-fever
In 2019, it was reported that Dream11 held 90% of the domestic fantasy sports market18. In 2018,
Dream11 signed MS Dhoni, India’s ex-captain as their new brand ambassador and subsequently
launched a media campaign under the banner “Dimaag se Dhoni” during Indian Premier League
201819.
HalaPlay was founded by the students of Birla Institute of Technology (BITS) Pilani in 2017. In
2019, HalaPlay reported 4 million users and 10 times growth over the last 12 months20.
MyTeam11, a Jaipur based startup was founded in 2016 by two engineers and had around 10
million+ users.

ONLINE GAMBLING LAWS ACROSS THE WORLD

In USA, fantasy sports were exempted from federal law concerning the Unlawful Internet
Gambling Enforcement Act (IGEA)21. It was left to a state-level decision; each state applied
various standards to determine if fantasy sports were a game of skill or chance. The decision
between skill and chance was to be taken based on whether the skill-based element was
dominant over chance in determining the outcome of the game22.

18
Source: The Rise of Fantasy Sports in India and Startups to Watch for in the Space, available at
https://yourstory.com/2019/04/fantasy-sports-startup-dream11-mpl-ipl
19
Source : https://en.wikipedia.org/wiki/Dream11
20
https://yourstory.com/2019/04/fantasy-sports-startup-dream11-mpl-ipl
21
Jonathan Griffin, “The Legality of Fantasy Sports”, National Conference of State Legislatures”, 23(33),
September 2015.
22
Source : https://en.wikipedia.org/wiki/Gambling_in_the_United_States
Most European countries recognized fantasy sports as a legal business; it was legal and highly
regulated in around 30 countries in Europe. In Sweden, Croatia and Norway, though fantasy
sports was legal, it was not very regulated. However, in countries such as Switzerland, Ukraine,
Macedonia and Belarus, online gaming was considered illegal23.

Sports & Entertainment and Gambling & Betting have been a state subject in India and respective
states can formulate laws for various gaming activities. Most of the laws governing the subject of
gambling in India in 2019 were old. The said laws used a yardstick of 'Skill vs. Chance' and
predominance of skill over chance to identify if a given game falls within the ambit of gambling.
Advent of online games coupled with lack of awareness about them, created confusion as to how
this yardstick of skill versus chance or game of mere skill could be applied. The two main
enactments dealing with gaming in India were the Pre-Independence Public Gambling Act, 1867
(“PGA”) and the Prize Competitions Act, 1955 (“PCA”). In general, gambling laws of most states
prohibited gambling, but the laws created a carve out for games involving 'mere skill' from the
applicability of provisions of gambling Act.' Historically, Indian laws have differentiated between
the games of skill and games of chance. While the game of skill was permitted by Indian laws,
game of chance was strictly prohibited.

SKILL BASED VS. CHANCE BASED

Games of skill are the games which involve a person’s skill, knowledge and judgment and they
rule out chance aspect from the game. On the other hand, games of chance include games that
are determined by mere luck, completely uncertain and the players cannot apply their skill to
estimate the result24.

While deciding whether a game is a game of chance or skill, the Indian courts while interpreting
the term “mere skill” have adopted the methodology used by the US government, that is, the
“dominant factor test”. It says that while classifying a game, it becomes important to check whether
chance or skill is the dominant factor in determining the result of the game. Dominant factor test
was applied to Rummy and Horse Racing. The Supreme Court of India classified Rummy as a
game of skill, as the fall of cards need to be memorized and considerable skill is required while
deciding whether to hold or discard a card25. On similar lines, horse racing requires objective
assessment of fitness and skill of the horse and the jockey.

23
Source : https://blog.vinfotech.com/fantasy-sports/european-laws-on-fantasy-sports
24
Source : https://www.ifsg.in/wp-content/uploads/2018/03/FantasySportsPublication.pdf
25
Source: K R Lakshmanan Vs State of Tamil Nadu, available at https://www.sci.gov.in/jonew/judis/16203.pdf
In 2018, the Supreme Court of India concurred with earlier verdicts by Punjab and Haryana High
Courts that Fantasy sports involves substantial amount of skill.26 The High Court of Punjab and
Haryana ruled that fantasy sports was predominantly a skill-based game. It is pertinent to note
here that it is difficult to conceive of any game of skill as a game of pure skill and no chance.
Some element of chance is always present in any game of skill. For instance, almost all forms of
popular sports involving very high degrees of skill, such as chess, cricket, football, hockey, etc.
also involve some element of chance. It becomes important to obtain evidence from data by
developing statistical models to test and classify fantasy games as a game of skill or game of
chance. If a fantasy sports is chance based, then every user should have an equal probability of
winning, whereas if it is skill based, then one should see consistent performance among the users.
The only way to check this is to use the data from a fantasy sports company.
To undertake this study Dream11 was approached with a request to share anonymized data for
testing whether fantasy sports involving cricket was skill dominant or chance dominant. Dream11
agreed to share their data and rules related to fantasy sports based on a cricket series, which are
described in sections below.

