Designing a Stated Preference Study for Analysis of Demand for High-speed Rail Using Efficient Choice Experiment Design

Designing a Stated Preference Study for Analysis of Demand for High-speed Rail Using Efficient Choice Experiment Design


A H Halse, S Fluegel, Institute of Transport Economics, NO


We present a stated preference study used to analyze choices beteeen high-speed train and other modes of transport. We demonstrate and discuss the implementation of efficient choice experiment design when attribute values are based on an actual trip.


The possibilities for building high-speed rail (HSR) networks are currently being investigated and discussed in several European countries. In Norway, there are plans for connecting the major cities through HSR, but no decisions have been made so far. In this abstract we present the methodology used in an independent study of the demand potential for HSR in two of the most relevant corridors.

As conventional trains in Norway run at a relatively slow speed, the introduction of HSR will imply a major reduction in travel times for those going by train. The high-speed service is also likely to be perceived as more modern and comfortable and travelers will hence probably regard it as a different mode than conventional rail. This implies that traditional transport models based on the existing modes of transport in Norway (car, airplane, train, bus) are not suitable for analyzing travel demand in a scenario where HSR exists. The analysis therefore has to be based on data both on existing travel demand and data on stated preferences for HSR.

Our study is a web survey among 889 Norwegian long distance travelers. The survey includes hypothetical choice experiments where respondents choose between high-speed train and their mode of travel (reference mode) on a trip which they have previously undertaken. The trip attributes are travel costs, in-vehicle travel time, access and egress time, frequency and the share of the trip spent in tunnels.

The attribute values of the reference mode alternative are based on actual trip characteristics reported by the respondents. The attributes of the high-speed train alternative are designed such that their values would be realistic for the trip in question if HSR existed. In the first choice experiment, all attributes of the reference mode alternative are fixed. In the second choice experiment, their values can differ from those in the current situation. In the two choice experiments there are eight and six repeated choice tasks, respectively.

We first conducted a pilot study among 217 of the respondents. In this study, the attribute values were combined randomly ("orthogonal design") and a large number of combinations were used. The estimated utility parameters from the choice model used to analyze this data were then used as prior parameter values to generate an efficient design with 12 different choice tasks in the first choice experiment and 12 different choice tasks in the second choice experiment. These are divided into three and two blocks of eight and six choice tasks which are distributed evenly to the respondents.

The designs are different depending on the current mode of travel and the trip purpose. The designs are generated not only based on the prior expectations of the utility parameters but also taking into account the average attribute values in both alternatives for each mode and trip purpose segment.

The efficient designs were tested in a second pilot study with 67 respondents. The prior parameter values from the choice model estimated on the data from both pilot studies were then used when generating the final efficient designs for the main study.

Our study has the important feature that the attribute values are different for each respondent and based on the reported characteristics of the individual reference trips. It is important to investigate how efficient designs perform in this context and which factors affect the efficiency. When generating the designs, we assume parameter values and attribute values representing a "typical" respondent. How much the respondents in our actual sample deviate from these assumptions with respect to both preferences and trip characteristics is likely to be important.

Furthermore, the design was generated to be efficient when estimating a multinomial logit (MNL) model with a certain utility specification where utility is linear in all attributes. It is hence of interest to evaluate how the design performs if we assume non-linear utility or/and estimate using a mixed logit model with random parameters.

We are able to evaluate the performance of the experiment design in several ways. One is to compare the results from the pilot study with orthogonal design with the results from the main study. Another is to compare results from the main study for different segments. We also give some recommendations for future studies of this kind.


Association for European Transport