Discussion Papers, Department of Economics, Universität Bern 13-04
This paper studies the identification of coefficients in generalized linear predictors where the outcome variable suffers from non-classical measurement errors. Combining a mixture model of data errors with the bounding procedure proposed by Stoye (2007), I derive bounds on the coefficient vector under different non-parametric assumptions about the structure of the measurement error. The method is illustrated by analyzing a simple earnings equation.
Generalized linear predictor Non-classical measurement error Contaminated sampling Corrupt sampling Multiplicative mean independence Stochastic dominance Nonparametric bounds