USF Libraries
USF Digital Collections

Modeling the impacts of an employer based travel demand management program on commute travel behavior

MISSING IMAGE

Material Information

Title:
Modeling the impacts of an employer based travel demand management program on commute travel behavior
Physical Description:
Book
Language:
English
Creator:
Zhou, Liren
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Travel behavior
Transportation demand management
Compressed work weeks
Telecommuting
Mode choices
Dissertations, Academic -- Civil and Environmental Engineering -- Doctoral -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Travel demand Management (TDM) focuses on improving the efficiency of the transportation system through changing traveler's travel behavior rather than expanding the infrastructure. An employer based integrated TDM program generally includes strategies designed to change the commuter's travel behavior in terms of mode choice, time choice and travel frequency. Research on TDM has focused on the evaluation of the effectiveness of TDM program to report progress and find effective strategies. Another research area, identified as high-priority research need by TRB TDM innovation and research symposium 1994 Transportation Research Circular, 1994, is to develop tools to predict the impact of TDM strategies in the future. These tools are necessary for integrating TDM into the transportation planning process and developing realistic expectations.Most previous research on TDM impact evaluation was worksite-based, retrospective, and focused on only one or more aspects of TDM strategies. That research is generally based on survey data with small sample size due to lack of detailed information on TDM programs and promotions and commuter travel behavior patterns, which cast doubts on its findings because of potential small sample bias and self-selection bias. Additionally, the worksite-based approach has several limitations that affect the accuracy and application of analysis results. Based on the Washington State Commute Trip Reduction (CTR) dataset, this dissertation focuses on analyzing the participation rates of compressed work week schedules and telecommuting for the CTR affected employees, modeling the determinants of commuter's compressed work week schedules and telecommuting choices, and analyzing the quantitative impacts of an integrated TDM program on individual commuter's mode choice.The major findings of this dissertation may have important policy implications and help TDM practitioners better understand the effectiveness of the TDM strategies in terms of person trip and vehicle trip reduction. The models developed in this dissertation may be used to evaluate the impacts of an existing TDM program. More importantly, they may be incorporated into the regional transportation model to reflect the TDM impacts in the transportation planning process.
Thesis:
Dissertation (Ph.D.)--University of South Florida, 2008.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by Liren Zhou.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 152 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 002007349
oclc - 402526632
usfldc doi - E14-SFE0002309
usfldc handle - e14.2309
System ID:
SFS0026627:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 002007349
003 fts
005 20090618124449.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 090618s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002309
035
(OCoLC)402526632
040
FHM
c FHM
049
FHMM
090
TA145 (Online)
1 100
Zhou, Liren.
0 245
Modeling the impacts of an employer based travel demand management program on commute travel behavior
h [electronic resource] /
by Liren Zhou.
260
[Tampa, Fla] :
b University of South Florida,
2008.
500
Title from PDF of title page.
Document formatted into pages; contains 152 pages.
Includes vita.
502
Dissertation (Ph.D.)--University of South Florida, 2008.
504
Includes bibliographical references.
516
Text (Electronic dissertation) in PDF format.
520
ABSTRACT: Travel demand Management (TDM) focuses on improving the efficiency of the transportation system through changing traveler's travel behavior rather than expanding the infrastructure. An employer based integrated TDM program generally includes strategies designed to change the commuter's travel behavior in terms of mode choice, time choice and travel frequency. Research on TDM has focused on the evaluation of the effectiveness of TDM program to report progress and find effective strategies. Another research area, identified as high-priority research need by TRB TDM innovation and research symposium 1994 [Transportation Research Circular, 1994], is to develop tools to predict the impact of TDM strategies in the future. These tools are necessary for integrating TDM into the transportation planning process and developing realistic expectations.Most previous research on TDM impact evaluation was worksite-based, retrospective, and focused on only one or more aspects of TDM strategies. That research is generally based on survey data with small sample size due to lack of detailed information on TDM programs and promotions and commuter travel behavior patterns, which cast doubts on its findings because of potential small sample bias and self-selection bias. Additionally, the worksite-based approach has several limitations that affect the accuracy and application of analysis results. Based on the Washington State Commute Trip Reduction (CTR) dataset, this dissertation focuses on analyzing the participation rates of compressed work week schedules and telecommuting for the CTR affected employees, modeling the determinants of commuter's compressed work week schedules and telecommuting choices, and analyzing the quantitative impacts of an integrated TDM program on individual commuter's mode choice.The major findings of this dissertation may have important policy implications and help TDM practitioners better understand the effectiveness of the TDM strategies in terms of person trip and vehicle trip reduction. The models developed in this dissertation may be used to evaluate the impacts of an existing TDM program. More importantly, they may be incorporated into the regional transportation model to reflect the TDM impacts in the transportation planning process.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
590
Co-advisor: John J. Lu, Ph.D.
Co-advisor: Steven E. Polzin, Ph.D.
653
Travel behavior
Transportation demand management
Compressed work weeks
Telecommuting
Mode choices
690
Dissertations, Academic
z USF
x Civil and Environmental Engineering
Doctoral.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.2309



PAGE 1

Modeling The Impacts Of An Employer Base d Travel Demand Management Program On Commute Travel Behavior by Liren Zhou A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Civil and Environmental Engineering College of Engineering University of South Florida Co-Major Professor: John J. Lu, Ph.D. Co-Major Professor: Steven E. Polzin, Ph.D. Xuehao Chu, Ph.D. Joseph S. DeSalvo, Ph.D. Philip L. Winters, B.S. Date of Approval: March 26, 2008 Keywords: travel behavior, transporta tion demand management, compressed work weeks, telecommuting, mode choices, nested logit model, generalized ordered logit model Copyright 2008, Liren Zhou

PAGE 2

ACKNOWLEDGMENTS I am deeply indebted to those who assisted me in the various stages of this work. In particular, I wish to express sincere a ppreciation to my doctoral committee members, John Jian Lu, Steve E. Polzin, Xuehao Chu, Joseph S. DeSalvo, and Philip L. Winters, for their advice and support. I would also like to thank Dr. Edward Mierzejewski for his invaluable support and encouragement. I gratefully acknowledge financial support from CUTR Transportation Demand Management program for my Ph.D. studies at the University of South Florida. I am deeply grateful to Mr. Philip Winters, who provided me with many opportunities to work on different interesting and challenging project s. I also want to thank Mr. Ed Hillsman and Brian Lagerberg in the Washington State Department of Transportation for providing me the data used in this study.

PAGE 3

i TABLE OF CONTENTS LIST OF TABLES...............................................................................................................v LIST OF FIGURES...........................................................................................................vii ABSTRACT.....................................................................................................................vi ii CHAPTER 1 INTRODUCTION.........................................................................................1 1.1 TDM Development............................................................................................1 1.2 Employer Based Travel Demand Management Program..................................3 1.3 Effectiveness Evalua tion and Forecasting of an Employer-based TDM...........4 1.4 Focuses of This Dissertation..............................................................................6 1.4.1 Compressed Work Week....................................................................7 1.4.1.1 Determinants of Employees Work Week Choice...............7 1.4.1.2 Compressed Work Week Participation Trend.....................8 1.4.1.3 Modeling the Compressed Work Week Choice...................9 1.4.2 Telecommuting.................................................................................10 1.4.2.1 Previous Empirical Studies on Telecommuting.................11 1.4.2.2 Contribution of This Study................................................12 1.4.2.3 Determinants of Telecommuting Choice...........................14 1.4.2.4 Telecommuting Participation Trend Analysis and Telecommuting Choice Modeling.....................................15 1.4.3 TDM impacts on Commuting Mode Choice....................................15 1.4.3.1 Limitations of Employer-based Method............................16

PAGE 4

ii 1.4.3.2 Modeling the Impacts of an Integrated TDM Program on Employees Journey to Work Mode Choice .....................17 CHAPTER 2 TDM LITERATURE REVIEW..................................................................20 2.1 What is TDM...................................................................................................2 0 2.1.1 TDM Definition................................................................................20 2.1.2 TDM Components Studied...............................................................22 2.2 Why TDM is Important...................................................................................23 2.3 TDM Strategies and Their Effects...................................................................25 2.3.1 Strategies to Change Travel Behavior by Changing Travel Cost.....25 2.3.1.1 Impacts of Parking Cost on Travel Behavior.....................26 2.3.1.2 Impacts of Out-of-the-Pocket Cost on Travel Behavior....28 2.3.2 Strategies to Change Travel Behavior by Changing Travel Time....29 2.3.3 Employer TDM Support Strategies..................................................31 2.3.4 Overall Effectiveness of TDM Strategies.........................................33 2.4 Modeling Framework of TDM Strategies........................................................34 2.4.1 Wa shington State TDM Effectiv eness Estimation Methodology (TEEM) Model.................................................................................. 35 2.4.2 Envi ronmental Protection Agency (EPA) COUMMUTER Model..36 2.4.3 CUTR Worksite Trip Reduction Model...........................................38 2.5 Summary........................................................................................................ ..38 CHAPTER 3 AN EMPIRICAL ANALYSIS OF COMPRESSED WORK WEEKS CHOICE .................................................................................................. ....39 3.1 Introduction................................................................................................... ...39 3.2 Data........................................................................................................... .......43

PAGE 5

iii 3.3 CWW Participation Tre nd for Employees Affected by the WA CTR Laws...45 3.4 Determinants of Employees Work Schedule Choice.....................................49 3.5 Multinomial Logit Modeling of Work Schedule Choices...............................51 3.5.1 Methodology.....................................................................................5 1 3.5.2 Model Specification..........................................................................52 3.5.3 Regression Results............................................................................55 3.6 Ordered Log it Modeling of Work Schedule Choices......................................62 3.7 Conclusion..................................................................................................... ..65 CHAPTER 4 MODELING OF TELECOMMUTING CHOICES....................................68 4.1 Introduction................................................................................................... ...68 4.1.1 Previous Researches..........................................................................68 4.1.2 Contribution of This Study...............................................................71 4.2 Telecommuting Choices Trend Analysis.........................................................74 4.3 Determinants of Telecommuting Choices.......................................................77 4.4 Modeling the Telecommuting Choices............................................................79 4.4.1 Methodology.....................................................................................7 9 4.4.2 Model Specification..........................................................................82 4.4.3 Regression Result..............................................................................88 4.5 Conclusion..................................................................................................... ..94 CHAPTER 5 AN INTEGRATED MODEL OF TDM IMPACTS ON JOURNEY TO WORK MODE CHOICE.............................................................................97 5.1 Introduction................................................................................................... ...97 5.1.1 TDM Strategies Evalua tion Literature Review.................................97

PAGE 6

iv 5.1.2 Li mitations of the Worksite -based Analysis Method.....................100 5.1.3 Modeli ng the Impacts of an Integr ated TDM Program on Mode Choice........................................................................................ .....101 5.2 Methodology..................................................................................................10 4 5.2.1 Nested Logit Model........................................................................104 5.2.2 Elasticities of Logit Model..............................................................108 5.3 Data and Variable Definition.........................................................................111 5.3.1 Mode Shared Trend for the CTR Affected Employees..................111 5.3.2 Variable Definition.........................................................................113 5.4 Model Specification ......................................................................................118 5.5 Regression Results ........................................................................................121 5.6 Elasticities Analysis.......................................................................................131 5.7 Conclusion .................................................................................................... 134 CHAPTER 6 CONCLUSION..........................................................................................137 6.1 Contributions.................................................................................................. 137 6.2 Major Findings...............................................................................................14 0 6.3 Future Research.............................................................................................142 REFERENCES................................................................................................................144 ABOUT THE AUTHOR..End Page

PAGE 7

v LIST OF TABLES Table 3.1 Percent of Employees by Work Schedule from 1993 to 2005......................46 Table 3.2 Participation Rate for CWW by Job Title from 1993 to 2005......................47 Table 3.3 Percent of Employees on CWW by Employer Primary Business Type from 1993 to 2005.........................................................................................48 Table 3.4 Participation Rate for Each Work Schedule by Job Title in 2005................48 Table 3.5 Participation Rate for Each Work Schedule by Primary Business Type in 2005..........................................................................................................49 Table 3.6 Selected Variables Definition.......................................................................54 Table 3.7 Empirical Results for CWW Choices...........................................................59 Table 3.8 Comparisons of Model Predictions and Survey Results...............................62 Table 3.9 Ordered l Logit Model for CWW Choice.....................................................64 Table 4.1 Telecommuting Rate by Teleco mmuting Days per Two Weeks from 1993 to 2005.................................................................................................. .......74 Table 4.2 Telecommuting Rate by Job Title from 1993 to 2005..................................75 Table 4.3 Telecommuting Rate by Employer Primary 2005........................................76 Table 4.4 Telecommuting Rate by Job Titl e and Telecommuting Days per Two Weeks in 2005..........................................................................................................76 Table 4.5 Telecommuting Rate by Primar y Business and Telecommuting Days per Two Weeks in 2005......................................................................................77 Table 4.6 Selected Variable Definitions.......................................................................84 Table 4.7 Ordered Logit Model (POM) for Telecommuting Choices..........................86 Table 4.8 Brant Test of Parallel Odds Assumptions.....................................................87

PAGE 8

vi Table 4.9 Generalized Ordered Mo del for Telecommuting Choices............................92 Table 4.10 Comparisons of the Model Predictions and Survey Results.........................93 Table 5.1 Mode Shares Trend for CTR Affected Employees in King County...........113 Table 5.2 Mode Share Trend for th e Entire CTR Affected Employees......................113 Table 5.3 Variable Definition.....................................................................................116 Table 5.4 Summary Description of Data (Continuous variables)...............................117 Table 5.5 Nested Logit Regression Results................................................................126 Table 5.6 Elasticities for Sele cted Continuous Variables...........................................133 Table 5.7 Effects for Selected Dummy Variables......................................................133

PAGE 9

vii LIST OF FIGURES Figure 3.1 CWW Program Year Effect on CWW Choices............................................58 Figure 5.1 Nested Logit Model Structure.....................................................................120 Figure 5.2 Mathematical Specificati on of the Nested Logit Model.............................120

PAGE 10

viii MODELING THE IMPACTS OF AN EMPLOYER BASED TRAVEL DEMAND MANAGEMENT PROGRAM ON COMMUTE TRAVEL BEHAVIOR Liren Zhou ABSTRACT Travel demand Management (TDM) focuse s on improving the efficiency of the transportation system through changing travelers travel behavior rather than expanding the infrastructure. An employer based in tegrated TDM program generally includes strategies designed to change the commuters travel beha vior in terms of mode choice, time choice and travel frequency. Research on TDM has focused on the evaluation of the effectiveness of TDM program to report progr ess and find effective strategies. Another research area, identified as high-priority research need by TRB TDM innovation and research symposium 1994 [Transportation Rese arch Circular, 1994], is to develop tools to predict the impact of TDM strategies in the future. These tools are necessary for integrating TDM into the transportation planning process and developing realistic expectations. Most previous research on TDM impact ev aluation was worksite-based, retrospective, and focused on only one or more aspects of TDM strategies. That research is generally based on survey data with small sample size due to lack of detailed information on TDM programs and promotions and commuter travel behavior patterns, which cast doubts on its findings because of potential small sample bias and selfselection bias. Additionally, the worksite-bas ed approach has seve ral limitations that affect the accuracy and app lication of analysis results.

PAGE 11

ix Based on the Washington State Commute Trip Reduction (CTR) dataset, this dissertation focuses on analyzing the part icipation rates of compressed work week schedules and telecommuting for the CT R affected employees, modeling the determinants of commuters compressed work week schedules and telecommuting choices, and analyzing the quantitative imp acts of an integrated TDM program on individual commuters mode choice. The major findings of this dissertation may have important policy implications and help TDM practitioners be tter understand the effectiveness of the TDM strategies in terms of person trip and vehi cle trip reduction. The models developed in this disse rtation may be used to evaluate the impacts of an existing TDM program. More importantly, they may be incorporated into the regional transportation model to reflect the TDM imp acts in the transportation planning process.

PAGE 12

1 1 CHAPTER 1 INTRODUCTION 1.1 TDM Development After the Second World War until the mid-1970s, American public policy focused on the construction of new highway f acilities to meet th e transportation needs from continued urban expansion and accelera ting automobile trav el. During that time period, a contention held by most transportatio n policy makers was that land use patterns and economic growth were two major sources of traffic. As a consequence, they believed that more roads should be built to reach adequate capacity to accommodate growing transportation needs and handle future demand [Wachs, 1990]. Em pirical evidence, however, suggests that because of so-called in duced travel, building more roads leads to more automobile travel. A st udy conducted by Fulton et al. [2000] finds that, on average, every ten percent increase in lane-miles result s in two to six percent increase of vehicle miles traveled (VMT). Lomax and Schrank [2005] conclude in The 2005 Urban Mobility Report of Texas Transportation Institute (TTI) th at This analysis shows that it would be almost impossible to attempt to maintain a constant congestion level with road construction alone. On the other hand, compared with th e rapid growth of population, licensed drivers, and even faster growth of vehi cle miles traveled, the highway supply has experienced a much lower growth for the last thirty years. For example, from 1976 to 1996, while the population increased by about 22 percent, the number of the licensed

PAGE 13

2 2 drivers increased by 34 percent, the vehicle miles traveled increased by 77 percent. In contrast, while the highway capital outlay ad justed by inflation increased by 56 percent from 1976 to 1996, the road mileage only incr eased by 2 percent. In fact, highway expenditures by all levels of government in 1996, after inflation adju sted, were about 56 percent of what they were for each vehicl e mile of travel in 1976 [Winters, 2000] Facing ever increasing congestion desp ite the expansion in road capacity, coupled with growing limits of transportation budgets and environmental concerns from air pollution to global warming, transportatio n policy makers gradually changed their view that building more roads is the only e ffective way to reduce congestion. Starting in 1980s, transportation professionals and public policy makers started to look at the demand side for solutions. The travel demand management (TDM) strategies were first implemented in early 1980s. A variety of incentives and subsidy programs aimed at increasing ride sharing and transit use by commuters were introduced. Since then, more and more communities ar e recognizing TDM as an essential part of the overall effort to effectively address transportation congestion [Bhattacharjee et al., 1997; Nozick et al., 1998]. They look for ways such as carpooling, vanpooling, parking charges and financial incentives to lower congestion by redu cing the number of vehicle trips on the road and increasing the number of passengers in each vehicle [Wilson, 1992; Parkhurst, 1995, 2000; Rose, 2002]. They also us e new technologies such as smart card and advanced traveler informati on systems to change the amount of time that vehicles use the road, thus lowering the load of the road network during the peak-hour periods [Winters, 2000]. Road pricing lowers traffi c volume during peak hours [Thorpe et al., 2000; Viegas, 2001; Nakamura and Kockelma n, 2002]. Alternative work schedules,

PAGE 14

3 3 including compressed work weeks, telecommuti ng, and flexible work hours, shift traffic out of the peak hour period and reduce comm uting personal trips [Giuliano and Golob, 1990; Tanaboriboon, 1994; Bhattacharjee et al., 1997; Nozick et al., 1998; Ory and Mokhtarian, 2005]. Travel Demand Management focuses on improving the efficiency of the transportation system through changing the tr avelers travel behavior rather than expanding the infrastructure. Traditionall y, TDM focuses on changing commuters mode choice by providing incentives or disincentives and su pport services. Contemporary TDM addresses not only mode c hoice, but also route choice, time choice, location choice, and travel frequency. 1.2 Employer Based Travel Demand Management Program Most of the TDM programs are employer ba sed, either mandatory or voluntary. Generally, the basic objective of a typical employer-based TDM program, such as the Washington State Commute Trip Reduction (CTR) program, is to reduce traffic congestion, reduce air pollution, and petrol eum consumption thr ough employer-based programs that decrease the number of commu te trips made by people driving alone [Washington State DOT, 2007]. The employers pa rticipating in the TDM program are required to implement programs that encourag e alternatives to drive-alone commuting to their worksites. Specifically, the general strategies that th e employers can choose to implement to achieve their CTR goals include:

PAGE 15

4 4 Alternative Work Schedules: Compressed work weeks and telecommuting. In terms of TDM, this group of strategi es functions to reduce personal trips and change the travel time. Employer TDM Support Strategies: Non-monetary inducements to encourage employees to use alternative modes rather th an drive-alone. These include preferential parking for high occupied vehicle (HOV) pa rking, rideshare matching services, employer transportation coordinators, and guaranteed ri des home. It also includes the activities focusing on promoting the TDM program, such as the regularly posting or distributing of CTR promotion material, conducting transporta tion events, and so on. In terms of TDM, this group of strategies functions to re duce the drive-alone trips by encouraging employees taking alternative modes. Travel Cost Changes: Measures such as imposition of parking fees, differential rates or discounts for carpools or va npools parking, transit fare subs idies, or in specific modal incentives or disincentives to any or all modes. In terms of TDM, this group of strategies functions to reduce the driving alone trips by increasing travel cost of driving alone or decreasing the travel cost of alternative modes. 1.3 Effectiveness Evaluation and For ecasting of an Employer-based TDM Research on TDM has focused on the evaluation of the effectiveness of TDM to report progress and find effective strategies. Another research area, identified as a highpriority research need by TRB TDM innovation and research symposium 1994 [Transportation Research Circular, 1994], is to develop tools to predict the impact of TDM strategies in the future. These tools are necessary for integrating TDM into the transportation planning process and developi ng realistic expectati ons [Winters, 2000].

PAGE 16

5 5 The direct measurement of the effectiven ess of the CTR program is the vehicle trips or peak period vehicle trips reduction. Based on the numb er of reduced vehicle trips, other measurements, such as the reduction of delay, travel time, and fuel consumption and emission, can then be derived. Generally, a compre hensive employer based TDM program achieves the goal of vehicle trip reduction through implementing worksite-based TDM strategies that focus on changing the commuters mode choice, travel time, and travel frequency. Specifically, through the compressed work week and telecommuting program to change the commute travel freque ncy and the time the commute trip occur, the employer TDM supports strategies and fi nancial incentives or disincentives to encourage employees to use alternative mode s to drive-alone, therefore reducing the vehicle trips. An integrated procedure of the employer-based TDM effectiveness evaluation, therefore, consis ts of estimating the number of employees working on compressed work week and telecommuting and percentage of employees shifted from driving alone to the alternative modes. The evaluation of the eff ectiveness of the employer-b ased TDM program can be categorized to evaluating an existing progr am based on the employee travel behavior survey and predicting or estimating the impacts of a program based the employer program implementation data. For an existing employer-based TDM program for which both the employer promotion data and empl oyee travel behavior information are available, such as the Washington State CT R program, the evaluation process generally consists of calculating and comparing the vehicl e trip rate or vehicle miles traveled for the program affected employer before and af ter the implementation of the program based on the employee travel behavior data. For mo st of other employer-based TDM programs,

PAGE 17

6 6 where the employee travel behavior information is not available, the program assessment normally includes applying the TDM models, such as EPAs COMMUTER model, to estimate or predict the vehicle trip rate change based on employer program implementation data. 1.4 Focus of This Dissertation Most previous research on TDM impact evaluation was worksite-based, retrospective, and focused on only one aspect of TD M strategies. Such research is generally based on survey data with small sample size due to lack of detailed information on TDM programs and promotions and commuter travel behavior patterns, which cast doubts on its findings because of potential sm all sample bias and self-selection bias. Additionally, the worksite-based approach, as I will elaborate later, has several limitations that affect the accuracy and application of analysis results. The Washington State Commu te Trip Reduction database provide a detailed information on both TDM strategies implemen ted by employer, worksite characteristics and employees travel behavi or and their job related characteristics, which makes a employee based systematic analysis of TDM ef fectiveness possible. This database tracks more than 1,000 worksites and around 300,000 individual employees from 1993 to 2005, which enables this research to avoid the problems of self-selec tion and small sample biases. Using this unique dataset, this disse rtation intends to analyze the TDM effectiveness and develop tool s to predict the impact of TD M strategies by addressing three issues: (1) For the CTR affected em ployees, what are the overall trends of compressed work week (CWW) schedule partic ipating rate, what are the factors that

PAGE 18

7 7 determine employees CWW choices; (2) For th e CTR affected employees, what are the overall trends of telecommuting participation rate, what are the determinants affecting employees telecommuting choices; (3) how TDM strategies, including the program promotion activities, parking management, and financial incentives or disincentives can affect commuters modal choice. The results fr om this dissertation ma y be directly used to evaluate the impacts of an existing TDM program and to identify the effective strategies based on the worksites charact eristics. More importantly, it may be incorporated into the regional transporta tion forecasting model to provide realistic prediction of the TDM impacts in the future and, at the same time, to improve the accuracy and predictability of the travel forecasting model. 1.4.1 Compressed Work Week The first issue involves compressed work weeks (CWW) schedule, which allows employees to work their regular number of hours in shorter-than-normal days per week or per pay period. In terms of TDM, comp ressed work week functions to reduce the commuters travel frequency and change the time the work trips occur. For example, if an employee works 4 days a week, she has to work 10 hours per day. This means she needs to leave home earlier and leave the office la ter. Therefore, peopl e working on compressed work weeks not only reduce the number of work trips, but also shift the work trips from peak period to non-peak period. 1.4.1.1 Determinants of Employees Work Schedule Choice Earlier studies on the compressed work week focus on the benefits and problems associated with its implementation [Alle n and Hawes, 1979; Nollen, 1981; Ronen and Primps, 1981; Wachs, 1990]. More recent studies focus on the impacts of CWW on

PAGE 19

8 8 vehicle trip reductions [Giuliano and Gol ob, 1990; Ho and Stewart, 1992; Hung, 1996] and individual activity travel patterns [Sundo and Fujii, 2005]. Restricted by data availability, there are no analysis on th e CWW participation tr end and no studies on examining the factors that determine commut ers decision to take the CWW. With Washington State CTR dataset, these important questions will be answered for the first time, which helps enrich the literature and provides new insight on TDM strategies to meet the goal to reduce trip rates and traffic congestion. There is no previous theoretical model or empirical work discussing the drive or constraints for CWW choices. Mokhtarian an d Salomon [1994], however, presents a conceptual framework for modeling telecommu ting choices, which I believe may also be suitable for modeling the work schedule choice. Following this guideline, the determinants that affect commuters choice of telecommuting would include (1) the commuters job characteristics, (2) the commute rs journey-to-work trav el characteristics, (3) the commuters socio-demographic characte ristics, (4) the att itudes of the employer towards CWW, and (5) the commuters personal preference. 1.4.1.2 Compressed Work Week Participation Trend Chapter 3 analyzes the trend of CWW participating rate from 1993 to 2005 and identifies the factors that influence co mmuters CWW choice. The analysis of the longitudinal CTR data indicates that for th e employees affected by the CTR program, the participation rates of CWW increase steadil y from 14.5 percent in 1993 to 20 percent in 2005. While the major pattern of CWW is sti ll working four days 40 hours per week (4/40) (7.3 percent in 2005), th e percentage of employees working on nine days 80 hours per two weeks (9/80) doubled from 1993 (2.9 percent) to 2005 (5.85 percent).

