U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-RD-98-166
Date: July 1999
Guidebook on Methods to Estimate Non-Motorized Travel: Supporting Documentation
2.2 Aggregate Behavior Studies
Descriptive Criteria: What is It?
Authors and Development Dates:
Ashley and Banister (1989); Epperson, Hendricks, and York (1995); Ridgway (1995); Nelson and Allen (1997).
Aggregate behavior studies or models attempt to predict mode split and/or other travel behavior characteristics for an aggregate population, such as residents of a census tract or metropolitan area. Prediction is based on characteristics of the population and of the area. An example of an aggregate model would be a regression equation to predict the bicycle mode splits of individual census tracts in a metropolitan area, based on the average income of the tract and on the total length of bikeways in the tract. Aggregate behavior models can be contrasted with disaggregate models, which predict an individual's behavior and then aggregate individual decisions across a population to obtain overall travel characteristics.
Aggregate models can be used for the following purposes:
1. Identifying which factors influence overall levels of bicycling or walking in an area.
2. Predicting the change in levels of bicycling or walking caused by a change in one of these factors.
3. Predicting the amount of bicycling or walking in other areas, based on data collected in one area.
4. Developing data for use in a travel demand model.
Linear regression equations are commonly used to predict an independent variable (bicycle mode split, number of trips, etc.) as a function of various dependent variables.
A model can be developed based on one dataset and then applied to another dataset to check its validity. Attempts to do this, however, have yielded less-than-satisfactory results (c.f. Ashley and Banister, 1989; Ridgway, 1995).
All data must be obtained at the level of the unit of analysis (census tract, employment center, metropolitan area, etc.) A wide variety of data can be used in developing aggregate behavior models. Both the data used and the unit of analysis are generally constrained by what data can be obtained from available sources or collected with little additional effort.
Ashley and Banister obtained data at the ward level in the United Kingdom on characteristics of the population, trip distances, availability of other modes, traffic levels, and local climate/topographical factors. Some data were obtained from census records while other data required additional collection and analysis efforts. They also identified a number of variables that were desirable to have but could not be collected because of resource limitations.
Potential Data Sources:
Census Data: Population characteristics (socioeconomic and demographic), journey-to-work mode, density
Land use data bases
Topographic maps: topography
Roadway network data bases: traffic volumes, road characteristics
Analysis can be conducted with spreadsheets or standard statistical software packages.
An ability to construct statistical models such as linear regression is required.
It is assumed that travel behavior at an aggregate level can be predicted with relative accuracy given the data available. The implications of this assumption are discussed under "Comments."
Facility Design Factors:
Ashley and Banister considered terrain (hilliness) and traffic levels. Availability of bicycling facilities and terminal facilities were not included because of lack of data.
Inclusion of facility design factors in aggregate demand models would require measures of facility availability/quality which can be constructed at the area level. These might include miles of bike path or lane, miles of sidewalk, percent of road network in good cyclable condition, etc. Further development of road/facility network data bases using GIS techniques should allow easier incorporation of facility design factors. Pedestrian environment factors, such as those developed in Portland, OR, are an example of area-level facility design variables.
Nelson and Allen included per capita miles of bikeway in an analysis of work-trip bicycle use at the metropolitan area level.
Output is mode split or total trips by mode for an area as a function of variables describing the area.
Ashley and Banister (1989) used UK census data to (1) evaluate factors influencing bicycling to work; (2) develop a model to predict the proportion of residents bicycling to work; and (3) test the model. A variety of factors were tested including personal characteristics, trip distance, availability of bicycling facilities, availability of other modes, traffic levels, and local climate/topographical factors.
Epperson, Hendricks, and York (1995) analyzed NPTS data to develop nationwide bicycle trip generation rates for 12 categories of people (stratified by age, gender, and race). These trip rates were applied to census tracts based on the number of people in each category by tract.
Nelson and Allen (1997) conducted a cross-sectional analysis of 18 U.S. cities to predict work trip bicycle mode split (from census data) based on weather, terrain, number of college students, and per capita miles of bikeway facilities. A positive association was found between the presence of bikeway facilities and bicycle work trip mode split.
Ridgway (1995) developed a regression model to estimate bicycle mode split at the city and census tract levels based on available data. Candidate variables were screened based on correlation with bicycle mode split. Those selected included age (percent of population under 25 years), and mean population travel time (a proxy for travel distance), and percent of student population.
Chris Banister: Department of Planning and Landscape, University of Manchester, UK.
Bruce Epperson: Miami Metropolitan Planning Organization, Hollywood, FL.
Matthew Ridgway: Fehr and Peers Associates, Lafayette, CA.
Ashley, Carol A. and Chris Banister. Bicycling to Work from Wards in a Metropolitan Area. Traffic Engineering and Control, Vol. 30, nos. 6-8, June - September 1989.
Epperson, Bruce, Sara J. Hendricks, and Mitchell York. Estimation of Bicycle Transportation Demand from Limited Data. (University of South Florida). Compendium of Technical Papers from the Institute of Transportation Engineers 65th Annual Meeting, pp. 436-440, 1995.
Nelson, Arthur C. and David Allen. If You Build Them, Commuters Will Use Them: Cross-Sectional Analysis of Commuters and Bicycle Facilities. City Planning Program, Georgia Institute of Technology, submitted to the Transportation Research Board, 76th Annual Meeting, Washington, DC (preprint), January 1997.
Ridgway, Matthew D. Projecting Bicycle Demand: An Application of Travel Demand Modeling Techniques to Bicycles. 1995 Compendium of Technical Papers, Institute of Transportation Engineers 65th Annual Meeting, pp. 755-785, 1995.
Evaluative Criteria: How Does It Work?
Aggregate demand models to predict bicycling and walking mode shares tend to have low-explanatory power; that is, most of the factors which influence mode shares have not been accounted for in the model.
Ashley and Banister found that "while it is possible to isolate some factors in the form of a model for particular areas, when the model is applied elsewhere the fit is not so good." Also there are significant difficulties involved with developing a transferable model.
Ridgway found that while his model based on census data adequately predicted bicycle mode split using data from 18 California cities, it did not perform so well at predicting mode split for census tracts in Berkeley.
Use of Existing Resources:
Aggregate models can be constructed largely using existing data on population and land use characteristics. Aggregate-level data on network characteristics may require additional data collection and analysis, although further development of road/facility network data bases using GIS techniques should allow easier incorporation of facility design factor.
Travel Demand Model Integration:
Aggregate models are frequently used in the travel modeling process to predict total number of trips by trip purpose at the zonal level. The models discussed here differ primarily in that they attempt to predict only total bike or walk trips, rather than total trips by all modes. In travel demand models, mode choice is usually predicted separately at a later stage of the travel modeling process.
Applicability to Diverse Conditions:
Aggregate models have not yet been developed which have been demonstrated to be transferable to other situations or areas.
Usage in Decision-Making:
No information is available.
Ability to Incorporate Changes:
Models can be re-estimated with relative ease if new data become available.
An ability to construct statistical models such as linear regression is required.
It is assumed that travel behavior at an aggregate level can be predicted with relative accuracy given the data available. Some of the drawbacks of this assumption include: