The Consequences of Non-Classical Measurement Error for Distributional Analysis

This paper analyzes the consequences of non-classical measurement error for distributional analysis. We show that for a popular set of distributions negative correlation between the measurement error (u) and the true value (y) may reduce the bias in the estimated distribution at every value of y. For other distributions the impact of non-classical measurement di¤ers throughout the support of the distribution. We illustrate the practical importance of these results using models of unemployment duration and income.