1 Battle of the Sexes: The Role of Gender in Survivor Sophie Truscott * Advised by Sara Hernández MMSS Senior Thesis, Northwestern University June 2017 Abstract This paper utilizes data from the first 33 seasons of the American reality television show Survivor to analyze the role of gender within competitive environments. Specifically, I use contestant-level voting and performance data to model how gender impacts contestants’ voting decisions, both in eliminating other contestants throughout the game and voting for a winner in the finals. I also observe potential differences in elimination order between male and female contestants with similar performance attributes. I find that when the majority of contestants voting against a contestant are female, the contestant is more likely to be male, and vice versa. Furthermore, increasing the number of women voting against a contestant increases the likelihood of the eliminated contestant being male. Additionally, I find that the voting decisions of jurors in the finals do not seem to be affected by the juror’s gender. While my paper finds some significant results regarding the role of gender in Survivor, I also conclude that many of the most important variables impacting a contestant’s success in Survivor are nuanced qualitative interpersonal factors not captured in my dataset that may be quite difficult to measure quantitatively. * I would like to thank my advisor, Professor Sara Hernández, for her guidance, support, and encouragement throughout this process; our teaching assistant, Aniket Panjwani, for helping to provide structure and direction to my analysis; Professor Joseph Ferrie for guiding our class through the thesis process; and Nicole Schneider for her administrative assistance. Thank you also to Jeff Pitman of True Dork Times for generously sharing his dataset and personal insights with me. Finally, thank you to my family and friends for their endless love and support.
37
Embed
Battle of the Sexes: The Role of Gender in Survivormmss.wcas.northwestern.edu/thesis/articles/get/958...1 Battle of the Sexes: The Role of Gender in Survivor Sophie Truscott* Advised
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Battle of the Sexes:
The Role of Gender in Survivor
Sophie Truscott*
Advised by Sara Hernández
MMSS Senior Thesis, Northwestern University
June 2017
Abstract
This paper utilizes data from the first 33 seasons of the American reality television show
Survivor to analyze the role of gender within competitive environments. Specifically, I use
contestant-level voting and performance data to model how gender impacts contestants’
voting decisions, both in eliminating other contestants throughout the game and voting for a
winner in the finals. I also observe potential differences in elimination order between male
and female contestants with similar performance attributes. I find that when the majority of
contestants voting against a contestant are female, the contestant is more likely to be male,
and vice versa. Furthermore, increasing the number of women voting against a contestant
increases the likelihood of the eliminated contestant being male. Additionally, I find that the
voting decisions of jurors in the finals do not seem to be affected by the juror’s gender. While
my paper finds some significant results regarding the role of gender in Survivor, I also
conclude that many of the most important variables impacting a contestant’s success in
Survivor are nuanced qualitative interpersonal factors not captured in my dataset that may
be quite difficult to measure quantitatively.
* I would like to thank my advisor, Professor Sara Hernández, for her guidance, support, and encouragement
throughout this process; our teaching assistant, Aniket Panjwani, for helping to provide structure and direction to
my analysis; Professor Joseph Ferrie for guiding our class through the thesis process; and Nicole Schneider for her
administrative assistance. Thank you also to Jeff Pitman of True Dork Times for generously sharing his dataset and
personal insights with me. Finally, thank you to my family and friends for their endless love and support.
2
1 Introduction
While a number of researchers have conducted studies to examine the gender
differences in individuals’ competitive behavior, most studies utilize simple,
controlled tests to model competitive situations which often differ greatly from the
competitive situations that individuals encounter in real life. This paper aims to
minimize the impact of this environment discrepancy by studying the role of gender
within the reality television show Survivor, a game that more accurately models the
nuances of real world competition. Survivor is an American reality television series
in which contestants are stranded in a remote location and compete against each
other until one contestant remains. The game requires contestants to utilize their
social, strategic, and physical skills, and the $1 million prize provides a strong
incentive for players to play the game to the best of their abilities. As such, Survivor
acts a useful natural experiment to analyze how patterns of competitive behavior
differ between men and women.
There are, of course, drawbacks associated with using a televised reality
show to analyze human behavior. First, observed insights are likely to be impacted
by producers’ editing decisions intended to make the show as entertaining as
possible, and since each season’s 39 days are compressed into 13 hour-long episodes,
the majority of the footage from the game is not available to viewers. To overcome
the potential effects of these editing biases, this paper relies only on the most
objective observable variables on the show that are unaffected by editing decisions.
Second, game show contestants tend to share certain qualities; people who self-select
to participate in such competitions are usually more competitive, self-confident, risk-
seeking, and attention-seeking than the average person. This paper does not
formally account for this discrepancy, but it is an important point to keep in mind
when interpreting results. Finally, one might consider how a reality show’s
viewership might impact contestant’s behavior within the game—if contestants
know that their behavior in the game will be widely viewed, they may choose to
change their true behavior for fear of being viewed negatively by the show’s
3
audience.1 There is, however, little evidence to believe that Survivor contestants
play the game with this mindset, and if nothing else, the $1 million prize acts as a
powerful incentive for contestants to act on their true intentions rather than
“playing a role” to create a particular image of themselves for viewers.
My analysis focuses on the role of gender within three competitive aspects of
Survivor: (1) the elimination of contestants at Tribal Councils, (2) the voting
decisions made by jurors at the Final Tribal Council, and (3) the elimination of
contestants with similar performance qualities. First, I examine how the gender of
the contestants voting against a contestant is related to the targeted contestant’s
gender; in other words, I consider if there is a gender-based influence on which
contestants eliminate male and female contestants. Second, I investigate how jurors
of different genders are affected by their relationships with each of the finalists;
specifically, I test whether there is a statistically significant difference between male
and female jurors’ propensity to forgive finalists for a personal betrayal. Finally, I
provide diagrams to observe potential differences in the elimination order of
similarly skilled male and female contestants.
