Partial Identification with Proxy of Latent Confoundings via Sum-of-ratios Fractional Programming

TL;DR

This paper introduces PI-SFP, a novel method for causal effect estimation under partial observability of confoundings, using sum-of-ratios fractional programming to provide valid bounds without strong assumptions.

Contribution

It develops a general single-proxy negative control approach that handles partial information of confounding transition matrices, expanding applicability beyond previous methods.

Findings

PI-SFP yields promising numerical results in simulations.

It effectively handles partial information scenarios.

Fills gaps in causal inference literature for irreversible transition matrices.

Abstract

Due to the unobservability of confoundings, there has been widespread concern about how to compute causality quantitatively. To address this challenge, proxy-based negative control approaches have been commonly adopted, where auxiliary outcome variables $\bm{W}$ are introduced as the proxy of confoundings $\bm{U}$. However, these approaches rely on strong assumptions such as reversibility, completeness, or bridge functions. These assumptions lack intuitive empirical interpretation and solid verification techniques, hence their applications in the real world are limited. For instance, these approaches are inapplicable when the transition matrix $P(\bm{W} \mid \bm{U})$ is irreversible. In this paper, we focus on a weaker assumption called the partial observability of $P(\bm{W} \mid \bm{U})$. We develop a more general single-proxy negative control method called Partial Identification via Sum-of-ratios Fractional Programming (PI-SFP). It is a global optimization algorithm based on the branch-and-bound strategy, aiming to provide the valid bound of the causal effect. In the simulation, PI-SFP provides promising numerical results and fills in the blank spots that can not be handled in the previous literature, such as we have partial information of $P(\bm{W} \mid \bm{U})$.