THE UNIVERSITY OF CHICAGO

Self-fulfilling Stigmatization: An Evolutionary

Game of Online Opinion Dynamics in Social
Movements

By

Hongding Zhu

July 2022

A paper submitted in partial fulfillment of the requirements for the

Master of Arts degree in the

Master of Arts Program in the Social Sciences

Faculty Advisor: Zhaotian Luo

Preceptor: Yan Xu
Self-fulfilling Stigmatization: An Evolutionary Game

of Online Opinion Dynamics in Social Movements

Hongding Zhu, University of Chicago

July 29, 2022

Abstract

To illustrate how the stigmatization of social movements arises from its micro-

foundations—emotional confrontation, I model the process of such stigmatization

on social media using an evolutionary game with two sub-populations—the par-

ticipants of a social movement and outsiders in the general public. Based on the

critical assumption of “hostility as retaliation”, the members of the movement, who

are hostile toward the general public, and haters in the general public reproduce

each other. Suppose there are enough such hostile members so that it exceeds the

stigmatization threshold. Then, the social movement is stigmatized and locked in

conflicts with haters outside the movement. The stigmatization of a social move-

ment can be a self-fulfilling prophecy. When there are initially enough haters, even

if a social movement is peaceful at the beginning, the speed at which the movement

becomes hostile under outgroup pressure will outpace the rate at which it can win

public support, and finally induces the stigmatization equilibrium.

Keywords: Social movement; stigmatization; evolutionary game; Echo Chamber;

censorship.
MA Thesis 1

1 Introduction

More and more social movements—from Black Lives Matter to MeToo, from the climate

movement to Occupy Wall Street—are being initiated and are happening online (Hara

and Huang, 2011). The ease of setting up groups, the global reach of networked com-

munities, and the convenient facilitation of organizational activities all contribute to the

widespread use of social media sites for social movements (Wall, 2007). Moreover, un-

der the tremendous political risk of offline activism, social movements in autocracy rely

on online discourse even more than their counterparts in democracy (Feng, 2017; Tan,

2017). In cyberspace, people can quickly obtain information about a social movement and

interact with its participants, forming their opinions about the movement accordingly.

Stigmatization refers to that social movements are misunderstood by society as the

advocated value are distorted, and the participants are tagged with negative character-

istics. We see significant conflicts between participants of a social movement and the

general public. Previous studies Harel et al. (2020); Bar-Tal (2007) notice that ideolog-

ical divergence and emotional confrontation are the micro-foundations of social conflicts

in cyberspace. Conflicts exist in both intergroup and intragroup interactions, and out-

group pressure leads to further polarization inside a specific social group. Meanwhile,

social media amplify the emotional factors that drive the conflicts. I follow this micro-

interaction-based and emotion-driven approach to study the process of stigmatization of

social movements in a dynamic process, discussing how the macro outcome of stigma-

tization emerges from micro-interactions between participants and non-participants of a

social movement online.

I first go through the informal literature related to the conflicts in social movements,

narrowing down the analytical scope of stigmatization to its micro mechanism—people’s

2 Hongding Zhu

online interaction, and arguing that conflicts can manifest themselves through the device

of emotional antagonism. Second, I discuss the limitation of existing formal models

on their application to the question of interest. Then, using evolutionary game theory,

I build a model to capture the dynamics of opinion evolution and conflict between two

populations—the participants of a social movement and the general public. In the model,

participants are either hostile or moderate towards the general public, while the outsiders

of the movement are either haters or sympathizers. Conflicts happen inside the movement

between hostile and moderate participants of a social movement; outside the movement,

only moderate participants and sympathizers maintain peace. The assumption that a

peaceful person loses more utility in conflicts than an emotionally hostile person drives

all the implications of the model.

The model has several theoretical implications. First, hostile participants of a social

movement and haters in the general public conflict with but also reproduce each other.

Second, there exists a stigmatization threshold so that if there are enough hostile mem-

bers of a social movement and haters in the general public, the evolutionary outcome is

the stigmatization equilibrium in which all players choose the hostile/hater strategy. The

larger the initial proportion of haters in the general public, the less hostile a social move-

ment is needed to induce this stigmatization equilibrium. Third, the stigmatization of

social movements is a self-fulfilling prophecy. In a specific situation, the social movement

is not hostile at the beginning and is winning more sympathizers outside the movement,

but enough level of initial stigmatization (haters in the general public) is creating more

hostile participants of a social movement. If the latter effect is stronger than the former

effect, then hostility toward the social movement still crosses the stigmatization thresh-

old. In the situation above, severe stigmatization at the beginning produces a social
MA Thesis 3

movement that is antagonistic to the general public in the end. In others words, when

the general public expect the movement to be extreme at the beginning, the movement

will become extreme at the end.

2 Literature review

2.1 Conflicts, stigmatization, and emotions in social movements

Some social movements suffer severe stigmatization among the public (?Kim, 2018; Bav-

oleo and Chaure, 2020; Yang, 2014; Ka, 2021). Take the feminist movement worldwide as

an example. South Korea’s backlash against radical feminism became a solid voting base

for conservative parties in the 2022 presidential election(Kim and Lee, 2022). In China,

feminist expression online easily suffer from trolling, (Mao, 2020) and the antagonism

between men and women has become a salient phenomenon in Chinese cyberspace 1 . In

India, feminism is widely perceived as a western ideological invasion and hated among

male anti-feminist Nicholas and Agius (2018); Rothermel (2020). Conflicts are inevitable

between the thriving, progressive force and reactionary social structure (Pachter, 1974).

The reactionary structure and its apparatuses stigmatize their challengers and have many

names—racism, sexism, homophobia, and so on. On the individual level, stigmatization

happens when some people hold false beliefs that a social movement and its participants

are negative. Therefore, stigmatization is ideological instead of material. Scholars dis-

cuss who stigmatizes and why they stigmatize. The privileged group usually has limited

empathy for the oppressed (DiAngelo, 2018). It thus holds negative perceptions of social

movements that try to address social injustice and stigmatize them as“seeking privilege”

and “creating oppression ”(Dignam and Rohlinger, 2019; Van Valkenburgh, 2021).
1
see https://www.bbc.com/zhongwen/simp/chinese-news-55571627
4 Hongding Zhu

However, social movements also find their allies in the general public. Whether the

privileged can be recognized as participants of a social movement has always been de-

bated. For example, they De Beauvoir (2010) argued that men could not be called fem-

inists because they lacked life experience as women. Putting aside the naming problem,

apart from those who already identify themselves as participants of the movement, some

sympathizers are generally pro-movement or open to its ideas. For instance, the sup-

port of African American men is essential in the practice of Black Feminism—empathy

and support, if not “participation”, were made possible by the unity legacy of shared

participation in the civil rights movement (hooks, 2000).

