Effect Size Reporting

A p-value tells you whether an effect is detectable given your sample size. An effect size tells you how large the effect actually is. Publishing p-values without effect sizes produces results that cannot be synthesized in meta-analyses and cannot be used to power replication studies.

STEP 1 —The Pitfall

The majority of published figures in biology, psychology, and medicine report p-values and significance stars but omit effect sizes. This is not a minor omission —it makes the result uninformative. A significant p-value with n=500 may reflect a trivially small effect. A non-significant result with n=6 may reflect a clinically meaningful effect that the study was underpowered to detect.

Effect size is not a post-hoc addition. It is the core measure of whether a result matters, and its omission is increasingly cited as a condition for rejection by journals following APA or ICMJE reporting standards.

[Image placeholder: Two bar charts with identical p=0.03. Left: n=400, Cohen's d=0.15 (trivial). Right: n=12, Cohen's d=1.2 (large). Demonstrates that p-value alone is uninterpretable without effect size.]

Identical p-values, radically different effect sizes. p alone does not convey the magnitude of the result.

STEP 2 —Journal Requirement

The APA Publication Manual (7th ed.) requires effect size for all inferential statistics. Nature's reporting checklist asks: "For each comparison, is the effect size reported?" NEJM and Lancet require effect estimates (not just p-values) with 95% CIs. Most biostatistics-literate reviewers now flag the absence of effect size as an automatic revision request.

STEP 3 —Licklider Solution

Input

Statistical test selected (t-test, ANOVA, chi-squared, correlation, regression, and other verified paths; rank-based names may appear in selection metadata — see Non-Parametric Alternatives)
Group structure and outcome variable from Variable & ID Mapping
n per group from Observation Unit Declaration

Output

Cohen's d for two-group mean comparisons (t-test; Mann–Whitney only when that rank path is actually executed — see Non-Parametric Alternatives)
eta² / partial eta² for ANOVA-family tests
r (rank-biserial correlation) when a rank-based comparison is actually executed (product support is limited — see Non-Parametric Alternatives)
Cramﾃｩr's V for chi-squared / Fisher's exact
Odds ratio / relative risk for binary outcomes
Effect size automatically embedded in figure legend and Methods text
Validity Score dimension: Fail if effect size absent from publication-ready export

Guard

Publication-ready export is blocked if the effect size for any inferential comparison is absent. The guard fires regardless of whether the result is significant —effect size is required for both significant and non-significant results. An override mechanism exists for exploratory analyses explicitly marked as non-inferential via Outcome Type Lock.

[Image placeholder: Licklider figure export panel showing effect size (Cohen's d = 0.82, 95% CI [0.41, 1.23]) displayed alongside the p-value in the figure legend, with the Validity Score badge showing this dimension as Pass.]

Effect size auto-inserted into the figure legend. The Validity Score dimension shows Pass once effect size is present.

STEP 4 —Draft Output (Draft / Needs review)

Group A vs. Group B: mean difference = 3.2 units (95% CI [1.8, 4.6]),
Cohen's d = 0.82 (large effect), t(24) = 4.11, p = 0.0004,
two-tailed independent samples t-test.