Primary Submission Category: Causal Inference in Networks
Improving causal inference controls using network theory in discrete choice data
Authors: Bernardo Modenesi,
Presenting Author: Bernardo Modenesi*
Many datasets in health and social sciences result from agents making repeated choices over time, each choice leading to an observable outcome. Researchers often aim to model the causal impact of covariates on the outcome variable using various estimation strategies (e.g. fixed effects regression, difference-in-differences, instrumental variables, etc). I propose a new way to increase control in these estimation procedures by applying network theory models motivated by a discrete choice framework. I suggest representing these datasets as a bipartite network, where agents are nodes on one side and choices are nodes on the other. Edges in this network represent a choice made by an agent at a certain time, stemming from a discrete choice problem. I argue that the structure of connections in this choice-network allows the researcher to further improve controls when modeling the outcome variable. For instance, I use the choice-network to project agents into a multidimensional latent space that captures each agent’s choice-profile. Distances between agents in this latent space represent a metric of similarity between them. By exploring the high-dimensional choice-profile of agents, I propose several ways to enhance causal inference exercises, as well as to compute heterogeneous treatment effects.