It is well known that residuals from stated preference (SP) experiments are not distributed as are residuals from revealed preference studies, which reflect actual behaviour. That means that SP parameters usually need rescaling before they can be used in forecasting. However, it is less well understood that estimates of the unknown parameters in the deterministic part of the SP model will also be biased in certain circumstances. Examples of these include: inertia, where starting point bias will favour the (exact) current position if offered as an alternative to a changed position; small time savings, which theoretically should be valued at the same unit value as large time savings but, which are observed to have low unit values in SP experiments (presumably because respondents cannot think how they would benefit from them); and gains versus losses, where it is quite clear that changes must be reversible, but results from SP experiments may seem tocontradict that. These effects are analysed in relation to value of time estimation. Illustrative utility functions are constructed, and displayed both mathematically and graphically. Bounds for slopes and curvature are set. Implications for short term SP responses are derived, and compared to the long run assumptions that underlie the analysis. Strategies for avoiding these resulting difficulties will then be proposed. In brief, these include (a) avoiding setting one alternative in a SP choice set equal to the exact current position; (b) avoiding estimating unit values of travel timesavings (or losses) from analysis of SP choices (only) involving small travel time differences; and (c) avoiding, where possible, offering SP alternatives all slower (or all faster) than the current travel time even if any proposed change is known to be towards slower (or faster) travel times. Examples from actual studies will be provided and discussed. For the covering abstract see ITRD E145999
Abstract