Polarization and Protests: Understanding Complex Social ...Complex Social and Political Processes...
Transcript of Polarization and Protests: Understanding Complex Social ...Complex Social and Political Processes...
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Polarization and Protests: UnderstandingComplex Social and Political Processes Using
Spatial Data and Agent-Based ModelingSimulations
Lefteris AnastasopoulosPhD, UC Berkeley (Political Science)
MA, Harvard (Statistics)
Democracy Fellow, Ash Center for Democratic Governance and Innovation,Harvard Kennedy School of Government
April 23, 2015
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
What are agent-based models?
1 Agents - Entities with goals and preferences.2 Utility - Preferences usually expressed in the form of a
utility function.Eg. Uprotest(G,C,P) = G + P
C3 Interactions in space and time.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
What are agent-based models?
1 Agents - Entities with goals and preferences.2 Utility - Preferences usually expressed in the form of a
utility function.Eg. Uprotest(G,C,P) = G + P
C3 Interactions in space and time.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
What are agent-based models?
1 Agents - Entities with goals and preferences.2 Utility - Preferences usually expressed in the form of a
utility function.Eg. Uprotest(G,C,P) = G + P
C3 Interactions in space and time.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
What are agent-based models?
1 Agents - Entities with goals and preferences.2 Utility - Preferences usually expressed in the form of a
utility function.Eg. Uprotest(G,C,P) = G + P
C3 Interactions in space and time.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Why agent-based models?
Emergence - Complex micro-level interactions can result ifmacro-level patterns.eg) Residential Segregation (Schelling)Reality is DynamicReal world social processes involve dynamic interactionsbetween individuals.Because we can!Parallelization of code and multi-core computing systemsallow large scale realistic simulations.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Why agent-based models?
Emergence - Complex micro-level interactions can result ifmacro-level patterns.eg) Residential Segregation (Schelling)Reality is DynamicReal world social processes involve dynamic interactionsbetween individuals.Because we can!Parallelization of code and multi-core computing systemsallow large scale realistic simulations.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Why agent-based models?
Emergence - Complex micro-level interactions can result ifmacro-level patterns.eg) Residential Segregation (Schelling)Reality is DynamicReal world social processes involve dynamic interactionsbetween individuals.Because we can!Parallelization of code and multi-core computing systemsallow large scale realistic simulations.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Why agent-based models?
Emergence - Complex micro-level interactions can result ifmacro-level patterns.eg) Residential Segregation (Schelling)Reality is DynamicReal world social processes involve dynamic interactionsbetween individuals.Because we can!Parallelization of code and multi-core computing systemsallow large scale realistic simulations.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Why agent-based models?
Emergence - Complex micro-level interactions can result ifmacro-level patterns.eg) Residential Segregation (Schelling)Reality is DynamicReal world social processes involve dynamic interactionsbetween individuals.Because we can!Parallelization of code and multi-core computing systemsallow large scale realistic simulations.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Why agent-based models?
Emergence - Complex micro-level interactions can result ifmacro-level patterns.eg) Residential Segregation (Schelling)Reality is DynamicReal world social processes involve dynamic interactionsbetween individuals.Because we can!Parallelization of code and multi-core computing systemsallow large scale realistic simulations.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G = H(1− L)P = 1− exp[−k(C/A)]N = RP
Simple ABM of civil violence w profound implications.What factors determine rebellion against a centralauthority?G = Grievance,H = Hardship,L = Legitimacy .P = Prob of arrest. C/A = Cop to protester ratio.N = Product of risk aversion R and arrest prob. P.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G = H(1− L)P = 1− exp[−k(C/A)]N = RP
Simple ABM of civil violence w profound implications.What factors determine rebellion against a centralauthority?G = Grievance,H = Hardship,L = Legitimacy .P = Prob of arrest. C/A = Cop to protester ratio.N = Product of risk aversion R and arrest prob. P.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G = H(1− L)P = 1− exp[−k(C/A)]N = RP
Simple ABM of civil violence w profound implications.What factors determine rebellion against a centralauthority?G = Grievance,H = Hardship,L = Legitimacy .P = Prob of arrest. C/A = Cop to protester ratio.N = Product of risk aversion R and arrest prob. P.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G = H(1− L)P = 1− exp[−k(C/A)]N = RP
Simple ABM of civil violence w profound implications.What factors determine rebellion against a centralauthority?G = Grievance,H = Hardship,L = Legitimacy .P = Prob of arrest. C/A = Cop to protester ratio.N = Product of risk aversion R and arrest prob. P.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G = H(1− L)P = 1− exp[−k(C/A)]N = RP
Simple ABM of civil violence w profound implications.What factors determine rebellion against a centralauthority?G = Grievance,H = Hardship,L = Legitimacy .P = Prob of arrest. C/A = Cop to protester ratio.N = Product of risk aversion R and arrest prob. P.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G − N > 0⇒ AOtherwise⇒ Q
States of the world: A = Active, Q = Quiet .G − N - Strength of grievance vs. subjective likelihood ofarrest.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
G − N > 0⇒ AOtherwise⇒ Q
States of the world: A = Active, Q = Quiet .G − N - Strength of grievance vs. subjective likelihood ofarrest.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
Predict outbreaks of ethnic cleansing.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Example: Modeling Civil Violence (Epstein 2002)
Effect of safe haven establishment during ethnic cleansingoutbreaks.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Enter Spatial Data...
