Experimental results from the test on the new Stated Preferences valuation method, the Continuous Attribute-Based Stated Choice, in the web based surveys

In the last fifty years the transport research has spent many millions dollars to collect data for transport planning and for developing and using travel-demand models based on such data. This fact emphasizes a crucial aspect that is the need of defining standards and guidelines for the survey design as also of harmonizing the collected data to obtain a higher surveys’ quality level. The research purpose is to provide a first insight on the standards to be adopted at international level to obtain comparable and high quality data. The authors’ purpose is to report results from one out of the ten tests based upon experimental comparisons in which one or more of the verbal, numerical, graphical and symbolic languages are manipulated to determine whether answers to survey questions were affected by these manipulations. At first time a qualitative pre-test was developed on a group of 53 people. A self-administered questionnaire has been given out to the group of people and the researcher was present to guarantee the right drawing up and to avoid that people go up and down through the questionnaire and correct the answers, as also for observing the respondents problems during the compilation. The respondents were people working at the Transport Department of the Politecnico di Torino and students following the courses of Transport Planning (thus, a sample familiar with the subject). Afterwards, on the basis of the results obtained by pre-test, a web-based survey was designed, including a test on the on the Stated Preference (SP) method, and given out to the reference population working or studying at the Politecnico di Torino (3,420 employees and 18,861 students); 2840 persons answered, equal to the 13% of the universe. The framework of the questionnaire is based on the investigation of the different forms of languages, thus the following topics have been faced: scalar questions, skip instructions, reversing the order of answer categories, check-all-that-apply format and Stated Preference (SP) survey. In this paper, only the last topic, concerning the SP method is presented. The innovation in the tests’ design is that to each unit all the versions of the tests have been given, ad hoc distributed along the questionnaire, so that a comparison among the answers of each person has been possible. In addition, the tests have been done in the same time to investigate if the changes of the versions could to influence the respondents in a so short time range. Differently, in the previous researches, different tests’ versions were administered to each group of people and the average values of the different groups’ answers were compared. In our case, each unit represents a reliable result concerning the people understanding of the question. This aspect can be really innovative allowing to define some standards with a minor number of experiments. The defect of this method is an increasing tiredness of the respondents. The issue of question complexity in SP studies has been the subject of debate for some time. In a typical stated choice format, the respondent chooses from a set of two or more options; each option differs in its levels of the quality attributes and in the cost to the respondent. A respondent will often make eight or more of these choices. Based on the stated choices, marginal utility of each of the attributes is estimated using a random utility model. With these estimated marginal utilities, it is possible to calculate willingness to pay (WTP) for any proposed change. A limitation of the stated choice format is that it too generates discrete data, in the sense that each stated choice reveals only ordinal preferences among the options presented in that choice. Many stated choices from each respondent and a large sample size of respondents are needed to identify marginal utility for each of the attributes. The purpose of the proposed test is to develop a new elicitation method, called the continuous attribute-based stated choice method that values attributes but collects continuous data. In an interactive computer-based survey, respondents choose levels of each attribute. The total cost of the package of attributes is continuously updated according to a cost function selected by the researcher. Thus, on the screen, respondents can choose any level of time of trip from -40% to + 40% of their time, any level of set availability from always and never, any level of comfort from better and worse, and any level of punctuality from guaranteed to the minute to indefinite delay. As sliders are moved, ticket’s cost moves with them according to a pre-programmed cost function. Each respondent’s choice reveals his or her marginal WTP for each attribute. In our large sample (n=2840) internal validity tests have been conducted for the continuous attribute-based stated choice method and, in the same sample, responses to the continuous attribute-based stated choice method have been compared to responses from a more traditional discrete attribute-based stated choice method survey. In Continuous Attribute-Based Stated Choice, respondents are reacting to the cost function, a key assumption behind the method. However, responses were sensitive to the range of opportunities presented. It does raise questions about whether respondents are reacting to absolute levels of attributes and cost or relative levels. The comparison between the Continuous Attribute-Based Stated Choice survey method and the more traditional one showed that the two methods generate similar results. This fact is somewhat surprising. There does not appear to be a similar systematic difference between how respondents answer Continuous Attribute-Based Stated Choice questions and how they answer SP questions. Comparing each respondents’ choice in the two methods, they do not show, on average, contradictions in their responses. The frequency of reversals in the SP method creates about the 7% of invalid answers caused by the respondents’ difficulty to recognize the differences among the various scenarios. The results allow to consider the Continuous Attribute-Based Stated Choice a promising method for preference elicitation as the tasks are easier to be completed, each response generates more information about the preferences than a single SP response and the respondents can complete more quickly the survey. The method does need to be administered on a computer, though it can be done through the web. Possible further endeavours are the variation of the cost function, identifying a marginal WTP function varying with the attribute levels. Interaction among the attributes in the utility function can also be identified.