Content uploaded by Vincent Berthet
Author content
All content in this area was uploaded by Vincent Berthet on Aug 18, 2014
Content may be subject to copyright.
34 VOL. 23, NO. 3, 2010
Second, at one of the four levels of play, all players have folded
their hands except one, who is then the winner. In such cases,
the player who wins the pot does not have to reveal his cards
(i.e., no showdown).
Because of the growth of poker’s popularity and its eco-
nomic importance, the issue of whether poker is governed by
skill or chance has become recurring and turned into a social
debate, as evidenced by Celeste Biever’s recent paper in The
New Scientist. As a card game, poker would be governed by
chance if the outcome of most games were determined by the
value of the cards—the distribution of the cards being random.
In other words, the player who wins the pot would also be the
one who holds the best hand in most cases.
Poker would be governed by skill if the outcome of most
games were determined by something other than the value of
the cards, such as the betting technique, reading of opponents,
or psychology, all of which are grouped under the generic
term “skill.” The poker industry strongly promotes the idea
that poker involves skill more than chance.
The purpose of our study was to provide new empirical
evidence to this debate. This work was done in response to a
recent study by Cigital, Inc., a consulting firm, that reported
two major findings over 103 millions hands: that games did
not end in a showdown 75.7% of the time and only 50.3% of
those in the 24.3% of showdowns were won by the best hand.
Both results suggest the outcome of games is determined by
something other than the value of the cards. Thus, Cigital
concluded that poker is governed by skill.
In our study, we show that this interpretation is not
relevant when data are further analyzed. Moreover, our
research applied inferential statistics, which allowed a clear
test of the theoretical hypothesis that poker is governed by
chance, while the Cigital research provided only descriptive
statistics. Our findings provide clear statistical evidence cor-
roborating this hypothesis.
Method
Data Acquisition
Given the nature of the analysis to be performed, we needed
a database that met two criteria: it was relative to cash-game
poker, rather than four-tournament poker, and it included the
hands of all players in each game, even folded hands. Even
though such data are difficult to obtain, they can be found in
poker television programs. Two met our criteria: “High Stake
Poker” (season five only) and “Poker After Dark” (season four,
“Cash Game #1” and “Cash Game #2”; season five, “Nets vs. Vets
Cash Game,” “Hellmuth Bash Cash Game I,” and “Hellmuth
Bash Cash Game II”; season six, “Top Guns Cash Game #1”
and “Top Guns Cash Game #2”). The coding of these programs
resulted in a 678-game database.
D
uring the last several years, poker has grown in popu-
larity so much that some might even say it’s become
a social phenomenon. Whereas poker was not played
much by nonspecialists a few years ago, it has since democra-
tized and become one of the most practiced card games.
Accordingly, the stereotype of the poker player has moved
from the old guy who smokes a cigar and hides an ace in his
sleeve to the 21 year old who plays online.
While the five-card draw used to be the most famous variant
of poker; the variant widely practiced today is Texas hold ‘em.
In Texas hold ‘em, each player is dealt two cards face down.
This level is called preflop, and play begins with a first betting
round in which each player can check (i.e., stay in the play
without having to bet), bet, raise (i.e., bet over a bet), or fold
(i.e., leave the play).
The dealer then deals a flop, three face-up community
cards. The flop is followed by a second betting round. After
the flop betting round ends, a single community card called
the turn is dealt, followed by a third betting round. A final
single community card called the river is then dealt, followed
by a fourth betting round.
A play can be completed in two ways. First, at least two
players have reached the river having bet the same amount.
In this case, the remaining players reveal their hands and the
winner is the one who holds the best hand (i.e., showdown).
Best Hand Wins: How Poker Is
Governed by Chance
Vincent Berthet
CHANCE 35
Data Coding
For each game, 13 variables were coded for each hand, together
with the two cards of each player and the cards on the board.
