Instrumental variable analysis with categorical treatment

Area of Science:

• Econometrics
• Biostatistics
• Causal Inference

Background:

• Existing instrumental variable methods primarily address dichotomous or ordinal treatment variables.
• Causal inference for multiple unordered treatments is underexplored due to increasing assumption complexity.
• There is a need for robust methods to compare effects of several interventions simultaneously.

Purpose of the Study:

• To derive causal point-estimators for head-to-head comparisons of multiple unordered treatments.
• To develop a methodology using plausible rationality assumptions and ordinal instruments.
• To address the limitations of current instrumental variable approaches in complex treatment scenarios.

Main Methods:

• Developed a novel instrumental variable methodology for comparing multiple unordered treatments.
• Utilized a set of well-defined rationality assumptions and ordinal instruments.
• Employed simulation studies to assess asymptotic unbiasedness in the presence of unobserved confounding.

Main Results:

• The proposed methodology yields asymptotically unbiased estimators, even with unobserved confounding.
• Demonstrated effectiveness in a simulation study comparing multiple treatment alternatives.
• Successfully applied to compare five anti-inflammatory drugs for rheumatoid arthritis using observational data.

Conclusions:

• The developed methodology extends instrumental variable analysis to scenarios with more than two unordered treatments.
• Provides a valuable tool for causal inference in comparative effectiveness research.
• Enhances the ability to compare multiple interventions when an instrumental variable is available.