ABMs give us general predictions in space and time.Demographic and spatial data at very fine levels ofgeography.Feed real spatial and demographic data into ABM togenerate predictions.Best of both worlds!
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Enter Spatial Data...
ABMs give us general predictions in space and time.Demographic and spatial data at very fine levels ofgeography.Feed real spatial and demographic data into ABM togenerate predictions.Best of both worlds!
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Enter Spatial Data...
ABMs give us general predictions in space and time.Demographic and spatial data at very fine levels ofgeography.Feed real spatial and demographic data into ABM togenerate predictions.Best of both worlds!
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Enter Spatial Data...
ABMs give us general predictions in space and time.Demographic and spatial data at very fine levels ofgeography.Feed real spatial and demographic data into ABM togenerate predictions.Best of both worlds!
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Current Projects
SimPolSeg: An Agent-Based Simulation of PoliticalMigration Dynamics and Geographic Polarization.Modeling Violent Protests
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Current Projects
SimPolSeg: An Agent-Based Simulation of PoliticalMigration Dynamics and Geographic Polarization.Modeling Violent Protests
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Migration and Geographic Polarization
Geographic urban-suburban polarization rising since the 1950s.
Highway development + suburbanization speculated to be a majorcause (Nall 2014).
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Migration and Geographic Polarization
Geographic urban-suburban polarization rising since the 1950s.
Highway development + suburbanization speculated to be a majorcause (Nall 2014).
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Migration and Geographic Polarization
-.1
-.05
0.0
5.1
% C
hang
e in
Sub
urba
niza
tion
(One
Yea
r)
1970 1980 1990 2000 2010Year
Major highway development and suburbanization ceasedaround 1992...Yet geographic polarization increasing at an increasingrate.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Migration and Geographic Polarization
-.1
-.05
0.0
5.1
% C
hang
e in
Sub
urba
niza
tion
(One
Yea
r)
1970 1980 1990 2000 2010Year
Major highway development and suburbanization ceasedaround 1992...Yet geographic polarization increasing at an increasingrate.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Push Migration, Diversity and Demographic Change
Causal estimates suggest that white flight resulting from the SecondGreat Migration was largely responsible for post-warsuburbanization. (Boustan 2010)
White flight responsible for suburbanization and polarization?
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Push Migration, Diversity and Demographic Change
Causal estimates suggest that white flight resulting from the SecondGreat Migration was largely responsible for post-warsuburbanization. (Boustan 2010)
White flight responsible for suburbanization and polarization?
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political and Racial Segregation
Schelling showed that weak preferences for similarneighbors create segregated urban spaces.Complete Integration or Segregation of an area dependsupon “tolerance” of residents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political and Racial Segregation
Schelling showed that weak preferences for similarneighbors create segregated urban spaces.Complete Integration or Segregation of an area dependsupon “tolerance” of residents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Migration and the Migration-Polarization (MP)Theory
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Ideology and Schelling Tolerance: MCSUIData
DV 1: Would you feel comfortable in the (7%/20%/33%/53% minority)neighborhood?
DV 2: If uncomfortable, would you be willing to move out of the(7%/20%/33%/53% minority) neighborhood?
IV Political ideology.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Ideology and Schelling Tolerance: MCSUIData
DV 1: Would you feel comfortable in the (7%/20%/33%/53% minority)neighborhood?
DV 2: If uncomfortable, would you be willing to move out of the(7%/20%/33%/53% minority) neighborhood?