Understanding how men and women behave in the face of competition can
help to identify how particular gender-based characteristics and behavior may help
or hinder individuals’ competitive performance. There are endless possible
applications for these insights, but perhaps the most relevant use of gender-based
competitive behavior patterns in today’s society is to help adjust the competitive
nature of the modern workplace to create a more even playing field for professional
men and women.
2 Literature Review
This section provides an overview on existing literature on the topics of reality
television and the relationship between gender and competition. I will also describe
where my research fits in among this existing work.
1 “Negative” in this case would likely differ from contestant to contestant. For example, some
contestants may worry about being perceived by viewers as domineering, while other
contestants may not want the audience to view them as passive.
4
2.1 Existing Research on Gender and Competition
Since the intention of this paper is to investigate the relationship between gender
and competitiveness, using Survivor as a natural experiment, it is relevant to
examine existing literature and theories concerning gender and competition.
Lee, Kesebir, and Pillutla (2016) find that women view same-gender
workplace competition as less desirable than their male counterparts do, and in the
presence in same-gender competition, women’s relationships with one another often
suffer. Lee et al. cite gender socialization research that women are taught to value
equality and harmony while hierarchical competition is a critical aspect of
masculinity; therefore, in competitive workplace situations that create hierarchy,
men welcome competition while women shy away from competition, particularly in
situations that pit them against other women. On the other hand, Burow, Beblo,
Beninger, and Schröder (2017) find in an online experiment that women prefer to
enter competitions when their competitors are known to be women. Burow et al. find
that women who correctly or underestimated their task abilities were more likely to
display this behavior, while women who overestimated their abilities enter
competitions regardless of their opponents’ genders (women’s actual abilities,
however, did not affect their willingness to compete in different gender
environments). While these two studies from Lee et al. and Burow et al. may seem to
be at odds with each other, the studies together present a more thorough, complex
picture of gender and competition. When faced with same-sex competition that
yields hierarchical rankings, Lee et al. find that women shy away for fear of
disrupting equality with their female peers, but if women have to compete, Burow et
al. suggest that they prefer to do so among female competitors with whom they feel
less intimidated to reveal their true abilities.
Dato and Nieken (2014) conducted an experiment to investigate competitive
gender differences in games where players had the opportunity to sabotage their
opponents. Dato and Nieken find that men are more likely to sabotage their
opponents, which led to a higher probability of winning. However, since players in
the study incurred a cost when they chose to sabotage, men and women had the
same earnings on average. Dato and Nieken believe that men are generally more
status-seeking than women, so men are more willing to “invest” in victory by
5
incurring a sabotage cost, even if doing so decreases their eventual earnings. In an
experiment with Swedish adolescents, Dreber, von Essen, and Ranehill (2014) find
that girls in competitive settings are more altruistic and risk-averse than boys.
Finally, Halladay (2017) conducted a two-stage experiment in which subjects
first play a game in teams and rate how positive, negative, and neutral their feelings
are toward each of their teammates. Then subjects play a second game against one
of their teammates from the previous round. Halladay finds that women in
competitive situations respond more strongly to negative emotions than men do;
women’s performance in the second stage was significantly higher when they
competed against an opponent toward whom they had negative feelings, whereas
male performance was unaffected by personal emotions toward an opponent.
It is important to note that Survivor is a self-selecting, coed game that
naturally attracts highly confident and competitive individuals, since the game
features a high-stakes $1 million reward and is broadcasted for an average of 16.2
million viewers per episode.2 As such, the studies from Lee et al. and Burow et al.
regarding women’s likelihood to enter into same- and mixed-gender competitions is
not wholly relevant to Survivor.3 More relevant to my analysis is Dato and Nieken’s
research into men and women’s propensity to sabotage in competitive situations and
Dreber et al.’s analysis of gendered altruism and risk-taking in competitions. While I
will not examine how contestants’ emotional opinions of one another affect their
personal performance in the game, Halladay’s research into the effect of emotions in
competition is certainly relevant to my examination of the effect of betrayal and
personal relationships on voting behavior at the Final Tribal Council.
2 Overall Survivor viewership average for all 34 seasons calculated from average episode
ratings per annual television season, as reported by CBS Entertainment. Average Survivor
viewership per episode for the 2016-2017 TV season was 10.32 million viewers per episode.
Source: The Nielsen Company. 3 Given the coed nature of Survivor, the research from Burow et al. suggests that women who
choose to compete in Survivor tend to overestimate their abilities. Since, however,
individuals’ self-confidence is not a factor in my analysis, this observation is not relevant to
my paper.
6
2.2 Existing Research on Survivor and Game Shows
While the internet is home to hundreds of active fan communities, websites, and
blogs dedicated to quantitative and qualitative discussions of Survivor, there has
been relatively little academic research dedicated to the show. For my purposes, I
have identified three academic publications concerned with Survivor and one paper
on another reality game show, The Weakest Link, which employs a similar mode of
analysis to my own.
Mixon (2001) utilizes game theoretical notions of cartel behavior to analyze
contestants’ incentives and behavior from the first season of Survivor. Mixon
observed qualitative interpersonal factors across the 13-episode season as well as
the Tribal Council voting history to break the contestants into various voting
“cartels” and analayze their incentives and decisions. Mixon’s results are somewhat
difficult to extrapolate to Survivor gameplay more broadly as they focus on the
specific, unique personality traits and relationships of the first season’s contestants.
