Closed-form moment-based estimator for the quadratically-weighted Conger / Fleiss / Brennan-Prediger family. Treats categorical ratings as numeric scores and uses per-rater means and covariances.
Use this when the quadratic weighting is the right loss and a parametric moment-based estimate is preferred over the U-statistic version (`kappa(x, estimator = "pairwise")`). The two agree under the usual asymptotics; the closed form can be faster on large `n`.
With missing ratings, every rater must be observed at least once and every rater pair must be jointly observed by at least one subject.
Arguments
- x
A subjects-by-raters numeric matrix of category scores; `NA` marks missing entries.
- values
Length-C numeric vector of category scores. The quadratic loss is `(values[i] - values[j])^2 / (max - min)^2`. The covariance is the empirical covariance of the reduced quadratic moment summaries from their row-wise estimating equations.