DREAM11: LET’S PLAY (Rules and constraints)

The following steps were to be adopted to play a game in Dream11.

● Select A Match: Select any of the upcoming matches from any of the current or upcoming
cricket series.
● Create Your Team: Create a team with all available players for a match with the following
constraints.
o Each player comes at a price, a budget of 100 credits is available to create a team
of 11 players.
o User can create multiple teams and choose to join a contest with any of the teams
created.
o After creating your team on Dream11 platform, choose a Captain and Vice-Captain
of your choice for the team.
▪ Captain – Gets 2x points scored by him in the actual game
▪ Vice-Captain – Gets 1.5x points scored by him in the actual game
o There will be multiple contests for a match, User needs to select a contest and
assign a team.

26
Source: G Gokhale and Rishabh Sharma, “The skill element in the fantasy sports games”, The Sports
Law & Policy Centre”, available at
http://www.nishithdesai.com/fileadmin/user_upload/pdfs/NDA%20In%20The%20Media/News%20Articles/180406_A_
Legality_of_Fantasy_Sports_India.pdf
 o You can make as many changes to your teams created on Dream11 platform as
you wish until the deadline of that match. You can also change your Captain or
Vice-Captain before the deadline of the match.
o Contests would be open for submitting a team till the deadline of that match. There
can be no changes to submitted team or possibility of adding new team after the
start of a match.
o Team submission can be done till slots are available in a contest.
o Create your Dream11 team by picking 11 players as per the following combination
C2, ns (C1, C2, C3...) within a budget of 100 credits. Maximum number of players
from a team cannot exceed 7. In other words, your team should necessarily consist
of players from both the real game participating team.

Player type C1 C2 C3 C4 C5 C6 C7
Wicket Keeper 1 1 1 1 1 1 1
Batsman 5 5 4 4 4 3 3
All Rounder 2 1 1 2 3 2 3
Bowler 3 4 5 4 3 5 4
Total 11 11 11 11 11 11 11

Dream11 :DATA HIERARCHY

The data available for analysis was provided in various tables. The overall hierarchy of data is as
follows.
- Various rounds/match were played. For example, one IPL match would be one round.
- There were players who were available to be picked up for a round or match. (This
number is more than the number of players who would play in that match, so it was
possible that a player selected by a user in his team may not actually play.)
- For every round, multiple contests were opened. The contests were of different
categories, from free to paid, and various types of playing and winning options (public,
private, special).
- User selected a team for a round and for a contest.
- There was a player round performance table which indicated how the player performed
in the specific round.
- Teams selected by users were scored based on the selected player’s performance in
a contest and those team level scores were provided in the contest user’s table.
Data Description:

Brief description of the data provided by Dream11

● Data tables
o CONTEST MASTER – Provides details of a contest such as if the contest is free
or paid, public or special or private. It also has details on the size of the contest.
o CONTEST USER RELATION – Provides details points scored by a user in a
contest for the team which has been fielded.
o PLAYER ROUND PERFORMANCE – It has details on performance of a player
in a round.
o ROUND MATCH – Association between roundid and matchid which denotes
number of matches played for a specific round or game. The scores are given
for a match, hence to get the scores for a user team, we need to map roundid
with matchid using this table.
o USER ROUND PLAYER – This table has all the players selected by user or the
team which is formed by the user for a round.
o USER ROUND TEAM – This table contains the date on which the user formed
the team, this table is mainly used for joining all other tables since it contains all
the required information about a round→user→team.
o AGGREGATE_DATA – This table has aggregate data which provides details of
user’s performance across multiple rounds.

● Attributes description
● roundid – Unique identifier for round played
● matchid – Unique identifier for matches played in a round
● contestid – Unique identifier for a contest. A round can have multiple contests.
● Contesttype – Public/Special/Private
● Contestcategory – Free/Paid
● teamid – Unique identifier for a team. User can create multiple teams for a round.
● currentpoints – Points scored by the user for the selected team in a contest for a
round
● userid – Unique identifier of a user
Details of data in all the tables provided by Dream11:

