Missing observations in regression: a conditional approach

Area of Science:

• Statistics
• Biostatistics
• Sociometrics

Background:

• Missing covariate data poses challenges in regression analysis.
• Existing methods like multiple imputation may not fully address ancillarity considerations.

Purpose of the Study:

• To present an alternative regression analysis approach for missing covariate data.
• To develop a method more aligned with ancillarity principles in regression.
• To evaluate the sensitivity of regression parameters to missing data.

Main Methods:

• Utilizing factorial and fractional factorial arrangements.
• Assessing the sensitivity of regression parameter inference to missingness in explanatory variables.
• Illustrating the approach with medical and sociological examples.

Main Results:

• The proposed method provides an alternative to multiple imputation.
• Sensitivity analysis offers insights into the impact of missing data on regression parameters.
• The approach is applicable to diverse fields, including medicine and sociology.

Conclusions:

• The recommended method is more faithful to ancillarity in regression analysis.
• This approach enhances the robustness of regression inference with missing covariate data.
• The method demonstrates utility in analyzing complex datasets in healthcare and social sciences.