Besides stigmatization and conflict outside the social movements, there are also in-

tergroup conflicts between participants with different ideologies in a social movement.

Harel et al. (2020) Shows that unaligned ideology in reciprocal interactions is an im-

portant foundation for conflicts to happen. Ideological divergence translates into hostile

emotion and behavior, and intragroup ideological pressure further leads to opinion po-

larization and friction inside the socio-political group. Work Bar-Tal (2007) argues that

in intractable social conflicts, people justify their own group and stigmatize other groups

as “evil”. Hence, ideological divergence causes conflict and stigmatization. In social

movements mainly, conflict happens between the moderate and radical factions. The

Baader-Meinhof (Aust, 2009) complex was one of the most extreme parts of the radical

left in West Germany, and it finally broke up with the more moderate left-wing, resort-

ing to terrorism to achieve its cause. The Scum ManifestoSolanas and Avital (2004)

broadcasts the voice of radical feminism, and this legacy is appreciated and criticized by

different feminist practices.

The discussion above indicates that first, there are several different strategies in the
MA Thesis 5

framework of a social movement—participants with varying ideologies in the movement,

sympathizers in the general public, and enemies in the general public; second, ideological

divergence is the micro foundation of conflicts and stigmatization in social movements.

Then, what kind of ideologies divergence mainly cause conflict and stigmatization? Koo

(2020) understands conflicts and stigmatization in the South Korean radical feminist

movement as the result of trolling behavior—which is mainly emotion-driven and does

not involve clear political agenda. Feminists in such forums like Megalia and Womad

were trolled because they were not aggressive to biological men enough or did not agree

with the tactic of “mirroring misogyny”. Kim (2018) argues that the trolling and gender

essentialism of such feminist practice is criticized as nonconstructive by its opposition. Ka

(2021) also documents that rather than political blueprints, stigmatization and conflicts

are driven by hostile sentiments. She also notices that even though there is a high diversity

of ideologies in a social movement, people use reduced tags to label themselves as “radical”

and “moderate” feminists, which is an emotional division of how hostile you are towards

men and patriarchy rather than their political agendas. For instance, participants of

the 6B4T feminist movement in China and South Korea usually identify themselves as

radical feminists due to their strong criticism of patriarchy. However, they only adopt

individualistic resistance such as “no marriage” to resist without radical demand for social

reform de facto. Hence, the conflicts in social movements may not necessarily be between

radical and moderate groups but between people with hostile emotions and less hostile

emotions towards the status quo. Similarly, people in the anti-movement forces also have

a higher emotional hostility towards the movement, while self-identified sympathizers are

less hostile.

I consider this emotion-driven approach valuable to help explain conflicts and stigma-
6 Hongding Zhu

tization of social movements in the social media era. I then discuss how emotion and

hostility are amplified by social media infrastructure.

2.2 Social movements in social media era

In the era of social media, some mechanisms make online communities more prone to in-

terpersonal emotional hostility and conflict. First, social media reduces the coordination

cost of social movements Van Laer and Van Aelst (2010); Kelly Garrett (2006), facilitating

the spread of opinion and creating more chances for online interaction related to a par-

ticular social movement. For example, Crossley (2015) argues that social media not only

helps the mobilization of feminism but also amplifies the voice of feminist influencers and

helps information diffusion. However, social media also has the same effect on the anti-

movement force, intensifying the conflict between the two forces. Second, information

segregation and the logic of community feedback help social media amplify the impact

of group polarization (Sunstein, 1999), making homophily and opinion polarization is a

common feature on the internet (McPherson et al., 2001). This opinion polarization is

more likely to happen when the topic is political (Barberá et al., 2015), and it naturally

translates into sentiment polarization (Harel et al., 2020)—more extreme emotional sta-

tus of people overall. Empirical evidence shows that negative emotion (Del Vicario et al.,

2016) is infectious, and someone who receives negative feedback tends to send more of

such feedback to others in the futureCheng et al. (2014). Therefore negative emotion

among the population reinforces itself on social media platforms. Finally, the anonymous

online environment reduces the cost of trolling and other violence and amplifies the voices

of fight-pickers in real life, contributing to conflicts (Bor and Petersen, 2022).

This branch of literature conveys that social media amplify ideological and emotional
MA Thesis 7

polarization. It also supports the leverage of an emotional-driven approach to under-

standing political phenomena on the internet. These studies concentrate on technological

aspects and platform interaction logic, emphasizing the interaction between people and

the dynamic nature of opinions and emotions. Next, I will show models related to the

dynamic evolution of opinions and discuss where my model can build on.

2.3 Related Models

Linking macro and micro level of analysis is an important task for social scientist(Coleman,

1994; Sawyer and Sawyer, 2005; Raub et al., 2011). It is usual to set up a formal model to

discuss how macro outcomes emerge from micro-interactions. Literature on the stigmati-

zation of social movements and social media suggest that the micro foundation of stigma-

tization and conflicts in social movements is ideological divergence—the divergence of

people’s opinion and attitudes. Accordingly, we need a model that can explain stigmati-

zation by showing the process of opinion evolution, linking the macro outcome with the

micro mechanism.

Previous models that involve opinion evolution incorporate one single population only

and consider “persuasion” in multiple forms are the mechanism. In other words, people

change their minds because of social feedback (Banisch and Olbrich, 2019; Bertotti and

Delitala, 2008; Chen et al., 2020; Di Mare and Latora, 2007; Yuan et al., 2021). Some

benchmark models try to find the condition of convergence or explain the formation

of opinion polarization. In contrast, others extend them in different social networks,

under uncertainty, or in a multi-dimensional environment. No model so far involves two

divergent populations with different interaction rules.

Furthermore, no model considers the effect of censorship, which is a crucial and om-
8 Hongding Zhu

nipresent feature in autocracy. Therefore, incorporating censorship into the model of

opinion evolution can enrich the knowledge of the effect of censorship. Currently, stu-

dents of authoritarian politics have discussed the mechanism of censorship as Bayesian

persuasion (Gehlbach et al., 2021) and its role in preventing collective action (King et al.,

2013) and discouraging opposition but also bringing about self-censoring, which reduces

the information capacity of an autocrat. Nevertheless, no study has yet examined the

effect of censorship on the evolution of social movements.