“Showdown” indicates whether a given game ended in a show-
down, whereas “Level” indicates the level reached in the game
(i.e., preflop, flop, turn, river).
“Progress” indicates whether a hand was still in progress
when the game ended. To illustrate this variable, let’s consider
a game in which two players were to the flop. One of the play-
ers bet; the other player folded. In this simple case, these two
players were in progress when the game ended, while players
who folded their hands before the flop were not in progress.
There are more subtle cases, however, in which the coding of
the progress variable is less obvious. For example, consider a
game in which three players—A, B, and C—flop. A was the
first player to act and he or she bet. Then B folded, C raised,
and A folded. In this case, despite there being three players
to flop, there were only two involved when the game ended.
Indeed, B was no longer in progress when C raised, which
resulted in A folding and the game ending.
“Status” indicates whether a hand was the best among the
hands in progress when the game ended. “Result” indicates the
objective outcome of a hand (i.e., won or lost).
The next four variables refer to which hand was the best
at each level of the games among the hands of all players. For
instance, for a game that reached the turn, we coded which
hand was the best at the preflop, which hand was best at the
flop, and which hand was best at the turn.
The last four variables refer to the strength of hands
at each level of the games. The strength of a hand was
defined as an ordinal variable with three levels: weak,
marginal, and strong (see supplemental material at www.
amstat.org/publications/chance/supplemental).
Contrary to classical poker databases, which allow comput-
ing of global statistics only (e.g., proportion of games won at
showdown, proportion of games won at preflop), our database
offered real possibilities for further analysis. Indeed, know-
ing the hand of each player and which hand was the best at
each level of the games allowed us to probe the underlying
mechanisms of poker.
Results
Three kinds of games were excluded from the analysis: games
that ended in a slip pot (1.47%); those that did not end in a
showdown, but in which players held the same hand (0.15%);
and those in which the board was run more than one time
(1.03%). This resulted in a 660-game sample.
We basically estimated the same three proportions as Cigital,
Inc. Our estimations (and 95% confidence intervals) were:
73.5% (70.2,76.8) of games without showdown
15.3% (12.6,18.0) of games that ended in a showdown in
which the best absolute hand did not win
11.2% (8.8,13.6) of games that ended in a showdown in
which the best absolute hand won
Based on their 103-million hand sample, Cigital, Inc.
reported 75.7%, 12.1%, and 12.2% for the first, second, and
third proportions, respectively.
Although both sets of estimations are similar, the two samples
differ slightly [x
2
= 6.6, df = 2, p < .05], especially regarding the
proportion of games that ended in a showdown in which the
best absolute hand did not win. This difference could be because
Cigital’s sample included hands played online by a majority
of amateur poker players, whereas our sample included hands
played live by professional poker players. Since professional
players tend to bluff more than amateur players, it seems normal
that best absolute hands won less often in the latter sample.
More fundamentally, we do not claim that our sample is rep-
resentative of Texas hold ‘em as a whole. Indeed, one should first
define all the relevant features of the population to construct a
representative sample of Texas hold ’em, and it seems Cigital did
not. As a result, one cannot claim that Cigital’s sample is more
representative of Texas hold ‘em than ours, because online
poker played by amateurs is not more representative of Texas
hold ‘em than live poker played by professionals.
Games with Showdown
One of the arguments advanced by Cigital to support the
hypothesis that poker involves skill more than chance is that
the best hand wins only 50.3% of the time at showdown.
At first glance, such a proportion seems surprising, as one
would logically expect the best hand to win every time at
showdown. Actually, this finding relies on a particular mean-
ing of the “best hand.”
In each game, while the best “absolute” hand is the one held
by the player who would have made the best five-card hand
at showdown, the best “relative” hand is the best hand among
players who were in progress when the game ended. In the
latter, the best hand is determined by taking into account all
players, regardless of whether they folded. In the former, the
best hand is determined with respect to players who were in
progress when the game ended. Based on these considerations,
it seems Cigital’s finding relies on the notion of best absolute
hand (i.e., best relative hand wins every time at showdown
by definition).
