Data analysis and modeling for complex social systems:
I am interested in problems that involve individuals coming together to create rich group dynamics, whether the individuals are cells in a developing tissue, cars in a traffic jam, or voters in a political election. Alongside my work on complex biological systems, I have become more involved in complex social systems over the last several years. The majority of my research in this area has focused on forecasting elections. The projects below involve data-driven, agent-based, and network modeling; data analysis; and parameter estimation.
Click each image for more details.
Forecasting elections using a dynamical-systems perspective
Election forecasting is characterized by uncertainty, incomplete information, and subjective choices (e.g., related to defining likely voters). Interestingly, polling errors tend to be correlated in states that have similiar demographics, adding further complexity. With the goal of shedding light on the forecasting process and raising questions from a dynamical-systems perspective, we developed data-driven models to forecast U.S. elections. Our compartmental approach tracks the number of Democratic, Republican, and undecided voters in each state, and we fit our parameters using polling data. Our forecasts for 2004-2018 are about as accurate as those of popular analysts. We applied the model to forecast the 2020 U.S. elections, and all of our code is available here.
Collaborators:
Daniel F Linder (Augusta University), Mason A Porter (University of California, Los Angeles), Grzegorz A Rempala (Ohio State University), Samuel Chian* (Northwestern University), William L He* (Northwestern University), Christopher M Lee* (Northwestern University), Emma Mansell* (Northwestern University)
Publications:
- A Volkening, DF Linder, MA Porter, GA Rempala.
Forecasting elections using compartmental models of infection.
SIAM Review, 62(4):837-865, 2020.
- S Chian, WL He, CM Lee, A Volkening.
2020 U.S. Election Forecasts with a Compartmental Model, 2020.
*Asterisk denotes undergraduate students mentored
Data analysis and modeling of social-movement dynamics on Twitter
Through data analysis and modeling, I am working to better understand the structure of the gun-violence conversation on social media. In particular, for different ideologies, is the conversation self-sustaining, or does it depend on external coverage of news events by the media? To address these questions, we are developing a model for Twitter activity and investigating the gun-violence conversation among 1,000 sampled accounts. This project also involves text analysis to identify the dates of mass shootings through abrupt changes in the words used in tweets.
Collaborators:
Heather Z. Brooks (Harvey Mudd College), Sarah Iams (Harvard University), Joseph Tien (Ohio State University)
Social-force modeling of student movement in large lecture halls
Pedestrian crowds exhibit a range of collective behaviors, including lane formation in corridors, stop-and-go waves at high density, and bi-directional movement at doorways. In this project, we focus specifically on pedestrian movement in large lecture hall settings. Motivated by new 600-person classrooms at U.C. Davis, we are extending social-force agent-based models to explore how lecture-hall size impacts class turnover times. Is it possible for students to travel from their previous classes, enter a lecture hall, and find seats during the time between class?
Collaborators:
Joseph Benson (Macalester College), Mariya Bessonov (NYC College of Technology), Korana Burke (University of California, Davis), Simone Cassani (University at Buffalo), Maria-Veronica Ciocanel (Duke University), Daniel Cooney (University of Pennsylvania)
Parameter identification and investigating state–state relationships across election years
Because there are many parameters in my models of election dynamics, I am interested in determining how the methods that we used for fitting parameters impact their values and our forecasts. As part of a summer undergraduate research project, Brian Hsu, Niall Mangan, and I have started to apply SINDy methods for data-driven model selection to our election work. We have also been analyzing the strength of state–state relationships across election years using statistical techniques.
Collaborators:
Brian Hsu* (Northwestern University), Niall M Mangan (Northwestern University)
*Asterisk denotes undergraduate students mentored