My model contributes in three aspects. First, I involve the dynamics of opinion

evolution of two heterogeneous population, participants of the social movement, and

outsiders in the general public, rather than a single population. This captures the reality

that intergroup and intragroup interaction both exists (Harel et al., 2020), and allows me

to examine the effect of outgroup pressure (emotional hostility) on the opinion evolution

of the social movement. Second, my model provides another possible approach to opinion

evolution that results from netizens’ population replacement and culture inheritance in

a given environment, complementary to the approach of persuasion via social feedback

in the existing literature. This setting is a weaker assumption that allows people to be

stubborn—opinions as strategies are assigned and unchangeable once the players enter the

game, rather than assuming that they easily change their minds due to others’ opinions.

Third, I incorporate the effect of censorship and Echo Chamber into my model as two

extensions, generating new theoretical implications.

2.4 Comments on previous literature

The stigmatization of social movements happens due to ideological divergence between

the participants of social movements and the general public. The macro outcome of
MA Thesis 9

stigmatization emerges from ideological conflicts in the micro-interaction of people online,

and this requires a model that is able to link both macro and micro levels. In the social

media era, such ideological conflicts happen on digital platforms, and emotional hostility

is amplified by the technological properties of social media. Hence, emotional hostility

affects the stigmatization of social movements and online conflicts more in the social

media era, which requires the academia to incorporate the emotion-driven approach in

understanding the dynamic process of the stigmatization of social movements.

Several models of opinion dynamics relate to my research question in that they discuss

how people’s opinion change and how macro pattern is produced through this dynamic.

However, they are not enough to settle my question. First, they only contain one single

population, neglecting the reality that conflicts exist in both intergroup and intragroup.

Second, they also do not involve censorship, which widely exists in authoritarian regimes

and has a significant impact on the stigmatization of social movements. Third, previ-

ous models only consider the persuasion effect of feedback as the driving force of the

opinion dynamics, ignoring alternative approaches such as population replacement under

evolutionary pressure. To build on previous literature, I set up an evolutionary game

theory model to capture the opinion dynamics of social movements online. My model

contains two populations —the participants and outsiders in the general public, consider-

ing evolutionary fitness and population replacement as a new approach to frame opinion

dynamics and discuss the effect of censorship and Echo Chamber as model extensions.

The implications of my model help the understanding of my research questions—how the

stigmatization of social movements emerges from micro-level ideological and emotional

conflicts among people.

10 Hongding Zhu

3 Method

This paper is the first one to leverage evolutionary game theory as a tool to analyze the

opinion dynamics of social movements. Hence, a basic introduction to evolutionary game

theory and its suitability for my analysis is needed.

In traditional non-cooperative games, scholars assume players to be rational and

strategic (Osborne et al., 2004; McCarty and Meirowitz, 2007). However, the strong

assumption of rationality is not always the case. Apolitical random events impact voters’

behavior, such as the occurrence of shark attacks influencing election outcome (Achen

and Bartels, 2012); ignorance is an important explanatory variable of why voters support

trade protectionism (Rho and Tomz, 2017). These examples suggest that in discussing

decentralized group behavior, it is more realistic to set up weaker assumptions about the

rationality of players. In evolutionary game theory, players are considered to be non-

strategic, so their strategies (genes) are assigned once they enter the game, and it never

changes. Therefore, in a basic evolutionary game theory setup, a player’s strategy is the

same as her type. The proportions of different strategies in a certain population change

according to natural selection pressure—different strategies have different levels of fitness

(payoffs) as a function of the overall strategy profile (the proportions of all strategies of

all populations), and the probability that a certain strategy leave ”offspring” is positively

correlated with their fitness. Hence, the fittest survive in the evolutionary game, and

strategies with low payoffs decline.

As a modeling method initiated in biological studies, is it feasible for social scientists

to apply evolutionary game theories? Weibull (1997) and Friedman (1998) both introduce

the evolutionary game to social science and argue that evolutionary game theory is of

great value for analyzing large-scale decentralized behavior of players (individuals, firms,
MA Thesis 11

and so on). Different from biology, the inheritance of strategies in social science is through

norms, culture, and social networks. For the research question of this paper—how the

stigmatization of social movements emerges from online interaction—the targeted pop-

ulation (netizens) is vast, and it is natural to consider them non-strategic. Empirical

evidence shows that the behavior and opinion of new users of a specific social media plat-

form are strongly decided by the pattern of current existing users (Cheng et al., 2014),

which satisfies the “inheritance” approach of evolutionary games. Furthermore, the evo-

lutionary game model assumes random matching between players, which is common in

social media subgroups where everyone has access to interact with other members. It

is easy to incorporate the effect of homophily to make the matching rule more realistic.

Therefore, an evolutionary game is suitable to depict the dynamics of opinions in a spe-

cific online community. Readers should restrict the interpretation of my model to small

to medium-size online communities where netizens intensively interact with each other.

The next section is a detailed introduction of the baseline model setup, including the

basis of specific settings and the justification of the assumptions.

4 Model

There are two sub-populations in the game: participants of a social movement A and

outsiders in the general public B. The whole population is normalized to 1, so that

A + B = 1. Among the participants exists two strategies, hostile A1 and moderate A0

and among the general public exists two strategies, haters B1 and sympathizers B0 . Table

1 shows the conflict-peace relations between the four strategies.

The two strategies of participants of a social movement is divided by their emotional

attitudes—be friendly/hostile to whom. Hostile participants (A1 ) consider moderate

12 Hongding Zhu

Table 1: Conflict Relations Between The Four Strategies

A1 Hostile A0 Moderate B1 Hater B0 Sympathizer
A1 Hostile Peace Conflict Conflict Conflict
A0 Moderate Conflict Peace Conflict Peace
B1 Hater Conflict Conflict
B0 Sympathizer Conflict Peace

participants and all individuals outside the movement evil and only maintain peace with

their kind. Examples of this strategy could be gender essentialist feminism discussed

in Koo (2020). Readers should notice that this is an ideal type. Meanwhile, moderate

participants (A0 ) can maintain peace with their own kind and moderate outsiders but

will conflict with any hostile individual. Among the general public, the sympathizers

B0 can interact peacefully with moderate participants of the social movement but not

the hostile participants; the haters B1 hate the movement and conflict with both types

of participants. The alternative rightists make ideal examples of the “hater” strategy

in online interactions (Van Valkenburgh, 2021; Dignam and Rohlinger, 2019). The level

of stigmatization of a social movement is represented by the proportion of players in

population B that adopt the hater strategy.

