Who are the most influential senators in Congress? Is there a small coalition of senators who are influential enough to prevent filibusters? In a different setting of microfinance markets, can we predict the effects of interventions to help policy makers? In order to pursue such diverse questions, we propose causal strategic inference, a game-theoretic counterpart of causal probabilistic inference. Using this general framework, we study two different sets of problems, broadly on social networks and networked microfinance economies.;In the first study, we introduce a new approach to the study of influence that captures the strategic aspects of the complex interactions in a network. We design influence games, a new class of graphical games, as a model of the behavior of a large but finite networked population. Influence games can deal with positive as well as negative influence without having to consider network dynamics. We characterize the computational complexity of various problems on influence games, propose effective solutions to the hard problems, and design approximation algorithms, with provable guarantees, for identifying the most influential individuals in a network. Our empirical study is based on the real-world data obtained from congressional voting records and Supreme Court rulings.;Our second study is on microfinance economies. It is motivated by the challenge of formulating economic policies without the privilege of conducting trial-and-error experiments. First, we model a microfinance market as a two-sided economy. We then learn the parameters of the model from real-world data and design algorithms for various computational problems. We show the uniqueness of equilibrium interest rates for a special case and give a constructive proof of equilibrium existence in the general case. Using data from Bangladesh and Bolivia, we show that our model captures various real-world phenomena and can be used to assist policy makers in the microfinance sector.;Despite contrasting application areas, these two studies bear a common signature that is prevalent in many other domains as well: the actions of the entities in a network-structured complex system are strategically inter-dependent. This dissertation presents a computational game-theoretic framework for studying causal questions in such scenarios.
展开▼
