Abstract:Non-compliance is common in real world experiments. We focus on inference about the sample complier average causal effect, that is, the average treatment effect for experimental units who are compliers. We present three types of inference strategies for the sample complier average causal effect: the Wald estimator, regression adjustment estimators and model-based Bayesian inference. Because modern computer assisted experimental designs have been used to improve covariate balance over complete randomization, we discuss inference under both complete randomization and a specific computer assisted experimental design - Mahalanobis distance based rerandomization, under which asymptotic properties of the Wald estimator and regression adjustment estimators can be derived. We use Monte Carlo simulation to compare the finite sample performance of the methods under both experimental designs. We find that under either design, the Bayesian method performs the best because it is stable, it yields smallest median absolute error and smallest median interval length. The improvement by the Bayesian method is especially large when the fraction of compliers is small. We present an application to a job training experiment with non-compliance.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to infer the Sample Complier Average Causal Effect (sample CACE) in experiments under Complete Randomization (CRE) and computer - aided balanced improvement designs (such as Mahalanobis Distance Based Rerandomization, ReM). Specifically, the paper focuses on how to estimate the average treatment effect of individuals who comply with treatment allocation in the presence of non - compliance. The paper proposes three inference strategies to estimate the sample CACE: 1. **Wald estimator**: This is a method based on Intention - to - Treat (ITT), which estimates CACE by comparing the difference in average outcomes between the treatment group and the control group with the difference in the proportion of those receiving treatment. 2. **Regression - adjusted estimator**: This method improves the estimation precision by adding covariates to the model to adjust for imbalance. 3. **Model - based Bayesian inference**: Using Bayesian methods, CACE is estimated by simulating missing data, and this method performs particularly well in small - sample situations. The paper also discusses the asymptotic properties of the Wald estimator and the regression - adjusted estimator under complete randomization and Mahalanobis - distance - based rerandomization designs, and compares the performance of different methods in finite samples through Monte Carlo simulations. The results show that under any design, the Bayesian method performs best due to its stability, the smallest median absolute error, and the shortest median interval length, especially when the proportion of compliers is small. In addition, the paper also provides an application example, an employment training experiment with non - compliance, demonstrating the application of these methods to practical problems.

Inference of Sample Complier Average Causal Effects under Experiments with Completely Randomized Design and Computer Assisted Balance-Improving Designs

Inference of Sample Complier Average Causal Effects in Completely Randomized Experiments

Model-based Bayesian inference under computer assisted balance-improving designs

Bayesian Inference for Causal Effects in Randomized Experiments with Noncompliance

Model-assisted complier average treatment effect estimates in randomized experiments with non-compliance and a binary outcome

Model-assisted complier average treatment effect estimates in randomized experiments with non-compliance and a binary outcome

Bayesian Causal Inference in Sequentially Randomized Experiments with Noncompliance

A Two-stage Inference Procedure for Sample Local Average Treatment Effects in Randomized Experiments

Inference in Experiments with Matched Pairs and Imperfect Compliance

Semiparametric Inference of the Complier Average Causal Effect with Nonignorable Missing Outcomes

Efficient Covariate Balancing for the Local Average Treatment Effect

Inference for Two-stage Experiments under Covariate-Adaptive Randomization

Synthetic Principal Component Design: Fast Covariate Balancing with Synthetic Controls

MODE: Minimax Optimal Deterministic Experiments for Causal Inference in the Presence of Covariates

Sampling-based randomized designs for causal inference under the potential outcomes framework

Estimating the Complier Average Causal Effect with Non-Ignorable Missing Outcomes Using Likelihood Analysis

Synthetic estimation for the complier average causal effect

A Comparison of Methods of Inference in Randomized Experiments from a Restricted Set of Allocations

Causal Inference in Rebuilding and Extending the Recondite Bridge Between Finite Population Sampling and Experimental Design

Leveraging Random Assignment to Impute Missing Covariates in Causal Studies

Instrumental variable approaches for estimating complier average causal effects on bivariate outcomes in randomised trials with non-compliance