The initial proportions of all strategies are decided by factors not included in the

model, such as culture, propaganda, or historical memory. The strategies of players when

T = 0 refers to their initial expectation of how other players’ strategies are, because it is

before actual interactions.

The participants interact with both participants and outsiders, while outsiders only

play with participants. This setup is justified that a participant may spend much time

getting information from and chatting with both their peers in the movement (for alliance

or opinion exchange) and outsiders (for propaganda and voicing for the movement). In

contrast, a typical outsider in the general public spends limited time deciding whether to
MA Thesis 13

hate or sympathize with the movement based on their interaction with the participants.

Outsiders do not spend much time discussing the movement with their peers (there are

so many social movements, but so much work and so little time!), which means the game

between outsiders is trivial.

The baseline model adopts random matching between players. The chance that a

player meets a certain kind of strategy depends on the global proportion of that strat-

egy. The definition of “interaction” in this study is general—it can be direct and dyadic

communication between two players, one-way observation of information from another

player, or small group discussion—and could be seen as an abstraction of online informa-

tion flows.

Table 2: The Strategic Form of Fitness

A0 A1 B0 B1
A0 1, 1 1 − e, 1 − m 1, 1 1 − e, 1 − m
A1 1 − m, 1 − e 1, 1 1 − m, 1 − e 1 − γe, 1 − γe

The fitness form (Table 2) of an evolutionary game is similar to the strategic form

of payoffs in traditional non-cooperative games. The fitness is substantively ordinal,

decided by several assumptions. First, people lose psychological utility from conflicts

online. Second, in conflicts, people with moderate attitudes (A0 and B0 ) lose more

utility than hostile people (A1 and B1 ), which is the critical assumption of this model.

This assumption is inspired from the theory of emotional contagion Del Vicario et al.

(2016)—the contagious nature of antagonism shows the supermodularity of antagonistic

emotion so that it should be a dominant strategy when facing another player with hostile

strategy. The second assumption means that m < e ∈ (0, 1) and γ ∈ (0, 1). Let aα,β
14 Hongding Zhu

denotes the fitness when strategy α meets β. The fitness of a strategy is:

X X
(1) uAi = Aj aAi ,Aj + Bj aAi ,Bj

X
(2) uBi = Aj aBi ,Aj

After clarifying the fitness form, we can calculate the replicator dynamics—the growth

rate of a particular strategy in terms of the proportion in its own sub-population with

respect to continuous time parameter t. The decision of signs of the replicator dynamics

are simple: the proportions of those strategies with higher-than-average fitness in their

own sub-population grow. The following equations specify the growth rate of a strategy:

∂Ai A0 uA0 + A1 uA1

(3) = Ai (uAi − ) = Ai A−i (uAi − uA−i )
∂t A0 + A1

∂Bi B0 uB0 + B1 uB1

(4) = Bi (uBi − ) = Bi B−i (uBi − uB− i )
∂t B0 + B1

The sequence of the game is:

1. Nature chooses the initial xt0 =(A1t0 , B1t0 ) by a generic distribution F , so that the

proportions of all four strategies are decided.

2. All players in the whole population interaction with each other in time t according

to a certain matching rule (pure random matching in the baseline model).

3. The fitness of the strategies is realized by aggregating the fitness of players that

adopt it.

4. The proportion of the strategies grow/decline according to the replicator dynamics

MA Thesis 15

until x converges to any evolutionary stable state (ESS).

5 Analysis

By looking at the fitness form, we infer that this game is similar to a coordination game

where players across populations want to match their types. In other words, if new

entering players in time t anticipate other players more likely to be moderate (A0 or

B0 ), they are more likely to inherit the moderate/sympathizer strategy, otherwise they

are more likely to inherit the hostile/hater strategy. There exists three Nash equilibria

in a basic coordination game, but the unique interior solution is not an ESS because it

does not have a uniform barrier of invasion (Weibull, 1997) so that any external shock at

the interior Nash equilibrium will lead to convergence to the other two ESSs. Therefore,

there are two ESSs in the game.

Proposition 1 (Evolutionary Stable States) There exists two evolutionary stable state.

First, the peace equilibrium xP∗ = {A1 = 0, B1 = 0} in which all players are moder-

ate participants or sympathizers of the movement; second, the stigmatization equilibrium

xS∗ = {A1 = 1, B1 = 1} in which all players are hostile participants or haters of the

movement.

The proof of which is trivial. Proposition 1 shows that in any ESS, both participants

and outsiders match their type. The first ESS in which all players adopt the moderate

or sympathizer strategies is the peace equilibrium and there is no conflict and stigma-

tization, while the second ESS in which all players adopt hostile or hater strategies is

the stigmatization equilibrium, and there are conflicts and stigmatization between the

movement and the general public. The result is not intuitive since it is common that
16 Hongding Zhu

opinion divergence exists in the movement and the general public empirically. However,

readers should notice that, first, the interaction environment in the model is where all

players can interact with each other, hence the equilibrium result is restricted to certain

online communities, forums, or platforms, rather than the internet as a whole; second,

the equilibrium can be interpreted as the tendency and direction of evolution according

with respect to time t, therefore what we observed in real life can be still in-progress of

the evolution.

Now I analysis the dynamics of evolution. Before the global mixed strategy x con-

verge to any ESS, the proportion of strategies change according to the signs of replicator

dynamics. Because there are only two strategies in a certain sub-population, if the pro-

portion of one strategy grows, the other one must decline. Therefore, the conditions for

the proportions of moderate participants A0 and sympathizers B0 to grow, so that the

proportions of hostile players A1 and B1 decline, are given by:

(5) uA0 > uA1

(6) uB0 > uB1

which solve four domains divided by two curves, C1 and C2 so that:

m
(7) C1 : A1 =
e + m − eγ

(ep + mp)A1 m
(8) C2 : B1 = − +
(1 − p)(e + m − eγ) (1 − p)(e + m − eγ)
MA Thesis 17

and the four domains are shown in Figure 1. In figure 1, point (A∗1 , B1∗ ) is the interior

Nash equilibrium of the game, C1 satisfies uB0 = uB1 and C2 satisfies uA0 = uA1 . Denote

the proportions of strategies in time t as xt . If xt is in Area 1, A1 and B1 both grow,

and the game converges to the stigmatization equilibrium xS∗ , else if xt is in Area 2, A1

and B1 both decline, and the game converges to the peace equilibrium xP∗ . This result is

intuitive and reflect the situations in reality that when both participants and outsiders of

a social movement is hostile (moderate) enough, the relationship between the movement

and the general public is high-conflict (peaceful). The property of replicator dynamics in

Area 1 and 2 is formalized as Lemma 1, which is important for analysis of Area 3 and 4.

