ABSTRACT
The conventional approach to regression modeling is to assume that explanatory variables are measured without error. The defining feature of an errors-in-variables problem is that rather than observing the explanatory variables without error, at least one is measured as an error-prone "surrogate". This situation is also sometimes referred to as a measurement error problem. Following Carroll, Ruppert and Stefanski (1995) we use X to denote the true value of the variable, W the surrogate and Z explanatory variables that are measured without error; Y will denote the response.
