PredictiveCheckTest.pvalues

property PredictiveCheckTest.pvalues: NDArray[float64]

(m, k) tail probabilities \(\Pr(T(y^{\text{rep}}) \ge T(y))\).

The share of replications whose statistic exceeds the observed one, plus half the share that equals it, so a statistic the model reproduces exactly reads \(0.5\) and a discrete statistic – min or max on integer-valued data – is not pushed to one by ties. A cell is nan when the statistic is undefined on any replication. Rows follow statistics and columns names. Recomputed on each access from observed and replicated; the cost is one comparison per cell and replication.

Example

>>> import numpy as np
>>> rng = np.random.default_rng(0)
>>> y = rng.standard_normal((100, 2))
>>> check = replication_check(y, rng.standard_normal((200, 100, 2)))
>>> p = check.pvalues
>>> p.shape, bool(np.all((0.0 <= p) & (p <= 1.0)))
((5, 2), True)