Lemma 1 (Replicator Dynamics in Area 1 and 2) In any time t, if xt = {A1t , B1t }

m (ep+mp)A1t m
satisfies that A1t > e+m−eγ
and B1t > − (1−p)(e+m−eγ) + (1−p)(e+m−eγ) , the evolutionary out-

come converges to xS∗ when t → +∞; if xt = {A1t , B1t } satisfies that A1t < m
e+m−eγ
and
(ep+mp)A1t
B1t < − (1−p)(e+m−eγ) + m
(1−p)(e+m−eγ)
, the evolutionary outcome converges to xP∗ when

t → +∞.

It is more interesting to discuss dynamics in Area 3 and 4, In these two areas, the A1

and B1 evolve in the opposite direction. In Area 3, there are too many haters B1 in the

general public so that the proportion of hostile participants grows as retaliation, while at

the same time, because there are enough moderate participants to interact peacefully with

the general public, the proportion of sympathizers of the movement also grows. Situation

in Area 4 mirrors the dynamics in Area 3 so that the social movement is becoming

moderate while the proportion of its haters in the general public grows. Clearly, the

evolutionary outcome in Area 3 and 4—whether it converges to the peace equilibrium

xP∗ or the stigmatization equilibrium xS∗ —depends on the relative speed at which two

sub-populations evolve in opposite directions. Take dynamics in Area 3 for example,

18 Hongding Zhu

if the speed at which the movement becomes more hostile overwhelms the trend of the

general public to be more friendly to the movement, xt will reach Area 1 at some t, and

according to Lemma 1, the game ends with xS∗ .

Figure 1: Absorption Domain of ESSs

To complete the discussion of dynamics in Area 3 and 4, I now formally derive the

absorption domains of the two ESSs, which is equivalent to derive the solution of the

initial value problem for an ordinary differential equation system f (t, x).



 ∂A1




 ∂t
∂B


 1


(9) f (t, xt0 ) = ∂t

A1 (0) = A1t0









B1 (0) = B1t0


MA Thesis 19

Particularly, the initial value problem f (t, (A∗1 , B1∗ )) solves a unique curve C ∗ that an-

nounces the absorption domains of the two ESSs. Therefore, C ∗ is the Stigmatization

Threshold.

Proposition 2 (Stigmatization Threshold) There exists a unique curve C ∗ that goes

through the interior Nash equilibrium (A∗1 , B1∗ ). Any xt0 below C ∗ leads to xP∗ and any

xt0 above C ∗ leads to xP∗ .

The proof of Proposition 2 is in the appendix. The baseline model generates three the-

oretical implications. First, the dynamic of large enough hostility and stigmatization

between the movement and the general public is a process of reinforcing each other. Sec-

ond, the dynamics in Area 3 shows that stigmatization of a social movement can be a

self-fulfilling prophecy when the proportion of haters are large enough. Haters in the gen-

eral public stigmatize and attack all participants of the movement, which in turns creates

more hostile participants in the movement and if this “hostility as retaliation” effect is

strong enough, everyone becomes hostile to each other in the end. Third, the shape of

the stigmatization threshold C ∗ indicates that if one side of the game, participants or

outsiders, is more hostile/hating, it lowers the threshold of sufficient hostility of the other

side to predict a stigmatization equilibrium.

6 Extensions

6.1 Echo Chamber

Homophily is a salient phenomenon in today’s society, especially in cyberspace (Boutyline

and Willer, 2017; Barberá et al., 2015; McPherson et al., 2001; Fiore and Donath, 2005;

Sánchez et al., 2016). People tend to interact with other people of similar ideas, creating
20 Hongding Zhu

an Echo Chamber that exacerbate the homogeneity of opinions locally. Echo Chamber

effect is strongly correlated to opinion polarization. Intuitively, the Echo Chamber effect

seems relate itself with more stigmatization and conflict online. But is it the case in this

model?

The baseline model predict two ESSs, which present convergence instead of polariza-

tion within a single population, however, the two equilibrium are different in terms of

the intragroup convergence-polarization outcome. In the peace equilibrium, both partici-

pants and outsiders are inclusive to each other without conflict, thus meaning convergence

between the two sub-populations; in the stigmatization equilibrium, though both popu-

lation are internally convergent, they attack each other in intragroup interaction, thus is

more close to polarization. To be clear, in this model, the stigmatization equilibrium is

equivalent to polarization. In this equilibrium, players have the right beliefs about other

players in the end, but the self-fulfilling nature of the bad equilibrium indicates that there

can be misunderstandings at the beginning of the game.

To incorporate the Echo Chamber effect, it is convenient to set a fixed probability

hi that the strategy Ai meet itself in time t, and probability 1 − hi that Ai is randomly

matching with all other strategies just like in the baseline model. The fitness of B is

unchanged. The new fitness function for A is:

X X
(10) uhAi = hi + (1 − hi )( Aj aAi ,Aj + Bj aAi ,Bj )

which is an affine transformation of the fitness function in baseline. If h0 = h1 = h,

∂Ai
the sign of ∂t
is decided by uhAi − uhA− i = (1 − h)(uAi − uA− i ), which means that

when both strategies have the same level of homophily, the Echo Chamber effect only

changes the velocity of replicator dynamics, making the elimination of one strategy slower,
MA Thesis 21

but does not change the evolutionary result. Only when h1 > h0 , which means hostile

participants A1 are more like “birds of a feather” than moderate participants A0 , will the

Echo Chamber effect contributes to stigmatization.

Proposition 3 (The Effect of Echo Chamber) There is a unique stigmatization thresh-

old Ch∗ that goes through the interior Nash equilibrium {A∗1h , B1h
∗
}. Ch∗ is below C ∗ if and

only if h1 > h0 . Echo Chamber enlarges the strategy space that induces xS∗ if and only if

the homophily effect of A1 is stronger than A0 .

