Improving efficiency in transporting average treatment effects.

Area of Science:

• Statistics
• Epidemiology
• Causal Inference

Background:

• Estimating treatment effects in new populations (transportability) is crucial when effects vary.
• Differences in average treatment effect (ATE) often stem from varying distributions of effect modifiers.
• Existing methods may lack efficiency or require extensive covariate data.

Purpose of the Study:

• To develop flexible, efficient semiparametric estimators for transporting the average treatment effect (ATE) to a target population.
• To offer estimators with varying data requirements, accommodating situations with incomplete covariate information in the target population.
• To evaluate the performance of proposed methods against existing techniques and under practical challenges like positivity violations.

Main Methods:

• Proposed a collaborative one-step semiparametric estimator requiring all covariates in the target population.
• Developed two one-step semiparametric estimators assuming knowledge of effect modifiers and their differential distribution.
• Utilized simulation studies to compare finite sample performance and applied methods to a housing trial.

Main Results:

• The proposed estimators offer potential efficiency gains for transported ATE estimation.
• Estimators requiring less covariate information in the target population are viable when data is limited.
• Methods demonstrated performance in simulations, including scenarios with positivity assumption violations.

Conclusions:

• Flexible semiparametric methods enhance the transportability of average treatment effects.
• The choice of estimator depends on data availability and knowledge of effect modifiers.
• These methods provide valuable tools for causal inference in diverse populations and settings.