2 Rounds 5365 & 6349 have been considered. These 2 rounds are having high numbers
of users playing the game.
● 5365 - Is a T20 Premier League match played on 29/05/16?
● 6349 - Is a ODI IND vs NZ match played on 29/10/16?
File: USER_ROUND_PLAYER.CSV
This table has information on team formed and fielded (11 players along with captain
and vice-captain) by the user for a round/match
● It has 531425 rows with 17 columns.
● 2 unique round ids
● 3006757 unique users
● Users having as many as 4 teams per round
File: CONTEST_MASTER.CSV
This table has information on the contest which is open for a round/match. It also has
details on type and size of each contest.
● It has 74220 rows and 5 columns.
● 2 unique round ids
● 74220 unique contests
o 5365 has 42247 unique contests
o 6349 has 31973 unique contests
● 4 unique contest types (public, private, special, grand)
● 2 contest categories (paid and free)
File: CONTEST_USER_RELATION.CSV
This table has information on points scored by a user for a team formed, in a
round/match and contest combination.
● It has 962602 rows and 6 columns.
● 2 unique round ids
● 72652 unique contest
● 239670 unique user ids
● 4 unique contest types (public, private, special, grand)
o special - 699330 records
o public - 160271 records
o grand - 84051 records
o private - 18950 records
● Max number of points scored by a user is 447.25, minimum being 0.
● Max number of teams formed by a user is 4.
File: PLAYER_ROUND_PERFORMANCE.CSV
This table has information on points scored by a player in a round/match. It has detailed
break-up of how many points player scored for bowling, batting and fielding.
● It has 44 rows and 43 columns.
● 2 unique match ids
● 43 unique players
 ● Highest point scored by a player is 72, minimum being -1.
File: round_match.csv
This table has mapping between matchid and roundid.
● It has 2 rows and 2 columns.
● 2 unique match ids
● 2 unique round ids
File: USER_ROUND_TEAM.CSV
This table has information on date of match, user and round/match.
● It has 285630 rows and 4 columns.
● 2 unique round ids
File: AGGREGATE_DATA.CSV
This table has aggregated information of user’s performance across multiple rounds.
● It has 385235 rows and 17 columns.
● 385235 unique user ids

METHODOLOGY

The main methodology used to test if Fantasy Sports is a game of skill or chance using data
provided by Dream11 was to create a couple of hypothesis which tests if Fantasy sports users
are applying any skill in choosing a team for a game and which gets reflected in their scores. Is
there any learning affect which is reflected in the scores of players playing regularly? These
assumptions are tested by creating various hypothesis tests and getting appropriate data sets
from the entire data provided by Dream11. Testing those hypotheses on these data sets created
by slicing and dicing the data on various parameters using different hypothesis tests help us reach
conclusions which are data driven.
Few possible hypotheses which helps us test skill vs chance are:
1. Users playing free contests are scoring lower than users playing paid contests. This can
prove that when users play paid contests, they play more cautiously and strategically, and
do not select teams at random.
2. Scores of randomly selected players can be tested against scores of the teams based on
a specific strategy such as selecting players who have performed well in the recent
matches.
3. Is the selection of captains and vice captains of the team random (equal probability)?
4. Selection of players and winning or getting high scores are dependent on each other.
5. As the user plays more games, his chance of winning increases (learning effect).
Various hypothesis tests and inferences

First of all this question of “Skill vs Chance” can be converted into a data analytics problem where
techniques like hypothesis testing helps in solving the dilemma. The output of fantasy sports is in
the form of scores of individual users based on their team selection, which eventually gets
converted into the money they earn. Scores will be our key variable (outcome variable), which will
drive whether the scores are random or are an output of a more strategic team selection. The
following are the possible approaches for testing skill vs. chance.
a. Consistency in performance of users of fantasy sports, that is, if it is skill dominant, the
scores will be consistent.
b. Selection of players with best performance in the recent matches and the scores of users.
c. Use of strategy such as selecting players based on past performance, as vice-captain and
captain.
d. Improvement in scores as the number of games played by users increases.
e. Average scores whenever a player is chosen or not (dependency of scores on selection of
a specific player).

Few sets of hypotheses as listed below can help us test the concept of “Skill Vs Chance”. Below
table explain how these hypotheses are linked to our problem and how the data provided can
help us test these hypothesis
Table 1 - Hypotheses for checking whether fantasy sports is skill based or chance based

Hypothesis Link to skill vs. chance Constructing the hypothesis

test using the data
Performance of In a game of chance, Users in the top quartile of
users of fantasy performance will be multiple rounds can be chosen
sports is inconsistent, whereas in a game and tested on the percentage of
consistent (either of skill, the performance will be times they have been in the top
good or bad) consistent. quartile.
Scores of the Continuous learning is a very This hypothesis can be tested
users improves as important aspect of skill. We can using two approaches. (1) Build
they play more check whether there is evidence a regression model and check if
number of games. that with more number of games scores of a user improve with
played, the score of a user number of games played. (2)
improves. Conduct a chi-square test of
independence for scores and
number of rounds by creating
appropriate bins.
 Player This can be used to prove that Data on player performance can
performance has the selection of player is not be correlated with number of
an impact on his random. users selecting the player.
selection in a
team.
Average This can be used to prove that If fantasy sports is a game of
percentile score of the fantasy sports is not a game chance, then the average
users is not 0.5. of chance. percentile score will converge
towards 0.5 for all users.
Selection of a Skill is required to choose the ANOVA can be used to check
player as a right player as captain or vice- average scores of users and
captain or vice- captain their selection of captain/vice-
captain increases captain.
the scores of
users.

These are few of the claims, there can be many more. We will test a few using the data provided.

1. First Hypothesis : As we know Dream11 platform has both free and paid users, the users who
play games for free with no return and users who pay a fee and obtain returns at the end of
the game based on their relative performance. If we can statistically test that scores of paid
users are significantly higher than the scores of free users, then we can say users who play
for money play a more strategic game and score more. To test this hypothesis, we will
compare the average scores of paid and free contest users and check if there is a statistically
significant difference between their average scores.

We will create two samples of data, one of the scores of paid contest users and another of scores
of free contest users and conduct a 2-sample t-test with unequal variance to test the hypothesis.
Ho: No difference in average scores of paid and free contestants (⎧1 = ⎧2)
Ha: Average scores of paid and free contestants are different (⎧1 ≠ ⎧2)

where ⎧1 is the average score of players who play free contests; and ⎧2 is the average score of
players who play paid contests. The number of users playing free and paid contests is shown in
Figure 1. The distribution of scores of paid and free contests are shown in Figure 2.The
hypothesis result output is shown in Table 2. The p-value is very small; so we reject the null
hypothesis, which means the average scores of paid users playing paid contests is more than the
average scores of players playing free contests.
Figure 1
Number of users of free and paid contests for rounds 5365 and 6349

Figure 2

Distribution of scores of free (red line) and paid contests (blue line)
 Table 2
Hypothesis Test Result

t-stat value = 18.35175195627833

p value = 3/2088574221489635e-75
Mean of Paid contest type = 236.87613622680416
Mean of Free contest type = 235.16216597368555
Standard Deviation for Paid contest type = 133.3065137899964
Standard Deviation for Free contest type = 130.34312313429582

Key inference out of this test is - Since average scores of free and paid contestants are
statistically different, we can infer that paid contestants are using some strategy (such as selecting
high performers) as money is involved and hence are scoring more than free contestants.

2. Second Hypothesis is if the scores of users who use some strategy to select players such as
recent performance of players is higher than users who select players randomly. This
hypothesis is to test whether scores obtained by teams selected based on strategy and
knowledge of players’ past performance are higher than scores obtained by teams with
random selection of players.

In order to test this, we need to create two samples as described below:

- One sample is based on random selection of players from the available list of players for the
game and applying all the constraints of fantasy sports while creating the teams.
- Second sample is created by selecting the team based on the past performance of the
available players. Data of performance of two teams (in this instance, Gujarat Lions and
Sunrisers Hyderabad IPL teams were considered). Data on performance of all the players of
these two teams in the IPL matches was collected from the portal CricInfo. Based on the
performance, top players were selected for batsman, bowler, allrounder, wicket keeper,
captain and vice-captain categories.

For the second sample, we formed teams based on the following strategies:

o Team composition: 1 All-rounder, 5 Batsmen, 4 Bowlers, 1 wicket keeper, captain and vice-
captain
o Team composition: 3 All-rounder, 4 Batsmen, 3 Bowlers, 1 wicket keeper, captain and vice-
captain
- Top players based on their performance in the previous matches were randomly selected and
teams were formed using the selected strategy. In this sample, the teams were formed with all
the constraints, but players were selected only from top performers of the previous matches.
- During the next match of these two teams (Gujarat Lions and Sunrisers Hyderabad) based on
the performance of these players in the subsequent match, scores were given to all the teams
in the two samples (teams with randomly selected players and teams with players selected
based on knowledge).
- 2-sample t-test was conducted on these two samples of scores obtained by these teams.

Ho: Scores of strategic selection and random selection are same.

Ha: Scores of strategic and random selection are not same.
Number of records in strategy users selection set = 36
Number of records in random selection set = 36
Results are as follows:
t-stat value = 11.331975795254104
p-value = 2.056343133916253e-17
Mean of strategy-based selection dataset = 416.6666666666667
Mean of random selection dataset = 240.95833333333334
Standard deviation for strategy-based selection dataset = 68.490368341
Standard deviation for random selection dataset = 61.02314849938323

The p-value of the hypothesis tests is very low; hence, we reject the null hypothesis and conclude
that the average scores of selecting players based on the recent performance is more than the
random selection of players.

Key Inference out of this test is - that knowledge and skill have statistically significant effect on
scores compared to random selection of players to form a team.

3. Third hypothesis - If fantasy-sports is a game of skill, then player performance plays a major
role in the player getting selected to a team as well as selection of captain or vice-captain. If
by using the data, we can test if selection of players in a team and getting high scores are
dependent on each other we will be able to prove that users are selecting their team players
based on their skill of knowing and understanding past player performance.

