Causal Effect Estimation with Hidden Confounding and Aggregated Data

Yuchen Zhu (Ph.D. Student)

Causal effect estimation is ubiquitous in many social science and public health domains. In such domains treatment variables and outcome variables are often confounded, and sometimes the confounding is hidden, requiring methods capable of correcting for these. Moreover, in many social science domains the variables of interest are aggregates of finer-grained variables; for example, Gross Domestic Product is an aggregated variable of individual level economic activities. Causal effects are ill-defined in macro-variables since the implementation of a given intervention on a given macro-variable is not unique. This thesis addresses both of these aspects. On one hand, methodolgies are developed for non-parametric effect estimation under heterogeneity, hidden confounding and measurement error; these methods come with theoretically guaranteed correctness such as consistency and convergence rate. On the other hand, theoretical foundations are developed for causal reasoning with aggregated data, including a framework for constructing consistent compression of large scale fine-grained causal models, and a framework for optimising for implementations with respect to given constraints.

Primary Advisor:	Matt J. Kusner (University College London)
Industry Advisor:	Dominik Janzing (Amazon Research Tübingen)
PhD Duration:	01 October 2020 - 01 August 2025