CAREER: Design and analysis of experiments for complex social processes

Modern scientific inquiry from the social to the health sciences centers around answering foundational “what if?” questions with a special emphasis on understanding the effects of changing complex social processes such as the arrangement of individuals into networks or groups and communication of information via text. The causal inference literature has heralded the role of randomization in getting answers to such “what if?” questions, treating randomized controlled experiments as a gold standard for testing simple causal hypotheses.

However, hidden behind this powerful tool are a series of assumptions and design decisions that are difficult to control for and are untenable in the context of complex and changing social processes. The PI will develop novel theory and methodology that will directly address the role of these social processes in causal inference. The new tools can be used across disciplines to study interventions whenever social processes are present such as in the study of important societal questions relating to vaccines, non-pharmaceutical interventions, implications of different education policies and the like.

The PI's education plan integrates the research products from this project into courses that will engage students from a wide range of academic backgrounds, presenting and linking the methodological contributions to substantive applications. Research products will be widely disseminated to the scientific community and the general public through popular and scientific publications, presentations, and open-source software.

This project addresses the nascent areas of causal inference in the presence of network information and text data. While much of the work in these areas has concentrated on the analysis of existing experimental designs, little work has gone into designing experiments specifically for such complex social processes. The research will demonstrate the inadequacy of classical designs and the importance of developing specialized experimental designs that adapt to underlying complex social processes.

For network and text data, the PI will provide a comprehensive framework for defining causal quantities of interest that exploit the structure in such data. The PI will work on three main thrusts: (1) Conditional design in the presence of network information: this thrust will develop restricted randomizations to target the testing and estimation of peer effects, total effects and other network quantities; (2) Unconditional design where the experimenter can control the social process: this thrust will draw on results from graph sampling to develop experimental designs that simultaneously design an interaction graph (that can represent how study participants will be allowed to interact) and a treatment allocation; and (3) Text as a social process: this thrust will provide guidance and tools for extracting causal quantities from text data that can play the role of treatment, outcome, confounder or mediator in a causal analysis.

The output of the research will include practical guidelines for experimental design as well as adaptable tools and algorithms that can be deployed to address a wide range of social processes and applied problems beyond those studied in this project.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.