In this hypothesis, we are assuming that the right selection of players increases the chance of
higher scores for a user and hence selection of players and scores are not independent of each
other. Similarly, users with high scores have chosen high performing players as captain or vice-
captain.
Ho: Selection of players and winning (high scores) are independent of each other.
Ha: Selection of players and winning (high scores) are not independent.

We can perform a Chi-square test of independence to test whether selection of players and scores
are independent. In this instance, we select one player and create a contingency table listing the
number of times the player was selected in a team by users and number of times the scores of
these users were in the top quartile. Similarly, number of times a player was not selected by users
and the score was in the top quartile was listed.

Player ID 635: (We have chosen a specific high-performance player and tested the hypothesis
for this player being selected as a team member and also as a captain and its impact on the
scores of the users who selected him). The contingency table is provided in tables 3-4 below.

Table 3

Selection of player ID 635 and scores

Table 4

Player ID 635 selected as captain

The chi-square statistic values from above tables are very high. Hence, we reject the null
hypothesis that the scores of fantasy sports users and selection of specific players in the team
are independent of each other.

Similar chi-square test of independence was carried out for several players and chi-square
statistic values are shown in table 5 below. In all these instances, the null hypothesis is rejected.

Table 5
Chi-square test of independence for several players
Key Inference out of this test is - This hypothesis test establishes that player selection has a
huge influence on achieving high scores. Top scorers are playing strategically and choosing
players with skill rather than randomly to achieve high scores.

4. Fourth Hypothesis – We can also test the reverse hypothesis than what we tested in the
previous test. If choosing a poor performing player as a captain or vice captain has any affect
on the scores of the users. We can test the above hypothesis by checking that users who
selected poorly performing players as captain or vice captain and their scores are dependent
on each other.

In this hypothesis, we are assuming that the poor selection of players reduces the chance of
higher scores for a user and hence selection of players and scores are not independent of each
other. Similarly, users with low scores have chosen poor performing players as captain or vice-
captain.

Ho: Selection of players and loosing (poor scores) are independent of each other.
Ha: Selection of players and loosing (poor scores) are not independent.

We can perform a Chi-square test of independence to test whether selection of players and scores
are independent. In this instance, we select one player who performed poorly in previous games
and create a contingency table listing the number of times the player was selected in a team by
users and number of times the scores of these users were in the top quartile. Similarly, number
of times a player was not selected by users and the score was in the top quartile was listed. For
this test we chose poor performing players from past data and tested hypothesis on sample data
where those players are selected in the team and selected as captain of the team.

Player ID 111: (We have chosen a specific poor-performance player and tested the hypothesis
for this player being selected as a team member and its impact on the scores of the users who
selected him). The contingency table is provided in tables 6 below.
Table 6 :

Round
id 6349
player
Id 111
Not
Selecte Selecte
d d Total
TopQ 101 52921 53022
Others 2708 164149 166857
Total 2809 217070 219879

Segme (Oij -
nt Class Oij Eij Eij)^2/Eij
Selecte 490.42686
TopQ d 101 677.367088 93
Selecte 155.84250
Others d 2708 2131.63291 86
NotSele 6.3463817
TopQ cted 52921 52344.6329 01
NotSele 2.0166840
Others cted 164149 164725.367 5
Chi-square 654.63244
Statistic = 36
DF = 1

Player ID 731: (We have chosen a specific poor-performance player and tested the hypothesis
for this player being selected as a captain and its impact on the scores of the users who selected
him). The contingency table is provided in tables 7 below.

Round id 5365
Captain id 731
Not
Select Selecte
ed d Total
TopQ 0 58336 58336
Others 65 187424 187489
Total 65 245760 245825
 (Oij -
Segment Class Oij Eij Eij)^2/Eij
Select 15.42495
TopQ ed 0 15.4249568 68
Select 4.799376
Others ed 65 49.5750432 38
NotSe 0.004079
TopQ lected 58336 58320.575 68
NotSe 0.001269
Others lected 187424 187439.425 37
Chi-square 20.22968
Statistic = 22
DF = 1

The chi-square statistic values from above tables are very high. Hence, we reject the null
hypothesis and can confirm that scores of fantasy sports users and selection of specific players
in the team are dependent of each other. In both cases poor performing players were selected
and we saw that users who selected those players in their team and their chances of being in the
top quartile is very low, especially in case of captain selection. First of all, very few people have
selected poor performing player as captain and whosoever has selected has never been in top
quartile.

Key Inference out of this test is - This hypothesis test establishes that player selection has a
huge influence on achieving high scores. Top scorers are playing strategically and choosing
players with skill rather than randomly to achieve high scores. Poorly selected teams can never
bring a user in top quartile of scores.