PAGE 20

9 9 1.4.1.3 Modeling the Compressed Work Week Choice A multinomial logit (MNL) model is first applied to analyze the determinants of CWW choices using the CTR da ta in 2005. From the MNL model, I find that employers promotion level of TDM programs is one of the key determinants of commuters decision of CWW choices. Commuters are more likely to participate in CWW programs with the increase in the promotion level, a meas ure of supportiveness of employer on TDM programs aimed at reducing vehicle trip rates. I also find that distan ce from home to work is another key factor that influences co mmuters decision of CWW choices. The longer the distance from home to work, the higher the probability to c hoose alternative work schedules. People using a single mode of transi t and shared ride are more likely to work on compressed work schedules compared with those using a single mode of driving alone. Another important finding is that the number of CWW program years, defined as the number of years the CWW program has been implemented by the employer since 1995, has significant, positive, and non-constant impact on the commuters CWW choices. The CWW program implementation ye ar has increasing effect on CWW choices until it reaches its peak in year 5. After year 5, its marginal effect falls until year 8, after which, it goes flat. This may suggest that it takes time for the employees to understand the benefits and the feasibility of CWW based on their personal information and job characteristics. Employees decision to partic ipate in CWW programs are also affected by their job title and their employers major business type. There are arguments, however, that the employees choice of work schedules, including working 6 days (3/36), 7 days (7/80) 8 days (4/40), 9 days (9/80), and 10 days (regular hours) per two weeks, is ordinal discrete choice. For an ordinal dependent

PAGE 21

10 10 variable, the appropriate model is ordered l ogit or probit regression. There differ from the multinomial logit model, which is based on random utility theory. In the ordered logit or probit model, the ordinal choice variable is assumed as the discrete realizations of an underlying, unobserved (or latent) continuous rando m variable. The choice set for each of the alternatives for the ordinal logit or probit model, therefore, is fixed. This constitutes the major drawback for its application in modeling employees work schedule choice since most of the employees do not have th e full options of the compressed work week schedules (less than 10 percent of CTR af fected employees have the full options of compressed work week schedules). To further examine the technical feasibility of the model, an ordered logit model is estimated based on the sub-sample of the employees with full options of work schedules and the results are compared with th at of the MNL model. Overall, the results from the ordered logit model are consistent with the major findings from the MNL model. 1.4.2 Telecommuting The second issue addresses telecommuting choices. Telecommuting is designed to allow commuters to use telecommunication t echnology to work at home or at a location close to home during regular work hours, rather than commuting to a conventional worksite at regular work hour s, thus saving their drivi ng time to work and, more importantly, eliminate vehicle trip s, which helps reduce congestion. 1.4.2.1 Previous Empirical Studies on Telecommuting Researchers interest in telecommuting ha s been continuous a nd growing since its first implementation as a part of public pol icy to address transportation congestion in

PAGE 22

11 11 1988 in California. Most earlier research focused on the impacts of telecommuting on household travel behavior. Many hypothese s have been formulated and tested [Mokhtarian, 1991; Pendyala et al., 1991]. Although the impact of telecommuting remains an unresolved issue because of conflicting findings, it seems that most researchers agree that, on net, telecommuting reduces total trips, especially peak-period trips, and generates a positive effect on th e environment [Hamer, 1991; Sampath et al., 1991; Quaid and Lagerberg, 1992, Choo et al., 2005]. Most of the empirical analyses of te lecommuting adoption and frequency have been based on either stated preference (SP) [Bernardino et al., 1993; Mahmassani et al., 1993; Mokhtarian and Salomon, 1995] or reveal ed preference (RP) data [Mannering and Mokhtarian, 1995; Mokhtarian and Salom on, 1997; Drucker and Khattak, 2000; Popuri and Bhat, 2003]. The findings, from both SP based and RP based analyses, however, seem to be inconsistent. Those inconsistenc ies may derive from the wide gap between preferring to telecommute and actually tele commuting. As discussed in Mokhtarian and Salomon [1995], while 88 percent of the total of 628 responde nts preferred to telecommute, only 13 percent actually did. One of the common drawbacks shared by most earlier empirical studies on telecommuting is data limitation. Most previo us empirical studies are based on small samples and have not clear definition either the telecommuters or their telecommuting frequency. For example, in most studies that apply the discrete choice model, the choice set is defined as frequently, infrequently, and rarely telecommuting, rather than number of telecommuting days per time period. The commuters are not distinguished between

PAGE 23

12 12 those self-employed or those who do not have or need a conventional office rather than home and those who have a fixed office but telecomm ute regularly. 1.4.2.2 Contributions of This Study To further strengthen the findings on telecommuting choices, this dissertation develops an ordered logit model to estim ate telecommuting choices based on a unique dataset with more than 200,000 observations. The employees choices of telecommuting are made from a set of mutually exclusiv e and collectively exhaustive alternatives, including not telecommuting, telecommuti ng one day, telecommuting two days, and telecommuting three or more days per two weeks. To model the telecommuting choice and its frequency through a discrete choice mode l, the dependent variable, therefore, is an ordinal discrete choice. Although multinomial logit and probit models have been widely used in discrete choice modeling and in se veral earlier studies on telecommuting choices, they may not appropriate because they fail to account for the ordinal nature of outcomes [Greene, 2000]. For an ordinal dependent variab le, ordered logit or probit regression is more appropriate. The data, collected from the Washingt on State Commute Trip Reduction (CTR) program, in program year 2005, has more th an 200,000 observations that have detailed information on employers characteristics a nd employees telecommuting patterns. The dataset includes only those employees who work in a worksite with at least 100 full-time employees with regular working schedules starting between 6:00 a.m. and 9:00 a.m. (inclusive) on two or more w eekdays for at least twelve continuous months [Washington State Legislature, 2007]. This indicates that the sample excludes the self-employed and other types of employees w ho do not have or need an office other than home.

PAGE 24

13 13 Furthermore, in this sample, the telecomm uters are defined as those who regularly telecommute one or more days per two w eeks. In other words, the employees who randomly or casually telecommute are not counted as telecommuters. This probably can explain why the telecommuting rate reported by the WA CTR data is dramatically lower than that reported by other studies. For ex ample, Drucker and Kha ttak [2000] reported a total telecommuting rate of 14.3 percent from the 1995 National Personal Transportation Survey, while based on the WA CTR data base, the telecommuting rate was only 1.51 percent in 1995. In another study conducted by Popuria and Bhat [2003] based on 1997 1998 Regional Transportation Household Inte rview Survey in New York, the total telecommuting rate was 15.4 percent, compared with the results from WA CTR data in 1997 of 2.21 percent. I believe this strict de finition may help generate more reliable results. Finally, this study focuses on examining th e effectiveness of telecommuting as a component of an integrated TDM program a nd predicting the telecommuting rate in the future. The empirical evidence may be applied to evaluate or predict the effectiveness of a TDM program. It may also be incorporated into local or regional travel demand forecasting models to better measure the overa ll performance of transportation system. The findings from this dissertation may also help policy makers when they consider alternative combinations of TDM strategies to be implemented. 1.4.2.3 Determinants of Telecommuting Choice Mokhtarian and Salomon [1994] develop a behavioral model of the individual choice to telecommute, in which they identi fy the possible constraints and drives of telecommuting choices. They define constraint as a factor th at prevents the choice to

PAGE 25

14 14 telecommute while drive is a factor that motivates commuters to begin telecommuting. Key constraints on telecommuting choices relate to awareness of telecommuting options, the organization, job, and psychological factors. The authors identify the key drives as work related, family related, leisure related, ideology related, and travel rela ted. Work related drives incl ude the desire to be more productive, independent, and flexible. Family a nd leisure related drives include the desire to spend more time with family and have mo re leisure time for othe r non-work ac tivities. Ideology related drive include certain peoples belief that te lecommuting can help protect the environment by reducing auto travel. If a commuter lives a long distance from work, or if the work related commute is burdens ome, then these two factors both work as drives. Given data availability, the variables in cluded in my empirical analysis include most of constraints and drives identif ied by Mokhtarian and Salomon. I use TDM promotion activities, the allowance of flexib le start/end work time, and the time the employer transportation coordinator spends on TDM promotion to measure supportiveness from employers, which may captu re organization related constraints. The number of years telecommuting has been a llowed at the worksite may capture the awareness constraint. I include employees job titles and work sche dules to capture job related constraints. The commute mode choice will be used to capture the travel related drive. The variables of commute distance, wh ether the worksite is located downtown, and the average property value by ZIP code in which the commuter resides can measure the family and leisure related drives. I believe the variables employers major business type and the existence of multiple shifts at the worksite can measure the work related drives.

PAGE 26

15 15 1.4.2.4 Telecommuting Participation Trend Analysis and Telecommuting Choice Modeling The data analysis from the Washington St ate CTR database indicates that, overall, the absolute number and per centage of telecommuters are small. In 2005, 5.83 percent of employees affected by the CTR law actually ch ose to regularly telecommute at least one day per two weeks. Compared with 1993, tw o years after the CTR law was passed, however, the number and percenta ge of telecommuters increased by at least five times by all job titles and employers business types. This suggests that telecommuting is a TDM program strategy with growing support and acceptance from both employers and commuters. I estimate the relationsh ip between telecommuting choices and a group of explanatory variables using a generalized ordered logit model. Telecommuting is categorized into not telecommuting, telecommu ting one day, two days, and three or more days per two weeks. To evaluate the mode l, I estimate the model again on a randomlyselected 80 percent sample and use the remaining 20 percent to test the models predictability. The model is further evaluated using 2003 data. 1.4.3 TDM Impacts on Commuting Mode Choice One of the major objectives of the em ployer-based Commuter Trip Reduction (CTR) program is to reduce vehicle trip s by implementing programs that encourage alternatives to drive-alone commuting to worksites [Washington State DOT, 2007]. Therefore, the impacts of the implemented TDM programs on commuters modal choices could be an important measure of TDM effec tiveness. The third goal of this dissertation

PAGE 27

16 16 is to address the impact of an integrat ed TDM program on journey to work modal choices. 1.4.3.1 Limitations of Employer Based Method Most previous research on TDM impacts have been worksite based, focused on one or more aspects of TDM strategies, a nd based on small samples [Mehranian et al., 1987; Brownstone and Golob, 1991; Peng et al., 1996; Cervero, 1996; Kuppam et al., 1999; Washbrook et al., 2006] The worksite-based approach estimates changes in mode split at an aggregate, worksite level by treating the worksite as th e analysis unit. Although most commute trip reduction programs are employer-based, using worksite as the analysis unit has limitations. Firstly, calculation of the aggregate mode split is highly affected by some factors that are hard to control or measure, for example the survey response rate. The nonrespondents are generally treated as having the same distribution of mode shares as that of valid respondents. It can be argued, however, that people driving alone are less likely to answer the questionnaire. Based on th is assumption, some studies treat the nonrespondents as driving alone, or treat the non-respondents as driving alone when the response rate is less than a certain amount, e.g. 70 percent. Since the impact of TDM on the worksites mode split is relatively low, the bias induced by the calculation could be significant. Secondly, some of the important determinants of mode choice, such as travel time and travel cost, can only take average values at the worksite level, while those variables are meaningful only from the perspective of individuals. The worksite-based approach

PAGE 28

17 17 also fails to catch varieties of individual trips, which is cr itical when the study focuses on quantifying the impact of reduced individua l trips. In addition, the worksite-based approach reduces the number of observatio ns available from which to make the estimates. This is especially important when the study ar ea is a sub-area, such as downtown or corridor. 1.4.3.2 Modeling the Impacts of an Integr ated TDM Program on Employees Journey to Work Mode Choice An employer-based TDM program generally includes different strategies. For most of those strategies, their impacts ar e more interactive th an independent. For example, an internal or external ride matc h program will be more effective if combined with reserved high occupancy vehicle (HOV) parking space or an HOV parking charge discount program. Focusing on only one aspect of TDM strategies without controlling for the availability of other TDM programs may result in omitted variable bias. Although the commuters travel behavior in terms of travel mode choice has been studied extensively, there is no empirical work that estimates the combined effects of a TDM program on an individuals modal choices. Among the various methodologies applied in human behavior study, the discrete choice model has been widely used in th e transportation community to study travelrelated human behavior, specifically the tr avelers mode choice and departure time choice. In chapter 5, a nested logit model is applied to estimate the determinants of employees modal choices based on a sample of more than 60,000 observations. I use a two-level nested logit model. The first nest in cludes motor, transit, and non-motor travel.

PAGE 29

18 18 In the second nest, motor is divided into driving alone and shared riding. The mode shares of each of the altern ative are: motor, 76.71 percen t (driving alone, 63.22 percent; Shared ride, 13.47 percent); transit, 15.25 percent; non-motor, 3.61 percent. Based on the nested logit model, the elasticity and marginal effects of financial incentives and TDM support and promotion progra ms are further calculated to evaluate the quantitative impacts of various TDM strategies on the modal choices. The variables in the utility function of the nested logit model include (1) characteristics of the commuter, including job title and work schedule; (2) characteristics of the connections between the commuters home zip code and the commuters worksite, including the commuting distance by mode, transit in-vehicle time, transit out-vehicle time, and transit number of transfers; (3) la nd-use characteristics of the commuters home ZIP code, including the average property valu e; (4) characteristics of the employer, including business type, total number of empl oyees, and the existence of multiple shifts at the worksite; (5) parking management at the worksite, including parking charge for SOV and HOV, ratio of onsite parking spa ces and total number of employees, and existence of reserved parking spaces for HOV; (6) financial subsidies for alternative modes, including the subsidy for transit, carpool, vanpool, bike, and walk; (7) employer TDM support/promotion strategi es/activities, including the av ailability of a guaranteed ride home program and the promotion act ivities of distributing program summary material, sending program information thr ough email, conducting tr ansportation events, and publishing TDM articles in employee news letters; (8) land-use characteristics at the worksite, including area type (downtown, rural, or other), existence of sidewalk, bike-

PAGE 30

19 19 lane, and onsite restaurant, a nd existence of onsite covered bi ke racks, clothing lockers, and showers. The results of this part of the study will not only provide a comprehensive, reliable quantitative and qualitative assessm ent of the impacts of TMD programs on the affected commuters mode choice, but it wi ll also explore the framework of a mode choice model that includes the TDM compone nts. This mode choice model may further be incorporated into the regional transporta tion model to reflect the impact of the TDM on the regional transportation planning process. The rest of this dissertation is organized as follows: Chapter 2 provides a brief literature review on travel demand management and its overall impact and effectiveness, Chapter 3 analyzes the CWW choices, Ch apter 4 addresses telecommuting choices, Chapter 5 analyzes the impact of TDM program s on journey to work modal choices, and Chapter 6 provides the conclusion.

PAGE 31

20 20 CHAPTER 2 TDM LITERATURE REVIEW Transportation Demand Management (TDM) programs were first introduced in the urban and transportation planning fields in the 1970s, aimed at providing alternatives for single occupancy commuter travel to save energy, improve air quality, and reduce the increasing congestion in most urban areas during the peak hours [Berman, 2002]. In the years since TDM was introduced, popular conc ern with travel demand management has grown. By the year 2000, at least 11 states had adopted substan tive regulations to implement TDM. The purpose of this chapter is to define TDM and to discuss the policy implications of its research. 2.1 What is TDM? 2.1.1 TDM Definition The Federal Highway Administration (FHW A) describes transportation demand management [FHWA, 2004] as follows: t o some, the realm of demand management applications is limited primarily to encourag ing alternatives to si ngle occupant vehicle travel for the commute to work. In practice, however, this narrow view is no longer consistent with the broad applications of demand-side strategi es currently underway across the country. Todays applic ations are not only limited to facilitating shifts in travel modethey also address shifts in travel r outes and travel departure-times (for all travelers, including single-occ upant vehicle drivers). Today s applications also extend beyond a focus on commute trips. At nation al parks, sports stadiums, university

PAGE 32

21 21 campuses, and other diverse destinations, transportation and facility managers are implementing demand-side strategies as part of coordinated efforts to reduce congestion. On bridges, and along corridors undergoing roadway reconstruction programs, demandside strategies are helping tr avelers avoid congestion by utiliz ing alternative travel routes, travel times and/or travel modesor by reduc ing the need for some trips altogether by facilitating work from home options a fe w days a month. A full understanding of demand-side strategies must recognize the reasonable limits of these applications. Demand-side strategies should not be consid ered total solutions to regional traffic congestion problems. Rather they should more often be implemented as part of an integrated set of solutions that balance supply-side infrastructure investments and demand-side strategies. Nationwide, there is no single definition of TDM. Here, I list a few definitions from the leaders in the field. The Victoria Transport Policy Institute [Victoria Transport Policy Institute, 2007] refers to TDM as a genera l term for strategies that result in more efficient use of transportation resources. TDM is a combination of various strategies that change travel behavior (how, when, and where people travel) to serve two purposes: increase transport system efficiency and ach ieve specific objectives ranging from reduced traffic congestion, road and parking cost savi ngs, increased safety, improved mobility for non-drivers, energy conservation, to pollu tion emission reductio ns. Winters [2000] defines TDM as the all-inclusiv e term given to measures to improve the efficiency of transportation systems. Washington State De partment of Transpor tation [WSDOT, 2002] has a working TDM definiti on. WSDOT defines TDM as a broad range of strategies that reduce or shift use of the roadway, ther eby increasing the efficiency and life of the

PAGE 33

22 22 overall transportation system. TDM program s influence travel behavior by using strategies that accommodate more person-trips in fewer vehicles, shift the location or time of day at which trips are made, or reduce the need for vehicle trips. From the above discussion, we can see how TDM developed from its traditional perspective to its contemporary one. The traditional TDM focuses on commute trips because they are the causes of peak-hour congestion. The primary mission of traditional TDM, thus, is to get commuters away from drive-alone into carpool, vanpool, transit, or other alternative modes [Ber man, 2002]. This may be achie ved through the provision of incentives, disincentives, and support services to change commuters travel behavior. Generally used tools include flexible work hours, compressed work weeks, preferential parking, transit subsidies, carpools and vanpool match services, and telecommuting. The contemporary TDM model broadens its tradi tional mission and incorporates policies and programs to address not only mode choice, but time choice, locati on choice, and route choice through technology, improved inform ation flow, and financial mechanisms [Winters, 2000; Berman, 2002]. 2.1.2 TDM Components Studied This research focuses on the employer based commuter trip reduction program. Specifically, the TDM components th at will be studied include: Alternative Work Schedules: Compressed work weeks and telecommuting works. In terms of TDM, this group of strategi es functions to reduce person trips. Employer TDM Support Strategies: Non-monetary inducements to encourage employees to use modes other than drive-al one. These include pref erential parking for high occupied vehicle (HOV) parking, rideshare matching services, employer

PAGE 34

23 23 transportation coordinators, and guaranteed ri des home. It also includes the activities focusing on promoting the TDM program, such as the regular posting or distributing of CTR promotion material, conducting transporta tion events, and so on. In terms of TDM, this group of strategies functions to re duce the drive-alone trips by encouraging employees taking alternative modes. Travel Cost Changes: Measures such as the imposition of parking fees, differential rates or discounts for carpool or vanpool parki ng, transit fare subsidies, or in specific modal incentives or disincentives In terms of TDM, this group of strategies functions to reduce the number of driving alone trips by increasing travel cost of driving alone or decreasing the travel cost of alternative modes. 2.2 Why is TDM Important? The past thirty years have witnessed a significant attitude change toward transportation planning. Afte r the Second World War until the mid-1970s, American public policy focused on expanding new highway facilities and transi t capacity to meet the transportation needs from continued ur ban expansion and accelerating automobile travel. During that time period, transportati on policy makers believe that land use patterns and economic growth were two major sources of traffic. As a consequence, they concluded that more roads should be built to reach adequate capac ity to accommodate growing transportation needs and handle future demand [Wachs, 1990]. Empirical evidence, however, suggests that, because of so-called induced travel, more road building leads to more automobile travel. A study c onducted by Fulton et al. [2000] finds that, on average, every ten percent increase in lane-miles results in a two to six percent increase in vehicle miles traveled (VMT). Lomax and Schrank [2005] conclude in The 2005 Urban

PAGE 35

24 24 Mobility Report of the Texas Transportation Institute (TTI), This analysis shows that it would be almost impossible to attempt to main tain a constant congestion level with road construction alone. On the other hand, the highway supply ha s experienced a mu ch lower growth during the last thirty years than that of population, license d drivers, and even faster growth of vehicle miles traveled. Fo r example, from 1976 to 1996, the population increased by about 22 percent, the number of licensed drivers in creased by 34 percent and vehicle miles traveled increased by 77 pe rcent. In contrast, over the same period highway capital outlays adjusted for inflati on increased by 56 percent, and road mileage only increased by 2 percent. In fact, highway expenditures by all le vels of government in1996, adjusting for inflation, were about 56 pe rcent of what they were for each vehicle mile of travel in 1976 [Winters, 2000]. There is no doubt that more severe urban congestion is th e direct result of these traffic and highway growth trends. The Ur ban Mobility Report documents that urban congestion has increased subs tantially from 1982 to 2003. In 2003, travel time during the rush-hour period in twenty-eight urbanized ar eas was at least 30 percent longer than that during the non-peak period, compared to only one such urban area having this severe congestion in 1982. Congestion caused 3.7 bil lion hours of travel delay and 2.3 billion gallons of wasted fuel with an estimated cost of more than $ 63 billion [Shrank and Lomax, 2005]. Facing ever increasing congestion despit e expansion in road capacity, coupled with slowing growth of tran sportation budgets and environmen tal concerns ranging from air pollution to global warming, transportatio n policy makers gradually changed their

PAGE 36

25 25 view that building more roads was the only effective way to reduce congestion. More and more communities are recognizi ng TDM as an essential part of the overall effort to effectively address transporta tion congestion [Bhattacharjee et al., 1997; Nozick et al., 1998]. These communities look for ways to lo wer congestion by reducing the number of vehicle trips on the road and by increasing th e number of passengers in each vehicle. They also use new technologies such as smart cards and advanced traveler information systems to change the amount of time that vehi cles use the road, thus lowering the load of the road network during the peak-hour periods [Winters, 2000]. 2.3 TDM Strategies and Their Effects As discussed above, TDM programs include va rious strategies that work together to reduce congestion. The question is whether th e TDM strategies are e ffective or not. In other words, what combination of strategies works better to serve certain objectives? Earlier work on the evaluation of the effectiv eness of TDM has reli ed on both aggregate data at the regional level and di saggregate data at the individual site level. A brief review of existing research is provided below. To better present the various TDM strategies, I divide the strategies into three major categorie s: strategies to change travel behavior by changing travel cost; strategies to change tr avel behavior by changing travel time; and other strategies. 2.3.1 Strategies to Change Travel Beha vior by Changing Travel Cost The basic idea behind this group of strate gies stems from economic principles. Like most normal goods, the demand for travel by any mode is not fi xed. If travel cost increases, people respond by traveling less. If the relative price of a substitute mode changes enough, people may switch to anothe r mode. The key question here is how

PAGE 37

26 26 commuters or travelers estimate their daily trip cost. There are a lot of studies on this question and they have obtaine d consistent findings: travelers treat the capital cost of owning a car, such as purchase price, intere st payments on car loans, maintenance costs, and insurance premiums as so -called sunk costs. On the ot her hand, travelers generally count as trip costs only their out-of-pocket costs, gasoline, parking, tolls and transit fares [Johnson, 1975; Louviere, et al. 1981; Adiv, 1980]. Based on the above assumption, TDM stra tegies aimed at changing commuters behavior by changing their travel cost generally include changing parking cost, and providing subsidies to transit use and ot her alternative modes such as carpooling, vanpooling, and road pricing. 2.3.1.1 Impacts of Parking Cost on Travel Behavior It is well known that as many as 95 pe rcent of American workers receive free parking from their employers [Vaca and Kuzmyak, 2005]. Free parking could be considered as a subsidy that encourages people to drive alone since it lowers travel cost [Shoup and Pickrell, 1997]. There are several studies on the effect on parking cost changes on journey-to-work mode choice. Francis and Groninga [1969] analyze the journey to work mode choice by employees wo rking in the Los Ange les Civic Center and find that when parking is paid by the c ounty and provided to county employees at no cost, 72 percent of the county employees chose to drive to work alone. At the same site, only 40 percent of federal employees drove to work alone when they had to pay for their parking. In another study, Shoup and Pick rell [1980] find that 20 percent fewer employees drive alone to work when they pa y to park than when the employer provides

PAGE 38

27 27 free parking. Comparing components of travel cost, Shoup & Pickrell conclude that free parking is a greater incen tive to drive alone than an offer of free gasoline. Based on a survey of over one hundred TDM programs at suburban mixed land use centers, Higgins [1989] concludes that the key difference between the most successful programs and those with little e ffect on commuting be havior rests on the increase in the cost of employee parking. In the state of California state law requires all employers with more than 50 employees to of fer commuters the option to choose cash in lieu of any parking subsidies offered. In a case analysis based on eight firms, Shoup [1997] reports that for the affected 1,694 empl oyees in the eight firms he studied, drivealone dropped by 17 percent afte r cashing out, while carpool s, transit use, and other alternatives such as walking and bicycling in creased by 64 percent, 50 percent, and 39 percent respectively. Additionally, vehicle miles traveled by affected employees fell by 12 percent. More importantly, he concludes that providing subs idies to people rather than free parking benefits employees, employers the community, and the environment. The impact of parking fees can also be reflected in the price elasticity of demand, which is the percentage change in the number of autos parking per 1-percent change in parking price. This price elasticity of de mand obtained either through empirical analysis or from simulation models ranges from 0.1 to 0.6, with 0.3 being the most frequently cited value [Vaca and Kuzmyak, 2005]. 2.3.1.2 Impacts of Out-of-the-pocket Cost on Travel Behavior There is another way to change commuter s travel cost: provide subsidies to transit users. It reduces the out-of-the-pocke t cost for transit users and increases the relative attractiveness of transit, which ma y change commuters tr avel behavior. There

PAGE 39

28 28 are many economic studies on the price elasticity of transit fare. It is estimated to range from .3 to .4, which indicates that 10 percen t decrease in transi t fare leads to an increase in ridership of 4 percent [S ullivan, 2003]. Based on extensive research, Transport Research Library [TRL, 2004] calcu lates that bus fare elasticities average around .4 in the short-run, -0.56 in the me dium run, and -1.0 over the long run, while metro rail fare elasticities are .3 in the short run and .6 in the long run. Bus fare elasticities are lower (-0 .24) during peak than off-peak (-0.51). The ways to change the travel cost for ot her alternative mode, such as carpooling, vanpooling, bicycling, and walking, can be direct financial subsidies or indirect incentives. Examples include the discount parking charge for vanpooling or carpooling and free bikes provided to commuters who ride bicycle to work. Analysis by Wambalaba et al. [2004] indicates that th e parameter of vanpool ridership with respect to fees is 0.026 to .148, which indicates that a one dollar decrease in vanpool price is associated with a 2.6 percent to 14.8 perc ent increase in the predicted odds of choosing vanpool with respect to drive-alone. York and Fabricat ore [2001] estimate the price elasticity of vanpooling at about 1.5, meaning that a 10 percent reduction in vanpool fares increases ridership by about 15 percent. 2.3.2 Strategies to Change Travel Beha vior by Changing Travel Time Travel time is identified as one of the most important variables that affect peoples mode choice, rout e choices, and departure time. Many TDM programs are designed to first change peoples travel time with intention to change travel behavior ultimately. Many studies, however, consistently demonstrate that commuters response to travel time are quite complex. The findings s uggest that people put much more weight on

PAGE 40

29 29 travel time reliability than the simple measure of total time elapsed. In other words, it seems that people care more about arriving their destination on time than minimizing their travel time [Wachs, 1990]. TDM strategies that address changi ng commuters travel time include: preferential parking for car pool and vanpool, compressed work week (CWW), flexible working hours, telecommuting, and high o ccupancy vehicle (HOV) lanes implemented by state or local government. Preferential parking for vanpools or car pools is one strategy widely implemented. HOV lanes are designed to provide advant age to carpool, vanpool or buses by allowing them to bypass congestion on adjacent lanes for all other uses, which may save the travelers in-vehicle time and provide desira ble time reliability. Combined with priority parking locations, the total in-vehicle and out -of-vehicle time can be shortened although carpool, vanpool or buses ha ve higher collection time. It is well known that some commuters are willing to depart to work very early to avoid congestion on the way to work. Some ot hers prefer to stay at work until the afternoon peak-hour is over. Many TDM stra tegies such as flexible work hours, compressed work week, and various other work hour variations are designed to reflect this phenomenon. Such programs are found to be very effective in some settings, able to reduce peak period congestion up to 20 percen t in some applications [Barton-Aschman and Associates, 1981]. On the other side, W achs [1990] argues that the benefits from these strategies could be quite localized si nce the effects of those programs on traffic stream dropped rapidly with the increase in distance from the worksite affected.

