Minimax Slope Estimation and Support Recovery under Weak Gaussian Selection
In the authors' words
In sample selection models, identifying a regression slope does not determine how accurately it can be estimated or distinguished from zero. We establish sharp uniform limits for outcome slopes in a fixed-dimensional Gaussian/probit selection model with unknown scale and an unknown covariate density bounded above and away from zero on a cube. We prove a uniform inverse bound between slope differences and observed-law Hellinger distance, allowing every nuisance parameter to differ between the two models. A matching path lies entirely within positive-correlation nonignorable selection. On parameter classes containing a feasible weak-selection neighborhood, the minimax squared slope risk is of order . For each prespecified minimum-signal sequence , uniform exact support recovery is possible if and only if , with rules allowed to depend on that sequence. The same estimator attains uniform root- slope accuracy on for each fixed finite , including the exact zero-selection-slope layer. A testing-based construction establishes statistical attainability without estimating the covariate density; computational efficiency is not asserted.
Appeared: Tuesday, October 20. arXiv. Preprint, not yet peer-reviewed.
Authors' comment: 41 pages, 4 figures. Complete proofs in Appendices A-G; numerical experiments in Appendices H-I; code, data and detailed simulation protocols in ancillary files