
Nonparametric efficient causal mediation with intermediate confounders
Interventional effects for mediation analysis were proposed as a solutio...
read it

A unifying approach for doublyrobust ℓ_1 regularized estimation of causal contrasts
We consider inference about a scalar parameter under a nonparametric mo...
read it

Asymptotic normality of the timedomain generalized least squares estimator for linear regression models
In linear models, the generalized least squares (GLS) estimator is appli...
read it

Efficient Least Squares for Estimating Total Effects under Linearity and Causal Sufficiency
Recursive linear structural equation models are widely used to postulate...
read it

Estimating Treatment Effects with Observed Confounders and Mediators
Given a causal graph, the docalculus can express treatment effects as f...
read it

The Variance of Causal Effect Estimators for Binary Vstructures
Adjusting for covariates is a well established method to estimate the to...
read it

RKL: a general, invariant Bayes solution for NeymanScott
NeymanScott is a classic example of an estimation problem with a partia...
read it
Distributional Robustness of Kclass Estimators and the PULSE
In causal settings, such as instrumental variable settings, it is well known that estimators based on ordinary least squares (OLS) can yield biased and nonconsistent estimates of the causal parameters. This is partially overcome by twostage least squares (TSLS) estimators. These are, under weak assumptions, consistent but do not have desirable finite sample properties: in many models, for example, they do not have finite moments. The set of Kclass estimators can be seen as a nonlinear interpolation between OLS and TSLS and are known to have improved finite sample properties. Recently, in causal discovery, invariance properties such as the moment criterion which TSLS estimators leverage have been exploited for causal structure learning: e.g., in cases, where the causal parameter is not identifiable, some structure of the nonzero components may be identified, and coverage guarantees are available. Subsequently, anchor regression has been proposed to tradeoff invariance and predictability. The resulting estimator is shown to have optimal predictive performance under bounded shift interventions. In this paper, we show that the concepts of anchor regression and Kclass estimators are closely related. Establishing this connection comes with two benefits: (1) It enables us to prove robustness properties for existing Kclass estimators when considering distributional shifts. And, (2), we propose a novel estimator in instrumental variable settings by minimizing the mean squared prediction error subject to the constraint that the estimator lies in an asymptotically valid confidence region of the causal parameter. We call this estimator PULSE (puncorrelated least squares estimator) and show that it can be computed efficiently, even though the underlying optimization problem is nonconvex. We further prove that it is consistent.
READ FULL TEXT
Comments
There are no comments yet.