PAGE 41

30 30 Telecommuting is another strategy that may fall under this category. For some commuters who live far way from their worksite, telecommuting is designed to allow them to use telecommunication technology to wo rk at home or at a location close to home during regular work hours, rather than commuting to a conventional worksite, thus saving their driving time to work, and more importantly, eliminate some vehicle trips. Salomon [1985] concludes that the potential impact of tele commuting is complex and not necessarily completely beneficial. Later rese arch, however, seems to reach consensus that on net, telecommuting reduces total trips, es pecially peak-period trips, and generates positive effect on the environment [Sampath et al., 1991; Hamer et al ., 1992]. Choo et al. [2005] find that telecommuting has reduced an nual VMT by less than 0.8 percent. They believe that, even with such small impacts, telecommuting appears to be far more costeffective than public transit in terms of public sector expenditures for the same level of reduction of vehicle-trips achieved. Shafizadeh et al. [2007] studied the cost and bene fits of telecommuting and illustrate the conditions under which the busin ess case for telecommuting is supported or weakened. Conditions for the employee (the te lecommuter) are generally most favorable when: (1) the employer bears the equipmen t cost; (2) commute distances are above average; (3) the commute vehicle has belowaverage fuel economy; (4) travel time is highly valued; and (5) telecommuting is frequent. Conditions for the employer are most favorable when: (1) the telecommuter bear s the equipment cost; (2) there is low telecommuter attrition; (3) th e employee is highly productive on telecommuting days; (4) the employee's time is highly valued; and (5) telecommuting is frequent. For the

PAGE 42

31 31 employer, telecommuting is also favorable if parking and office space savings are realized. 2.3.3 Employer Support TDM Strategies Most TDM programs are employer based and can be either mandatory or voluntary. In a mandatory TDM program, empl oyers are required by th eir state or local governments to set up reduction goals of vehicl e miles traveled (VMT) and to implement specific support strategies to achieve the goa ls. One way in which employers can try to persuade employees to consider traveling by alternative modes, rather than drive-alone, is to provide various types of s upport that make it easier and more attractive to use those modes. These support programs typically consist of measures that he ighten awareness of the availability of other modes, provide info rmation on their service or use, or generally make it easier and more attractive to employ ees to consider their use. These employer support programs generally do not include measurable time or cost incentives or disincentives. Rather, they se rve to provide an improved set of conditions for employees to use an alternative, and provide incentives that are ta ngible and important, but not necessarily quantifiable by the employee [EPA 2005]. Many of the strategies in this category of programs are specific to the needs of the particular mode. These strate gies include ride matching a nd preferential parking for carpooling and vanpooling, on-site transit inform ation booths and pass sales for transit, sidewalk and shower facilities for people w ho bicycle to work. In addition to the modespecific types of strategies, there are actions employers can take that are almost universal in their applicability across all of the altern ative modes. Examples of these strategies include the following: (1) Employee Transpor tation Coordinators, ge nerally persons who

PAGE 43

32 32 are trained to provide informa tion or advice to employees re garding use of any alternative mode, in terms of where to go for informa tion, company policy and benefits, etc. (2) Guaranteed Ride Home, a program to help an employee go back home by alternative means if it is necessary to work late or in event of a personal emergency. (3) Flexible Work Hours, a formal or informal policy that allows employees some flexibility over the official office hours in order to meet the sc hedule of the chosen alternative mode. (4) Promotions through marketing and other methods to increase awareness of a given mode or employer incentive or to pr ovide prizes or awards for meeting some usage challenge. As discussed above, most support-type stra tegies do not translat e into changes in travel time or travel cost. While the impacts of those strategies on travel behavior are important and significant, they are comple mentary and interactive, rather than independent, and are relatively difficult to measure. In the Federal Highway Administration (FHWA)s TDM model, the support-type strategies are therefore estimated by categorizing them to different program levels for each employer and then associating the program level with an incremen tal change in the mode share of the mode to which the program is applied [FHWA, 1993]. Through a case study of thirty worksites in the Puget Sound region affected by the Washington State Commute Trip Reduction (CTR) program, Hendricks [2004] finds that management support and an effective employee transportation coordinator (ETC) are not necessary for a successful work site trip reduction program if the work site is lo cated in an area with access to high quality public transportation and employs lower-inc ome staff who must choose transportation cost saving over time savings and conveni ence. They are necessary, however, for a

PAGE 44

33 33 successful work site trip reducti on program if the work site is not located in an area with access to high quality public transportation. 2.3.4 Overall Effectiveness of TDM Strategies As discussed above, the effectivene ss of TDM strategies depends on the relationship between the incentives and/or disin centives to change travelers travel cost and/or travel time and the propensity of travel ers response in a particular travel market. Effectiveness evaluations consist of empiri cal studies of TDM programs using aggregate data at the regional level or disaggregat e data at the individual site level. In a study conducted by the Environmenta l Defense Fund on the potential effect of a comprehensive package of demand manage ment strategies (including road pricing) on vehicle miles traveled (VMT), re searchers conclude that VMT levels could be lowered to 1990 levels by the year 2000 and another 10 pe rcent reduction is e xpected by the year 2010 [Replogle, 1993]. Meyers [1997] reviews se veral successful a pplications of TDM actions to reduce urban congestion and enhance mobility. It seems that these studies agree that certain financial incentives or disi ncentives are the key for travelers to change their travel mode or travel behavior. The Maryland Department of Transporta tion adopted a Concurrency Management Systems (CMS) approach that focused on implementing a package of congestion reduction and mobility enhancement actions in targeted transportation corridors. The actions implemented include: (a) transpor tation demand strategies (b) transportation systems management strategies that consist primarily of traffic operations improvements, (c) public transit improvement s, (d) highway capacity impr ovements, (e) high occupancy vehicle lanes, (f) measures to encourage the use of non-moto rized modes, and (g) growth

PAGE 45

34 34 management and activity center strategies th at related to land us e and development [ITE, 1997]. They find that the strategies that work best in the targ eted corridor are road pricing and parking cost change while other strategies targeted at relatively small travel markets have relatively small effect [GAO, 1997]. In summary, this section briefly revi ews the origin a nd development of transportation demand management and focu ses on the empirical evidence on various TDM programs and strategies. Current availa ble empirical findings on the effectiveness of TDM programs suggest that TDM strategi es have potentially important effects on travel demand. It seems that those strategies ai med at changing traveler s travel cost have noticeable effect. More empirical evidence, however, is needed to estimate the effectiveness of TDM programs based on real da ta collected at both wo rksite and corridor levels over a relatively longer period. 2.4 Modeling Framework of TDM Strategies The state-of-the-art in travel demand anal ysis is quite advanced. The interest in travel demand management was originally spurred by a sustained national program of new facility constructions. Before the 1970s, travel demand analysis was used to improve the ability to make choices between large capital investments in different corridors and to evaluate highway projec ts of different capaci ties and operational characteristics. As the focus of transpor tation planning was shifted from highway capacity increase to travel demand manageme nt, the traditionally used travel demand models are not sensitive to travel time and co st variables and to transit, walk, and bike accessibility variables. Additionally, because th ose models are designed to predict traffic volume, they do not require a high level of accuracy. This makes the traditionally used

PAGE 46

35 35 model even less sensitive to TDM strategies [Wachs, 1990; Johnston and Rodier, 1994]. Because of the strict modeling requirements of th e Clean Air Act, some of the Metropolitan Planning Organiza tions are updating their travel demand models to increase their sensitivity to TDM policies. The following models are the widely used in TDM evaluation in the United States. A brief review of the models is provided as below. 2.4.1 Washington State TDM Effectivene ss Estimation Methodology (TEEM) Model The purpose of developing this model was to produce an analytical tool that could quantify the effectiveness of TDM and land us e strategies in the Central Puget Sound Region. The model was created based on local data sources and can estimate the effectiveness of 20 TDM and land use strategies at a corri dor or sub-area level. Each strategy is evaluated separately using diffe rent methodologies. The combined impacts can be evaluated based on the assumption of th e interaction of different strategy. The evaluation of the combined impacts of different strategies depends on the assumption of the interaction of the strategies There are four main categories of strategy combination. In some cases, the cumulative e ffect of combining most strategies can be found by sequentially predicting th e effect of one, then adjus ting the baseline data and applying the next one. Strategies such as these are referred to as multiplicatively additive. Other strategies, when combined, affect different markets and the results can be combined directly. These are referred to as di rectly additive. This could include a strategy affecting only employee trips being combined with a strategy affe cting only residential non-work trips. The third type of combination is strategies that conflict in ways that are

PAGE 47

36 36 not accounted for by readjusting the base shar es. These are referred to as conflicting strategies and a correction factor must be specified to be able to estimate the combined effect of both. The final category of strategy combination is referred to as synergistic. When combined, they produce greater results because of their supportive nature than a direct addition of their impacts would suggest. TEEM is designed to apply se nsitivity factors to base mode shares incrementally when more than one strategy is being teste d. By readjusting the ba se mode shares, the methodology can accurately represent the first tw o types of interactions above: directly additive and multiplicatively additive. If the strategies do not interact or affect the same markets and are directly additive, then no adjustment of the predicted changes is necessary at all. If they are multiplicatively additive, the readjusting of the base mode share provides an accurate assessment of the combined affect but the individual effects cannot be identified. The order in which they are tested does not aff ect the results. Only the conflicting and synergistic affects are not directly accounted for in TEEM. Users of TEEM need to be aware of when such in teraction may be occurring and certain adjustments need to be made [WA DOT, 2006]. 2.4.2 Environmental Protection Agency (EPA) COMMUTER Model This is a model developed by Cambridge Sy stem, Inc. for the U.S. Environmental Protection Agency (EPA). The first version of the model was released in 2000 and the model was updated in 2005. The basic objective of the model is to assess or evaluate the emission impacts of various transportation control measure strategies. The methodology and procedure of the model are based on th e Federal Highway Administrations Travel Demand Management Evaluation Model (FHWA TDM model).

PAGE 48

37 37 In the COMMUTER model, the TDM stra tegies are classified into four categories: employer TDM support strategies, alternative work sc hedule, travel time improvement, and travel cost changes. Th e first two categories are analyzed using relational factors in look-up tables, with a normalization procedure applied to the adjusted shares to ensure that changes are proportionate across the available alternatives and do not allow final choices to exceed 100 pe rcent. The strategies that involve changes to either travel time or cost are analyzed through the more rigor ous logit pivot-point procedure [EPA, 2006]. The COMMUTER model estimates the co mbined impacts of different TDM strategies by performing the calculation through a se quencing order. The order in which the COMMUTER model performs its calculations of travel changes is as follows: It first calculates the changes due to Altern ative Work Hours. This serves to readjust the travel population baseline to determine how many trips will be shifted to the off-peak period, and how many will remain in the peak period and be subject to application and analysis of the mode-choice oriented strategies. Next, mode shares of the remaining peak trip s are readjusted to reflect the effects of the employer TDM support strategies. All time and cost related stra tegies are tallied up and brought into the logit pivot-point procedure, which is then applied to the revi sed mode share starting point from step 2. 2.4.3 CUTR Worksite Trip Reduction Model This model was developed by the Center for Urban Transportation Research (CUTR) at the University of South Florida in 2004 using data ba sed on several thousand worksite trip reduction programs from three urban areas in the United States (Los

PAGE 49

38 38 Angeles, Tucson, and Washington State) th at have had trip reduction requirements on employers for many years. Two approaches we re used for the model building process: linear statistical regression models and nonlinear neural network models. The linear statistical regression models were used as a benchmark for the validity and accuracy of the neural network models. Several phases we re followed to build the neural network models. Models were built for each of the thr ee datasets using a vari ety of approaches of handling the data, including variable selec tion, grouping of incentives and the treatment of outliers. Models were also built after combining the data from the three urban areas into a single dataset. The only model to ge t better results simultaneously on all three cities validation sets was a neural networ k model built with no va riable selection on equally sampled combined data [CUTR, 2004]. 2.5 Summary This chapter reviews the origin a nd development of Transportation Demand Management, followed by a discussion of various strategies implemented by TDM programs. Empirical evidence regarding each category of strategies aimed at changing travel time, travel cost and other purposes is presented. Overall evaluation of TDM strategies is also reviewed. The last part of this chapter reviews the modeling framework of TDM programs and provides a brief disc ussion of three leading TDM evaluation models.

PAGE 50

39 39 CHAPTER 3 AN EMPIRICAL ANAL YSIS OF COMPRESSED WORK WEEK CHOICE 3.1 Introduction The compressed work week (CWW), one of the alternative working schedule programs designed to reduce vehicl e trip rates, is the focus of this chapter. Compressed work week allows employees to work thei r regular number of hours in shorter-thannormal days per week or per pay period. For example, employees may choose four 10hour days with one day off, or nine ninehour days with one day off every other week [Ronen and Primps, 1981; Gi uliano and Golob, 1990; Tanabo riboon, 1994; Bhattacharjee et al., 1997]. In terms of TDM, compressed work week functions to reduce the commuters travel frequency and change the time the work trips occur. If an employee works 4 days a week, 10 hours a day, she needs to leave home earlier and leave office later. Therefore, people working on compre ssed work week not only reduce the number of work trips, but also shift the work trips from peak period to non-peak period. The first CWW program was implemented in Southern California in 1982. The interest in CWW was later reinforced by two pu blic policy implementations in California. The first one was the 1989 Air Quality Management Plan, which proposed to reduce work trips by 30 percent by the year 2010 us ing CWW and other t ools. The second one was Regulation XV of the South Coast Air Quality Management, which requires employers with more than 100 employees at a single work site in Los Angeles, Orange

PAGE 51

40 40 County, and the urban areas in Riverside and San Bernardino counties to submit plans to achieve higher vehicle occupancy ratios (r anging from 1.3 to 1.75 depending on areas) and lower vehicle trip rates, thus reducing air pollution and congestion [Mokhtarian, 1991]. The states of Washington and Or egon also passed similar laws. The Commute Trip Reduction (CTR) Law was passed by the Washington State legislature and incorporated into the Washington Clean Air Act in 1991. The goals of the program are to reduce traffic congestion, air pollution, and petroleum consumption through employerbased programs that decrease the number of commute trips made by people driving alone. It calls for a statewide multimodal plan and re quires all state agencies to aggressively develop substantive programs to reduce commute trips by state employees. According to the CTR laws, the state's nine most populated counties (including the cities within those counties) are required to adopt CTR ordinanc es and provide support for local employers in implementing CTR. Employers are require d to develop a commuter program designed to achieve reductions in vehicle trips and may offer benefits such as subsidies for transit fares, compressed work schedules, teleco mmuting opportunities, and more. More than 1,110 worksites and more than 560,000 commuters statewide participated in the CTR Program in 2005 [Washington State De partment of Transportation, 2007]. Earlier studies on the compressed work w eek focus on the benefits and problems associated with its implementation [Alle n and Hawes, 1979; Nollen, 1981; Ronen and Primps, 1981; Wachs, 1990]. In 1993, the Fe deral Highway Administration (FHWA) issued FHWA TDM Evaluation Model, wh ich provides a guideline on evaluating the impacts of CWW. This model assumes a CW W participation rate of 22 percent for

PAGE 52

41 41 eligible office employees [Federal Highway Administration, 1993]. More recent studies focus on the impact of CWW on vehicle trip reductions [Barton-Aschman and Associates, 1981; Giuliano and Golob, 1990; H ung, 1996] and individual activity travel patterns [Sundo and Fujii, 2005]. Given the options of CWW from employers what are the factors that determine commuters decision to take the CWW is an important question that remains unanswered. This chapter identifies those factors that in fluence a commuter choices of whether they participate in CWW or not. This chapter also analyzes the trend and participating rate based on a large sample. This study may help policy makers evaluate the effectiveness of TDM strategies and choose the most efficient ways to cut trip rates. The results from this study also have important applications in regional travel demand forecasting. By incorporating CWW into those models, their pred ictability of trip ra tes can be improved. Analysis of the employee commute travel behavior survey data from 1993 to 2005 indicates that for the employees affected by the CTR program, the pa rticipation rate in CWW increased steadily from 14.5 percent in 1993 to more than 20 percent in 2005. While the major pattern of CWW is still work ing four days for 40 hours per week (4/40) (7.3 percent in 2005), the percen tage of employees working fo r nine days at 80 hours per two weeks (9/80) doubled from 2.9 percen t to 5.85 percent from 1993 to 2005. To identify the factors that determine co mmuters choice of CWW, I first apply a multinomial logit (MNL) model based on the 2005 data from the Washington State CTR database to estimate the employees work sc hedules choices. I find that an employers promotion level of TDM programs is one of th e key determinants of a commuters choice to become involved in CWW. Commuters are more likely to participate in CWW

PAGE 53

42 42 programs the more that employers support and promote it. I also find that the number of CWW program years, a measure of how l ong the CWW program has been implemented, has a positive but not constant effect on CWW choices. Distan ce from home to work is another key factor that influences a comm uters decision about CWW. The longer the distance from home to work, the higher th e probability that the employee will choose alternative work schedules. Employees mode c hoices of the journey-to-work affect their choices of working on CWW schedules. Commute rs using a single mode of transit and a shared ride are less likely to work on CWW sc hedules than those who simply drive alone to work, while commuters using mixed mo des are more likely to work on a CWW schedule than those who drive alone. Additiona lly, employees decision s to participate in CWW programs are also affected by their j ob titles and their em ployers major business types. To further examine the technical feasibility of the model, an ordered logit model is estimated based on the sub-sample of the employees with full options of work schedules and the results are compared with th at of the MNL model. Overall, the results from the ordered logit model are consistent with the major findings from the MNL model. The rest of this chapter is organized as follows. Section 2 introduces the dataset. Section 3 provides a brief desc riptive analysis of particip ation rate trend of CWW. Section 4 presents the discussion of the determinants of the employees work schedule choice. Section 5 presents the multinomial lo git modeling of the CWW choice, including methodology, model specification, and discussion of the main results. Section 6 present the results of the ordered logit model. Section 7 presents the conclusion.

PAGE 54

43 43 3.2 Data The Washington State Commute Trip Re duction (CTR) program is an employerbased regional Transportation Demand Mana gement (TDM) effort initiated in Washington State in 1991. The CTR law requi res employers to implement programs that encourage alternatives to drive-alone co mmuting to their worksites. The CTR Law applies to all employers with 100 or more full-time empl oyees arriving at work between 6:00 and 9:00 a.m. located in a county w ith a population greater than 150,000. By 2005, more than 560,000 employees working for more than 1100 worksite s living in nine counties in Washington State we re affected by this law. Em ployers affected by the CTR law are required to submit an Employer A nnual Report & Program Description form to report the summary information on the pr ograms they implemented. The affected employers are also required to measure empl oyee commute behavior every two years to measure their progress toward their CTR goals. The data are from the Washington State CT R Database. This database is designed to systematically organize and store the in formation collected in the Washington State CTR annual employer reports and biennial em ployee commute travel behavior survey conducted by the Washington State Department of Transportation. The employer annual report provides detailed information on employers characteristics and the TDM programs implemented by the employer, such as: Worksite and employer information, including the organization name, worksite street address, Employer Transportation Coordina tors (ETC) information, total number of employees, total number of affected employees, business type, etc.

PAGE 55

44 44 Program promotion information, including list of CTR programs implemented or promoted by the employer, such as distribute CTR summary information, conduct transportation events, publish CTR articles, etc. Worksite characteristics, including the information of the accessibility the worksite has to a list of facilities such as bus stops, shopping, child care, etc. Worksite parking information and parking management, including the total number of onsite/offsite parking spaces, parking charge for solo and HOV driver, availability of reserved/preferential HOV parking spaces, etc. Financial incentive and subsid ies, including ince ntives/subsidies for transit, vanpool, carpool, walking, bicycling, etc. Site amenities, including the availabi lity of covered/uncovered bicycle spaces/racks/lockers/cages, clothes lockers, showers, etc. Work schedule policy, including the availa bility (allowance) of compressed work week, telecommuting, and flexible work hours Other TDM programs availability. Include th e availability of guaranteed/emergency ride home program, internal match service, etc. The employee biennial commute travel behavior survey collects detailed information on employees commuting travel behavior, such as: Work schedule Commute modal choice, including driv ing alone, carpool, vanpool, transit, motorcycle, walking, and bicycling Commute distance, including one way dist ance from home to worksite in miles

PAGE 56

45 45 Compress work week schedule, includi ng work schedule of 5/40, 4/40, 3/36, 9/80, 7/80 Telecommuting schedule, including the da ys regularly telecommuting per two weeks Job title Home zip code The Washington State CTR database cont ains employers data from 1995 to 2005 and employees data for 1993, 1995, 1997, 1999, 2001, 2003, and 2005. The Washington State CTR database is the only dataset that provides detailed information on TDM strategies and the corresp onding employee commute travel behavior over time for tens of thousands of employees According to the annual report issued by the Washington Department of Transpor tation on CTR law implementation, the coordinator in each worksite randomly send paper surveys to the employees every two years. The target response rate is 70 percent. The data analysis indicates that although the response rate varies, the overall response ra te was as high as 77 percent in 2005. Among the total of 1100 worksites, more than 50 pe rcent has a response rate above 80 percent. The total valid number of indivi dual respondents is more than 200,000. Therefore, it is reasonable to believe the sample is representa tive of the population affected by the CTR laws. The relationship revealed between the explanatory variables and the CWW choices are also considered as reliable. 3.3 CWW Participation Trend for the Em ployees Affected by the WA CTR Laws In this section, I provide a brief descri ptive analysis of the CWW participation trend for the commuters affected by the Washington CTR programs from 1993 to 2005.

PAGE 57

46 46 The results from this analysis may be used directly to evaluate the effectiveness of TDM strategies and help decision makers choose the most efficient ways to cut trips rates based on the primary business type of the empl oyer and the job title of the employee. The CWW participation rate, as shown in table 3.1, increase d steadily from 14.5 percent in 1993 to about 20 percent in 2005. While the major pattern of CWW is still working four days 40 hours per week (4/40) the percentage of employees working on 9/80 (nine days 80 hours ev ery two weeks) doubled from 2.9 percent in 1993 to 5.85 percent in 2005. The percentages of employ ees working on 3/36 (three days 36 hours per week), 7/80 (seven days 80 hour for very two weeks), and other CWW schedules have relatively slight increase or remain stable from 1993 to 2005. This may suggest that the options of 3/36, 7/80, and other CWW schedules are more jobs relate d. In other words, people working on 3/36, 7/80 and other CWW sc hedules are more likely to choose those schedules because of their job characteristics. For example, the regular work schedule of a firefighter consists of two 24-hour days pe r week, for an average of 8 days per month. The regular work schedule for a hospital nur se is three days 36 hours per week. For an average employee, the actual possible opti ons of CWW schedules are 4/40 and 9/80. Table 3.1 Percent of Employees by Work Schedule from 1993 to 2005 Percent of Employees by Work Schedules (%) Program Year Num of Employees 5/40 3/36 4/40 7/80 9/80 Other 1993 188714 85.53 1.82 6.30 0.66 2.90 2.78 1995 204832 83.47 1.73 7.64 0.75 3.64 2.77 1997 256510 81.53 2.35 7.99 0.95 4.15 3.03 1999 238113 82.87 2.07 8.06 0.63 3.55 2.82 2001 246322 82.01 2.16 9.39 0.78 3.51 2.14 2003 247239 80.87 2.50 8.10 0.61 4.94 2.99 2005 273957 79.97 2.29 7.34 0.66 5.85 3.89

PAGE 58

47 47 In terms of job title, on average from 1993 to 2005, 17.2 percent of the sample is administrative support, 13.5 percent is craft/p roduction/labor, 15 per cent is management, 4 percent is sales/marketing, 6.9 percent is customer service, 35 percent is professional/technical, and ar ound 8 percent is the other. For all job categories, CWW program participation rate ha s been steadily increasing. The employees with job title of craft/production/labor, professiona l/technical, and the other ha ve the highest percentage of working on CWW schedules. Over the 12year period from 1993 to 2005, the growth rate for participation of CWW schedule progr ams ranged from 14 percent to 57 percent, representing annual increase rate of 1.2 pe rcent to 4.7 percent. The percentages of employees work on CWW schedules by job titl e from 1993 to 2005 are reported in Table 3.2. Table 3.2 Participation Rate for CWW by Job Title from 1993 to 2005 Avg. Num of Employees Percent of Employees on Compressed Work Week (%) Job Title N % 1993199519971999 2001 20032005 Administrative Support 3205217.159.35 10.88 11.62 10.50 11.86 12.66 14.25 Craft/Production/Labor 2513513.4515.28 20.14 22.73 22.81 23.19 22.27 23.94 Management 2798314.977.77 9.78 10.68 9.38 10.18 10.45 11.73 Sales/Marketing 74593.996.91 8.21 8.73 6.69 7.43 7.04 9.79 Customer Service 127996.8513.33 13.91 15.25 12.75 15.48 16.09 17.46 Professional/Technical 6633735.49 18.97 21.34 23.39 21.68 22.15 24.54 23.84 Other 151568.1121.02 20.73 22.00 19.04 17.64 20.95 24.00 In terms of primary business type of the employer, on average from 1993 to 2005, around 19.9 percent of the sample works for government, 17.7 percent works for manufacturing, 14.3 percent works for health care, and 9.4 percent works for financial service industry. All other busin ess types have lower than 10 percent of the sample size. For all of the business types with large sample size, the employees work for health care have the highest percentage of worki ng on CWW schedules, 33.6 percent in 2005. The

PAGE 59

48 48 employees work for manufacturing have highest growth of participation rate, more than doubled from 10.8 percent in 1993 to 22.7 percen t in 2005. The only business type that experienced a decrease in CWW pa rticipation rate is retail/trade. The participation rates of CWW by major business type from 1993 to 2005 are presented in Table 3.3. The participation rates for each work schedule by job title and by primary business type for the year 2005 are reporte d in Table 3.4 and Table 3.5 respectively Table 3.3 Percent of Employees on CWW by Employer Primary Business Type from 1993 to 2005 Avg. Num of Employees Percent of Employees on Compressed Work Week (%) Primary Business N % 1993199519971999 2001 20032005 Agriculture, Forestry, Fishing, Mining 775 0.45 7.5914.049.047.64 10.66 3.219.45 Finance, Insurance, Real estate 14632 9.38 5.405.796.658.25 10.51 11.179.90 Information Services/ Software/ Technical 11577 7.42 3.669.1717.856.89 6.51 5.365.51 Professional/ Personal Services 8174 5.24 6.767.958.739.25 9.99 10.1411.72 Retail/Trade 6796 4.36 11.111.879.287.26 7.45 6.637.96 Manufacturing 27540 17.66 10.7615.0718.3619.12 21.22 22.6624.29 Health Care 20413 14.28 26.3327.4629.1630.80 29.82 30.7933.61 Public Utility 7092 4.55 11.778.8211.799.67 9.70 12.2213.48 Military 11639 7.46 17.9922.4721.3516.56 13.89 17.1119.71 Construction 259 0.17 39.8151.6935.6312.46 17.80 18.0712.97 Transportation 3343 2.14 15.3914.114.717.6 26.1 25.7526.51 Government 28428 19.89 16.1421.0423.5923.27 24.33 26.1928.07 Education 5302 3.40 7.308.259.9412.15 11.79 13.7614.89 Other 5602 3.59 14.2014.5713.7214.32 13.59 10.6110.77 Table 3.4 Participation Rate for Each Work Schedule by Job Title in 2005 Num of Employees Percent of Employees by Work Schedules (%) Job title N % 5/40 3/36 4/40 7/80 9/80 Other Administrative support 33718 13.09 85.75 1.95 4.67 0.33 4.89 2.41 Craft/Production/Labor 26582 10.32 76.06 1.34 13.35 1.14 4.17 3.94 Management 34644 13.45 88.27 0.78 3.60 0.27 4.79 2.30 Sales/Marketing 9179 3.56 90.21 1.50 2.35 0.58 2.93 2.43 Customer service 21640 8.40 82.54 2.28 7.17 0.67 4.43 2.91 Professional/Technical 115039 44.66 76.16 2.92 8.05 0.76 7.46 4.65 Other 16769 6.51 76.00 3.27 7.96 0.82 5.02 6.93

PAGE 60

49 49 Table 3.5 Participation Rate for Each Work Schedule by Primary Business Type in 2005 Num of Employees Percent of Employees by Work Schedules (%) Primary Business N % 5/40 3/36 4/40 7/80 9/80 Other Agriculture, Forestry, Fishing, Mining 1228 0.51 90.55 0.49 4.80 0.33 2.12 1.71 Finance, Insurance, Real estate 23783 9.79 90.10 1.19 4.16 0.13 2.83 1.59 Information Services/ Software/Technical 24644 10.14 94.49 0.66 2.02 0.13 0.72 1.98 Professional/ Personal services 14467 5.95 88.28 1.61 3.61 0.35 3.35 2.81 Retail/Trade 12196 5.02 92.04 0.72 2.57 0.20 2.55 1.92 Manufacturing 39393 16.21 75.71 1.01 11.30 0.78 5.59 5.61 Health Care 32919 13.55 66.39 8.73 8.62 2.20 6.35 7.72 Public Utility 7944 3.27 86.52 0.77 8.13 0.39 2.51 1.69 Military 9781 4.02 80.29 0.48 2.86 0.76 11.50 4.11 Construction 563 0.23 87.03 0.53 8.17 0 2.66 1.60 Transportation 5813 2.39 73.49 3.91 12.42 0.58 6.26 3.34 Government 51229 21.08 71.93 1.81 10.95 0.57 11.74 3.01 Education 10190 4.19 85.11 2.16 4.75 0.42 2.41 5.14 Other 8879 3.65 89.23 0.83 4.14 0.27 2.77 2.75 3.4 Determinants of Employees Work Schedule Choice There is no previous theoretical model or empirical work discussing the drive or constraints for CWW choices. Mokhtarian and Salomon (1994), however, presents a conceptual framework for modeling telecomm uting choices, which I believe may also suitable for modeling the work schedule choice. Following this guideline, the determinants that affect commuters choice of telecommuting would include (1) the commuters job characteristics, (2) the commute rs journey-to-work trav el characteristics, (3) the commuters socio-demographic characte ristics, (4) the att itudes of the employer towards CWW, and (5) the commuters personal preference. For employees working on certain type of jobs or for certain type of employers, working on compressed work schedules is manda tory rather than optional. For example, the regular work schedule for a hospital nurse is three days 36 hours per week. One of the

PAGE 61

50 50 typical work schedules for fire fighters is working three cycles of 24 hours on duty and 24 hours off duty followed by a 96-hour off pe riod year round including weekends and holidays. Most emergency medical responders work a fixed 12-hour schedule, but some of them are assigned to a 24hour on duty, 48-hour off duty sc hedule. An analysis of WA CTR data shows that, even for average empl oyees, the participation rates of CWW vary dramatically for employees work on different jobs and/or for different industries. It is highly expected that the longer the distance from home to work, the higher the probability that employee will choose a lternative work schedules. The commuters travel pattern, specifically the mode choice, is expected to affect the commuters choice of CWW as well. The commuters work schedule choices affect not only the frequency of the home-based work trip, but also the time at which the travel occurs. The employees working on compressed work week have to leave home earlier and leave the office later every workday. This may make the transit and shared-ride options less attractive, especially for those working on 3/36 and other CWW schedules. The personal or family characteristics may also affect the employees CWW choice. Because working on compressed work week leads to leavin g home earlier and leaving the office later every workday, CWW ma y less attractive to employees that are responsible for taking care of a family. For the same reason, people from a family with young children may be less likely to work CWW. The other important factor that direct ly affects the adoption of CWW is the supportiveness of employer. Whether the CWW is encouraged, or it is only allowed is expected to play a significant role for commuters to make the decision whether to participate in CWW or not.

