Partitioning estimation of local variance based on nearest neighbors under censoring (Q725665)

Let \(X\) be a random vector of covariates with some distribution \(\mu\), and \(Y\) be the dependent random variable. Let \(m(x)=\mathbb{E}(Y\,|\,X=x)\), \( x\in\mathbb{R}^d,\) be the regression function, and \[ \sigma^2(x)=\mathbb{E}\big((Y-m(x))^2\,|\, X=x\big) \] be the conditional local variance. The author of the paper proposes new estimators of \(\sigma^2(x)\) based on the first and the second nearest neighbors under censoring. For the proposed estimators their consistence and rate of convergence are discussed. The local variance estimation is demonstrated on the basis of real survival data.