Stated Preference (SP) experiments are an established and useful method for gaining insights into an individual's utility or worth of different transport characteristics. Whilst interest has been directed at improving the efficiency and accuracy of such experiments, in practice the design is still most frequently based on assumptions rooted in standard linear regression theory. The most basic and frequently applied assumption is that SP design should be orthogonal. Previous work in this area reported at PTRC has contended that even using a regression analogy, orthogonality may be sub optimal. As part of an EPSRC funded project investigating the theoretical basis of SP work, exact expressions for the variance of logit model parameters and the variance of the parameter ratio (of Value of Time) have been derived. Using these, experiments similar to those previously carried out using the regression analogy, but based on the more appropriate logit model theory, have been conducted. Comparisons between a traditional orthogonal design and alternatives containing varying degrees of both negative and positive correlation have shown substantial reductions in variance for both parameter estimates and their ratio using non-orthogonal designs, thus challenging one of the most fundamental assumed aspects of SP design. The variance formula also highlight other interesting characteristics of SP design which are discussed here.
Abstract