Nevertheless, Mixon’s paper demonstrates how Survivor can be used as a platform
for observing economic and game theoretical principles in a competitive setting.
Hedges’s (2014) paper examines how viewers perceive the gender identities of
contestants from four seasons of Survivor using Q-Methodology, a psychological
research method to capture individuals’ subjectivity. Her experiment finds that
viewers use more criteria to evaluate the masculinity of male contestants than the
femininity of female contestants. While Hedges’s focus on viewers’ perception of
gender differs from my focus on contestants’ perception of gender within the game,
her research reveals how both the game of Survivor and the show’s portrayal of
contestants can be useful in gender-based analyses.
Dilks, Thye, and Taylor (2010) analyze Survivor Tribal Council voting history
from the show’s first 17 seasons to test economic models of taste-based
discrimination and social identity theory. Dilks et al. assign contestants to either
“low status” (women, minorities, and elderly) or “high status” groups, and find that
low status contestants are more likely to be eliminated in early stages of the game—
when contestants have an incentive to eliminate less competent players—and high
status contestants receive more votes in later stages when competency is considered
a threat. My analysis will build upon this notion of voting discrimination in
7
Survivor—specifically in terms of gender discrimination—by analyzing not only
which players types may be discriminated against but also by whom.
Although Antonovics, Arcirdiacono, and Walsh’s (2005) paper focuses on The
Weakest Link, a British reality television show, the principles they examine and the
identification strategies they employ bear resemblance to those utilized in this
paper. Antonovics et al. use contestants’ voting history to observe discrimination
between contestants of different genders and race. Their results do not find any
evidence of males discriminating against females or of whites discriminating against
blacks, but they do observe that women discriminate against men in early stages of
the game. Antonovics et al. substantiate the notion that voting patterns in reality
game shows can be extrapolated to identify patterns of social behavior.4
In a similar approach to Dilks et al. and Antonovics et al., my research aims
to use Survivor as a platform to quantitatively investigate gender-specific patterns
of social behavior in competitive environments. Both Dilks et al. and Antonovics use
contestants’ voting history to illuminate patterns of discriminatory decision-making.
I expand this approach by considering additional factors that might affect Survivor
contestants’ opinions of and relationships with one another, including the presence
of alliances and betrayals, and their perception of other’s abilities based on
performance within the game.
3 Survivor Overview and Strategies
Survivor first aired on CBS television network in May 2001. On May 24, 2017, the
most recent Survivor season (Season 34 – Game Changers) finished airing on CBS.
In May 2017 the show was renewed by CBS for the 2017-2018 television season.5
The show’s leadership has remained relatively unchanged since its inception;
4 It should be noted that contestants’ incentives in Survivor and The Weakest Link vary
rather significantly, since contestants in The Weakest Link compete as a team for the
entirety of the game, while Survivor contestants only compete as a team for half of the game
(which affects the team-centric mentality even in early stages of the game). Nonetheless,
Anthonovics et al. provide a foundation for evaluating voting differences amongst groups
that will be very useful in my analysis. 5 CBS Broadcasting Inc., CBS Entertainment. (2017, March 23). CBS Renews 18 Series for
2017-2018 Season [Press release]. Retrieved from https://cbspressexpress.com/cbs-
entertainment/releases/view?id=47362.
8
notably, current creator/executive producer Mark Burnett and host/executive
producer Jeff Probst have worked on Survivor since the first season.
While numerous Survivor seasons have included a number of gameplay
twists, the basic format of the game has remained the same throughout the show’s
history.6 Each season, 16 to 20 contestants (known as “castaways”) are brought to a
remote location and divided into 2 to 4 tribes. The tribes live at separate camps with
limited resources but meet regularly to compete in reward or immunity challenges.
In reward challenges, the winning team receives amenities (often fire, food, fishing
supplies, blankets, or more extravagant gifts). In immunity challenges, tribes
compete to win immunity, thereby avoiding attending Tribal Council. The losing
tribe goes to Tribal Council, where one member of the tribe is voted out. After
approximately half of the contestants have been eliminated, the separate tribes
merge into one tribe, and the remaining contestants compete as individuals in
reward and individual immunity challenges. Contestants who win individual
immunity attend and vote at Tribal Council but are exempt from elimination. The
gameplay continues until 2 or 3 contestants remain (the “Final 2” or “Final 3”), at
which point the jury (composed of the last 7 to 10 contestants eliminated before the
finalists) vote for one of the finalists to win.7 The finalist who receives the most jury
votes at the Final Tribal Council is named the winner (“Sole Survivor”) and is
awarded the $1 million cash prize.
Fans of Survivor will likely note that the interpersonal complexities of the
game make it difficult to prescribe a singular rational gameplay strategy. With that
said, I identify a basic rational strategy for gameplay under the single assumption
that a contestant’s goal is to win the game, which necessitates that a contestant is
not eliminated before the finals. Rational strategy for contestants varies rather
significantly between the tribe and individual stages of the game. I will refer to the
6 The following is a non-exhaustive list of notable twists to the standard Survivor game
format: initial tribes divided by ethnicity, gender, age, or beauty/intelligence/physical
strength; players compete against family members; contestants from previous seasons return
to play in a new season. Table IV lists a number of these twists by season. 7 The use of the Final 2 or Final 3 depends on the season and is decided by the Survivor
producers. The Final 2 was the norm until the Final 3 was introduced in Survivor: China
(Season 13). Host Jeff Probst explained that a Final 3 prevents a universally respected
finalist facing off against an unlikeable finalist, which yields an anti-climactic Final Tribal
Council.
