Pingouin Cheat Sheet

2T-teststwo groups

pg.ttest(x, y, paired=False)★
Independent-samples t-test. Returns T, dof, alternative, p-val, CI95%, Cohen's d, power, and BF10 (Bayes factor) in one row.
pg.ttest(pre, post, paired=True)★
Paired t-test for repeated measures. correction="auto" applies Welch's for unequal variances in the unpaired case.
pg.ttest(x, 0.5) # one-sample vs a population mean
Pass a scalar y for a one-sample test. alternative="greater"/"less" for one-sided.
pg.tost(x, y, bound=0.5) # equivalence test
Two one-sided tests (TOST) for statistical equivalence — proving groups are "close enough".

3Non-parametric testsno normality

pg.mwu(x, y) # Mann-Whitney U (unpaired)★
Rank-based alternative to the unpaired t-test. Returns U, p-val, and effect sizes (RBC, CLES).
pg.wilcoxon(x, y) # Wilcoxon signed-rank (paired)★
Paired non-parametric test.
pg.kruskal(data=df, dv="score", between="group")
Kruskal-Wallis: non-parametric one-way ANOVA (3+ independent groups).
pg.friedman(data=df, dv="score", within="time", subject="id")
Friedman: non-parametric repeated-measures ANOVA. pg.cochran for binary outcomes.

4Pairwise & post-hocwhich pairs differ

pg.pairwise_tests(data=df, dv="score", within="time", subject="id", padjust="bonf")★
The workhorse post-hoc: all pairwise comparisons with multiple-comparison correction. Handles within/between/interaction and switches parametric↔non-parametric via parametric=.
padjust: "bonf" · "holm" · "fdr_bh" · "sidak" · "none"★
Correction methods for the family of p-values; the corrected column is p-corr.
pg.pairwise_tukey(data=df, dv="score", between="group")
Tukey HSD for a balanced one-way design. pg.pairwise_gameshowell for unequal variances.
pg.pairwise_corr(data=df, columns=["a","b","c"])
Pairwise correlations across many columns, with correction — a correlation matrix with p-values.

5ANOVA family3+ groups

pg.anova(data=df, dv="score", between="group", detailed=True)★
One-way / N-way between-subjects ANOVA. Returns SS, DF, MS, F, p-unc, and np2 (partial eta-squared). Pass a list to between for factorial designs.
pg.rm_anova(data=df, dv="score", within="time", subject="id")★
Repeated-measures ANOVA (SciPy can't). Reports sphericity (correction) automatically.
pg.mixed_anova(data=df, dv="score", within="time", between="group", subject="id")★
Mixed-design ANOVA (one within + one between factor) — the classic pre/post × treatment design.
pg.welch_anova(...) · pg.sphericity(...) · pg.epsilon(...)
welch_anova for unequal variances; sphericity (Mauchly) & epsilon corrections for RM designs.

6ANCOVA & regressionwith covariates

pg.ancova(data=df, dv="score", between="group", covar="age")★
ANOVA controlling for one or more continuous covariates. covar takes a list.
pg.linear_regression(X, y)★
OLS returning coefficients, SE, T, p-val, r2, adj-r2, and CI per predictor — a tidy alternative to statsmodels. X can be multi-column.
pg.logistic_regression(X, y)
Logistic regression with coefficients, SE, z, p, and CIs.
pg.rm_corr(data=df, x="x", y="y", subject="id")
Repeated-measures correlation — the correct within-subject association (not pooled across subjects).

7Correlationtwo variables

pg.corr(x, y)★
Pearson by default. Returns n, r, CI95%, p-val, power, and BF10 — and flags outliers.
pg.corr(x, y, method="spearman")★
Also "kendall", "bicor" (biweight midcorrelation), "percbend", "shepherd", "skipped" — robust options resistant to outliers.
pg.corr(x, y, alternative="greater")
One-sided correlation test.

8Partial & matrix correlationcontrol for covariates

pg.partial_corr(data=df, x="a", y="b", covar="c")★
Correlation between a & b controlling for c (a list of covariates works). x_covar/y_covar for semi-partial.
df.rcorr() # pandas accessor★
Pingouin adds a .rcorr() DataFrame method: a correlation matrix with significance stars in the upper triangle.
df.pairwise_corr() · df.partial_corr(...)
Several functions are also exposed as pandas DataFrame methods for fluent chaining.

9Mediation & contingencyindirect effects

pg.mediation_analysis(data=df, x="X", m="M", y="Y", n_boot=500)★
Simple & multiple mediation with bootstrapped confidence intervals for the indirect effect.
expected, observed, stats = pg.chi2_independence(data=df, x="a", y="b")★
Chi-square test of independence for a contingency table, with several effect sizes (Cramér's V). pg.chi2_mcnemar for paired.
pg.multivariate_normality(X) · pg.box_m(...)
Multivariate normality (Henze-Zirkler) and homogeneity of covariance matrices.

10Assumption checksbefore you trust it

pg.normality(df, dv="score", group="grp")★
Shapiro-Wilk (default) per group; returns a pass/fail normal boolean column. method="normaltest"/"jarque_bera".
pg.homoscedasticity(df, dv="score", group="grp")★
Levene (default) / Bartlett test for equal variances — decides whether you need Welch's correction.
pg.qqplot(x, dist="norm")
Q-Q plot to eyeball normality. Pair with the numeric test above.
# p < .05 on normality/levene = assumption VIOLATEDread it right
A significant assumption test means the assumption fails — switch to a robust/non-parametric test or a correction.

11Effect size & powerbeyond p

pg.compute_effsize(x, y, eftype="cohen")★
Cohen's d, Hedges' g, Glass's delta, eta-square, CLES, and more. pg.convert_effsize converts between them.
pg.power_ttest(d=0.5, n=None, power=0.8, alpha=0.05)★
Power analysis: leave exactly one of n/d/power/alpha as None and it solves for it. Also power_anova, power_corr, power_chi2.
pg.compute_esci(...) · pg.compute_bootci(x, func="mean")
Analytic and bootstrapped confidence intervals for effect sizes and arbitrary statistics.

12Plotting & extrasvisualize

pg.plot_paired(data=df, dv="score", within="time", subject="id")
Before/after paired plot showing each subject's change.
pg.plot_blandaltman(x, y)
Bland-Altman agreement plot for two measurement methods.
pg.plot_rm_corr(...) · pg.plot_circmean(...)
Repeated-measures correlation plot; circular-statistics helpers (pg.circ_mean, circ_r, circ_corrcc) for angular data.
df.groupby("grp").apply(lambda d: pg.normality(d.score))
Because results are DataFrames, they compose with pandas groupby/apply and export straight to CSV/LaTeX.

Pingouin cheat sheet