• Federal Highway Administration >
• Publications >
• Research Publications >
• 98166 >
• Sec2.6.Cfm >
• Guidebook on Methods to Estimate Non-Motorized Travel: Supporting Documentation

Guidebook on Methods to Estimate Non-Motorized Travel: Supporting Documentation

2.6 Discrete Choice Models: Route Choice

Descriptive Criteria: What is It?

Categories:

Bicycle

Pedestrian

Facility-Level

Area-Level

Authors and Development Dates:

Bovy and Bradley (1985); Hopkinson and Wardman (1996); Hunt and Abraham (1997); Hyodo, Suzuki, and Takahashi (1998).

Purpose:

Discrete choice modeling techniques can be applied to predicting bicycle route choice as well as mode choice. Discrete route choice models have a number of possible applications:

• Determining the relative preferences of bicyclists for different route characteristics, e.g., separate path, bicycle lanes, or mixed traffic. One advantage of discrete choice models over other methods is that the tradeoffs between attributes can be quantified (for example, a change in pavement quality from "fair" to "good" can be equated to a travel time improvement of X minutes).

• Developing elasticities, which can be used to relate the change in a particular factor to the expected percent change in number of users.

• Predicting actual route choice on a bicycle or pedestrian network. The output would be the distribution of trips over the network, given a set of origins and destinations. A route choice model developed by Hunt and Abraham(1997) is being applied to a network in Edmonton, with the ratios among coefficients for each facility type being used to weight travel time in the mode choice submodel (see "Regional Travel Models" discussion). Discrete route choice models may be a key element in future development of bicycle demand forecasting models, particularly in developing models which predict both mode and route choice as a function of route characteristics.

Structure:

Three of the four references reviewed here use a logit model to predict route choice as a function of route characteristics and other factors. The resulting coefficients on the route characteristics are used to compare the relative importance of these characteristics.

Hyodo, Suzuki, and Takahashi (1998) use a slightly different approach. Bicyclists are surveyed and asked to map their trips on a road network. The frequency of actual trips on each link is compared to the frequency of trips under a shortest-distance path assignment, and parameters which affect the "cognitive travel time" are estimated based on facility design factors. The authors include sidewalk width (under or over 2.5 m) and street type (dummy for a shopping arcade closed to traffic) as facility design factors, although other factors could be included if data were available. The "genetic algorithm" method, a specialized technique for parameter estimation, is used to estimate parameter values. The authors' method is unique in that it uses revealed-preference data (observed trip routes) and link-specific characteristics to derive parameters which can be used to include facility characteristics in a route choice model. However, it also points out the computational and methodological difficulties in estimating more than one or two parameters. While including design factors improves the model fit compared to a simple shortest-path assignment, the method has not yet been developed for use in demand forecasting.

Calibration/Validation Approach:

If the route choice models were applied to predicting network flow distributions, predicted distributions could be compared to actual flows as determined from counts.

Inputs/Data Needs:

The four models reviewed here are based on data from stated-preference surveys of bicyclists. Respondents are asked to choose between pairs of hypothetical bicycling links with specified attributes.

Potential Data Sources:

Development of a discrete route choice model usually requires stated-preference survey data. Revealed-preference data could also be used but would require extensive real-world data collection on facility characteristics and user trip patterns, although GIS techniques and data bases may make this easier in the future. Estimation of coefficients using revealed-preference data also presents some technical problems (see Hyodo, Suzuki, and Takahashi, 1998). Other problems with the use of revealed-preference data are discussed in Hunt and Abraham (1997).

Computational Requirements:

See general entry for "Discrete Choice Models," Method 2.5.

User Skill/Knowledge:

Requires knowledge of discrete choice modeling and stated-preference survey techniques.

Assumptions:

See general entry for "Discrete Choice Models," Method 2.5.

Facility Design Factors:

Bovy and Bradley:

• Facility type (physically separated, reserved on-street, non-existent).
• Surface quality (smooth, moderate, rough).
• Traffic level (light, moderate, heavy).
• These three design factors are traded off against travel time (9, 12, or 15 minutes).