9
initial stage of the game, in which contestants compete in tribes, as the “Pre-Merge
Stage” and the secondary stage of the game, in which contestants compete as
individuals, as the “Post-Merge Stage.” In the Pre-Merge Stage, contestants want
their tribe to be as strong as possible so they can win challenges. (Recall that
winning immunity challenges excuses the tribe from attending Tribal Council,
thereby guaranteeing that every contestant on the winning tribe is spared from
elimination that round.) As such, when a tribe attends Tribal Council, contestants
should vote to eliminate the weakest member of their tribe to increase their tribe’s
chance of winning future challenges. A contestant’s “strength” is usually evaluated
primarily by one’s performance in challenges (usually physical strength and agility),
although other skills, such as providing food and shelter around camp, may
contribute to the tribe’s overall strength and thus increase a contestant’s perceived
strength. In the Post-Merge Stage, in which contestants compete as individuals,
contestants should vote off the strongest challenge competitors in order to increase
their own chances of winning individual immunity challenges and thus
guaranteeing their progression in the game.
The jury component somewhat complicates this basic strategy. Since the jury
selects the overall winner, contestants may also consider who will accompany them
to the finals. There are no rules on the evaluation criteria used by jurors to vote for a
winner, but jurors generally vote for a winner based on their perception of the
quality of the finalists’ gameplay, as well as more personal and interpersonal
qualities such as their perception of the finalists’ integrity and the jurors’ personal
relationships with each of the finalists. Therefore in the Post-Merge Stage of the
game, particularly as contestants near the finals, contestants may consider
eliminating not only contestants who perform well in challenges but also contestants
who they believe would be likely to receive numerous jury votes at the Final Tribal
Council.
While these strategies and incentives may oversimplify a number of
contestants’ considerations throughout the game (including, but not limited to,
personal relationships/alliances, mistrust of contestants, and dislike of or annoyance
with contestants), this basic strategy will be useful in motivating some decisions in
my analysis later in this paper.
10
4 Data
I compiled a relational database on every Survivor season excluding Season 34 using
data from two websites: Survivor Wiki and the True Dork Times.8, 9 The Survivor
Wiki data was retrieved directly from the website and includes demographic
information on all contestants and contestant-level voting and elimination history
for every Tribal Council and Final Tribal Council. The editor-in-chief of the True
Dork Times, Jeff Pittman, shared his entire dataset with me via email (some, but
not all, of the data in this dataset appears on the True Dork Times website). The
data from the True Dork Times includes contestant-level performance data from
every challenge and Tribal Council.
All data in my database is observed exclusively from Survivor episodes (in
other words, potential data from post-Survivor interviews, contestants’ social media,
etc. is not included). In an effort to avoid potential bias from editing decisions made
by Survivor producers, contestant data such as testimonials, airtime, and character
portrayal is not included in the database.
The database contains four main data tables: data by contestant, data by
Tribal Council, data by juror, and data by finalist. The data by contestant includes a
row for each contestant with columns for variables pertaining to each contestant. It
should be noted that this table treats returning players as unique players (i.e.
players who return to the game have a row for each gameplay), but the table
includes variables noting whether a contestant is a returning player. The data by
Tribal Council includes rows for each Tribal Council across every season; this data
can be thought of as panel data where the ordered Tribal Council within a season
(which can be used to track “stages” or periods of a season) is the individual
dimension and the season is the time dimension. The data by juror is a pooled cross
section that includes a row for each juror/finalist pair and includes variables
capturing whether the juror voted for the finalist at the Final Tribal Council and to
what degree the juror and finalist interacted throughout the game. The data by
8 Survivor Game Changers (Season 34) aired its finale on May 24, 2017, one week before the
submission of this thesis. Given this time constraint, this season was not included in any of
my analysis. 9 Survivor Wiki can be accessed at survivor.wikia.com; the True Dork Times can be accessed
at truedorktimes.com.
11
finalist lists each finalist across seasons and includes variables concerning number
of jury votes receives and comparing the finalists’ jury votes to their overall
performance in the game.
Table I includes summary statistics for all Survivor contestants in the
sample, Table II includes summary statistics for Survivor winners, and Table IV
provides summary statistics for the first 33 Survivor seasons. Table III provides
summary statistics for contestants’ performance disaggregated by gender. The first
variable in Table III, “Overall Finish,” captures a contestant’s rank amongst the
other contestants within a given season, where the winner’s rank is coded as 1 and
each subsequent finisher’s rank increases by 1. The “Mean Percent Finish in
Individual Challenges” captures how contestants placed on average in individual
challenges. For example, in an individual challenge with five competitors, the
challenge winner’s percent finish would be coded as 1, the second-place finisher’s
percent finish would be coded as 0.8, the third-place finisher’s percent finish would
be coded as 0.6, and so on. These percent finishes are averaged for each contestant
across every individual challenge in which they participate to generate a
contestant’s mean percent finish. Note that individual challenges, with very few
exceptions, take place only after the tribe merge, so contestants who never
participated in an individual challenge are excluded from this statistic. “Votes
against Boot Percentage” counts at how many Tribal Councils a contestant cast a
vote against the contestant who was eliminated at that Tribal Council (known as the
“bootee”) and divides it by the total number of Tribal Councils the contestant
attended. This variable captures, at least to a certain degree, a contestant’s strategic
strength, as it measures how often the contestant effectively eliminated another
contestant; the higher a contestant’s score, the more often he or she voted for the
bootee. Since a rational contestant’s goal is to win the game, he or she must survive
every Tribal Council without being eliminated. A contestant’s best Tribal Council
strategy is therefore to ensure that the person who they vote for is the person
ultimately eliminated at that Tribal Council.10 Under this strategy, a higher “Vote
10 Contestants usually attempt to ensure this by speaking with their peers and alliance
before Tribal Council and either convincing others to vote for a particular contestant or
learning for whom the majority of tribe members are voting and voting the same way.