IV Political ideology.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Ideology and Schelling Tolerance: MCSUIData
DV 1: Would you feel comfortable in the (7%/20%/33%/53% minority)neighborhood?
DV 2: If uncomfortable, would you be willing to move out of the(7%/20%/33%/53% minority) neighborhood?
IV Political ideology.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Comfort In Minority Neighborhood: MCSUI
Pr(Comfortc |Ideology ,X ) =1
1 + exp − (αc + βc Ideology + XΓc + mj )(1)
White Respondent Ideology v. Probability of Discomfort, N = 2407
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Moving From Minority Neighborhood: MCSUI
Pr(Move|Ideology ,X ) =1
1 + exp − (αc + βc Ideology + XΓc + mj )(2)
White Respondent Ideology v. Prob. of Moving from Any NeighborhoodConditional on Discomfort, N = 1695
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
GSS 2002 Data: Background
Image presented to Respondents in the 2000 General Social Survey.
Respondents asked to indicate ethnic/racial background of each of the14 houses surrounding them.
Choices were: Asian, Black, Hispanic or White.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
GSS 2002 Data: Background
Image presented to Respondents in the 2000 General Social Survey.
Respondents asked to indicate ethnic/racial background of each of the14 houses surrounding them.
Choices were: Asian, Black, Hispanic or White.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
GSS 2002 Data: Background
Image presented to Respondents in the 2000 General Social Survey.
Respondents asked to indicate ethnic/racial background of each of the14 houses surrounding them.
Choices were: Asian, Black, Hispanic or White.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
GSS 2002 Data: Average Preferred % White byIdeology
Average % White (of 14 surrounding houses) by IdeologyAmong White Respondents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
GSS 2002 Data: Preferred % White by Ideology
%White = α + βIdeology + XΓ + cj + ε
DV: Neighborhood % WhiteVARIABLES Original Coefficients Standardized Coefficients
Ideology 0.036*** 0.192***(0.006)
Age 0.002** 0.136**(0.001)
Education -0.001 -0.014(0.002)
Income 0.000 0.018(0.001)
Sex -0.036 -0.056(0.021)
Marital -0.001 0.004(0.007)
Employment 0.003 0.020(0.005)
Children 0.014** 0.078**(0.007)
Observations 770R-squared 0.138
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Ideology and Tolerance
ηi =√
Ii − Di
Di =√∑K
k=1∑J
j=1 ρji(E [αjk ]− αji)2
0 < Di < Ii < 1
ηi = ToleranceDi= Social DistanceIi= Ideology (0 = Very Conservative, 1 = Very Liberal)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Ideology and Tolerance
ηi =√
Ii − Di
Di =√∑K
k=1∑J
j=1 ρji(E [αjk ]− αji)2
0 < Di < Ii < 1
ηi = ToleranceDi= Social DistanceIi= Ideology (0 = Very Conservative, 1 = Very Liberal)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Political Ideology and Tolerance
ηi =√
Ii − Di
Di =√∑K
k=1∑J
j=1 ρji(E [αjk ]− αji)2
0 < Di < Ii < 1
ηi = ToleranceDi= Social DistanceIi= Ideology (0 = Very Conservative, 1 = Very Liberal)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Utility of Residing in a Neighborhood
URin = ηimn −m2
n + f (pn, εn) = (√
Ii − Di)mn −m2n + f (pn, εn)
0 ≤ mn ≤ 10 < ηi < 10 < Di < Ii < 1
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Utility of Residing in a Neighborhood
URint = (
√Ii − Di)mnt −m2
nt + f (pnt , εnt)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Agent Moving Decisions: When to Move?
If mnt > m∗i and ∃n′ ∈ A s.t. mn′ t ≤ m∗
i :{If UR
nt < URn′ t− δn,n′ Move from n
Else Remain in n
δn,n′ = cost of moving from n to n′.
Euclidean distance between two neighborhood centroids.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Agent Moving Decisions: When to Move?
If mnt > m∗i and ∃n′ ∈ A s.t. mn′ t ≤ m∗
i :{If UR
nt < URn′ t− δn,n′ Move from n
Else Remain in n
δn,n′ = cost of moving from n to n′.
