Audit Missingness Patterns Across Variables

Define the variables and observation unit

A missingness pattern describes which selected fields are observed or missing within one record. Decide whether that record represents a participant, visit, or specimen, and name the fields in a fixed order. Without those definitions, two pattern tables can look different merely because their columns or populations differ.

Start from a documented missing-value map. The official pandas missing-value documentation illustrates that a software detector recognizes particular missing markers; empty strings and other special values are not automatically interpreted as missing. Research-specific codes need an explicit cleaning rule.

Count records as well as missing cells

Calculate each field's missing count, the number of records missing any selected field, and the number complete across the selected fields. These are different quantities. Adding field-specific missing counts counts missing cells and can count the same record several times.

Preserve a mutually exclusive table of observed/missing combinations. Write the pattern legend prominently, since some software uses one to mean observed while other tools use it to mean missing. Never interpret a binary pattern without its coding definition.

Recalculate a fictional overlap

Imagine six fictional participant records and fields A and B. Two records have both fields observed, two have only A missing, one has only B missing, and one has both missing. A is missing in three records and B in two, giving five missing cells across twelve possible cells.

Four participants have at least one missing field, while two are complete. The sum three plus two exceeds the four incomplete participants because one person contributes to both field counts. In the missing-pattern tool, 1 means missing and 0 means observed. Enter patterns 00, 10, 01, and 11 with counts 2, 2, 1, and 1 respectively. It requires unique, equal-length patterns and positive whole-number counts.

Investigate the process behind the pattern

Use the missing-pattern worksheet to retain the population, field order, coding map, pattern counts, and unresolved questions. Compare patterns with the collection schedule: a skipped module, unavailable specimen, and missed follow-up may require different explanations.

Do not automatically combine structural non-applicability with an unobserved eligible measurement. Retain the original reason where available even if a downstream analysis uses a common missing marker. Check patterns by relevant study group or time point without exposing identifying records in a demonstration tool.

A pattern table cannot establish that missingness is random or determine an appropriate imputation model. It is a descriptive audit that helps formulate questions about the data-generating process. Preserve that distinction when handing results to the analyst.

Finish by checking that pattern counts sum to the selected record count and that their implied field totals match the individual missing counts. A reconciled table makes overlap visible and prevents percentages from silently switching between cells, records, and participants.

Source checked on 3 November 2026; fictional pattern counts independently enumerated.