N Disclosure & Missing Data Trail

The figure shows p = 0.03, but 30% of the uploaded observations were silently excluded. The reader has no way to know. Licklider's N Disclosure system surfaces the input N vs. analysis N gap for every figure, and for every group when a grouping variable is declared, and requires the researcher to acknowledge material discrepancies before claim-bearing export.

STEP 1 - The Pitfall: The Invisible Exclusion Gap

When a researcher uploads 60 observations and publishes a figure based on 42, the Methods section might say "n = 21 per group." What it rarely says is "n = 30 per group were uploaded; 9 per group were excluded due to missing outcome data." The reader sees the analysis N; the input N disappears.

This matters for several reasons:

• Missing data is not neutral. Data can be missing completely at random (MCAR), missing at random (MAR), or missing not at random (MNAR). MNAR, where the probability of missingness depends on the missing value itself, can introduce systematic bias that is undetectable from the analysis N alone.
• LOD/LOQ handling is easy to hide. In assay data, values below the limit of detection are often set to zero, half-LOD, or excluded entirely. These decisions are rarely made explicit in the Methods unless they are carried into the disclosure record.
• Group-level N imbalance is obscured. A figure comparing two groups with n = 21 vs. n = 18 after exclusions looks balanced. A figure that started as n = 21 vs. n = 30 before exclusions tells a very different story.

STEP 2 - Journal Requirement

The CONSORT 2010 checklist mandates a participant flow summary showing numbers enrolled, allocated, followed up, and analyzed, with reasons for exclusions. The STROBE statement requires equivalent disclosure for observational studies. Nature and related journals expect the number of independent samples or animals in each experimental group to remain visible in the figure-facing record.

For preclinical and basic science research, NC3Rs ARRIVE 2.0 guidelines require reporting of the number of animals used in each experiment and the number analyzed. For in vitro data, the number of biological replicates must be distinguished from technical replicates.

A reviewer who asks "how many samples were excluded and why?" is exercising a standard journal requirement. Without automated N tracking, the honest answer often requires reconstructing information from multiple sources and may still be incomplete.

STEP 3 - Licklider's Solution

Input

• The uploaded dataset (CSV), from which Licklider records the initial row count and, when available, the initial row count per group.
• Grouping variable and timepoint variable as resolved at figure render time, used to track row counts at input and analysis stage.
• All preprocessing operations and outlier exclusions applied during the session, carried forward from the preprocessing and outlier logs.

Output

• N disclosure summary surfaced via guard cards, template drafts, and chat disclosure. Input N and analysis N for the current figure are surfaced through these review surfaces, plus per-group breakdowns when a group variable is available.
• Missing data pattern review. Licklider checks whether missingness is concentrated in one group or time point. When the observed attrition pattern raises bias risk, the researcher must declare the assumed missingness mechanism (MCAR / MAR / MNAR) before claim-bearing export.
• LOD/LOQ disclosure guard. LOD-related exclusion counts and fractions are captured when lod_suspected is flagged in the Data Contract. Handling choices such as set-to-zero or half-LOD substitution are not currently recorded automatically and should be described manually in the disclosure text.
• Figure legend N annotation. A reviewable legend draft is generated with the analysis N and, when available, the group-wise N summary.

Guard

STEP 4 - Current Draft Output (Draft / Needs review)

Example: Legend draft with N disclosure

[Draft / Needs review] Add the dataset and analysis-unit description.
N=46; 14 rows were excluded before analysis (23.3%).
Group-wise n summary: control (input n=30, analysis n=25, dropped=5);
treatment (input n=30, analysis n=21, dropped=9).
[Needs review] Add the final inferential details if this legend supports
a claim-bearing figure.

Example: Methods blocks

Data
The figure used 46 analyzed rows from 60 input rows, with 14 rows
excluded before analysis (23.3%). Group-wise attrition summary:
control (input n=30, analysis n=25, dropped=5); treatment
(input n=30, analysis n=21, dropped=9).

Caveats
missingness may affect conclusions: overall dropout 23.3%;
group dropout imbalance 13.3%.
[Not resolved yet] Confirm the attrition disclosure before publication.

The examples above reflect the current implementation. Licklider generates draft methods blocks and legend text that require editing before publication. Note: MAR/MNAR assumption text and sensitivity analysis references are not generated automatically and should be added manually before publication.

Related Features