Generalized M-estimators for high-dimensional Tobit I models (Q668611): Difference between revisions

Consider the design of statistical methods in case of censored data which may occur e.g. if the measurement device in physical sciences has a certain value limit. For coping with censored data, robust methods are discussed here, especially for high-dimensional left-censored linear models: \[ y_t = \max \{0,\beta^Tx_t + \epsilon_t, t=1,2,\dots,n \}, \] where $x_t, y_t$, $t=1,\dots,n$, are the input $p$-vectors, scalar output variables, resp., $\epsilon_t$, $t=1,\dots,n$, denote independent error vectors, and $\beta$ is the unknown parameter $p$-vector to be estimated. For high-dimensional data, robust estimation methods and confidence intervals are presented for left-censored regression problems. Furthermore, consistency and asymptotic normality results are derived. Numerical results on simulated data sets are shown.