exekiel179/pingouin-psych-stats/pg-mean-tests

PG Mean Tests

Use for t tests and post hoc pairwise comparisons where the dependent variable is approximately continuous.

Load

Read:

• ../../references/supervision-gates.md
• ../../references/pingouin-api-quickref.md
• ../../references/pingouin-optimization.md
• ../../references/apa-output-template.md if writing results.

Decision Rules

• One sample against a known value -> pg.ttest(x, y=<value>).
• Two independent groups -> pg.ttest(x, y, paired=False, correction="auto").
• Same participants measured twice -> pg.ttest(x, y, paired=True).
• More than two group levels or multiple pairwise contrasts -> pg.pairwise_tests.
• Non-parametric pairwise comparisons -> pg.pairwise_tests(..., parametric=False).

Required Inputs

• Outcome column.
• Group or condition column.
• Subject ID for paired/repeated comparisons.
• Which comparisons are planned versus post hoc.
• Multiple-comparison correction: default to holm for post hoc families unless the user specifies another correction.
• Alternative hypothesis: default to two-sided.

Code Patterns

Independent t test:

x = df.loc[df["group"].eq("A"), "score"]
y = df.loc[df["group"].eq("B"), "score"]
res = pg.ttest(x, y, paired=False, correction="auto",
               alternative="two-sided", confidence=0.95).round(3)
pg.print_table(res)

Paired t test from wide columns:

res = pg.ttest(df["pre"], df["post"], paired=True,
               alternative="two-sided", confidence=0.95).round(3)
pg.print_table(res)

Pairwise tests:

res = pg.pairwise_tests(data=df, dv="score", between="group",
                        parametric=True, padjust="holm",
                        effsize="hedges").round(3)
pg.print_table(res)

Repeated pairwise tests:

res = pg.pairwise_tests(data=df, dv="score", within="condition",
                        subject="id", padjust="holm",
                        effsize="hedges").round(3)

Output Checks

Inspect res.columns before writing prose. Expected t-test columns often include T, dof, p-val, CI95%, cohen-d, BF10, and power.

Guardrails

• Do not use deprecated pairwise_ttests.
• Do not call independent tests for paired data.
• Do not omit multiplicity correction for exploratory multi-comparison families.
• Report Welch/student choice when independent groups are imbalanced.
• If the outcome is ordinal or severely non-normal with small n, mention non-parametric alternatives.
• End result-bearing answers with one compact S0-S5 audit line.