Determining the change in welfare estimates from introducing measurement error in non-linear choice models

Observed and unobserved characteristics of an individual are often used by researchers to explain choices over the provision of environmental goods. One means for identifying what is typically an unobserved characteristic, such as an attitude, is through some data reduction technique, such as factor analysis. However, the resultant variable represents the true attitude with measurement error, and hence, when included into a non-linear choice model, introduces bias in the model. There are well established methods to overcome this issue, which are seldom implemented. In an application to preferences over two water source alternatives for Perth in Western Australia, we use structural equation modeling within a discrete choice model to determine whether welfare measures are significantly impacted by ignoring measurement error in latent attitudes, and the advantage to policy makers from understanding what drives certain attitudes.