The Impact of Measurement Errors on Discrete Choice Models

Econometricians have known for almost a century that using variables subject to measurement errors in regression models always biases inference and frequently leads to inconsistent estimation. Nevertheless disaggregate modelers frequently use variables with substantial measurement error. Revealed preference discrete choice models all require information about non-chosen alternatives, and these data are frequently imputed with substantial error. This talk reviews work in economics on measurement error in income and employment status (a common exogenous variables in demand models), and then gives some results from attempts to model the measurement process in route choice and vehicle choice applications.