12
for Boot Percentage” indicates that the contestant has more regularly achieved this
voting strategy.
5 Gender-Based Voting and Elimination Analysis
5.1 Identification Strategy
My first research question concerns the relationship between the gender of an
eliminated contestant and the gender(s) of the contestants who eliminate him or her.
I am interested to test how the expected gender of an eliminated contestant might be
affected by the number of female and male contestants voting against the
contestant.
Before laying out the model, it is useful to define a few terms. First, “bootee”
refers to the eliminated contestant at a particular Tribal Council. Second,
“elimination Tribal Council” refers to the Tribal Council at which a particular bootee
was eliminated. Third, an “eliminator” is a contestant who votes for the bootee at the
bootee’s elimination Tribal Council, thereby contributing to the bootee’s elimination.
Finally, an “eliminator group” is the group of all eliminators who vote for the bootee
at the bootee’s elimination Tribal Council.
The data for this model lists each contestant from all 33 seasons and
quantifies how many male eliminators (“Total No. Male Eliminators”) and female
eliminators (“Total No. Female Eliminators”) voted for the contestant at the
contestant’s elimination Tribal Council, as well as the percent of females in the
eliminator group (“Eliminator Group Gender”). The sample for this model excludes
all contestants who progressed to the finals (including winners), since these
contestants were not eliminated in a Tribal Council. Additionally, contestants who
were eliminated by means other than a traditional Tribal Council are excluded from
the sample.11 Finally, contestants who were eliminated while members of single-
gender tribes (as assigned by Survivor producers as a gameplay “twist”) were
excluded from this sample.12
11 Here, “traditional Tribal Council” refers to a Tribal Council that did not result in a tie (tie
votes require a re-vote) and did not involve the successful use of a hidden immunity idol.
Furthermore, contestants who quit or were medically evacuated from the game are not
eliminated by traditional Tribal Council, and are therefore excluded from this sample. 12 See Table IV for seasons with initial tribes divided by gender,
13
Since my dependent variable (“Bootee is Female”) is a binary dummy
variable, I selected a conditional fixed effects logistic regression (“logit model”) to
test this question. My initial regression, Equation (1) captures the effect of the
number of female and male eliminators and the gender-mix composition of the
eliminator group on a bootee’s gender, including fixed effects for each Survivor
season:13
(Bootee is Female)i = β0 + β1(Total No. Female Eliminators)i + β2(Total No. Male
Eliminators)i + β3(Eliminator Group Gender)i + εi
As discussed earlier in Section 3, contestants’ voting incentives may vary
significantly between the pre- and post-merge stages of the game. Equation (2)
includes a dummy variable and interaction term for Post-Merge (coded as 0 if the
bootee’s elimination Tribal Council was before the merge, 1 if the bootee’s
elimination Tribal Council was after the merge) to account for the potential impact
of this strategic difference:
(Bootee is Female)i = β0 + β1(Total No. Female Eliminators)i
+ β2(Total No. Male Eliminators)i + β3(Eliminator Group Gender)i
Table VII limits the sample to only female jurors (the gender of the finalists,
however, is unconstrained) and finds that a female juror are unlikely to significantly
consider a finalist’s role in her elimination when voting at the Final Tribal Council,
but that female jurors do consider a finalist’s performance in individual challenges.
Note that all of the correlation coefficients for “Finalist’s Mean Percent Challenge
16 These three variables together function to quantify a finalist’s betrayal of a juror. If a
finalist booted a juror, and the finalist and juror had spent a majority of the game on the
same tribe and had voted the same at the majority of Tribal Councils, the juror would be
likely to feel betrayed by the finalist’s role in his/her elimination. In other words, if Finalist
Booted Juror equals 1, as the variables % of Shared Tribes Pre-Merge and % Same Votes at
Shared Tribal Councils increase, the more betrayed by the finalist the juror is likely to feel. If
the finalist did not eliminate the juror, the juror is not likely to feel betrayed by the finalist;
in this case, it is still interesting to examine the effect of amount of tribes and votes shared
by the juror and finalist (i.e. the strength of the juror and finalist’s personal and strategic
interactions) on the juror’s decision to vote for the finalist.
(3)
(3)
19
Finish” are all positive and statistically significant at the 0.01 level, which suggests
that an increase in the finalist’s mean percent challenge finish by one percent
increases the log odds of a female juror voting for the finalist by at least 3.614. The
only other significant correlation coefficients are on the “Finalist Eliminated Juror”
variable in the first and second columns. The negative value of these coefficients
suggest that if the finalist eliminated a female juror, the log odds of the female juror
voting for the finalist decrease by 0.945 and 0.934, respectively. However, since
these coefficients are not significant in columns 3 and 4, it is difficult to conclude
that a finalist’s elimination of a female juror significantly impacts her decision to
vote for the finalist.
