When we fit statistical demand functions to data about individuals, we are limited by the degree of modelling error present. Such modelling error might derive, for example, from day to day variation in individual behaviour, variation in behaviour over the population, fallibility in respondents' memories when answering our questions and the simplification involved when we concentrate only on the main influences on quantity demanded, in particular price. This paper reports on theoretical analysis of situations where modelling error has had undesirable results. One example to be considered is the often noted finding of asymmetric response, for example when asking travellers how they would react to improvements as against worsenments of service quality, or fare increases as against fare decreases. In Contingent Valuations Methodology this even has its own terminology - `Willingness to Pay' and against `Willingness to Accept'. This paper argues that too little attention is paid to the influence of modelling errors on these findings. A second example studied is from the area of consumer surplus calculations in transport models. Here the demand functions for new modes might be adequately estimated for the purpose of demand modelling, but not for the purpose of consumer surplus estimation, where the quantity demand at very high price is very critical to the calculation, but which is very poorly estimated by logit models. (A)
Samenvatting