Analytical Travel Forecasting Approaches for Project-Level Planning and Design (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Background P A R T 1

3 The project-level traffic forecasting guidelines presented herein are intended to • Help standardize the traffic forecasting process for high- way projects, • Give practical guidance to practitioners, • Give a high-level understanding to forecast users, and • Help define the current state of traffic forecasting practice. These guidelines are important since highway projects con- stitute a large portion of the U.S. transportation system’s infrastructure and, indeed, affect the entire U.S. economy. Experience has shown that traffic forecasting is a fundamen- tal part of planning for, and developing, highway projects. This report may be thought of as a revision of NCHRP Report 255: Highway Traffic Data for Urbanized Area Project Planning and Design (1). A tool box of techniques for directly creating project-level forecasts or for post-processing travel demand model results for use in the planning and design of highway projects was originally published in NCHRP Report 255. This introduction will help define what a “project” is within the context of NCHRP Report 255 and these new guide- lines and provide an overview of the organization of the guidelines. 1.1 What Are Projects? The guidelines presented herein are intended to be used for the planning, design, and operation of highway system ele- ments. Examples of highway projects include the following: • Traffic impact studies of new or modified land use activities. • Improvements to increase reliability and reduce travel time variability in support of state or regional operational goals and objectives, such as reversible lanes, hard shoulder run- ning, ramp metering, signal timing modifications, and vari- able speed limits. • Operational studies of highway facilities, such as lane clo- sure analyses, traffic control plans, signing plans, work zone traffic plans, and traffic incident management plans. • The planning of new highway facilities such as corridor studies, new alignments, needs studies, transportation improvement plans, air quality analysis, tolling analyses, and access management. • Construction of new highway facilities or expansions of existing facilities; additions including lane widening, added turn lanes, facilities on new alignments or new rights-of- way; and reconstructed pavements. NCHRP Project 08-83 includes analysis methods to support performance goals and target setting by states and metropoli- tan planning organizations (MPOs). The guidelines do not apply to highway projects with non- automobile elements. While they are multimodal to the extent that transit and non-motorized modes are part of a high- way’s design, the guidelines do not address stand-alone tran- sit projects. Figure 1-1 shows that traffic forecasting occurs in the early stages of a planning study, as well as in the environmental analysis and the design stages of a project’s development. Accurate and timely traffic forecasts are crucial for ensuring the success of highway projects through each stage, including construction and operations. 1.2 Context of NCHRP Report 255 Traffic forecasting has a rich history that parallels the development of the U.S. Interstate system and transporta- tion planning. NCHRP Report 255 was the first comprehensive practitioner’s guide to traffic forecasting that incorporated the use of computerized travel demand forecasting and provided a standardized set of procedures for translating raw traffic estimates into forecasts suitable for planning and design. C H A P T E R 1 Introduction

4NCHRP Project 08-83 follows the spirit and intent of NCHRP Report 255 in that it demonstrates how the large menu of fore- casting practices that are currently available can be applied to produce reasonable forecasts for planning, design, and opera- tions. Table 1-1 gives some perspective on how traffic fore- casting has developed over time and where NCHRP Report 255 fits within this timeline. As can be seen, NCHRP Report 255 came with the advent of the microcomputer. It has survived for over 30 years—a period of time that has seen the development of the Internet, much faster computers, and many innovations in travel fore- casting theory and software. New traffic forecasting guidelines are important since high- way projects, piece by piece, constitute a large portion of our transportation system’s infrastructure and, indeed, affect the entire U.S. economy. Experience over the past 30 years has shown that traffic forecasting is a fundamental part of devel- oping and accessing highway projects. 1.3 Traffic Forecasting Guidelines Project-level traffic forecasting has an underlying codifica- tion at the federal, state/MPO, and industry guideline level. In most cases, traffic forecasting procedures are the product of recommended guidelines rather than strict policy require- ments. Thus, there is tremendous variation in practitioner procedures. At the federal level, traffic forecasting is required for air qual- ity analysis, major investment projects, and highway design projects undertaken by the federal government (Special Report 288, FHWA [22]). Traffic forecasting is also an integral part of several standard transportation processes, such as highway design—see the AASHTO Policy on Geometric Design (121), and the Highway Capacity Manual (21). At the state/MPO level, traffic forecasting is also required by many states and MPOs for a variety of applications. Florida’s Project Traffic Forecasting Handbook (3) guides all traffic fore- casts made in Florida. Similarly Ohio, North Carolina, and several other states have manuals that guide the preparation of traffic forecasts. NCHRP Report 255 has been an authoritative source for this required traffic forecasting in the past 30 years. Section 1.4 gives a chapter-by-chapter review. 1.4 Chapter-by-Chapter Review of NCHRP Report 255 NCHRP Report 255 was published in 1982. Despite its age, several of its sections remain relevant, and these sections have been retained in this report. Appendix A of this report presents a detailed review of NCHRP Report 255 showing what sections remain intact in this report, which have been updated, and which were eliminated due to obsolescence (report appendices are provided on CRP-CD-143). Table 1-2 summarizes the review of NCHRP Report 255. Turning movement procedures and directional distribution procedures are still, for the most part, valid and of current interest. Procedures that have value but are in need of major updating include the following: • Screenline refinement, • Interpolating between forecast years, • Extrapolating existing forecasts, • Time-of-day refinement, and • Vehicle classification. 1.5 Traffic Forecasting State of the Practice The traffic forecasting state-of-the-practice contains three sections that are contained in the appendices to this report: • Source documents—a review of the literature and a review of the most important current traffic forecasting reports and sources in Appendix C. State-of-the-practice docu- ments were reviewed as part of this research. A total of 41 source documents were reviewed and summarized. Figure 1-1. Project development process.

Year Key Transportaon Organizaon (s) PlanningMilestones Technology Traffic Forecasting State of the Art Forecasng RelatedManuals 1930s Bureau of Public Roads (BPR 1915) Traffic counters Toll road studies American Associaon Of State Highway Officials (AASHO-1914) AASHO Geometric design of rural highways 1940s Federal Works Agency (1939-49) Travel surveys Count trends 1950s BPR resumes Housing Act of 1954 Chicago Area Transportaon Study six-step planning Analytical methods Highway Capacity Manual (HCM 1950) 1956 Interstate Highway Act 20 yr forecast period 1960s Federal Highway Administraon (FHWA - 1966) Naonal Environmental Protecon Act 1969 First Computer Models Modeling HCM 1965 Fed. Aid Highway Act (1962) Planpac documentation Planning studies 1970s Naonal Highway Transportaon Survey Commercial soware Modeling Ismart Validation Report Environmental Protecon Agency (EPA 1970) Clean Air Act Amendment (1970) ITE Trip Generaton Report, 1st edition (1972) American Associaon of State Highway and Transportaon Officials (AASHTO - 1973) 1980s Microcomputers Systemac modeling Naonal Cooperave Highway Research Program 255 (1982) NCHRP 187 (1978) HCM 1985 1990s ISTEA (1991) Faster microcomputers Air quality modeling NCHRP 365 (1998) Model Validaon and Reasonableness (1997) Quick Response Freight (1996) 2000s Acvity based modeling Activity based modeling HCM 2000 Conference Special Report 288 Freight Analysis Framework 2010s Probes/cell phone studies Performance based NCHRP 706 (2012) planning Quick Response Freight II (2007) ITE TG Report, 9th Edition (2012) HCM 2010 Source: Urban Transportation Planning in the United States (139). Table 1-1. Traffic forecasting history highlights.

6Table 1-2. NCHRP Report 255 review summary.

7 There are 19 national sources; 20 state, county, and state/ MPO sources; and 3 other sources. The documents range from handbooks to guidelines to policy manuals to peer exchanges. The topics include travel estimation and forecast- ing techniques, traffic data, tools and resources, and travel demand model techniques. Additionally, several of these documents contain case studies, request forms, spreadsheets/ workbooks, and transferrable parameters that could be of use to practitioners. • Traffic forecasting survey—a summary of results from a survey of forecasting practitioners in Appendix D. To guide the development of NCHRP Project 08-83, a survey was developed to understand the state-of-the-practice for the following: – Project-level traffic forecasting methods and techniques currently employed, – Limitations of current methods, and – Needs and deficiencies in the practice that must be addressed. The most frequently cited uses of traffic forecasting appli- cations in planning were corridor planning (80%), Long- Range Transportation Plans (LRTPs) (74%), and site impact analysis (63%). Sixty-five percent (65%) of respondents reported using at least one NCHRP Report 255 technique for forecasting. • Expert panel—the findings and conclusions from a dis- cussion with an expert panel on the form and content of NCHRP Report 255 can be found in Appendix E, which also summarizes the panel’s comments about the survey results and suggestions about various study aspects. The overall study methodology and the survey results were pre- sented in a webinar to the expert panel. 1.6 Report Organization This report is arranged into two distinct parts: Part 1: Back- ground and Part 2: Guidelines. Appendices A through I are available on CRP-CD-143, which is bound into the report. An .iso image of CRP-CD-143 and instructions for burning this image onto a CD-ROM are available on the TRB website. The intent of the report is to be an easy-to-use and frequently ref- erenced resource for transportation planning practitioners. 1.6.1 Part 1: Background Part 1 contains the chapters described below. Chapter 1: Introduction defines and describes project- level traffic forecasting and how it fits in with the overall project development process. This chapter is a review of the traffic forecasting state-of-the-practice. A survey of practi- tioners and feedback from an expert panel provided input for this section. Chapter 2: Overview of the Fundamentals of Traffic Forecasting contains a review of the fundamental traffic forecasting parameters, introduces traffic forecasting mea- sures of effectiveness, and describes source data/basic traf- fic forecasting tools. It also recommends key resources—the essential bookshelf—that all forecasters need. Chapter 3: Overview of Traffic Forecasting Tools and Methodologies describes a proposed standard for traffic forecasting. It also summarizes the state-of-the-practice of travel forecasting models (including four-step models), traf- fic simulation models, advanced four-step modeling prac- tices, advanced modeling practices, modeling inputs/outputs, and other forecasting tools. 1.6.2 Part 2: Guidelines Part 2 contains the chapters described below. Chapter 4: The Project-Level Forecasting Process will be the starting point for most users of the guidelines. This sec- tion covers the forecasting process from inception through development of project-level traffic forecasts and documen- tation/communication of results. There are example forecasts contained in this section and in the appendices, along with a tool selection matrix that helps determine the correct tool to be used for a number of forecasting applications. Chapter 5: Working with a Travel Model provides a variety of information on the proper use of travel demand models for traffic forecasting. Topics include a model component check- list; validation methods; speed/volume data errors; input data fixes; and understanding model outputs including outliers, the role of professional judgment, and computational issues. Chapter 6: Model Output Refinements describes refine- ments to travel demand model outputs. Origin-destination (OD) matrix estimation methods, turning movement refine- ment methods, and screenline volumes adjustments using the traditional NCHRP Report 255 (1) ratio and delta spread- sheets are included, as well as information on how to refine directional splits from travel demand models. Chapter 7: Refining the Spatial Detail of Traffic Models provides guidance on specialized tools that can be developed from travel demand models, including focusing, windowing, refining directional splits from travel demand models, and multiresolution models. The chapter also discusses refin- ing external-external trip tables and integrating statewide, regional, and local travel demand models. Chapter 8: Improving Temporal Accuracy of Traffic Fore- casts includes information that helps the forecaster understand and develop traffic forecasts for different time periods and at different levels of temporal detail. Key topics include activity- based modeling, dynamic traffic assignment, peak spreading, pre-assignment/post-assignment factoring, day-of-the-week

8factors and month-of-the-year factors, and vehicle classifica- tion considerations. Chapter 9: Traffic Forecasting Methods for Special Purpose Applications contains a host of topics: basic traffic forecast- ing deliverables, interpolation of traffic forecasts, vehicle mix accuracy, equivalent single axle loads, benefit/cost analysis, toll/ revenue forecasts, work zone congestion, environmental jus- tice, and traffic impact studies. Chapter 10: Tools Other Than Travel Models acknowl- edges that not all traffic forecasts use a travel demand model and provides guidance on when to use travel demand mod- els and when to use other methods. Techniques covered include time-series models, a sketch-planning technique (“manual gravity”) for traffic diversion, elasticity methods, post-processing using the edition of the Highway Capac- ity Manual published in 2010 (HCM2010) (21), combining data from multiple models (“stitching models together”), and an overview of a simplified highway forecasting tool for low-risk traffic forecasts. Chapter 11: Case Studies uses case studies to illustrate the application of multiple techniques simultaneously. There are case studies for a suburban arterial network, a network win- dow, a small city, an activity-based model application for a large city, a time series on a link, and blending travel forecast- ing with a traffic simulation model. 1.7 CRP-CD-143 The following material is available on CRP-CD-143, which is bound into this report. An .iso image of CRP-CD-143 and instructions for burning this image onto a CD-ROM are avail- able on the NCHRP Report 765 web page on the TRB website. Included on CRP-CD-143 are Appendices A through I from the contractor’s final report for NCHRP Project 08-83. These appendices supplement this report by providing a sub- stantial amount of companion data and information. The appendices also include the extended literature review, the detailed NCHRP Report 255 review, supplementary tables, and glossary/acronyms. Also included on CRP-CD-143 are spreadsheet demon- strations. These spreadsheets include tools for screenline refinement, OD table estimation, time-series methods, and sketch-planning methods. Finally, CRP-CD-143 includes, for reference purposes, a tool developed by the North Carolina Department of Transportation to assess annual average daily traffic.

9 C H A P T E R 2 2.1 Traffic Forecasting Data and Parameters Traffic data are the fundamental unit of information required for traffic forecasting. This includes average annual daily traf- fic (AADT) counts, design hour volumes, and percent trucks. Baseline traffic data can be factored by growth rates to produce forecasts, used for validation or comparison against alterna- tive sources of information, or used as a control total against which more detailed information is developed. Several of these data can be further referenced in the Ohio Certified Traffic Manual (2) and the 2012 Florida Project Traffic Forecasting Handbook (3). Parameters are numeric constants and/or weights used in mathematical expressions and are integral to the generation of baseline traffic data and traffic forecasts. The parameters used in project-level traffic forecasts vary depending on the study and tools used to perform the forecast and local area objectives. Most parameters are obtained by applying sta- tistical analysis to travel behavioral data or traffic data that are obtained locally. When statistical analysis cannot be per- formed, then borrowed or asserted parameters may be used in the application, if appropriate and defensible. The use of bor- rowed parameters rather than project-specific information is more acceptable for planning studies than for engineering or design studies, where the tolerance for error is substantially lower and the need for accuracy higher. This project-level forecasting should minimize the use of borrowed parameters. 2.1.1 Basic Traffic Forecasting Definitions Traffic counts/classification counts are used for capacity analyses and pavement design. It is necessary to have high- quality traffic count data to estimate traffic forecasting param- eters of AADTs, design hourly volumes (DHVs), design hour factors (K), directional distribution factors (D), seasonal factors, the daily truck factor (T) and turning movements. Usually the state departments of transportation (DOTs) have a monitoring system responsible for programming, collect- ing, analyzing, and reporting traffic volume and vehicle clas- sification data on Interstates and highways throughout the state. Permanent traffic count information provides a statis- tical basis for estimating AADT, D, and T for all other traffic counts where short-term traffic counts are obtained. Short- term traffic count data are useful for capturing peak-hour intersection turning movement counts. Historical count data are useful in determining traffic growth trends. Average daily traffic (ADT) reflects the average number of vehicles that travel through a segment of roadway over a short-term period. The ADT is the raw, non-factored count data obtained by a short-term traffic count (usually a 24- to 48-hour period) typically collected on Tuesday, Wednesday, and Thursday. The ADT is an important, basic unit of traffic monitoring and is essential for developing traffic forecasts. Average annual daily traffic (AADT) is the estimate of typi- cal daily traffic on a road segment for all days of the week, Sun- day through Saturday, over the period of 1 year. A true AADT is developed by utilizing a full year of traffic count data, such as data generated from a permanent traffic counter; however, for the purposes of the guidelines presented herein, the term AADT is used to show a factored ADT. The AADT adjustment factors include daily factors, weekly factors, seasonal adjust- ment factors, and axle correction factors developed from auto- matic traffic recorders (ATRs) that collect data continuously throughout the year. The AADT is the best measure of the total use of a road because it includes all traffic for an entire year. It is important to always use AADT for traffic forecast- ing since a simple ADT may not be representative of the average traffic at the site being measured. The formula for calculating AADT is ×=AADT ADT Adjustment Factor Design hourly volume (DHV) is the number of vehicles that travel through a segment of roadway during the design hour. The DHV is used for making roadway structural and Overview of the Fundamentals of Traffic Forecasting

10 capacity design decisions because traffic volume varies by hour and from day to day throughout the year. The formula for calculating the DHV is =DHV K AADT× Design hour factor (K) is the design hour factor that repre- sents the proportion of AADT occurring in an hour. K factors can change due to a number of factors such as peak spreading and altered traffic patterns and should be updated frequently. The AASHTO Green Book states that the 30th highest hour of the year is best suited for the design hour used to design high- ways in non-urban settings and that the K factor varies only slightly from year to year even though the ADT itself might change significantly. Frequently, states are not allowed to design to the 30th highest hour because of financial constraints or large seasonal variations. Several states use the 50th highest hour or even the 100th highest hour of the year to reflect the design hour volume. Additionally, Florida uses standard K val- ues which define “factors within a rural, transitioning, urban or urbanized area that are based on a ratio of peak hour volume to annual ADT. Multiple standard K factors may be assigned depending on the area type and facility type and applied state- wide” (2). The formula for calculating the K factor is =K DHV AADT Directional design hour volume (DDHV) is the amount of traffic moving in the peak direction during the design hour. It is a critical design volume with a directional compo- nent. The formula for calculating the DDHV is =DDHV DHV D× Directional distribution factor (D) represents the propor- tion of traffic moving in the peak direction during the design hour. The directional factor is derived from ATR data and estimated with short-term count data. D is a critical number that helps determine the geometric design of a road. Traffic on most roads is not evenly split (50%/50% in both direc- tions) during the design hour; the D factor represents this asymmetry and the possibility of different geometric designs by direction. The directional distribution factor is used to determine the DDHV: =D DDHV DHV Daily truck volume (DTV) is the volume of heavy and commercial trucks on a roadway segment. DTV can be cal- culated by multiplying the total daily volume by T: =DTV AADT T× Daily truck factor (T) is the critical value for pavement design. It represents the percentage of ADT that is heavy and commercial trucks (Categories 4 through 13). Since trucks take up more space and are heavier vehicles than passenger cars, T is an important component in the design of pavement thickness in highway design projects. The daily truck percent- age is derived from vehicle classification counts on the actual facility or functional class averages: =T DTV AADT Design hour truck percentage (DHT) is defined as the percentage of DHV that is heavy and commercial vehicles. Typically this percentage is less than the daily truck percentage since the percentage of truck traffic is not evenly distributed throughout the day. For example, in Florida, DHT is assumed to be half of T because it is assumed that the proportion of trucks present in peak-hour traffic is half that present for ADT. Examples of other parameters used in traffic forecasting and baseline traffic data estimates include the following: • Percentage of local and through traffic; • Percentage of traffic by trip purpose; • Trucks as a percentage of total traffic; • Vehicle occupancy; • Values of time; • Hourly (diurnal) distribution of traffic as a percentage of total traffic; • Rates, including trip generation rates, traffic growth rates, emission rates, and accident rates; and • Intersection analysis inputs, including traffic progression, signal timing, and phasing. 2.1.2 Other Traffic Forecasting Data Data required to produce project-level traffic forecasts vary depending on the study and tools used to perform the forecast and local area objectives. Maps/aerial photos/site data are used to gather current traf- fic information that is not readily available from other sources and to determine existing and future land uses that contrib- ute traffic that would use the proposed facility. The Manual on Uniform Traffic Control Devices (MUTCD) is a good reference to follow when collecting new data: http://mutcd.fhwa.dot. gov/kno-2003r1.htm (4). For some projects, it may be neces- sary to collect additional new data via surveys, including spe- cial travel time information. Existing travel demand models are used to estimate exist- ing and future traffic volume on various facilities and include origin-destination (OD) and routing information. Travel demand models are data intensive and require such inputs as socioeconomic data, population, employment, network geog- raphies and attributes, survey data, freight data, and so forth for existing and future conditions and often must be validated and adjusted accordingly. Inventory databases/reports, housed by many states and local agencies, are useful for project-level traffic forecasting.

