Abstract:Instrumental variables (IVs) are widely used to estimate causal effects in the presence of unobserved confounding between exposure and outcome. An IV must affect the outcome exclusively through the exposure and be unconfounded with the outcome. We present a framework for relaxing either or both of these strong assumptions with tuneable and interpretable budget constraints. Our algorithm returns a feasible set of causal effects that can be identified exactly given relevant covariance parameters. The feasible set may be disconnected but is a finite union of convex subsets. We discuss conditions under which this set is sharp, i.e., contains all and only effects consistent with the background assumptions and the joint distribution of observable variables. Our method applies to a wide class of semiparametric models, and we demonstrate how its ability to select specific subsets of instruments confers an advantage over convex relaxations in both linear and nonlinear settings. We also adapt our algorithm to form confidence sets that are asymptotically valid under a common statistical assumption from the Mendelian randomization literature.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to use instrumental variables (IVs) to estimate causal effects in the presence of unobserved confounding factors. Traditionally, instrumental variables must satisfy two key assumptions: First, the instrumental variable affects the outcome variable only through the exposure variable; second, there are no unobserved confounding factors between the instrumental variable and the outcome variable. However, in practical applications, these two assumptions are often difficult to be strictly satisfied. Especially in Mendelian Randomization (MR) studies, when genetic variation is used as a candidate instrumental variable, these assumptions may be violated due to pleiotropy or linkage disequilibrium. For this reason, the paper proposes a new framework - BudgetIV, which is used to relax these strong assumptions and partially identify causal effects through adjustable and interpretable budget constraints. Specifically, this framework allows the exclusivity and exogeneity assumptions of instrumental variables to be violated to a certain extent, and returns a feasible set of causal effects through an algorithm. This set can accurately determine the causal effects under given relevant covariance parameters. This method is applicable to a wide range of semi - parametric models and has advantages in selecting specific subsets of instrumental variables, especially in linear and nonlinear settings. In addition, this method can also form asymptotically valid confidence sets under common statistical assumptions. In general, the paper aims to provide a more flexible and practical method to deal with the limitations of instrumental variable assumptions in practical applications, especially in genetic epidemiology.

BudgetIV: Optimal Partial Identification of Causal Effects with Mostly Invalid Instruments

Bounding Causal Effects with Leaky Instruments

Valid Causal Inference with (Some) Invalid Instruments

Sequential Underspecified Instrument Selection for Cause-Effect Estimation

Nonparametric identification and efficient estimation of causal effects with instrumental variables

Identifiability of Sparse Causal Effects using Instrumental Variables

Identification of causal effects using instrumental variables

Identification and Inference with Invalid Instruments

Identification and multiply robust estimation of causal effects via instrumental variables from an auxiliary heterogeneous population

Confidence Intervals for Causal Effects with Invalid Instruments using Two-Stage Hard Thresholding

Identifying Causal Effects using Instrumental Time Series: Nuisance IV and Correcting for the Past

Control Function Instrumental Variable Estimation of Nonlinear Causal Effect Models

Two Robust Tools for Inference about Causal Effects with Invalid Instruments

Discovering Ancestral Instrumental Variables for Causal Inference From Observational Data

Identifying Causal Effects of Discrete, Ordered and ContinuousTreatments using Multiple Instrumental Variables

The Instrumental Variable Model with Categorical Instrument, Treatment and Outcome: Characterization, Partial Identification, and Statistical Inference

Instrumental Variables Estimation with Some Invalid Instruments and its Application to Mendelian Randomization

Bounded, efficient and multiply robust estimation of average treatment effects using instrumental variables

Identification of Causal Effects Using Instrumental Variables: Comment

Selective Machine Learning of the Average Treatment Effect with an Invalid Instrumental Variable

Learning causal effects from many randomized experiments using regularized instrumental variables