Overcoming Measurement Error Problems in the Use of Survey Data on Expectations

Abstract

Survey data frequently requires conversion from qualitative responses to quantitative series and it is often asserted that the conversion procedures introduce measurement errors that render the series unusable in structural modelling. We investigate the nature and treatment of the measurement error that arises when conversion procedures are used to obtain direct measures of expectations. We use simulation experiments to demonstrate the need for the adequate treatment of conversion errors and show that a procedure proposed to 'purge' the series of conversion error is extremely successful in circumventing the problems. We illustrate the procedures in empirical applications using business survey data.