11 Such inventory databases include historical traffic counts, intersection control, roadway classification, accident history, previous traffic forecasts, and so forth. The state and local area reports are useful to reference guidelines, local parameters and traffic conditions, and spreadsheets for use in traffic forecasting analysis. Such reports include the following: • State and MPO source documents of K and D factors, hourly percentages, and traffic counts, and “FHWA Vehicle Clas- sification Scheme F Report” (http://www.dot.state.oh.us/ Divisions/Planning/TechServ/traffic/Reports/Scheme_F_ Report/SchemeF.pdf). • State level of service manual. • Existing forecasts from corridor, long-range planning, and thoroughfare studies. Environmental data are used for environmental analysis of air quality, noise impact analyses, and energy consump- tion analyses. Environmental data are not typically captured by traffic counts or traffic models, but include cold/hot starts distributions and vehicle age distributions. 2.2 Traffic Forecasting Tools There are several methods available to produce traffic fore- casts. The most appropriate method depends on factors such as the size and complexity of the project, the desired time frame, the resources available to produce the forecasts, the availability of data, and staff resources. Tools that are rou- tinely used include the following: • Growth rates, • Trend line analyses, • Time-series analyses, • Turning movement analyses, • Travel demand models, and • Traffic simulation models. At one end of the spectrum, where forecast needs are rela- tively simple and where external factors (e.g., land use) remain relatively constant, tools such as growth rates, trend lines or time-series analyses, and turning movement analyses are relatively simple to apply and yield reasonable results. These tools were documented in NCHRP Report 255 and have been successfully applied for many years. They have been updated in this report. Forecasting needs have become more complex, however, resulting in a need for more sophisticated tools and tech- niques. This has occurred for numerous reasons: • Changing land use patterns have intensified travel demand in suburban and developing areas; • Increasing traffic congestion has resulted in the need to address peak-period spreading and oversaturated roadways; • Traditional infrastructure funding mechanisms are no lon- ger sufficient to keep pace with growing travel demand, and user-funded alternatives must be considered in traffic forecasting; • Applications of traffic forecasting have expanded with the increasing availability of complex tools like microscopic traffic simulation; • Increasing policy regulations related to the environment, fuel economy, and safety have introduced new needs for traffic forecasts; and • Travel choice and accessibility policies have resulted in the need to address transit and non-motorized forms of travel. The result is that the set of available tools has become both more robust and more complex. Correspondingly, the required understanding of these tools and skill with their application is greater than ever. 2.3 Measures of Effectiveness Analysts and planners produce traffic forecasts to assess transportation performance under different assumptions about transportation supply and demand. The forecasts pro- vide information that describes the condition and perfor- mance of the transportation system and guides policy and investment decision-making. The vast quantities of data that traffic forecasts often produce need to be synthesized into credible, digestible information for a variety of audiences, including non-technical decision-makers, stakeholders, peer analysts, and individuals and groups seeking to understand how a particular decision was reached at some point in the past. There are a number of industry-standard measures of effectiveness (MOEs) that provide this information. The con- text of the study or analysis and the audience for the MOEs should guide the analyst toward the appropriate MOE, level of required detail, and particular style of presentation. MOEs describe the condition and performance of the trans- portation system, but do not address the adequacy of the tool or the methodology employed to forecast future conditions. Understanding the validity, sensitivity, and accuracy of a fore- casting procedure or tool is the province of calibration and validation. By the time MOEs are extracted or developed from a fore- casting process, the forecasting process should have been properly configured and accepted as appropriate for the pur- pose for which it is being applied. Analysts should be cognizant of the context in which the information they produce will be used and come to an under- standing about the MOEs to be published and presented at the outset of a study. Often, highly detailed information about a traffic forecast will be published in technical appendices,

12 far from the executive summary most consumers will read. Or, the work of several months will be crystallized into a few short presentation slides. Anticipating how information will be used early on will help avoid misunderstandings and miscommunications. Analysts should also try to describe any underlying factors or assumptions to which the measures are most sensitive. In many instances, traffic forecasts are heavily influenced by assumptions about population or employment growth. Lastly, and within the purview of traffic forecasting, analysts should describe any trade-offs that may result from alternative courses of action beyond those that can be reliably described by MOEs, including environmental, financial, and social impacts. There are a number of industry-standard MOEs that provide this information, as shown in Table 2-1. Table 2-1. Typical MOEs by forecasting application. Traffic Forecasng Applicaon Typical Volume based MOEs Typical Time based MOEs Accessibility MOEs Air quality conformity analysis Area wide Vehicle Miles of Travel Speeds Asset management, including bridge and pavement needs Link specific volumes Capital Improvement Program, priorizaon Benefit/cost, Level of Service Congeson management process Corridor volumes Speeds Corridor mobility studies Intersecon Level of Service, intersecon turning movements, traffic volumes Segment travel mes Demand management plans Number of peak-hour trips, Level of Service Vehicle Hours of Delay Environmental impact statements Vehicle Miles of Travel, emissions, accidents Vehicle Hours of Travel Evacuaon plans Hourly traffic volumes, throughput Travel mes Facility design and operaons Design hour traffic volumes Highway feasibility studies Benefit/cost, screenline volumes, Level of Service Vehicle Hours of travel Access to labor market and jobs Interchange jusficaon requests Traffic volumes, Level of Service Roadway (general and freight) long range planning Vehicle Miles of Travel, Level of Service Vehicle Hours of Travel Access to labor market and jobs Traffic impact studies Intersecon turning movements, Level of Service, delay per vehicle

13 There are three categories of MOEs covered in this sec- tion: volume-based measures, travel-time-based measures, and accessibility measures. 2.3.1 Volume-Based Measures The definition of volume-based measures, their strengths and weaknesses, and their application and purpose are given in the following: • Definition—Volume-based measures deal with the quan- tity of use of transportation facilities at a point on a trans- portation network, along several points on a screenline (an imaginary line that cuts across several roadways), or between origins and destinations. Demand is associated with a specific time frame, typically ranging from 15 min- utes to a 365-day annual period. • Strengths/Weaknesses—Traffic volumes are the most basic, readily understood, readily accepted, and thus most important output of the traffic forecasting process. How- ever, traffic volumes do not convey the traveler’s experi- ence of congestion or delay, per se. • Application/Purpose—Traffic volumes describe the num- ber of vehicles at a point on a roadway. Traffic forecasts are often developed for specific travel markets, such as passen- ger vehicles and trucks. Special-purpose studies may focus on additional vehicle classes, such as buses or combination and single unit trucks. Volumes can describe how the distribution and magnitude of demand change across different supply/demand scenarios. Volumes can describe throughput—a measure of the quan- tity of transportation activity that can be accommodated at a single point or multiple points on a transportation network. Volumes are also a critical input to assessments of congestion and economic and environmental impact. There are several ways to quantify demand at multiple locations. One is to quantify demand crossing boundaries such as screenlines or cutlines or to show demand between regions using OD tables. Another is to sum the total vehicle miles traveled (vehicles × segment length, summed over all segments in the area of interest). With the exception of VMT, each vehicle (or person) should be counted only once (to cal- culate the total distance a vehicle travels, it must be counted multiple times). Level of service (LOS) can be considered a traffic-volume- based measure. In transportation planning and preliminary engineering applications, LOS is a widely used and instantly recognizable qualitative measure of roadway and, in particu- lar, vehicular congestion (this discussion does not address bicycle, pedestrian, or transit quality/LOSs). LOS distills the calculation of the quality of roadway service—which is the product of dozens of input variables and parameters—into a simple letter grade, from A to F. The intuitive appeal of the measure is evident in Figure 2-1. Stop and go conditions are indicative of LOS F, for example. The Highway Capacity Manual (HCM) is the definitive source for calculating LOS for operational and planning appli- cations. The HCM and the HCM software provide separate methodologies for different roadway types, including free- ways, multi-lane highways, two-lane highways, and signalized and non-signalized intersections. LOS calculation methodol- ogies are provided for interchanges, freeway weaving sections, roundabouts, and multimodal applications as well. LOS is a measure of the congestion produced by the inter- action of supply and demand. To calculate LOS, the analyst supplies information about geometric conditions (e.g., num- ber or lanes), traffic characteristics (e.g., free flow speed), and traffic demand (e.g., vehicles in peak 15-minute period). LOS and volume-to-capacity ranges are shown Table 2-2. See Section 10.4 of this report for more details on the use of the HCM for project-level traffic forecasting. Figure 2-1. Traffic conditions described by LOS. Source: 2009 FDOT Quality/Level of Service Handbook (78). LOS A/B LOS C/D LOS E/F

14 2.3.2 Travel-Time-Based Measures The definition of travel-time-based measures, their strengths and weaknesses, and their application and purpose are given in the following: • Definition. Travel-time measures describe the trip dura- tion of travel by a vehicle between two points or the total effort incurred in travel by a group of travelers between or within regions. • Strengths/Weaknesses. Travel-time-based measures more closely describe a traveler’s experience than do volume-based measures. However, most forecasting applications estimate travel time less accurately than travel demand volumes. • Application/Purpose. Travel-time information is often compared to some norm or standard to describe perfor- mance. The standard or baseline against which travel-time performance is being measured should be realistic and clear. When comparing travel times between two points, often the point of comparison is a no-build scenario. A related measure, travel-time delay, is expressed as the increment of time incurred in travel, over and above some expected level. Often the expected level is defined as travel in uncon- gested conditions (free flow times). However, uncongested conditions may be an unrealistic standard in some urban- ized areas and during peak periods of travel. This is less of an issue when delay is used as a relative measure, as a way to compare alternatives. Travel-time contours have been used to illustrate the effects of congestion on access to destinations. Travel-time contour maps are similar to topographical maps, with lines encircling a destination that radiate outward and do not touch. The spac- ing between the lines corresponds to the travel times needed to traverse them for a given distance. Thus, closely spaced lines correspond to slower speeds than do widely spaced lines. Tools and forecasting techniques that include stop delay (at intersections) and queuing delay (at bottlenecks) are more accurate than those that do not. Highway capacity techniques and microsimulation models account for these travel-time components, and microsimulation does so for individual vehicles, over small time slices and at a high level of geographic detail. Regional travel demand models measure travel times for peak-hour conditions at best, and their speed models estimate times for all vehicles on a link, regardless of vehicle type or the lane of travel. Vehicle hours of travel describe the level of activity in a region. When applied to travelers’ values of time, they are the most significant component of a benefit/cost analysis. Lastly, reliability measures use the variability of travel times between two points to describe traveler inconvenience or the addi- tional time needed as a cushion to ensure an on-time arrival. 2.3.3 Accessibility Measures The definition of accessibility measures, their strengths and weaknesses, and their application and purpose are given in the following: • Definition. Accessibility measures the proximity of people to places. Regions that offer more transportation supply to areas with more and denser land use activity score more highly on accessibility measures, other factors being equal. Since accessibility combines elements of land use and trans- portation supply conditions, it is used in a variety of analy- ses, from economic impact studies to social impact studies. • Application/Purpose. Accessibility measures can be used to assess how environmental justice populations are affected by transportation investments and development and to estimate the proximity of facilities such as ports, airports, warehouses, and distribution facilities to shippers and site retail and other commercial developments, using measures such as the population within a reasonable distance of a development site. • Strengths/Weaknesses. Higher levels of accessibility corre- spond to increased destination choices and modal choices and thus better economic and social outcomes. Basic mea- sures of accessibility are readily understandable when the choices or opportunities available are simply counted. When used at large geographic scales, the measure can be relatively insensitive to small to moderate changes in trans- portation capacity. Aggregate measures of accessibility that produce unitless results (such as change in utility) can be difficult to explain and understand. • Application/Purpose. Typical applications include the relationship between a smaller area and a larger region: – Number of households (larger region) within 30/45/ 60 minutes of an employment or activity center (smaller area), as a measure of access to labor. – Number of jobs (larger region) within 30/45/60 minutes of a residential area (smaller area), as a measure of access to jobs. LOS Volume to Capacity Rao Range Percent of Free Flow Speed (Peak Hour) A 0.50 and below 90% or greater B 0.60 to 0.69 70% to 90% C 0.70 to 0.79 50% D 0.80 to 0.89 40% E 0.90 to 0.99 33% F 1.00 and above 25% or less Source: http://www.kitsapgov.com/pw/translos.htm. Table 2-2. LOS and volume-to-capacity ranges.

15 This type of accessibility measure can be averaged over several zones comprising a region as: 1 1 A = A W W i ii = n ii = n ∑ ∑ Where W is the weighting factor, such as households or jobs (5), i is the origin traffic analysis zone (TAZ), j is the destina- tion TAZ, and n is the number of zones. Researchers and planners have developed more generalized measures of accessibility, suitable for application at a regional level. Such measures rely on formulations similar to the grav- ity model, or the logsum of a mode choice or destination choice model. A gravity-type formulation is the following: ∑ −βln 1 A = O ei j Tij j = n where Ai = the accessibility from region or origin i, O = opportunities (such as jobs) at destination j, Tij = travel time between origin i and destination j, and b = friction factor parameter for exponential function (values can range from 0.05 to 0.15). The denominator of a destination or mode choice model, the logsum, can also be used as a measure of accessibility across all modes represented. A typical formulation is the following: 1 A lne ei V k m k∑=   = Where Vk is a linear combination of utilities corresponding to different trip components (e.g., parking, in-vehicle time, out-of-vehicle time) for mode k. 2.4 Essential Bookshelf The essential bookshelf is an accumulation of documents fundamental for project-level traffic forecasting. The docu- ments include federal guidance documents, NCHRP pub- lications, and state DOT resources. These documents are summarized and reviewed for their usefulness related to traffic forecasting. The techniques, parameters, and applica- tions described in these reports are identified as they relate to project-level traffic forecasting. 2.4.1 NCHRP Report 716: Travel Demand Forecasting: Parameters and Techniques The objectives of NCHRP Report 716 (6) were to revise and update NCHRP Report 365 (7) with current travel characteris- tics and guidance on forecasting procedures and applications. NCHRP Report 716 was developed to address a broad range of planning issues and to be a user-friendly guidebook with a range of approaches and references to more sophisticated techniques. Data sources include the 2001 National Household Travel Sur- vey (NHTS), the 2009 NHTS, and an analysis of MPO docu- mentation. The MPO information was gleaned from 70 MPOs (ranging from small to large) through direct contact or using publicly available reports. Information includes model param- eters (trip attraction rates, friction factor parameters, mode choice parameters, and volume-delay parameters) and model methods used. Variables were selected based on ease of transfer- ability. Trip purposes include home-based work, home-based school, home-based other, and non-home-based. The key empirical findings reported in NCHRP Report 716 are largely contained in Chapter 4, “Model Components,” which contains all the recommended parameters by model component. The components of the four-step modeling paradigm are described in chapter sections and include the following: • Time of day. This section provides factors for post- distribution factoring and for post-mode-split factoring. The factors are provided for both “from” and “to” trips for each trip purpose. • Vehicle availability. This section provides four different formulations of the multinomial logit model utility equa- tion, with each formulation having many independent variables. • Trip productions. This section provides options as to the independent variables for the cross-classification model and parameters stratified by city size for some trip purposes. • Trip attractions. Parameters are given for linear relation- ships of trip attraction as a function of activity levels (employment in various industrial sectors, households, and school enrollments). • Trip distribution with the gravity model. This section gives sample friction-factor parameter values and mean trip lengths for cities of various sizes, which could be used to calibrate a single-parameter friction-factor function. • External trips. This section recommends against transfer- ring external-to-external (E-E) models from other locales and for synthesizing an E-E table from traffic counts. • Mode choice. The section presents several mode-split mod- els, some multinomial logit and some nested logit, that have been developed for cities throughout the United States. • Automobile occupancy. This section recommends fixed automobile occupancy factors. • Highway traffic assignment. The report gives parameters for a volume-delay function (VDF) based on the Bureau of Public Roads (BPR) curve as well as a few values of time for converting tolls to impedance.

16 • Transit assignment. Rather than give specific values for an impedance function, the report gives historical ranges of ratios of out-of-vehicle time values to in-vehicle time values. • Freight. This section outlines the steps of a commodity- based freight model and gives sample parameters for these steps. Advanced modeling practices and model validation are also discussed in NCHRP Report 716. Additionally, case stud- ies are provided illustrating the application of parameters for conventional travel forecasting. However, the report does not focus on the concept of transferable parameters but rather on the variability in parameters used by MPOs throughout the United States. Because of the way the parameters are presented, they mostly should be used only as guidelines to ensure that locally developed parameters fall within a reason- able range. 2.4.2 NCHRP Report 365: Travel Estimation Techniques for Urban Planning NCHRP Report 365 (7) is an update to NCHRP Report 187 (8) and has been superseded by NCHRP Report 716 (6). NCHRP Report 365, for the most part, provides default parameters for three- and four-step travel forecasting models for urban areas in the United States. The philosophy of NCHRP Report 365 was similar to that of NCHRP Report 187—travel forecast- ing parameters are transferable between urban areas, and parameters established through national data sources can be applied to specific urban areas. NCHRP Report 365 retained few methodologies from NCHRP Report 187, but did retain the North Carolina E-E model and a highway spacing meth- odology. There is no mention of freight in NCHRP Report 365. NCHRP Report 365 makes key recommendations in the following areas: • Many parameters differ by urban area size. Four urban area sizes are defined. • There are three standard trip purposes: home-based work, home-based-other, and non-home-based. • The trip attraction equation is linear and has five inde- pendent variables: dwelling units, central business dis- trict (CBD) retail employees, non-CBD retail employees, service employees, and non-retail/non-service employ- ees. Parameters differ by trip purpose, but not by urban area size. • Trip productions for all trip purposes are done using cross-classification in terms of persons per household and number of automobiles (or perhaps income). The split to different trip purposes is based on income. • Gravity model-friction factors are given in terms of a “gamma” function. Parameters vary by trip purpose. • A logit model, with coefficients, is provided for mode split. • Parameters are provided for the relationship between vol- ume and travel time in the form of a BPR curve. The availability of transferable parameters means that a generic travel model can be assembled rather quickly, as the need arises, without the need for a local travel survey. In addition, transferable parameters can help provide a seed OD table for synthetic OD table estimation from ground counts. 2.4.3 Highway Capacity Manual 2010 (HCM2010) The HCM contains concepts, guidelines, and computa- tional procedures for computing the capacity and quality of service of various types of highway facilities, including free- ways, signalized and unsignalized intersections, and rural highways, as well as the effects of transit, pedestrians, and bicycles on the performance of these systems. HCM2010, TRB’s fifth edition of the Highway Capacity Manual (21), incorporates results from more than $5 million of research completed since the publication of the 2000 edition of the HCM. It provides more in-depth analysis of signalized road- way segments and introduces a multimodal LOS methodol- ogy that considers the needs of all users of urban streets at the intersection, link, segment, and facility levels. The com- panion software suite contains all of the HCM methodologies and vastly facilitates capacity and LOS calculations. The HCM provides service volume tables with which the capacity of any type of roadway can be calculated. The ser- vice volume tables enable an analyst to quickly determine the number of lanes needed to achieve a target LOS according to the demand and supply characteristics of the planned facil- ity (e.g., free flow speed, AADT, the demand peaking factor, and the directional demand factor). The document provides guidance on the appropriate use of defaults and the construc- tion of agency-specific service volume tables. For planning applications, the HCM estimates LOS based on look-up tables of traffic density (vehicles/lane/hour), which in turn relate to speed. The HCM also provides methods for evaluating isolated intersection volume/capacity ratios. It provides recom- mended performance measures for the evaluation of systems of facilities. It identifies six dimensions of system perfor- mance: quality of service, intensity of congestion, duration of congestion, extent of congestion, variability, and accessi- bility. In HCM2010, for the first time, guidance is provided on adapting its capacity and free flow estimation methods for improved travel demand model traffic assignment appli- cations. The HCM also provides case studies, examples, and

17 guidance on the capacity and LOS analysis, and covers addi- tional, related topics such as speed and volume data collec- tion to quantify traffic demand and performance measures such as saturation flow and stopped delay. The HCM is a tool for estimating roadway and traffic per- formance and uses project-level traffic forecasts as an input. HCM2010 has the ability to integrate easily with Freeval and Transyt7f to simulate the operational impacts of alternatives. Lastly, HCM2010 recognizes variability in traffic demand (seasonal, monthly, and daily; recreational versus commuter; and hourly [peak versus non-peak, peak hour versus analysis hour]) but offers no methods for quantifying it. 2.4.4 Travel Model Validation and Reasonableness Checking Manual, Second Edition The Travel Model Validation and Reasonableness Checking Manual (9) provides a framework for testing and updating travel demand models to achieve an acceptable level of per- formance in terms of accuracy and sensitivity. The manual consists of many simple techniques that have been found useful for performing travel model validation or for under- standing when a model is behaving realistically. The manual provides a unified structure of validation techniques that can be applied to a fairly broad range of travel model designs (from three- or four-step models to activity-based models, regardless of platform). Validation refers to a set of adjustments to model param- eters and corrections when errors in coding are encountered that produce an improved system-wide statistical fit to a set of observed data. Ideally these data are separate and distinct from the data used to estimate a model in the first place. Both model inputs and outputs can be validated. Each step in the model has its own comprehensive set of validation checks: • Amount of travel (trip generation), • Trip distribution or location choice, • Mode choice, • Vehicle occupancy, • Time of day, and • Traffic assignment. It is noted that the validation and reasonableness checking processes did not have a process for validating traffic speeds. The manual lists troubleshooting strategies for improving model performance when a model is quantitatively or intui- tively incorrect. The manual encapsulates considerable prac- tical knowledge on how to build and execute a travel model and reflects an understanding of how to build confidence in technical analyses. The manual makes specific mention of NCHRP Report 255 in a section called “Acceptable Methods for Achieving Valida- tion Thresholds.” The manual cautions that [NCHRP Report 255] procedures have been used frequently and have helped improve traffic forecasts for project plan- ning and design. The techniques, however, are applied for a specific planning context and are not generally acceptable for all planning studies. In general, the following guidelines should be used to determine acceptable methods for achiev- ing an improved match between modeled and observed travel characteristics: • Model adjustments or refinements are frequently made at a small-area or link level because many regional models do not have the requisite accuracy needed for detailed link level traf- fic forecasts. • The adjustments should reflect transportation supply or trav- eler behavior rather than simple arithmetic, • The adjustments should be reproducible, and • The reasons for adjustments should be clearly documented. 2.4.5 Quick Response Freight Manual II (QRFM II) The Quick Response Freight Manual II (QRFM II), pub- lished by FHWA in September 2007 (10), provides informa- tion and guidance for developing freight vehicle trip tables and general information about freight. As a resource for esti- mating freight vehicle trips at the project level, QRFM II is potentially useful in that it • Provides a framework for employing simple techniques to analyze and forecast freight vehicle trips, • Identifies the available data sources and methods of data collection and validation, and • Provides information on many case studies covering meth- odological and data issues. There are five different methods of forecasting freight vehi- cle trips discussed in QRFM II: • Simple growth factor methods, • Incorporating freight into “four-step” travel forecast, • Commodity models, • Hybrid approaches, and • An economic activity model. QRFM II has abundant information on freight genera- tion rates, techniques, and parameters for freight distribution, mode split, and assignment. While the techniques are mostly useful to develop freight forecast models, these techniques could be applicable to forecasting freight vehicle trips with an existing model and site-specific freight forecast.

18 Validation techniques are also available for trip generation, trip distribution, mode split, and assignment. Case studies provided in QRFM II include the Los Angeles freight fore- cast model, Portland metro truck model, Florida state freight model, and Texas state analysis model. 2.4.6 Institute of Transportation Engineers Trip Generation Manual and Trip Generation Handbook The Institute of Transportation Engineers (ITE) Trip Gen- eration Manual (11) provides trip generation rates for more than 170 land use types and building types. Users can cus- tomize the standard tables of rates by adding local adjustment factors and mixing uses and rates. Ongoing work adds to the database of site-specific traffic and land use information; over 4,000 traffic studies are aggregated for the current edition. ITE procedures estimate the number of trips entering or exiting a site at a given time. The basic relationship is described by mathematical relationships between a dependent variable (either daily or peak-hour trip ends) and independent vari- ables such as gross leasable square footage; number of employ- ees; or land-use-specific variables such as restaurant seats, hotel rooms, hospital beds, and so forth. These mathematical relationships have been developed through numerous studies from which data have been collected and submitted to ITE. Mathematical relationships typically assume the form of either an average trip rate or a regression equation. No guidance is given on future growth of trip generation estimates. This manual helps in conducting site impact studies, deter- mining on-site circulation patterns, forecasting travel demand, performing access management studies, determining traffic signal timing, and conducting environmental assessments. Few travel demand models use ITE rates for trip generation since they treat each site, rather than the vehicle or traveler, as an independent trip generator and typically produce far higher numbers of trips than rates estimated from household surveys. Also, in most cases, ITE rates are not specific to a trip purpose. 2.4.7 NCHRP Synthesis 406: Advanced Practices in Travel Forecasting NCHRP Synthesis 406 (12) describes advanced practices in travel forecasting, including some of the basic elements of tour-based models, activity-based models, land use micro- simulation models, and freight and statewide travel demand models. This publication includes an overview of dynamic traf- fic assignment (DTA) procedures and the advantages that they offer over traditional static assignment methods. A review of advanced practices in a number of metropolitan areas around the country and interviews of more than 30 practitioners who work with, or plan to use, newer methods for travel modeling are the most important elements of this report. The description of linking microsimulation-based activ- ity model systems with DTA and traffic simulation models in NCHRP Synthesis 406 is quite relevant to the practice of project -level traffic forecasting. In addition, the report alludes to the feedback between transportation and land use and notes the importance of considering the effects of project-level improvements on land use and, consequently, traffic patterns. Compared to traditional models, the outputs of activity-based travel models can be more effectively fed into traffic simula- tion models, thus allowing a greater level of fidelity in analyz- ing traffic patterns at the level of individual projects, facilities, or locations. NCHRP Synthesis 406 details the advantages of advanced models for answering policy and modal questions, compared to traditional models. The report also notes that there are several project-level policies (such as high-occupancy toll [HOT] lanes and traffic operations improvements) and land use contexts where the advanced models offer capabilities that the traditional methods do not. Case studies are pro- vided in Chapter 6 of NCHRP Synthesis 406. 2.4.8 Dynamic Traffic Assignment— Dynamic Traffic Assignment: A Primer A number of publications describe the application of DTA, a technique for producing a sequence of time-specific (such as hourly) traffic forecasts. Dynamic Traffic Assignment: A Primer (13) is useful for project-level traffic forecasting. Dynamic Traffic Assignment helps practitioners determine the appro- priateness of DTA for project analysis and provides guidance on how to implement DTA and interpret DTA results. Since DTA is described as being “mesoscopic,” that is, com- bining elements of the macroscopic techniques of regional transportation planning and the microscopic techniques of traffic operational analysis, DTA is suitable for many project- level traffic forecasts. The primary areas useful for project- level traffic forecasting are operational planning and real-time operational control. Operational planning is for making plan- ning decisions on major operations, construction, or demand management actions that are likely to shift transportation patterns on facilities, as well as capacity-increasing strate- gies. Real-time operational control is for large-scale, real-time traffic management and/or information provision problems such as traffic incidents and intelligent transportation systems (ITS). The primary topics of specific interest to project-level traffic forecasting in the reports are the following: • Advantages of DTA over static traffic assignment, • Assignment convergence (achieving full equilibrium assignment) issues,

19 • Data requirements and preparation (e.g., OD trip table estimation from ground counts), • Calibration targets and performance measures, • Error and model validity checking, and • Frequently encountered issues and suggestions for resolv- ing them. 2.4.9 Dynamic Traffic Assignment— Utilization of Dynamic Traffic Assignment in Modeling Utilization of Dynamic Traffic Assignment in Modeling (14) is useful for project-level traffic forecasting. This guidebook describes how to apply DTA within a modeling framework. The first part of the report provides background on DTA modeling principles and software capabilities, and the second part pro- vides guidance on how to apply DTA using traffic models. The second section describes data requirements and techniques for developing, calibrating, validating, and applying a DTA model. 2.4.10 FHWA’s Traffic Analysis Toolbox FHWA has sponsored the development of several analyti- cal tools relevant to traffic forecasting, which are described in Traffic Analysis Tools (15): • STEAM (Surface Transportation Efficiency Analysis Model) is a benefit/cost tool suitable for project or program evalua- tion. Using economic principles, this tool estimates the trans- portation efficiency benefits of a transportation investment and considers capital costs, travel time, operating cost, safety, noise, and emissions costs. STEAM takes input directly from the traditional four-step model, and can post-process traffic assignment outputs for more accurate estimation of congested speeds. • SPASM (Sketch Planning Analysis Spreadsheet Model) is a sketch-planning tool for multimodal corridor analysis and evaluation. As a spreadsheet-based version of STEAM, this tool is only useful for corridor-level analysis, and it is based on economic efficiency analysis of cross-modal and demand management strategies. • SMITE (Spreadsheet Model for Induced Travel Estima- tion) is a spreadsheet application that evaluates the effect of urban highway capacity expansion projects, especially considering induced demand. It is used when the four-step model is unavailable or unable to forecast the full effect of the induced demand. SMITE-ML (Spreadsheet Model for Induced Travel Estimation-Managed Lanes) is a specially modified version of SMITE to evaluate the effects of pric- ing alternatives with managed lanes such as HOT lanes and express toll lanes in urban areas. It has been used as a quick- response sketch-planning tool for pricing policy evaluation. • Intelligent Transportation System Deployment Analysis System (IDAS) is a sketch-planning analysis software tool developed by FHWA to estimate the benefits and costs of ITS investments. It is designed to help public agencies and consultants integrate ITS into the transportation planning process. IDAS evaluates various impacts including changes in user mobility, travel time/speed, travel time reliability, fuel costs, operating costs, accident costs, emissions, and noise as well as benefits (value of time saved, value of acci- dent reductions, and so forth) and benefit-cost ratio. The ITS benefits module reflects values that have been averaged from past studies. 2.4.11 Other Forecasting Guidelines Various traffic analysis tools and methods are reported for multiple project types. The resources include focusing tools, sketch-planning tools (many of them spreadsheet based), OD estimation methodology, and guidebooks for tolling projects and the National Environmental Policy Act (NEPA) process. These resources are discussed below. Network Focusing—A Tool for Quick Response Subarea Analy sis (16) describes the practice of introducing greater net- work and zonal detail into a project area of interest and reduc- ing detail elsewhere. When done properly, this practice speeds model execution time without an unacceptable loss in accu- racy. Subarea focusing is useful when the regional model is seriously deficient for project-level work, such as failing to con- tain traffic-controlled intersections or failing to be dynamic. TCRP Synthesis 66: Fixed-Route Transit Ridership Forecast- ing and Service Planning Methods (134) examines the state of the practice in fixed-route transit ridership forecasting and service planning. This report also explores forecasting meth- odologies, resource requirements, data inputs, and organiza- tional issues. In addition, the report analyzes the impacts of service changes and reviews overall effectiveness of forecast- ing methods based on an extensive survey of transit service agencies and operators. Estimation of Origin-Destination Matrices Using Traffic Counts (17) provides a more recent update of various meth- ods for estimating an OD matrix using traffic counts on links in a transportation network. The paper provides a survey of generic approaches as well as an annotated bibliography of some individual contributions. The author notes that the treatment of congestion effects is an important distinguishing property among the various methods used for OD estimation. There have been three primary approaches to estimating OD matrices: • Traffic-modeling-based approaches, • Statistical inference approaches, and • Gradient-based solutions.

20 Toll Road Traffic & Revenue Forecasts: An Interpreter’s Guide (18) provides practical information that could be used to help interpret toll and revenue projections. Topics of specific interest to project-level traffic forecasting are the following: • Traffic modeling and forecasting, • Traffic risk empirical evidence, • What to look for in a traffic and revenue study, • Often-seen issues and suggestions on how they may be resolved, and • Appendices on traffic risk indices. Other sections of interest in this guide for project-level traffic forecasting are the 20 tips on how to inflate traffic forecasts, tools to aid the “keep it simple” philosophy, and common sources of forecasting error (e.g., unrealized land use growth assumptions, time savings less than anticipated, unanticipated improvements to toll-free routes, less off-peak traffic, and many others). Travel Forecasting Resource (19) is a website in wiki format intended to provide comprehensive coverage of topics related to travel forecasting. The committee has worked on populating topics such as conformity, data, networks, transit, pricing, and climate change. While the resource is under development, var- ious reports are available now, including an assessment of vari- ous methods and tools covering both four-step travel demand models as well as a variety of sketch-planning tools. (http:// tfresource.org/Travel_Forecasting_Resource_-_Home) Interim Guidance on the Application of Travel and Land Use Forecasting in NEPA (20) describes project-level forecasting in the context of the NEPA process. The guidance does not address the need to improve the actual technical methods used to forecast land use and travel behavior as applied to NEPA processes, but tries to fill the gap between the technical guidance for producing forecasts and application of forecasts in the NEPA process. 2.4.12 State-Specific Forecasting Guidelines Over the years, many states have published forecasting guide- lines that set standards for forecasting and serve as resources of tools, methodologies, and data for those states. Due to the com- prehensiveness of these sources, they can provide insight and guidelines for other areas. Several of these guides contain case studies, request forms, spreadsheets/workbooks, and transfer- rable parameters that could be of use to practitioners. Table 2-3 lists 20 published guideline documents that are described in Organizaon Author Document Florida DOT Florida DOT Florida Project Traffic Forecasng Handbook 2012 Minnesota DOT Minnesota DOT Minnesota Procedure Manual for Forecasng Traffic 2010 Ohio DOT Ohio DOT/ Ohio Cerfied Traffic Manual 2007 Texas DOT Texas DOT Traffic Data and Analysis Manual 2001 Florida DOT Florida DOT Traffic Impact Handbook 2010 Kentucky Transportaon Cabinet Kentucky Transportaon Cabinet Traffic Forecasng Report 2008 North Carolina DOT (NCDOT) Stone, Saur & Letchworth Guidelines for NCDOT Project Level Traffic Forecasng Procedures 2002 North Carolina DOT North Carolina DOT Project Level Traffic Forecasng: Administrave Procedures Handbook Oregon DOT Oregon DOT Oregon Analysis Procedure Model 2006 Wisconsin DOT Wisconsin DOT Facilies Development Manual CALTRANS (California DOT) California Transportaon Commission California Regional Transportaon Plan Guidelines Delaware Valley Regional Planning Commission (DVRPC) Delaware Valley Regional Planning Commission 2000 and 2005 Validaon of the DVRPC Regional Simulaon Models Georgia DOT Georgia DOT Design Policy Manual Parsons Brinckerhoff Parsons Brinckerhoff New York Best Pracce Model for Regional Travel Demand Forecasng Utah DOT Utah DOT Utah Project Specific Modeling Info Arizona DOT Arizona DOT Website Contents Ohio DOT Ohio DOT Website Contents Kentucky Transportaon Cabinet (KYTC) Kentucky Transportaon Cabinet KYTC Permit Guidance Manual—Traffic Impact Study Requirements Kentucky Transportaon Cabinet Kentucky Transportaon Cabinet KYTC Design memorandum No. 03 11, Traffic Engineering Analysis Montgomery County (MD) Planning Department Montgomery County (MD) Planning Department Local Area Transportaon Review and Transportaon Policy Area Review Guidelines, 2013 Table 2-3. State and MPO forecasting guidelines.

21 detail in Appendix C. The notable elements of the guidelines published by a selected number of states are described below: • Florida provides guidelines and techniques to forecast traf- fic and assess the impacts of land use on the transportation system. • Minnesota provides a comprehensive, step-by-step pro- cedural manual for traffic forecasting, while the Kentucky Transportation Cabinet provides trip generation rates and data resources to forecast traffic. • North Carolina and Georgia provide procedures for request- ing project-level traffic forecasts as well as guidelines for traffic forecasting. • Oregon provides procedures for conducting long-term analyses of plans and projects. • Wisconsin provides policies and requirements regarding the facilities development process. • California addresses the air quality and land use compo- nents of the planning process. • Delaware and New York provide model development strate- gies, suggested validation targets, and model implementation. • Utah provides traffic forecasting requirements and pub- lishes standards for the tools and procedures used in the traffic forecasting projects. These documents include state-of-the-practice techniques, resources, and case studies for state traffic forecasting. A list of the specific state forecasting guidelines based on the source doc- ument review in the interim report for NCHRP Project 08-83 is presented in Table 2-3.

22 C H A P T E R 3 This section presents an overview of traffic forecasting tools and methodologies. Of course, the most frequently used (and perhaps the best) tool is a well-validated travel demand model. Various aspects of travel demand models are discussed in this chapter. Section 3.6 covers tools and method- ologies used for traffic forecasting other than travel demand models. The traffic forecasting tool/methodology topics include the following: • The travel forecasting model ideal, which, although rarely achievable, is worth identifying as a “gold standard”; • State of the practice of travel forecasting models, covering the current basic travel demand modeling premises and modules, advanced modeling topics, and trend line analy- sis techniques; • State of the practice of data inputs for travel forecasting models, covering socioeconomic data, network data, traffic counts, household travel surveys, origin-destination sur- veys, and freight/heavy vehicles; • State of the practice of data outputs for travel forecasting models, covering volumes, speeds, turning movements, measures of effectiveness, origin-destination information, and post-processors; • Default data versus locally specific data; and • c parameters, which is a discussion of model parameters. Table 3-1 summarizes the topics in a matrix format with the accompanying section number in the chapter. 3.1 The Travel Forecasting Model Ideal The starting point for establishing an ideal standard for fore- casting is to recognize the need to represent traffic flows over large areas and time slices, while representing traffic conditions over small geographic areas and in small time slices. The most feasible approach for achieving this currently is the hybrid travel demand model. A hybrid model involves two or more distinctly different traffic modeling software packages that are executed sequentially, each with its own strengths and weaknesses. The survey conducted for this report, as described in Appen- dix D, shows a modest trend toward the creation of hybrid models for evaluating highway projects. A typical example of a hybrid model would be a conventional four-step travel forecasting package feeding results to a dynamic traffic assign- ment (DTA) package that feeds results to an operations-level traffic modeling package. Some analysts have recognized that it may be necessary also to provide a feedback loop between the last modeling platform and the first, so that delays discov- ered by the traffic operations model can be incorporated into the travel forecasting model. Although possible, such hybrid models are often awkward to implement because of incom- patibilities among the software packages and an inability to ensure convergence to a traffic equilibrium solution. There- fore, hybrid models will likely be applied only to situations requiring highly precise speed estimates that go well beyond the current capabilities of today’s travel forecasting packages. It is possible to conceive of a travel forecasting model- ing package that contains many of the elements of a hybrid model, but is completely self-contained for project-level work and satisfies all the convergence requirements of a planning model. Such a model would resolve most of the issues that have arisen when using conventional four-step models by them- selves and would obviate the need for various refinements. Some practitioners are already developing/implementing soft- ware that is close to the ideal, and therefore do not need to routinely perform refinements on their model outputs. The project-level model ideal has these properties: • For long-term travel forecasts, the ability to estimate demands between all origins and all destinations through behavioral principles. Overview of Traffic Forecasting Tools and Methodologies

Table 3-1. Traffic forecasting tool/methodology topics. Topic Descripon Locaon Traffic forecasng ideal Characteriscs and performance capabilies of a hypothecal, opmal forecasng tool 3.1 State of the pracce in forecasng models Characteriscs and performance capabilies of a typical, good pracce forecasng tool, described for each of the common submodels (trip generaon, trip distribuon, model split, traffic assignment, and reporng) 3.2–3.2.1.6 Enhancements, improvements, and alternaves to standard four step modeling Alternave and/or beer pracce model approaches and capabilies for non acvity based models. Topics include (1) Origin desnaon trip table esmaon from ground counts (2) Intersecon turn penales from delay calculaons (3) Travel me feedback between model steps (4) Mulclass traffic assignment (5) Commodity based freight forecasng (truck component is included) (6) Intersecon delays within standard network modeling approach (7) Car ownership and fleet mix modeling (8) Mulresoluon (detail in me and geography) plaorms (9) Induced travel (10) Enhancement to link impedances in tolls [FORTHCOMING] (11) Esmaon of intersecon and ramp delay outside of standard network modeling approach including simulaon 3.2.2–3.2.2.11 Non highway models, including non motorized models 3.2.3 Advanced topics, including (1) Integrated land used models (2) Spreading of the peak hour/period (3) Time of Day (TOD) choice (4) Tours and tour based models (5) Acvity based models (6) Dynamic traffic assignment (DTA) (7) Travel me reliability (8) Economic modeling (9) Impacts (10) Land use modeling 3.2.4.1– 3.2.4.10 Microscopic traffic simulaon, what it is, how/when to use it; Comparison of measures of effecveness from Highway Capacity Manual (HCM) and microscopic traffic simulaon 3.2.5–3.2.5.3 Data inputs for travel forecasng models Types and sources of data used by travel forecasng submodels: (1) Socioeconomic data (2) Network data (3) Traffic counts (4) Household travel surveys (5) Origin desnaon studies (6) Freight and heavy vehicle data 3.3–3.3.6 Travel forecasng outputs Types of results produced by forecasng and modeling process: (1) Traffic volumes (2) Speeds (3) Intersecon turning movements (4) Measures of effecveness (5) Origin desnaon informaon 3.4–3.4.5 Analysis derived from model outputs: (1) Emissions (2) Economic impact (3) Benefit/cost analysis (4) Bridge and pavement deterioraon analysis 3.4.6 Defaults vs. locally specific parameters Use of data and informaon derived from naonal sources or other non local sources 3.5 Other traffic forecasng tools and methodologies Traffic forecasng that does not rely on full travel demand or simulaon modeling and with modest data requirements: 1) Kentucky Transportaon Cabinet traffic forecasng 2) NCHRP Report 255 sketch planning 3) Origin desnaon matrix growth factoring 4) Time series modes 5) Traffic impact study tools including Instute of Transportaon Engineers trip generaon approaches 6) Elascies 3.6–3.6.6

24 3.2 State of the Practice of Travel Forecasting Models The state of travel forecasting has been described in many textbooks and reports, so this discussion is intended only to establish a baseline that can be referenced later when describ- ing supplementary techniques or enhancements. The reader is directed to NCHRP Report 716 (6), NCHRP Report 365 (7), TRB Special Report 288 (22), and various textbooks for meth- odological details. 3.2.1 Basic Travel Demand Modeling (Four-Step Modeling) There are a few common elements across travel forecasting models, but there are many differences, too. Travel forecasting models can legitimately differ because of planning require- ments, availability of data for calibration and operation of the model, and the philosophy and preferences of the modeler. It is difficult to say that a model is “good” or “bad” by simply looking at an outline of the model steps. Indeed, an evalua- tion of a model can be quite subjective, as indicated by the many peer reviews of metropolitan planning organization (MPO) models in recent years. The following discussion is framed in the context of key con- cepts and components encountered in many basic or baseline travel demand models. In each instance, a “travel forecasting model” is defined as being one that is being used at least par- tially for highway project-level work. 3.2.1.1 Basic Premises Basic premises are the following: • A travel forecasting model is capable of estimating pas- senger and vehicle demand between all relevant pairs of origins and destinations. • A travel forecasting model should be able to calculate pas- senger and vehicle demand from behavioral principles. • A travel forecasting model should be sensitive to those policies and project alternatives that the model is expected to help evaluate. • A travel forecasting model should be capable of satis- fying validation standards that are appropriate to the application. • Validation standards are described in the Travel Model Validation and Reasonableness Checking Manual, Second Edition (9). • The model should be subject to frequent recalibrations to ensure that validation standards are continuously met. • For short-term travel forecasts, the ability to estimate an origin-destination (OD) table, either through analysis of traffic counts or through behavioral principles or both. • For long-term forecasts, the ability to make adjustments to the OD table to reflect differences between base year traffic counts and base year forecasted volumes. • The ability to perform dynamic equilibrium traffic assign- ments, with appropriate feedback to earlier steps, if necessary. • The ability to calculate delays for through and turning move- ments (separately) at intersections, such as signals, stop signs, and roundabouts, in accordance with accepted traf- fic engineering principles, such as those found in the 2010 edition of Highway Capacity Manual (HCM2010) (21). • The ability to incorporate delays from turning movements into traffic assignments. • Sensitivity to traffic control operational strategies, includ- ing work zones. • The ability to apply time-of-day (TOD) factors applied prior to traffic assignment for peak-hour assignments or to create a dynamic OD table for DTAs. TOD factors could be deter- mined through historical data or through behavioral prin- ciples such as a departure-time choice model. Furthermore, it is implied that an ideal project-level model would have these features, which are commonly available in today’s modeling packages: • A fine-grained zone system that eliminates lumpy traffic assignments (i.e., large changes in traffic volumes on con- nected links) on links that are relatively free of congestion and largely eliminates unassigned intrazonal trips. • A high level of network detail that allows for good paths connecting zones to the arterial system and allows traffic to be assigned to streets of lower functional classes. • Multiple vehicle classes to correctly track trucks and to cor- rectly incorporate trucks into estimates of delay. • Multiple driver classes to correctly represent the effects of pricing on different income groups or the effects of other policies that would have varying impacts on different pop- ulation segments. An ideal travel forecasting model would be capable of responding to a variety of planning needs, including transpor- tation systems management (TSM), transportation demand management (TDM), multimodal infrastructure changes, pricing strategies (including tolling), and policy initiatives. The behavioral principles would be robust enough to be sen- sitive to policies that would encourage or discourage travel. An assumption of this report is that the analyst does not possess an ideal project-level model.

25 are the ends of trips at the location where the trip purpose is satisfied (completed): – A commonly accepted method of calculating trip attrac- tions is a set of linear equations of levels of zonal activ- ity. Levels of activity may include number of households at the zone of residence, number of workers by industry at the zone of employment, and school enrollments at the zone of the school. – It is commonly accepted that trip attractions are in units of person trips over a 24-hour period of time. • A travel forecasting model may use “special generators” to account for unique or unconventional trip generators: – Special generators may include productions and/or attractions. – Activities covered by special generators have included universities, military bases, parks, beaches, amusement parks, airports, large shopping malls, and hospitals. – Occasionally, special generators are used at external sta- tions to draw external-internal trips that are not gener- ated by the region’s households out of the model during model validation. – Trips generated by special generators are often static and may impact forecasting results if the forecaster is unaware of their existence. • A travel forecasting model may have the ability to reconcile (or “balance”) differences between totals of trip produc- tions and trip attractions. A single trip must have a pro- duction end and an attraction end. Thus, the total of all productions and the total of all attractions within a trip purpose must be equal. While many practitioners prefer to balance (normalize) attractions to match productions, it is helpful when the model has the option for balanc- ing productions to match attractions. For example, if an imbalance occurs in a small region that relies on importing workers to fill regional jobs, reducing attractions to match productions would not be appropriate. The model should be capable of excluding from balancing productions and attractions at external stations and at special generators. • A travel forecasting model may have the ability to calculate zonal “attractiveness” values in lieu of trip attractions, in which case balancing is unnecessary. Zonal attractiveness is a measure of the size of the zone. It is preferred when either a destination choice model or a singly constrained gravity model is chosen for the trip distribution step. In a singly con- strained gravity model, either the sum of the trips across all columns matches the total number of productions, or the sum of the trips across all rows matches the number of attractions for a zone. In a doubly constrained model, both conditions are met. • Commercial vehicle demand should be incorporated within the model. A typical commercial vehicle component has 3.2.1.2 Trip Generation The following list describes some aspects of a trip genera- tion step: • A travel forecasting model relying on behavioral principles should be organized by trip purpose. There may be one or many trip purposes: – A commonly adopted set of trip purposes, from NCHRP Report 187 (8), is home-based work, home-based non- work, and non-home-based. – Other trip purposes can be added, as needed, by sub- dividing one or more of these three purposes (or travel markets) such as school trips, shopping trips, and rec- reational trips. – In some cases, project-scale travel demand models may focus only on a single trip purpose to analyze a particu- lar kind of travel behavior such as work trips or shop- ping trips. – Trip purposes include both trips from productions to attractions (“to”) and from attractions to productions (“from”). The return-to-home trip purpose has fallen into disuse. – In a “trip-based” model, individual trips are unlinked, but tours are inherently accounted for through the use of the non-home-based trip purposes. – In some models, an effort is made to differentiate trip generation rates by area type or location. This assists with capturing differences in trip generation behavior, par- ticularly along a rural/urban divide, and with addressing unique land use activity patterns, such as mixed-use or transit-oriented-design. • A travel forecasting model may have the ability to calculate trip productions from behavioral principles. Trip produc- tions are trip ends at the location where the need for the trip has been established: – Some trip production calculations are aided by an auto- mobile availability substep. – A commonly accepted method of calculating trip pro- ductions is cross-classification. – Trip productions may be estimated from zonal data on the number of persons per household, number of workers per household, automobile availability, or income either alone or in combination. – Non-home-based trip productions that may have been generated at the zone of residence should be relocated to zones where there are high levels of non-home activities. – It is commonly accepted that trip productions are in units of person trips over a 24-hour period of time. • A travel forecasting model may have the ability to calculate trip attractions from behavioral principles. Trip attractions

26 require that the user supply an initial (or “seed”) OD trip table, as well as traffic counts. There is also no consensus as to how a “seed” OD table may be built; although an historical OD table is recognized as a good starting point, when available. Experimentation and common sense are required for good results. 3.2.1.4 Mode Choice A travel forecasting model may have the ability to split trips according to travel modes: • Two commonly accepted methods of mode split are multi- nomial logit and nested logit. Nested logit models group modal choices with similar characteristics in a separate cal- culation step and overcome the potential of multi nomial models to overestimate trips by such similar modes. Mode split within these analytical methods is often a function of travel time, travel cost, and convenience and socio economic factors. Trips can be split into few or many modes. Poten- tial modes include automobile, truck, bus, rail transit, non- motorized travel, and carpool. • In locations where the plan does not require information on transit ridership, it is possible to use historical factors to apportion trips to the automobile mode. • In addition to splitting between transit and automobile modes, a mode choice model can be used to subdivide automobile modes into single-occupancy vehicles (SOVs) and high-occupancy vehicles (HOVs). These can be fur- ther subdivided by willingness to pay tolls. • A travel forecasting model may have the ability to convert person trips to automobile trips in a step referred to as automobile occupancy. A commonly accepted method of determining automobile occupancy is to apply fixed his- torical occupancy factors to person trips in order to obtain vehicle trips. Automobile occupancy factors can, option- ally, vary by TOD. 3.2.1.5 Trip Assignment A travel forecasting model is capable of assigning highway traffic to road segments (links): • This process involves summing the number of vehicles between each origin and destination that traverse a road segment. • The most common division of vehicles in traffic assign- ment is between automobiles and trucks; although some models further subdivide trucks into light, medium, and heavy duty. In travel demand models that explicitly count trucks during the highway assignment process, a steps similar to a passenger component. Many models have commercial vehicle components similar to the truck model described in the Quick Response Freight Manual II (10). A true freight model—that is, a model based on com- modity flows—is still unusual in urban locations because of the difficulty of obtaining the necessary data. 3.2.1.3 Trip Distribution A travel forecasting model may have the ability to calculate the number of person trips between origins and destinations for each trip purpose: • OD person trips may be obtained with a doubly constrained gravity model. • OD person trips may be obtained with a destination choice model or a singly constrained gravity model: – Gravity model friction factors may be calculated with a “gamma” function, an exponential function, a power function, or an empirically derived table of numbers. – Gravity models primarily use impedances between origins and destinations as a measure of spatial sepa- ration. Impedances are expressions of total travel costs that incorporate the effects of travel time, travel costs, and distance. The units of impedance are theoretically unimportant, but impedances are often expressed in units of minutes. – Destination choice models are often formulated as multi nomial logit models. – Through trips, that is those passing entirely through the study area, are most appropriately handled by supplying an empirically derived external-external (E-E or X-X) vehicle trip table between external stations. • When using either a gravity model or destination choice model, it may be possible to find composite impedances across modes by taking the log-sum of highway and transit impedances (and any other available modes). • A travel forecasting model may have the ability to appor- tion demands into time periods shorter than one full day. A commonly accepted method of TOD factoring is to use historical fixed percentages of trips to and from the pur- poseful end of trips. Factors are most conveniently applied just after trip distribution, but they can be applied after trip generation or after mode split. • In the absence of the ability to estimate demand from behavioral principles, a travel forecasting model should be able to calculate vehicle demand from traffic counts, turn- ing movements, or other on-the-road empirical evidence. There are many available methods for estimating vehicle trip tables from traffic counts, and there is no consensus as to which method is best. Commonly adopted methods

27 • Person trips from a mode other than the automobile mode (e.g., transit) may be assigned to a modal network. • Traffic from a commercial vehicle component may be added to automobile traffic when assigned to the network: – Heavy truck traffic is commonly weighted by PCE fac- tors, similar to those found in HCM2010 (21), when calculating delays. – Some models track trucks separately from automobiles during the assignment step through a process known as multiclass traffic assignment. • Traffic assignments should achieve a close approximation of equilibrium conditions. This means that at the conclusion of the traffic assignment process, travel times or impedances on alternative feasible paths between origins and destina- tions will be very nearly equal. Without achieving equilib- rium or nearly equilibrium conditions, the results from the traffic assignment process will not be nearly as realistic in replicating real-world conditions. 3.2.1.6 Outputs/Reporting The following output/reporting functions are desirable for travel forecasting models: • A travel forecasting model should be capable of report- ing results in a form that is sufficiently detailed, precise, and diagnostic to understand the impacts of a project on the street system and provide reliable information to understand the need for and efficacy of additional road- way capacity and to calculate measures of effectiveness, as appropriate. • Outputs should be consistent with each other and with equilibrium conditions on the network, if an equilibrium traffic assignment technique is used. • A travel forecasting model should be able to generate traffic volumes by link, turning movements at important inter- sections, and delays (or loaded travel times) by link, for any time period of interest. • A travel forecasting model may provide diagnostic reports about the trip patterns that make up a traffic assignment. Such diagnostic reports include vehicle trips in and out of each zone, select link analysis, select zone analysis, and link- to-link flow analysis. • A travel forecasting model may provide diagnostic reports about the calculation of system demands. Such diagnostic reports include trip productions, trip attractions, average trip lengths, and trip length distributions. 3.2.2 Four-Step Modeling Enhancements The following modeling topics are considered to be signifi- cant enhancements to four-step travel demand modeling, but passenger car equivalent (PCE) factor will be used. The PCE is used to convert trucks into an equivalent number of passenger cars for the model’s congestion calculations and for pavement thickness calculations for highway design. • A basic model includes only “static” traffic assignment, which assigns all traffic within a single time period to the network in a single loading. • If congestion is not anticipated or cannot be determined with good precision, then highway traffic volumes may be estimated with an all-or-nothing traffic assignment or a “multipath” assignment method that has the ability to split trips across many paths between any given origin and destination. • If congestion is anticipated, then highway traffic volumes may be estimated with a method that satisfies principles of traffic equilibrium. Acceptable methods of equilibrium traffic assignment include the method of successive aver- ages (MSA) and Frank-Wolfe decomposition. • If congestion is moderate to severe, then delays that are found in the assignment step may be fed back to any earlier steps that use highway travel times, particularly mode split and trip distribution. It is commonly accepted that feed- back processes should be run until an equilibrium state is achieved throughout the whole modeling sequence. Spe- cifically, this includes achieving a correspondence between the input travel times used for trip distribution and the travel times output from the traffic assignment model. This feedback process might require several iterations. A well-established way of obtaining consistent highway travel times throughout the model is the MSA, although other, less-efficient iterative schemes are possible. The MSA produces an unweighted average of many all-or-nothing traffic assignment volumes. Each successive all-or-nothing traffic assignment uses delays that are computed from the previous iteration’s average. • Delays (or re-estimated travel times) may be calculated for all roads in the region. Delays may occur along road segments (links) or at intersections (nodes). Delays along road segments are commonly estimated by a “volume- delay function” (VDF), which uses the estimated volume- to-capacity (V/C) ratio. VDFs can vary by functional class. VDFs are sometimes used for delays at intersections, but more sophisticated methods that are sensitive to conflict- ing and opposing traffic levels are growing in popularity. For example, some models use relationships drawn from HCM2010 (21). • Path building may be made sensitive to turn restrictions and turn penalties that are specified ahead of time. A com- monly accepted method for properly accounting for turn restrictions and penalties is vine building.

28 each origin zone and each destination zone that start at a particular time. Dynamic OD tables require dynamic traffic counts and a dynamic seed OD table. An important consideration, often missed by practitioners, is the amount of error inherent in the traffic counts. Algo- rithms are fully capable of tightly fitting the measured traf- fic counts, even though traffic counts have their own errors, some of which are substantial. A particular application of synthetic OD table estimation is the refinement of an OD table that has been created within a travel forecasting model. In such an application, the model’s OD table is adjusted to match, or almost match, ground counts. Additive or multiplicative factors are retained so that they can be applied to any forecast done later with the model. 3.2.2.2 Turn Penalties from Delay Calculation Mesoscopic models are able to differentiate delays by turn- ing movements, as shown in the operational analysis proce- dures of HCM2010 (21). It is possible to incorporate those delays in subsequent path buildings within a model, so that especially long delays, say from left turns at signals, can be accounted for in short- est paths. 3.2.2.3 Feedback to Earlier Steps Early implementations of four-step models were often criticized for doing the steps in a particular order and never resolving inconsistencies between those steps. Most troubling were inconsistencies in the path travel times that were used for trip distribution and mode split when compared to the output travel times from the traffic assignment step. In addi- tion, travel models with an integrated land use step had travel time inconsistencies when allocating activities to zones. Work on “combined” models from the 1970s, particularly a well- known study by Evans (137), showed that it was possible to neatly resolve those inconsistencies. The idea of resolving path travel time inconsistencies has been dubbed “feedback,” to suggest an iterative process where delays from traffic assignment are incorporated into earlier steps in the model. Many modelers found it diffi- cult to implement Evans’s theories directly into their soft- ware, so many modelers relied on unsophisticated heuristic algorithms until it was demonstrated that MSA could find the same solution as Evans’s method. MSA is iterative and follows Evans’s theories in overall structure, but replaces a complex optimization substep with an elementary averag- ing of traffic volumes from all previous all-or-nothing traffic assignments. not quite in the category of advanced modeling. These topics include the following: • OD trip table estimation from ground counts, • Turn penalties from delay calculations, • Feedback, • Multiclass traffic assignment, • Commodity-based freight component (truck component is included), • Intersection delays, • Car ownership modeling, • Multiresolution platforms, • Induced travel, and • Estimation of node delay. 3.2.2.1 Origin-Destination Table Estimation from Traffic Counts Research on OD table estimation from traffic counts dates from the 1970s, but a clear consensus has not been reached as to how it should be accomplished. Because of the topic’s importance, numerous software packages have introduced their own algorithms for finding synthetic OD tables. Unfortunately, different algorithms will find different solutions, so the process is not usually repli- cable across software packages. The idea is simple: find a reasonable OD table that will reproduce known traffic counts. On large networks, there are many different OD tables that will reproduce traffic counts with equal quality, so there is a need for additional informa- tion to help choose an OD table. Most algorithms available today supplement the traffic counts with a “seed” OD table that is a best-guess approximation of the desired result. A seed OD table may be one that has been observed in the past, one that has been observed recently but imprecisely, or one developed from principles of driver behavior. Most of the published methods for OD table estimation formulate the algorithm as a constrained optimization prob- lem. That is, they find the best solution to maximize some index of quality or minimize some index of error. Inherent or explicit weighting is done as a compromise between fitting the “seed” OD table well and fitting the traffic counts well. The most rigorous algorithms are computationally inten- sive and can require a very large amount of computer memory. Optimization problems require searching for a solution, and the time necessary to achieve a solution increases rapidly with the number of variables. In the worst case, there can be as many variables as the number of cells in the OD table. OD table estimation can also be made dynamic. In this case, a dynamic OD table gives the number of trips between

29 The main hindrance to building a commodity-based model in urban regions is the absence of commodity flow data for zones smaller than counties. 3.2.2.6 Intersection Delay in Traffic Assignment Means for calculating intersection delay in travel forecasting have been rapidly evolving. Older travel forecasting models estimated delay only on links with an elementary mathemati- cal relationship between travel time, volume, and capacity. Such relationships are referred to as VDFs. There are several forms of VDFs, but the most prevalent VDF is the “BPR curve.” (“BPR” stands for the Bureau of Public Roads, which ceased to exist in 1967 and was the predecessor agency to FHWA.) VDFs are insensitive to levels of conflicting and opposing traffic, so they do poorly at estimating delays at intersections. They are also insensitive to queuing due to congestion. Some newer travel forecasting models have moved toward implementation of sophisticated delay estimation procedures to produce more realistic estimates of travel times through intersections. The complexity of intersection delay relationships adds to the data requirements of the model. Intersections need to be described carefully in terms of their lane geometry and sig- nalization timing. It is possible, and often desirable, to simu- late signalized intersection timing so that inputting full timing information can be avoided. Intersection delay relationships interfere with some popu- lar methods of finding equilibrium solutions, especially those that formulate traffic assignment as an optimization prob- lem, such as Frank-Wolfe decomposition. None of the opti- mization formulations in network assignment algorithms can handle situations where delay on one link is affected by traffic on other links. Current optimization formulations require a VDF in only one variable (namely link volume), and the VDF must be presented in a form for which an integral can be con- veniently and quickly found. Intersection delay relationships do not satisfy these two requirements. A major impetus for the adoption of better methods for intersection delay was the publication of the 1985 Highway Capacity Manual (26) that contained, for the first time, delay estimation procedures for signalized and one-way and two- way stop intersections. There are other ways to calculate intersection delay, but the wide acceptance and availability of the Highway Capacity Manual (HCM) in the traffic engineering community has made it a particularly attractive source document for cal- culating delay in travel models. The variety of intersections that can be handled by HCM procedures has increased over the years. The HCM2010 (21) has means for calculating delays for one- way stops, two-stops, all-way stops, signals, and roundabouts. Delays, thus link and node travel times, are calculated from the averaged volumes. According to Wardrop (24), the equi- librium conditions for a traffic assignment are reached when no travelers can improve their origin-destination travel times by changing paths. MSA obtains a solution that satisfies the conditions of “user-optimal” equilibrium traffic assignment, as described by Wardrop (24), and ensures complete consis- tency between path travel times across all steps. 3.2.2.4 Multiclass Traffic Assignment Multiclass traffic assignment, which separately tabulates volumes for each vehicle class, is a modest extension of stan- dard traffic assignment algorithms. Vehicles can be classified by body type (passenger cars, trucks, buses, etc.), by occu- pants (SOV, carpools, etc.), or by characteristics of the driver that would suggest different path-finding behaviors. Multi- class traffic assignments are usually used only when it is sus- pected that path choice behavior differs significantly across classes. Typical applications of multiclass traffic assignments are determining heavy truck volumes on roads, determining the impact of road pricing, and determining the utilization of HOV lanes. 3.2.2.5 Commodity-Based Freight Component There are two methods of forecasting freight traffic within travel models: truck based and commodity based. Truck-based components are most often seen in urban regional models. Commodity-based components are often seen in statewide models. The development of a commodity-based model was described succinctly in 10 steps in the Guidebook on Statewide Travel Forecasting (25). • Obtain freight modal networks. • Develop commodity groups. • Relate commodity groups to industrial sectors or economic indicators. • Find base year commodity flows. • Forecast growth in industrial sectors. • Factor commodity flows. • Develop modal cost for commodities. • Split commodities into modes. There are three categories of methods for splitting commodities into modes: – Mode split models. – Tables. – Expert opinion. • Find daily vehicles from load weights and days of operation. • Assign vehicles to modal networks.

30 clean vehicle penetration in the marketplace, policies aimed at influencing vehicle purchase decisions of consumers can be analyzed using vehicle fleet composition models. (There is a distinction between fleet mix in the context of modeling for air quality analysis and truck percentages used for an actual forecast. The analyst needs to be aware of the subtle implica- tions of the difference.) Recent developments in the formulation, estimation, and application of the multiple discrete-continuous extreme value (MDCEV) model offer considerable promise in the ability to model and forecast vehicle fleet composition (defined by mix of vehicle body types, fuel types, and vintage) for transporta- tion policy analysis (31). Corridor-specific strategies that aim to promote use of alternative fuel vehicles (e.g., allowing alter- native fuel vehicles with a single occupant to use the HOV lane) can be analyzed with respect to their potential to influence vehicle ownership and usage patterns among the population. 3.2.2.8 Multiresolution Platforms Practically speaking, there is little distinction between hybrid models and multiresolution platforms. Hybrid mod- els involve two or more modeling platforms that have fun- damentally different ways of obtaining their results, such as mixing a macroscopic travel forecasting model with a traf- fic microsimulation. Multiresolution models involve two or more modeling platforms that have different levels of preci- sion. However, the need to achieve highly detailed network flows usually coincides with the need for highly accurate delay estimations that may only be achievable with microsimulation techniques. See the work of Shelton and colleagues (32) and Burghout and Wahlstedt (33) as examples. The concept of multiresolution, itself, deals with the issues of spatial scale and network precision. Mulitresolution plat- forms are particularly attractive when highly precise results are desired, but the scale of the study area is too large for all of the software steps at the finest level of detail. Multiresolution models enable the following: • Multiple network structures in varying levels of precision, • Detailed spatial representation (known as spatial decom - position), • Detailed temporal representation (known as temporal decomposition), and • More precise forecasts than can be achieved with a coarser model. A particular form of multiresolution platform involves spatial detail. An example of this is the interface between a statewide model and MPO/regional models, in which the statewide model has a streamlined node, link, and zone con- Delays can be calculated separately for each lane group, which allows for separate movements and phases to be analyzed. These procedures are formulated as a variational inequality problem for which solution algorithms have been developed. A troubling matter, however, is that it is entirely possible for there to be more than one equilibrium solution. The presence of multiple solutions is considered a minor issue for most plan- ning studies, but it can distort the comparison of almost similar alternatives that might arise within traffic operations studies. MSA is a traffic assignment method that does not rely on optimization and can find solutions to the equilibrium traf- fic assignment problem when delay on one link is affected by traffic on other links. VDFs remain popular in planning models for estimating delays on uninterrupted facilities. 3.2.2.7 Car Ownership and Vehicle Fleet Mix Modeling Car ownership or vehicle ownership has been a key explan- atory variable affecting estimates of travel demand in various model components. Car ownership is an explanatory vari- able that appears in numerous trip production models and mode choice models, and more recently, in several destina- tion choice models as well. It is sometimes viewed as a sur- rogate for household income, a variable that is often difficult to measure accurately in travel surveys. As car ownership is itself a function of several socioeconomic and demographic variables, various types of models of car ownership have been developed and incorporated into travel demand model sys- tems in numerous metropolitan areas. Car ownership models may take the form of the following: • Linear regression models, • Count models such as poissons and negative binomial models, • Unordered discrete forms such as the multinomial logic and nested logic models, or • Ordered discrete forms such as the ordered profit and ordered logic models (27, 28, 29). These models are capable of predicting vehicle ownership distributions in zones as a function of other socioeconomic and demographic characteristics (such as household size, income, number of workers, number of children, and age). More recently, with the recognition that car ownership has reached a point of saturation for several years now, attention has turned to modeling not only car ownership but also the fleet composition or mix (30). Vehicle fleet mix has important implications for energy and air quality analysis, and with the recent interest in enhancing

31 While induced travel may result exclusively from changes in network level of service measures due to a project-level improvement, it is also important to consider the longer term effects arising from land use changes triggered by enhance- ments in network accessibility (38). In the longer term, new land use developments may come into existence following an improvement in highway facilities. Changes in land use devel- opment patterns will, in turn, inevitably lead to changes in activity-travel patterns across the spectrum of travel choices (activity generation, destination choice, mode choice, and route choice). Integrated land use–transport models that account for the cyclical relationship between land use and transporta- tion are capable of accounting for such longer run effects as long as the land use forecasting models are sensitive to mea- sures of network performance. According to Kuzmyak and colleagues (38), individual studies have generally reported elasticities of vehicle miles traveled (VMT) with respect to roadway capacity of +0.1 to +0.9. There continues to be considerable debate surrounding the magnitude of induced travel effects as evidenced by the split opinion reported in TRB Special Report 245 (39). There do not appear to be stan- dard methods or tools to account for possible induced travel effects in project-level forecasting efforts. The application of elasticity values reported in the litera- ture may provide a basis on which to apply adjustments to project-level forecasts to account for such effects (for example, the forecast from a model may be amplified by 10 or 20% to account for possible induced travel effects). As activity-based microsimulation travel models continue to gain ground in a variety of project-level analysis contexts, it is envisioned that standard methods or tools to quantify induced travel demand will evolve to meet local needs. 3.2.2.10 Estimates of Node Delay Model Application. In a model network, a node joins two or more links. Nodes represent the physical intersec- tion of streets or roads. In a highway system, traffic control is used to assign right-of-way at these junctions, and char- acteristically there is delay associated with traffic control devices. Most travel demand model software platforms have the capability to compute and/or assign node delay. The delay can be calculated internally within the modeling software or computed/estimated externally using other tools and methods and applied as part of the trip distribution and traffic assign- ment steps. Whether computed internally or externally, node delay can be applied as travel time penalties of separate values assigned to individual intersection movements—left turns, through movements, right turns, and so forth. figuration that allows for the insertion of one or more local area networks with no additional network coding. A two-way interface between those models allows the state- wide model to provide external flows for the regional models and allows the regional models to provide better impedance values for the statewide model. This same concept has been used experimentally for purely local models, where the urban area has been decomposed into many small networks and care is taken to ensure consistency of flows across all the sub- networks (34). Models with spatial decomposition can make efficient use of multicore computers. 3.2.2.9 Induced Travel The notion of induced travel is often a key consideration in developing project-level forecasts. Induced travel refers broadly to the additional traffic that a facility will experience follow- ing the implementation of a project or policy that improves travel time or reduces generalized travel costs. Induced travel may arise from a number of sources that have been identified in the literature (35, 36, 37). When a particular facility is improved, thus reducing travel time or cost on the facility, some travelers may divert from other routes to the improved route, thus increasing traffic vol- umes on the improved facility. Changes in destination choice may also result from a project-level roadway improvement as travelers may be able to visit more preferred destinations farther away without experiencing longer travel times follow- ing the improvement to a facility. Yet other travelers may shift mode choice, choosing to use the automobile mode (from transit or carpool) in light of the reduced congestion that often results from a project-level improvement. All of these shifts result in increased traffic volumes on the improved facil- ity and constitute induced traffic. The shifts identified above are often well captured by travel demand models. Destination choice or trip distribution models capture changes in destina- tion choice patterns in response to changes in system capacity, mode choice models capture shifts in mode usage, and traffic assignment models account for route shifts. One of the sources of induced travel that is not captured well by current four-step travel demand models (albeit with a few exceptions) is that of net new trips that are undertaken by travelers following the improvement of a facility. Most trip generation models in four-step models are not sensitive to accessibility measures or other network level of service variables; as such, changes in travel time or cost do not impact trip generation estimates. More recent tour-based and activity-based models, on the other hand, have begun to incorporate approaches whereby activity and tour generation are sensitive to accessibility measures (often represented by the log-sum terms of mode choice models downstream).

32 applied to model traffic assignments. For each approach leg and the intersection as a whole, an aggregate node delay can be computed as the volume-weighted average of control delays for approach links to the node. 3.2.2.13 Signalized Intersections The HCM method for signalized intersections represents average control delay for all vehicles during an analysis period, including those vehicles still in a queue when the analysis period ends. For an approach lane group, control delay is expressed as follows: + +d d d d1 2 3= where d = control delay (s/veh), d1 = uniform delay (s/veh), d2 = incremental delay (s/veh), and d3 = initial queue delay (s/veh). The uniform delay term, d1, is based on the original model developed by Webster (40) to represent delay under uniform arrivals and departures. As an alternative to the Webster method, the “incremental queue accumulation” procedure is offered to compute a more accurate uniform delay term where traffic moves in platoons, where movements occur over multi- ple green periods, and where there are movements with multi- ple saturation flow rates (e.g., protected-permitted left turns). The incremental delay term, d2, accounts for (1) delay due to the effect of random cycle-by-cycle fluctuations in demand that occasionally exceed capacity and (2) delay due to sustained oversaturation during the analysis period. The initial queue delay term, d3, accounts for additional delay incurred due to an initial queue at the beginning of the analysis period that is caused by unmet demand in the previ- ous period. All three terms incorporate the ratio of demand traffic vol- ume to capacity for intersection approach lane groups. As the demand V/C ratio increases, approach delay increases mod- estly in a linear form until the V/C ratio approaches 1.0, then increases rapidly beyond that point. Intersection delay also is a function of traffic signal timing parameters such as cycle length, phase duration and sequenc- ing, treatment of left turns (protected, permitted, or protected- permitted phasing), and coordination with other signals. Variations in these parameters can have significant effects on intersection delay. Requiring a complete signal timing plan can be onerous, especially for a planning application where timing parameters are unknown or must be assumed. The HCM provides a quick estimation technique that can be used to develop reasonable timing plans for estimation Demand models typically aggregate total delay on a link basis, where the total delay equals the mid-block delay along the link plus the control delay associated with the node at the downstream end of the link. Control delay can have a significant effect on travel time and route choice, but it is often ignored at the node level in travel demand models. As delay is a component of total travel time and a key parameter in route choice, node delay should be considered in the traffic forecasting process. Types of Node Delay. Node (intersection) delay occurs in one of two ways: • Control delay, a result of traffic control devices; and • Geometric delay, in association with the physical charac- teristics of an intersection or junction. Traffic control devices causing node delay include signals, STOP signs, and YIELD signs. In addition to being located at street intersections, traffic signals can be placed on free- way entrance ramps to meter flow, thus resulting in an asso- ciated delay. Geometric delay is delay caused by the geometric features of a facility that cause drivers to reduce their speed in nego- tiating the facility. Geometric delay is relatively small when compared to control delay and is typically ignored. 3.2.2.11 Tools for Estimating Node Delay The tools most commonly used to estimate control delay are the HCM [an updated version was published in 2010 (21)] and microscopic traffic simulation; however, sometimes there are special cases where conventional tools or methods don’t apply. The HCM, microscopic traffic simulation, and special cases are discussed below. 3.2.2.12 HCM The HCM provides procedures to estimate node delay for • Signalized intersections, • Unsignalized intersections (two-way STOP-control [TWSC], all-way STOP-control [AWSC], and roundabouts), and • Interchange ramp terminals. The HCM defines control delay as the increase in travel time due to a traffic control device. On an approach link, it includes delay associated with initial deceleration, queue move-up time, stopped time, and final acceleration. HCM methods estimate control delay for individual approach lane groups at an intersection, and these can be

33 3.2.2.15 Interchange Ramp Terminals Interchange ramp terminals represent a special case of at-grade (service) interchanges of freeways with surface streets. The HCM contains a methodology that addresses interchanges with signalized intersections, interchanges with roundabouts, and impacts of interchange ramp ter- minals on closely spaced downstream intersections. It does not address oversaturated conditions, nor does it specifi- cally address alternative configurations like diverging dia- mond (double crossover) or continuous flow interchanges. The HCM method recognizes that interchanges operate as a system and that there are operational effects associated with the close spacing of ramp terminal intersections with the surface street. Example interchange configurations include diamond interchanges and two-quadrant partial cloverleaf interchanges. The distance separating the two intersections and/or the proximity of adjacent downstream intersections can have a significant effect on queuing and delay, especially under con- gested conditions. For signalized interchanges, average control delay for each lane group and movement is estimated using the HCM sig- nalized intersection methodology. Average control delay for each movement through the interchange is estimated as the total delay experienced by the OD combination that defines the movement. If the OD passes through both intersections, its average control delay is the sum of delays experienced along its path. 3.2.2.16 Microscopic Traffic Simulation Microscopic traffic simulation basics are discussed in more detail in Section 3.2.5. Whereas HCM methods for estimating node delay are deterministic (i.e., they involve no randomness and always produce the same result for a given set of inputs), microscopic simulation is stochastic and accounts for inher- ent randomness in driver decisions and actions such as car following, lane changing, and gap acceptance. To account for this randomness, multiple simulation runs are made and average delays are computed to produce a “typical” estimate. On an approach link, delay is generally defined as the excess travel time spent on a roadway segment compared with free- flowing travel time on a link representing zero-delay con- ditions. Node delay from microscopic simulation models includes control delay and geometric delay. Total delay on the link includes the excess travel time plus node delay at the downstream end of the link. Some simulation models employ HCM methods for esti- mating control delay at intersections. Where this is the case, model documentation should be reviewed and/or model devel- opers should be consulted to identify any possible departures from the HCM methods that may be incorporated into the simulation. of node delay. A detailed documentation of the method for an operational analysis is located in Volume 3, Chapter 18 of HCM2010. 3.2.2.14 Unsignalized Intersections There are three types of unsignalized intersections: TWSC, AWSC, and roundabouts. The HCM TWSC method is based on gap-acceptance theory and includes the following basic elements: • Size and distribution (availability) of gaps on the major street, • Usefulness of these gaps to minor street drivers and their willingness to accept them, and • Relative priority of various movements at the intersection. Capacity for STOP-controlled approaches is computed as a function of the conflicting flow rate, the major movement critical headway (i.e., the minimum headway or gap that drivers are willing to accept), and follow-up headway for the minor movement. Control delay is computed as a function of the demand flow rate and capacity. The TWSC method also can be applied at two-stage cross- ings, where the major street is divided by a median or two-way, left-turn lane (TWLTL) that allows minor street drivers to cross or enter the major street in separate, staged movements. The HCM does not include a detailed method for estimat- ing delay on YIELD-controlled approaches, but the TWSC method can be applied at YIELD-controlled intersections with appropriate changes to key parameters (e.g., critical headway and follow-up headway). The HCM AWSC method recognizes that every driver approaching the intersection must stop before proceeding and that the decision to proceed is a function of traffic condi- tions on the other approaches. Node delay is computed as a function of the service time (difference between the departure headway and move-up time) and the degree of utilization, which is the product of the arrival rate and mean departure headway. The method also recognizes that giving right-of- way to the driver on the right doesn’t always occur, and it includes a probability of conflict model in its computation of capacity and delay. The HCM roundabouts procedure applies to roundabouts containing one or two circulating lanes. The methodology incorporates a combination of lane-based empirical regres- sion models and analytical gap-acceptance models for single- lane and double-lane roundabouts. Roundabout control delay is computed similarly to TWSC control delay, with a modi- fication to account for YIELD-control on the approach leg, which does not require drivers to come to a complete stop when there is not conflicting traffic.

34 ramp metering is to achieve a smoother flow (and faster travel speeds) along the freeway. 3.2.3 Non-Highway Models Increasing demands to forecast demand for a broader spec- trum of the transportation system have contributed to a grow- ing interest in modeling non-automobile modes of transport, including bicycle and pedestrian modes, high-speed rail, light rail, and airplane. Traditional travel demand models have been used to fore- cast travel demand by various modes of transportation. Mode choice models that have been estimated using revealed pref- erence travel survey data, or stated preference data (if revealed preference data are not available and cannot be collected), may be applied to determine the total demand for non-highway modes of travel. High-speed rail and air transportation mar- kets comprise long distance travel demand; statewide models and special studies may be conducted to estimate long dis- tance travel demand, and mode choice models may be used to estimate the demand for these modes of transportation. In the urban context, light rail is a non-highway mode for which considerable attention has been paid to forecasting demand due to the scrutiny that light rail projects receive through the New Starts and Small Starts processes at the fed- eral level (41). It is important to collect appropriate data that can form the basis for estimating mode choice models, and specific atten- tion should be paid to the relative magnitude of the alternative specific constants in the utility functions for various modes. If there are competing bus- or express-bus-lines that serve the same OD pairs as the light rail mode, then the use of light rail for a particular trip may constitute as much a path choice as it does a mode choice. The treatment of non-motorized modes of transport (bicycle and pedestrian) generally differs from that for other mechanized means of transportation due to natural and phys- ical constraints that prevent the use of non-motorized modes for all types of travel. Bicycle and pedestrian trips tend to be shorter in length, and although they are often lumped together into a single non-motorized category, there are important dif- ferences between bicycle and pedestrian trips that should be recognized when evaluating bicycle- and pedestrian-oriented projects. The estimation of non-motorized mode travel demand may be accomplished using a variety of approaches. A good overview of various approaches is provided in the Guidebook on Methods to Estimate Non-Motorized Travel published by FHWA in 1999 (42). On the demand estimation side, several possible approaches are adopted to estimate non-motorized mode usage. Com- parison studies allow the estimation of demand for a new 3.2.2.17 Special Cases Special cases sometimes exist where conventional tools or methods like the HCM don’t easily apply. While the HCM TWSC method can be applied to estimate node delay under YIELD-control, it may be more appropriate to use simulation for cases like a merge from an entrance ramp to a congested freeway. While there might be no delay at this junction under low- flow conditions, there would be associated delay with this merge when the freeway is congested that might play a part in driver route choice (and thus in forecasted traffic). For this example, the simulation model would be run for (1) low flow and (2) congested flow conditions on the freeway. The delay difference between the scenarios could be assigned as node delay (under congested freeway conditions only) at the freeway ramp/mainline junction. Ramp metering represents another special case where delay is intentionally introduced under specific traffic flow condi- tions. For this case, the ramp (link) capacity is set to the meter rate in vehicles per hour. Alternatively, the HCM chapter on advanced traffic management provides guidance for estimat- ing capacity for single-lane and two-lane metered ramps. For this application, two possible scenarios exist with respect to the ramp demand volume-to-capacity ratio: 1. When the ramp demand volume is less than capacity (V/C < 1), queues occasionally form due to randomness of vehicle arrivals, or 2. When the ramp demand volume is greater than capacity (V/C > 1), queue storage occurs over an extended period of time. Deterministic queuing analysis can be performed to esti- mate node delay for these scenarios. Requirements for queu- ing analysis include the following elements: • Mean arrival rate, • Arrival distribution, • Mean service rate, • Service distribution, and • Queue discipline. There are several published sources that document pro- cedures for queuing analysis and estimating delay, including those that apply to toll plazas. The analyst is advised to select the one that most directly applies to the case at hand. Alternatively, microscopic traffic simulation can be used to estimate delay associated with ramp metering. When using simulation, estimation of delay should include the entire trip (within the analysis area) for all affected origins and desti- nations and not just the localized delay, as the objective of

35 3.2.4.1 Integrated Land Use—Transport Models Integrated land use—transport models are used to rep- resent the interactions between land use changes and travel demand (43, 44). Generally, land use characteristics serve as inputs to transport models with demand estimates influenced by variables representing the built environment. It is now well recognized that land use and transportation interact in a cyclic manner with changes in transport system attributes affecting land development patterns over time (45). Land use models generally involve some of the following elements: • Models of housing development, residential location choice, and associated building stock; • Models of business or commercial development, business location choice models, and associated non-residential building stock; and • Models of land values and building stock prices that help establish market prices and bid rents. Most disaggregate microsimulation-based land use models incorporate a series of choice models capable of simulating the location choices of individual households, businesses, and other agents in the urban system (46, 47). These models attempt to replicate the market mechanisms that exist in the real world. Land use models are generally responsive to changes in transport accessibility measures. Changes in modal level of ser- vice attributes contribute to the growth or decline in land use development and influence the mix of land uses that will occur. Thus, integrated land use—transport models incorporate feedback loops where travel time and accessibility measures feed back into land use models to reflect the effects of trans- portation supply attributes on land use dynamics (47, 48). While some suggest that this iterative, cyclical process should be executed until “equilibrium” is reached, others argue that the urban system is in a constant state of flux and that feedback processes between transport and land use mod- els actually capture the longitudinal dynamics in urban system evolution over time. In any event, integrated land use— transport models are increasingly finding their way into main- stream practice and are used for forecasting traffic associated with individual projects (49). 3.2.4.2 Peak Spreading Peak spreading, an adjustment in the temporal character- istics of travel in response to worsening traffic congestion, is a phenomenon that is observed in major metropolitan areas around the world (50). As peak-period traffic conges- tion worsens, travelers tend to shift their time of departure to the shoulders of the peak period in order to experience pedestrian or bicycle facility by examining usage patterns at similar peer facilities with socioeconomic and land use contexts that compare well with the proposed new facility. Sketch-planning or simple spreadsheet models can be used to implement comparison studies or other approaches that are not data intensive—such as elasticity-based methods or use of descriptive characteristics from large travel surveys similar to the National Household Travel Survey (NHTS) that provide trip lengths, rates, TOD distributions, and trip purpose distributions for non-motorized trips. Full-scale travel models often include a mode choice step capable of providing estimates of non-motorized travel. These models tend to be elaborate and potentially data intensive as the esti- mation of a mode choice model that includes non-motorized modes as explicit choice alternatives requires that adequate non-motorized trip samples exist in the travel survey used for model development. Non-motorized travel demand may also be estimated using off-model approaches; a separate model that is specific to non- motorized mode usage may be developed and applied outside of the regional travel demand model. Such a model would estimate non-motorized trip generation and perform trip dis- tribution specific to non-motorized trips. Microsimulation models are beginning to incorporate the ability to simulate pedestrian movements such as in downtown environments. An important issue associated with non-motorized and transit travel demand estimation is that there is substantial potential for induced demand. When a new light rail line or non-motorized mode facility (sidewalk or bicycle path) is built, people may be induced to make new transit or non- motorized trips that they did not make previously. One poten- tial way to estimate the induced demand is to introduce the change in accessibility created by the new facility into a land use model that will estimate the change in development quan- tity and mix. The increased development can be imported into the model and will produce additional person trips. 3.2.4 Advanced Travel Demand Modeling Topics This section covers the most frequently cited and used areas of advanced travel demand modeling topics: • Integrated land use, • Peak spreading, • TOD choice, • Tours and tour-based models, • Activity-based models, • DTA, • Travel time reliability, • Economic modeling, and • Land use modeling.

36 Each of the above methods has certain advantages and dis- advantages, and caution is to be exercised in deploying TOD choice models, particularly in the context of transit model- ing, as transit service tends to vary by the time of day. Cur- rently, most TOD models split trips after trip distribution. TOD choice models have generally focused on predicting trip departure times. More recent work in modeling TOD choice has focused on modeling departure time based on pre- ferred arrival time, expected and experienced travel time, and the notion of schedule arrival delay penalty (disutility). Tour- based models that jointly model arrival time and departure time from a tour primary stop are also being implemented. 3.2.4.4 Tours and Tour-Based Models Tours are a series of interlinked trips that are chained together. Depending on the definition used to describe tours, they may (or may not) be considered synonymous with trip chains. For example, one may consider a journey from home to work with an intermediate stop to drop off a child at school or pick up a cup of coffee as a tour. On the other hand, one may argue that a tour should be a closed chain, in which the origin of the first trip of the tour and the destination of the last trip of the tour are the same. Regardless of the defini- tion that one adopts, the important aspect to note is that the concept of “tours” involves a recognition that trips are not independent entities; rather, there are spatial, temporal, and modal interdependencies across trips in a tour. Due to the importance of this interdependency in model- ing travel demand and the increasing prevalence of complex trip chaining patterns in the real world, there has been a move toward the development of tour-based models. In tour-based models, the tour is the unit of analysis (as opposed to a trip), and many choice processes are modeled at the tour level. Some of the key model components in a tour-based model include (but are not necessarily limited to) the following: • Tour generation models (frequency of tours by type—such as work tours and non-work tours); • Subtour generation models that capture smaller tours that take place within the context of a larger tour (for example, a tour from work to eat lunch); • Primary activity and intermediate stop destination choice models; • Tour mode choice, subtour mode choice, and, where appli- cable, stop-level mode choice; • TOD choice model for tours, subtours, and intermediate stops; and • Tour accompaniment models to capture joint trip making versus solo trip making. The idea behind these models is that one will be able to cap- ture constraints and interdependencies across travel choices shorter and/or more reliable travel times (less variance on experienced travel times). As a result, and especially in con- gested areas, the peaks in diurnal distributions of travel tend to widen and flatten over time (51). Increasing flexibility with work schedules and the ability to work anytime or anywhere are further contributing to peak spreading. Peak spreading may be accounted for in several ways in the context of travel modeling (52): • The use of post-processing techniques where simple peak- spreading factors obtained from traffic observations or household surveys are applied to determine the amount of traffic that remains within the peak period versus the amount of traffic that shifts to the shoulders. • The use of peak-spreading factors within the four-step travel demand model (prior to the assignment step) to determine the travel demand associated with the peak period and the peak-period shoulders separately. Separate assignments are carried out for these different periods to reflect traffic volumes that result from the peak-spreading phenomenon. • The use of more stand-alone-type, peak-spreading models that often take the form of TOD choice models (discrete choice models). These TOD models are sensitive to changes in system attributes and the socioeconomic characteristics of travelers in a region. They may be introduced as a sepa- rate step in the four-step travel modeling process (53). 3.2.4.3 Time-of-Day Choice As concerns about alleviating peak-period traffic conges- tion increasingly dominate the attention of policymakers, there is a growing interest in the deployment of TOD choice modeling techniques capable of accurately representing and capturing the diurnal distributions of travel demand. Many travel demand management strategies and pricing measures are specifically targeted at reducing peak-period traf- fic and alleviating congestion. The measures aim to influence the temporal patterns of travel demand and the choices people make with respect to the scheduling of their activities and trips. TOD choice modeling techniques may be of two basic types: • TOD factors can be derived from observations of traffic volumes or from household travel survey data. These fac- tors can then be applied at various stages of the four-step travel modeling process to determine travel demand by TOD. These factors are based on observed data and are not sensitive to any explanatory variables. • TOD choice models may be estimated using household travel survey data. These choice models may take the form of discrete choice models (multinomial logit models) where the number of alternatives is equal to the number of choice periods of interest.

37 In contrast, tour-based models—although also incorpo- rating heuristics and rules to some extent—tend to be deeply nested logit model systems with a series of log-sum terms feeding up the chain to account for interdependency across choice processes. 3.2.4.6 Dynamic Traffic Assignment DTA refers to a class of mesoscopic traffic simulation methods in which travel demand between origins and des- tinations is routed through a network in a time-dependent way, and movements of individual vehicles are simulated to capture various traffic phenomena, such as delays, queues, and bottlenecks. DTA procedures need time-varying or time- dependent travel demand estimates. These estimates may be in the form of TOD trip tables or in the form of trip lists with time stamps. One benefit of DTA is that it allows one to measure delays, queues, and other traffic phenomena as they change over time. DTA models involve two major elements. The first is a rout- ing element in which trips are routed from each origin to each destination along time-dependent, shortest paths. As traffic builds on a network, the shortest path between an OD pair may change, and there may be multiple shortest paths that are chosen by different travelers to execute their trips. The time-dependent shortest path algorithms built into DTA models accomplish two tasks. First, shortest paths are constantly updated as the day evolves and traffic volumes build on the various links in the network. As traffic volumes build up, the travel times on links are updated based on macroscopic speed-flow relationships (and not based on microscopic simu- lations). Second, the shortest paths are computed in a time- dependent way. In other words, the shortest path reflects the fact that the travel time on a link downstream in the path may be different by the time the traveler reaches that link as opposed to when the traveler actually started the trip at the origin. DTA can be used to refine OD travel demand using feed- back. Updated skim trees from a DTA model can be fed back into the demand steps to re-estimate OD demand, and this iterative process can be continued through feedback loops until convergence is achieved. 3.2.4.7 Travel Time Reliability Travel time reliability is a newer enhancement to travel models that takes into consideration the uncertainty of reach- ing a destination in a predetermined amount of time. Research has shown that drivers perceive uncertainty as a separate com- ponent of trip disutility (or impedance). The more common formulations within disutility expressions use the standard deviation of trip time as the indicator of the amount of uncer- tainty. Uncertainty can theoretically affect destination choice, mode choice, or route choice. for trips that belong to a tour, thus providing the ability to better estimate behavioral response to a wide range of policy measures. 3.2.4.5 Activity-Based Models Activity-based models are often considered synonymous with tour-based models. These two terms are often used inter- changeably. Activity- and tour-based models are generally implemented via a microsimulation framework. In a microsimulation framework, a synthetic population of the entire model region is generated using statistical proce- dures that employ census data sets. The activity- and tour-based models are then applied to the entire synthetic population to simulate activity-travel patterns of each and every person in the synthetic population, effectively returning individual activity-travel records that are similar to household travel sur- vey records. These activity-travel records can then be aggre- gated into trip tables for traditional static network assignment procedures or they can be fed into DTA procedures. Activity-based models are sometimes distinguished from tour-based models on the basis of the focus on the continuous time representation of activity engagement patterns of indi- viduals in activity-based models. These models consider time to be an all-encompassing entity in that activity durations are explicitly modeled. In contrast, the TOD choice of activ- ity engagement and travel episodes in activity-based models is determined by modeling the activity and travel durations along the continuous time axis. Activity-based models start the simulation of a daily activity- travel pattern for an individual at the beginning of the day and then sequentially simulate activity after activity and travel epi- sode after travel episode, to build an entire activity-travel pat- tern for the day in an “emergent” manner. The activity-travel episodes that come later in the day are influenced by activity- travel episodes that took place earlier in the day, thus capturing daily history dependency in activity-travel engagement. Activity-based models often include the following components: • Activity type choice models (activity generation models), • Activity duration models, • Constrained destination choice models, • Constrained mode choice models, • Activity accompaniment models, and • Work and school activity schedule models. Some activity-based models also introduce a greater level of heuristics and rule-based behavioral principles to bring about consistency in activity-travel patterns, simulate certain choices in a more qualitative way, and account for interdepen- dencies in activity-travel choices across household members.

38 grams or systems. The jobs-created-per-dollar investment figure sometimes heard around discussions of proposed pro- grams or infrastructure improvements are often derived from the types of economic impact analyses discussed here. There is no industry standard or broadly accepted approach to conducting an economic impact analysis of transporta- tion projects and programs. This lack of consistency and, in some cases, rigor, creates challenges for assessing the merits of different project assessments conducted by different teams. The FHWA has produced some guidance on economic impact analysis for its Transportation Investment Generating Eco- nomic Recovery grant program and for general applications (54). Prior NCHRP publications have provided a framework for transportation economic impact analyses as well (55). Foundation. Over the years, a number of studies have attempted to isolate the relationship between infrastructure spending—and more specifically, transportation spending— and economic outputs. In 1994, Nadiri and Mamuneas pub- lished a widely cited paper on the positive relationship between publically financed infrastructure investment and manufactur- ing productivity and performance (56). The paper also found a significant variation in the impact by industry. A 2004 paper found a positive, though diminishing, rate of return for high- way capital investment for consumers and producers (57). A 2011 Rand Corporation meta-analysis of 35 papers describing the relationship between public infrastructure expenditures (including highway investment) and economic outcomes also found a positive relationship. These and other studies have cited a number of factors that may contribute to an apparent variability in outcomes, especially at the subnational level. Such factors include the regional economic and geographic compo- sition of private industry and households, the maturity and coverage of the existing transportation system, and the existing level of demand in the region (58). Analysis Framework. Transportation investments can create competitive advantages for a region by decreasing travel costs, increasing travel reliability, and increasing travel choices. This means that households can enjoy free time not spent in travel or use time and cost savings to consume goods and services of greater benefit to the regional economy. It also means that they may have access to more opportunities for work, as well as health care, shopping, and recreation. Businesses may be able to produce goods and services more cheaply. Businesses may also be able to reach a more desirable market, have access to a greater variety of inputs for the goods and services they produce, and enjoy lower cost shipping opportunities. Regions with these advantages can retain and attract business and household growth more readily. These consequences, which are highly dependent on the economic and geographic context within which the investment is made, lead to increased business activity and job creation. Adding an uncertainty term to the destination and mode choice steps is reasonably straightforward, but adding uncer- tainty to route choice has presented some challenges to modelers. The issue with route choice is that standard algorithms for finding shortest paths do not perform well when there is uncer- tainty in link travel times because the standard deviation of path travel time is not the simple sum of the standard deviations of all the component (link and node) travel times. Suggested methods of overcoming this limitation have been finding the k-shortest paths, even though only one path is being selected, or iterating to improve the fidelity of the path over several trials. 3.2.4.8 Economic Modeling Transportation Application of Economic Impact Analysis. Economic impact analysis is a special applica- tion of macroeconomic analysis, in which the benefits and costs associated with transportation investments are stated in terms such as jobs, income, and gross regional product (GRP). The principal role of traffic forecasting in estimating the long-term economic benefits of transportation invest- ments is to estimate changes in VMT, vehicle hours traveled, and changes in accessibility. Often, economic impact analy- ses require specialized transportation analyses or adapta- tions of existing transportation methods. Economic impact analysis extends and reinterprets user benefit analysis in terms of private-sector efficiency benefits and, more broadly, in terms of regional and national economic competitiveness. Economic impact analyses may consider con- sequences such as relocation, hiring, and increased or decreased household spending. These types of consequences transcend the monetary value of changing travel time and vehicle oper- ating and safety costs at the level of directly impacted origin and destinations. Long-term economic impacts are distinguished from short- term construction impacts. Construction impacts include the additional jobs created in building new roads or creating other new transportation services. This type of economic impact creates additional spending that cycles through the regional economy through direct wages, the spending of wages, and the additional jobs that this spending may create. In contrast, effective transportation investments reduce travel costs and increase accessibility, and the ripple effects of these impacts are long-lasting. It is these latter impacts that economic impact analysis tries to capture. A common application of an economic impact analysis is to communicate the value of large public expenditures in terms that are readily understandable to decision-makers and the public. Because of the cost and complexity of economic impact analyses, their application is mostly limited to evalu- ations of proposed new highway corridor alignments, large transit systems, or communicating the value of existing pro-

39 ing, it is useful to understand the types of travelers that use the roadway system in sufficient detail to assign values that correspond to the economic use of travel. Travel markets can include personal travel, commuter travel, on-the-clock automobile travel (travel on the way to conduct business), on-the-clock truck travel, tourism travel, and other leisure travel. If a travel demand model does not include all the passenger travel markets needed for the analysis, data from local household interview surveys and national surveys, such as the NHTS, may supplement the analysis. Freight travel may be separated from truck travel and further disaggregated by com- modity and/or industry, using sources such as the Transearch database and the Vehicle Inventory and Use Survey (VIUS). Estimating Changes in Other Transportation Costs. Total maintenance, operating, and accident costs require unit costs that are applied to miles traveled and do not require dis aggregation into travel markets. Generally, unit values for vehicle types such as automobile and single and combi- nation unit truck suffice. There are numerous academic and practitioner-level data sources offering suggested values for these unit values; one of the most widely used sources is the AASHTO Red Book. Improvements to transportation networks and modal options can also lead to an increase in lower cost shipping, ware- housing, and delivery configurations, for a given good. Logis- tics models can analyze these options, although they are highly complex and are not commonly used in regional analyses. Industry-Specific Costs and Input/Output Models. Input/output models provide information that an analyst can use to translate transportation costs into industry-specific costs. Transportation is one of many inputs that industries use to produce their goods and services. What types of inputs and how much of each input a particular industry uses to pro- duce a dollar of output are expressed in an input/output table. The U.S. Bureau of Transportation Statistics (BTS) has devel- oped a special type of input/output table, the Transportation Satellite Accounts (TSA), which describes how much trans- portation is used as an input by industries. The TSA includes transportation needs that are met by a private industry’s own transportation fleets as well those provided by other private carriers (see Figure 3-2). Accessibility Impacts. The reach of business services is influenced by the transportation supply and demand charac- teristics of each region. Geographic information system (GIS) techniques can be used to calculate the size of the labor market or customer market that falls within a normal delivery or trip time, which affects the ability of businesses to attract business activity. Access plays a role in productivity as well, and these effects are sometimes captured in production function models. A typical economic impact analysis captures one or more of these effects, in terms of economic outcomes, as shown in Figure 3-1. Transportation investments create economic impacts whose magnitude can be estimated using a combination of tools and conversions that capture the linkage among trans- portation service and infrastructure spending and long-term economic growth. In broad terms, three steps in this estima- tion can be identified: (1) developing transportation impacts, (2) developing intermediate impacts, and (3) developing eco- nomic impacts. 3.2.4.9 Developing Transportation Impacts Estimating Vehicle Miles Traveled and Vehicle Hours of Travel Impacts. The primary outcome of the transporta- tion impact analysis is an estimate of the change in mobility created by a transportation investment (i.e., a project). Prop- erly specified travel demand models or analytical approaches will capture many of the immediate effects of capacity and mobility enhancements. These effects include, in descending order of likelihood: (1) changing routes (trip assignment), (2) changing depar- ture times (departure time choice), (3) changing mode of travel (mode choice), and (4) changing origin or destination (trip distribution/destination choice). These impacts can be captured by travel demand models, extra-model estima- tion methods, or some combination of the two. The result of these effects is a difference in VMT and vehicle hours of travel (VHT) in comparison to baseline (i.e., without the invest- ment) conditions. Estimating Intermediate Impacts. Often, a substantial amount of post-processing is required to convert changes in VMT and VHT into a form usable by a macroeconomic model. These conversions produce transportation cost changes spe- cific to industry types and other sectors of the economy. Value of Time and Disaggregating Travel Markets. Travel between each origin and destination represents a vari- ety of economic values of time. For economic impact model- Transportation Impacts Travel Time Costs Vehicle Operating Costs Safety Costs Environmental Costs Intermediate Impacts Costs of Doing Business Access to Markets Household Costs Other Travel Market Costs Economic Impacts Employment Income Population Gross Regional Product Figure 3-1. Illustrative impact framework.

40 3.2.4.10 Land Use Modeling Travel demand in a region or along a corridor is strongly influenced by the changes in land development patterns that occur over time. A host of land use variables serve as inputs for travel demand models, and the accuracy of project-level traffic forecasts is inextricably tied to the accuracy with which land use variables are predicted into the future. Land use measures may include housing and population by dwelling unit type, firms and employees by industry sector, parks and recreational spaces, and economic indicators such as prices and rents for residential or commercial space. It is generally believed that land use defines the five “Ds” that are impor- tant determinants of human travel behavior and activity choices: • Density, which is generally measured in terms of units (or employees or inhabitants) per unit area; higher density of development is generally associated with greater levels of non-motorized and transit mode usage. • Diversity, which represents the mix of land uses that are present in a zone or spatial unit; a greater mix of land uses is considered conducive to transit use, non-motorized mode use, and shorter trips. • Destination accessibility, which focuses on the ease with which alternative destinations may be reached by various modes of transportation; this aspect also incorporates net- work design aspects because grid networks are considered to provide greater accessibility than cul-de-sac type designs. • Design, which focuses on the design of the built environ- ment and the provision of green space; for example, wide sidewalks, bicycle pathways, and designs in which store- fronts and building facades are close to the street are con- sidered conducive to promoting bicycling and walking. Industries that rely on specialized inputs that are not read- ily substituted are more sensitive to changes in access than are businesses and industries that can readily substitute the inputs they use to produce their goods and services. This idea has been explored in NCHRP Report 463: Economic Implica- tions of Congestion (59). Estimating these cost and productiv- ity effects by industry requires information about travel flows and the inputs various industries use to produce their goods and services. Estimating Economic Impacts. At the intermediate evaluation stage, transportation costs are translated into industry and household costs in a way that is consistent with the economic structure of a region. The economic impact analysis inputs these changes in industry and household cost structures to estimate the broader benefits to the regional economy. Regional Macroeconomic Models. Economic models estimate the advantage gained or lost by an economic region from a change in costs, compared to other regions. Some of these models estimate the impact of changes in business and household costs on factors that influence each other, such as the price of goods, the demand for goods and services, the demand for and cost of labor, and business activity in a region. The outcome of these model estimates are GRP, income, and employment. These models may be dynamic, in that they show how changes in one factor influence the others over time. An example of this is how increased economic activity might cause wages and prices to increase. Other models focus on the economic relationships between industries and con- sumers in a static way, using input/output tables or regression- like functions that show the relationship between costs and productivity. Figure 3-2. Transportation cost as a share of output. Source: BTS, RITA.

41 3.2.5 Microscopic Traffic Simulation Some forecasts require information about the detailed interactions of traffic in small time increments over short dis- tances (such as at driveways and intersections) and so require the use of tools more detailed than travel demand models. Complex tools like microscopic traffic simulation models have emerged to fill this need. Microscopic traffic simulation is the simulation of individual vehicles moving through a roadway system. This kind of simulation incorporates math- ematical models for basic relationships that simulate driver behavior related to: • Car following, • Lane changing, and • Traffic stream entry gap acceptance. Vehicles typically enter a roadway network based on a statis- tical arrival distribution and are tracked through the network over intervals of time (these are usually small; e.g., 1 second). At each point of origin into the network, vehicles are assigned a destination, a vehicle type, and driver type. By their nature, traffic microsimulation models are stochastic—they incorporate the randomness reflective of real-world uncertainty and variability into the modeling process. Mathematical models typically use probability distribu- tions and random number generators to initiate network entry and other events. Vehicles then “obey” the rules for car follow- ing, lane changing, gap acceptance, and right-of-way as they are defined in the models. Model variability increases when operational parameters like actuated signal control also vary. Because they are stochastic, traffic microsimulation models require multiple runs to define the likely range of results within which performance measures (delay, travel speed, etc.) may fall. The number of required model runs (i.e., sample size) varies, depending on the level of congestion in the network and the desired error or level of confidence about the mean value for the performance measure in question. Statistical analyses to derive the mean, standard deviation, and desired confidence intervals are needed to estimate the likely range of results, due to variability. The required num- ber of runs for model calibration may be different than for model application, depending on the desired error for each. 3.2.5.1 Guidelines on the Use of Microscopic Traffic Simulation Tools There are several widely used, commercially available microsimulation programs in use. They differ in the ways that they apply models to simulate driver behavior. There is also a considerable amount of published literature on the various • Distance to transit, which measures the ease with which transit stops, terminals, and stations can be reached from anywhere in space. By developing forecasts of the housing and employment markets, land use models provide the critical inputs and built environment descriptors needed for accurately model- ing travel demand. Using accurate measures of land use is a necessary condition for producing realistic travel demand forecasts. Land use forecasting processes come in a variety of forms. There is increasing use of land use models in the profession, largely motivated by the increasing availability of accurate land use data (in electronic databases), particularly at the parcel level. Since land use is highly dependent on local zon- ing policies and regulations, building permits, and economic cycles, there is a continued use of Delphi methods in the development of future land use projections. However, these methods are being increasingly informed by computational models that use a series of economic and socio-demographic projections, existing land use development patterns, data on developable/vacant land, and bid-rent price equations to determine the likely patterns of land use change that will occur over time. A review of land use models is provided by Wegener (45). Early land use models, such as the Lowry model (60), took the form of aggregate spatial distribution models and used gravity- based approaches to allocate housing and employment to var- ious sectors across space. Another well-known aggregate land use model system is that comprising the DRAM (disaggregate residential allocation model) and EMPAL (employment allo- cation model) models (61); this model system and variants of it continue to be used in a number of jurisdictions. MEPLAN is another extension of the aggregate land use model series, where housing location is modeled with greater rigor and behavioral validity by recognizing the trade-off between housing prices and transportation costs. More recently, there has been considerable work in the development of disaggregate land use models that simulate the dynamics of housing and employment markets in a detailed fashion; examples include, but are not limited to, UrbanSim (46) PECA (62), TRANUS (63), RELU (64), ILUTE (48), and LEAM (65). Many of these models use discrete choice modeling methods to simulate choice processes, hedonic bid-rent functions to pre- dict market prices, and market clearing mechanisms to fore- cast land use change over time. There are a growing number of case studies in which these land use models have been inte- grated with transport models (both four-step travel demand models and activity-based microsimulation models) such that land use changes are sensitive to changes in network condi- tions and accessibility measures over time (64).

42 has been coded correctly. Analysts’ expectations also should be field verified. It is possible that some residual errors may exist after error checking steps have been performed. These may be due to software limitations or an error in the software itself. The error checking step should be performed prior to calibration. Calibration. Calibration is the adjustment of model parameters to improve the model’s ability to reproduce driver behavior and traffic performance characteristics. It is necessary because no single model can accurately account for all pos- sible traffic conditions and because models must be adapted to local conditions. Calibration is a process that involves adjusting model parameters so that local traffic conditions are reasonably reproduced. Calibration involves the adjust- ment first of global parameters, then local (link-specific) parameters. A three-step strategy is recommended: 1. Calibrate capacity parameters, 2. Calibrate route choice parameters, and 3. Calibrate overall model performance. A calibration objective function (mean square error or MSE, for example) should be chosen and the analyst should seek to minimize the error between model output and field measurement. Application and Evaluation. Alternatives analysis includes several steps: 1. Development of baseline demand forecasts, 2. Generation of project alternatives, 3. Selection of measures of effectiveness, 4. Model application (runs), 5. Tabulation of results, and 6. Evaluation of alternatives. Microscopic traffic simulation models, like travel demand models, include traffic distribution and assignment steps typi- cally associated with travel demand models. Different soft- ware platforms perform these in different ways with varying levels of complexity. At one end of the spectrum, vehicles enter the network and are assigned as they approach each node, based on user-input turning volumes or percentages. This occurs at each successive node, where the vehicle assign- ment is repeated as a stochastic process of the pre determined turning movement proportions. For this approach, the “dis- tribution” is the accumulation of vehicles at the network exit nodes. At the other end of the spectrum, several simulation pro- grams perform a trip distribution–traffic assignment process similar to that employed in travel demand models. These pro- cesses may include the capability to perform equilibrium-based traffic assignments that model flow under near-saturated or oversaturated conditions. Advanced microsimulation models aspects of microsimulation. One notable source is the FHWA Traffic Analysis Toolbox Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software (66). Guidance on data collection, model development, error checking, calibra- tion, application, and evaluation of results from this source is summarized below. Data Collection. The specific data required by a micro- simulation model will vary, depending on the software being used and the model application. Required data typically include geometric data (number of lanes, lane use, turn bay lengths, posted speed limit, grade, etc.), traffic control data (signal timing plans, for example), demand data (traffic entry volumes or counts, turning movements, and OD tables), and vehicle classification information. In addition to basic input data, microsimulation models require data on vehicle and driver characteristics. These data often are difficult to mea- sure in the field and default values typically are provided in the software. Additional data typically are collected for calibrating the model. These data include travel times, turning movement counts, observed queues, and saturation flow rates. Model Development. The blueprint for constructing a microsimulation model is the link-node diagram. The dia- gram identifies which streets and highways will be included in the model and how they will be represented. Nodes represent the intersection of two or more links and typically are located using x-y (and sometimes z) coordinates. They represent an intersection/junction or a change in link geometry. Physical and operational link characteristics input into the model include number of lanes, lane width, link length, grade, and curvature. Other link-based parameters that may be entered include pavement condition, sight distance, and locations of bus stops, crosswalks, or other pedestrian facilities. Traffic control data are entered at the node level and include type of control (no control, YIELD signs, STOP signs, traffic signals, and ramp meters). Other model development elements include traffic operations and management data, traffic demand data, driver behavior data, event and scenario information, and simula- tion run controls. Error Checking. Error checking a simulation model is essential so that the calibration process does not result in dis- torting model parameters to compensate for coding errors. Error checking should be performed in three sequential stages: (1) software error checking, (2) input coding error checking, and (3) animation review. Software error checking should be performed to identify the latest known “bugs” and work- arounds. Input coding should be checked for the link and node network, demand input, and traveler behavior and vehi- cle characteristics. The animation then should be reviewed to ensure that vehicle behavior is as expected and that the network

43 3.3 State of the Practice of Data Inputs for Travel Forecasting Models Four-step travel demand models make use of a variety of input data sources. These data inform the model develop- ment process and establish the descriptive characteristics of the region being modeled. This section discusses the state of the practice with regard to model input data. 3.3.1 Socioeconomic Data Socioeconomic data describe the population and economic activity occurring in an area and are correlated with the mag- nitude, location, and mode of travel demand. These data are placed into a database organized by traffic analysis zones (TAZs), small units of geography that subdivide the model area. Each record in a TAZ database represents a unique TAZ in the model and contains the socioeconomic data required by the model. TAZs are significant in the modeling process since all trip ends are generated at the TAZ level, and all trips travel from one TAZ to another. 3.3.1.1 Demographics Demographic data are the key independent variables in trip generation models. Socioeconomic and demographic information is relatively easy to obtain through sources such as the decennial Census and the American Community Sur- vey (ACS). Similarly, a significant portion of the behavioral data that are used to define trip-making characteristics can be obtained through household travel surveys. This con- nection between trip-making behavior and socioeconomic and demographic characteristics is widely regarded as being one of the most documented and robust aspects of travel demand models. Typical variables include households, dwelling unit type, occupancy and vacancy rates, automobile ownership, income, and population. These variables are described below. • Household. The household typically serves as the base unit upon which all other demographic characteristics are built. Attributes such as population and automobile ownership are often expressed as persons per household or vehicles per household. The typical cross-classification trip gen- eration model used in many conventional travel demand models will stratify households according to some combi- nation of variables, such as population in the household, income, automobile ownership, and/or dwelling unit type. • Dwelling unit type. This variable is used in some areas with high seasonal variability in dwelling unit occupancy or with a mix of residential building types as the basic unit of socioeconomic information in trip generation. Dwelling may include DTA functionality, where paths between OD pairs vary in real time as a function of congestion and traffic control. 3.2.5.2 Microsimulation and Highway Capacity Manual Measures of Effectiveness Microsimulation and the HCM employ significantly differ- ent computational procedures to estimate delay, congestion, and other measures. Therefore, when using either methodol- ogy, analysts should carefully document and cite the source of the method used to estimate a measure of effectiveness. Generally, HCM-based performance measures are the result of macroscopic, deterministic methods. Vehicular demand is quantified in terms of an average or maximum flow rate (in vehicles per hour) and measures such as speed and density are representative for all vehicles traversing a facility during the analysis period. Microsimulation uses a trajectory analysis of individual vehicles to define and esti- mate performance measures. The current logic is that any comparison of results between the two approaches is possible only through analysis of vehi- cle trajectories as the “lowest common denominator.” The HCM advises that trajectory-based performance measures can be made consistent with HCM definitions through field measurement and calibration. Operational performance measures fall into five basic groups: • Speed-related measures, • Queue-related measures, • Stop-related measures, • Delay-related measures, and • Density-related measures. Table 3-2 provides a comparison of the five groups of performance measures and the basic computational differ- ences between the macroscopic deterministic methods and microsimulation trajectory-based methods. These are for uninterrupted-flow facilities like freeways and rural highways and for interrupted-flow facilities like urban streets and sig- nalized intersections. In addition to similar performance measures computed by HCM-based methods, some microsimulation tools also estimate additional environmentally related performance measures such as fuel consumption and emissions. For more information on the differences between HCM- based performance measures and those computed by micro- simulation tools, the user is advised to consult the FHWA Traffic Analysis Toolbox Volume I: Traffic Analysis Tools Primer (67). In Section 11.6 of this report, Case Study #6—Blending a Regional Travel Forecasting Model with a Traffic Microsimu- lation illustrates the use of a microscopic traffic simulation model for developing traffic forecasts.

44 Determinisc Methods (e.g., HCM) Vehicle Trajectory Based Methods (e.g., Microsimulaon) Speed Related Measures Average speeds are computed on the basis of free flow speed and determinants such as demand volumes, weaving speeds, proporon of heavy vehicles, grades and link delay (where applicable). For some facility types, there are different procedural methods for undersaturated and oversaturated condions. Speed and travel me related measures are treated together because they are closely related. The average speed of an individual vehicle is computed by dividing the segment (link) length by the travel me. Space mean speed for all vehicles traveling on a segment is esmated by dividing the number of vehicle miles of travel for the segment by the number of vehicle hours of travel me. Queue Related Measures Measures are defined for both interrupted and uninterrupted flow facilies. Queues may be defined in terms of the number of vehicles in a queue or the distance of the last vehicle in the queue from the end of the segment (i.e., back of queue). The probability of the back of queue reaching a specified point where it will cause operaonal problems (e.g., turn lane spillback) is of parcular interest. Microsimulaon tools have the ability to produce queuing measures more robust than those produced using HCM based methods, but these measures are difficult to compare to the HCM. Of parcular importance is what defines a queued state and this definion typically varies across various microsimulaon tools. Vehicles generally are considered to have le‡ the queue when they have le‡ the link on which they entered the queue; thus, the link definion is an important parameter. Microsimulaon models are able to establish an instantaneous back of queue at each point in me, so the queson of how to process these instantaneous values in a meaningful manner is parcularly important. Queue length analyses are treated differently for (1) undersaturated noncyclical operaon; (2) undersaturated cyclical operaon; and (3) oversaturated operaon, either cyclical or noncyclical. Stop Related Measures Measures include an esmated number of stops on an approach or stop rate (in stops per mile), computed using determinisc procedures. Most microsimulaon tools provide their own definion for what constutes a stopped state, including when a stop begins and when it ends. It is important for the analyst to understand how the stopped state is defined for the tool at hand and to what extent the parameters can be adjusted to be consistent with the HCM. Esmang the number of stops can be problemac, depending on the microsimulaon tool, in that it generally relies on varying arbitrary thresholds. The accumulaon of mulple stops (i.e., subsequent stops a‡er the first one) poses problems with microsimulaon models, and it is difficult to compare values produced by various microsimulaon tools and the HCM. Delay Related Measures There are mulple definions and thresholds for delay across the various computaonal methods. For uninterrupted flow facilies, delay is computed as the difference between free flow speed and calculated operaonal space mean speed for undersaturated condions. For oversaturated condions, the space mean speed is esmated from the prevailing density on a segment. For interrupted flow facilies, control delay is defined as the delay that results from a traffic control device (compared with an uncontrolled condion). In microsimulaon models, delay is generally defined as the excess me spent on a roadway segment (link) when compared with a me at an ideal speed represenng zero delay condions. Various simulaon models have different definions of this ideal or target speed. Delay may be aggregated (usually expressed in vehicle hours) or unit delay (usually expressed in seconds per vehicle). With respect to vehicle trajectories, delay elements include stopped delay (me a vehicle is actually stopped), queue delay (which reflects the me spent in a queue), control delay (delay resulng from a traffic control device), and segment delay (delay experienced by each vehicle upon leaving the upstream node). Density Related Measures Density is expressed in terms of vehicles per mile per lane. For undersaturated uninterrupted flow condions, it is computed by dividing the adjusted flow rate by the esmated speed. For oversaturated condions, it is determined by queue tracking procedures defined in the HCM. Density is not reported for interrupted flow facilies. Density is simply the sum of vehicles on a roadway secon at a specific me. The queson, therefore, is how to apply the definion of density to the proper roadway secon at the proper me. If the microsimulaon tool reports an average vehicle spacing, then an equivalent density can be computed as the segment length divided by the average vehicle spacing. For more information, please consult Chapter 7 of the HCM 2010 (21). Table 3-2. Comparison of measures of effectiveness from microsimulation and HCM methods.

