Nonparametric Berkson regression under normal measurement error and bounded design

Regression data often suffer from the so-called Berkson measurement error which contaminates the design variables. Conventional nonparametric approaches to this errors-in-variables problem usually require rather strong conditions on the support of the design density and that of the contaminated regression function, which seem unrealistic in many cases. In the current note, we introduce a novel nonparametric regression estimator, which is able to identify the regression function on the whole real line under normal Berkson error although the location of the design variables is restricted to some bounded interval. The asymptotic properties of this estimator are investigated and some numerical simulations are provided.