PG ANOVA
Use when the outcome is continuous and predictors are categorical factors, optionally with covariates or repeated measures.
Load
Read:
../../references/supervision-gates.md../../references/pingouin-api-quickref.md../../references/pingouin-optimization.md../../references/apa-output-template.md if writing results.
Function Choice
- One between-subject factor ->
pg.anova(data=df, dv=..., between=..., detailed=True).
- Multiple between-subject factors ->
pg.anova(data=df, dv=..., between=[...], detailed=True).
- Unequal variances in one-way between design -> consider
pg.welch_anova.
- One or more within-subject factors ->
pg.rm_anova(..., within=..., subject=..., detailed=True).
- One within-subject factor plus one between-subject factor ->
pg.mixed_anova.
- Between-subject factor plus continuous covariate ->
pg.ancova.
- Follow-up contrasts ->
pg.pairwise_tests with padjust.
Required Inputs
- Dependent variable.
- Between-subject factor(s).
- Within-subject factor(s).
- Subject ID for repeated/mixed designs.
- Covariates for ANCOVA.
- Planned contrasts or post hoc intent.
- Desired effect size, if not Pingouin default.
Code Patterns
One-way ANOVA:
aov = pg.anova(data=df, dv="score", between="group", detailed=True).round(3)
pg.print_table(aov)
Repeated-measures ANOVA:
aov = pg.rm_anova(data=df, dv="score", within="condition",
subject="id", detailed=True).round(3)
pg.print_table(aov)
Mixed ANOVA:
aov = pg.mixed_anova(data=df, dv="score", within="time",
between="group", subject="id").round(3)
pg.print_table(aov)
Follow-up:
posthoc = pg.pairwise_tests(data=df, dv="score", within="time",
between="group", subject="id",
padjust="holm", effsize="hedges").round(3)
pg.print_table(posthoc)
ANCOVA:
aov = pg.ancova(data=df, dv="score", between="group", covar="baseline").round(3)
Interpretation Checklist
- Identify omnibus effects before pairwise claims.
- For repeated-measures effects, check and report sphericity or corrected p-values if returned.
- Report effect size column present in output, commonly
np2.
- Report correction used for post hoc comparisons.
- For interactions, do not interpret main effects as simple group differences unless appropriate.
Guardrails
mixed_anova is for a specific mixed design; do not use it for arbitrary multi-level nesting.- Pingouin is not a full mixed-effects model package. For random slopes/nested clusters, recommend statsmodels/R lme4 instead.
- Do not run ANOVA on wide repeated-measures data until reshaped or a Pingouin function explicitly accepts that shape.
- If cell sizes are very small or missing cells exist, inspect design balance before trusting output.
- End result-bearing answers with one compact S0-S5 audit line.
1---2name: pg-anova3description: Run or generate Pingouin code for one-way, factorial, repeated-measures, mixed, Welch ANOVA, ANCOVA, and follow-up pairwise tests.4---56# PG ANOVA78Use when the outcome is continuous and predictors are categorical factors, optionally with covariates or repeated measures.910## Load1112Read:1314- `../../references/supervision-gates.md`15- `../../references/pingouin-api-quickref.md`16- `../../references/pingouin-optimization.md`17- `../../references/apa-output-template.md` if writing results.1819## Function Choice2021- One between-subject factor -> `pg.anova(data=df, dv=..., between=..., detailed=True)`.22- Multiple between-subject factors -> `pg.anova(data=df, dv=..., between=[...], detailed=True)`.23- Unequal variances in one-way between design -> consider `pg.welch_anova`.24- One or more within-subject factors -> `pg.rm_anova(..., within=..., subject=..., detailed=True)`.25- One within-subject factor plus one between-subject factor -> `pg.mixed_anova`.26- Between-subject factor plus continuous covariate -> `pg.ancova`.27- Follow-up contrasts -> `pg.pairwise_tests` with `padjust`.2829## Required Inputs3031- Dependent variable.32- Between-subject factor(s).33- Within-subject factor(s).34- Subject ID for repeated/mixed designs.35- Covariates for ANCOVA.36- Planned contrasts or post hoc intent.37- Desired effect size, if not Pingouin default.3839## Code Patterns4041One-way ANOVA:4243```python44aov = pg.anova(data=df, dv="score", between="group", detailed=True).round(3)45pg.print_table(aov)46```4748Repeated-measures ANOVA:4950```python51aov = pg.rm_anova(data=df, dv="score", within="condition",52 subject="id", detailed=True).round(3)53pg.print_table(aov)54```5556Mixed ANOVA:5758```python59aov = pg.mixed_anova(data=df, dv="score", within="time",60 between="group", subject="id").round(3)61pg.print_table(aov)62```6364Follow-up:6566```python67posthoc = pg.pairwise_tests(data=df, dv="score", within="time",68 between="group", subject="id",69 padjust="holm", effsize="hedges").round(3)70pg.print_table(posthoc)71```7273ANCOVA:7475```python76aov = pg.ancova(data=df, dv="score", between="group", covar="baseline").round(3)77```7879## Interpretation Checklist8081- Identify omnibus effects before pairwise claims.82- For repeated-measures effects, check and report sphericity or corrected p-values if returned.83- Report effect size column present in output, commonly `np2`.84- Report correction used for post hoc comparisons.85- For interactions, do not interpret main effects as simple group differences unless appropriate.8687## Guardrails8889- `mixed_anova` is for a specific mixed design; do not use it for arbitrary multi-level nesting.90- Pingouin is not a full mixed-effects model package. For random slopes/nested clusters, recommend statsmodels/R lme4 instead.91- Do not run ANOVA on wide repeated-measures data until reshaped or a Pingouin function explicitly accepts that shape.92- If cell sizes are very small or missing cells exist, inspect design balance before trusting output.93- End result-bearing answers with one compact S0-S5 audit line.
Exekiel179