A0487
Title: Estimating nonparametric Berkson measurement error models
Authors: Ji-Liang Shiu - Jinan University (China) [presenting]
Abstract: Non-parametric identification and estimation of Berkson measurement error models with conditional mean independent regression error is shown. The identification result is achieved without requiring side information and specifying the distribution of the measurement error. We first apply techniques of integral operators associated with the distributions of observable and unobservable variables to identify the moments of the measurement error. Based on the identification of the moments, the moments of the regression function can be uniquely determined.Additional identification restrictions include conditional distribution $f_{Y|X*}$ is complete and non-vanishing Fourier transforms of the regression function. We then use the identification result to construct a sieve minimum distance (MD) estimator to estimate the regression function and the distribution of the measurement error. We investigate the finite sample properties of the proposed sieve MD estimator through a Monte Carlo study.
