Description

If jackknife_stats gets a numpy.ma.MaskedArray, for example a light curve with flagged bad points, it returns no error and no warning, but the estimate, bias and standard error are wrong. The full-sample statistic honours the mask. The leave-one-out resamples do not, because the masked values are copied into them as ordinary data. The result therefore depends on the value hidden under the mask.

Expected behavior

The result should match the one for the unmasked values, estimate = 2.5, bias = 0, se = 0.6455. Alternatively, jackknife_stats could raise an error saying that masked input is not supported.

How to Reproduce

import numpy as np
from astropy.stats import jackknife_stats

# 5 flux measurements; the last one is flagged bad and masked
data = np.ma.array([1.0, 2.0, 3.0, 4.0, 1000.0], mask=[0, 0, 0, 0, 1])

est, bias, se, ci = jackknife_stats(data, np.mean)
print("masked input :", est, bias, se)

est, bias, se, ci = jackknife_stats(data.compressed(), np.mean)
print("compressed   :", est, bias, se)

masked input : -795.5 798.0 199.50062656543213
compressed   : 2.5 0.0 0.6454972243679028

Versions

import astropy
astropy.system_info()

# Paste the result here