5. Fifth Hypothesis – In order to test that selecting top performing players in your team can
increase the chances of getting high scores we need to check if the top scorers of Fantasy
sports are choosing top performing players and that will prove that skill is needed to get to the
top scorer spot.

This hypothesis aims to test if the average scores of users selecting top players and not selecting
top players are equal. In a game of skill, this hypothesis will be used to establish the fact that high
score is dependent on the team which is selected by the users and not random.

For one top player, we will divide the dataset into two samples; one with the scores of users who
selected the player in the specific rounds where he played and the other sample with the scores
of the users who did not select the player in the specific round and we perform a 2-sample t-test.
Figure 3 shows average scores of users who have selected player ID 635 as Captain/Vice-
Captain. Figure 4 shows the number of users who have selected player ID 635 as Captain/Vice-
Captain.

Figure 3
Average scores of users when player ID 635 was either selected (Y) or not selected (N) as
captain/vice-captain for rounds 6349 and 5365

Figure 4
Number of users who have selected (Y) or not selected (N) player ID 635 as captain/vice-
captain for rounds 6349 and 5365

Ho: Average scores of users who selected the top player and not selected are equal.
Ha: There is a difference between the scores of the users who selected the top players and who
did not select them.
To test this hypothesis, the following process is used

- Top 20 high scoring players are selected

- For each round
o Matches or Round in which the players are selected
o For each round in which the top players played
▪ Select users who participated
▪ Two samples created with the scores of users who selected the players and users who
did not select the players
▪ 2-sample t-test conducted for the 2 samples created from multiple rounds

The results of hypothesis test are shown in table -7 below.

Table 7

Results of hypothesis test

From this table, it is evident that for all the top 20 players, the p-value for 2-sample t-test is very
small which means we reject the null hypothesis. We can infer that the average score of users
selecting top players is not equal to the average score of users not selecting top players.

Key Inference out of this test is - The average scores of users who choose top players and who
do not choose the top players are statistically significantly different. This implies that the users
have to identify the high performers using their skill in order to win or score high.
Sixth Hypothesis- Choosing of right captain and vice-captain is very key to getting high scores in
this game since the scores of captain and vice-captain are multiplied two times and one and a
half times respectively. If we can test that high scorers have chosen captain and vice-captain
based on the past performance of the players, then we can claim that skill is a dominant factor in
forming a team mainly the captain and vice-captain.

This hypothesis is to find if there is a difference between average scores of users who are
selecting top 4 players as captain or vice-captain and others who are not. There will be three
samples of average scores of users who selected three top players as captain or vice-captain and
one for users who did not choose any of those three players.
This hypothesis test uses the Analysis of Variance (ANOVA) to find the average scores of users
selecting top players as captain/vice-captain and users not selecting them as captain/vice-
captain.

Hypothesis

Ho: No difference in average scores of users

Ha: Not all the user scores are same

To carry out ANOVA, we will create 3 samples: 1. Scores of users who selected this top player
as captain, 2. Scores of users who selected top player as vice-captain, and 3. Scores of users
who did not select the top player either as captain or vice-captain.

We have identified top 4 players who may be chosen as captain and vice-captain. The following
process was used to perform ANOVA.

- For each of these players

o All the rounds in which these players played are selected.

o All the users who played in the selected round are chosen.
o For all those users based on the factor if one of the top 4 players is chosen as captain or
vice-captain or not, three groups are created.
▪ Group 1: Scores of users who selected the player as captain
▪ Group 2: Scores of users who selected the player as vice-captain
▪ Group 3: Scores of users who did not select the player as captain or vice-captain

ANOVA was performed between these three groups for all the 4 players separately. The results
are provided below:
Result

1. Player: 635
a. F-statistic = 712301.1032301941
b. P-value = 1.1102230246251565e-16
2. Player: 119
a. F- statistic = 165083.02614741513
b. P-value = 1.1102230246251565e-16
3. Player: 342
a. F- statistic = 217211.11978071192
b. P-value = 1.1102230246251565e-16
4. Player: 6
a. F- statistic = 21118.936498018324
b. P-value = 1.1102230246251565e-16

ANOVA results shows that p-value is very small for all 4 instances; so, we need to reject the null
hypothesis that the average scores are not equal when top 4 players are either selected or not
selected as captain/vice-captain.

A couple of other hypothesis tests were performed on selection of players as captain by high and
low performing users. The results, elucidated below, evidence that there is statistically significant
difference between the high performers and low performers when it comes to selection of captain.

Hypothesis: Captaincy selection and performance are dependent

The following contingency table is created for two sets of users: 1. High performers (scored
more than 350 points) and 2. Low performers (scored less than 150 points)
The high/low performers were then checked for selection of a specific player as captain (player
number 342). The corresponding contingency table is shown below:
Contingency Table
Player 342 Selected as Captain Total
Yes No
Uses with scores more than 350 12 27801 27813
Uses with scores less than 150 275 2456 2731
287 30257 30544

For the above contingency table, we can perform a chi-square test of independence. Chi-
square statistic value is 2685 and the corresponding p-value is almost 0. That is, there is
dependency between selection of captain and performance.

