Rejoinder to the letter: "Standard and reference-based conditional mean imputation: Regulators and trial...

Area of Science:

• Statistics
• Biostatistics
• Missing Data Analysis

Background:

• The ongoing debate on variance estimation for reference-based imputation methods, specifically "multiple imputation" versus "conditional mean imputation," warrants clarification.
• Both methods share fundamental assumptions and yield comparable treatment effect estimates, challenging the presented dichotomy.

Discussion:

• This work clarifies that conditional mean imputation, while typically associated with frequentist variance, can be adapted for variance estimation similar to Rubin's approach.
• The equivalence in assumptions and estimates between multiple imputation and conditional mean imputation is highlighted.

Key Insights:

• Conditional mean imputation coupled with jackknife resampling provides a robust and adaptable deterministic method for missing data under missing-at-random or reference-based assumptions.
• Frequentist variance accurately reflects the assumptions inherent in reference-based imputation methods.

Outlook:

• Further investigation into the theoretical underpinnings of Rubin's variance in frequentist and Bayesian frameworks is needed.
• Promoting a unified understanding of imputation methods and variance estimation is crucial for advancing statistical practice.