The proof of Proposition 3 is in the appendix. When the Echo Chamber effect among

the hostile participants is stronger than it among the moderate participants, the strategy

space that induces the peace equilibrium is smaller than in the baseline model because the

stigmatization threshold Ch∗ in the Echo Chamber model is strictly below the threshold

in the baseline model (see figure 2). Hence, surprisingly, in our model, homophily does

not necessarily help stigmatization, but different level of homophily between different

strategies does. The intuition is that while Echo Chamber among the hostile participants

secures more psychological payoff for new entering users to be hostile, it has the same

effect of encouraging users to be moderate. When people talk about Echo Chamber and

polarization, they assume if people have more exposure to diverse information, there

will be less polarization, so people’s opinion is a balance between different source of

information. However, in this model, I show that the interaction with outside information

does not necessarily make a person similar to them, but instead, when the interaction

is not satisfactory, it strengthens the existing biases. The counter-intuitive result arises

because if moderate people have more “protection” from emotionally hostile content, they

are more like to stick to their strategy. This is the case when social media is not founded

and people with extreme stances are hard to find their allies. Therefore, Echo Chamber
22 Hongding Zhu

effect on social media helps more for extreme opinions rather than moderate opinions

because such Echo Chamber does not widely existed before social media era.

Figure 2: The Effect of Echo Chamber and Censorship on Stigmatization

6.2 Censorship

What is censorship? Usually censorship is considered as information manipulation that

selectively delete information or even prevent the generation of information. Yet in the

spirit of Adena et al. (2015), censorship could also be a form of repression that warning

someone that her opinion is not appreciated. I incorporate both effects of censorship by

setting up a probability b that interaction with a certain kind of strategy is blocked (in-

formation manipulation) and a dead weight cost c on the fitness of strategies (repression).

Similar to the previous extension, we know that if the effect of censorship on the fitness

is symmetric for both A1 and A0 , then the replicator dynamics is unchanged. Therefore,

for parsimony and clarity, I only add the treatment on the ”more censored” strategy.

In autocracy, censorship mainly targets posts with collective action potential (King
MA Thesis 23

et al., 2013) and alternatives of the official ideology (Shue, 1990) which is generated

more by the moderate participants of a social movement, rather than hatred speech,

trolling, or other forms of offense and language violence generated by hostile participants.

Take the Douban forum, the headquarter of Chinese feminist movement, as an example,

censorship targets at Douban groups that systematically criticize patriarchy and voice

clear political agenda, rather than the groups that express hatred towards biological men

or troll their enemies2 , which makes moderate participants instead of hostile participants

more vulnerable to censorship because they are emotionally moderate but politically

radical while participants with hostile emotions are the opposite.

First consider the effect of censorship as repression:

X X
(11) ucA0 = Aj aA0 ,Aj + Bj aA0 ,Bj − c

(12) ucA1 = uA1

and then as information manipulation:

X
(13) ubAi = A1 aAi ,A1 + (1 − b)A0 aAi ,A0 + Bj aA0 ,Bj

(14) ubBi = A1 aBi ,A1 + (1 − b)A0 aBi ,A0

Both setups reduce the fitness of moderate participants. In time t, the fitness of ucA0 in the

2
see working paper by Hongding Zhu and Jose E. Trinidad, available on
https://docs.google.com/document/d/1stlpfjHY8wUfXEM0Q anE9LcMmB6AEQJRIzd4qYphbY/edit?usp=sharing
24 Hongding Zhu

repression extension is smaller than in the baseline model and accordingly the replicator

dynamics of A1 is larger than in the baseline model. So censorship as repression is in favor

of stigmatization. Intuitively, the fitness of moderate strategy is reduced by censorship,

and more participants of the social movements choose to be hostile than in the baseline

model, which in turn results in more stigmatization (more haters) of the movement.

Proposition 4 (The Effect of Censorship) For any b, c > 0, there is a unique stigma-

tization threshold Cc∗ that goes through the interior Nash equilibrium {A∗1c , B1c
∗
}. Cc∗ is

below C ∗ . Censorship enlarge the strategy space that induces xS∗ .

The proof of Proposition 4 is trivial because it uses exactly the same logic and steps as

Proposition 3. When participants of a social movement suffer extra cost for articulating

their ideas and agenda in peaceful language, they gradually turn to trolling and emotional

hostility. This story is the case in the Chinese cyberfeminist movement. Feng (2017);

Shen (2021) shows that censorship creates a narrower space for feminists in China to

articulate themselves while stories of “how censorship drives me to become an extreme

feminist” are common on the Douban forum and Twitter.

7 Conclusion

I set up a two-population evolutionary game to model the opinion evolution and dy-

namics of stigmatization of a social movement in cyberspace. The interaction is among

participants of the movement and between the participants and the general public.

Based on the critical assumption of “hostility as retaliation”, hostile participants and

haters in the general public tend to reproduce each other. If the initial proportion of

hostile participants of the movement exceeds the stigmatization threshold, the evolution-
MA Thesis 25

ary outcome is the stigmatization equilibrium, where all participants are hostile to the

public, and all outsiders are haters of the movement. The stigmatization threshold is

pinned down by the proportion of haters in the general public so that the higher the

initial stigmatization, the lower the threshold to induce the stigmatization equilibrium.

I also discuss the typical situation in which stigmatization of a social movement is a self-

fulfilling prophecy. In this situation, the movement itself is moderate at the beginning;

however, severe enough social stigmatization (haters in the general public) will drive the

movement to become more and more hostile, and finally, the conflict between the public

and the movement is inevitable.

The effect of the Echo Chamber and censorship on the stigmatization of social move-

ments are discussed in the extensions. The analysis shows that Echo Chamber only helps

stigmatization and polarization when hostile participants in a social movement have a

larger tendency toward homophily. Moreover, censorship in autocracy usually targets

collective action potential and alternative ideologies other than the official one instead

of trolling and language violence, which means that moderate participants, instead of

hostile participants, are more vulnerable to censorship. Consequently, censorship exac-

erbates the stigmatization of social movements in autocracies, no matter whether it is

interpreted as repression or information manipulation.

These results contribute to the theoretical discussion from several aspects. First, it

analyzes the process of stigmatization of social movements through a dynamic approach,

which complements the static analysis in the literature. The mechanism of stigmatization

as a self-fulfilling prophecy helps explain how an initially moderate and peaceful social

movement can descend into bitter conflict with its opponents in society. Second, in addi-

tion to complex ideological and structural factors, stigmatization of social movements can
26 Hongding Zhu

also arise from micro-emotional conflicts in online interactions, which is a crucial feature

in the social media era and is supported by empirical observations (Koo, 2020; Ka, 2021).