Euclidean distance between two neighborhood centroids.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Agent Moving Decisions: Where to Move?
maxn′
L(URnt ,U
Rn′ t) = (UR
nt − δn,n′ + URn′ t)
2
Choose n′
to maximize utility gain from moving.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 1.0
SimPolSeg(Neighborhood ,MinorityPop0,MinorityPop1,WhitePop,TotPop, Ideology ,Sims)
Initial SimPolSeg 1.0 software takes arguments above.Simulates political effects of change in minority population.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 1.0
SimPolSeg(Neighborhood ,MinorityPop0,MinorityPop1,WhitePop,TotPop, Ideology ,Sims)
Initial SimPolSeg 1.0 software takes arguments above.Simulates political effects of change in minority population.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Simulation
Variable Symbol Initial ValueNeighborhoods A N = 20
Mean Between Neighborhood Ideology It0 µIt0 = 0.5σI
t0 = 0.1It0 =rtnorm(n=20,mu =0.5,sd=0.1)
Minoritypop Bt0 µBt0 = 5σB
t0 = 5Bt0 = rtnorm(n=20,mu =5,sd=5)
Majoritypop Wt0 µWt0 = 100σW
t0 = 10Wt0 = rtnorm(n=20,mu =100,sd=10)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Simulation Results
Movers as Percent of Total Population,t = 0 to t = 151
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Simulation Results
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Simulation Results
Average neighborhood ideology, t = 0 to t = 151
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Simulation Results
Polarization and Segregation BetweenNeighborhoods,t = 0 to t = 151
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0
SimPolSeg2(TractID,MinorityPopT0,MinorityPopT1,WhitePopT0,TotPopT0,DemocratT0,RepublicanT 0,Sims,Lat ,Long)
SimPolSeg 2.0 under development.New Features :
1 Inputs real Census tract data and demographics.2 Parallelized code to efficiently handle interactions between
millions of agents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0
SimPolSeg2(TractID,MinorityPopT0,MinorityPopT1,WhitePopT0,TotPopT0,DemocratT0,RepublicanT 0,Sims,Lat ,Long)
SimPolSeg 2.0 under development.New Features :
1 Inputs real Census tract data and demographics.2 Parallelized code to efficiently handle interactions between
millions of agents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0
SimPolSeg2(TractID,MinorityPopT0,MinorityPopT1,WhitePopT0,TotPopT0,DemocratT0,RepublicanT 0,Sims,Lat ,Long)
SimPolSeg 2.0 under development.New Features :
1 Inputs real Census tract data and demographics.2 Parallelized code to efficiently handle interactions between
millions of agents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0
SimPolSeg2(TractID,MinorityPopT0,MinorityPopT1,WhitePopT0,TotPopT0,DemocratT0,RepublicanT 0,Sims,Lat ,Long)
SimPolSeg 2.0 under development.New Features :
1 Inputs real Census tract data and demographics.2 Parallelized code to efficiently handle interactions between
millions of agents.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0: Simulate Political Effects of KatrinaMigration
Over 100,000 African Americans migrated to Harris Countyshortly after Katrina.Profound political effects.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0: Simulate Political Effects of KatrinaMigration
Over 100,000 African Americans migrated to Harris Countyshortly after Katrina.Profound political effects.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0: Simulate Political Effects of KatrinaMigration
Increase in Democratic voting in Harris.Increase in polarization between Harris and surroundingcounties.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
SimPolSeg 2.0: Simulate Political Effects of KatrinaMigration
Increase in Democratic voting in Harris.Increase in polarization between Harris and surroundingcounties.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks
Goals:1 Discover common themes leading up to violent
protests using topic models of Twitter data.2 Simulate spatial diffusion, spread and containment of
protest activity using an agent-based model.3 Test model using geocoded Tweets and public
transportation data.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks
Goals:1 Discover common themes leading up to violent
protests using topic models of Twitter data.2 Simulate spatial diffusion, spread and containment of
protest activity using an agent-based model.3 Test model using geocoded Tweets and public
transportation data.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks
Goals:1 Discover common themes leading up to violent
protests using topic models of Twitter data.2 Simulate spatial diffusion, spread and containment of
protest activity using an agent-based model.3 Test model using geocoded Tweets and public
transportation data.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks
Goals:1 Discover common themes leading up to violent
protests using topic models of Twitter data.2 Simulate spatial diffusion, spread and containment of
protest activity using an agent-based model.3 Test model using geocoded Tweets and public
transportation data.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks and Spread:Common Themes
α θd zd ,n
βkψ
wd ,n
K
ND
Extract latent themes preceding violent protests (topics) using LatentDirichlet Allocation (LDA).
