For MAR missingness, which imputation approach is commonly recommended?

For MAR missingness, which imputation approach is commonly recommended?

• A. Mean imputation for MAR.
• B. Multiple imputation is commonly recommended for MAR.
• C. Complete-case analysis.
• D. Listwise deletion.

Answer: (B)

Missingness at random means the fact that data are missing can be explained by information we observe, so we can use the observed variables to predict the missing values. In this setting, multiple imputation is widely recommended because it creates several plausible imputations for each missing entry based on a model that uses observed data, analyzes each completed dataset, and then pools the results. This process captures the uncertainty about what the true values might be and preserves relationships among variables, leading to less biased estimates and proper standard errors under MAR. Mean imputation fills in missing values with a variable’s average. It ignores relationships with other variables, underestimates variability, and can bias relationships and inferences. Complete-case analysis uses only records with no missing values. Under MAR, this can bias results and wastes information, reducing power. Listwise deletion is essentially the same idea as complete-case analysis and carries the same drawbacks under MAR.

Missingness at random means the fact that data are missing can be explained by information we observe, so we can use the observed variables to predict the missing values. In this setting, multiple imputation is widely recommended because it creates several plausible imputations for each missing entry based on a model that uses observed data, analyzes each completed dataset, and then pools the results. This process captures the uncertainty about what the true values might be and preserves relationships among variables, leading to less biased estimates and proper standard errors under MAR.

Mean imputation fills in missing values with a variable’s average. It ignores relationships with other variables, underestimates variability, and can bias relationships and inferences.

Complete-case analysis uses only records with no missing values. Under MAR, this can bias results and wastes information, reducing power.

Listwise deletion is essentially the same idea as complete-case analysis and carries the same drawbacks under MAR.