Third, different from previous studies (Harel et al., 2020) that claim outgroup pressure

intensifies opinion polarization inside a social group, my model suggests that in a micro

online community, stigmatization from the public induces only intragroup polarization

and creates intergroup convergence—members of both groups choose internally consistent

but externally hostile strategies. Hence, we should reconsider the does the previous claim

hold only at a macro level. And finally, the logic of evolutionary games suggests that in

studying the opinion dynamics, population replacement and ideological inheritance from

old users to new users is a complementary mechanism other than persuasion.

Readers should also be aware of the limitations of this study. First, the assumption

of random matching in the model restricts its analytical level to a micro online commu-

nity where the discourse of a social movement circulates rather than the global dynamics

on the internet. Second, the four strategies in the game are ideal types, and this ty-

pology captures only limited aspects of individual opinions in social movements. Third,

the theoretical implications of the model require further empirical evidence to prove its

credibility.
MA Thesis 27

References
Christopher H Achen and Larry M Bartels. Blind retrospection: Why shark attacks
are bad for democracy. Center for the Study of Democratic Institutions, Vanderbilt
University. Working Paper, 2012. 10

Maja Adena, Ruben Enikolopov, Maria Petrova, Veronica Santarosa, and Ekaterina Zhu-
ravskaya. Radio and the rise of the nazis in prewar germany. The Quarterly Journal
of Economics, 130(4):1885–1939, 2015. 22

Stefan Aust. Baader-Meinhof: The inside story of the RAF. Oxford University Press,
USA, 2009. 4

Sven Banisch and Eckehard Olbrich. Opinion polarization by learning from social feed-
back. The Journal of Mathematical Sociology, 43(2):76–103, 2019. Publisher: Taylor
& Francis. 7

Daniel Bar-Tal. Sociopsychological foundations of intractable conflicts. American Behav-

ioral Scientist, 50(11):1430–1453, 2007. 1, 4

Pablo Barberá, John T Jost, Jonathan Nagler, Joshua A Tucker, and Richard Bonneau.
Tweeting from left to right: Is online political communication more than an echo
chamber? Psychological science, 26(10):1531–1542, 2015. Publisher: Sage Publications
Sage CA: Los Angeles, CA. 6, 19

Bárbara Inés Bavoleo and Desireé Chaure. Korean cyberfeminism: emergence, charac-
teristics and results. Anuario en Relaciones Internacionales del IRI, 2020, 2020. 3

Maria Letizia Bertotti and Marcello Delitala. On a discrete generalized kinetic approach
for modelling persuader’s influence in opinion formation processes. Mathematical and
computer modelling, 48(7-8):1107–1121, 2008. Publisher: Elsevier. 7

Alexander Bor and Michael Bang Petersen. The psychology of online political hostility:
A comprehensive, cross-national test of the mismatch hypothesis. American political
science review, 116(1):1–18, 2022. 6

Andrei Boutyline and Robb Willer. The social structure of political echo chambers:
28 Hongding Zhu

Variation in ideological homophily in online networks. Political psychology, 38(3):551–

569, 2017. Publisher: Wiley Online Library. 19

Tinggui Chen, Qianqian Li, Peihua Fu, Jianjun Yang, Chonghuan Xu, Guodong Cong,
and Gongfa Li. Public opinion polarization by individual revenue from the social
preference theory. International journal of environmental research and public health,
17(3):946, 2020. Publisher: Multidisciplinary Digital Publishing Institute. 7

Justin Cheng, Cristian Danescu-Niculescu-Mizil, and Jure Leskovec. How community

feedback shapes user behavior. In Eighth International AAAI Conference on Weblogs
and Social Media, 2014. 6, 11

James S Coleman. Foundations of social theory. Harvard university press, 1994. 7

Alison Dahl Crossley. Facebook feminism: Social media, blogs, and new technologies of
contemporary us feminism. Mobilization: An International Quarterly, 20(2):253–268,
2015. 6

Simone De Beauvoir. The second sex. Knopf, 2010. 4

Michela Del Vicario, Gianna Vivaldo, Alessandro Bessi, Fabiana Zollo, Antonio Scala,
Guido Caldarelli, and Walter Quattrociocchi. Echo chambers: Emotional contagion
and group polarization on facebook. Scientific reports, 6(1):1–12, 2016. 6, 13

Alessandro Di Mare and Vito Latora. Opinion formation models based on game theory.
International Journal of Modern Physics C, 18(09):1377–1395, 2007. Publisher: World
Scientific. 7

Robin DiAngelo. White fragility: Why it’s so hard for white people to talk about racism.
Beacon Press, 2018. 3

Pierce Alexander Dignam and Deana A Rohlinger. Misogynistic men online: How the
red pill helped elect trump. Signs: Journal of Women in Culture and Society, 44(3):
589–612, 2019. 3, 12

Feng. Hard times for feminists in China. SupChina, 2017. URL https://supchina.
com/2017/03/08/hard-times-feminists-china/. 1, 24
MA Thesis 29

Andrew T Fiore and Judith S Donath. Homophily in online dating: when do you like
someone like yourself? In CHI’05 extended abstracts on Human factors in computing
systems, pages 1371–1374, 2005. 19

Daniel Friedman. On economic applications of evolutionary game theory. Journal of

evolutionary economics, 8(1):15–43, 1998. 10

Scott Gehlbach, Zhaotian Luo, Anton Shirikov, and Dmitriy Vorobyev. A Model of
Censorship, Propaganda, and Repression. 2021. 8

Noriko Hara and Bi-Yun Huang. Online social movements. 2011. 1

Tal Orian Harel, Ifat Maoz, and Eran Halperin. A conflict within a conflict: intragroup
ideological polarization and intergroup intractable conflict. Current Opinion in Behav-
ioral Sciences, 34:52–57, 2020. 1, 4, 6, 8, 26

bell hooks. Feminist theory: From margin to center. Pluto Press, 2000. 4

Fei Ka. douban nvquan, jiafeng zhong xingcheng de ”zhongguo tese nvquan”.
Initium Media, March 2021. URL https://theinitium.com/article/
20210308-opinion-china-douban-feminism-awakens-or-failure/?utm_source=
twitter&utm_medium=twitter&utm_campaign=twpost. 3, 5, 26

R Kelly Garrett. Protest in an information society: A review of literature on social

movements and new icts. Information, communication & society, 9(02):202–224, 2006.
6

Bo-Myung Kim. Late modern misogyny and feminist politics: The case of ilbe, megalia,
and womad. Journal of Korean Women’s Studies, 34(1):1–31, 2018. 3, 5

Hannah June Kim and Chungjae Lee. The 2022 south korean presidential election and
the gender divide among the youth. Pacific Affairs, 95(2):285–308, 2022. 3