To illustrate the difference between best absolute hand
and best relative hand, let’s consider the following example.
Three players—A, B, and C—start a play. A holds the five of
diamonds and five of spades (a pair of fives), B holds the ace
of hearts and the king of spades (two high cards), and C holds
the ace of clubs and the 10 of diamonds.
The flop is the ace of diamonds, the king of hearts, and
the queen of clubs. The value of the hands is now a pair of
“
Phil Hellmuth said:
‘Poker is 100% skill and
50% luck.’ That quote is
100% dubious!
”
- Anonymous (Two Plus Two poker forum)
36 VOL. 23, NO. 3, 2010
fives for A, two pairs (aces and kings) for B, and a pair of aces
for C. In terms of hand strength, B is better than C, which is
better than A.
In the first betting round, A checks, B bets, C calls, and A
folds. The card at the turn is the five of clubs. If A had remained
in the game, A would have had the strongest hand (three fives).
B bets and C calls. The card at the river is the two of hearts,
which does not help any of the players. B bets and C calls, so
the game ends in a showdown. B holds the best relative hand
(top two pairs), whereas A holds the best absolute hand (three
of a kind). In this play, the best relative hand (B) wins because
A folded. Player C loses the most money.
Given that the best absolute hand does not win at show-
down in half the cases, Cigital suggested that this hand was
beaten by skill. This rationale is valid only under the assump-
tion that the best absolute hand is the best hand at all levels of
a game. Instead, if the best absolute hand is not the best hand
from stem to stern in most cases, then this hand is dominated
at a certain level in the play. Therefore, it could be that the best
absolute hand does not win at showdown half the time because
it is dominated before showdown half of the time.
To test this alternative interpretation, we analyzed the
statistical properties of best absolute hands that did not win
at showdown. Table 1 presents the contingency table of best
absolute hand strength by level categories at time of fold.
Two main observations can be drawn from this table.
First, best absolute hands that did not win at showdown
were folded preflop 79% of the time. Second, even though
these hands would have won at showdown, they were actu-
ally weak hands when they were folded (80% of the time).
Moreover, regarding their status, these hands were inferior
when they folded 82% of the time. Taken together, these
three observations lead to a clear conclusion: Best absolute
hands that did not win at showdown were actually largely
dominated when they were folded.
Accordingly, the best absolute hand does not win at show-
down half the time because it was beaten before showdown
half the time—not because it faced a skilled player who
bluffed. In other words, it is the value of the cards in this
set of games, rather than skill, that determines the outcome.
Moreover, these findings highlight the importance of how the
best hand is defined. We claim that when determining the best
hand, only considering the best relative hand makes sense.
Games Without Showdown
The second argument advanced by Cigital is that games that
do not end in a showdown are governed by skill, and that such
a scenario occurs in the large majority of cases (75.7% of the
time). At first glance, this argument makes sense. Indeed, a card
game is considered to be governed by chance if the outcome of
most games is directly related to the value of the cards. There-
fore, one could conclude that games without showdown cannot
be governed by chance, since no private card is revealed.
It is worth noting that games without showdown are the
core of poker. In fact, the possibility of winning a game without
having to reveal one’s cards is poker’s trademark. This feature
seems to eliminate chance for the benefit of skill and opens
the door to bluff.
