Causal Modeling of Affective Polarization
Abstract
Causal modeling provides us with the powerful counterfactual reasoning and interventional mechanism to generate predictions and reason under various what-if scenarios. It helps planners and decision-makers to gain insights into causal factors behind polarization, substance abuse, mental health, and the will-to-fight of military units. The gold standard approach in establishing cause-effect relationships is a randomized controlled trial. Unfortunately, this is not always practical and/or ethical, leaving us with discovering the causal structure from observational data. The fundamental challenge is that causal discovery using observation data remains a nontrivial task due to unobserved confounding factors, finite sampling, and changes in the data distribution. To mitigate these limitations in practice, we propose an interactive causal modeling framework that provides domain experts with a high-level view of the problem to iteratively learn, test, and correct relationships in the causal graphs identified from observational data. The focus is to equip state-of-the-art causal deep learning models with features that allow domain experts to induce problem constraints and to understand the limitations of the model and receive suggestions for remedial interventions. The anticipated results of this project include a general implementation of a causal deep learning model that constrains the causal graph to be invariant across various datasets and accommodate expertsÕ inputs in the form of known direct and indirect causal and non-causal relations. The framework will be augmented with a predictive model to identify variables/concepts that share unobserved common causes to inform the expert of the need of acquiring new data. The proposed expert-in-the-loop causal modeling framework will be applied in this project to understand affective polarization. There is a growing concern in recent years that the affective polarization in the United States hampers the ability of the country to effectively project power and forge domestic and international agreements. The events we propose to examine in this project are critical events where online polarization has bridged the gap into real-world consequences. By examining the case studies highlighted in this project, we take the first critical steps in understanding the relationship between polarization and its most serious repercussion, radicalization. The current study offers the potential to intervene early in the polarization process and suppress radical behavior.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 24, 2022
- Source ID
- W911NF2210035
Entities
People
- Gabriel Terejanu
Organizations
- Army Contracting Command
- United States Army
- University of North Carolina at Chapel Hill