Gary King, Jennifer Pan, and Margaret E Roberts. How censorship in china allows gov-
ernment criticism but silences collective expression. American political science Review,
107(2):326–343, 2013. 8, 22
30 Hongding Zhu

JiHae Koo. South korean cyberfeminism and trolling: The limitation of online feminist
community womad as counterpublic. Feminist Media Studies, 20(6):831–846, 2020. 5,
12, 26

Chengting Mao. Feminist activism via social media in china. Asian Journal of Women’s
Studies, 26(2):245–258, 2020. 3

Nolan McCarty and Adam Meirowitz. Political game theory: an introduction. Cambridge
University Press, 2007. 10

Miller McPherson, Lynn Smith-Lovin, and James M Cook. Birds of a feather: Homophily
in social networks. Annual review of sociology, 27(1):415–444, 2001. Publisher: Annual
Reviews 4139 El Camino Way, PO Box 10139, Palo Alto, CA 94303-0139, USA. 6, 19

Lucy Nicholas and Christine Agius. # notallmen,# menenism, manospheres and unsafe
spaces: Overt and subtle masculinism in anti-“pc” discourse. In The persistence of
global masculinism, pages 31–59. Springer, 2018. 3

Martin J Osborne et al. An introduction to game theory, volume 3. Oxford university

press New York, 2004. 10

Henry M. Pachter. The idea of progress in marxism. Social Research, 41(1):136–161,

1974. ISSN 0037783X. 3

Werner Raub, Vincent Buskens, and Marcel ALM Van Assen. Micro-macro links and
microfoundations in sociology. The Journal of Mathematical Sociology, 35(1-3):1–25,
2011. Publisher: Taylor & Francis. 7

Sungmin Rho and Michael Tomz. Why don’t trade preferences reflect economic self-
interest? International Organization, 71(S1):S85–S108, 2017. 10

Ann-Kathrin Rothermel. Global–local dynamics in anti-feminist discourses: An analysis

of indian, russian and us online communities. International Affairs, 96(5):1367–1385,
2020. 3

R Keith Sawyer and Robert Keith Sawyer Sawyer. Social emergence: Societies as complex
systems. Cambridge University Press, 2005. 7
MA Thesis 31

Du Shen. kongjian jiya yu jixing shengzhang: zhongguo nvquan de neiy-

ouwaikun. Initium Media, April 2021. URL https://theinitium.com/article/
20210415-opinion-china-feminisim-statism/. 24

Vivienne Shue. The reach of the state: sketches of the Chinese body politic. Stanford
University Press, 1990. 23

Valerie Solanas and Ronell Avital. SCUM manifesto. Verso, 2004. 4

Cass R Sunstein. The law of group polarization. University of Chicago Law School, John
M. Olin Law & Economics Working Paper, (91), 1999. 6

Daniel López Sánchez, Jorge Revuelta, Fernando De la Prieta, Ana B Gil-González,

and Cach Dang. Twitter user clustering based on their preferences and the Louvain
algorithm. In International Conference on Practical Applications of Agents and Multi-
Agent Systems, pages 349–356. Springer, 2016. 19

Jia Tan. Digital masquerading: Feminist media activism in China. Crime, Media, Culture,
13(2):171–186, 2017. 1

Jeroen Van Laer and Peter Van Aelst. Internet and social movement action repertoires:
Opportunities and limitations. Information, Communication & Society, 13(8):1146–
1171, 2010. 6

Shawn P Van Valkenburgh. Digesting the red pill: Masculinity and neoliberalism in the
manosphere. Men and Masculinities, 24(1):84–103, 2021. 3, 12

Melissa A Wall. Social movements and email: expressions of online identity in the glob-
alization protests. New media & society, 9(2):258–277, 2007. 1

Jörgen W Weibull. Evolutionary game theory. MIT press, 1997. 10, 15

Yuke Yang. Jijin de nvquan biaoqian: Nvquan zhuyi ruhe zai meijie pingtai bei wum-
inghua. Journalism and Communication, 2014:94, 2014. 3

Jiangjun Yuan, Jiawen Shi, Jie Wang, and Weinan Liu. Modelling network public opinion
polarization based on SIR model considering dynamic network structure. Alexandria
Engineering Journal, 2021. Publisher: Elsevier. 7
32 Hongding Zhu

Appendix

Proof of Proposition 2

Because the first-order ODEs f (t, x) are polynomials about x, they satisfy the Lipschitz

condition and have one unique solution {A1 (t), B1 (t)} according to Picard-Lindelöf The-

orem. Which means any initial value problem has a unique evolutionary track—a unique

curve that goes through it. Therefore, there exists a unique curve C ∗ in Figure 1 that

goes through the interior Nash equilibrium (A∗1 , B1∗ ) and any xt0 not on C ∗ never cross

C ∗ during the replicator dynamics. Hence, any xt0 below C ∗ must arrive at Area 2 at

some t and any xt0 above C ∗ must arrive at Area 1 at some t. And according to Lemma

1, the evolutionary outcome after arriving at Area 1 or 2 is clear. ■

Proof of Proposition 3

Denote the two curves which is the solution of uhB0 = uhB1 and uhA0 = uhA1 as C1h and

C2h , it solves:

m
(15) C1h : A1 =
e + m − eγ

(16)
(1 − h)m p[e + (1 − h)m]A1
C2h : B1 = −
(1 − p)[e + (1 − h)m − (1 − h)eγ] (1 − p)[e + (1 − h)m − (1 − h)eγ]

in which C1h remains the same as C1 while C2h is strictly below C2 when h1 > h0 .

Therefore, the interior Nash equilibrium (A∗1h , B1h

∗
) is strictly below (A∗1 , B1∗ ) of the base-

line model. Accordingly, there is a unique stigmatization threshold Ch∗ that goes through

(A∗1 , B1∗ ). Suppose C ∗ h crosses C ∗ at some point, then at the intersection point (Ã1 , B̃1 ),
MA Thesis 33

the replicator dynamics (growth rate) of B1 is the same in the baseline and the Echo

Chamber model, while the replicator dynamics of A1 must be smaller than in the base-

line model in the two situation in Figure 3. However, in the setup we know that on

any point (A1 , B1 ), the replicator dynamic of A1 is larger than in the baseline model

since uhA1 > uA1 on any point (A1 , B1 ). There are contradictions. So the intersection

situations in Figure 3 does not exist, and accordingly Ch∗ is always below C ∗ . ■

Figure 3: Suppose the Intersection Exists