Assumption & Robustness Guard

Most statistical errors are not calculation errors —they are assumption errors. The Assumption & Robustness Guard is the unified pipeline that checks all estimation-relevant assumptions before analysis proceeds and routes the analysis to the appropriate method based on what it finds.

STEP 1 —The Pitfall

Assumption checking is typically done manually, inconsistently, and selectively. A researcher might check normality for one analysis but not another. They might check equal variance only when they remember to. They might check independence only if someone asks. The result is an inconsistent standard applied to the same dataset across different figures in the same paper.

The inconsistency is itself a validity concern: if normality was checked and reported for Figure 2 but not for Figure 4, a reviewer cannot know whether Figure 4 passed or whether it was simply not checked. Uniform assumption checking across all figures eliminates this ambiguity.

[Image placeholder: Decision flowchart starting at "Is data normally distributed?" —branching to parametric / non-parametric / permutation paths. Each branch also shows the CI estimation method (analytical / bootstrap) and the effect size estimator appropriate for that path.]

Assumption-to-method routing: the decision tree that determines which estimation method is appropriate for a given dataset and question.

STEP 2 —Journal Requirement

Journal reporting guidelines generally require that the basis for test selection be stated. "Appropriate statistical tests were selected based on data distribution and variance homogeneity" is acceptable; the assumption checks that support this statement should be reportable on request. A uniform assumption-checking record satisfies this requirement for all figures simultaneously.

STEP 3 —Licklider Solution

Input

All active figures in the session
Analysis type and group structure for each figure
Declared analysis intent (exploratory / confirmatory / descriptive) from Outcome Type Lock

Output

Per-figure assumption check summary: normality (pass/warn/fail), equal variance (pass/warn/fail), independence (pass/warn/fail —linked to Pseudoreplication Detection), sample size adequacy
Method recommendation: parametric / non-parametric / permutation / bootstrap —based on the assumption check results
Assumption-method alignment indicator: flags figures where the selected method does not match the recommended method
Unified Methods text: single assumption-checking statement covering all figures in the export package
Validity Score contribution: assumption violations that are unresolved reduce the overall Validity Score

Guard

For confirmatory analyses, if the selected method does not match the recommended method and no justification is provided, the guard sets a Warning on the relevant Validity Score dimensions. The export package includes a flag noting the assumption-method misalignment. For exploratory analyses, the flag is advisory only.

[Image placeholder: Licklider session-level assumption summary table showing all figures as rows and assumption checks as columns (Normality / Equal Variance / Independence / Sample Size Adequacy). Color-coded: green pass, yellow warn, red fail. Recommended method for each figure shown in the final column.]

Session-level assumption summary: all figures, all checks, all method recommendations in one view.

STEP 4 —Draft Output (Draft / Needs review)

Statistical test selection was informed by assumption checking performed
prior to analysis. Data were tested for normality (Shapiro-Wilk for n 竕､ 50,
D'Agostino-Pearson for n > 50) and for equal variance (Levene's test).
Parametric tests (t-test, one-way ANOVA) were applied to data meeting
normality and equal variance assumptions. Non-parametric tests
(Mann-Whitney U, Kruskal-Wallis) were applied where assumptions were
violated. Assumption check results per figure are provided in
Supplementary Table 1.

Example journal-style Methods text. Rank-based tests named here are illustrative; verify in-app execution against Non-Parametric Alternatives.