Hunt and Abraham:

• Secure parking.
• Availability of showers.
• Facility type (mixed traffic, bike lanes, or bike paths shared with pedestrians). The interactions of facility type variables with the experience and comfort levels of the bicyclist are given particular attention.

Hyodo, Suzuki, and Takahashi:

• Sidewalk width; and
• Dummy variable for a pedestrian mall.

Output Types:

Outputs include the relative importance of various route attributes.

Real-World Examples:

Bovy and Bradley (1985) used stated-preference surveys to develop a discrete route choice model. Route factors included facility type, surface quality, traffic level, and travel time (each described qualitatively at three levels).

Hopkinson and Wardman (1996) used stated-preference techniques to obtain valuations of improvements to cycle facilities, forecast the effects of such facilities on route choice, and provide a partial cost-benefit analysis of alternate bicycle routes.

Hunt and Abraham (1997) developed a discrete route choice model based on a hypothetical-choice stated-preference survey of cyclists in Edmonton, Canada. Facility factors included time spent cycling on three different facility types and the availability of showers and secure bicycle parking. Socioeconomic data and indicators of experience and comfort level were also used in model development.

 

Hyodo, Suzuki, and Takahashi (1998) proposed a bicycle route choice model in which facility characteristics (e.g., road width or sidewalk) affect the impedance function in route choice. Development of the model was based on a survey of bicyclists in which they were asked to map their trip on a network. Parameters were estimated based on actual vs. minimum-path routes, using the Genetic Algorithm method.

Contacts/Source:

John Abraham: T.J. Modeling, Calgary, Alberta, Canada.

John Hunt: University of Calgary, Department of Civil Engineering, Calgary, Alberta, Canada.

Tetsuro Hyodo: Tokyo University of Mercantile Marine, Tokyo, Japan.

Publications:

Bovy, Piet H.L. and Mark A. Bradley. Route Choice Analyzed with Stated-Preference Approaches. Transportation Research Record 1037 (1985).

Hopkinson, P. and M. Wardman. Evaluating the Demand for New Bicycle Facilities. Transport Policy, Vol. 3 (1996).

Hunt, J.D. and J.E. Abraham. Influences on Bicycle Use. Submitted for presentation at the 1998 Transportation Research Board Annual Meeting, July 1997.

Hyodo, Tetsuro; Norikazu Suzuki and Yoji Takahashi. Modeling Bicycle Route Choice Behavior on Describing Bicycle Road Network in Urban Area. Presented at the 1998 Transportation Research Board Annual Meeting, Paper #980353, January 1998.

Evaluative Criteria: How Does It Work?

Performance:

No information is available.

Use of Existing Resources:

Survey and modeling efforts are required.

Travel Demand Model Integration:

Route choice models have not widely been integrated with travel demand models. However, bicycle route choice models could theoretically be included in the traffic assignment step. Hunt and Abraham note that their route choice model is being applied to the development of a network-based travel demand model in Edmonton, with the ratios among coefficients used to develop a utility function for bicycling in the mode choice submodel.

Discrete route choice models may be a key element in future development of bicycle demand forecasting models, such as in developing combined models which predict both mode and route choice as a function of route characteristics.

Applicability to Diverse Conditions:

The results of current route choice models have been based on generic, hypothetical route characteristics and thus should be applicable to various locations and conditions. Nevertheless, the survey responses may to some extent have been conditioned by the environment with which the respondents are familiar, so transferring results should be done with caution.

The validity of these models can be assessed by comparing the relative preference results to results obtained from other studies. Bovy and Bradley found consistency between their results and earlier studies by Axhausen (1984), and also found a reasonable amount of internal consistency comparing the results of different evaluation methods used in their study.

Usage in Decision-Making:

No information is available.

Ability to Incorporate Changes:

See entry for "Discrete Choice Models," Method 2.5.

Ease-of-Use:

Generally requires survey efforts and a knowledge of discrete choice modeling.

FHWA-RD-98-166