ec_diffnet_test(): a differential network without a
Gaussian assumption, following the permutation route of NetCoMi’s
diffnet and the max-over-edges statistic of the Network
Comparison Test, with exchangeability blocks in the sense of Winkler et
al. (2014). The edge statistic is the heterogeneity of the
Fisher-transformed correlations across groups, the same quantity the
Gaussian theory uses, referred to the law obtained by permuting the
group labels instead of to a chi-square; on ranks the procedure is
distribution-free. One p-value for the whole network (sum or maximum
over edges) with the per-edge p-values from the same permutations for a
closed follow-up, so a p(p-1)/2-edge network pays no multiplicity
correction, and a cluster permutation for interviewer or ward effects.
Measured cost against the Gaussian route on Gaussian data: about five
points of power. Measured gain off it: the Gaussian null rejects a true
null 28 per cent of the time on t(3) data and 29 per cent under
clustering, where this holds 5 and 4.ec_multi_test(): one permutation test for a battery of
categorical items across groups. This is Pesarin’s nonparametric
combination with the direct (sum) and Tippett (maximum) combining
functions, and the maximum with per-item follow-up is a step of Westfall
and Young’s maxT; the sum or the maximum over items of a per-item
Pearson or score-dispersion statistic referred to the law obtained by
permuting the group labels, across respondents or across clusters. One
p-value for the whole battery, no multiplicity correction, and per-item
p-values from the same permutations for a closed follow-up. Provided
because no package in this programme’s toolchain offered these with
exchangeability blocks in one place; on the prevention-behaviour paper’s
data the maximum-statistic form matched or beat a Bonferroni-corrected
Fisher battery at its own size.First release.
ec_rc_test() and ec_rc_null(): exact
conditional inference in a two-way table given both margins, sampled by
random permutation of the column labels. Statistics are Goodman’s
linear-by-linear association, the dispersion of the row mean scores for
nominal rows against an ordinal column, Pearson’s statistic and the
likelihood ratio.ec_rc_two(): exact conditional test that two tables
share a linear-by-linear parameter.ec_sphere_maxquad(): exact null for a quadratic form
maximised over candidate structures, valid after selection by fit.ec_pvalue(): tail probabilities with a mid-p companion
and a Wilson interval for the Monte Carlo error, so an estimated p-value
is not quoted as though it were exact arithmetic.