Corpus - Collection of Tweets one day before violent protests break out.
Documents - Tweets, one, three and five hours before protests breakout.
Common themes - Use topic chains to find topic similarity for eachprotest.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks and Spread:Common Themes
α θd zd ,n
βkψ
wd ,n
K
ND
Extract latent themes preceding violent protests (topics) using LatentDirichlet Allocation (LDA).
Corpus - Collection of Tweets one day before violent protests break out.
Documents - Tweets, one, three and five hours before protests breakout.
Common themes - Use topic chains to find topic similarity for eachprotest.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks and Spread:Common Themes
α θd zd ,n
βkψ
wd ,n
K
ND
Extract latent themes preceding violent protests (topics) using LatentDirichlet Allocation (LDA).
Corpus - Collection of Tweets one day before violent protests break out.
Documents - Tweets, one, three and five hours before protests breakout.
Common themes - Use topic chains to find topic similarity for eachprotest.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks and Spread:Common Themes
α θd zd ,n
βkψ
wd ,n
K
ND
Extract latent themes preceding violent protests (topics) using LatentDirichlet Allocation (LDA).
Corpus - Collection of Tweets one day before violent protests break out.
Documents - Tweets, one, three and five hours before protests breakout.
Common themes - Use topic chains to find topic similarity for eachprotest.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Model andSimulations
P(A) =f (G,N,C,Di,i ′ ,Di,s)
P(Q) =P(A′) = 1− P(A)
Model agent behavior building on Epstein (2002).P(A) = probability of protesting, P(Q) = probability of notprotesting.Traditional Components: G = Grievance, N = likelihoodof being arrested, C = law enforcement.Spatial Components: Di,i ′ = distance between agent andlikeminded agents. Di,s = distance between agent and siteof protest s.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Model andSimulations
P(A) =f (G,N,C,Di,i ′ ,Di,s)
P(Q) =P(A′) = 1− P(A)
Model agent behavior building on Epstein (2002).P(A) = probability of protesting, P(Q) = probability of notprotesting.Traditional Components: G = Grievance, N = likelihoodof being arrested, C = law enforcement.Spatial Components: Di,i ′ = distance between agent andlikeminded agents. Di,s = distance between agent and siteof protest s.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Model andSimulations
P(A) =f (G,N,C,Di,i ′ ,Di,s)
P(Q) =P(A′) = 1− P(A)
Model agent behavior building on Epstein (2002).P(A) = probability of protesting, P(Q) = probability of notprotesting.Traditional Components: G = Grievance, N = likelihoodof being arrested, C = law enforcement.Spatial Components: Di,i ′ = distance between agent andlikeminded agents. Di,s = distance between agent and siteof protest s.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Model andSimulations
P(A) =f (G,N,C,Di,i ′ ,Di,s)
P(Q) =P(A′) = 1− P(A)
Model agent behavior building on Epstein (2002).P(A) = probability of protesting, P(Q) = probability of notprotesting.Traditional Components: G = Grievance, N = likelihoodof being arrested, C = law enforcement.Spatial Components: Di,i ′ = distance between agent andlikeminded agents. Di,s = distance between agent and siteof protest s.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Model andSimulations
Model protest spread as a spatial diffusion process from publictransportation hubs or central meeting nodes.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Test ModelUsing Data
Ferguson and Arab Spring protests in Egypt as test cases.Ferguson acquittal announced around 9:22(ET)/8:22 (CT)/6:22 (PT).Use BART data andgeocoded Tweets to explore protest diffusion fromstations and test model.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Test ModelUsing Data
Ferguson and Arab Spring protests in Egypt as test cases.Ferguson acquittal announced around 9:22(ET)/8:22 (CT)/6:22 (PT).Use BART data andgeocoded Tweets to explore protest diffusion fromstations and test model.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Understanding Violent Protest Outbreaks: Test ModelUsing Data
Ferguson and Arab Spring protests in Egypt as test cases.Ferguson acquittal announced around 9:22(ET)/8:22 (CT)/6:22 (PT).Use BART data andgeocoded Tweets to explore protest diffusion fromstations and test model.
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Geocoded #Ferguson on November 24, 2014 - 6:30(CT)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Geocoded #Ferguson on November 24, 2014 - 7:30(CT)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Geocoded #Ferguson on November 24, 2014 - 8:30(CT)
1: Introduction 2: SimPolSeg and Geographic Polarization 3: Violent Protests
Questions and Comments Appreciated!