Skip to contents

For an outcome vector, the Stephenson score of a unit at ascending rank j is choose(j - 1, r - 1). Ties take average ranks. Larger r concentrates weight on the top ranks; r=2 reduces to the (rank-1) Wilcoxon score.

Usage

riposte_stephenson_scores(y, r)

Arguments

y

numeric vector (a single block's outcomes).

r

the Stephenson tuning value, a whole number; r=2 is the Wilcoxon (rank - 1) score and r=1 gives a constant (uninformative) score. Must be a whole number: a fractional r would silently round inside choose().

Value

numeric vector of scores the same length as y.