Hypothesis: High performers’ selection of captain is different from low performers

The following Table shows proportion of times a player is selected as a captain by high and low
performing users:

Player Code Proportion of users selecting the player as captain

High Performers Low Performers
5 0.002 0.009
109 0.002 0.018
309 0.40 0.0004
342 0.0004 0.1159
612 0 0.0615
635 0.3622 0.0307

From the above table, we can see that more than 76% of high performing users have chosen
either player 309 or 635 as Captain, whereas only 3% of the low performing users have chosen
players 309 or 635 as captain. 11.5% of low performers have chosen player id 342 as captain,
whereas less than 1% of high performers chose player 342 as captain.

Key Inference out of this test is - Based on the ANOVA results, we can claim that choosing
the right player as captain and vice-captain has an impact on the average scores of the users and
hence again skill of choosing a captain and vice-captain is a dominant factor than by chance
choosing these players.

6. Seventh Hypothesis – Learning effect is also a very important factor to prove skill as a
dominant factor. Since learning from my past team selections and then improving my future
selections show that analysis and intelligence goes in team selection and it is not random.

This question will enable us to understand that in a game of skill, there will be a learning effect,
that is, the users will learn the skill required for selecting the best fantasy team based on their
past experience. This can be checked by developing a simple regression model in which the
outcome variable is the number of wins (defined as top percentile or quartile) and the dependent
variable is the number of rounds played by the user.

The null hypothesis for the regression model is defined as follows:

H0: There is no relationship between winning a round and the number of rounds played.

The following pre-processing of the data was carried out for developing the linear regression
model.
 1. Winners are identified as users who are in top 5, 10 or 20 percentiles.
2. Data was extracted for users who have played at least 57 rounds.

The regression outputs are shown in below tables

Figure 5
Plot between number of rounds played and number of rounds in which the player was in
top 20 percentile.

A linear trend model is computed for Winn perc 20 (user is part of top 20 percentile score) given
Valid Rounds. The model may be significant at p ≤ 0.05. The regression output is shown in table
8 below and the corresponding plot is shown in Figure 5.
Table 8

Regression output in which winning is defined as top 20 percentile

(Valid Rounds +
Model formula:
intercept)
Number of modeled
55272
observations:
Number of filtered
0
observations:
Model degrees of freedom: 2
Residual degrees of freedom
55270
(DF):
SSE (sum squared error): 1.64241e+06
MSE (mean squared error): 29.7162
R-Squared: 0.0221533
Standard error: 5.45125
p-value (significance): < 0.0001

Individual trend lines

Panes Line Coefficients

Row Column p-value DF Term Value StdErr t-value p-value
Winn perc 2 Valid Round < 5527 Valid 0.005972 0.000168 35.385 <
0 s 0.0001 0 Rounds 8 8 7 0.0001
0.035541 <
intercept 20.2513 569.79
8 0.0001
Figure 6
Plot between number of rounds played and number of rounds in which the player was in
top 10 percentile

Trend Lines Model

A linear trend model is computed for Winn perc 10 (user is part of top 10 percentile score) given

Valid Rounds. The model may be significant at p ≤ 0.05. The regression output is shown in table
9 and the corresponding plot is shown in Figure 6.
Table 9
Regression output in which winning is defined as top 10 percentile

(Valid Rounds +
Model formula:
intercept)
Number of modeled
55272
observations:
Number of filtered
0
observations:
Model degrees of freedom: 2
Residual degrees of freedom
55270
(DF):
SSE (sum squared error): 773864
MSE (mean squared error): 14.0015
R-Squared: 0.0216551
Standard error: 3.74186
p-value (significance): < 0.0001

Individual trend lines

Panes Line Coefficients

Row Column p-value DF Term Value StdErr t-value p-value
Winn perc 1 Valid Round < 5527 Valid 0.004052 0.000115 34.976 <
0 s 0.0001 0 Rounds 5 9 7 0.0001
0.024396 410.15 <
intercept 10.0064
7 4 0.0001
Figure 7
Plot between number of rounds played and number of rounds in which the player was in
top 10 percentile

Trend Lines Model

A linear trend model is computed for Winn perc 5 (user is part of top 5 percentile score) given
Valid Rounds. The model may be significant at p ≤ 0.05. The regression output is shown in table

10 and the corresponding plot is shown in Figure 7.

