Pg Correlations

PG Correlations

Use for association analyses between continuous or ordinal variables.

Load

Read:

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

Function Choice

• Two continuous variables, linear association -> pg.corr(..., method="pearson").
• Ordinal or monotonic/non-normal association -> method="spearman" or method="kendall".
• Association controlling covariates -> pg.partial_corr.
• Many variables -> pg.pairwise_corr with padjust.
• Repeated observations per participant -> pg.rm_corr.
• Nonlinear dependence screening -> pg.distance_corr if appropriate.

Required Inputs

• Variables to correlate.
• Whether observations are independent.
• Covariates, if partial correlation is requested.
• Multiple-comparison family and correction.
• Hypothesis direction; default to two-sided.

Code Patterns

Basic correlation:

res = pg.corr(x=df["x"], y=df["y"], method="pearson",
              alternative="two-sided").round(3)
pg.print_table(res)

Partial correlation:

res = pg.partial_corr(data=df, x="x", y="y",
                      covar=["age", "baseline"],
                      method="pearson").round(3)

Pairwise correlations:

res = pg.pairwise_corr(data=df, columns=["x", "y", "z"],
                       method="spearman", padjust="holm").round(3)

Repeated-measures correlation:

res = pg.rm_corr(data=df, x="x", y="y", subject="id").round(3)

Reporting

Report n, r, CI, p-value, Bayes factor if returned, and method. Include a scatterplot recommendation when interpretation depends on linearity or outliers.

Guardrails

• Correlation is not causation.
• Do not use ordinary pg.corr when observations repeat within participants.
• If outliers drive the effect, report sensitivity analysis or robust alternative.
• If many correlations are explored, use and report p-value adjustment.
• Do not interpret partial correlation as causal control.
• End result-bearing answers with one compact S0-S5 audit line.