Metaphorically, games without showdown are the king-
dom of skill and the bluff is king. To illustrate this idea, let’s
consider the summary of a promotional video distributed by
“Poker After Dark.” The narrator is Doyle Brunson, a poker
living legend. Here is what he says: “It all begins with a raise
before the flop, then a re-raise. You fold, then a call. Jack,
Table 1—Contingency Table of Best Absolute Hand
Strength by Level at Time of Fold
Level
Hand
Strength
Preflop Flop Turn River Row total
Weak
Observed 78 3 0 0 81
Column % 97.5 33.3 0.0 0.0 (80.2%)
Marginal
Observed 2 5 1 1 9
Column % 2.5 55.6 100.0 9.1 (8.9%)
Strong
Observed 0 1 0 10 11
Column % 0.0 11.1 0.0 90.9 (10.9%)
Column total
80 9 1 11 101
(79.2%) (8.9%) (1.0%) (10.9%)
CHANCE 37
four, 10, with two hearts at the flop. Check, bet, and call. The
Ace of heart[s] is the turn. Bet, and a big raise. Now the real
game begins.”
Then, the other player folds his hand—a pair of aces—
which was probably the best hand. The remaining player
wins the pot without showing his cards. “That’s poker, folks,”
says Brunson.
This promotional video shows what advocates of poker
think to be the core of the game: When cards are not revealed,
one can put the pressure on the player who holds the best hand
and force him to fold. This reasoning is valid at an explicit
level, but one also could consider an implicit level. Indeed, that
cards are not revealed is not sufficient to discard their influence
on the outcome. One has to consider the possibility that most
games without showdown are nevertheless won by the player
who actually holds the best hand.
We therefore analyzed games without showdown (485
games) in more detail. We tested the hypothesis that poker is
governed by chance by examining the distribution of winning
hands in those games. According to what we suggested previ-
ously regarding best hand determination, we considered best
relative hands (i.e., the best hand among players who were
in progress when the game ended), rather than best absolute
hand (i.e., the best potential five-card hand that would have
won when the game ended).
If poker is governed by skill, winning hands should be
uniformly distributed into the inferior hand and best (rela-
tive) hand categories (i.e., null hypothesis). Indeed, claiming
that poker is governed by skill means winning or losing with
a particular hand does not depend on its objective value. In
other words, one could win as much with inferior hands as
with best hands.
However, if poker is governed by chance, winning hands
would be best relative hands most of the time (i.e., alternative
hypothesis). This means the outcome of a particular hand
would be directly related to its objective value: one would win
with best hands and lose with inferior hands.
A one-way chi-square test revealed that winning hands
were best hands in most cases [x
2
= 99.8, df = 1, p < .001, after
Yates’s correction]. Indeed, of the 485 winning hands, 353
were best hands, corresponding to 72.8% (69.5,76.1). Thus,
even when hands were not revealed, best hand won almost
75% of the time.
Moreover, this tendency was observed at each level of the
games. Though the dominance of best hands was slightly
modulated by level, [x
2
= 13.9, df = 3, p < .005], the best
hand tended to win whichever level was reached. Indeed,
best hand won 62.7% (59.1,66.3) of the time at the preflop,
74.2% (70.9,77.5) of the time at the flop, 68.9% (65.4,72.4) of
the time at the turn, and 84.8% (82.1,87.5) of the time at the
river. Table 2 presents the contingency table of winning hand
status by level categories. Considering the process of getting
a winner as a process of survival with each level being a new
obstacle, these findings show that the best existing hand tends
to survive to the winning stage most of the time.
One could argue that winning hands being best hands
almost 75% of the time does not demonstrate that poker is
governed by chance. In fact, when hands are not revealed,
holding the best hand is not sufficient to win. Holding the
best hand is related to chance, but knowing that one holds the
best hand is related to skill. However, knowing that one holds
the best hand might not involve much skill, since it could be
that best hands are strong hands in most cases. Indeed, the
probability of holding the best hand is directly related to the
strength of one’s hand, and there is no need to be a skilled
Table 2—Contingency Table of Winning Hand Status by Level
Level
Winning
Hand Status
Preflop Flop Turn River Row total
Inferior
Observed 38 41 37 16 132
Column % 37.3 25.8 31.1 15.2 (27.2%)
Best
Observed 64 118 82 89 353
Column % 62.7 74.2 68.9 84.8 (72.8%)
Column total
102 159 119 105 485
(21.0%) (32.8%) (24.5%) (21.6%)
38 VOL. 23, NO. 3, 2010
player to know that. Thus, we further tested the hypothesis
that poker is governed by chance by analyzing the distribution
of winning hands as a function of hand strength.
