| Summary: | Stated Preference (SP) methods have been used extensively in transport research and elsewhere both for demand forecasting purposes and to value the importance attached to different product features and travel attributes. Alongside the broader acceptance and wider application of SP methods, some practitioners (Bates, 1998; Ampt et al., 2000; Wardman and Shires, 2001) have argued for greater openness in discussing what they see as significant concerns surrounding SP. The present study is motivated by the desire to analyse and reduce biases in the SP application, specifically addressing the issue of the strategic biasing of SP responses. The review of biases observed in the previous SP applications explored the sources of bias, which can be categorized as unrealistic design, incentive to strategic bias and task complexity effects. Amongst these, the issues of design/scenarios specification and task complexity have received a considerable amount of attention. On the other hand, and despite serious concerns in the early literature, the strategic biasing of responses tends to have been overlooked in recent times, particularly within the SP methodology. This study is motivated by the desire to investigate the incentives for respondents to bias their answer in the SP survey and methods to amend the bias. This study reviewed and summarised concerns surrounding the extent to which the SP responses to hypothetical questions reliably reflect individuals' true preferences when there is an incentive to bias responses. The discussion was illustrated with examples from research in transport field, environment science and marketing. In an empirical demonstration using data obtained from 1222 respondents (10885 preference observations) on the valuation of the improved rolling stock in Greater Manchester, UK, this study presented results for different designs. Based on the review of studies on rolling stock in recent years, a suite of SP experiments were designed to investigate the effects of different designs on responses. Two factors were introduced into the experiment, a `cheap-talk' script and `adding more attributes to mask the research aim', to amend incentives to bias. In the experiment, post-questionnaire questions on respondents' perception of experiments were introduced. More specifically, respondents' perceptions of the task load, familiarity of experiment alternatives together with their perceptions of the attribute change were added to probe the decision making process and the impact of perception on the decision making. Standard logit models were used to demonstrate the overall effects of variables for the whole sample. The segmentation model, based on the incremental factors, was used to identify respondents' taste variations. The heteroskedastic multinomial logit (HMNL) model was used to incorporate the impact of design factors, respondents' characteristics and perceptions into the scale parameter, which were unable to be captured by the standard logit model. This study found that the cheap-talk script decreased the valuation of the improved rolling stock by 20% on average, through increasing respondents' sensitivity to the cost attribute in the SP survey. However, this impact was not significant at the 5% significance level. This indicates that the warning message will help individuals to amend the incentive to strategic bias in the SP experiment; however bias may remain in our study. This study did not detect significant impact of the complex design on the valuation of the improved rolling stock, although task complexity effects were detected where a large error variance was found in the complex SP design. Individuals' perceptions have significant impacts on the valuation and model estimation precision. Individuals' familiarity with alternatives in the experiment increased the value of the improved rolling stock and improved the estimation precision. Individuals' perceptions of potential price increase have an impact on the valuation and estimation precision. The more likely respondents perceived the potential price increase, the fewer preferences were given to the improved rolling stock and respondents were observed to be more consistent in their choice making. In brief, this study suggests that incentives to strategic bias exist in the SP experiment due to its hypothetical nature. Warning message such as a CT script is helpful to amend individuals' incentive to strategic bias. Attention should be made to the complexity of the experiment, as respondents are subjected to certain cognitive ability. In the SP analysis, individuals' perceptions can be incorporated into the model analysis.
|
|---|
