Abstract
Using administrative records on earnings, hours worked, and hourly wages for hourly employees of a single manufacturing firm, we found standard assumptions about errors in survey measures to be violated to varying extents. Errors are correlated with true scores, and errors in reports of earnings and hours in different periods are generally positively correlated with one another, particularly for reports given in a single interview. Although a high proportion of reporting errors are from an approximately normal distribution, a small proportion comes from a distribution with a much greater variance, and these cases often have considerable influence on estimates of relationships between variables.