Invited commentary: mixing multiple imputation and bootstrapping for variance estimation.

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Methods

Background:

• Missing data can introduce bias in statistical analyses.
• Multiple imputation (MI) is a common technique to address missing data.
• Propensity score methods are used to adjust for confounding.

Purpose of the Study:

• To review current approaches for variance estimation when using multiple imputation with propensity scores.
• To complement the work of Nguyen and Stuart on the statistical consistency of MI and propensity score integration.
• To address the lack of consensus on implementing MI and nonparametric bootstrapping.

Main Methods:

• Review of existing literature on variance estimation techniques.
• Focus on nonparametric bootstrapping as a method for variance estimation.
• Discussion of different approaches for combining MI and propensity score analysis.

Main Results:

• Variance estimation for integrated MI and propensity score methods requires further development.
• Nonparametric bootstrapping offers a flexible approach for variance estimation when closed-form solutions are unavailable.
• Several approaches exist for combining MI and nonparametric bootstrapping, but consensus is lacking.

Conclusions:

• Further research is needed to establish best practices for variance estimation in MI and propensity score analyses.
• Nonparametric bootstrapping is a viable option for valid inference in complex statistical models.
• Standardizing the implementation of MI and bootstrapping is crucial for reliable results in epidemiological studies.