Complex systems

Complexity and political science

Politics is inherently relational in that individuals, groups, and states are competing for influence and power. In turn, within a political system the behaviors and actions of agents impacts the decisions of other agents within the system. To better understand these process, my methodological research focuses on network science, machine learning, and text as data. Specifically, I am working on developing new statistical models, software, and analytic approaches which will help address several of my theories on conflict. Below I discuss some completed projects and ongoing research in methodology.

Ego-network of Islamic State Recruits into Syria.

Ego-network of Islamic State Recruits into Syria.

Network science
I am interested in how social networks influence actor behavior; specifically, how the social structures of states and individuals impacts conflict and political violence. I recently finished a project studying the effect of exogenous shocks on team performance. In the project, we randomly removed an experiment participant from a real life team during a game. This allowed us to precisely measure the individual contributions of the player, how the structure of a team impacted its performance, and how teams adapt to exogenous shocks.

I am currently working on adapting the exponential random graph model so that it models network formation as a two-stage process. This allows us to see how groups within a network effect the actors within the network. The R package is currently under development and can be found on my github page.

Image classification for bomb craters in Cambodia.

Image classification for bomb craters in Cambodia.

Machine learning
Unexploded ordnance in Southeast Asia negatively impacts agricultural output and economic growth. Current UXO removal practices rely on in-person enumeration, which can be costly and dangerous. I worked on a research team that identified bomb craters from a satellite image of Cambodia in the Prey Veng province. In a paper currently under review, we used two stage classification to identify candidate patched and then predict UXO from the patches. The image below shows bomb craters and along with several confusing classes in the satellite image. I am currently adapting the two-stage approach on a project that forecasts terrorist attacks. We are using the Gallup survey data to see if we can more accurately predict regions which are likely to experience terrorism.

president_words_time_series_v2.png

Text as data
I am currently working on a project which explores how white nationalism is discussed on the major news networks. The data includes all utterances for hosts, speakers and contributors for the major news networks between 2000 and 2017. This allows us to examine how issues or race, gender, and politics are discussed on these programs. The figure below displays the words spoken by each president on each network over time.