While the skill hypothesis states that winning hands should
be uniformly distributed into the weak, marginal, and strong
categories (i.e., one could win whatever the hand strength), the
chance hypothesis assumes winning hands should be strong
hands most of the time. A one-way chi-square test confirmed
that winning hands were strong in most cases [x
2
= 72.9,
df = 2, p < .001]. In fact, of the 485 winning hands, 27.0%
(23.7,30.3) were weak, 21.6% (18.5,24.7) were marginal, and
51.3% (47.6,55.0) were strong. This supports our claim that
knowing one holds the best hand does not require much skill,
as winning hands are strong most of the time.
Conclusion
Our study might suffer from two limits. The first concerns our
sample, which included games played by top professionals only.
Since such players have relatively the same amount of skill, it
could be that luck determines the outcome of games in such
cases. It is more likely that skill would play a significant role
when skilled players play against beginners. Further research is
needed to investigate how properties of winning hands depend
on the level of players.
The second limit is that our analysis did not take into
account betting history. Betting influences the outcome of
games along with the value of the cards. While we showed
that the latter plays a massive role in the process of getting
a winner in each game, it seems likely that betting also plays
a significant role. For instance, a weak hand is more likely to
survive when accompanied by proper betting. In the same
way, a strong hand is more likely to make money if the bet-
ting pattern is adequate. Betting could be used to measure the
amount of skill with which a hand was played, so that one
could determine which of the chance (i.e., the value of the
cards) or skill (i.e., betting) factors best predicts the outcome
of hands. This would address further the chance vs. skill issue
and extend our findings. We plan to construct a database that
includes betting history for the games.
When addressing the chance vs. skill issue, one should
always keep two guidelines in mind. First, such an issue is
unlikely to be resolved by a study alone. As a result, legal
decisions regarding the status of poker should be made by
examining all available scientific evidence.
Second, games of pure chance and games of pure skill
are located at the extremities of a continuum, and poker is
located at a certain point on this continuum. No one can
deny that poker involves a certain degree of chance. The
mere existence of “bad beats” (i.e., the favorite hand finally
loses) is a direct consequence of this random component of
the game. However, some studies reported that skill also plays
a significant role.
For example, Rachel Croson, Peter Fishman, and Devin
Pope showed that poker is similar to golf, which is a typical
game of skill, in a recent CHANCE article. Moreover, Michael
DeDonno and Douglas Detterman reported in an article pub-
lished in the Gaming Law Review the findings of learning effects
in poker. Such effects would not be observed if poker were a
game of pure chance. As a result, poker is neither a game of
pure chance nor a game of pure skill. Thence, the problem is to
determine whether poker is dominated by chance or by skill.
By showing that it is the value of the cards that mostly
determines the outcome, our findings demonstrate that poker
is truly governed by chance. In most cases, the player who
wins the pot is the one for whom the cards were the most
favorable. Best hand wins with and without showdown. That’s
poker, folks.
Further Reading
Biever, C. 2009. Poker skills could sway gaming laws. The New
Scientist 202(2702):10.
Cigital, Inc. 2009. Statistical analysis of Texas hold ’em. www.
cigital.com/resources/gaming/poker/100M-Hand-AnalysisReport.pdf
Croson, R., P. Fishman, and D. G. Pope. 2008. Poker super-
stars: Skill or luck? Similarities between golf—thought to
be a game of skill—and poker. CHANCE 21(4):25–28.
DeDonno, M. A., and D. K. Detterman. 2008. Poker is a skill.
Gaming Law Review 12(1):31–36.
Sklansky, D., and M. Malmuth. 1999. Hold ’em poker for advanced
players. Henderson, NV: Two Plus Two Publishing.
