Missing Data Disclosure

Missing data is not a nuisance to hide; it is information. The pattern, mechanism, and handling of missing values can fundamentally change the validity of conclusions. Journals increasingly treat non-disclosure of missingness as a reportable methodological problem.

STEP 1 - The Pitfall

Complete-case analysis, simply dropping rows with missing values, is the default in most statistical software and the default choice of many researchers. But complete-case analysis is only valid when data are Missing Completely At Random (MCAR). When missingness is related to the outcome or an unmeasured variable (MAR or MNAR), complete-case analysis can produce biased estimates.

Common non-obvious missingness patterns include below-limit-of-detection (LOD) values coded as zero or NA, dropout in longitudinal data related to disease progression, and non-response bias in surveys. Each mechanism requires different handling and disclosure.

[Image placeholder: Three panels showing different missingness patterns: (1) random scatter (MCAR), (2) missingness correlated with group (MAR), and (3) missingness correlated with the missing value itself, for example high-value measurements below LOD (MNAR).]

MCAR / MAR / MNAR patterns and the direction of bias they can introduce in complete-case analysis.

STEP 2 - Journal Requirement

CONSORT and STROBE require reporting of missing data rates and handling methods. ARRIVE 2.0 requires disclosure of excluded animals and reasons. Journals following ICMJE guidance expect a clear description of how missing data were handled in Methods.

STEP 3 - Licklider Solution

Input

Raw data with missing value indicators (NA, NaN, blank, sentinel values)
Group structure from Variable & ID Mapping
LOD thresholds if applicable (declared in the Data Contract or preprocessing metadata)

Output

Missing value rate per column and per group (displayed in the data summary panel)
Declared missingness mechanism (MCAR / MAR / MNAR) captured in the disclosure record when bias risk is elevated
LOD disclosure guard: declared LOD-related exclusions stay visible in the draft Methods and legend output
N Disclosure: input N vs. analysis N displayed automatically, with per-group breakdown when available
Methods and legend draft text include the missingness summary and the declared mechanism assumption
Validity Score dimension: Fail if analysis N differs from input N and no missingness disclosure is provided

Guard

If the analysis excludes rows relative to the input dataset and no missing data disclosure text is present, the Validity Score dimension is set to Fail and claim-bearing export is blocked. When missingness risk is elevated, the guard is satisfied only after the researcher declares the assumed mechanism and carries that disclosure into the draft output.

[Image placeholder: Licklider missing data summary panel showing column-level missingness rates, group-level N comparison (input vs. analysis N), and the declared missingness mechanism.]

Missing data summary panel: per-column and per-group missingness with mechanism declaration and LOD disclosure status.

STEP 4 - Draft Output (Draft / Needs review)

Of 48 input observations, 44 were included in the primary analysis
(4 excluded: 2 below limit of detection, 2 quality control failures
declared at study design). Missing data rate = 8.3%. A missing-at-random
(MAR) mechanism was declared for the primary analysis. Complete-case
analysis was performed; sensitivity analysis including the declared LOD
handling rule produced consistent results (Supplementary Table 1).