class: center, middle background-color: white .title[Devils in the Details:] .subtitle[Learning and Diffusion in Networks of Violent Groups] .author[Laila A. Wahedi, PhD] .date[July 12, 2018] .institution[McCourt School of Public Policy, Georgetown University]
.center[] .footnote[Created with [{Remark.js}](http://remarkjs.com/) using [{Markdown}](https://daringfireball.net/projects/markdown/) + [{MathJax}](https://www.mathjax.org/) + [{Liminal}](http://www.jmlilly.net/liminal.zip)] --- class: center background-color: .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu] # What are violent groups going to do next? -- .left-column[
  What they
want
to do      
] -- .right-column[
  What they
can
do            
] --- class: center background-color:#AF002B .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
What they
want
to do:
Threat from Security Forces
Security Forces Learn
Adapt to Survive
--- class: center background-color:#01BAEF .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
What they
can
do:
Mistakes Are Dangerous
Sharing is Risky
Learning Is Hard
--- class: center, background-color: #F2DC5D .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Groups need to adapt...
But learning is hard...
How do violent groups learn their craft?
-- background-color: #9BC53D
Groups mitigate risk of adoption by learning from
trusted parters
--- class: center, background-color: #F2DC5D .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Implications:
If groups learn from their partners, expect:
The probability of a group
adopting
a given tactic should
increase
if a group has a
trusted partner using
the tactic.
--- class: center, background-image: url(../figures/seg_arch1.png) .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Building Blocks: Data
Big Allied And Dangerous I
Cooperative Partnerships
Rivalries
Group-level covariates
Time-static
--- class: center, background-image: url(../figures/seg_arch2.png) .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Building Blocks: Data
Terrorism Knowledge Base
Attack Event Data
Tactical Information
Event-level covariates
--- class: center, background-image: url(../figures/seg_arch3.png) .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Building Blocks: Data
All Together
Whether a partner has
used a tactic in
the past three years
--- class: center, background-image: url(../figures/seg_arch4.png) .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Building Blocks: Unpooled
Probability of adopting
a given tactic increases
when partners use the tactic in:
16/17 tactics
11/17 tactics significant
Small n, multiple tests
--- class: center, background-image: url(../figures/seg_arch5.png)
Empirical Building Blocks: Pooled
.3 increase in P(adoption) when partners use it
Multiple dependent observations
--- class: center, background-image: url(../figures/seg_arch6.png) .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Keystone: Hierarchical TNAM
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Results
--- class: center, background-color: #F2DC5D .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Implications:
If groups learn from their partners
rather than observation,
expect:
The probability of a group
adopting
a given tactic should
not
increase when rivals use that tactic
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Results
--- class: center, background-color: #F2DC5D .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Empirical Implications:
If groups learn from their partners
rather than observation,
expect:
The probability of a group
adopting
a given tactic should
increase
when embedded in a network structure that facilitates trust and coordination
--- class: center background-color: .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Decentralized:
Redundant Paths:
Autonomy
Robust to Random Removal
Inefficient
Free Riding
Vulnerable to Unraveling
--- class: center background-color: .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Centralized:
Information Shortcuts
Efficient
Robust to Random Removal
Costly for Center
Periphery Loses Freedom
Vulnerable to Targeted Removal
--- class: center background-color: .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Centralized:
Information Shortcuts
Efficient
Robust to Random Removal
Specialization
Periphery Loses Freedom
Vulnerable to Targeted Removal
--- class: center background-color: .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Centralized:
Information Shortcuts
Efficient
Robust to Random Removal
Specialization
Periphery Trades For Efficiency
Vulnerable to Targeted Removal
--- class: center background-color: .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Centralized:
Information Shortcuts
Efficient
Robust to Random Removal
Specialization
Periphery Trades For Efficiency
Evolves Around Strong Core
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Results
--- class: center background-color:#FAFF75 # Beyond Tactics
Al Qaeda and IS Networks
## Endogenous Process of network effects and formation .footnote[ Follow along at Wahedi.US, under Current Presentation Laila A. Wahedi, PhD -- @lwahedi -- law98@georgetown.edu] --- class: center background-color:#FAFF75 .footnote[Follow along at Wahedi.US, under Current Presentation Laila A. Wahedi, PhD -- @lwahedi -- law98@georgetown.edu]
When do local groups join global movements?
Zero inflated negative binomial Hierarchical TNAM
## Groups in centralized local networks kill more people, conduct more attacks, and survive longer --- class: center background-color:#FAFF75 .footnote[Follow along at Wahedi.US, under Current Presentation Laila A. Wahedi, PhD -- @lwahedi -- law98@georgetown.edu]
When do local groups join global movements?
Global Actors
Local Actors
## When they have no other choice. --- class: center background-color:#FAFF75 .footnote[Follow along at Wahedi.US, under Current Presentation Laila A. Wahedi, PhD -- @lwahedi -- law98@georgetown.edu]
How does competition impact these global networks?
Global Actors
Local Actors
--- class: center background-color:#FAFF75 .footnote[Follow along at Wahedi.US, under Current Presentation Laila A. Wahedi, PhD -- @lwahedi -- law98@georgetown.edu]
How do power and ideology interact to shape network evolution?
Cooperative Network
Rivalry Network
## Constructing new network data from Syrian Civil War --- class: center background-color:#FAFF75 .footnote[Follow along at Wahedi.US, under Current Presentation Laila A. Wahedi, PhD -- @lwahedi -- law98@georgetown.edu]
How do you even measure soft power
Al Qaeda Influence
### By tracking diffusion of topics from official statements to discussions on social media --- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Backup
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Centralization
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Eigenvector Centrality
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Lethality
--- class: center, .footnote[ Laila A. Wahedi -- Follow along at Wahedi.US, under Current Presentation -- law98@georgetown.edu]
Structure Interaction Results
Groups in centralized local networks kill more people, and conduct more attacks
* Additive effect, no interaction ---