Using Balancing Weights to Target the Treatment Effect on the Treated when Overlap is Poor.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Inverse probability weights (IPW) are standard for estimating causal effects in observational studies.
• Extreme weights from poor covariate overlap in IPW can bias estimates of average treatment effect (ATE) or average treatment effect on the treated (ATT).
• Overlap weights offer an alternative but can yield difficult-to-interpret causal estimands.

Purpose of the Study:

• To investigate if balancing weights can effectively target the ATT when IPW methods produce biased estimates due to poor covariate overlap.
• To compare the performance of balancing weights against IPW and overlap weights in scenarios with limited covariate overlap.

Main Methods:

• The study employed three simulation studies to assess different weighting strategies.
• An empirical application was conducted to validate findings in a real-world dataset.
• Balancing weights were explored as an alternative to IPW for targeting the ATT.

Main Results:

• Balancing weights demonstrated effectiveness in targeting the ATT, even in the presence of poor covariate overlap.
• Estimates using balancing weights showed reduced bias compared to IPW in simulated scenarios with poor overlap.
• Overlap weights proved useful but balancing weights offered a way to target more familiar causal estimands.

Conclusions:

• Balancing weights provide a viable alternative to IPW for estimating the ATT, particularly when covariate overlap is limited.
• While overlap weights are valuable, balancing weights can facilitate the estimation of well-understood causal effects in challenging data conditions.
• The findings suggest that balancing weights enhance the robustness of causal effect estimation in observational epidemiology.