Comparing Instrumental Variable Methods for Continuous Exposure and Binary Outcome
Author Information
Author(s): Shunichiro Orihara, Atsushi Goto
Primary Institution: Tokyo Medical University, Tokyo, Japan; Yokohama City University, Yokohama, Japan
Hypothesis
This study aims to conduct a simulation for a detailed comparison of various instrumental variable methods focusing on valid scenarios.
Conclusion
The findings indicate that no single method is best for addressing bias in instrumental variable methods, and using multiple methods is recommended.
Supporting Evidence
- Six instrumental variable methods were classified into three groups based on their performance.
- The first group showed clear bias due to outcome model misspecification.
- The second group performed well with strong instruments but suffered bias with weak instruments.
- The third group produced conservative results and was less affected by weak instruments.
Takeaway
This study looked at different ways to analyze data when some important information is missing, and found that using more than one method is usually better.
Methodology
The study compared six instrumental variable methods under 32 simulation scenarios using an artificial dataset.
Potential Biases
The study noted significant biases in certain methods due to model misspecification.
Limitations
The study only explored 32 specific situations and was limited to 200 simulation iterations.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website