Empirical Analysis of Hypothetical Bias in Stated-Preference Experiments

The stated-preference (SP) survey is widely used to obtain data of choice prediction, especially for new transport service systems. However, the hypothetical nature of SP experiment determines that it is usually taken under a 'what-if' situation. The difference between the 'stated' and 'real' responses is named as the hypothetical bias (HB), which may further lead to great errors of the following model analysis. While HB has been largely confirmed in the area of economics, few studies have proved the existence of HB in transportation planning, not to mention measuring it. This paper presents a new viewpoint to study HB problem in terms of the mode split prediction. An SP survey and a revealed-preference (RP) survey were conducted before and after the opening of a new metro line in Chengdu, China. To reduce requirements on high accurate sampling, a selection-calibrationsimulation method is proposed to transfer the individual choice to aggregate market share of alternatives. The multinomial logit (MNL) is selected as the basic model structure in this paper. By calibrating two separate MNL models using SP and RP data, we ran the simulation under a uniform input. The result presents the valid evidence on HB's existence and also indicates the possible direction of deviating it. Furthermore, an improved value of time (VOT) model that can accommodate multiple time and cost variables is proposed. The calculation results of VOT confirm the simulation result and demonstrate how large HB is.