Separate structurally skipped values, unknown responses, refusal and invalid entries. A field that was never asked is different from a response that was lost. Review the instrument and eligible population before interpreting a missing-cell count.
Define missingness first
Separate structurally skipped values, unknown responses, refusal and invalid entries. A field that was never asked is different from a response that was lost. Review the instrument and eligible population before interpreting a missing-cell count.
Look beyond the overall percentage
Summarise patterns by relevant groups, sites, time and outcomes where permitted. Similar overall completeness can hide systematically missing follow-up in one group. These patterns inform questions; they do not prove that data are missing at random.
Worked example: follow-up loss
Illustrative example: people with worse baseline outcomes are less likely to return. A complete-case result may describe returners rather than the original target population. Ask how that selection changes interpretation and what assumptions alternative methods would need.
Original decision log
Use Variable | Eligible count | Missing count/reason | Pattern by group | Plausible mechanism | Candidate handling | Assumptions | Sensitivity check | Reviewer. Record why deletion, weighting or imputation was considered; do not choose an option because it produces a preferred p-value.
Report what remains uncertain
Compare conclusions under plausible alternatives where feasible and explain unavailable information. Imputation is not recovery of observed truth. A defensible report makes its assumptions and uncertainty visible rather than describing the completed dataset as error-free.
Review this research decision
We can review your actual form, dataset and reporting requirement, then scope the smallest service that solves the risk.
Explore the service ↗