Table VIII repeats the same regressions as Table VII does but limits the
sample to only male jurors. In this table, the only statistically significant coefficients
are on the “Finalist Eliminated Juror” variable in columns 1, 2, 3, and 4. These
coefficients indicate that if a finalist eliminated a male juror, the log odds of the
male juror voting for the finalist in the Final Tribal Council decrease by at least
0.903. In other words, a male juror’s voting decision at the Final Tribal Council does
appear to be impacted by a finalist’s role in his elimination. Male jurors, however, do
not appear to be affected by when in the game the finalist eliminated them, since
none of the correlation coefficients on “Finalist Eliminated Juror (Scaled)” are
statistically significant.
Table IX displays the regression results for all juror/finalist pairs and
includes an interaction dummy for the juror’s gender, which allows us to observe
differences between male and female jurors. First, since none of the correlation
coefficients on any of the interaction terms are statistically significant, we cannot
conclude that male and female jurors differ in how the included variables impact
their decision to vote for a finalist; as such, in this instance there is no evidence for a
gender difference. However, the table can still be used to understand which
variables affect the decision of all jurors (regardless of gender) to vote for a finalist.
Note that the correlation coefficients on “Finalist Eliminated Juror” in columns 1
through 4 are negative and statistically significant, which indicates that a juror is
less likely to vote for a finalist if the finalist eliminated the him or her. Second,
notice that the correlation coefficients for the “Finalist’s Mean Percent Challenge
Finish,” are positive and statistically significant at the 0.05 level in all columns,
20
indicating that jurors are more likely to vote for a finalist with a better individual
challenge performance record. Among all the finalist qualities explored in this
section, it appears that a relatively strong individual challenge performance record
is the only quality that will significantly improve a finalist’s chance of receiving jury
votes and thus winning Survivor.
6.3 Discussion
I hypothesized that there is a difference in how male and female jurors view
betrayal and finalists’ game performance at the Final Tribal Council. My results,
however, fail to confirm statistically different reactions to betrayal by male and
female jurors. With that said, my results do suggest that all jurors are less likely to
vote for a finalist who previously eliminated them and are also likely to regard more
highly finalists with relatively strong performances in individual challenges.
It is possible that female and male jurors do differ in their propensity to
forgive finalists but that the variables in my model do not capture the aspects of a
betrayal that are meaningful to contestants. Furthermore, my results do not
indicate that male and female jurors definitely employ the same decision-making
strategies at the Final Tribal Council; further qualitative analysis might yield more
insight into possible differences between the decisions of male and female jurors at
the Final Tribal Council.
While regression results with only a few statistically significant correlation
coefficients can often be disappointing, these results nonetheless suggest something
important about juror’s Final Tribal Council voting behavior. Since jurors do not
appear to rely heavily on any if the variables used in these regressions to make their
vote at the Final Tribal Council, there must be a number of other factors that are
significantly important to jurors across seasons, some of which may be correlated
with a juror’s gender. Based on my results, I hypothesize that the most significant
factors impacting a juror’s vote at the Final Tribal Council are qualitative factors,
such as a finalist’s social abilities, strategic accomplishments, and perceived
integrity. More thorough analysis of the qualitative factors in Survivor episodes may
help to substantiate this hypothesis.
21
7 The Impact of Abilities on Elimination Order
7.1 Identification Strategy
In my final question, I investigate how male and female contestants of particular
ability types differ in their order of elimination. I analyze contestants’ ability across
two metrics used previously in this paper: (1) contestants’ mean percent finish in
individual challenges and (2) contestants’ vote for boot percentage at Tribal
Councils. In addition, I created a third variable, the simple sum of contestants’ mean
percent challenge finish and vote for boot percentage, to capture a contestants’
overall performance, both in competitions (generally associated with physical
strength) and Tribal Councils (generally associated with strategic and social
strength).
I decided to show contestants’ average overall finish by ability in bar charts.
“Mean Overall Finish” is a variable calculated by dividing a contestant’s order of
elimination by the total number of contestants that season. This variable provides a
way of representing a contestant’s order of elimination, but is better than simply
using a contestant’s boot order because it creates a more uniform metric across
seasons with varying numbers of contestants. To create the first table, displaying
contestants’ average overall finish by their ability in individual challenges, I first
divided all contestants with non-zero “Mean Percent Challenge Finish” scores into
quartiles.17 I then calculated the “Average Overall Finish” for male and female
contestants within each “Mean Percent Challenge Finish” quartile (see Figure I). I
repeated the same procedure based on contestants’ “Vote for Boot Percent” (see
Figure II) and the sum of contestants’ “Mean Percent Challenge Finish” and “Vote
for Boot Percent” (see Figure III). My objective here is to give an overview of how the
overall finish of male and female contestants of similar abilities differ. Note that
these charts do not explain a causal relationship between male and female
contestants’ abilities and their overall finish, since the charts show averages within
groups rather than correlations between variables.
17 Recall that not all Survivor contestants participate in individual challenges since the vast
majority of individual challenges take place after the tribal merge. Contestants who do not
compete in any individual challenges have a mean percent challenge finish value equal to 0,
and are thus excluded from this first chart.
