- Published
- Updated
Selecting a missing-data method is more than choosing a software option. Reasons for absence, previous measurements and treatment history help assess whether the assumptions behind an analysis are plausible in the particular trial setting.
Define the question and assumptions
Inspect patterns by time, arm and known reason. Missingness explainable by observed information differs from missingness that also depends on an unobserved outcome. The data cannot verify every assumption; clinical colleagues should understand the conditions under which the chosen statistical approach supports its intended interpretation.
Hypothetical study example
In a hypothetical study, participants reporting worsening symptoms more often miss the following questionnaire. The team reviews earlier measurements and recorded reasons. It considers possible dependence on the unobserved outcome rather than treating a low overall missingness percentage as sufficient reassurance about the resulting estimate.
Design review checks
- Describe missingness through patterns and reasons rather than one overall percentage.
- Confirm that observed variables supporting the assumption were actually collected reliably.
- Plan sensitivity assessments representing meaningful departures from the main assumption.
Record the decision and its limits
An accepted plan connects mathematical terminology to clinical scenarios. Report uncertainty when missingness reasons themselves are incomplete or unreliable. Successful model execution does not prove its assumptions correct. Explain how conclusions change under plausible alternatives rather than choosing every alternative specifically to preserve the preferred primary finding.
Frequently asked questions
- Is a small amount of missing data always unimportant?
- No. Reasons and distribution between groups can change its impact.
- Can the assumption be proved from the dataset?
- Usually not completely; clinical rationale and sensitivity assessment remain necessary.