stats.comparison.pairwise_paired_t

stats.comparison.pairwise_paired_t(panel, *, correction='holm')

Pairwise two-sided paired t-tests over the columns of a panel.

Runs a paired t-test for every pair of entries (columns) over the shared blocks (rows) and returns the p-values as a square symmetric matrix — the pairwise stage of the blocked comparison protocol, rendered with spotforecast2.plots.comparison.plot_critical_difference.

Parameters

Name	Type	Description	Default
panel	pd.DataFrame	Complete block-by-entry frame (e.g. one row per target day, one column per entry, daily MAE values). Must contain no NaN — subset to shared blocks first.	required
correction	str	Multiple-testing correction applied across the pairs. Options: - `"holm"`: Holm step-down adjustment (`holm_adjust`). - `"none"`: raw p-values. Defaults to `"holm"`.	`'holm'`

Returns

Name	Type	Description
	pd.DataFrame	pd.DataFrame: Square symmetric p-value matrix over
	pd.DataFrame	`panel.columns` with 1.0 on the diagonal.

Raises

Name	Type	Description
	ValueError	When `panel` has fewer than two columns or rows, contains NaN, or `correction` is unknown.

Examples

import pandas as pd
from spotforecast2_safe.stats.comparison import pairwise_paired_t

days = pd.date_range("2026-06-10", periods=8, freq="D")
panel = pd.DataFrame(
    {
        "alpha": [500.0, 520, 480, 510, 490, 505, 515, 495],
        "beta": [540.0, 565, 515, 555, 530, 545, 560, 535],
        "gamma": [502.0, 519, 483, 508, 493, 504, 517, 496],
    },
    index=days,
)
p = pairwise_paired_t(panel)
print(p.round(4).to_string())
assert (p.values == p.values.T).all()

       alpha  beta  gamma
alpha  1.000   0.0  0.247
beta   0.000   1.0  0.000
gamma  0.247   0.0  1.000