Transparent effect-size conversions
Choose the analysis context first. Effi only performs conversions that have a defensible mathematical relationship and explains when a conversion is approximate, restricted, or not identifiable.
Required parameters
Conversion result
How to use and how to interpret
A statistic is not automatically an effect size. A test statistic also needs its design and degrees of freedom, and some conversions cannot recover direction or the original information.
Step 1 — choose the analysis context
The context controls which statistics are offered and prevents treating a paired measurement or an omnibus ANOVA as an independent two-group comparison.
Step 2 — enter only the parameters that belong to that analysis
Do not use a planned sample size in place of the analysed N. For independent groups, N should equal n₁ + n₂. For ANOVA, df₁ can be obtained from k − 1 and df₂ from N − k only when those values are genuinely applicable.
Important limitations
- Cohen’s d, Hedges’ g, and t conversions are for two independent groups unless the paired dz option is selected.
- The OR–d relationship is a logistic-distribution approximation, not an identity.
- χ² cannot recover a sign; a direction must be supplied if a signed φ, r, or d is requested.
- A single omnibus F does not identify a unique pairwise d. Effi allows F → d only for df₁ = 1 and marks the direction as user-supplied.
- Cohen’s f and η²-family measures describe ANOVA-type effects and should not be interpreted as if they were correlations.
Supported relationships
| Metric | Meaning | Main requirements |
|---|---|---|
| Cohen’s d / Hedges’ g | Standardised mean difference | n₁ and n₂ for conversions involving t, r or g |
| r / OR | Association measures | r↔d uses n₁ and n₂; OR↔d is approximate |
| F / η² / partial η² / ω² / f | Omnibus or ANOVA-type effects | df₁ and df₂ for F and ω² pathways; k may supply df₁ |
| χ² / φ / V | Contingency-table association | N; rows and columns for V; 2×2 for φ |
Bulk conversion from CSV
Upload a CSV, select the numeric source-value column, and use the same settings above for every row. Invalid rows are retained with an explanation.