A screening method for a nonparametric model (Q1086953)

Let (W,Z) be a two-dimensional r.v. such that W and Z are positively dependent in the sense that the function \(\Theta\) defined by \(\Theta (t,K)=P(W>K| Z=z)\) is a continuous, strictly increasing function of t for any \(K\in {\mathbb{R}}\). We consider estimation of the so-called screening threshold, whose definition is motivated by some discrimination problems, on the basis of a completely observable learning sample. The construction of the estimator utilises the order restrictions imposed on the model. Strong consistency of the proposed estimators is proved and their behaviour for moderate sample sizes is investigated via simulation.