PAGE 62

51 51 3.5 Multinomial Logit Modeling of Work Schedule Choices The ultimate goal of this part of study is to develop and estimate a model that can be used to forecast the possibility of wo rking on CWW schedules of an employee based on the characteristics of the employee and empl oyer. This model can then be used to predict the number of employees work on differe nt schedules at the levels of worksite or traffic analysis zone (TAZ). This model, bui lt on a large sample, therefore, can not only be used to predict or evaluate the impact of a TDM program, but will al so be able to used to improve the accuracy of the travel demand forecasting for a regional transportation model. 3.5.1 Methodology The discrepancies of the work schedule for different employees are essentially the results of choice making from a set of mutu ally exclusive and collectively exhaustive alternatives. The theoretical framework that underpins the modeling of the choice made among or between a set of mutually ex clusive options is random utility model [McFadden, 1973]. The decision maker is assumed to maximize her utility by evaluating the attributes associated with each of alternatives. The choice made by the decisionmaker is determined by his preferences, attribut es of alternatives, and other constrains. In the random utility model, each individuals utility for each choice is a function of observed influences and random infl uences. For example, individual i s utility from choice j can be expressed as ij j i ijv b x U (4.1) where x is the vector of observed attributes an d individual characteristics influencing

PAGE 63

52 52 choice for each option, and ijv represents relevant but unobs erved influences. Individual i choose option j if ijU> ikUfor all j k Under the three assumptions: (1) ijvis independently distributed, (2) ijv is identically distributed, and (3) ijv is Gumbeldistributed with a location parameter and a scale parameter the probability of individual i choose option j is given by j i j ib x j b x ije e P (4.2) The model is normalized by setting the coefficien ts of base option to be zero to remove the indeterminacy of the model [Green e, 2000]. The log-odds ratio is given by ) / ln(j i Base ib x P P (4.3) 3.5.2 Model Specification I used the 2005 employee data to estimat e the logit model of CWW choices. The choice set includes: (1) work ing 5 days for 40 hours per week, denoted as 5/40, (2) working 3 days for 36 hours per week, denoted as 3/36, (3) working 4 days for 40 hours per week, denoted as 4/40, (4) working 9 days for 80 hours every two weeks, denoted as 9/80, (5) other CWW schedules, denoted as other. The observed influences in the model included those variables available from both employer and employee surveys. From the employee survey, I used three variables, including commute distance from home to work, employees job title, and employees journey-to-work mode choice, to capture in dividual differences. The commute distance from home to work measures the one-way di stance from an employees home to his or her usual work site, including miles for errands or stops made daily on the way to work.

PAGE 64

53 53 The employees journey-to-work mode choi ce was divided into the single mode of driving alone, transit, shared rides, and mixed modes. WA CTR data report seven job titles: administrative support, craft/produc tion/labor, management, sales/marketing, customer service, professi onal/technical, and other. Th us, I created seven dummy variables to reflect an individuals job title. Four groups of variables were from th e employer survey: business type, employers TDM program promotion level, number of CWW program years, and the existence of multiple shifts at a worksite. 10 dummy variables were created to reflect the primary business of the employer: finance/real estate/insurance, information service or software, professional/personal service, re tail/trade, manufact uring, health care, transportation, government education, and other. Employers TDM promo tion level is an index used to measure the supportiveness of employers on employees choice of TDM programs, and more specifically in this cas e, the choice of CWW schedules. This index was constructed to reflect the overall implementation of TDM strategies. The number of CWW program years, de fined as the number of years the CWW program has been implemented since 1995, was used to capture the effect of time on CWW choices. The effect of this variable wa s expected to be posit ive since it takes time for employees to understand the benefits of CWW programs and make transitions accordingly. The time effects were also expect ed to be not constant. Therefore, I created 11 dummy variables to reflect the number of years that had passe d since the initial implementation of CWW programs. If a worksi te started the CWW programs in 2005, the number of years would be zero, which is the base value for this variable and excluded from the regression The last control variable is the existence of multiple shifts, a dummy

PAGE 65

54 54 variable reflecting whether a worksite requires multiple shifts. The detailed definitions of selected variables are presented in Table 3.6 (other variables are self-explanatory, therefore not reported). TABLE 3.6 Selected Variable Definitions Variable Definition Distance One way distance in mile commute from home to work location Shift Does this worksite have multiple shifts? Shift=1, if Yes; Shift=0, Otherwise Drive alone =1, if drives alone to work for the whole week =0, otherwise Transit =1, if takes public transit to work for the whole week =0, otherwise Shared rides =1, if carpools or vanpools to work for the whole week =0, otherwise Mixed modes =1, if takes at least two different modes to work for the whole week =0, otherwise Promotion Level Employer TDM program promotion level =0, No CTR promotion =1, Post CTR promotional materials for employees, OR provide information about the worksite CTR program during new employee orientations or in hiring packets =2, All the above, PLUS: Conduct transpor tation events/fairs and/or participate in county/state CTR promotions/campaigns, OR send electronic mail messages about the CTR program =3, All the above, PLUS : Publish CTR articles in employee newsletters =4, All the above, PLUS: Distribute CT R information with employee paycheck In 2005, there were about 273,000 valid observations from the employee commute travel behavior survey. There was, however, inconsistency about the availability of the CWW sc hedules between what was re ported by employers and the choices made by employees. For example, fo r a certain worksite, an employer reported that no CWW program was available, while ce rtain employees indicated they were on one of the CWW schedules. To determine the actual availability of each of the CWW options, I calculated the total number of employees working on each of the work

PAGE 66

55 55 schedules for each worksite based on the employee survey data. The results of this calculation were then compared with the in formation provided in the employer survey. For each of the CWW schedules, if the numbe r of employees working on a particular CWW schedule was zero and the employer re ported that this CWW schedule was not allowed, I assumed it was not available for al l the employees working for this employer. Based on the results of this procedure, if none of the CWW schedules were allowed at the worksites, all of the observations (employ ees) working for this employer were excluded from the sample. The final sample size was 181,009 3.5.3 Regression Results The model was estimated based on 2005 sample. The results of multinomial logit regression model are presented in Table 3.7. Columns 2 to 5 report the coefficients for the CWW schedules of 4/40, 3/36, 9/80, and other. The base is the regular schedule of 5/40. The value of the log likelihood function at its maximum, Log-L( ), is -102,164.4. The R-Squared, an informal goodness-of -fit index (reported by LIMDEP, the software I use to run the regression) that measures th e fraction of an init ial log likelihood value explained by the model, defined as 1-Log-L( )/Log-L(0), is 0.5614. The chi-square, a statistic used to test the null hypothesis that all the pa rameters are zero, defined as 2(Log-L(0)Log-L( )), is 277,23.02, which indicates that I can reject the null hypothesis that all the parameters are zero at the level of 0.001 or better. Examining the coefficients in the models for the choices of CWW, it was first observed that the constant terms for the ch oices of 3/36, 4/40, 9/80, and other were all negative, suggesting that the average effect of those unobser ved influence variables was in the direction of not participating in C WW. This was expected since around 80 percent

PAGE 67

56 56 of employees in the sample chose the regu lar schedule when they had the option to participate in the CWW program in 2005. The coefficients of the one-way distance fr om home to work for all the choices of CWW schedules were positive and statistically significant at the le vel of 0.01 or better. This result suggests that all ot her things being equal, commut ers have a higher probability to choose participating in CWW when the home-to-work distance is longer. The coefficients of TDM promotion level for all the choices listed in the model were positive and statistically significant at the level of 0.01 or better. This finding suggests that employers suppor t of TDM strategies plays a very im portant role in commuters choice of compressed work week schedules. As discussed above, the TDM program promotion is not specifically for the CWW promotion but for the whole TDM program. For CWW, the index of the empl oyer TDM program promotion may be more likely to serve as the reflection of the att itude of the employer toward the employees participation in CWW. In other words, th is result shows that whether the CWW is encouraged or it is only allowed does matter for commuters trying to make the decision whether to participate in CWW or not. It also shows that the coefficient for the TDM promotion level in the util ity of 4/40 is greater than th e one for 9/80, which in turn, is greater than the one for 3/36. This means that increasing the TDM promotion level is associated with an increased preference for 4/40 and 9/80 compared with 3/36. This may again support the expectation that the 3/36 sc hedule is more job ch aracteristics related, and therefore is less likely to be imp acted by the employers TDM promotion. The coefficients of transit and shared ride were all negative a nd significant for all of the CWW choice categories at the level of 0.1 or better. This finding suggests that,

PAGE 68

57 57 compared with those who drive alone, pe ople taking transit or carpooling or vanpooling are less likely to work on C WW schedules. It also shows that the coefficients of 3/36 and other are greater than those of 4/40 a nd 9/80. This suggests that among the CWW schedules, transit-commuters or carpoolers or vanpoolers are less likely to work on 3/36 than other CWW schedules. The commuters work schedule choices affect not only the frequency of the home-based work trip, but also the time at which the travel occurs. The employees working on compressed work week ha ve to leave home earlier and leave the office later every workday. This may make the transit and shared-ride options less attractive, especially for those working on 3/36 and other CWW schedules. It is interesting to see that the employees usin g mixed modes are more likely to work on CWW schedules compared with those who drive alone. The coefficients of the number of CWW program years were all positive and statistically significant at the level of 0.01 or better. The non-constant coefficients also confirm the expectation that the time effect on CWW choices is not equal. Figure 3.1 illustrates the time effect, suggesting that CWW program implementation has increasing effect on CWW choices until it reaches its peak in year five. After year five, its marginal effect falls until year eight, after which, its marginal effect goes flat. It seems that CWW programs have larger effects during the first fi ve years. This finding suggests that when evaluating the impacts of CWW on person tr ip reductions, how long the CWW programs have been implemented should be incorporated and their effects should not be the same. The coefficients of multiple shifts for all the choices were positive and statistically significant at th e level of 0.001 or better. Th is variable controls the characteristics that cannot be captu red by job title and business type.

PAGE 69

58 58 Figure 3.1 CWW Program Year Effect on CWW Choices 0.000 0.200 0.400 0.600 0.800 1.000 1.200 01234567891011 Number of CWW program yearsCoefficient There are six dummy job title variab les used in the regression. For those individuals who worked as administrative staff and managers, th e coefficients were consistently negative and stat istically significant, sugges ting that commuters working under the above mentioned job title have a lo wer likelihood of choosing alternative work schedules than choosing a regular schedule co mpared with those with other job titles. This is not surprising considering their job characteristics. Managers supervise other peoples work. When they are not around, some decisions may not be made on time and performance of other members under superv ision may not be c onsistent, which may lower the overall efficiency of the worksi te. Employees working as administrative support provide supportive work for managers. It is expected to see supporting staff work the same schedule as managers. This s uggests that when Employer Transportation Coordinators (ETCs) at each worksite decide which programs to implement, they may consider the restrictions of job characteristics to improve the effectiveness of the programs they implement.

PAGE 70

59 59 TABLE 3.7 Empirical Results for CWW Choices Variable 3/36a, b 4/40a,b 9/80a,b Other a.b Constant -6.257*** (36.0) -4.503*** (38.8) -4.148*** (35.0) -4.437*** (34.7) Distance 0.007*** (5.1) 0.011*** (14.7) 0.005*** (5.9) 0.003*** (3.0) Promotion Level 0.117*** (4.5) 0.138*** (8.9) 0.127*** (7.5) 0.138*** (6.9) Shift 0.405*** (6.8) 0.357*** (11.9) 0.365*** (12.2) 0.602*** (13.8) Transit -1.438*** (8.9) -0.469*** (8.4) -0.296*** (6.2) -0.989*** (11.7) Shared Ride -1.185*** (8.6) -0.130*** (3.1) -0.073* (1.8) -0.837*** (12.3) Mixed Modes 1.782*** (46.1) 1.455*** (65.0) 0.720*** (27.6) 0.906*** (31.7) Administrative Support -0.484*** (6.0) -0.521*** (10.0) -0.143** (2.5) -1.054*** (17.3) Production/Labor -0.091 (1.0) 0.576*** (11.9) -0.248*** (3.9) -0.572*** (9.5) Management -1.385*** (13.5) -0.854*** (15.6) -0.225*** (3.9) -1.187*** (19.3) Sales/Marketing -0.335** (2.3) -0.990*** (9.0) -0.207** (2.1) -0.776*** (7.5) Customer Service -0.309*** (3.6) -0.058 (1.1) -0.173*** (2.7) -0.716*** (11.1) Professional/Technical 0.070 (1.1) 0.122*** (2.9) 0.479*** (9.8) -0.262*** (5.9) Finance/Real Estate/Insurance 0.484*** (3.8) -0.143** (2.6) -0.741*** (12.4) -0.493*** (6.2) Information Service/Software -0.359** (2.2) -0.970*** (12.3) -2.156*** (19.5) -0.374*** (4.3) Personal Service 0.984*** (7.3) -0.202*** (2.9) -0.431*** (6.3) 0.207** (2.5) Retail/Trade 0.125 (0.7) -0.412*** (5.1) -0.659*** (8.1) -0.152 (1.6) Manufacturing 1.250*** (10.2) 0.925*** (20.9) 0.160*** (3.5) 0.936*** (16.3) Health Care 2.198*** (20.3) 0.418*** (8.8) -0.065 (1.4) 1.079*** (19.0) Transportation 1.637*** (11.5) 0.529*** (7.3) -0.306*** (3.2) 0.255** (2.4) Government 1.121*** (10.0) 0.926*** (21.7) 0.798*** (19.9) 0.413*** (7.1) Education 0.919*** (6.8) -0.133* (1.9) -1.062*** (11.7) 0.547*** (7.3) CWW Year 1 0.675*** (6.9) CWW Year 2 0.590*** (6.1) CWW Year 3 0.770*** (8.2) CWW Year 4 0.890*** (9.1) CWW Year 5 0.974*** (10.1) CWW Year 6 0.572*** (5.3) CWW Year 7 0.438*** (4.5) CWW Year 8 0.445*** (4.7) CWW Year 9 0.565*** (6.1) CWW Year 10 0.604*** (6.5) N(R-Squared) Log-L Chi-squared[94] 181,009 (0.5614) -102164.4 27723.02*** a absolute value of z-statistics in parentheses. b *2-tail significance at = 0.10. **2-tail significance at = 0.05. ***2-tail significance at = 0.01. There were nine business type variables used in the regression. The coefficients of information service/software were consiste ntly negative, while those of manufacturing and government were consistent ly positive. Other business types had coefficients with mixed signs.

PAGE 71

60 60 The coefficients of information service/ software were negative and statistically significant at the level of 0.01 or better. This suggests that employees working for the information service industry, compared with people working in othe r business types, are less likely to choose alternative work schedul es. A possible explanation for this finding may be that information service commuter s are more likely to choose other TDM programs such as telecommuting since their jobs may be done at home or other locations close to their home. A data analysis using the WA CTR database confirms that, overall, 15 percent of commuters in this industry choose telecommuti ng at least one day per week compared with the average of less than 6 percent. The coefficients of manufacturing were positive and statistically significant at the level of 0.01 or better on al l CWW work schedules. This result suggests that people working in manufacturing have a higher pr obability of choosi ng alternative work schedules. This also implies that, fo r manufacturing employers, among the TDM programs that are aimed at reducing journeyto-work person trips, a CWW program may be a very effective method since manufacturing employees are more likely to be required to physically work at their worksites. The coefficients on health care were posit ive and statistically significant at the level of 0.01 or better for the choice of 3/36, 4/40, and other CWW schedules, which suggests that employees working in the health care industry ha ve a higher probability of choosing alternative compressed work schedu les of 3/36, 4/40, an d other schedules, compared with other business type s. It also shows that the coefficient of 3/36 is greater than that of other and 4/40, which means among the CWW schedules, they are more likely to choose 3/36. As discussed above, in the health care industr y, employees such as

PAGE 72

61 61 registered nurses are required to work the 3/36 schedule. The result captures this requirement. In the model, I only focus on th e commuters actual c hoices since I am not able to tell whether a commuters choice of CWW is required by the employer or selected discretionally by the employee. From the above analysis, it is reasonable to conclude th at overall, an employees job title and an employers major business type are important determinants of an employees decision about CWW. For certain business types, such as manufacturing, health care, and transportation, since their employees are more likely to be required to physically work at their work sites, CWW could be a very effective way to reduce person trip rates. For an industry like information serv ice/software, since the jobs can be done at home or other places rather than a works ite, telecommuting may be a better program to reduce journey-to-work person trips. As a final check on the analysis, I estima te the model using 80 percent of the 2005 sample and test the predictability of the model using the 20 percent of excluded sample. By comparing the predicted probability with th e survey result, I feel confident that the model is able to predict the choices about CWW fairly well. The model predicts that 21.11 percent of employees would participat e in CWW in 2005, 0.68 percent lower than the survey results. The predictio n for 9/80 is 6.34 percent, wh ile the survey result is 6.72. For all of the other CWW choices, the diffe rence between model pr ediction and survey results are less than 0.15 percent. The coefficients obtained from the multi nomial regression were then applied to the 2003 survey data to predict the likelihood for CWW choices. Overa ll, I predicted that around 19.89 percent of employees who have th e option chose CWW, compared with the

PAGE 73

62 62 survey result of 20.97 percent, which is very close. The model, however, over-predicts the choice of 9/80 and under-predicts the c hoice of 4/40. These diffe rences are expected as a result of different da tasets and changing CWW partic ipation trends. As shown in Table 1, from 1993 to 2001, 4/40 had been the most popular CWW choice. After 2001, however, the choice of 9/80 had increased fr om 3.5 percent to 5.9 percent, while the participation rate for 4/40 decreased from 9.4 percent to 7.3 percent. Using 2003 data may not capture this trend accurately. Overall, I believe the models predictability is satisfactory and the coefficients obtained fr om the regression may be incorporated to regional travel demand models for future trip rate forecasting. A Detailed comparison is reported in Table 3.8. TABLE 3.8 Comparisons of the Model Predictions and Survey Results Average Compressed Work Week Percentage (%) Program Year 2005* Program Year 2003** CWW schedule Model Survey Model Survey 3/36 2.34 2.38 2.30 2.60 4/40 7.96 8.09 7.61 9.02 9/80 6.34 6.72 5.73 5.68 Other 4.47 4.60 4.25 3.66 Total CWW 21.11 21.79 19.89 20.97 *Based on the 20 percent of 2005 sample that is excluded from the model estimation **Based on all of the 2003 sample 3.6 Ordered Logit Modeling of Work Schedule Choices There are arguments, however, that the employees choice of work schedules, including working 6 days (3/36), 7 days (7/80) 8 days (4/40), 9 days (9/80), and 10 days (regular hours) per two weeks, is ordinal discrete choice. Fo r ordinal dependent variable, the appropriate model is ordered logit or probit regression. Differs from the multinomial logit model, which based on the random utility theory, in the ordered logit or probit model, the ordinal choice variable is assu med as the discrete realizations of an

PAGE 74

63 63 underlying, unobserved (or latent) continuous ra ndom variable (The detailed introduction about the ordered logit model is included in chapter 4). The choice set for each of the alternatives for the ordinal logtit or probit m odel, therefore, is fixed. This constitutes the major drawback for its application in mode ling employees work schedule choice, since most of the employees do not have the full op tions of the work sc hedules (less than 10 percent of CTR affected employees have the full options of compressed work week schedules). To further examine the technical feasibility of the model, an ordered logit model is estimated based on the sub-sample of the employees with full options of work schedules and the results are compared with that of the MNL model. The regression results of the ordered logit model are reported in table 3.9. The coefficients of commute distance, TD M promotion level, and the existence of multiple shifts are all positive and significant at the confidential level of 99 percent of better. These findings suggest that the longer commute di stance, higher TDM promotion level, and the existence of multiple shifts at the worksite are likely to increase the possibilities the employee make the transition from working on regular hours to working on CWW and from working more days to work ing less days per tw o weeks (from 9/80 to 4/40 to 3/36). The coefficients for transit and shared ride ar e negative, for mixed modes is positive, while all of them are significant at the confidential level of 99 percent of better. Once again, it indicates that, compared with driving alone, the pe ople using the single mode of transit and shared ri de are less likely to work on CWW, while those using the mixed modes are more likely to work on CWW.

PAGE 75

64 64 Although the interpretation of the coeffici ents of the ordere d logit model is different from that of the multiple logit mode l, overall, we can still see that the results from the ordered logit model are consistent with the major findings from the MNL model. Table 3.9 Ordered Logit Model for CWW Choices Variable Coefficient z-statistics Distance 0.009***15.22 Promotion Level 0.118***10.65 Shi ft 0.397***18.38 Transi t -0.421***-12.14 Shared Ride -0.192***-6.54 M ixed Modes 1.312***81.81 J ob titleA dministration Su pp or t -0.411***-11.81 J ob title-Production/Labo r 0.387***10.65 J ob titleM ana g emen t -0.683***-18.82 J ob title-Sales/Marketin g -0.580***-8.81 J ob title-Customer Service -0.179***-4.83 J ob title-Pro f essional/Technical 0.240***8.3 Business t yp eF inance/RealEstate/Insurance -0.304***-7.81 Business t yp eI n f ormation Service/So f tware -1.479***-23.14 Business t yp e-Personal Service -0.088*-1.89 Business t yp e-Retail/Trade -0.588***-10.04 Business t yp eM anu f acturin g 0.960***27.68 Business t yp eH ealth Care 0.771***24.75 Business t yp e-Trans p ortation 0.649***12.03 Business t yp e-Governmen t 0.803***27.26 Business t yp e-Education -0.220***-4.51 Tele Year 1 0.450***3.3 Tele Year 2 0.452***3.36 Tele Year 3 0.463***3.51 Tele Year 4 0.791***5.89 Tele Year 5 0.770***5.75 Tele Year 6 0.362**2.54 Tele Year 7 0.345***2.56 Tele Year 8 0.1411.06 Tele Year 9 0.403***3.07 Tele Year 10 0.354***2.7 Cut Off Point 1 3.356 Cut Off Point 2 3.953 Cut Off Point 3 5.333 Cut Off Point 4 5.591 N ( Pseudo R2 ) 13 637 ( 0.1089 ) Lo g likelihood ( LR chi2 ( 31 ) ) -85729.6 ( 20963.5*** ) *2-tail significance at = 0.10. **2-tail significance at = 0.05. ***2-tail significance at = 0.01.

PAGE 76

65 65 3.7 Conclusion This chapter analyzes the participation trend of CWW schedules and applies multinomial logit model to estimate the choices of CWW schedules using the Washington State CTR 2005 survey data The data analysis indicated that for the employees affected by the CTR program, th e participation rate in CWW increased steadily from 14.5 percent in 1993 to about 20 percent in 2005. While the major pattern of CWW was still working 4 days for 40 hours per week (4/40) (7.3 percent in 2005), the percentage of employees working 9 days fo r 80 hours per two week s (9/80) doubled from 2.9 percent to 5.85 percent from 1993 to 2005. A multinomial logit model is developed a nd developed to predict the employees choice of CWW schedules. The models predic tability was analyzed by comparing the predicted result with the survey results. The difference was very small. I also used the 2003 data to verify the model. Again, the difference between the prediction and the survey result was reasonable. I found that employers promotion level of TDM programs is one of the key determinants of a comm uters decision about CWW. Commuters are more likely to participate in CWW progr ams the more that employers support and promote it. Employees journey-to-work mode choices also affect their choices of working on CWW schedules. Compared with those who drive alone, the commuters using a single mode of transit or shared ride are less likely to work on CWW schedules, while the commuters using mixed modes are more likely to work on a CWW schedule. I also found that the number of CWW program years, a measure of how long the CWW program has been implemented, have a positive but not constant effect on CWW choices. Distance from home to work is another key factor that influences commuters decisions

PAGE 77

66 66 about CWW. Additionally, employees decisi ons to participate in CWW programs are also affected by their job titles and their em ployers major business types. Overall, those commuters whose jobs must be performed at their worksites are more likely to choose alternative work schedules when the option is available. To further examine the technical feasibility of the model, an ordered logit model is estimated based on the sub-sample of the employees with full options of work schedules and the results are compared with th at of the MNL model. Overall, the results from the ordered logit model are consistent with the major findings from the MNL model. The commuters work schedule choices affect both the frequency of their homebased work trips, but also the time at which the travels occur. The employees choice of working on a compressed work week schedule not only helps reduce th e number of work trips, but also helps shift the work trips from peak periods to non-peak periods. For example, an employee working four days a week will reduce his or her travel by two work trips per two weeks. Further, because of the expansion of daily work hours from eight hours to ten hours, the employee will ha ve to leave home earlier and leave the office later, thus shifting his or her work tr ips from peak periods to non-peak periods. If enough employees choose to participate in compressed work weeks, peak period congestion may be alleviated. The MNL model can easily be applied to evaluate the impacts of existing TDM programs. For metropolitan areas where a comprehensive commute trip reduction program is implemented but no detailed in formation on employee travel behavior is available, the MNL model can be applied to estimate the number of CTR affected

PAGE 78

67 67 employees working on compressed work schedules when employers and employees information on basic variables such as job title are readily available. Furthermore, the MNL model may be incorporated into the regional transportation model to reflect the TDM imp acts in the transporta tion planning process. For the area affected by the Washington Stat e CTR program, the model can be directly used to predict the percentage of employees working on compressed work week schedules at the TAZ level for CTR affected employees. For other areas where detailed employer and employee data are not availa ble, the model developed here may be simplified to use aggregate data at the TAZ level to predict the participation rate of compressed work week schedules. For example, I can use the average commute distance, the percentage of employers affected by TDM strategies, and the percentage of employees by job title and business type to estimate CWW participation rate. The projected percentage of employees working compressed work weeks then may be applied to adjust the number of home based work tr ips to reflect its impacts on the transportation system and, at the same time, to improve the accuracy of the regional planning model. For the promotion of TDM programs, the estimates of the determinants of the CWW choices have important applications TDM promotion agencies or ETCs should consider the job characteristic s of employees and major busin ess type of employers to identify suitable TDM strategies. Although CWW is not costly to implement, for certain industries, they may not work well.

PAGE 79

68 68 CHAPTER 4 MODELING OF TELECOMMUTING CHOICES 4.1. Introduction Telecommuting is designed to allow commuters to use telecommunication technology to work at home or at a location close to hom e during regular work hours, rather than commuting to a conventional works ite, thus saving their driving time to work, and more importantly, eliminating some vehi cle trips, which may help reduce congestion. The concept of the electric homeworker firs t appeared in automation literature in 1957, but did not receive public atte ntion until the oil crisis of the 1970s [Mahmassani et al., 1993]. As a feasible policy tool, telecommu ting opportunities were first available to commuters in southern California in 1988 [Mokhtarian, 1991]. With the rapid development and widespread application of information technology, telecommuting options are available to ma ny regular employees and are on the menu of TDM programs that more and more employers may choose to implement. 4.1.1 Previous Researches Researchers interest on telecommuting ha s been continuous and growing since its first implementation as a part of public pol icy to address transportation congestion in 1988. Most early research focused on the impact of telecommuting on household travel behavior. Many hypotheses have been fo rmulated and tested [Mokhtarian, 1991; Pendyala et al., 1991]. Since most journey to work trips are made during the peak hour periods, telecommuters can reduce their work trips. This may lead to more flexibility

PAGE 80

69 69 when it comes to time budgeting and activ ity scheduling. If the assumption that commuters have a fixed budget of time for travel is correct, then saved time and money for telecommuters may lead to some undesira ble effects, such as more home-centered trips or non-work trips. Another important im pact of telecommuting is related to journey to work modal choices. For telecommuters, rem oval of some job related trips also lowers their probability of carpool ing, vanpooling, or other alternative mode since telecommuters do not need to commute daily [Pendyala et al., 1991]. Although the impact of tel ecommuting remains an unsolved issue because of conflicting findings, it seems that most re searchers agree that, on net, telecommuting reduces total trips, especially peak-period trips, and generates a positive effect on the environment [Hamer, 1991; Sampath et al ., 1991; Quaid and Lagerberg, 1992]. This conclusion is supported by most recent evid ence obtained by Choo et al. [2005]. They find that telecommuting reduces annual VMT by 0.8 percent or less. Their finding is based on a multivariate time series analysis of aggregate nationwide data spanning 1966 1999 for all variables except telecommu ting and 1988-1998 for telecommuting. They conclude that although its impact is small, telecommuting appears to be far more costeffective than public transit in term s of public sector expenditures. Since the effectiveness of telecommuti ng as a strategy to reduce traffic congestion, energy consumption, an d air pollution depends largel y on the extent to which it is adopted by firms and accepted by employees it is important to address the demand side of telecommuting. Mokhtarian and Salo mon [1994] were the first to develop a conceptual model of individual choice in telecommuting. They illustrated the relationships between constraints, preferen ces, and choices faced by individuals and

PAGE 81

70 70 argued that individuals would choose to te lecommute only if the constraints are not binding. Most other studies are empirical analyses based on either the stated preference approach [Bernardino et al., 1993; Mahmassa ni et al., 1993; Mokhtarian and Salomon, 1995] or the revealed preference appr oach [Mannering and Mokhtarian, 1995; Mokhtarian and Salomon, 1997; Drucker a nd Khattak, 2000; Popur i and Bhat, 2003]. Findings based on the stated preference approach seem to be inconsistent. Mokhtarian and Salomon [1995] find that at titudinal factors are more important determinants than social-economic and de mographic characteristics for telecommuting choices. However, the findings from Bernar dino et al. [1993] suggest that attitudinal factors are not determinants of telecommuting choices. Their explanation is that since the employer decides to offer th e option of telecommuting, empl oyers are likely to make a telecommuting program more or less attractiv e based on their own interests. In addition to those arguments, Mahmassani et al. [ 1993] identify more factors that influence peoples telecommuting decision including in formation input from employer, job redesign, fair evaluation of job perfor mance, and promotion opportunities. The studies based on stated preferences pr ovide useful insight s into the factors affecting telecommuting choice, but give n the wide gap between preferring to telecommute and actually telecommuting, a better understanding of the telecommuting adoption decision would only be possible by analyzing the data from revealed preference surveys. As discussed in M okhtarian and Salomon [1995], while 88 percent of the total 628 respondents preferred to telecommute, only 13 percent actually did. Findings from studies based on the revealed preference appr oach, however, are not consistent either.

PAGE 82

71 71 Based on revealed preference survey data from three public agencies in California, Mannering and M okhtarian [1995] explore the individual's choice of telecommuting frequency as a function of de mographic, travel, work, and attitudinal factors through a multinomial logit model. They find that commuters are more likely to telecommute if they have a la rger household size, small child ren at home, more vehicles in the household, a higher degree of family devotion, preference for working alone, familiarity with other telecommuters, or are male. Job related characteristics, such as distance and travel time to work, the amount of work time spent in face-to-face contact, and occupation type, however, are insignificant in determining the telecommuting frequency. Mokhtarian and Sa lomon [1997], find that commuters awareness of telecommuting opportunities, management suppor t, job suitability, technology, and other job related drives play important roles in commuters choices of telecommuting. Based on a revealed preference survey collected in the New York metropolitan region, Popuri and Bhat [2003] apply a joint discrete choice model to estimate the home-based telecommuting choice and weekly home-base d telecommuting frequency simultaneously. They find that individual demographics work-related attri butes, and household demographics are all signifi cant determinants of telecommuting adoption and frequency. 4.1.2 Contribution of This Study One of the common issues faced by most empirical studies on telecommuting is the data availability. Most previous studies are based on small samples and do not have a clear definition of the telecommuters or their actual telecommuting frequency. For example, in most studies appl ying the discrete choice model, the choice set are defined as frequently, infrequent, and ra rely telecommuting, rather than the actual frequency. The

PAGE 83

72 72 commuters are not distinguished between thos e self-employed, those who do not have an office away from the home, and those who ha ve a fixed office but telecommute regularly. To strengthen the findings on telecommuting choices, th is chapter develops an ordered logit model to estimate telecommuti ng choices based on a unique dataset with more than 90,000 observations. The employees choices of telecommuting are made from a set of mutually exclusive and collectivel y exhaustive alternatives, including not telecommuting, telecommuting one day, two days, and three or more days per two weeks. To model the telecommuting choice and its fr equency through a discrete choice model, the dependent variable, therefore, is an ordinal discrete choice. Although multinomial logit and probit models have been widely used in discrete choice modeling and in several earlier studies on telecommuting choices, they may not be appropriate because they fail to account for the ordinal nature of outco mes [Greene, 2000]. For ordinal dependent variables, ordered logit or probit regression is more appropriate. The data was collected from the Wash ington State Commute Trip Reduction (CTR) program. In 2005, this dataset had more than 200,000 observations that have detailed information on employers characteri stics and employees tr avel patterns. The dataset includes only those employees who work at a worksite with at least 100 full time employees with regular working schedules starting between 6:00 a.m. and 9:00 a.m. (inclusive) on two or more weekdays for at least twelve continuous months. This means the sample excludes the self-employed and ot her types of employees who do not have an office away from home. Furthermore, in this sample, the telecommuters are defined as those who regularly telecommute one or more days per two weeks. In other words, the employees who randomly or casually telecomm ute are not counted as telecommuters.

PAGE 84

73 73 This probably explains why the telecommuti ng rate reported by the WA CTR data is dramatically lower than that reported by othe r studies. For example, Drucker and Khattak [2000] reported a total telecommuting rate of 14.3 percent from the 1995 National Personal Transportation Survey data, while the telecommuting rate based on WA CTR database was only 1.51 percent in 1995. In a nother study conducted by Popuria and Bhat [2003] based on 1997-1998 Regional Transportati on Household Interview Survey in New York, the total telecommuting rate was 15.4 per cent, compared with the results from WA CTR data in 1997 of 2.21 percent. I believe this strict defini tion may help generate more reliable results. Finally, this study focuses on examining th e effectiveness of telecommuting as a component of an integrated TDM program a nd predicting the telecommuting rate in the future. The empirical evidence may be applied to evaluate or predict the effectiveness of a TDM program. It may also be incorporated into local or regional travel demand forecasting models to better measure the overa ll performance of tr ansportation systems. The findings from this chapter may also help policy makers when they consider implementing alternative combinations of TDM strategies. The rest of this chapter is organized as follows: section 2 analyzes the telecommuting choices trend. Section 3 presents a brief discussion of the determinants of the telecommuting choice, followed by th e introduction of mode ling methodology, the model specification and results discussion in section 4. Section 5 provides conclusion.

PAGE 85

74 74 4.2 Telecommuting Choices Trend Analysis This section reports results from the data analysis of Washington State CTR data on telecommuting choices from 1993 to 2005. The results are based on valid observations in each year. Table 4.1 Telecommuting Rate by Telecommuting Days per Two Weeks from 1993 to 2005 Percent of Employees by Telecommuting Days Per Two Weeks (%) Year Num of Employees 1 2 3 4 5 6 7 8 9 10 Total 1993 186467 0.36 0.270.090.080.060.020.010.02 0.01 0.050.97 1995 202965 0.66 0.450.100.100.060.030.010.03 0.01 0.051.51 1997 253653 0.81 0.770.150.160.090.050.020.04 0.02 0.112.21 1999 234343 1.16 0.920.290.300.210.100.050.12 0.06 0.283.49 2001 239969 1.13 1.000.250.280.140.090.040.07 0.02 0.223.24 2003 239882 1.46 1.490.370.450.200.140.050.11 0.04 0.254.57 2005 260992 1.68 1.800.490.600.270.220.100.19 0.09 0.385.83 Table 4.1 presents an overall picture of telecommuting choices made by CTR law affected employees. Although it is clear th at the overall par ticipation rate for telecommuting is still pretty low (5.83 percent in 2005), te lecommuting has been gaining popularity consistently. In 1993, two years after the CTR law was passed, less than 1 percent of employees affected by this law c hose to telecommute regularly, while in 2005, 5.83 percent of employees made the telecommut ing choices, an increase of more than 500 percent. Table 4.2 reports the participation rate for telecommuting by job title from 1993 to 2005. It is clear that the telecommuting ra tes vary dramatically for employees with different job titles. The employees working as sales/marketing have the highest telecommuting rate (10.57 percent in 2005), followed by professi onal/technical and management (8.72 percent and 6.88 percent in 2005). Employees working as

PAGE 86

75 75 administrative support, craft/production/labor, and customer service have telecommuting rates below 2 percent. They all experienced consistent growth of telecommuting choices from 1993 to 1995. Table 4.2 Telecommuting Rate by Job Title from 1993 to 2005 Avg. Num of Employees Percent of Employees on Telecommuting (%) Job Title N % 1993 1995 1997 1999 2001 2003 2005 Administrative support 35837 15.98 0.24 0.40 0.81 1.77 1.28 1.39 1.80 Craft/Production/Labor 31110 13.87 0.14 0.21 0.27 1.03 0.35 0.50 0.60 Management 30857 13.76 1.33 2.33 3.33 4.74 4.60 5.81 6.88 Sales/Marketing 7966 3.55 3.31 2.73 4.80 5.38 7.08 8.93 10.57 Customer Service 14791 6.60 0.19 0.49 0.69 2.12 1.27 1.51 1.53 Professional/ Technical 84808 37.82 1.49 2.41 3.41 4.70 4.70 6.89 8.72 Other 18855 8.41 1.29 1.74 2.36 2.91 2.31 3.03 3.95 Table 4.3 presents the participation rate for telecommuting by employers primary business type from 1993 to 2005. For those working for information service/software/technical, the telecommuti ng rate is 14.82 percen t in 2005, more than double the average rate (5.83 percent in 2005). Th is is highly expected since they not only have the technology needed for telecommuting but they also have a lower requirement of worksite presence and personal interaction based on their job characteristics. Manufacturing is noteworthy for its unexpectedly high te lecommuting rate and its highest growth rate. This may be explaine d by the fact that industry evolution and globalization have changed th e definition and nature of manufacturing. For one thing, more and more manufacturing jobs that re quire a physical worksite presence are moving overseas. Furthermore, the manufacturing industr y is more and more high-tech related, making it more suitable for telecommuting.

PAGE 87

76 76 Table 4.3 Telecommuting Rate by Pr imary Business Type from 1993 to 2005 Avg. Num of Employees Percent of Employees on Telecommuting (%) Primary Business N % 1993199519971999 2001 20032005 Agriculture, Forestry, Fishing, Mining 9580.450.000.510.500.23 0.33 0.371.72 Finance, Insurance, Real estate 212469.960.821.221.954.19 4.74 4.965.94 Information Services/ Software/Technical 131186.153.604.557.748.72 7.29 14.2514.82 Professional/Personal Services 111675.231.362.783.905.07 5.05 5.127.69 Retail/Trade 91594.290.430.690.962.08 3.36 3.744.93 Manufacturing 3649317.100.350.700.901.94 1.62 4.397.31 Health Care 3015114.130.731.071.782.66 2.44 2.883.23 Public Utility 85183.993.913.295.157.58 6.24 6.417.10 Military 177448.320.160.530.561.39 0.75 0.690.72 Construction 3320.160.332.030.300.68 0.85 2.010.72 Transportation 48472.270.841.471.372.13 2.01 2.392.32 Government 4503821.111.221.932.283.03 2.65 3.203.23 Education 73583.452.173.163.965.04 6.01 7.068.71 Other 72433.391.051.031.672.79 2.90 4.013.99 Table 4.4 reports telecommuting par ticipation rates by job title and telecommuting days per two week for pr ogram year 2005. Table 4.5 presents the telecommuting choices participation rate by employers major business type and telecommuting days per two weeks for program years 2005. Table 4.4 Telecommuting Rate by J ob Title and Telecommuting Days per Two Weeks in 2005 Num of Employees Percent of Employees on Telecommuting (%) Job title N % 1 2 3 4 5 6 7 8 9 10 Total Administrative support 3332013.150.400.520.120.140.08 0.080.04 0.12 0.02 0.281.80 Craft/Production/Labor 2592010.230.080.110.040.050.11 0.030.01 0.01 0.02 0.140.60 Management 3413613.472.772.140.600.510.27 0.150.11 0.10 0.05 0.186.88 Sales/Marketing 90393.573.322.910.971.060.66 0.290.22 0.30 0.09 0.7410.57 Customer Service 213268.420.240.280.110.090.18 0.050.04 0.10 0.03 0.411.53 Professional /Technical 1000044.752.402.790.730.990.36 0.370.16 0.28 0.15 0.488.72 Other 162426.410.921.070.320.470.25 0.160.04 0.21 0.06 0.443.95

PAGE 88

77 77 Table 4.5 Telecommuting Rate by Primary Business Type and Telecommuting Days per Two Weeks in 2005 Num of Employees Percent of Employees on Telecommuting (%) Primary Business N % 1 2 3 4 5 6 7 8 9 10 Total Agriculture, Forestry, Fishing, Mining 11630.490.690.770.260.000.00 0.000.00 0.00 0.00 0.001.72 Finance, Insurance, Real estate 233849.821.641.670.410.470.27 0.190.09 0.27 0.12 0.835.94 Information Service/ Software/Technical 2445010.274.774.261.381.580.60 0.500.23 0.35 0.24 0.9114.82 Professional/ Personal Services 142115.972.602.670.660.610.28 0.180.04 0.13 0.05 0.477.69 Retail/Trade 119845.032.391.330.320.440.10 0.080.07 0.06 0.04 0.124.93 Manufacturing 3813516.011.421.750.650.990.57 0.500.26 0.43 0.25 0.507.31 Health Care 3246813.630.950.970.220.280.13 0.110.05 0.13 0.04 0.363.23 Public Utility 77873.272.322.250.510.720.22 0.310.13 0.19 0.06 0.397.10 Military 95524.010.170.200.030.120.09 0.020.00 0.01 0.02 0.060.72 Construction 5570.230.360.180.000.000.00 0.180.00 0.00 0.00 0.000.72 Transportation 57012.390.951.020.140.070.04 0.020.00 0.00 0.00 0.092.32 Government 5055221.230.941.380.220.320.10 0.080.04 0.07 0.01 0.073.23 Education 94183.962.003.310.881.190.42 0.270.12 0.22 0.04 0.258.71 Other 87633.681.151.270.300.240.19 0.090.08 0.15 0.02 0.503.99 4.3 Determinants of Telecommuting Choices Mokhtarian and Salomon (1994) developed a behavioral model of the individual choice to telecommute. In their paper, they identified the constraints and drives of telecommuting choices. They defi ned constraint as a factor that prevents the choice to telecommute while drive is a factor that motivates commuters to begin telecommuting. Key constraints for telecommuting choices are categorized as relating to awareness of telecommuting options, the orga nization, job, and psychological factors. If employees lack awareness of telecommuting choices or misunderstand their options, they are not likely to telecommute. The organization rela ted constraints mainly involves lack of support from employers and/or managerial disapproval. The job related constraints include job unsuitability, unavailable technol ogy, and/or high cost. The above mentioned three categories are external constraints and may be changed through public policy,

PAGE 89

78 78 company policy, marketing strategies and technology improvement in telecommunication technology. Psychological cons traints are internal factors, thus individual related. Personal interaction need s, household interaction problems, lack of discipline, risk aversion, and perceived co mmute benefits can also prevent commuters from telecommuting. Among the potential factors that may mo tivate commuters to telecommute, they identify the key drives as work related, fam ily related, leisure related, ideology related, and travel related. Work rela ted drives include the desi re to be more productive, independent, and flexible. Family and leisure related drives include the desire to have more time with other family members and have more leisure time by saving time to drive to work. Ideology related drives include cer tain peoples belief that telecommuting can help protect environment by driving less. If a commuter has a long dist ance from home to work, or the work related commute is burdensom e, these two factors bot h work as drives. Given the data availability, the variable s included in this empirical analysis include the majority of constraints and dr ives. I use TDM promotion activities, the allowance of flexible start/end work tim e, and the time the employer transportation coordinator spends on TDM promotion to measure supportiveness from employers, which may capture organization related c onstraints. The number of years the telecommuting have been allowed at the work site may capture the awareness constrain. I include individual employees job titles and work schedules to capture job related constraints. The commute mode choice will be used to capture the travel related drive. The commute distance, whether the worksite is located in downtown area, and the average property value by ZIP code in which the commuter reside can measure the family

PAGE 90

79 79 and leisure related drives. I believe the vari ables of employers major business type and the existence of multiple shifts at the work site can measure the work related drives. 4.4 Modeling the Telecommuting Choices The major objective of this chapter is to examine the effectiveness of telecommuting as a component of an integrat ed TDM program and e xplore the possibility of estimating a model that can be used to predict the telecommuting rate in the future. I estimate the determinants of employees te lecommuting choices usi ng an ordered logit model. In this section, I begin with methodology, followe d by model specification and discussions of major findings. 4.4.1. Methodology The employees choices of telecommuting are made from a set of mutually exclusive and collectively exhaustive a lternatives, including not telecommuting, telecommuting one day, telecommuting two days and telecommuting th ree or more days per two weeks. To model the telecommuting c hoices through a discrete choice model, the dependent variable, therefore, is an ordinal discrete ch oice. For ordinal dependent variables, the appropriate model is orde red logit or probit regression. Although multinomial logit and probit models are widely used in discrete choice modeling, they may not be appropriate because they fail to account for the ordinal nature of outcomes [Greene, 2000]. For the computation simplicity, I use ordered logit in this chapter. In the ordered logit model, the or dinal choice variable, denoted as y, is assumed as the discrete realizations of an underlying, unobserved (o r latent) continuous random variable y*. The latent y* is a linear combination of some predictors, x, plus a disturbance term, which has a logit distribution.

PAGE 91

80 80 x y* (4.1) where is the coefficient vector. The observed choice variable y is assumed to be determined by the latent continuous variable y* as follows: J j y j yj j i,... 2 1 if 1 (4.2) where are unknown thresholds or cutoff points in the di stribution of y* with J and 0. In this study, the dependent variable telecommuting choices are ordered variable with four categories: (1) no telecommuting (j = 1); (2) telecommuting one day per two weeks (j = 2); (3) telecommuting two days per two weeks (j = 3); (4) telecommuting three or more days per two weeks (j = 4). Assume the probability that employee i reports her telecommuting choice of j given a vector of observed influence variables x is ) x j P(y Pi i i/ then ) ( ) ( ) P( ) P( ) ( ) / (1 1 1 1 i j i j i j i i j j i i j j j i i ix x x x x y P x j y P where ) ( is the cumulative probability distribution of To estimate this model use maximu m likelihood estimation (MLE), the loglikelihood is simply: N i J j j i j i j iQ L11 1 , ,) ln( ln where L is the likelihood function, j iQ, is an indicator variab le which equals 1 if yi=j and 0 otherwise, ) (, i j j ix and ) (1 1 i j j ix

PAGE 92

81 81 Based on the assumption of logit distribution of the so-called proportional odds model (POM) is then 4 ... 2 1 ), exp( ) / ( ) / ( j x x j y P x j y Pj (4.3) where x j y P / ( ) is the conditional probabil ity of choosing at most j-1 days per two weeks given a vector of observed influence variables x ) / ( x j y P is the probability of choosing more than j-1 days, is a column vector coefficients. This model assumes that does not depend on j In other words, the slope of log odds ratio are the same across the categories of dependent va riable. This means the separate equations for each category differ only in the intercepts. While the proportional odds model is easy to estimate and straight forward in interpretation, the assumption of parallel sl ope, also called proportional odds assumption, is not necessary realistic. For example, the impact of distance on telecommuting choice may vary by the number of telecommuting days The feasibility of the proportional odds assumption can be tested using the Wald Te sts, which tests the hypothesis that the coefficients in each independent variable ar e constant across categories of the dependent variable. If this assumption does not hold, generalized ordered logit model should be applied by allowing the slope chan ge in response to choices. The generalized ordered logit model can be written as 1 ...J 2, 1 j ) exp( 1 ) exp( ) ( j i j j i j ix x j y P (4.4) From (4.4), it can be shown that the probabilities that y will take on each of the values from 1 to J is given as below

PAGE 93

82 82 ) exp( 1 ) exp( 1 ) ( 1 J 2,..., j ) exp( 1 ) exp( ) exp( 1 ) exp( ) ( ) exp( 1 ) exp( 1 ) 1 (1 1 1 1 1 1 1 1 J i J J i J i j i j j i j j i j j i j i j i j j i j ix x J y P x x x x j y P x x y P (4.5) When J = 2, the generalized ordered logit model is the same as binomial logit model. When J >2, the generalized ordered logit model b ecomes equivalent to a series of binary logistic regressions where categories of dependent variable are combined [Williams 2007, pp. 2]. In this case, for j = 1, the generalized ordered logit model is equivalent to contrast choice 1 with choices 2, 3, and 4. For j = 2, the contrast is between sum of choices 1 and 2 against choices 3 and 4. For j = 3, it is choice 1, 2, and 3 versus choice 4. 4.4.2. Model Specification I apply the ordered logit model to estimate the relationship between telecommuting choices and a gr oup of observed influences to those decisions. The observed influences in the model include th ose variables availabl e from both employer and employee surveys. From the employee survey, except the telecommuting choices, I use four variables, includi ng commute distance from home to work, the employees job title, travel pattern, and work schedule to capture individual di fferences. The commute distance from home to work measures the one-way miles from employees home to her usual work site, including m iles of errands or stops made daily on the way to work. Washington State CTR data report seve n job titles: administrative support, craft/production/labor, management, sales/mark eting, customer service, professional/ technical, and the other. Thus, I created seven dummy variables to reflect individuals job title. The employees journeyto-work mode choice was divi ded into using the single

PAGE 94

83 83 mode of driving alone, transit, shared ride s, and using the mixed modes. The dummy variable of work schedule is defined to measure if the employee works on compressed work week. Seven groups of variables are from the empl oyer survey: total number of full time employees, primary business type, employers TDM program promotion activities, number of hours the Employer Transportation Coor dinator (ETC) spent on CTR program promotion, the existence of multiple shift at the worksite, the availability of flex time policy to allow employees to vary their start and end times and the number of years the telecommuting program has been implemente d since 1995. Nine dummy variables were created to reflect the primary business of the employer: finance service (including real estate and insurance), information service or software, manufacturing, health care, public utility, transportation, govern ment, education, and other. Through a factor analysis among the thir teen employer TDM pr omotion activities, four of them are selected to reflect the employers TDM promotion, including distributing CTR information, conducting transportation ev ents, publishing CTR articles, and sending electronic mail messages about the CTR program. Although the TDM program promotion activities and the time s the Employer Transportation Coordinator (ETC) spent on CTR promotion (ETC hours) are not specifically for telecommuting promotion but for the whole TDM program, it is reasonable to expect that the higher degree of supportiveness from the management for TDM choices, the hi gher participation rate for telecommuting from the employees.

PAGE 95

84 84 Table 4.6 Selected Variable Definitions Variable Definition Distance One way distance in mile co mmute from home to work location Total Employees Total number of full time employees TDM Promotion Activities Distribute Summary of TDM Program: Distribute a summary of your worksites CTR program to employees? =1, Yes =0, Otherwise Conduct Transportation Even ts: Conduct transportati on events/fairs and/or participate in county/state CTR promotions/campaigns? =1, Yes =0, Otherwise Publish CTR Articles: Publish CTR articles in employee newsletters? =1, Yes =0, Otherwise Send CTR info through email: Send ou t the CTR informatio n through email? =1, Yes =0, Otherwise ETC Hours The average number of hours the Empl oyer Transportation Coordinator spent on CTR promotion Average Property Value The average property value by ZIP code in which the commuter reside Shift Does this worksite ha ve multiple shifts? Shift=1, if Yes; Shift=0, otherwise Flex Time Does your organization offer flex time (allow employees to vary their start and end times)? Flex time=1; Flex time =0, otherwise Drive Alone =1, if drives alone to work for the whole week =0, otherwise Transit =1, if takes public transit to work for the whole week =0, otherwise Shared Rides =1, if carpools or vanpools to work for the whole week =0, otherwise Mixed Rides =1, if takes at least two different modes to work for the whole week =0, otherwise Work Schedule =1, if works 5 days per week =0, otherwise The number of telecommuting program years, defined as the number of years the telecommuting program has been implemented, or allowed, since 1995, is used to capture the effect of time on telecommuting choices. The effect of this variable is expected to be positive since it takes time for employees to understand the benefits of telecommuting

PAGE 96

85 85 program and make transitions accordingly. I also expect the time effect is not constant. Therefore, I created 11 dummy variables to reflect the numbe r of years from the initial implementation of telecommuting program rath er than use one cont inuous variable of number of years. If a worksite started th e telecommuting program in 2005, the number of the year is zero, which is the base valu e for this variable and excluded from the regression I also tried to capture the differe nce of employers lo cation by using a dummy variable to reflect whether the worksite is located in the downtown area or not. I use the average property value of the ZIP code in which the employee resides to serve as a proxy to capture employees social economic information. The property value includes the land value and the building value. The data ar e from King County appraisers web site [ King County Department of Assessment, 2005 ]. The detailed definitions of selected variables are presented in Table 4.6 (other variables are selfexplanatory, therefore not reported). In 2005, there are about 200,000 valid observations from employee commute travel behavior survey. There is, however, inconsistency a bout the availability of the telecommuting choices between that reporte d by employers and the choices made by employees. For example, for certain worksite, employer reports that no telecommuting program is available, while certain employees indicate that they regularly telecommute certain days per two weeks. To determine the actual availability of the telecommuting options, I calculated the total number of employees working on telecommuting for each worksite based on the employee survey data. The results of this calculation are then compared with the information provided in the employer survey. If the number of telecommuting employees is zero and the empl oyer reported that te lecommuting is not

PAGE 97

86 86 allowed, I assume that telecommuting is not allowed for all employees working for this employer. Otherwise, it is allowed. Table 4.7 Ordered Logit Model (POM) for Telecommuting Choices Variable Coefficient z-statistics Distance 0.0294***24.2 Total em p lo y ees -0.000025***3.92 Downtown 0.2150***5.21 Distribute Summar y o f CTR Pro g ram 0.4511***4.55 Conduct Trans p ortation Events 0.1868***3.38 P ublish CTR Articles 0.2021***5.53 Send CTR in f o throu g h email 0.2930***4.90 ETC hours 0.0025**2.39 A vera g e p ro p ert y value 1.91e-7**2.11 Shi ft -0.1788***5.22 F lex time 0.1785***2.91 Transi t -0.4988***8.82 Shared rides -0.3686***6.72 M ixed rides -0.2855***8.21 CWW schedule -0.7368***22.40 J ob titleA dministration Su pp or t -0.7987***7.87 J ob titleP roduction/Labo r -1.8804***12.26 J ob titleM ana g emen t 0.5176***6.56 J ob title-Sales/Marketin g 0.8784***9.85 J ob title-Customer Service -0.7583***6.44 J ob titleP ro f essional/Technical 0.7562***10.27 Business t yp eF inance/RealEstate/Insurance -0.1344**2.32 Business t yp eI n f ormation Service/So f tware 0.48249.00 Business t yp eM anu f acturin g 0.4725***9.52 Business t yp eH ealth Care -0.4834***7.52 Business t yp eP ublic Utilit y 0.2476***3.20 Business t yp e-Trans p ortation -0.5261***4.39 Business t yp e-Governmen t -0.4524***8.06 Business t yp e-Education 0.8701***10.48 Tele Year 1 1.0529***4.37 Tele Year 2 0.9400***3.99 Tele Year 3 1.0511***4.49 Tele Year 4 0.8712***3.75 Tele Year 5 0.9992***4.28 Tele Year 6 1.2317***5.36 Tele Year 7 0.2970 1.11 Tele Year 8 0.8654***3.65 Tele Year 9 0.7067***3.07 Tele Year 10 0.7319***3.24 Cutoff Poin t1 4.7481 CutoffPoint2 5.1611 CutoffPoint3 5.7935 N ( P seudo R2 ) 92,321 ( 0.0859 ) Lo g likelihood ( LR chi2 ( 117 ) ) -5503 ( 0.000 ) *2-tail significance at = 0.10. **2-tail significance at = 0.05. ***2-tail significance at = 0.01.

PAGE 98

87 87 Table 4.8 Brant Test of Parallel Odds Assumption Variable Chi-square p>Chi-square ALL 966.07 0.000 Distance 3.89 0.143 Total employees 21.21 0.000 Downtown 1.12 0.572 Distribute Summary of CTR Program 0.52 0.770 Conduct Transportation Events 3.41 0.182 Publish CTR Articles 30.45 0.000 Send CTR info through email 0.96 0.618 ETC hours 1.38 0.502 Average property value 2.03 0.363 Shift 121.60 0.000 Flex time 5.17 0.075 Transit 7.63 0.022 Shared rides 15.82 0.000 Mixed rides 65.73 0.000 CWW schedule 0.95 0.622 Job title-Administration Support 3.13 0.209 Job title-Production/Labor 0.16 0.921 Job title-Management 46.53 0.000 Job title-Sales/Marketing 6.14 0.046 Job title-Customer Service 9.66 0.008 Job title-Professional/Technical 15.81 0.000 Business type-Finance/Re al Estate/Insurance 1.67 0.433 Business type-Information Service/Software 1.40 0.496 Business type-Manufacturing 30.58 0.000 Business type-Health Care 1.95 0.377 Business type-Public Utility 0.94 0.626 Business type-Transportation 8.15 0.017 Business type-Government 23.98 0.000 Business type-Education 8.96 0.011 Tele Year 1 1.83 0.400 Tele Year 2 2.10 0.350 Tele Year 3 3.84 0.147 Tele Year 4 2.45 0.294 Tele Year 5 0.68 0.711 Tele Year 6 3.62 0.163 Tele Year 7 0.68 0.711 Tele Year 8 1.70 0.428 Tele Year 9 1.61 0.447 Tele Year 10 2.96 0.228 Based on the availability of the detaile d property value data, I only include the records for which the employ ees home zip code is with in the King County. After combining the two dataset, the final sample size is 92,321.

PAGE 99

88 88 The model specification is guided by a series of tests. The model is estimated using the econometric software STATA 9.0. I first run a proportional odds model (POM) based on equation (4.3). The re gression results of the POM a ppear in table 4.7. I then apply the Brant test (Wald Tests) in STATA to see whether the common slope assumption is violated. The results of Brant test are reported in table 4.8. It is clear, from the Brant test, that the parallel regression as sumption is violated fo r the overall model and for most of the variables. Therefore, I use ge neralized ordered logit model to estimate the model based on equation (4.4). The model is estimated based on 2005 sample. 4.4.3 Regression Results The generalized ordered log it model regression results ar e reported in Table 4.9. Columns 2 to 4 report the coefficients for the choices of telecommuting one or more days, two or more days, and three or more days per two weeks. The value of the log likelihood function at its maximum is -28835.7. Th e chi-square, a statisti c used to test the null hypothesis that all the parameters are zero, defined as -2(Log-L(0)Log-L( )) is 6361.91 with a degree of freedom of 117, whic h indicates that we can reject the null hypothesis that all the parameters are zero at the level of at least 0.001. Examining the coefficients in the models for the telecommuting choices, it is first observed that the constant term s are all negative, suggesting that the average effect of those unobserved influence variables is in the direction of not telecommuting or telecommuting fewer days. This is fully anti cipated since around 93 percent of employees in the sample chose commute to their worksites when they have the option to telecommute. Additionally, majority of telecommuters only telecommute one or two days per two weeks.

PAGE 100

89 89 Beginning with the effect of journey to wo rk distance, it is not surprising that employees commuting longer distances are more likely to make the transition from not telecommuting to telecommuting and from tele commuting one day to two days and from two days to three or more days per two weeks. The co efficients of distance are statistically significant at th e level of 0.001 or better. The coefficients for the three dummy variables that measure the employees journey-to-work mode choice are all negative and statistically significant at the level of 0.001, which suggests that compared with commuters who using the single mode of driving alone, employees using single mode of transit and shared ride, or using mixed modes are more likely to not telecommute or telecommute fewer days. I realize that employees journey-to-work modal choice an d telecommuting choice may be jointly determined by some unobserved influences. It may also be possible that an employees telecommuting choices may affect his journeyto-work modal choice. If this is the case, then the variables of employees modal ch oice may be endogenous. It is, however, very difficult to find suitable instrumental variable s to correct this potential problem. I feel confident that even these variables may bias the result, the bias is not significant given the model prediction results discussed later. The impact of an employees work sc hedule is significant. People working on compressed work weeks are less likely to work on telecommuting. The employers supportiveness toward th e CTR program, reflected by the three dummy variables representing the employers TDM promotion activities, the number hours the ETC spend on promoting CTR program, and the allowance of flexible start/end work time at the work site, as expected has positive and significant impacts on

PAGE 101

90 90 employees telecommuting choice. The th ree dummy variables of employers TDM promotion activities have positive coefficients and statistically significant at the level of 0.001 or better for all of the telecommuting c hoice categories. The coefficients of flexible time are positive for all choice categories and significant for telecommuting one or more days and two or more days. The numbe r of hours the Employer Transportation Coordinator works is significant for telecommu ting one or more days. The last variable in this group is the number of ETC hours spent on TDM programs. The results suggest that, although the employers TDM promotion ma y not be specifically focused on telecommuting but rather a reflection of the employers supportivene ss to the whole CTR program, the employers positive attitude to the TDM program does have significant impacts on employees telecommuting choice. It also suggests th at the impact of employers CTR supportiveness on employee s telecommuting choice most likely happens on encouraging employee from not te lecommuting to telecommuting rather than from telecommuting less frequently to more frequently. The eleven dummy variables, representing the number of years the telecommuting program has been implementing since 1995, are created to reflect the awareness of the employee on telecommuting program. Except year seven, which has a positive but not statistically significant coefficient, all other year dummy variables have a positive coefficient and statistically significant at the c onfidential level of 95 percent or better for the first two categories. This suggests that the number of program years has a positive impact on employees choice of telecommuting one or two days per two weeks. On the other hand, most coefficients for the last ca tegory, telecommuting th ree or more days, are not statistically significant. As to the valu es of the coefficients of the year dummy

PAGE 102

91 91 variables, as expected, they are not consta nt but follow a similar pattern for the three telecommuting categories. Overall, the coeffici ents are increasing for the first six years and begin to decrease from year seven whil e the coefficients are still positive. This suggests that telecommuting program implemen tation year has an increasing effect on telecommuting choices until it reaches its peak in year 6, after which its marginal effect falls or goes flat. The impact of worksite location is pos itive and significant which means the employees working on downtown area are more likely to telecommute. The impact of employer size, the total nu mber of employees, is ne gative and significant. It is interesting to see that people living in an area with higher property value area are significantly more likely to telecommute This may suggest that telecommuting is more suitable for the high-end job. Among the six job titles included in the regression, the coefficients of administrative support, production/labor, a nd customer services are negative and statistically significant at th e level of at least 0.001 for all categories. These results suggest that employees with th e job titles mentioned above ar e less likely to telecommute at all or telecommute fewer days if teleco mmuting. This finding may be explained by their job characteristics. While administ rative staffs provide direct support for management and production workers at factor ies or other facilities produce goods, both have a need to physically work at the worksite. The coefficients of management have positive signs but ar e not statistically significant for the choice of telecommuting thr ee or more days per two weeks. This result suggests that managers are more likely to ma ke the transition from not telecommuting to

PAGE 103

92 92 telecommuting but are less likely to telecommute three or more days, which is reasonable considering their job characteristics. Table 4.9 Generalized Ordered Logi t Model for Telecommuting Choices Variable One Day +a,bTwo-Day +a,b Three-Day+a,b Distance 0.0292 ( 23.87 ) ***0.0305 ( 21.88 ) *** 0.0302 ( 16.78 ) *** Total em p lo y ees -0.00002 ( 3.57 ) ***-0.00004 ( 5.35 ) *** -0.00005 ( 4.89 ) *** Downtown 0.2075 ( 5.02 ) ***0.2437 ( 4.90 ) *** 0.2342 ( 3.28 ) *** Distribute Summar y o f CTR Pro g ram 0.4500 ( 4.54 ) ***0.4788 ( 4.15 ) *** 0.5962 ( 3.80 ) *** Conduct Trans p ortation Events 0.1902 ( 3.44 ) ***0.2213 ( 3.29 ) *** 0.0585 ( 0.64 ) *** P ublish CTR Articles 0.1942 ( 5.30 ) ***0.2797 ( 6.32 ) *** 0.4745 ( 7.77 ) *** Send CTR in f o throu g h email 0.2910 ( 4.86 ) ***0.2676 ( 3.73 ) *** 0.2152 ( 2.15 ) *** ETC hours 0.0021 ( 2.03 ) **0.0018 ( 1.53 ) 0.0021 ( 1.26 ) A vera g e p ro p ert y value 2.07e-7 ( 2.34 ) **7.35e-8 ( 0.67 ) 2.77e-7 ( 0.21 ) Shi ft -0.2151 ( 6.28 ) ***-0.4394 ( 0.28 ) -0.1005 ( 1.85 ) F lex time 0.1800 ( 2.93 ) ***0.1602 ( 2.13 ) ** 0.0229 ( 0.23 ) Transi t -0.4888 ( 8.63 ) ***-0.5674 ( 8.17 ) *** -0.6693 ( 6.70 ) *** Shared rides -0.3521 ( 6.41 ) ***-0.4416 ( 6.77 ) *** -0.6261 ( 6.72 ) *** M ixed rides -0.2617 ( 7.50 ) ***-0.3851 ( 9.17 ) *** -0.5575 ( 9.22 ) *** CWW schedule -0.7288 ( 22.04 ) ***-0.7279 ( 19.71 ) *** -0.7793 ( 17.23 ) *** J ob titleA dministration Su pp or t -0.7959 ( 7.84 ) ***-0.8506 ( 7.27 ) *** -0.7842 ( 5.47 ) *** J ob titleP roduction/Labo r -1.8421 ( 12.01 ) ***-1.9084 ( 11.44 ) *** -1.8998 ( 9.77 ) *** J ob titleM ana g emen t 0.5403 ( 6.84 ) ***0.2852 ( 3.10 ) *** 0.0277 ( 0.24 ) J ob title-Sales/Marketin g 0.8859 ( 9.92 ) ***0.7448 ( 7.18 ) *** 0.6348 ( 4.93 ) *** J ob title-Customer Service -0.7587 ( 6.44 ) ***-0.6851 ( 5.32 ) *** -0.4241 ( 2.77 ) *** J ob titleP ro f essional/Technical 0.7590 ( 10.30 ) ***0.6837 ( 8.13 ) *** 0.4564 ( 4.33 ) *** Business t yp eF inance/Real -0.1373 ( 2.37 ) **-0.1418 ( 2.00 ) ** -0.0295 ( 0.30 ) Business t yp eI n f ormation 0.4934 ( 9.18 ) ***0.4707 ( 7.37 ) *** 0.5002 ( 5.77 ) *** Business t yp eM anu f acturin g 0.4483 ( 9.00 ) ***0.5208 ( 8.72 ) *** 0.8096 ( 10.16 ) *** Business t yp eH ealth Care -0.4723 ( 7.34 ) ***-0.5600 ( 7.06 ) *** -0.6157 ( 5.49 ) *** Business t yp eP ublic Utilit y 0.2456 ( 3.17 ) ***0.2432 ( 2.58 ) *** 0.3379 ( 2.59 ) *** Business t yp e-Trans p ortation -0.5208 ( 4.35 ) ***-0.4560 ( 3.11 ) *** -1.0614 ( 3.74 ) *** Business t yp e-Governmen t -0.4419 ( 7.86 ) ***-0.3568 ( 5.41 ) *** -0.6130 ( 6.02 ) *** Business t yp e-Education 0.8723 ( 10.46 ) ***0.8959 ( 9.46 ) *** 0.6930 ( 5.28 ) *** Tele Year 1 1.0475 ( 4.34 ) ***1.3732 ( 4.05 ) *** 0.9070 ( 2.06 ) *** Tele Year 2 0.9530 ( 4.04 ) ***1.1797 ( 3.56 ) *** 0.6440 ( 1.50 ) *** Tele Year 3 0.9895 ( 4.22 ) ***1.4584 ( 4.43 ) *** 1.2423 ( 2.93 ) *** Tele Year 4 0.8501 ( 3.66 ) ***1.2714 ( 3.88 ) *** 1.0401 ( 3.47 ) *** Tele Year 5 1.005 ( 4.31 ) ***1.1541 ( 3.48 ) *** 0.6601 ( 1.54 ) Tele Year 6 1.1884 ( 5.71 ) ***1.6267 ( 5.01 ) *** 1.1663 ( 2.80 ) *** Tele Year 7 0.2883 ( 1.07 ) 0.5720 ( 1.56 ) 0.2101 ( 0.45 ) Tele Year 8 0.8459 ( 3.56 ) ***1.2656 ( 3.79 ) *** 1.0058 ( 2.34 ) ** Tele Year 9 0.7077 ( 3.08 ) ***1.0173 ( 3.12 ) *** 0.6440 ( 1.54 ) Tele Year 10 0.7283 ( 3.22 ) ***1.1112 ( 3.45 ) *** 0.6173 ( 1.49 ) Constan t -4.73 ( 18.00 ) ***-5.5337 ( 15.37 ) -5.4984 ( 11.78 ) N ( P seudo R2 ) 92,321 ( 0.0994 ) Lo g likelihood ( LR chi2 ( 117 ) ) -28835.725 ( 0.000 ) a absolute value of z-statistics in parentheses. b *2-tail significance at = 0.10. **2-tail significance at = 0.05. ***2-tail significance at = 0.01.

PAGE 104

93 93 Among the eight employers major business types, the coefficients of information service/software, manufacturing, public utility, and education are positive and statistically significant at the level of 0.01 or better. The coefficients on transportation, health care, and government are negative and statistically significant. This finding suggests that employers major business type can be consid ered either as a drive or a constraint affecting commuters telecommuting choices. As a final check, I estimated the model ba sed on randomly selected 80 percent of the total sample and used the other 20 percent to test the models predictability. The model is also tested using 2003 data. Th e results are reported in Table 4.10. From Table 4.10, it is clear that, overall, the prediction is very close to the survey result. The model predicts th at 7.461 percent of commuter s choose to telecommute at least one day per two weeks in 2005, while the survey resu lt is 7.102 percent. When using the 2003 data, the model over-predicts the percentage of telecommuters. This is fully anticipated since the telecommuting programs have been more and more acceptable to both employers and employees over time, a trend illustrated in Table 4.1. Table 4.10 Comparison of the Model Predictions and Survey Results Average Percentage of Employees on Telecommuting (%) Program Year 2005 Program Year 2003 Days/2 Weeks Model Survey Model Survey 0 Day 92.539 92.898 92.814 94.313 1 Day 2.315 2.075 2.217 1.936 2 Days 2.149 2.195 2.079 1.966 3 + Days 2.997 2.832 2.890 1.785 Total Telecommuting 7.461 7.102 7.186 5.687

PAGE 105

94 944.5 Conclusion This chapter analyzes the participation trend for telecommuting and applies a generalized ordered logit model to estimate th e impacts of journey to work distances and mode choice, employers supportiveness towa rds the CTR program and the number of years telecommuting has been a llowed, employees job characteristics, work schedule, and average household property value, and em ployers major business types and worksite location on employees telecommuting choice. The data analysis indicates that alt hough, overall, the participation rate of regularly telecommuting one or more da ys per two weeks for the CTR affected employees is still pretty low (5.83 per cent in 2005), telecommuti ng has been gaining popularity consistently. During the period fr om 1993 to 2005, among those affected by the CTR law, the percentage of employees choosing to regularly telecommute at least one day every two weeks increased more than 5 times. I also find that the telecommuting rates vary dramatically for the employers with different primary business types and for the employees with different job titles. I apply a generalized ordered log it model to estimate the employees telecommuting choices. Telecommuting is categorized into not telecommuting, telecommuting one day, two days, and three or more days per two weeks. I find that commuters with a longer distance from home to work are more likely to make transition from not telecommuting to telecommuting and telecommuting more days if already choosing to do so. The people us ing the single mode of driving alone are more likely to telecommute compared with those using the si ngle mode of transit or shared rides, as well as those using mixed modes.

PAGE 106

95 95 The employers supportiveness toward th e CTR program, reflected by the three dummy variables representing the employers TDM promotion activities, the number hours the ETC spend on promoting CTR program, and the allowance of flexible start/end work time at the work site, as expected has positive and significant impacts on employees telecommuting choice. The employees awareness of the telecommuting program, represented by the number of year the telecommuting has been allowed, has positive but not constant impacts on the employees adoption of telecommuting. Job characteristics, including job title and work schedule, serve as either drive or constraint for employees telecommuting choice. The employees telecommuting choices ar e also affected by their employers other business characteristics, such as worksite location, total number of fulltime employees, the existence of multiple shifts at the worksite, and employers major business type. The model is evaluated by estimating the model using randomly selected 80 percent of 2005 data and comparing the models predictions with the survey results using the 20 percent of excluded sample. The model is also tested using 2003 data. For the 2005 data, the model prediction of the overall te lecommuting rate is 6.344, very close to the survey result of 6.377. The differences be tween the model predictions and the survey results for all three categories of telecommu ting are less than 2 pe rcent. For the 2003 data, the model over-predicts th e telecommuting rate (5.650 pe rcent versus 4.70 percent). Since the telecommuting rates of the CTR aff ected employees changed significantly from 2003 to 2005 (4.57 percent versus 5.83 pe rcent), the models over prediction of

PAGE 107

96 96 telecommuting is fully expected and suggests th at commuters preference is changing in favor of telecommuting over time, a factor that cannot be captured by the model. As elaborated in chapter 3, the results of this chapter can be used to evaluate the impacts of a telecommuting program, a com ponent of an integrated TDM program, and to identify the effectiveness of the TDM strategies. More importantly, they may be incorporated into the regional transportation model to reflection the impacts of TDM on transportation planning process and, at the same time, to improve the accuracy of the regional planning model.

PAGE 108

97 97CHAPTER 5 AN INTEGRATED MODEL OF TDM IMPACTS ON JOURNEY TO WORK MODE CHOICES 5.1 Introduction One of the major objectives of the employer-based Commuter Trip Reduction (CTR) program is to reduce vehicle trip s by implementing programs that encourage alternatives to drive-alone commuting to worksites [Washington State DOT, 2007]. Therefore, the impacts of the implem ented TDM programs on a commuters modal choice could be an important measure of TDM effectiveness. Although the commuters travel behavior in terms of trav el mode choice has been studied extensively, there is no empirical work that estimates the combined effects of a TDM program on an individuals modal choice. Most previous research of TDM impacts was worksite-based, retrospective, focusing on one or more as pects of TDM strategies, and based on small samples. 5.1.1 TDM Strategies Evaluation Literature Review In a case study conducted by Mehranian et al. [1987], two downtown companies are compared to clarify the effect of parki ng cost on journey-to-work modal choices. The two companies are located at th e same site, and their employ ees have access to the same parking facilities. The major differences ar e the employers policy on subsidization of parking cost. One company provided a partial parking subsidy to about one-third of its employees and no financial assistance to ca rpoolers, vanpoolers, transit users. The

PAGE 109

98 98 other company had a more complex system of subsidies to its employees, providing varying levels of support for solo drivers, carpoolers, vanpoolers, and transit riders. Although the second company sp ent far more money on promotion of ridesharing, the two companies have almost the same percentage of drive-alone. They find that the second companys complex subsidies to different modes shifted transit use to vanpooling and carpooling. Although the second company spent much more money on the promotion of ridesharing, its majority of commuter subsidie s are used to subsidiz e the parking costs of solo drivers, which counters the effectiven ess of its original effort of promoting ridesharing and transit use. They conclude th at, for the employers that already subsidize the parking of solo drivers, it is more cost-e ffective to promote ridesharing and transit use by eliminating parking subsidies to solo drivers than it is to offer additional subsidies to other alternative modes. Brownstone and Golob [1991] investig ate the effect of certain incentives implemented to increase journey to work ri desharing using the greater Los Angeles area data based on an ordered probit discrete choi ce model. They find that female full timers and those employees who have larger house hold sizes with multiple workers, longer commutes, and larger worksite are more likely to rideshare. Their simulation model suggests that policy tools su ch as providing all carpool and vanpool with reserved parking, ridesharing subsidies, guaranteed rides home, and high-o ccupancy vehicle lanes would reduce drive-alone commuting between 11 and 18 percent. Peng et al. [1996] investigate the effect of parking prices on urban journey-towork modal choices using travel activity data from Portland, Oregon. The results from their nested multinomial logit model suggest that parking prices have a significant

PAGE 110

99 99 influence on commuters' mode choices. They find that parking prices have divergent impacts on commuters using different modes a nd/or with different residential locations. Compared with central city transit users, s uburban transit users are more sensitive to parking price changes. Vanpoolers and carpooler s are less responsive to parking prices than solo drivers. For suburba n residents, those driving alone and ridesharing to work are less responsive to parking prices than are central city reside nts. Employment location also plays an important role in mode choice. Em ployees working in suburban areas are more likely to drive and less likely to use transit. While increased transit service alone has a fairly small effect on transit use, combin ed efforts of increasing parking price and improving transit service simultaneously provi des an effective means of reducing solo driving and increasing transit use. Several researchers look at the eff ect of land-use policies on modal choices. Cervero [1996] explores how mixed land-uses affect the commuting choices in large urban areas based on data from the 1985 Amer ican Housing Survey. The effects of landuse environments on mode choice are modeled using binomial logit analysis. It appears that mixed land-use policies may help to pr ovide alternatives to driving, although the effect is likely to be small. Bento et al. [ 2003] also look at the effect of urban form on journey to work mode choices using the 1990 na tional personal trans portation survey data and find most urban spatial characteristic s have no significant effect on commuters choice individually. Kuppam et al. [1999] carry out an anal ysis using the 1991 wave of the Puget Sound Transportation Panel data set to investig ate the effects of attitudinal and preference variables on commuters m ode choices. They find that demographic variables and

PAGE 111

100 100 attitudinal variables are extremely important in explaining mode-choi ce behavior, but the latter have more explanatory power. In a discrete choice experiment of road pricing and parking charges conducted in Greater Vancouver suburb areas based on a sa mple of 548 commuters who drove alone to work at the time of experiment, Washbrook et al. [2006] find that increases in drive-alone costs will bring about greater reductions in single-occupancy-vehicle (SOV) demand than increases in SOV travel time or improvements in the times and costs of alternatives beyond a base level of service. This study designs a customi zed discrete choice experiment and asks the partic ipants to choose between driv e-alone, carpools, or take a hypothetical express bus service. Attributes coefficients based on the experiments are then used in a predictive model to estim ate commuters responses to various policyoriented combinations of charges and incenti ves. The authors believe this is a costeffective way for policymakers to evaluate choices to lower SOV. 5.1.2 Limitations of the Worksite Based Analysis Method As stated in Chapter 2, most of the TDM models, including EPAs COMMUTER model, CUTRs worksite reduction model, and Washington States TEEM model, are worksite based. The worksite-based approach estimates changes in mode split at an aggregate, worksite level by treating the work site as the analysis unit. Although most of commute trip reduction programs are employer-bas ed, using worksite as the analysis unit to evaluate the effectiveness of the TDM strategies has limitations. Firstly, calculation of the aggregate mode split is highly affected by some factors that are hard to control or measure, for example the survey response rate. The nonrespondents are generally treated as having the same distribution of mode shares as that

PAGE 112

101 101 of valid respondents. There are other argumen ts, however, that people driving alone are less likely to answer the questionnaire. Based on this assumption, some studies treat the non-respondents as driving alone, or treating the non-respondents as driving alone when the response rate is less than certain amount e.g. 70 percent. Since the impact of the TDM on the worksites mode split is relativel y low, the bias induced by the calculation could be significant. Secondly, some of the important determinants of mode choice, such as travel time and travel cost, can only take average value at the worksite level, while those variables are meaningful only from the perspective of individuals. The worksite-based approach also fails to catch varieties of the individual trips, which is critical when the study focuses on quantifying the impact of reduced individua l trips. In addition, the worksite based approach reduces the number of observations (worksites) av ailable from which to make the estimates. This is especially important when the study area is a sub-area, such as downtown or corridor area. 5.1.3 Modeling the Impacts of an Integr ated TDM Program on Mode Choice An employer-based TDM program generally includes different strategies. For most of the strategies, their impacts are more interactive th an independent. For example, an internal or external ride match program will be more effective if combined with reserved high occupancy vehicle (HOV) pa rking space or HOV parking charge discount. Focusing on only one aspect of TDM strategies without control ling of the availability of other TDM programs may result in omitted variable bias. Among the various methodologies applied in human behavior study, the discrete choice model has been widely used in the transportation community to study the travel-

PAGE 113

102 102 related human behavior, specifically, the tr avelers mode choice and departure time choice. The Washington State CTR datase t provides detailed information on TDM strategies and the corres ponding employee commute travel behavior for hundreds of employers and tens of thousands of employees This makes it possible, for the first time, to perform a systematic discrete choice anal ysis of integrated TDM impacts on individual employees mode choices. In this chapter, a nested logit model is applied to estimate the determinants of modal choices for the CTR affected employees and evaluate the impacts of various TDM strategies on commuters modal choices based on a large sample of about 60,000 observations. The model is a two-level nest ed logit model. The first ne st includes motor, transit, non-motor. In the second nest, motor is divi ded into drive-alone a nd shared riding. The mode shares of each of the alternative are: motor, 74.18 percent (drive-alone, 60.61 percent; shared ride, 13.57 per cent); transit, 22.29 percent; and non-motor, 3.52 percent. Based on the nested logit model, the elas ticity and marginal effects of finance incentives and TDM support and promotion prog rams are further calculated to evaluate the quantitative impacts of various TDM strategies on the modal choices. Commuter mode choice has b een studied extensively. Generally, the factors that have been examined and proven to be rele vant to commute mode choice include (1) commuters sociodemographic characteristics, such as age, gender, income, household composition and car ownership, and so on; (2 ) connection information between the origin and destination, such as travel time and tr avel cost by modes, a nd so on; (3) land-use characteristics of the origin and destina tion, such as locati on, population density,

PAGE 114

103 103 accessibility to transit services and parking facili ty, availability of sidewalk and bike lane, and so on; and (4) other fact ors, such as travelers subjec tive perceptions of and feelings toward modes as well as their lifesty le. In this chapter, the data obtained from the employer annual report includes the characteristics of employer, the worksite la nd-use, and the TDM program implemented at the worksite. The data from the employee tr avel behavior survey include commuters mode choice, job title, commuting distance, and work schedule. The detailed travel information for both transit and auto be tween the commuters zip code and the commuters worksite within King Count y are extracted from Google transit (www.google.com/transit ) through a computer program. Th e average property value of the commuters home zip code, used as a pr oxy for the commuters household income, is obtained from King Count y appraisers website (http://www.metrokc.gov/asse ssor/download/download.asp ). Since the transit connection information is only available for King C ounty Metro, this study will focus on the worksites located within King County and employees residing King County, including the Seattle urban area and other suburban and rural areas. Since more than 60 percent of the employers affected by the Washington State CTR program are located in King County, the utilization of this sub-sample will not affect the accuracy of the model. The result of this part of study will not only provide a comprehensive, reliable quantitative and qualitative assessment of the impacts of TMD strategies on the affected commuters mode choice, but wi ll also explore the framew ork of a mode choice model that includes the TDM components. This mode choice model may further be incorporated

PAGE 115

104 104 into the regional transportation model to re flect the impact of the TDM on the regional transportation planning process. 5.2 Methodology 5.2.1 Nested Logit Model Utility-based choice or choice based on the relative attractiveness of competing alternatives from a set of mutually exclus ive alternatives is cal led a discrete choice situation. Discrete choice models are interprete d in terms of an underlying behavioral model, the so-called random utility maximization (RUM) model. The decision-maker chooses the alternative with th e highest utility. Characteristi cs of the decision-maker and of the choice alternatives determine the alternatives' utilities. Discrete choice decisions in the context of random utility theory are usually modeled and estimated with the multinomial logit model (MNL), or standard logit model, because of its closed choice pr obabilities and straightforward of interpretation. The MNL, however, assumes independence of irrelevant alternatives (IIA), i.e. the ratio of the choice probabilities of two alternatives is not dependent on the presence or absence of other alternatives in the model. In a standard logit model, for any two alternatives i and k the ratio of the probabilities of individual n choosing i over k is nk ni nk ni nj nk nj niV V V V J j V V J j V V nk nie e e e e e e p p / /1 1

PAGE 116

105 105 where J is the number of alternatives and Vnj is the utility of alternative j for individual n This ratio does not depend on any alternatives other than i and k That is, the relative odds of choosing i over k are the same no matter wh at other alternatives are available or what the attributes of the other al ternatives are. Since the ratio is independent other than alternatives, it is said to be inde pendent of irrelevant al ternatives, or IIA. The IIA property is the direct result of the basic assumption on which the MNL has been established, that is, the error te rm of the utility func tion is independently identically Gumbel-distributed. While the IIA property is realistic in some choice situations, it is clearly inappropriate in others. To overcome this restrictiv e substitution assumption betw een alternatives, various extensions of the MNL exist, all with the general solution of allowing correlations between the alternatives' erro r terms. The idea of the ne sted logit model lies in the grouping of similar alternatives into nests and thus creating a hierarchical structure of the alternatives (Ben-Akiva and Lerman, 1985; Train, 2003). The e rror terms of alternatives within a nest are correlated with each other, a nd the error terms of alternatives in different nests are uncorrelated. The nested logit model, also known as the generalized ex treme value (GEV) model, structured logit, and sequential logit, was first deri ved by Ben-Akiva [1973], as an extension of the multinomial logit model designed to capture correlations among alternatives. The nested logit model has b ecome an important tool for the empirical analysis of discrete outcomes and has been widely applied in transportation modal choice studies (Train, 1980; Bhat, 1997) Its popularity comes from tw o facts: (1) it relaxes the restrictive assumption of the independence of irrelevant alternatives (IIA) of conditional

PAGE 117

106 106 logit model, and (2) it uses a closed-f orm likelihood functi on, which enables straightforward and fast computation. Theref ore it is considered analytically tractable compared to multinomial probit and mixed logit. The nested multinomial logit model is a more generalized multinomial logit model. It allows researchers to specify a stru cture that categorizes the alternatives into groups (nests) by assuming that alternatives in each group are sim ilar in an unobserved way, thus creating a hierarchical structure of the alternatives (Ben-Akiva and Lerman, 1985; Train, 2003). The error terms of alternativ es within a nest are correlated with each other, and the error terms of alternatives in different nests are uncorrelated. For simplicity, assume a two-level nesting structure. Suppose there are J alternatives categorized into K nests: N1, N2 NK. Suppose y = j is the observed choice selected and alternative j is an element of nest Nk, then the probability that y = j for individual n is given by ) | ( ). ( ) (k n k n nN y j y P N y P j y P (5.1) where ) (kN y P is the marginal probability of choosing an alternative in nest NK, and ) | (kN y j y P is the conditional probabil ity of choosing alternative j given that an alternative in nest Nk is chosen. In other words, the probability of alternative j in the Nk nest results from the product of the ma rginal choice probability of nest kN and the conditional choice probab ility for alternative j within nest Nk. Both marginal and conditional choice probabilities are standard logit models. Without loss of generality, the observed co mponent of utility can be decomposed into two parts: (1) a part labeled x that is constant for all alternatives within a nest and (2)

PAGE 118

107 107 a part labeled that varies over alternatives with in a nest. Utility of alternative j for individual n can be written as j nj nk njx U (5.2) for j Nk where xnk depends only on variables that describe nest Nk. These variables differ over nests but not over alternatives within each nest. nj depends on variables that describe alternative j These variables vary over alternatives within nest Nk. The marginal and conditional pr obabilities can be expressed as K l I x I x k nnl l nl nk k nke e N y P1) ( (5.3) k k nj k njN j k ne e N y j y P / /) | ( (5.4) where k k njN j nke I /ln According to (5.1), the probability of y = j for individual n is given by k k nj k nj nl l nl nk k nkN j K l I x I x k n k n ne e e e N y j y P N y P j y P / / 1. ) | ( ). ( ) ( (5.5) where Ink, often called inclusive value of nest NK is the log of the denominator of the conditional probability model. It links the ma rginal probability model and the conditional probability model by bringing information from the lower model into the upper model. The coefficient k on Ink in the marginal model, often called the log-sum coefficient k,

PAGE 119

108 108 reflects the degree of independence among the unobserved portions of utility for alternatives in nest Nk with a lower k indicating less independence (more correlation). The parameters of a nested model can be estimated by standard maximum likelihood techniques. Substituting the choice probabilities of Expression (5.5) into the log-likelihood function gives an explicit function of the parameters of this model. Instead of performing maximum likelihood, nested logit models can be estimated consistently (but not efficientl y) in a sequential fashion, expl oiting the fact that the choice probabilities can be decomposed into marginal and conditional probabilities. This sequential estimation is performed bottom up. The lower models (for the choices of alternative within a nest) are estimated fi rst. Using the estimated coefficients, the inclusive value is calculated for each lower model. Then the upper model (for choice of nest) is estimated, with the inclusive value entering as expl anatory variables. 5.2.2 Elasticities of Logit Model Elasticities measures how the independent variables response to the change in the determining factors. For the discrete choice model, it is the percentage change in the probability of choosing one of th e alternatives due to a 1-per cent change in some attribute that is an independent variable in the model. For a discrete choice model, the coefficients are not directly tied to the elasticities a nd it is necessary to distinguish between disaggregate and aggregate, dir ect and cross elasticities. A disaggregate direct elasticity repr esents the percentage change in an individuals choice probabil ity of choosing alternative i due to a 1-percent change in the value of some attribute that is an independent variable in the utility function of alternative i

PAGE 120

109 109 The formula of a disaggregate direct elasticity is: k ink n ink n n ink ink n i Pn xx i P x i P i P x x i P Eink)] ( 1 [ log ) ( log ) ( ) () ( (5.6) where Pn(i) denotes the possibility of individual n choose alternative i, xink is the k attribute in the utility function of alternative n for the individual i. k is the coefficient of attribute k. A disaggregate cross elasticity repres ents percentage change in an individuals choice probability of choosing alternative i due to a 1-percent cha nge in the value of some attribute that is an independent vari able in the utility function of alternative j The formula of a disaggregate cross elasticity is k jnk n ink n n ink ink n i Pn xx j P x i P i P x x i P Eink) ( log ) ( log ) ( ) () ( (5.7) where j i It is clear that the logit model has unifo rm cross elasticities, that is, the cross elasticities of all alternatives with respect to a change in an attribute affecting only the utility of alternative j are equal for alternatives j i [Ben-Akiva and Lerman, 1985, pp. 111-112]. For a two-level nested logit model, the di rect elasticity, which is defined as the percentage change in individuals ch oice probability of choosing alternative k in branch l

PAGE 121

110 110 due to a 1-percent change in the value of some attribute r, that is an independent variable in the utility function of alternative k in the branch l is computed as: l r r l k P l k r xl k P l P l k P l k l k x E )] ( ) ( 1 [ )] ( 1 [ * ,) ( ) ( (5.8) where l k xr,is the value of the attribute r in the utility function of alternative k in branch l, l kr, is the coefficient of the attribute r P(k|l) is the conditional probability of choosing alternative k given branch l is chosen, P(l) is the probability of choosing branch l, l is the coefficient of Inclusive Value for branch l. The cross elasticity, which is defined as the percent change in an individuals choice probability of choosing alternative k in branch l du1 to 1-percent a change in the value of some attribute r that is a variable in utility f unction of alternative j in branch i is calculated as: ] ) ( 1 [ ) ( * ,) ( ) ( l r r l k P i j r xl P l k P i j i j x E (5.9) where i l i l j l or While the disaggregate marginal eff ects measure the responsiveness of an individuals choice probability to a change in the value of some attribute, in most cases people are more interested in the respons iveness of some group of decision makers. Aggregate elasticities measur e the summarized responsiven ess of a group of decision makers to the changes in the value of some at tribute, rather than that of any individual response. There are different ways, however to summarize the group responsiveness, including averaging the individual sample observations, using the sample means of the data, and calculating a weighted average. For the first method, the aggregate elasticities are simply the average of disaggregate elasti cities. Observations receive equal weigh in

PAGE 122

111 111 the average. One problem that can arise using th is method is that if an observation in the sample has an extreme configuration of attribut es for some reason, then the elasticities for that observation can be extremel y large, which in turn will cause the average to be huge. For the second method, the elasti cities are computed just once at the sample means of the attributes. The weighted average method ca lculates the aggregat ed elasticities as weighted average of the individual level el asticities using the c hoice probabilities as weights, that is, ) ( ) (1 ) ( 1 ) (i P E i P En N n i P x N n n i P xn jnk jk (5.10) where N is the number of decision makers in the group. By this construction, if an individual probability is very small, the resulting extreme value for the elasticities is multiplied by a very small probability weight, which offsets the extreme value [Greene, 2002, pp. N3 -24]. In this disserta tion, the elasticities are calculated using the weighted average method. 5.3 Data and Variable Definition 5.3.1 Mode Share Trend for The CTR Affected Employees Once again, the major data sources for th is part of study come from Washington State CTR dataset, which was described in deta il in Chapter 3. The dependent variable is journey-to-work mode choice. The mode shares from 1993 to 2005 based on CTR affected employee in King County and in all of the nine counties are presented in Table 5.1 and 5.2 respectively. In 1993, 64.79 percent of employees in King County and 74.52 percent of all of CTR affected employees drove alone. This share then declined to 58.27 and 68.64 percent respectively in 2001, af ter which, it rebounded to 63.28 and 70.02

PAGE 123

112 112 percent in 2005. Although the shares of dr ive alone were higher based on the whole sample of CTR affected employees, the tr end was the same. Despite the rebound, the shares of drive alone in King County and for the entire CTR affected employee are lower than the state share of 74.3 percent and nati onal average of 77 percent in 2005. It seems that the share of non-motor re mains relatively stable. The si gnificant changes come from carpool, vanpool, and transit, especially, from vanpool and transit. At the peak of 2001, for the CTR affected employees within King C ounty, the transit shar e increased 5 percent from 1993 while the vanpool share more than d oubled. This is expected since, for most commuters, the alternatives of walking and bi cycling are constraine d by some factors, such as commute distance and the weather in Puget Sound area, and is less sensitive to the CTR strategies. While the reason for the rebound of the driving alone share after 2001 remains unknown, from the perspective of CTR strategi es, it may suggest the existence of some kinds of Program Tiredness, which me ans as the CTR strategies are being implemented, people are getting used to the stimulation and they are becoming less and less sensitive. It may also suggest that the same type of CTR strategies may have different impacts at different time period. Table 5.1 Mode Shares Trend for CT R Affected Employees in King County Mode Share (%) Program Year Num of Employees Drive Alone Carpool Vanp ool Transit Non Motor 1993 71,691 64.79 14.68 0.97 16.40 3.16 1995 95,812 60.75 15.46 1.63 19.12 3.04 1997 104,013 59.55 15.59 2.63 19.11 3.12 1999 98,804 58.83 16.00 1.83 20.06 3.29 2001 109,671 58.27 15.21 2.11 21.41 2.99 2003 110,763 60.93 13.93 2.65 19.42 3.07 2005 133,681 63.28 13.09 2.48 18.12 3.02

PAGE 124

113 113 Table 5.2 Mode Shares Trend for the Entire CTR Affected Employees Mode Share (%) Program Year Num of Employees Drive Alone Carpool Vanp ool Transit Non Motor 1993 176,722 74.52 13.33 1.10 8.40 2.64 1995 192,525 69.67 15.06 1.40 10.96 2.91 1997 235,009 68.54 15.36 2.36 10.60 3.14 1999 194,975 68.78 14.99 1.41 11.50 3.32 2001 222,584 68.64 14.42 1.69 11.95 3.30 2003 208,486 69.71 13.47 2.04 11.93 2.86 2005 231,322 70.02 12.87 2.17 11.99 2.95 5.3.2 Variable Definition The variables in the utility function of the nested logit model include (1) characteristics of the employer, including bus iness type, total number of employees, and the existence of multiple shifts at the work site; (2) parking management information at the worksite, including parking charge for SOV and HOV, ratio of onsite parking spaces to total number of employees, and the existenc e of reserved parking spaces for HOV; (3) employer paid financial subsidies for alterna tive modes, including the subsidy for transit, carpool, vanpool, bike, and walk; (4 ) employer TDM support/promotion strategies/activities, including the availabi lity of guaranteed ride home program, the availability of company fleet vehicle for ca rpooling or vanpooling, and the promotion activities of distributing program summar y material, sending program information through email, conducting tran sportation event, and pub lishing TDM articles in employee newsletter; and (5) amenities and land-use characteristics at the worksite, including area type (downtown, rural, other), existence of sidewalk, bike-lane, and onsite restaurant, and existence of onsite covered bi ke locker, cloth locker, and showers. The above variables are obtained from the employer annual report.

PAGE 125

114 114 Parking management is one of the handy tools that the employers can lean on to achieve their CTR goal. I use the ratio of number of onsite parking spaces to the total number of employees, rather than the absolute number of parking spaces to measure the accessibility to the parking faci lity. In addition to the parking charge for SOV and HOV parking, I include the HOV parking charge discount, which is defi ned as the difference between the SOV parking charge and that of HOV, to measure the impacts of discounted HOV parking. The financial subsidies to the alternativ e modes include cash incentives, gift card incentives, free passes for transit, and reimburse ment of travel costs. It does not include the parking discount provided for HOV parking. The variables from the employee travel behavior survey include commuters job title, work schedule, and commuting distance. The detailed travel information for both transit and motor between the commuters home ZIP code and the commuters worksite are extracted from Google transit (www.google.com/transit ). A computer program is de signed to search the travel information between each of the worksite-home pair for total more than 44,000 pairs. For driving, the results of the searching incl ude travel time and dist ance. For transit the searching results include first walk time, fi rst in-vehicle time, transfer time and second in-vehicle time if need transfer, and last walk time. Currently, in Washington State, the Google transit service only is available for King County Metro, which includes the Seattle metropolitan area. Therefore, this st udy will focus on the work sites located within King County and the employees residing King County, including the Seattle urban area and other suburban and rural areas. Since more than 60 percent of th e employers affected

PAGE 126

115 115 by the Washington State CTR program are located in King County, the utilization of the sub-sample will not affect the accu racy of the model. Table 5.3 Variable Definitions Variable Description* Total Employees Total number of employees. Finance Info Service Personal Services Manufacture HealthCare Public Utility Military Transportation Government Education Other Primary business of the organization. =1, if Finance, Insurance, Real estate; =0, if Otherwise (13.3%) =1, if Information services/Software/Technical; =0, if Otherwise (9.4%) =1, if Professional/Personal Services; =0, if Otherwise (8.6%) =1, if Manufacturing; =0, if Otherwise (7.1%) =1, if Health Care; =0, if Otherwise (16.3%) =1, if Public Utility; =0, if Otherwise (13.7%) =1, if Military; =0, if Otherwise (2.4%) =1, if Transportation; =0, if Otherwise (2.9%) =1, if Government; =0, if Otherwise (17.4%) =1, if Education; =0, if Otherwise (2.7%) =1, if Other; =0, if Otherwise (5.6%) Shift Does this worksite have multiple shifts? =1, if Yes (78.8%) =0, if Otherwise Administrative Support Management Technical Production/Labor Customer Service Sales/Marketing Other Job title of the employee =1, if Administrative Support; =0, if Otherwise (13.7%) =1, if Management; =0, if Otherwise (14.1%) =1, if Professional/Technical; =0, if Otherwise (46.9%) =1, if Craft/Production/Labor; =0, if Otherwise (8.0%) =1, if Customer Service; =0, if Otherwise (7.6%) =1, if Sales/Marketing; =0, if Otherwise (4.0%) =1, if Other; =0, if Otherwise (5.7%) CWW Does the employee work on comp ressed work week schedules? =1, if yes (16.9%) =0, if otherwise Telecommuting Does the employee work on telecommuting? =1, if yes (5.7%) =0, if otherwise Commute Distance On way distance in mile commute from home to worksite *In the parenthesis is the percentage of the observations with observed value equal to 1 for the dummy variables

PAGE 127

116 116 Table 5.3 Variable Definitions (Cont) Transit In-Vehicle Time Transit travel in vehicle time Transit Out-Vehicle Time Transit travel out vehicle time Transit Transfer Times Transit travel number of transfers Avg. Property Value Average property value in dollar of the ZIP code in which the commuter resides Onsite Parking Charge Parking charge for single occupant vehicles ($/space/month) Onsite Parking Ratio The ratio of total number of onsite parking spaces to total number of employees Reserved HOV Parking Worksite reserved HOV park ing space(s) availability. =1, if reserved HOV parking space(s) available (71.7%) =0, if otherwise HOV Parking Discount Difference between SOV parking and HOV parking ($/space/month) Subsidy Monthly transit, HOV, or bicycle/ Walk subsidy employer paid per participating employee ($/employee/month) Flextime Does your organization offer flex time (Allow employees to vary their start and end times)? =1, if yes (90.8%); =0, if otherwise GRH Is the guaranteed emergency ride home program available at this worksite? =1, if yes (89.1%); =0, if otherwise Distribute Material Does the employer distribute a summary of the worksites CTR program to employees? =1, if Yes (96.4%); =0, if Otherwise CTR Events Does the employer conduct transportati on events/fairs and/ or participate in county/state CTR promotions/campaigns? =1, if Yes (87.8%); =0,if Otherwise CTR Email Does the employer send out the CTR information through email? =1, if Yes (91.5%); =0, if Otherwise CTR Article Dose the employer publish CTR articles in employee newsletters? =1, if Yes (42.9%); =0, if Otherwise *In the parenthesis is the percentage of the observations with observed value equals to 1 for the dummy variables

PAGE 128

117 117 Table 5.3 Variable Definitions (Cont) HOV Fleet Vehicle Availability of employer provided fleet vehicle for carpool =1, if Yes (2.8%) =0, Otherwise Downtown Is the worksite located in the downtown area? =1, if Yes (32.1%) =0, if No Sidewalk Worksite sidewalks availability. =1, if sidewalks is available onsite or within 1/4 mile (84.1%); =0, if otherwise Restaurants Worksite restaurants/cafeteria availability. =1, if restaurant/cafeteria is available onsite or within mile (75.7%); =0, if otherwise Covered Bicycle Racks Worksite covered bicycle spaces, cag es, racks or lockers availability =1, if available onsite (80.4%) =0, if Otherwise Lockers Worksite clothes lockers availability =1, if available onsite (76.4%) =0, if Otherwise Showers Worksite showers availability =1, if available onsite (78.3%) =0, if Otherwise *In the parenthesis is the percentage of the observations with observed value equals to 1 for the dummy variables Table 5.4 Summary Description of Data (Continuous variables) Variable Mean Std. Dev Minimum* Maximum Total Employees 1,842.24 2,560.74 53 11,488 Commute Distance 11.84 8.02 1 55 Onsite Parking Charge 56.17 85.55 0 (58.77%) 310 Onsite Parking Ratio 0.52 0.37 0 (9.27%) 1 HOV Parking Discount 16.88 34.63 0 (72.44%) 280 Transit Subsidy 36.52 28.12 0 (16.88%) 144 HOV Subsidy 11.31 20.88 0 (66.41%) 185 Bike/Walk Subsidy 5.87 13.33 0 (78.44%) 100 Avg. Property Value 347,925 143,427 109000 1,988,184 Transit In-Vehicle Time 35.50 21.80 1 143 Transit Out-Vehicle Time 19.10 13.31 2 101 Transit Transfer Times 0.71 0.71 0 (43.07%) 4 *In the parenthesis is the percentage of the observations with observed value equals to 0

PAGE 129

118 118 I use the average property value of the ZIP code in which the employee resides, combined with the employee job title, to serve as a proxy for the employees socioeconomic information. The property value in cludes the land valu e and the building value. The data are from the King C ounty appraisers web site [King County Department of Assessment, 2005]. The detaile d variables definitions are presented in Table 5.3. Descriptive data statis tics are reported in Table 5.4. 5.4 Model Specification A nested logit model is applied to estimate the determinants of modal choices for the CTR affected employees and evaluate th e impacts of various TDM strategies on commuters modal choices based on a larg e sample of about 60,000 observations. The model specified in equation 5.5 imposes no restriction on the inclusive value parameters. However, as one of the discrete choice models, the nested logit model is derived from the random utility theory. The th eory foundation of random utility theory is utility maximization. For nested logit estimatio n, the model is consis tent with the utility maximization if and only if the inclusive value parameter lies between zero and one. Some normalization is required for the nested logit model to satisfy the inclusive value parameter restriction. Normalization is simply the process of setting one or more scale parameters equal to unity, while allowing the other scale parameters to be estimated. Generally, a nested logit model can be norma lized in two different ways to produce the desired results. For a two level nested logit model, it can be normalized by either setting the scale parameter at the top level equal to one or setting the scale parameter at the bottom level equal to one. In this study, I apply the second way of normalization to specify the model, which is, normalizing the s cale parameter at the branch.

PAGE 130

119 119 The model is a two-level nest ed logit model. The first ne st includes motor, transit, and non-motor. In the second nest, motor is di vided into drive-alon e and shared riding. The mode shares of each of the alternative are: motor, 74.18 percent (drive-alone, 60.61 percent; shared ride, 13.57 percent); transit, 22.29 percent; and non-motor, 3.52 percent. Figure 5.1 exhibits the structure of nested l ogit model. Figure 5.2 de picts the mathematic specification of the model. Determining a traveler's choice set is al ways a problem that is theoretically straightforward but practically difficult. Theore tically, a travelers choice set consists of every mode whose probability of being c hosen exceeds zero. Practically, however, the travelers choice set only contains the mode s whose probabilities of being chosen are large enough to be practically significant. There are no rigorous analytic methods for assigning choice sets to travelers. The accuracy of the definition of a travelers choice set largely depends on the availability of the travelers personal information and the information on mode accessibility. The Washingt on State CTR data shows that less than one percent of travelers taking transit need to transfer three or more times, less than five percent need to transfer two or more times. Fo r the travelers riding bicycle, more than 97 percent commute 20 or fewer miles one way. Th erefore, in this dissertation, the drive alone and shared ride are assumed to be availabl e to all travelers. Tran sit is assumed to be available only to the travelers for whom th e maximum number of transfer times is less than two. Non-motor is assumed to be avai lable to the commuter for whom the one way commuting distance is less than 20 miles.

