What approach is recommended for MNAR (missing not at random) data?

• A. Complete-case analysis
• B. Multiple imputation
• C. Sensitivity analyses
• D. Listwise deletion

Answer: (C)

When data are MNAR, the likelihood of missingness depends on unobserved values, so methods that assume missingness at random aren’t reliable on their own. The recommended approach is to perform sensitivity analyses to see how conclusions would change under different MNAR assumptions. This means exploring plausible ways the missing data could behave—for example, using pattern-based or selection models, or applying delta adjustments in imputation—and re-running the analysis under those scenarios. If results remain consistent across a reasonable range of MNAR assumptions, you gain confidence in the findings; if they shift, you transparently report the range of possible conclusions and discuss implications. Complete-case analysis and listwise deletion ignore differences between observed and missing data and can lead to biased results when MNAR is present, while multiple imputation relies on MAR and may still be biased under MNAR unless supplemented with sensitivity analyses.