New V-fold jackknife offers efficient statistical inference alternative

Researchers have introduced the V-fold jackknife, a new statistical method designed as a computationally efficient and theoretically sound alternative to the bootstrap for semiparametric inference. This method requires only V leave-fold-out refits and uses the dispersion of jackknife pseudo-values to estimate uncertainty, bypassing the need for influence functions. For standard asymptotically linear estimators, the V-fold jackknife statistic converges to a t-distribution, yielding valid confidence intervals. The approach also extends to generalized estimators and has demonstrated reliable inference in simulations for average treatment effects, Kaplan-Meier survival curves, and highly adaptive lasso dose-response curves. AI

IMPACT Provides a more efficient and theoretically grounded method for statistical inference in machine learning contexts.

RANK_REASON Academic paper introducing a new statistical methodology. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv stat.ML →

• average treatment effect
• Bootstrap
• Highly Adaptive Lasso (HAL)
• Kaplan-Meier survival curves
• Studentized V-fold jackknife
• Student's t-distribution
• V-fold jackknife

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →