U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
|< Previous||Table of Contents||Next >|
Step 5 in the scalable risk assessment process is to select an analytic method (or methods) to estimate exposure. There are numerous analytic methods that can be used to estimate exposure, and the most appropriate method(s) depends upon several criteria, such as desired geographic scale (Step 2), desired exposure measures (Step 4), analysis scope, data availability, staff technical capabilities, and available analysis resources.
As indicated in Figure 1, there may be some iteration or concurrency between this step (selecting an analytic method) and the previous step (Step 4, selecting an exposure measure). For example, an analyst may want to calculate a specific exposure measure, but have no expertise in the most common analytic methods used to estimate that exposure measure. Or, a specific analytic method may not be able to accurately estimate a specific of exposure measure. Therefore, many analysts are likely to consider both the desired exposure measure and the most feasible analytic method in a concurrent or iterative manner.
This chapter introduces these analytic methods and provides guidance on selecting the most appropriate method(s) given various criteria. The analytic methods outlined in the guide are listed below and briefly described in the following pages.
Table 14 provides a method selection matrix to help analysts make informed choices about which analytic method(s) is best suited for them. It is important to note that local customization may be required for all these models to be useable.
|Analytic Method||Input Data Requirements||Technical Complexity||Popularity in Practice||Direct Usability||Accuracy|
|Demand Estimation Models||Direct demand models||/||/|
|Trip generation and flow models||/||/||/|
|Discrete choice models||/||/||/|
|Simulation-based traffic models|
Legend: = low suitability; = moderate suitability; = high suitability.
Note: For some categories, multiple ranges (e.g., /) are used since the corresponding criteria might vary significantly based on the specific characteristics of the model developed.
Site counts are a direct measurement of the number of pedestrians or bicyclists at a defined location for a defined time period. The counts may be gathered automatically from various technology-based sensors or manually from human observers. Site counts are taken at a point, but in some cases are applied along a street segment where the counts are not expected to vary along the segment length. Counts are more commonly used to estimate exposure when the desired facility coverage is limited, as count data collection for all facilities within a large network or region is cost-prohibitive (unless extensive sampling is used). In some cases, it is also a challenge to get representative pedestrian and bicyclist counts due to seasonal variation with these modes.
There are numerous estimation models that have been used to estimate pedestrian and bicyclist demand (i.e., counts) on specific facilities. The models range in complexity and input requirements, and some have been more commonly used than others. Several of the models rely on pedestrian and bicyclist count data for model development or calibration, representing somewhat of a hybrid approach that combines counts and models. In addition, while most of these models provide the volume estimate directly (e.g. direct demand models), some must be integrated with other methodologies to provide demand estimates (e.g. route choice models). The descriptions of the latter types of models have also been included in this step due to their potential role as non-motorized planning tools that can be used in exposure estimation.
Important to note that the models described here have an established history and been applied frequently in various fields. However, only a few of them have been used extensively in the context of estimating exposure for safety analysis, and the remaining models have been used infrequently or on a very limited basis. Direct demand models are noteworthy to mention since they played a major role in pedestrian and bicyclist volume estimation especially in supporting traffic safety studies. However, an overview of all potential models is provided here for completeness. Regardless of the model chosen, the analyst should keep in mind that the models might need to be customized and calibrated with respect to the local characteristics of the project (e.g. study area). The analysts are recommended to review the key considerations provided later in this section before model selection. This will help choose a feasible model that will provide an optimal solution with acceptable accuracy.
This section of Step 5 first provides an overview of the demand estimation models. A list of resource documents is then provided for analysts who are interested in learning more about these models. The section is concluded with discussions on key considerations when selecting a model.
Direct demand models are statistical models that estimate facility-specific pedestrian and bicyclist volumes based on observed volumes at a sample of locations and nearby context (such as land use and form, street type, etc.). Direct demand models are often based on regression analysis. These models are commonly used and are appealing due to their simplicity; however, they are limited in terms of capturing the underlying behaviors and travel patterns that produce higher and lower volumes in particular locations. Many of the existing direct demand models are also based on relatively small sample sizes. A detailed discussion on direct demand models and step-by-step instructions to develop them are provided in Step 6 given their practicality and widely use in the literature for pedestrian and bicyclist volume estimation modeling.
Regional TDM are computerized systematic processes that estimates existing and future travel demand at a regional basis given numerous inputs, such as the transportation network, population and demographic characteristics, and trip-making behavior. The end result of a regional travel demand model is traffic volume estimates on individual transportation network links, but several other model outputs can also be obtained, such as vehicle miles of travel, mode shares, origin, and destination of trips. Regional TDM constitute part of long-range transportation plans developed for MPOs or state or local agencies. Traditional TDM and enhanced models of tour- and activity-based are the most commonly known model structures of this category.
Traditional Regional Travel Demand Models. Traditional regional TDM are the current state-of-practice models for regional travel demand forecasting, which are mostly based on trip-based approach. The trip-based models generally consist of four main steps: trip generation, trip distribution, mode share, and traffic assignment. TAZs are the most commonly used geographic units to inventory existing and future demographic data required for modeling purposes. Therefore, trip-based models are particularly limited in estimating non-motorized travel due to their coarse level of spatial analysis structure. To overcome these limitations, several enhancements have been made to trip-based models over the last recent decades. Such enhancements can also be beneficial to increase the sensitivity of trip-based models to pedestrian and bicyclist trips, such as developing an enhanced trip generation model sensitive to land use factors or an enhanced auto ownership model as input to non-motorized trip production. On the other hand, these models are not yet adequate in estimating exposure for bicycle and pedestrian travel in various ways, such as in capturing the fine-grained differences in intersection-level bicycle or pedestrian activity or in capturing recreation trip purpose which is a key consideration in pedestrian and bicyclist trip rates.
Advanced Regional Travel Demand Models. Advanced regional TDM represent an emerging practice in regional travel demand forecasting models. They are used to overcome the limitations of traditional regional TDM. For example, tour- or activity-based models provide superior alternatives to traditional four-step models since they are based on individuals rather than trips, and the spatial resolution can be reduced to a smaller level of geography (such as parcels instead of TAZ). Particularly, as indicated by Sener et al. (2009), “by placing primary emphasis on activity participation and focusing on sequences or patterns of activity participation and travel behavior rather than travel, the activity-based approach recognizes the spatial and temporal linkages among the various activity-travel decisions of an individual, as well as the linkages between the activity-travel patterns of different individuals within a household”. Tour- or activity-based models simulate the activity travel decisions of households and individuals, which yields to activity-travel patterns that are composed of various decisions, such as to determine when or where an individual participates in a certain activity. Therefore, these models provide a more behaviorally realistic representation of travel through detailed information on individuals’ activity-travel patterns. Despite their advantages, the adoption of tour- and activity-based models have been relatively slow in practice especially due to several considerations, particularly on staff, time, and budget resources needed to transfer from traditional trip-based models to tour- or activity-based models. Therefore, while providing opportunities in estimating exposure for pedestrians and bicyclists, such advanced models have not yet become a practice in the safety field.
Trip generation and flow models can be considered as variations of regional TDM with the volume estimate developed as the model output. Such specially focused models that can be particularly beneficial to be applied at the corridor and subarea planning level. Although the models are applicable to both bicycle or pedestrian travel, there have been mostly pedestrian-based models of this category. The two common examples are pedestrian trip generation and flow models and network simulation models.
Pedestrian Trip Generation and Flow Models. These models are conceptually similar to regional four-step models, but focus on specific market segments particularly to improve the sensitivity of the models to bicycle and pedestrian travel. Pedestrian trip generation and flow models are among the examples of these focused models, which uses smaller geographical zones rather than TAZs. They generally include trip generation, trip distribution and network assignment steps. However, since they only focus on pedestrian trips, the models do not include a mode choice component. An example application of this type of modeling is the Model of Pedestrian Demand (MoPeD), which were developed by the University of Maryland’s National Center for Smart Growth. The model requires data on vehicle ownership, street connectivity, parcel-level land use, Census population and employment, and travel survey. MoPeD uses pedestrian analysis zones (which are block or street-level) as the level of analysis. The end result is an estimate of the numbers of pedestrians, or pedestrian volumes, which will occur on sidewalks and intersections in the study area over a 24-hr period (see for detailed procedural information: http://kellyjclifton.com/products/moped/).
Network Simulation Models. Network analysis models can also be categorized as variations of the four-step modeling approach, which are generally based on spatially driven network simulation procedures and use a representation of a network to estimate volumes for specific facility types over an entire area. The models use detailed network structures with complete links and nodes and with various other complementary data elements (e.g. street network density). Space Syntax is one of the most well-known example studies of network analysis models, which are described as “suite of modeling tools and simulation techniques used to analyze pedestrian movement and to predict pedestrian volume” (Raford and Ragland 2004). Although these models may have the potential to provide more appropriate exposure measures for non-motorized travel, Kuzmyak et al. (2014) indicated that two potential reasons for its relatively minimal usage: 1) the information on its special software is limited; and 2) the process is not intuitive to transportation planners since its process does not follow traditional trip generation and distribution steps. Instead, it uses spatial characteristics and relationships to explain the route chosen.
GIS-based models heavily use GIS tools and GIS-based measurements in determining activity levels. They can be described as spatial models of built environments and proximities. In the current context, these models are often used to estimate non-motorized travel under alternative land use and transportation investment scenarios. Important to note that, the models falling under this category are significantly dependent on GIS-based modeling and forecasting tools and use GIS as the main feature rather than as a tool to the modeling framework.
GIS-based Walk Accessibility Model. The GIS-based walk accessibility model (developed for NCHRP 770) provides an enhanced example to GIS-based tools, expanding their capability by estimating pedestrian trip tables using GIS-derived walk-accessibility scores. This model relies entirely on GIS tools and data to create relationships between land use activity, accessibility to opportunities defined by the shape and service of the transportation networks, and mode choice. It is similar to the Walk Score program, but it uses an enhanced methodology to estimate walk potential and mode choice. While generating walk trip tables, the accessibility model does not have an assignment module to estimate facility volumes. In addition, the model is limited to walk travel, but the structure might be applicable to bicycle travel if adequate bicycle data are available.
GIS-based Origin Destination Centrality Demand Model. Another example of GIS-based models is the methodology developed by McDaniel et al. (2014) to estimate bicycle volume, which has been recently automated via an online tool available at: http://uidaho.maps.arcgis.com/apps/webappviewer/index.html?id=88af7e023fd24d31965c4d0b62fdadd9. Although this model uses a combination of different techniques, it is included here due to its lineage with using GIS as the main feature of the modeling framework. Specifically, “the method uses network analysis to quantify travel patterns between origins and destinations through a new metric that is called origin–destination (O-D) centrality. The metric is then used as an explanatory variable in a direct demand model that is programmed as a tool for GIS software.” Once the input files are provided, the tool follows a systematic process including computing O-D centrality, calibrating the direct demand model and estimating bicycle volume across the entire street network.
Discrete choice models are used to analyze and predict decision makers’ preferences among several alternatives. The underlying principles of discrete choice models can also be used to determine bicyclists and pedestrian activity in the context of exposure analysis. In general, these models do not necessarily provide direct volume estimates but are included here due to their potential role as part of non-motorized planning tools that can be used in exposure estimation especially when integrated with other tools. For example, cross behavior models are used to develop information about the crossings and crossing behavior as an exposure measure. Route choice models provide significant information on the factors influencing an individual’s route choice (e.g., a bicyclist’s route), which can be then be integrated into a traffic assignment model to develop better volume estimates.
Simulation-based traffic models are built through the application of advanced computerized programs and mathematical modeling of transportation systems. They mainly use outputs of regional TDM as inputs into their algorithm to determine detailed activity levels. These models can be applied at microscopic, macroscopic, or mesoscopic levels. For example, agent-based microsimulation models have been used to capture pedestrian activity in an area through simulation of individual pedestrian movement in crowds using complex behavioral rules and environmental modeling. While providing quite accurate, detailed, and visually strong traffic flow, these models are quite complex and require significant input data and special resources, such as specialized software (e.g., VISSIM, PARAMICS) and unique technical expertise.
