Robust causal inference with continuous instruments using the local\n instrumental variable curve

2016/07/09 by Edward H. Kennedy, Scott A. Lorch, Kennedy, Edward H. +3 · 2 voices · 1 citation
Mathematics · Economics, Econometrics and Finance · #Advanced Causal Inference Techniques #Statistical Methods and Bayesian Inference #Healthcare Policy and Management

paper · pdf · doi:10.48550/arxiv.1607.02566

Abstract

Instrumental variables are commonly used to estimate effects of a treatment\nafflicted by unmeasured confounding, and in practice instruments are often\ncontinuous (e.g., measures of distance, or treatment preference). However,\navailable methods for continuous instruments have important limitations: they\neither require restrictive parametric assumptions for identification, or else\nrely on modeling both the outcome and treatment process well (and require\nmodeling effect modification by all adjustment covariates). In this work we\ndevelop the first semiparametric doubly robust estimators of the local\ninstrumental variable effect curve, i.e., the effect among those who would take\ntreatment for instrument values above some threshold and not below. In addition\nto being robust to misspecification of either the instrument or\ntreatment/outcome processes, our approach also incorporates information about\nthe instrument mechanism and allows for flexible data-adaptive estimation of\neffect modification. We discuss asymptotic properties under weak conditions,\nand use the methods to study infant mortality effects of neonatal intensive\ncare units with high versus low technical capacity, using travel time as an\ninstrument.\n

Cited by

• A Practical Introduction to Regression-based Causal Inference in Meteorology (II): Unmeasured confounders

Discussions

• Cool thread! MTE/LIV is pretty neat - very natural & advantageous w/ cts IVs I worked on semiparametric methods for this effect here: arxiv.org/pdf/1607.025... Also interesting statistically - a ra [bsky, 10 points, 1 comments]
• My intuition for MTE shape is pretty limited except in cases where you have a clear "incentive IV" like payment to take up treatment, in which case it gives you a kind of willingness-to-pay curve. Ed [bsky, 2 points, 0 comments]

Related

• Nonparametric Estimation of Local Treatment Effects with Continuous Instruments
• Local Effects of Continuous Instruments without Positivity
• Marginal Causal Effect Estimation with Continuous Instrumental Variables
• Doubly Robust Nonparametric Instrumental Variable Estimators for\n Survival Outcomes
• Identification and multiply robust estimation of causal effects via instrumental variables from an auxiliary population