Table 10

Regression output in which winning is defined as top 5 percentile

(Valid Rounds +
Model formula:
intercept)
Number of modeled
55272
observations:
Number of filtered
0
observations:
Model degrees of freedom: 2
Residual degrees of freedom
55270
(DF):
SSE (sum squared error): 363808
MSE (mean squared error): 6.58239
R-Squared: 0.017496
Standard error: 2.56562
p-value (significance): < 0.0001

Individual trend lines

Panes Line Coefficients
Row Column p-value DF Term Value StdErr t-value p-value
Winn perc Valid Round < 5527 Valid 0.002492 31.372 <
7.944e-05
5 s 0.0001 0 Rounds 3 4 0.0001
0.016727 <
intercept 4.94369 295.54
6 0.0001

In all 3 regression models, the regression coefficient for the independent variable and number of
rounds (valid_rounds) is positive and statistically significant indicating that there is a learning
effect.

A separate hypothesis test was also performed excluding the first 10 rounds to establish whether
learning effect persists even in latter stages.
Following are the results –

• Winning 5 percentile with Valid rounds > 10

Trend Lines Model

A linear trend model is computed for Winn perc 5 given Valid Rounds. The model may be
significant at p <= 0.05.

Model formula: ( Valid Rounds + intercept )

Number of modeled observations: 266660
Number of filtered observations: 0
Model degrees of freedom: 2
Residual degrees of freedom (DF): 266658
SSE (sum squared error): 5.80621e+06
MSE (mean squared error): 21.774
R-Squared: 0.0045978
Standard error: 4.66626
p-value (significance): < 0.0001

Individual trend lines:

Panes Line Coefficients

Row Column p-value DF Term Value StdErr t-value p-value

< Valid
Winn perc 5 Valid Rounds 266658 0.0037588 0.0001071 35.0955 < 0.0001
0.0001 Rounds

intercept 4.63058 0.0105502 438.908 < 0.0001

• Winning 10 percentile with Valid rounds > 10

Trend Lines Model

A linear trend model is computed for Winn perc 10 given Valid Rounds. The model may be
significant at p <= 0.05.

Model formula: ( Valid Rounds + intercept )

Number of modeled observations: 266660
Number of filtered observations: 0
Model degrees of freedom: 2
Residual degrees of freedom (DF): 266658
SSE (sum squared error): 1.1544e+07
MSE (mean squared error): 43.2915
R-Squared: 0.0069738
Standard error: 6.57963
p-value (significance): < 0.0001
 Individual trend lines:

Panes Line Coefficients

p- p-
Row Column DF Term Value StdErr t-value
value value

< Valid <

Winn perc 10 Valid Rounds 266658 0.0065353 0.000151 43.2745
0.0001 Rounds 0.0001

<
intercept 9.38334 0.0148762 630.76
0.0001

• Winning 20 percentile with Valid rounds > 10

 Trend Lines Model

A linear trend model is computed for Winn perc 20 given Valid Rounds. The model may be
significant at p <= 0.05.

Model formula: ( Valid Rounds + intercept )

Number of modeled observations: 266660
Number of filtered observations: 0
Model degrees of freedom: 2
Residual degrees of freedom (DF): 266658
SSE (sum squared error): 2.24237e+07
MSE (mean squared error): 84.0915
R-Squared: 0.0095228
Standard error: 9.17014
p-value (significance): < 0.0001

Individual trend lines:

Panes Line Coefficients

p- p-
Row Column DF Term Value StdErr t-value
value value

< Valid <

Winn perc 20 Valid Rounds 266658 0.0106573 0.0002105 50.6333
0.0001 Rounds 0.0001

<
intercept 19.0561 0.0207333 919.107
0.0001

The model results are significant (p value close to zero) at all wining conditions. This indicates
evidence of learning effect even at latter stages.

Key Inference out of this test – This test shows that number of rounds played has a significant
affect on the chances of a users to be in the top percentile of scores, hence there is a learning
affect and hence skill is again a dominant factor.
SUMMARY

Based on all the tests which are done using the users, score and player data from Dream11 there
are few key aspects of Fantasy Sports that we have been able to test and prove. Few of them
are:

1. Users playing for money and not a free contest get higher scores proving they are playing more
strategic game else.
2. In a game of chance every player has an equal chance of being selected in the team but the
test done to show that randomly selected players in team lead to low scores than the scores
which are obtained by Fantasy sports users prove that these users are playing more strategic
game than a random one.
3. Selection of players in a fantasy sports team involves skillful assessment of a players’ past
performance.
4. Moreover, people who skillfully assess the past performance of players in the Captain & Vice
Captain selection, tend to have higher average scores than other players.The learning effect
proves that the substantial skill improvement over time which is a characteristic of a skill based
game rather than a chance based game.

Based on various hypotheses tests and detailed analysis of the Dream11 data set, there is
sufficient evidence to conclusively establish that Dream11 format of fantasy sports is skill
dominant. Hence, we conclude that format of Online Fantasy Sports Games as offered by
Dream11 requires predominant skill to play and hence a ‘Game of Skill’.