Answers are grounded in this paper and cite their sources.
Demonstrates improved power in testing independence in nonparametric regression models, indicating enhanced performance.
We propose an information-weighted extension of the Bergsma–Dassios τ * statistic for testing independence between a covariate and the regression error in nonparametric regression models. The method incorporates deterministic weights based on Fisher information and Kullback–Leibler divergence to emphasize regions of the covariate space with high informational content. The resulting statistic remains a degenerate fourth-order U-statistic under the null hypothesis and preserves the same asymptotic null distribution as the classical τ * , enabling valid permutation-based inference. Under contiguous alternatives, the weighting amplifies the noncentrality parameter, yielding improved local asymptotic power. Simulation studies and a real-data application demonstrate substantial power gains over the unweighted τ * while maintaining accurate type I error control.
No takes yet. Share an insight, caveat, or question.
Sthitadhi Das (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: