8.3 Missing-data decisions

Missingness should be investigated as a process. Ask whether values are missing because the field was not collected, did not apply, failed during entry, was suppressed, or was lost during extraction. Examine whether missingness changes over time, by system, location, group, or outcome.

Possible responses include:

• retaining a meaningful unknown category;
• excluding a variable that is too incomplete for the intended use;
• limiting analysis to periods with comparable collection;
• imputing under a justified statistical process;
• conducting sensitivity analysis; or
• reporting that the question cannot be answered reliably.

Imputation creates estimated values. It does not recover the unknown truth, and it should not be performed only to satisfy software that rejects missing data.