22
7.2 Results & Discussion
Figure I shows the average overall finish of male and female contestants within
quartiles for their “Mean Percent Challenge Finish.” The total sample size for this
table is 392 contestants; 203 contestants were excluded from this sample because
they did not compete in any individual challenges. Note that the first quartile, or
“Q1,” is the group of contestants with the lowest “Mean Percent Finish” scores; these
contestants, on average, performed the worst in individual challenges relative to
their peers. The contestants in the fourth quartile, or “Q4,” represent the
contestants with the top 25% of “Mean Percent Finish” scores. Figure I shows that
women with challenge performances in the first and third quartiles progressed
further on average in the game than their male counterparts. Men and women
whose “Mean Percent Challenge Finish” falls in the second quartile progressed
equally far in the game, on average. But amongst contestants with the strongest
individual challenges (Q4), the average male contestant progressed further than the
average female. Among the four quartiles, women whose average challenge
performance fell in the third quartile of contestants progressed further on average
than other female contestants; the same is true of male contestants with average
percent challenge finish in the third quartile. However, women with the strongest
challenge performances (Q4) had the worst average overall percent finish among
women in each of the quartiles, whereas men with the worst challenge performances
(Q1) progressed the least on average compared to the average male contestant in the
other quartiles.
Figure II shows the “Mean Overall Finish” of male and female contestants
within quartiles by their vote for boot percent at Tribal Councils. The sample for this
chart includes all 595 Survivor contestants. In general, the chart shows that as both
male and female contestants’ “Vote for Boot Percentages” increase, the “Mean
Overall Finish” of the average contestant also increases. This is sensible, since
voting for the boot indicates that the contestant voted with the majority of his or
tribemates at that Tribal Council. If a contestant repeatedly successfully votes for
the boot, the contestant is likely part of a majority alliance; unless a majority
alliance decides to turn on one of its members, contestants within a majority alliance
should progress farther in the game than their tribe mates who are not included in
23
the majority alliance, since the alliance will eliminate these outlying contestants
before eliminating its own members. The table shows that male and female
contestants with the best “Vote for Boot Percentages” (Q3 and Q4) finished, on
average, in approximately the same place as each other. However, among
contestants with the worst Tribal Council voting histories (Q1 and Q2), the average
male contestant progressed further in Survivor than the average female contestant.
Finally, Figure III divides contestants into quartiles based on the sum of
their “Mean Percent Challenge Finish” and their “Vote for Boot Percent.”18 This
sample includes 392 contestants (the same sample size as Figure I) as it excludes all
contestants who never participated in an individual challenge. Women with scores
in the second, third, and forth quartiles had a higher average overall finish than
men in these quartiles. However, between men and women with scores in the lowest
quartile, the average male contestant progressed further than the average female
contestant.
On the whole, most of the average overall finishes between men and women
in the same ability quartile are relatively close, separated by at most two places in
Survivor.19 Based on this observation one might hypothesize that among contestants
with similar abilities, a contestant’s gender does not significantly impact how far he
or she will progress in Survivor.
8 Conclusions
This paper makes three primary insights about the role of gender within Survivor.
First, there is an inverse relationship between the average gender of the group who
votes to eliminate a contestant and the eliminated contestant’s gender. Second,
women are more likely to band together and vote out a man, while men who vote
together do not, as a rule, use gender as a criteria to identify a target. Third, male
18 The maximum sum of a contestant’s “Mean Percent Challenge Finish” and “Vote for Boot
Percent” is 2, since the maximum value for each variable is 1. 19 Among the calculations displayed in Figures I, II, and III, the greatest difference in mean
overall finish between men and women occurs amongst contestants whose mean percent
challenge finish falls in the first quartile (the difference between female and male average
overall finish here is 0.12). In seasons with 16 contestants, outlasting one additional
contestant adds 0.0625 to a contestant’s overall finish. Therefore, a difference in average
overall finish of 0.12 represents the progression of no more than two “places” in Survivor.
24
and female jurors do not appear to regard differently a finalist’s role in their
elimination when deciding their vote at the Final Tribal Council. Thus the most
valuable strategic finding from this paper serves as a warning to male Survivor
contestants: if female contestants are voting (and successfully eliminating
opponents) together, men would do well to heed the old adage, “if you can’t beat ‘em,
join ‘em,” and either align with or defeat this female group, as these women are
more likely to vote against male contestants than female contestants.
This paper’s most significant contribution to the body of academic work on
gender and competition is the insight that women who band together tend to target
male contestants. This pattern complements the finding from Burow et al. (2017)
that women prefer competing against other women than against men. Although all
women in Survivor have self-selected to compete in the game against both male and
female opponents, it may be the case that female contestants still feel more
comfortable competing amongst other women, and are thus more inclined to align
with other women and eliminate men.
The methods and results in this paper highlight both the interesting and
challenging aspects of using Survivor to evaluate patterns of social behavior. On one
hand, Survivor acts as a self-selecting multi-period game in which contestants
compete under conditions of deprivation for $1 million, yielding human decisions
that are generally unimpeded by the situational effects of more simplistic controlled
studies, including boredom and apathy. In this way, Survivor can be a platform for
studying a multitude of social and psychological phenomena including, but also far
beyond, behavioral gender differences. My results suggest that a purely quantitative
analysis of the objective components of Survivor may not fully capture the important
factors that influence contestants’ decisions. More thorough analyses of the role of
gender and other social science theories within Survivor might consider the
qualitative relationships and personalities of individual players alongside the type of
quantitative data I identified.
25
References
[1] Antonovics, Kate; Arcidiacono, Peter and Walsh, Randall. “Games and
Discrimination: Lessons from The Weakest Link”. Journal of Human
† Finalist Eliminated Juror is a binary variable equal to 1 if the finalist voted against the juror at the juror's elimination Tribal Council.
§ % of Pre-Merge Game Played Together gives the percentage of pre-merge Tribal Councils that the juror and finalist both attended.