45 category with an additional 3+ autos category. Larger num- bers of categories are used if the analysis of household travel surveys supports greater segmentation in automobile own- ership. Typically, households owning four or more func- tional automobiles have historically been a small enough segment of the population to not justify greater segmenta- tion. This may change if the number of multi-generational households increases in the future. • Income. Income is used by some models instead of auto- mobile ownership as a predictor of trip making. House- holds are typically divided into income categories. While specific categories vary from model to model, a typical stratification will include income ranges that describe low income, medium income, and high income. It is also pos- sible to encounter a five-tier system that includes very low income and very high income. As with automobile owner- ship, the key determining factor of the number of catego- ries used should depend on the results of a local household travel survey. While there is generally a correlation between income and automobile ownership, the relationship between automobile ownership and trip making is not as clear. • Population. Population data describe the number of individ- uals living in an area. While it is possible to associate popula- tion with a wide variety of attributes, most models tend to restrict themselves to those attributes directly related to the trip generation model’s structure. Most travel demand models tend to identify household characteristics and then disaggre- gate TAZ-level population into households. The population then adopts the characteristics of the households. Typically, the key attribute most models look at with regard to the population is the number of people living in the TAZ. Travel demand models using more sophisticated life- style trip generation models will also attempt to identify additional characteristics relevant to those models. Exam- ples include identifying school-aged population, retirees, and military personnel. At an even more sophisticated level, the emerging activity-based models being developed by some of the larger MPOs in the country require a greater level of detail in order to create data records for each individ- ual in the model along with their relationships to each other. Typically, these relationships are described at the household level and aim at trying to identify which household mem- bers travel together on which trips. Such relationships are usually derived from household travel surveys. Common sources of demographic data include the U.S. Census, U.S. Census Journey-to-Work data, the ACS, and mid-census estimates: • U.S. Census. The U.S. Census collects data on every indi- vidual living within the United States of America every tenth year. The decennial census is by far the greatest source units may either be given as a total number or may be fur- ther stratified. The most common stratification of dwell- ing unit types is to distinguish between single-family and multi-family dwelling units. There tends to be a correlation between dwelling unit types and access to an automobile, a strong predictor of trip-making activities. In general, resi- dents of single-family homes tend to have more access to automobiles than residents of multi-family homes (e.g., apartment buildings). Even when factors such as income and household size are corrected for, the fact that many apart- ment complexes and condominiums place a restriction on the number of parking spaces available to residents will tend to limit the number of automobiles available to each house- hold. Furthermore, given the aggregate nature of data at the TAZ level, some models make use of a population dis- aggregation technique to divide the households in a TAZ into distinct cohorts composed of specific household sizes. Further disaggregation of households into dwelling unit types provides for a much more precise application of a cross-classification trip generation model. • Occupancy and vacancy rates. These are used to deter- mine what percentage of dwelling units are occupied, and therefore make up the number of households in the data set. Occupancy rates describe the percentage of dwelling units that are occupied at the time that the population data were collected. Typically, this reflects the conditions expe- rienced in April once every 10 years, when the census is conducted. Often, models will make use of vacancy rates instead. Vacancy rates describe the percentage of dwelling units that remain vacant at a given time. When vacancy rates are used, it is common to see two separate rates given: permanently vacant units and seasonally vacant units. Per- manent vacancies describe units which remain unoccupied throughout the year and may be due to foreclosures, evic- tions, abandonment of property, or a weak housing and rental market. Permanent vacancies as a percentage of total dwelling units can also describe average vacancies result- ing from normal turnover in the housing market. Seasonal vacancies describe dwelling units that are occupied for only a portion of the year. These may include second homes and vacation homes. Beach houses and mountain cottages rep- resent common types of seasonal vacancies. • Automobile ownership. This describes the number of vehicles owned by a given household. The higher the num- ber of vehicles owned by a given household, the greater the access the residents of the household have to an automobile and the higher the likelihood that trips will originate from that household. It is common to see automobile ownership categorized, with a certain percentage of households in the TAZ falling within each of the categories. Common catego- ries include 0 autos, 1 auto, and 2+ autos; although one may encounter greater disaggregation such as a separate 2 autos

46 The ACS is a continuous survey conducted on a sample of the American population and supplements data collected during the decennial census. Questions asked by the survey touch on subjects such as housing costs and rent, utilities, race, gender, and other topics. The continuous nature of the survey makes it possible to track trends in the popula- tion on an annual level. • Mid-census estimates. These can be used when it is not possible to set a travel demand model base year to a census year. Many models have base years that are not defined con- sistently with census years. The need to validate a model to traffic ground counts often dictates the model’s base year, and not all areas will have traffic counts for a census year. Even when counts are available, it may have been too long since the last census to base a current model on the data. This is very common during the second half of a given decade. Mid-census estimates of population address this issue and are typically developed through a combination of population forecasting techniques and local knowledge of regional population patterns. An analysis of area-wide growth trends serves to establish reasonable mid-census control totals and general distribution patterns of the new growth. Many areas conduct a review of occupancy permits to identify where, specifically, growth is taking place and how many new households may have settled in the region. As with any estimation techniques, mid-census estimates assume that there has not been a recent disruption to his- torical trends. Disruptions such as the recent economic environment result in actual growth that can diverge greatly from the established trends. 3.3.1.2 Employment Employment data are the key input used by trip generation models to develop trip attractions. These data represent the economic activity occurring within a region. Most trip pur- poses rely on employment data for generating trip attractions. For employment data to be useful, they must contain three pieces of information: location of employment, number of employees, and an industrial code for the employer describ- ing the type of work being performed at the job site. Employ- ment data are usually acquired for a specific moment in time. Typical variables include location, number of employees, industrial classification codes, and employment categories: • Location. Location data are typically conveyed through street address data corresponding to the job location. Depending on the data source, addresses given may be for corporate offices and not actual job sites such as individual retail stores. Care must be taken in understanding the nature of the data source and how much cleaning of the data has taken place prior to use. Addresses can be geocoded to provide latitude for demographic data. The census expends every effort in its mission to count every individual. As such, it represents the most inclusive data set on the American population in the United States. The census collects data on every indi- vidual using a form (previously referred to as the short form) that captures the most vital statistics of concern to the census. Namely, this form focuses on the number of individuals residing in each dwelling unit. In the past, a sample of the population was also taken using a supple- mental long form that could be expanded to the popula- tion as a whole in order to get more detailed characteristics of each household. The long form has been discontinued and replaced by the ACS, which is now conducted on a continuous basis throughout the decade. The decennial census remains the primary source of population data for most travel demand models in the United States. During the course of a census data collection and post-processing effort, individual MPOs work with the U.S. Census Bureau to define TAZs. This makes it possible to disaggregate cen- sus data into the TAZs. • U.S. Census Journey-to-Work data. These data have been a tremendously valuable census-related resource for transportation modeling and transportation planning for decades. The Journey-to-Work data are at the TAZ level and are a rich source for home-based work trips from the origin zone (home) to the destination zone (employment zone). Laws and policies protecting the privacy of respondents dictate the level of disaggregation that can be applied to cen- sus data before they are provided to the public. The smallest amount of data is available at the smallest level of census disaggregation, the census block. As it pertains to modeling, the only relevant data that can be obtained at the block level is number of persons and dwelling units. Higher levels of aggregation can reveal income, automobile ownership, and other characteristics. Census block groups can be used to provide most of the demographic data needed for a travel demand model. It is not uncommon, particularly in rural areas, that cen- sus block groups are too large to serve as TAZs; however, TAZs can be designed so as to nest within census block groups. Block group characteristics can then be applied to the individual TAZs. Even when TAZs are delineated in conjunction with the census, the resulting TAZs may need to be larger and/or more oddly shaped than desired in order to ensure that the contained population data repre- sent observed census data and not synthesized population. These large TAZs may then be subdivided using a combi- nation of census block data, knowledge of the area, and professional judgment. • The ACS. The ACS replaces the census long form and serves the purpose of providing more detailed information on the American population than the census short form.

47 Links typically carry information about distance, travel time, and roadway capacities, and provide curvature to the network, while nodes establish intersections or opportuni- ties to change travel direction in the network. For traditional models, link networks describe network connectivity move- ments along a series of interconnected nodes, for example, from Node 101 to Node 102 to Node 103. Some modeling soft- ware packages describe network connectivity as a sequence of links instead of referencing network nodes, for example, from Link 1 to Link 2 to Link 3. All highway networks are used to define the transporta- tion system and can be edited to ensure an accurate reflection of all possible movements along the transportation system being represented in the model. Typical variables carried by highway network links include the following: • Functional classification. Functional classification describes a segment of road with regard to a hierarchy of roadway purpose which the road serves with respect to dif- ferent roadway characteristics. Common functional classes include freeways, principal and minor arterials, collectors, and local streets for both urban and rural areas. The most commonly encountered functional classification system in a travel demand model is the FHWA schema used by the Highway Performance Monitoring System (HPMS). In some cases, a model may employ a local variation of the FHWA system in an attempt to capture the distinctive roadway characteristics of the area. • Facility type. Similar to functional classification, facility type describes the nature of a road with regard to a hierar- chical classification of roadway purpose that the road serves. Unlike functional classification, facility types are typically defined without regard to the prevailing land use surround- ing a given segment of road or whether the land use abut- ting the road is urban or rural in character. Common facility types include collector, arterial, expressway, and freeway. • Area type. Area type describes the prevailing land use sur- rounding a given segment of road in relatively broad catego- ries (but more specific than merely distinguishing between urban and rural segments). Common area types include central business district (CBD), residential, sub urban, and rural. • Number of lanes. Number of lanes describes the number of through traffic lanes occurring on a given segment of road. Since most travel demand models handle each direc- tion of traffic as a distinct entity, the number of lanes is most commonly expressed as the number of lanes per direction. • Speeds. Speeds are expressed in travel demand model input networks as free flow speeds. Free flow speeds are the speeds at which traffic is expected to travel without regard to con- gestion. Some models make use of posted speeds (those speeds indicated on speed limit signs for given segments of and longitude (lat-long) coordinates. Errors in the develop- ment of lat-long data may occur. The most common error is the inversion of coordinate signs. These errors become apparent as soon as an attempt is made to plot the locations. Another common error is the truncation of decimal places in the coordinate. This can result in a more subtle shift in location that is difficult to detect. • The number of employees. The number of employees at a job site is crucial for developing travel demand model employment data. Number of employees should be given as an absolute number. • Industrial classification codes. These codes identify the type of work performed by the employment center. These codes may be provided either by the older Standard Indus- trial Classification (SIC) system, the newer North Ameri- can Industry Classification System (NAICS), or both. These codes are used to group employment by broad categories that are used by the travel demand model. • Employment categories. These are broad categories into which employment data that are put into travel demand models are usually aggregated. The categories are created by grouping together employment types based on indus- trial codes with similar types of activities. Similar activi- ties are assumed to have similar impacts on the number of trips attracted to a given TAZ based on the type of employ- ment found there. Three of the most common categories are retail, service (service employment may also be called office or commercial employment), and industrial. At times, these three categories can be supplemented by addi- tional employment categories. Common additional catego- ries include subdividing industrial into manufacturing and non-manufacturing, dividing service into separate service and office employment, distinguishing between standard and trip-intensive retail (fueling stations, fast food restau- rants, etc.), and including special categories for govern- ment, hospital, and/or military employment depending on the design and requirements of the travel demand model. A number of sources exist for the acquisition of employ- ment data. These data can be obtained both from private ven- dors and the government. Government sources of these data include Local Employment Dynamics (LED) data and the Quarterly Census of Employment and Wages (formerly known as ES-202). 3.3.2 Network Data Transportation networks represent the physical trans- portation system in a travel demand model. These models are typically constructed from a set of points called nodes on a coordinate plane. The linkages between the nodes are called links.

48 packages all now have features built into them that allow the user to import GIS or LRS data and translate them into a format suitable for travel demand modeling. At the very least, geographically accurate distances can be retained in the model network. The major commercially available modeling software also provides GIS environments in which users can edit and maintain highway network data as a GIS feature. Currently, the most common method of creating a high- way network from scratch is to import an already existing GIS dataset using a travel demand modeling software package and make edits to the network using a graphic user interface. It is now common to edit and maintain network data completely in a GIS/LRS environment. State departments of transporta- tion (DOTs) typically have or have access to mapping and data resource centers that maintain a database of their road- way inventory in a GIS format. The quality of these data var- ies from state to state and depends on how advanced each agency’s GIS practice is. U.S. Census TIGER line data can also serve as a starting point, but may contain more errors and lack the specific data desired to develop a highway network. Private vendors can also provide these data. GIS/LRS data used to develop highway networks will need to be carefully reviewed since such data are typically devel- oped originally for mapping purposes without regard to net- work connectivity or path building. Common errors include missed connections, missing pieces, and intersections occur- ring where they should not (such as an arterial intersecting with the mainline of a freeway instead of passing under it). 3.3.3 Traffic Counts Traffic counts are typically included in travel demand models to serve as a basis upon which to validate the model’s highway assignment. Additionally, traffic counts may serve as a benchmark for forecasting reasonableness by providing a point of comparison with future year model results. In some cases, the volume-to-count ratio from a validated highway assignment is used to develop an adjustment factor that is then applied to forecast model results to compensate for base year validation error. Traffic counts are also instrumental to establishing a model’s external trips. Traffic counts that are used in a travel demand model are consistent with the temporal scale of the model. Hourly traffic counts can be summed to any larger period to develop either daily or TOD models. However, in many cases, hourly count data are not available, just daily count data. While many travel demand models still only assign overall vehicle or person trips, more and more models are attempting to explicitly model truck traffic. Vehicle classification counts can assist with the development and validation of truck models. Traffic counts can usually be obtained from a state’s DOT. The quality of traffic count data varies from state to state. roads) as the basis of their free flow speeds. Other models may make use of speeds developed based on speed-delay studies conducted for an area. Typically, some attempt is made to include the impacts of signal delay and side friction into the free flow speed since most travel demand models in use today do not explicitly model these phenomena. Many models use some combination of facility types, area types, and/or functional classes to derive free flow speeds. Some models require the analyst to directly input model speeds for each link in the network. • Traffic counts. Traffic counts are typically included in travel demand models to help validate the model’s high- way assignment. Additionally, traffic counts may serve as a benchmark for forecasting reasonableness by provid- ing a point of comparison with future year model results. See Section 3.3.3 for a more detailed discussion of traffic counts in travel demand modeling. • Annual average daily traffic or average daily traffic. Models typically use an average weekday. Annual average daily traffic (AADT) is developed from traffic counts. Traf- fic forecasts usually use an AADT or a specific day. See Sec- tion 3.3.3 for a more detailed discussion of AADTs in travel demand modeling and project-level forecasting. • Capacities. Capacities express the number of vehicles that can be expected to travel along a given segment of road during a given segment of time. Capacities are primar- ily used during the highway assignment step of the travel demand model to measure congestion and the influence of traffic diversion. Capacities are typically expressed as the number of vehicles per lane per hour, but are typically converted by travel demand models into total hourly or peak-period capacities. Many models use some combina- tion of facility type, area type, and functional classes to derive the capacities for the model network. Some mod- els require the analyst to directly input capacities for each link in the network. In some cases, models may make use of more sophisticated equations to derive capacities using additional variables such as the presence of on-street park- ing and lane width. Network geometry describes the physical shape of the transportation network. Over the past decade, great advances have been made in network geometry. The greatest impact of inaccurate network geometry is the potential for large dis- crepancies between modeled link distances and real-world link distances. These discrepancies can translate into signifi- cant differences in travel times. Large differences in travel time can result in less accurate highway assignments, which lead to greater error in travel demand model results. Most new highway networks created for travel demand models today are created by importing data from established GIS or linear referencing systems (LRS) datasets. The major commercially available travel demand modeling software

49 3.3.5 Origin-Destination Studies OD data are used to develop trip distribution models, external trip tables, and baseline data for corridor studies. OD studies are a common feature of traffic and revenue studies conducted as part of a toll road financing study. Types of OD studies typically conducted include intercept surveys, license plate video capture, GPS tracking, and use of cellular phone data: • Intercept surveys. These surveys involve stopping respon- dents in mid-travel and asking them questions concerning where they are traveling from, where they are traveling to, the reason for their travel, and how they plan to get to their destination. This technique is most frequently used when attempting to get external trip data for a subarea or a travel demand model external boundary. It is also used for corri- dor studies to identify which trips may be more susceptible to trip diversion. Some states have passed laws or enacted policies to prohibit intercept surveys. Intercept surveys can be disruptive to travel and need to be coordinated with the appropriate government agencies and law enforcement. In some special cases, surveyors may use natural stops in traffic flow (such as signalized intersections) to distribute mail-back survey cards to minimize the impact to traffic. • License plate video capture. This is an alternative to the intercept survey. In this method, high-speed cameras are set up in strategic areas around the study area. License plates are recorded as the vehicles travel past the field of vision of the cameras. The captured data are then post processed, and matching license plates from the various cameras are noted. A distribution of trips across the cordon established by the cameras is then recorded. This method does not disrupt the flow of traffic, but limits the data that can be collected to only the observed OD pattern. Data on vehicle occupancy or trip purpose are unobtainable. • GPS tracking. This method is not as common as the other methods for collecting OD data, but more work is being done to leverage the capabilities of GPS tracking. For exam- ple, in California, it has become common to incorporate GPS tracking in OD studies. Another example of this is GPS data available from the American Transportation Research Institute (ATRI) showing freight truck movements. This information can be useful in calibrating and validating truck models, but is limited to long distance truck travel. Furthermore, since the data are continuous GPS data, all movements are captured. Long distance trucks often make periodic stops that are not their final destination. The data need to be reviewed carefully to verify true origins and destinations. • Cellular phone data. Use of cellular phone data is increas- ing in planning studies with long distance OD travel data Most state DOTs maintain at least some permanent auto- matic traffic recorders (ATRs or PTRs) to collect data along the state’s more crucial roadways. These usually include seg- ments of Interstate highways and key state highways. These are then supplemented by periodic temporary counts collected around the state. Temporary counts are usually collected for a period of a few days and then post-processed using seasonal adjustment factors to develop AADT. Some states may also utilize peak season weekday average daily traffic (PSWADT) for modeling and then convert the results into AADT. Some MPOs and municipalities also conduct supplemental count programs not covered by the state program. Additional counts may sometimes be collected as part of a corridor study or in conjunction with a model development project. 3.3.4 Household Travel Surveys Household travel surveys are conducted by contacting indi- viduals at their place of residence and asking them to answer some questions regarding their daily trip-making activity and household characteristics. The primary purpose of collecting these data is to have enough information to estimate the non- assignment components of travel activities, possibly includ- ing trip generation and trip chaining, trip distribution, time of trip departure, and mode choice. Travel diaries are usually sent to respondents as a follow- up to a household travel survey. Respondents are asked to take the diary with them as they travel throughout the day and make notes concerning where they are going, the mode of travel used to arrive at the destination, the departure and arrival times, reason for travel, number of passengers, and any other information that is considered desirable by the surveyor. In some cases, global positioning system (GPS) units may be loaned to respondents for more accurate data collection. Often, respondents fail to report short distance trips, view- ing such trips as being of negligible importance. This can skew travel behavior results related to short trips that could be represented in a travel demand model as either intrazonal trips or trips to neighboring zones. GPS units can be used to capture these trips. Whenever possible, models should make use of recently collected household survey data from the local area. How- ever, due to the expense of conducting a household travel survey, many models are using out-of-date or borrowed trip generation parameters. Those states that participated in the 2008 NHTS add-on may have a wealth of data that can be used for this purpose. Some states participating in the NHTS have focused on collecting data on undersurveyed popula- tions, typically those falling in rural areas outside of MPO boundaries. Other states purchased more comprehensive add-ons that are intended to develop trip generation rates for all MPOs throughout the state.

50 division of AM peak, PM peak, and off-peak or a four-part division of AM peak, mid-day, PM peak, and overnight are the most common TOD schemes encountered. Typical peak periods cover 3 hours each with the remaining periods mak- ing up the difference of the 24-hour day. Daily volumes for such models are usually generated by summing the period volumes. At a minimum, a model that estimates peak con- ditions is necessary to understand the effects of congestion. More recently, models that assign trips on an hourly level have been developed. This is accomplished either by subdividing some of the periods into hours (typically the peak periods) or subdividing the entire 24-hour day into 24 hourly segments. In almost every case, some level of post-processing will be required to make forecast traffic from travel demand mod- els suitable for project-level forecasting. Model volumes can also be used as inputs to microsimulation models. The finer the temporal resolution of the travel demand model’s assign- ment, the easier it is to incorporate the model volumes into a traffic microsimulation. 3.4.2 Speeds Congested speeds are produced by the model for the same temporal resolutions as the volumes developed by the model. Congested speeds are derived from the congested travel times produced by the model’s traffic assignment. Congested travel times are a function of the V/C ratio and produced by the VDF. The larger the temporal resolution of the model, the less the speeds developed by the model reflect actual travel speeds at any given time. Even at the smallest level of temporal reso- lution encountered in travel demand models, the hour, the congested speeds reflect at best an average speed. Furthermore, the VDFs used in most travel demand models are only rough approximations of actual congested travel time relationships. As a result, the congested speeds and travel times are themselves only rough approximations. Model speeds are an essential component to many analyses, such as benefit/cost analyses and air quality emissions modeling. 3.4.3 Turning Movements Most travel demand modeling software packages are capable of producing existing and forecasted turning movement volumes. These turning movements are taken directly from the model’s path building subroutine during highway assignment. Most travel demand models do not incorporate intersection modeling into their process. As a result, the turning move- ments developed by most models are insensitive to inter section capacity and delay characteristics, which have a significant influence on real-world turning movements. The commercially available software packages most com- monly used in the United States have the ability to incorporate requirements. Since cellular phones need to be connected to the cellular phone network to send and receive calls, it is possible to track the movement of the phone as it travels around the cellular phone service provider’s service net- work. These data, stripped of any information that could be traced to an individual, can then be post-processed to pro- vide OD data. While this limits data collection to users of cellular phones, these phones have become very prevalent. 3.3.6 Freight and Heavy Vehicles Data for heavy vehicle movements are typically derived from the FHWA Freight Analysis Framework (FAF). Although particularly suited to long distance Interstate movements, this information can be disaggregated using locally devel- oped data to be more suitable to regional planning efforts. The availability of the FAF series of data sets along with the Quick Response Freight Manual II (69) has coincided with an increase in the development of truck and freight models. 3.4 State of the Practice of Outputs for Travel Forecasting Models Travel demand models often supply the raw data needed for project-level decision-making. However, these data are rarely suitable for use directly and will typically need to be post-processed. This section discusses some of the common outputs of the travel demand models that are used at the project level. 3.4.1 Volumes The simplest and most basic forms of travel demand models represent daily (24-hour) travel corresponding to an average annual weekday. Daily models are sufficient for most planning applications when the purpose is to identify relative changes in demand between alternatives. Many daily models are still used for transportation planning in medium and small urban areas and rural areas, to help identify the needed improvements for roadway facilities. However, roadway design requires an assessment of traf- fic at peak conditions. Travel demand is distributed unevenly during the hours of the day, often with noticeable peaks dur- ing the morning and evening commuting hours. In many cases, congestion does not occur outside of these peak times. Because daily models do not model traffic specifically at these peak times, they are ineffective at explaining the impacts of congestion. While methods exist to post-process 24-hour volumes into peak volumes, many models in areas that expe- rience congestion attempt to make this process more accu- rate by specifically modeling smaller periods of a day using a TOD approach (see Sections 3.2.4.2 and 3.2.4.3). A tripartite

51 • Fuel consumption. Fuel consumption is typically a com- ponent of economic impact assessments and air quality analysis. The simplest methods of calculating fuel con- sumption involve applying fuel consumption rates to VMT. More sophisticated methods attempt to tie fluc- tuations in fuel consumption to changes in congestion. Fuel consumption rates are derived from sources such as the national CAFÉ standards and the Transportation Energy Data Book (68). The latest version of EPA’s emis- sions model, MOVES2010a, also includes a sophisticated fuel consumption model. • Vehicle hours of delay. Vehicle hours of delay measures the amount of time lost due to congestion. A common approach to developing hours of delay is to calculate VHT using both congested and free flow travel times. The differ- ence between the two is considered to be the delay. • Cost of delay. Cost of delay assigns a monetary cost to delay. This metric is often used in planning and benefit/cost analy- sis to provide a comparison of the relative values of trans- portation investments. Cost of delay can be calculated by multiplying the vehicle hours of delay by a value of time. The value of time is typically calculated for a local area as a function of the area’s average wage rate, which may be specified for each of a number travel markets. 3.4.5 Origin-Destination Information OD data are developed by all travel demand models as a normal part of the trip distribution and assignment pro- cesses. Common forms of OD data model output include the following: • Forecast OD tables. Forecast population and employ- ment data, other socioeconomic data properly representa- tive of future conditions, and future transportation supply conditions (for models using feedback) form the basis for developing forecast OD tables. Area-to-area flows show- ing the change in intra- and inter-regional trips, as well as trip length frequency distributions, are ways of sum- marizing and comparing OD tables between a base and forecast year. • Select link. Select link analyses identify the origins and destinations and the assignment path of trips traveling along particular segments of roads. Select link analyses are used in corridor studies and bridge projects to identify impacted populations and roads. These analyses are also used to determine the influence area (travel shed) of a par- ticular link or corridor to define study areas and to build OD matrices for subarea models. • Select zone. The select zone analysis identifies the vol- umes and the paths associated with travel to and from some level of intersection modeling as part of the travel demand model, including defining traffic control devices, cycle lengths, and the number of turn lanes. However, the incorporation of intersection modeling in regional travel demand models is relatively rare due to the data requirements and the additional complications that modeling intersection delay can produce for network assignment validation. Standard methods for develop- ing turning movements include taking the approach volumes at an intersection from a model and calculating the turning move- ments using off-model tools. 3.4.4 Measures of Effectiveness Measures of effectiveness (MOEs) provide analysts with statistics that can be used to evaluate the performance of a set of supply and demand characteristics associated with a traffic forecast. MOEs are often applied on a relative scale to compare various alternatives. These measures can also be compared to other non-model data to form a portion of a larger decision-making tool or process. More information on MOEs can be found in Section 2.3. Typical MOEs include the following: • Vehicle miles traveled. VMT is the product of the num- ber of vehicles traveling along a given segment of road and the distance those vehicles travel on the segment. These results are then summed to the regional level. VMT is a measure of travel demand and is also used in a variety of post-processing applications to measure the intensity of travel in corridors during alternatives testing, estimating regional fuel consumption, and estimating regional pollut- ant emissions. It is also sometimes used as a weighting fac- tor for calculating area-wide statistical averages over many highway segments. • Vehicle hours of travel. VHT is the product of the num- ber of vehicles traveling along a given segment of road in the transportation system and the travel time experienced by those vehicles on the segment. These results are then summed to the regional level. VHT is an additional mea- sure of vehicular activity, but it also can be used to estimate system-wide or subsystem delay. Like VMT, VHT is easily extracted from travel demand model highway assignment results. • Congested speeds. Congested speeds are derived from con- gested travel times produced by the model. Link-specific congested speeds are derived by dividing the link length by the congested travel times. System-wide average congested speeds are derived by dividing VMT by VHT, averaged over all roadway segments or links. While some attempt is made in most models to take into account intersection delay and side friction, the methods employed tend to be very aggregate and not well suited to operational analysis.

52 useful information for an agency’s maintenance, rehabili- tation, and reconstruction programs. 3.5 Defaults versus Locally Specific Parameters Ideally, travel forecasting models should be statistically calibrated to conditions for the area being analyzed. However, project-level travel forecasts are often done under severe time and budget constraints that preclude a rigorous calibration. In other cases, some data that would form the basis for cer- tain parameters do not exist. This report suggests the use of transferable parameters where appropriate and necessary, and when the analyst understands the implications of borrowing parameters from another location or from national defaults. Parameters are an important input to the travel forecasting process and are usually obtained by applying statistical analy- ses to travel behavioral or traffic data that are obtained locally. The quality of an individual parameter can be assessed by cer- tain statistics, such as a t-score, that are provided during the estimation process. These statistics give the analyst confidence that the model will provide good forecasts. When statistical analysis cannot be performed, then the analyst is forced to use borrowed or asserted parameters of unknown quality for the application. The notion of transferable parameters has a long history, dating to the release of NCHRP Report 187 (8). The authors of NCHRP Report 187 were working mainly on professional experience because there was little scientific evidence prior to this time that the concept of transferable parameters would work adequately. The ultimate success of NCHRP Report 187 (8) and its subsequent revision, NCHRP Report 365 (7), dem- onstrated the strength of the idea through many applications throughout the United States. The draft report for NCHRP Report 716 (6), which replaces NCHRP Report 365 (7), con- tained a review of academic literature on transferability. The authors of this review concluded that transferability worked well in some situations and did not work well in others. The situations where transferability did not work well tended to involve attempts to transfer parameters well into the future or instances in which there were inconsistencies in how socio- economic and network data were prepared. Often the perfor- mance of transferable parameters could be improved through scaling or other forms of adjustment. The concept of transferable parameters was extended to urban truck forecasting by the Quick Response Freight Manual (6) from FHWA in 1996 and its revision, the Quick Response Freight Manual II (10), in 2007. NCHRP Report 599: Default Values for Highway Capacity and Level of Service Analyses (70) provides an extensive look at the application, sensitivity, and transferability of default values used in highway capacity analyses. a selected group of zones. This information can then be used to calculate the distribution of trips due to specific developments. 3.4.6 Model Outputs for Other Analyses Travel demand models are outputs often used in eco- nomic and environmental impact analyses as part of a com- prehensive project forecasting analysis. As travel demand modeling software has become more sophisticated, flexible, and customizable, it has become more common to see post- processors built directly into the models. Post-processors are commonly developed for the following purposes: • Air pollution emissions. Air pollution emissions post- processors take model data such as VMT, VHT, and speeds and determine how much pollution is generated in an area due to transportation activity. EPA’s emission model, MOVES2010a, and its predecessor, MOBILE6.2, both use travel demand model data to determine the amount of air pollution due to mobile source emissions. Additional post-processors are typically written to integrate the data from the travel demand model with the data from the emissions model. Simpler emission post-processors have also been developed for assessing the carbon footprint of the transportation system by calculating the amount of greenhouse gases emitted by vehicles along the trans- portation system. • Economic impact analysis. There is growing interest by planning agencies in the economic development impact of transportation plans, alternatives, and policies. Post- processors address the question of the amount of business or jobs growth that might be achieved. Economic develop- ment is driven by improvements in accessibility resulting from travel time savings for commuters, customers, and commercial activity. • Benefit/cost analysis. Motivated by a desire to prioritize transportation investments, benefit/cost post-processors assign monetary values to the transportation activities forecast by the travel demand model. At the most basic level, the value of travel time and vehicle operating costs are compared to the capital and operating costs of a pro- posed improvement or new facility to determine whether the cost is justified on transportation efficiency grounds. • Bridge and pavement deterioration analysis. Bridge and pavement condition models estimate the effect of vehicle activity (and many other factors) on asset condition. Since bridge and pavement damage is a function of vehicle weight, the condition models require detailed information about vehicle types, which are commonly expressed as equivalent single axle loads (ESALs). A travel demand model’s fore- casts of future automobile and truck volumes can provide

53 • OD matrix growth factoring, • Time-series models, • Traffic impact study tools, and • Elasticity methods (see Section 10.3 of this report). A brief overview of these tools is provided below so that practitioners can both make use of them and improve them. 3.6.1 Manual Gravity Tool Procedure— Kentucky Transportation Cabinet This sketch-planning tool is developed for traffic forecast- ing on new facilities in an area not covered by a travel demand model. The tool is built around the methodologies provided in NCHRP Report 387 and those published by Caltrans (73). The procedure diverts trips to the new facility based on its attractiveness, which is a function of distance and travel time advantage over alternative routes. The required input data include infrastructure characteristics, traffic and control, no- build average daily traffic (ADT) for base and future years, and ADT turning movements at major intersections. There are five major steps involved in the procedure: • Estimate free flow speed. • Estimate roadway capacity using HCM approaches for various facility types. • Generate no-build traffic forecasts for base and future years (using growth factor) and turning movements at major intersections in the study area (using a program such as turns.bat). • Compute congested speeds using a modified version of the BPR curve. • Perform diversion analysis based on an equation developed by Caltrans that computes the percentage of diverted traf- fic as a function of travel time (based on congested speeds) and distance savings of the new facility for traffic volume between each OD pair. The procedure provides a quick forecast of the traffic vol- ume on a new facility. However, users are advised to check the reasonableness of the results of each of the above steps. The accuracy of the final forecast depends on the accuracy of the input data. 3.6.2 Sketch-Planning Tools— NCHRP Report 255 NCHRP Report 255 contains a number of sketch-planning tools, such as trend line analysis and turning movement fore- cast tools that can be used when modeling tools are not avail- able or are deemed unnecessary. Trend line analyses are based on the principle that a traffic volume trend can be established by analyzing land use patterns There are two classes of transferable parameters, bor- rowed and asserted, and their uses are somewhat different. Borrowed parameters are those prepared somewhere else and are thought to remain relatively constant when applied within a new locale. For example, NCHRP Report 365 (7) contains many parameters that were obtained from the 1990 National Personal Travel Survey (71), the forerunner of the NHTS. These parameters were stratified by city size to reflect assumed differences in urban geography. The analyst could reasonably expect that these parameters would be good enough in places of similar characteristics (e.g., metropolitan area size) as those for which the parameters were developed because the parameters were derived from a scientific sam- ple of all U.S. households. It is also possible to use borrowed parameters selectively when individual parameters from a local calibration fall outside well-established norms. Asserted parameters are also defaults, but they are imposed upon the analyst by agency policy to maintain consistency across forecasts. Asserted parameters are similar to borrowed parameters in that they are believed to be readily transfer- able, but they are most often created by identifying similari- ties in models in different locations. Asserted parameters are particularly useful for developing parity in competing appli- cations for funding at the state and national levels. Asserted parameters have also been used to ensure that all applications for new site developments are treated equally. Where possible, sensitivity analyses of default parameters should be conducted. This will help the analyst to determine whether the use of borrowed or asserted parameters is accept- able or whether locally determined parameters should be uti- lized instead. Specifically, if varying an input parameter for a particular model or method results in a wide range of results, then it may be concluded that the parameter in question is highly sensitive, and resources should be committed to data collection and accurate development of the parameter based on local conditions. 3.6 Other Traffic Forecasting Tools and Methodologies There are many tools and methodologies other than travel demand models used by traffic forecasters. These tools are very useful to agencies in providing traffic forecasts for a vari- ety of needs. A short list of other traffic forecasting tools and sketch-planning tools is provided: • The Kentucky Transportation Cabinet’s manual gravity tool procedure (72), • NCHRP Report 255 spreadsheet (trend line analysis, extra- polation, screenline adjustment, and model assignment adjustments), • Turning movement tools (several versions are publicly available),

54 ing movements are known. The non-directional volume procedure requires more judgment on the part of the ana- lyst as typically the turns are derived from knowledge of the non-directional approach link volumes and an estimate of the total turn percentage at the intersection. Both of these procedures (which typically use iterative proportional fit- ting or Fratar methods) require several iterations before an acceptable closure is reached. As these methods have been implemented in software, they are far less time consuming today than when they were applied originally by hand. These procedures are not intended to be used for design purposes. • “T” intersection procedures. These are a special applica- tion of the iterative procedures. The non-directional volume method will produce a unique solution if all three approach volumes are known. The directional volume method will produce a unique solution if directional approach volumes and one turning movement are known. The method allows analysts to use approach link volumes as a substitute for base year turning movements in cases where these are not available. 3.6.3 OD Matrix Growth Factoring This method involves a simplified application of a travel demand model. In this method, a model sketch network of a study area is developed and a base year OD matrix is cre- ated (typically using matrix estimation) from traffic counts. The base year OD matrix can be either a 24-hour matrix or a peak-period matrix. Growth factors are applied to the base year matrix to create a future year matrix or matrices. Growth factors can be based on various sources, such as trend analy- ses, population and/or employment projections, anticipated land use changes, and so forth. A single growth factor can be applied or separate factors can be applied to individual cell matrices (i.e., individual OD pairs). The future year matrices then are assigned to the model network. This factoring approach can be easily misapplied. The analyst is cautioned that changes in socioeconomic variables do not transfer directly to equivalent changes in automobile trips and therefore other variables such as productions and attractions should be considered. Inherent in the method is the assumption that travel patterns external to the study area will remain constant, so determination of the extent of the study area network is important. 3.6.4 Time-Series Models Time-series models extrapolate upon past trends to pre- dict future conditions. Time-series models can be used for the traffic forecast itself or for forecasting the inputs to travel forecasting models. There are only a few articles in the pub- lished literature on the use of time-series methods for travel forecasting. A review of existing literature and practice is con- tained in the Guidebook on Statewide Travel Forecasting (44), and/or historical traffic counts. Trend lines can be either linear, representing a constant growth rate over time, or non-linear. Non-linear growth curves fall into three basic categories: • Increasing growth (exponential), • Decreasing growth (logarithmic), and • Stepped growth. Trend line analyses use historical traffic volume data to develop a regression model. The model then is used either to interpolate growth between 2 years or to extrapolate growth from a single time frame. An important step is to determine the shape of the growth curve upon which the trend line analysis will be based. For the least sophisticated and least demanding analyses, linear growth is appropriate. For moderately sophisticated analyses, land use trends and both local and through traffic should be considered; either linear or non-linear trends may be applicable. For high-level analyses in which land use changes form the basis for select- ing the curve, stepped growth may be the most appropriate. Trend line analyses typically require a large number of data points, and outliers should be identified and understood before being discarded. Because trend line analysis assumes a continuation of past growth patterns, the analyst should always check the reasonableness of the output, especially when there is a change in land use, future roadway capacity, or geometry. The output should also be compared against other forecasts for the adjacent facilities. NCHRP Report 255 provides guidance on the use of trend line analyses for both interpolation and extrapolation. NCHRP Report 255 provides methods for estimating turn- ing movements for planning and designing highway intersec- tions and interchanges. NCHRP Report 255 recognized that traffic assignments rarely provide turning movement fore- casts that can be used directly and that therefore these fore- casts need significant refinement. The refinement procedures described in NCHRP Report 255 can be used to develop more reliable turning movement estimates from various sources and for various uses. NCHRP Report 255 discusses three sets of procedures: fac- toring procedures, iterative procedures, and “T” intersection procedures: • Factoring procedures. These procedures use either the ratio method or the difference method. The primary fea- ture of factoring procedures is that they are easy to apply but assume a constant growth rate consistent with past trends. Factoring procedures also require base year turn- ing movement counts, which may not be available. • Iterative procedures. These include separate directional vol- ume and non-directional volume methods. The directional volume method adjusts future year turning movements to match a predetermined estimate of turning movement per- centages and can be applied whether or not base year turn-

55 The basic relationship is described by mathematical relation- ships between the dependent variable (either daily or peak- hour trip ends) and independent variables (such as gross leasable square footage and number of employees) or land use–specific variables (such as restaurant seats, hotel rooms, hospital beds, and so forth). These mathematical relation- ships have been developed through numerous studies from which data have been collected and submitted to the ITE. Mathematical relationships typically assume the form of an average trip rate or a regression equation. No guidance is given on future growth of trip generation estimates. 3.6.5.2 Kentucky Transportation Cabinet Permits Manual—Traffic Impact Study Requirements (74) The Kentucky Transportation Cabinet’s procedures for traffic impact studies follow the ITE’s recommendations for site trip generation, distribution, and assignment. The Cabi- net has developed spreadsheets for performing trip genera- tion and for estimating future design year non-site traffic. The Cabinet specifies an analysis or design year 10 years after the year of opening. In the event that the full build-out of the devel- opment is anticipated to extend beyond the 10-year horizon, the design year is the year of full build-out. Future year forecasts are estimated using the weighted aver- age annual rate provided by exponential and linear growth models. Design-year, no-build traffic volumes are determined by applying the projected growth rate to existing traffic volumes. Final design-year total traffic volumes are determined by summing future (no-build) traffic, pass-by trips, and trips generated (entering and exiting). 3.6.5.3 Kentucky Transportation Cabinet Design Memorandum No. 03-11: Traffic Engineering Analysis (75) The Kentucky Cabinet has published a policy manual for traffic engineering analyses in highway design. The procedures and methodologies are consistent with the most recent edition of the HCM and can be used to (1) determine the basic num- ber of lanes on a facility and (2) determine the need for auxil- iary lanes on a facility. The policy manual specifies the need for design-year as well as current or opening-year traffic volumes. 3.6.5.4 Florida Department of Transportation Traffic Impact Handbook (76) FDOT’s Traffic Impact Handbook was prepared to assist state and local agencies in reviewing and providing comments on local government comprehensive plan amendments and development orders as they relate to transportation impacts on state and regional multimodal facilities. The two main published in 1999. At that time, practice was confined almost exclusively to the use of linear trend models and growth fac- tor models. The survey done for NCHRP Project 08-83 also showed that many agencies are still using linear trend models and growth factor models. The Wisconsin Department of Transportation, departing from the norm, reported on the use of Box-Cox regression analysis, which is still being actively used by that agency. The survey done for NCHRP Project 08-83 failed to find any other popular techniques. Time-series analysis is a very active area of statistics, and methods exist that greatly expand upon elementary tech- niques such as linear trend models or growth factor mod- els. Major statistical software packages contain full suites of time-series methods. There is much untapped potential for project-level traffic forecasting. The trend analysis method adopted by the Florida Depart- ment of Transportation (FDOT) performs traffic forecast for areas without a travel demand model. It relies primarily on historical traffic counts to establish growth trends through regression analysis; data on gas sales, land use, and popula- tion are used as a supplement when traffic counts are insuffi- cient. K30 and D30 factors can be used to forecast directional design hourly traffic, and T24 (a daily truck factor) can be used for facilities without existing truck traffic counts. The FDOT methods cite a series of resource documents, including NCHRP Report 255 (1) and NCHRP Report 187 (8), the Institute of Transportation Engineers’ (ITE’s) Trip Gen- eration Handbook (11), and so forth. 3.6.5 Traffic Impact Study Tools Traffic impact studies assess the local congestion conse- quences of a prospective land use site development. They are unique to traffic forecasting in that they combine forecasts of background traffic with a forecast of locally generated traffic. Background or non-site traffic forecasts typically are developed using travel demand models or manual tools like trend line analyses. Forecasts for site-generated traffic are prepared using accepted land use/trip generation relationships based upon quantifiable parameters such as gross leasable area, number of employees, or number of parking spaces. Site-related trips are distributed and assigned to the area roadway network (either manually or using a travel demand model), then summed with background traffic for the total traffic forecast. Guidance for forecasting background traffic varies from little or none to very specific (FDOT, for example). Section 9.9 of this report supplements the guidance materials listed below. 3.6.5.1 Trip Generation Handbook, 2nd Edition (11) The ITE’s Trip Generation Handbook provides guidance on estimating site-generated traffic stratified by land use types.

56 categories of reviews are local government plan reviews and development of regional impact (DRI) reviews. The Traffic Impact Handbook advises that future condi- tions for impact assessments can be estimated using “manual methods,” travel demand forecasting models, or a combina- tion of the two. Manual methods are those methods of trip generation done without large-scale travel demand models. The most common examples of manual methods are the use of trip generation factors to estimate trip generation and the application of growth factors or the addition of known trips from other developments to the surrounding road system to estimate background traffic growth. The Traffic Impact Handbook contains an entire section on projecting background traffic. Background (non-site) traffic is typically estimated using one of three methods based on local area needs and conditions: 1. Growth rate/trend methods relying on historic trends. These methods are typically appropriate in applications for – Small projects that will be built within 1–2 years. – Areas with at least 5 years of data showing stable growth that is expected to remain stable. 2. Build-up methods that use specific development infor- mation. These methods are typically appropriate in appli- cations for – Areas experiencing moderate growth. – Areas where multiple projects will be developed during the same period. – Project horizon years of 5 years or less. – Locations where there is thorough documentation of development approvals. 3. Model methods that involve the use of a large-scale travel demand model. Model methods are typically appropriate in applications for – High growth areas. – Large regional projects that may have multiple build- out phases. – Locations where there is sufficient information available to calibrate the model to current and future conditions. 3.6.6 Elasticity Methods In the absence of a highly sophisticated travel demand model, project planners need to rely on simple methods to determine the impacts of multimodal operational policies. A well-established method of doing so is in the application of elasticities. Elasticity is a concept from economics that describes the percentage change in travel demand resulting from a percentage change in an independent variable, such as cost or travel time. As an example, the transit industry elastic- ity standard known as the Simpson-Curtis rule held that a 1% increase in fare would reduce ridership by 0.33% (or that a 1% decrease in fare would increase ridership by 0.33%). The equivalent elasticity is -0.33. Often, only a single (constant) elasticity is available over the entire range of independent variables. Elasticities are handy shortcuts for estimating changes in travel demand where there is a reasonably close similarity between the project at hand and the case study that created the elasticity. Analysts applying elasticities should take care in application not to exceed the range of experience that the elasticity is derived from. As an example, using an observed elasticity derived from a change in travel cost of 20% to fore- cast the change in demand from a 100% increase in travel costs is inappropriate. Chapter 15 of TCRP Report 95: Trav- eler Response to Transportation System Changes Handbook (38) provides elasticities for a very large number of transportation system changes.