Data fusion is a process of integrating several data sources into a single one that provides a more accurate representation. Data fusion methodologies provide promising tools in estimating pedestrian and bicyclists’ activity especially considering the various data sources used in estimation as described above. The advancements in technology and heavy use of mobile devices have provided new opportunities in collecting crowdsourced data and extending the capability of data collection for non-motorized users (Lee and Sener 2017). For example, passively collected data through GPS-enabled smartphones, or actively collected data through user’s initiated smartphone applications (e.g., Strava) can be integrated with direct field counts to obtain a more informative pedestrian and bicyclists’ volume estimates. It is important to note that all these data sources have their own limitations and biases in sampling, and need careful processing to develop a final dataset which contains best estimates of exposure.
Analysts who are interested in learning more on demand estimation models will find many more procedural details in the following comprehensive guidance documents:
The following documents also provide information and examples of application on the demand estimation models described in Step 5. Since several examples of direct demand models are included in Step 6, they are not provided here.
Analysts should consider the following points when selecting a demand estimation model:
It is important that one carefully reviews the project goal or objectives together with the resources available (time, budget, data availability, expertise needed, etc.). While the most advanced form of modeling might be desired to answer most potential questions with relatively high level of accuracy, it is likely that there are limitations with the resources. It might be needed to focus on the most feasible and practical approaches and choose the one that will provide an optimal solution with acceptable accuracy.
A model is as good as its input data. Using a very advanced form of modelling with inadequate data may not result in any better estimations than a simple form of a model with very good data. It is essential that pm assesses currently available data and the feasibility of gathering additional data for a practical and informative analysis. A robust dataset will not only improve model performance at present, but will also be beneficial in validating various other analyses and models that may be needed in future.
Most of the demand estimation models may be available through different sources because of their use in other fields. For instance, an MPO of a large urban area might have already developed an activity-based travel demand model, which might provide detailed information on non-motorized trips. It is recommended to investigate all potential options available in the region to eliminate any duplicative work. While increasing the usability of what has been already done in the region, this might also help execute improvements to the existing models.
Before using or adapting any available model in the region, it is important to clearly understand what the models have been designed for, how well they perform, and if it would be suitable to use/adapt them for the question of interest. For almost all models, local customizations are expected to be made.
It is important to note that the end results of these models may embed unique characteristics of the region for which they were created and hence may not be directly transferable. Therefore, while the structure and general requirements are similar, the models may need to be redesigned, reimplemented, and calibrated with respect to local conditions.
The following checklist should also be helpful in evaluating the model selection/development decision:
A travel survey is a systematic effort to collect information about individual travel behavior. Travel surveys are typically collected from a statistical sample of travelers for a specified day or days (not an entire month or year), and typically gather aggregate trip information (travel mode, trip purpose, trip start and end location, trip length or time, etc.). Depending upon the number of travelers surveyed, trip information from travel surveys are often summarized into more aggregate geographic zones (not on specific facilities) to improve the statistical precision and accuracy of the survey data.
The ACS is a national ongoing survey of a sample of U.S. households by the U.S. Census Bureau that gathers a wide variety of information (e.g., demographic, social, economic, housing) in addition to their primary travel mode from home to work. Therefore, the ACS does not have trip information for non-commute trips (whereas NHTS does, but on a five- to seven-year cycle). Because the ACS only asks about the primary travel mode, it does not include modes of travel that may be considered secondary (such as walk trips to public transit).
The NHTS is a national survey of daily and long-distance travel that is conducted every five to seven years from a sample of U.S. households by the U.S. DOT. The survey provides estimates of trips and miles by travel mode (including walking and bicycling), trip purpose, and other household attributes and demographics.
A travel survey typically conducted by a MPO for the purpose of developing a regional travel demand forecasting model. The frequency of these surveys varies from city to city, with some planning agencies conducting household travel surveys every eight to ten years or longer.
|< Previous||Table of Contents||Next >|