Interactions and Independence in Stated Preference Modelling



Interactions and Independence in Stated Preference Modelling

Authors

ORTUZAR J de D, RONCAGLIOLO D A and VELARDE U C, Pontifieia Universidad Catolica de Chile, Chile

Description

Stated preference (SP) techniques have been probably the fastest groMng phenomenon in the travel demand forecasting field. Earlier concerns about the validity and inherent structural bias of the approach have diminished with the advent of mixed revealed (

Abstract

Stated preference (SP) techniques have been probably the fastest groMng phenomenon in the travel demand forecasting field. Earlier concerns about the validity and inherent structural bias of the approach have diminished with the advent of mixed revealed (RP)/SP modelling; nowadays most agencies happily hire masses of consultants ready to study almost any other problem with the aid of SP techniques.

However, research on the subject shows that many issues surrounding the experimental design, collection of SP data, and indeed the modelling of demand using such data, is plagued with unsatisfactorily solved problems. It is worrying that the client agencies seem not to be aware of these; it is also worrying that there is no way of certifying who knows how and who just does it without much care for quality. Unfortunately the results of the method always appear to be right so it is not easy to separate wheat from straw.

In this paper we consider two problems related to SP data alone. The first one concems a potential though seldom realised advantage of the approach, namely the possibility of estimating models with non-linear utility functions. The reason for not doing this in practice has been, as usual, one of convenience. SP experiments allowing for interaction terms and not just main effects are harder to design, collect data for and estimate. The second problem lies at the heart of the approach's attractiveness (i.e. the generation of multiple responses per interviewer) and concerns the fact that almost all practical applications to date consider the SP responses by a single individual as independent of each other as those given by the rest of the sample. This problem has been receiving more attention lately although it is still not well known outside the inner priesthood of knowledge, neither has it yet been satisfactorily solved.

The rest of the paper is organised as follows. In section 2 we present these problems formally and review what has been done or proposed so far in the literature to tackle them. In section 3 we briefly describe some data, either available or specifically collected to study these problems. We actually designed a small test bed to examine the problem of interactions, as data is not forthcoming in practice for this purpose; however, we used a series of different samples available from previous studies to test various ways to solve the problem of independence. In Section 4 we present some modelling results using these data sets, and in section 5 we summarise our main conclusions.

Publisher

Association for European Transport