‡‡ Finalist's Vote for Percent gives the frequency with which the finalist voted against the boot at Tribal Council.
Conditional Fixed Effects Logit Estimates for Effect of a Finalist's Elimination of a Female Juror on
the Female Juror's Decision to Vote for the Finalist to Win Survivor
‡ Finalist Eliminated Juror (Scaled) scales the previous variable by the stage of the game in which the juror was eliminated (a higher number means
the juror lasted longer in the game).
¶ % of Same Votes at Shared Tribal Councils takes the percentage of Tribal Councils the juror and finalist both attended at which the juror and finalist
voted against the same contestant.
†† Finalist's Mean Percent Challenge Finish gives the finalist's average finish in individual challenges (a higher number means the finalist performed
better on average in individual challenges).
33
TABLE VIII
Dependent Variable: Male Juror Voted for Finalist to Win (Binary)
† Finalist Eliminated Juror is a binary variable equal to 1 if the finalist voted against the juror at the juror's elimination Tribal Council.
§ % of Pre-Merge Game Played Together gives the percentage of pre-merge Tribal Councils that the juror and finalist both attended.
‡‡ Finalist's Vote for Percent gives the frequency with which the finalist voted against the boot at Tribal Council.
Conditional Fixed Effects Logit Estimates for Effect of a Finalist's Elimination of a Male Juror on the
Male Juror's Decision to Vote for the Finalist to Win Survivor
‡ Finalist Eliminated Juror (Scaled) scales the previous variable by the stage of the game in which the juror was eliminated (a higher number means
the juror lasted longer in the game).
¶ % of Same Votes at Shared Tribal Councils takes the percentage of Tribal Councils the juror and finalist both attended at which the juror and finalist
voted against the same contestant.
†† Finalist's Mean Percent Challenge Finish gives the finalist's average finish in individual challenges (a higher number means the finalist performed
better on average in individual challenges).
34
TABLE IX
Dependent Variable: Juror Voted for Finalist to Win (Binary)
† Finalist Eliminated Juror is a binary variable equal to 1 if the finalist voted against the juror at the juror's elimination Tribal Council.
§ % of Pre-Merge Game Played Together gives the percentage of pre-merge Tribal Councils that the juror and finalist both attended.
‡‡ Finalist's Vote for Percent gives the frequency with which the finalist voted against the boot at Tribal Council.
§§ Dummy for Female Juror equals one if juror is female.
Conditional Fixed Effects Logit Estimates for Effect of a Finalist's Elimination of a Juror on the Juror's
Decision to Vote for the Finalist to Win Survivor , with Interaction Term for Juror's Gender
‡ Finalist Eliminated Juror (Scaled) scales the previous variable by the stage of the game in which the juror was eliminated (a higher number means the juror lasted longer
in the game).
¶ % of Same Votes at Shared Tribal Councils takes the percentage of Tribal Councils the juror and finalist both attended at which the juror and finalist voted against the
same contestant.
†† Finalist's Mean Percent Challenge Finish gives the finalist's average finish in individual challenges (a higher number means the finalist performed better on average in
individual challenges).
35
FIGURE I Male and Female Overall Mean Percent Finish
by Average Challenge Finish Quartiles
Source: Survivor Wiki and True Dork Times
Note: all contestants were divided into quartiles based on mean percent finish in challenges, then
average finish for men and women was computed within each quartile. A contestant's average
finish is calculated by dividing the order in which a contestant was eliminated by the total
number of contestants that season. For example, in a season with 16 contestants, the last place
(i.e. 16th place) finisher would have an overall percent finish of 0.0625. The season winner always
has an overall percent finish of 1.
0.620.64
0.720.68
0.74
0.64
0.77
0.63
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
Q1 Q2 Q3 Q4
Mean
Overa
ll P
erc
en
t F
inis
h
Challenge Mean Percent Finish, Quartiles
Male Female
36
FIGURE II Male and Female Overall Mean Percent Finish
by Vote for Boot Percent Quartiles
Source: Survivor Wiki and True Dork Times
Note: all contestants were divided into quartiles based on vote for boot percent at Tribal Councils,
then average finish for men and women was computed within each quartile. A contestant's
average finish is calculated by dividing the order in which a contestant was eliminated by the
total number of contestants that season. For example, in a season with 16 contestants, the last
place (i.e. 16th place) finisher would have an overall percent finish of 0.0625. The season winner
always has an overall percent finish of 1.
0.31
0.52
0.58
0.80
0.21
0.47
0.60
0.80
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
Q1 Q2 Q3 Q4
Mean
Overa
ll P
erc
en
t F
inis
h
Vote for Boot Percent, Quartiles
Male Female
37
FIGURE III Male and Female Overall Mean Percent Finish by Sum of
Mean Percent Challenge Finish and Vote for Boot Percent Quartiles
Source: Survivor Wiki and True Dork Times
Note: all contestants were divided into quartiles based on the sum of their mean percent finish in
challenges and vote for boot percent at Tribal Councils, then average finish for men and women
was computed within each quartile. A contestant's average finish is calculated by dividing the
order in which a contestant was eliminated by the total number of contestants that season. For
example, in a season with 16 contestants, the last place (i.e. 16th place) finisher would have an
overall percent finish of 0.0625. The season winner always has an overall percent finish of 1.
0.55
0.63
0.70
0.79
0.53
0.71
0.75
0.82
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
Q1 Q2 Q3 Q4
Mean
OV
era
ll P
erc
en
t F
inis
h
Challenge Mean Percent Finish + Vote for Boot Percent, Quartiles