PAGE 131

120 120 Figure 5.1 Nested Logit Model Structure Figure 5.2 Mathematical Specifica tion of the Nested Logit Model 1. Utility Functions U(Drive Alone) = 1*Onsite Parking Charge + 2*Onsite Parking Ratio U(Shared Ride) = 0 + 1*Commute Distance + HOVP* HOV Parking Discount + ReservedHOVParking*R eserved HOV Parking + 2*Subsidy + 3*Total Employees + 4*Avg. Property Value + 5*Downtown + 6*CWW + 7*Telecommuting + 8*Shift + Flextime*Flextime + GRH*GRH + Distribute*Distribute Ma terial + CTREmail*CTR Email + CTREvents*CTR Events + CTRArticle*CTR Article + 9*Administrative Support + 10*Labor + 11*Manager + 12*Sales + 13*Technical + 14*Finance + 15*Information + 16*Manufacture + 17*Health + 18*Public Utility + 19*Transportation + 20*Government + 21*Education + Restaurant*Onsite Rest aurant + HOVFleet*HOV Fleet Vehicle U(Transit) = 0 + InVehicleTime*Transit In-Vehicle Time + OutVehicleTime*Transit Out-Vehicle Time + 2*Subsidy + 3*Total Employees + 4*Avg. Property Value + 5*Downtown + 6*CWW + 7*Telecommuting + 8*Shift + Flextime*Flextime + GRH*GR + Distribute*Distribute Ma terial + CTREmail*CTR Email + CTREvents*CTR Events + CTRArticle*CTR Article + 9*Administrative Support + 10*Labor + 11*Manager + 12*Sales + 13*Technical + 14*Finance + 15*Information + 16*Manufacture Motor (74.18%) Public (22.29%) Bike/Walk (3.52%) Drive Alone (60.61%) Shared Ride (13.57%) Choice Transit (22.29%) Non-Motor (3.52%)

PAGE 132

121 121 + 17*Health + 18*Public Utility + 19*Transportation + 20*Government + 21*Education + Restaurant *Onsite Restaurant + Sidewalk*Sidewalk U(Non-Motor) = 0 + 1*Commute Distance + 2*Subsidy + 3*Total Employees + 4*Avg. Property Value + 5*Downtown + 6*CWW + 7*Telecommuting + 8*Shift + Flextime*Flextime + GRH*GR + Distribute*Distribute Ma terial + CTREmail*CTR Email + CTREvents*CTR Events + CTRArticle*CTR Article + 9*Administrative Support + 10*Labor + 11*Manager + 12*Sales + 13*Technical + 14*Finance + 15*Information + 16*Manufacture + 17*Health + 18*Public Utility + 19*Transportation + 20*Government + 21*Education + Restaurant *Onsite Restaurant + CoveredBike*Covered Bicycle Rack + Lockers*Onsite Lockers + Showers*Onsite Showers + Sidewalk*Sidewalk 2. Conditional Probabilities P(Drive Alone | Motor) = ) exp( ) exp( ) exp( Ride Shared Alone Drive Alone DriveU U U P(Shared Ride | Motor) = ) exp( ) exp( ) exp( Ride Shared Alone Drive RideU U UShared P(Transit | Pubic) = 1 P(Non-Motor | Bike/Walk) = 1 3. Inclusive Values IVMotor= ln[) exp( ) exp( Ride Shared Alone DriveU U ] IVPulic= UTransit IVBike/Walk = UNon-Motor 4. Branch Probabilities P(Motor) = ) exp( ) exp( ) exp( ) exp(/ / Public Mootr Mootr Walk Bike Walk Bike Public Motor MotorIV IV IV IV P(Public) = ) exp( ) exp( ) exp( ) exp(Bike/Walk / Public Mootr PublicIV IV IV IVWalk Bike Public Motor Public

PAGE 133

122 122 P(Bike/Walk) = ) exp( ) exp( ) exp( ) exp(Bike/Walk / Public Mootr Bike/Walk /IV IV IV IVWalk Bike Public Motor Walk Bike 5. Choice Probabilities P(Drive Alone) = P(Motor)* P(Drive Alone | Motor) P(Shared Ride) = P(Motor)* P(Shared Ride | Motor) P(Transit) = P(Public) P(Non-Motor) = P(Bike/Walk) 5.5 Regression Results The model is estimated using LIMDEP 8.0. After combining all of the data from different source, the final sample size is 62,346. Table 5.5 reports the nested logit regression results. The value of the log likelihood function at its maximum is 50763.5. The chisquare, a statistic used to te st the null hypothesis that all the parameters are zero, is 26532.76 with 81 degrees of freedom, which i ndicates that we can reject the null hypothesis that all the parameters are zer o at the level of at least 0.001. The R-Squared an informal goodness-of-fit index that measur es the fraction of an initial log likelihood value explained by the model, defined as 1-Log-L( )/Log-L(0), is 0.495. The inclusive value coefficients of for alternatives to shared-ride, transit, and non-motor are 2.648, 2.567, and 2.166 respectiv ely and statistically significant at the level of at least 0.001. This statistic suggests the nested logit model is appropriate and necessary to estimate the commuters journey-to-work mode choice. Examining the coefficients in the models for the mode choices, it is first observed that the constant terms are all negative, s uggesting that the average effect of those

PAGE 134

123 123 unobserved influence variables is in the direction of drive-alon e. This is anticipated since more than 60 percent of commuters drive alone to work. Commute distance is a variable in the utility function of shared-ride and nonmotor used to measure the travel impedance. The coefficient on commute distance in the shared-ride equation is positive and statistically significant at the level of at least 0.001, which suggests that the longer the commute distance, the mo re likely commuters choose to share rides to work. On the other hand, the coefficient in the non-motor equation is negative and statis tically significant at the level of at least 0.001, indicating that commuters with longer commute distance are less likely to bicycle or walk. For transit, instead of using distance, three variables of transit in-vehicle time, transit out-vehicle time, and transit transfer times are used as the measurement of travel impedance. Out-vehicle time includes walking to a bus stop, waiting for a vehicle, and transfer time, and walking from a bus stop to the final destination. The coefficient of transit in-vehicle time is posit ive and statis tically significant at the level of at least 0.001. The coefficients of transit out-vehicle time and transit transfer times are negative and statistically significant at the level of at least 0.01. These findings suggest that commuters are less likely to use transit to work if they have to wait longer for a transit to come or walk long distance to transit stops or they ca nnot arrive at their destination non-stop. The positive sign suggests that transit use is not negatively affected by transit in-vehicle time. The positive sign on transit in-vehicle time, together with the findi ngs that out -vehicle time and transfer time have a negative impact on transit use, implies that commuters are more responsive to increases in out-of-vehic le time than in-vehicle time, a conclusion drawn by Domencich et al. [1972] and Small [1992]. The positive coefficient of transit

PAGE 135

124 124 in-vehicle may also be explained by the wide variety of transit services available to the residence. For people who have better acces s to transit service and can reach their worksite non-stop, tran sit use may not be a bad choice cons idering it is cost-effective and not very time-consuming. The number of total employees is used to control for the size of the worksite. The coefficients of this variable are positive and st atistically significant at the level of at least 0.1 for all of the three alternative modes. This result suggests that worksite size has a positive effect on commuters alternative mode choices, which may be explained by the facts that it is easier for commuters to have a good match to share ride in large companies or organizations and that commuters in larg e worksites may have better access to transit use since public transportation stops are ge nerally located in the places with high employment density. The dummy variable, the existence of multiple shifts, is used to control for whether the worksite has multiple shifts. The coefficients of shift for the choices of transit and non-motor are negative and statistically significant at the level of at least 0.01. It is negative but not significant for shared ride. This result suggests that commuters in worksites with multiple shifts are less likely to use transit, bicycle, or walk to work. Possible explanations are that transit services may be not available for certain shifts or it is not very safe to walk or bicycle during night shifts. There are four variables used to capt ure differences in employers policy on parking cost and parking space supply. Parking charge is the variable used only in the equation of drive-alone. The co efficient on parking charge is negative and statistically significant at the level of at least 0.001. Th is result is expected and once again confirms

PAGE 136

125 125 that parking charges do have positive and significant impacts on employers choice of using the alternative modes. HOV parking discount measures the differe nce of the SOV parking charge and HOV parking charge. It is only de fined for the alternative of shared ride. The coefficient of HOV parking discount is pos itive and statistically significan t at the level of at least 0.05. This result confirms that HOV parking discount has positive and significant impacts on commuters choices of carpool and vanpool. I use the onsite parking ratio, which is defined as the ratio of the total onsite parking spaces to the total num ber of employees, rather than the total number of onsite parking spaces, to measure the employers parking facility suppl y. It only enters into the utility function of drive-alone. The coefficien t on the onsite parking ratio is positive and statistically significant at the level of 0.001, which suggest s that the higher the ratio of onsite parking space to the total number of employees, the higher the likelihood for commuters to drive alone to work. If the onsi te parking space is limited, it may increase the time that commuters spending on locati ng an onsite parking space or even force the commuter to park on the offsite parking facility, which in turn increases the out-ofvehicle time for driving alone. Consequently, it may encourage the commuter to use the alternatives to driving alone. Reserved HOV parking is a dummy variable indicating whether reserved parking space is available for high-occupancy vehicles. It only enters into the utility function of shared ride. The coefficient of this variable is positive and statistically significant at the level of at least 0.01, which suggests that the existence of reserved parking spaces has significant positive impacts on a commuters choice of carpooling ad vanpooling.

PAGE 137

126 126 Table 5.5 Nested Logit Regression Results Variable Drive Alone a, b Shared Ride a,b Transit a,b NonMotor a.b Constant -1.881*** (20.5) -0.848***(10.0) -0.834***(8.0) Commute Distance 0.022*** (14.2) -0.158***(10.0) Total Employees 5.2e-5***(8.6) 5.0e-6*(1.7) 1.8e-5***(3.0) Shift -0.029 (0.8) -0.045***(2.8) -0.097***(2.9) Parking Charge -0.001***(-9.7) Onsite Parking Ratio 0.646***(16.9) HOV Parking Discount 0.001**(2.0) Reserved HOV Parking 0.0315*** (5.7) Subsidy 0.005***(11.0) 0.001***(4.0) 0.004***(4.6) Avg Property Value -8.0e-7***(7.5) 1.02e-8(0.2) 2.6e-7***(3.2) CWW -0.132***(3.9) -0.134***(7.3) -0.055*(1.8) Telecommuting -0.161***(2.9) -0.174***(6.0) -0.102*(1.8) Flextime 0.235*** (9.3) 0.235*** (9.3) 0.235*** (9.3) GRH 0.120***(5.7) 0.120***(5.7) 0.120***(5.7) Distribute Material 0.300***(7.1) 0.300***(7.1) 0.300***(7.1) CTR Events 0.159***(7.7) 0.159***(7.7) 0.159***(7.7) CTR Email 0.262***(9.3) 0.262***(9.3) 0.262***(9.3) CTR Article 0.014(1.0) 0.014(1.0) 0.014(1.0) HOV Fleet Vehicle 0.333***(5.3) Transit In-Vehicle Time 0.003***(7.7) Transit Out-Vehicle Time -0.004***(7.1) Transit Transfer Times -0.044***(4.0) Downtown 0.308***(7.5) 0.664***(12.8) 0.21***(5.4) Sidewalk 0.044***(3.1) 0.044***(3.1) Onsite Restaurant 0.118***(7.2) 0.118***(7.2) 0.118***(7.2) Covered Bicycle Racks 0.139***(2.9) Showers 0.088**(2.2) Lockers 0.096**(2.3) Job-Administrative Support 0.011(0.3) 0.066***(3.1) -0.091**(2.0) Job-Production/Labor 0.266***(5.1) 0.029(1.0) 0.082(1.4) Job-Management -0.518***(11.0) -0.530***(12.0) -0.351***(6.1) Job-Sales/Marketing -0.338***(4.9) -0.435***(9.5) -0.308***(4.1) Job-Professional/Technical -0.146***(3.9) -0.119***(6.1) 0.064*(1.8) Business-Finance/Real Estate/Insurance 0.073*(1.7) 0.051***(2.4) -0.088(1.9)* Business-Information Service/Software -0.227***(3.9) 0.008(0.3) -0.018(0.3) Business-Manufacturing -0.412***(9.1) -0.429***(9.6) -0.059(1.1) Business-Health Care -0.203***(4.5) -0.007(0.3) -0.001(0.0) Business-Public Utility -0.106(1.2) -0.050(1.4) 0.048(0.5) Business-Transportation -0.349***(4.5) -0.064*(1.7) -0.442***(3.5) Business-Government -0.153***(3.7)) 0.047***(2.4) -0.095**(2.3) Business-Education -0.251***(3.0) 0.104***(2.9) 0.093(1.5) 2.648***(15.2) 2.567***(13.2) 2.166***(10.3) N(R-Squared) Log-L Chi-squared[81] 62,346 (0.49515) -50763.5 26532.76*** a absolute value of z-statistics in parentheses. b *2-tail significance at = 0.10. **2-tail significance at = 0.05. ***2-tail significance at = 0.01.

PAGE 138

127 127 Alternative modes financial subsidy is one of the most popular strategies employers resort to achieve their CTR goals. Fo r the sample I used to estimate the model, 83.12 percent of employees work for an em ployer providing transit subsidy, 33.59 percent providing carpool or vanpool subsidy, 21. 56 providing bike or walk subsidy. The alternative modes financial incentives include cash incentives, gift card incentive, free pass for transit, and other reimbursements of out-of-pocket travel costs. It does not include the parking discount provided for HOV parking. Transit subsidy, HOV subsidy, and Bicycle-Walk subsidy are generic variables entering the utility functions for each of the alternative modes, the subsidy for drive alon e is assumed to be 0. The coefficients of the subsidies are positive and st atistically significan t at the level of at least 0.001 for all of the alternative modes, suggesting financia l subsidy could be an effective tool to encourage alternative modes use. The specif ic impacts of the financial incentive to a commuters mode choice will be discussed in the next section. The dummy variables of CWW and telecommuting are used to capture the employees difference of their work schedules They are both entered into the utility functions of the alternative modes. The coefficients of CWW and telecommuting are negative and statistically significant at the le vel of at least 0.01 fo r the shared ride and transit and at the level of at least 0.1 for the non-motor. This result suggests that when employers work on compressed work schedules or telecommuting, they are more likely to drive alone. When employees work on CWW, they need to work longer hours and leave home earlier and reach home late. This makes it harder for them to match the

PAGE 139

128 128 schedules of others to make a shared ride. Since most transi t service varies in peak-hour and non-peak periods, it is not surprising that CW W workers are less likely to use transit. Flexible time is a dummy variable to refl ect the availability of the option whether employers allow commuters vary their start and end work time. The coefficients of flexible time are positive and statistically sign ificant at the level of at least 0.001. The effect of flexible time is assumed to be the same across mode choices1. This result suggests that when offering flexibility to comm uters to start and end their work, they are more likely to use alternative modes to driv e alone. While this result is expected, it is important to know the positive impact is signific ant. We will also s ee in the next section that the elasticities of flexible on decreasing the drive alone use is even larger than some other TDM strategies, such as guaranteed ride home program. Since flexible time is easy to implement and cost-effective to empl oyers, it should be recommended to the employers as one of the effective and effi cient strategies to achieve the CTR goals. The variable of guaranteed ride home program enters the utility functions of all of the alternative modes and assumed to have the same impacts across the alternative models. The variable of HOV fleet vehicle measures the availability of employer provided fleet vehicles for carpool or vanpool. It only enters into the utility function of shared ride. As expected the coefficients for both of the variables are positive and significant at the level of at least 0.01. Four dummy variables are used to m easure the employers TDM promotion activities, including distributing CTR inform ation, conducting transportation events, publishing CTR articles, and sending electronic mail mess ages about the CTR program. 1 When allowing this variable to have different marginal effect across modes, the coefficients are very close.

PAGE 140

129 129 Their effects are assumed to be the same across the alternative modes. Except publishing CTR articles, which is positive but not significant, the coefficients of the other three variables are all positive and st atistically significant at th e level of at least 0.001. A dummy variable downtown is used to c ontrol for the location of the worksite. The coefficient on downtown is positive and statistically significant for all of the alternative modes, suggesting employees work ing in downtown areas are more likely to share ride, transit, bicycle, or walk to work. There are se veral reasons that induce the commuters working downtown to be more likel y to take the alternative modes. Firstly, the downtown area may have better access to the public transportation system and many other amenities. Secondly, because of peak period congestion, auto use to downtown areas may associate with it a high degree of travel time unreliability. The uncertainty about arrival time may induce downtown workers to leave from home earlier. This extra time may be viewed as a source of disutility to the urban traveler. Sidewalk and onsite restaurant are two dummy variables cont rolling for the landuse design of the worksite. Restaurant indicates whether a restaurant is available onsite or within 0.25 miles from the worksite. Sidewalk enters the utility function of transit and non-motor and its impact on these two modes are assumed to be the same. The coefficient on sidewalk is positive and statistically sign ificant at the level of at least 0.01. The coefficients on onsite restaurant are positive and statistically significant at the level of at least 0.01. These two findings suggest that landuse designs that are pedestrian friendly in the areas of high employment density may also play a positive role in helping achieving CTR goals.

PAGE 141

130 130 Covered bicycle racks, s howers, and lockers are the three non-motor specific variables to indicate whether employers provi de such facilities to help bicycling or walking employees. The impacts of the three variables on non-motor using are positive and statistically significant. Five dummy variables are used to reflect employees job title. The coefficients of management and sales/marketing are negative an d statistically significant at the level of at least 0.01, suggesting th at managers and employees with the job title of sales/marketing are more likely to drive alone to work. The coefficients of administrative support and professional/technical dummy variables have mixed signs. Employees working as administrative support are more likel y to use transit and less likely to bicycle or walk to work, while pr ofessional/technical commuters are more likely to use nonmotor and less likely to use shared ride and transit commut ing to work. The coefficients of production/labor are positiv e but only statistically si gnificant for shared ride, suggesting production/labor workers are more likely to share ride to work. From the above results, it is clear that a commuters j ourney to work mode choice is affected by his or her job characteristics. This suggests that when encouraging an alternative mode to drive-alone, employers should consider their job characteristics a nd provide incentives tailored to commuters with di fferent job characteristics. There are eight dummy variables used to capture employers major business type. The coefficients of manufacturing, health car e, and transportation are negative for all of the alternative modes, suggesting that compared to other industry, commuters in the three above industries are more likely to drive alone. The co efficients of other dummy variables have mixed signs.

PAGE 142

131 1315.6 Elasticity and effect Analysis One of the major objectives of this chap ter is to measure the quantitative impacts of the TDM strategies on a co mmuters journey to work mode choice. For the continuous variables, the quantitative impacts are measur ed as elasticities, which, for a discrete choice model, is the percentage change in the probability of choosing one of the alternatives due to a 1-percent of change in some attribute th at is an independent variable in the model. For the dummy variable, the e ffect is measured as the change in the probability of choosing one of th e alternatives with and wit hout the strategy. The direct and cross aggregate elasticities of the conti nuous variables on a commuters mode choice are calculated according to equation 5.8, 5.9, and 5.10. Tabl e 5.6 reports the elasticities of the strategies whose impacts ar e statistically significant. Th e effect of a dummy variable for each observation is simply the difference of the possibility of choosing an alternative when the dummy variable equals to one and equals to zero The aggregate impact of a dummy variable is the average of all of the observations. Table 5.7 reports the effects of the dummy variables whose impact s are statistically significant. Parking charge is the variable only entered into the utility function of driving alone. The direct elasticity of 0.281 and cr oss elasticities of 0.086, 0.307, 0.150 suggest that when the parking charge increases by 10 percent, commuters likelihood to drive alone falls by 2.81 percent; shared ride, tran sit, and non-motor increase by 0.86, 3.07, and 1.50 percent respectively. The empirically derived as well as modeled parking price elasticities of demand from various empirical an alyses generally range from 0.1 to 0.6, with -0.3 being the most frequently cited value [Vaca and Kuzmyak, 2005].

PAGE 143

132 132 Onsite parking ratio measures parking space supply from employer. Its direct elasticity for drive-alone is 0.317, while its cr oss elasticities for shar ed ride, transit, nonmotor are 0.190, 0.471, and 0.394 respectivel y. If an employer decreases the ratio of number of onsite parking spaces to total number of employees by 10 percent, a commuters likelihood to drive alone decrease by 3.17 percent; shared ride, transit, and non-motor increase by 1.90, 4.71, and 3.94 percent respectively. The direct elasticity of HOV parking disc ount is 0.065, suggesting that when the parking discount to high occupancy vehicl es increases by 10 pe rcent, a commuters likelihood to commute by shared-ride increase by 0.65 percent. Alternative modes financial subsidy is one of the most popular strategies employers use to achieve their CTR goals. In order to evaluate the cost effectiveness of a commute trip reduction program, it is essen tial to accurately measure the quantitative impacts of the alternative modes financial subs idies. It is interesting to see that the subsidy elasticities for transit and shared ride reported in this dissertation are relatively lower than the price elasticities reported in the literature. For example, the direct elasticity of subsidies provide d to share-riders is 0.509, suggesting that when the subsidies to shared riders increases by 10 pe rcent, the likelihood for commuters to share ride increases by 5.09 percent. York and Fabr icatore [2001] estimate the price elasticity of vanpooling at about 1.5. Th e direct elasticity of tran sit subsidy is 0.108, while Transport Research Library [TRL, 2004] calcu lates that bus fare elasticities average around 0.4 in the short-run, 0.56 in the medi um run, and 1.0 ove r the long run. Metro rail fare elasticities are 0.3 in the short run and 0.6 in the long run. This might be explained by fact that the subsidy only covers part of the out-of-thepocket travel cost of

PAGE 144

133 133 transit and shared ride. The direct elasticity of subsidies to non-mo tor is 0.233, suggesting that when the subsidies provided to non-moto r users increase by 10 percent, commuters likelihood to use non-motor to work increases by 2.33 percent. Reserved HOV parking and HOV fleet ve hicle are two dummy variables entered into the utility function of shared ride to measure the non-monetary supportiveness of the shared ride program. If the employer prov ides reserved parking to high occupancy vehicles, the commuters likelihood to share ri de to work increases by 2.248 percent and the likelihood of driving alone decrease by 0.747 percent. If HOV fleet vehicle is available to shared riders, commuters likelihood to share ride to work increases by 4.858 percent. If flexible time is allowed, a commuter s likelihood of using share ride, transit, and non-motor increases by 1.195 percent, 4.817 pe rcent, and 0.733 percent, respectively. Table 5.6 Elasticities for Se lected Continuous Variables Variable Drive Alone Shared Ride Transit Non-Motor Parking Charge -0.281* 0.086** 0.307** 0.150** Onsite Parking Ratio 0.317* -0.190** -0.471** -0.394** HOV Parking Discount -0.065** 0.109* -0.074** -0.041** Subsidy Shared Ride -0.359** 0.460* -0.236** -0.167** Subsidy Transit -0.080** -0.019** 0.066* -0.036** Subsidy Non-Motor -0.330** -0.069** -0.148** 0.252** *Direct elasticities, **Cross elasticities Table 5.7 Effects for Selected Dummy Variables Variable Drive Alone Shared Ride Transit Non-Motor Reserved HOV Parking -0.747 2.248 -1.334 -0.167 Flextime -6.745 1.195 4.817 0.733 GRH -3.563 0.574 2.597 0.392 Distribute Material -0.849 0.154 0.603 0.0918 CTR Events -0.775 0.122 0.567 0.086 CTR Email -0.472 0.075 0.345 0.052 HOV Fleet Vehicle -1.776 4.858 -2.730 -0.351

PAGE 145

134 134 5.7 Conclusion This chapter focuses on the impacts of an integrated commute trip reduction program on commuting mode choice. A nested l ogit model is develope d to estimate the employees commute mode choices. In particul ar, the effect of TD M promotion activities and support strategies, parking management, worksite amenit ies, and alternative modes subsidies are examined. Furthermore, the elasticities of the financial incentives, the parking management programs, the TDM su pport programs, and the TDM promotion activities are calculated to measure the quant itative impacts of the TDM strategies. A trend analysis of Washington State Commute Trip Reduction program shows that the mode shares of driving alone dec line from 74.52 percent in 1993 to 68.64 percent in 2001 for the CTR affected employee. After 2001, the share of driving alone rebounded to 70.02 percent in 2005. Despite the rebound, the shares of driving alone for the CTR affected employees are significantly lower th an the state average of 74.3 percent and national average of 77 percent in 2005. Th e significant differences come from carpool, vanpool, and transit, especially from vanpool and transit. Although what causes the rebound of shar e of drive-alone after 2001 remains unknown, from the perspective of CTR strategi es, it may suggest the existence of some kinds of program fatigue, which means as the CTR strategies being implemented, people are getting used to the stimulation and they are becoming less and less sensitive. It may also suggest that the same type of CTR strategies may have different impacts at different time period.

PAGE 146

135 135 From the nested logit model, it is conf irmed that, overall, the impacts of the TDM programs on commuters choice of the altern ative modes are positiv e and statistically significant. In particular, from the perspective of parking management, it shows that if the SOV parking charge increases by 10 percent, a commuters likelihood to drive alone falls by 2.81 percent, while shared ri de, transit, and non-motor in crease by 0.86, 3.07, and 1.50 percent respectively. If the employer decreas es the ratio of number of onsite parking spaces and total number of employees by 10 percent, the commuters likelihood to drive alone decrease by 3.17 percent, while shared ride, transit, and non-motor increase by 1.90, 4.71, and 3.94 percent respectively. Wh en the parking discount to high occupancy vehicles increases by 10 percent, a commut ers likelihood to co mmute by shared-ride increases by 1.09 percent. If employer pr ovides reserved parking to high occupancy vehicles, commuters likelihood to share ride to work increases by 2.248 percent and the likelihood of driving alone decrease by 0.747 percen t. If HOV fleet vehi cle is available to shared riders, a commuters likelihood to shar e ride to work increases by 4.858 percent. As for the alternative modes financial subs idies, the direct elasticities of shared mode, transit, and non-motor are 0.460, 0.066, and 0.252, respectively, and their elasticities on driving alone are 0.359, 0. 080, 0.330 The financia l subsidy elasticities for transit and shared ride reported in this di ssertation are relatively lower than the price elasticities reported in literature. A possible explanation might be that subsidies only cover part of the out-of-the-pocke t travel cost of transit and shared ride. This means a one percent increase in the subsidy is worth less than one percent increase in transit or

PAGE 147

136 136 carpool or vanpool fare and, consequently, has a smaller less impact on a commuters commuting mode choice. The quantitative impacts of TDM promoti on activities and support strategies are also examined. The findings suggest that when encouraging alternatives to drive-alone, it is important that employers provide certain associated services, which may not only provide the physical help the commuter needed but also signal th e supportiveness from the management and provide peace of mind to potential users. The model developed in this chapter can be applied directly to estimate or predict the mode shares for the TDM affected employ ee. The derived elasticities can be used to evaluate the quantitative impacts of the indivi dual strategies. Additionally, in order to achieve the goals to reduce th e share of drive-alone, employe rs may consider different combinations of strategies to implemen t depending on their major business type and detailed job characteristics.

PAGE 148

137 137CHAPTER 6 CONCLUSION This dissertation focuses on an overall evaluation of the Commute Trip Reduction program implemented in Washington State. In particular, I investigate the impacts of Travel Demand Management strategies im plemented by the employers on commuters travel behavior from three aspects. First, I examine the important determinants of employees compressed work weeks schedul e choice and how TDM promotions and strategies affect their decision to participate in CWW. Second, I develop models to examine the effectiveness of telecommuting as a component of an integrated TDM program and to predict the telecommuting rate and telecommuting frequency in the future. Finally, I apply a nested logit model to evaluate the quantitative impacts of various TDM strategies on commuters journey-to-work mode choices. 6.1 Contribution In this thesis, although I use different methodology to address the impact of TDM on commuters travel behavior from differen t perspectives, the models I build are an integrated modeling effort to evaluate and fo recast the effectiveness of an existing TDM. A direct measurement of the effectiveness of a CTR program is the number of vehicle trips or peak period ve hicle trips reduction. Based on the number of reduced vehicle trips, other measurements, such as the reduction of delay, travel time, and fuel consumption and emission, can then be derived. Gene rally, a comprehensive employer based CTR program achieves the goal of vehicle trip reduction through implementing worksite-based

PAGE 149

138 138 TDM strategies that focus on changing the commuters mode choice, travel time choice, and travel frequency choice. In particular compressed work week and telecommuting programs are aimed at changing commuters trav el frequency and travel time, which help reduce the number of person trips or peak period person trips. Employers TDM support strategies and financial incen tives or disincentives strategies are implemented to encourage employees to use alternative modes of drive-alone and reduc e the vehicle trips. An integrated procedure of the employer-based TDM effectiveness evaluation, therefore, consists of estimating the number of employees working on compressed work week and telecommuting and the percentage of employ ees shifted from driving alone to the alternative modes. The models developed in this dissertation can be applied to address these three issues. The evaluation of the eff ectiveness of the employer-b ased TDM program can be categorized to evaluating an existing progr am based on the employee travel behavior survey and predicting or estimating the imp acts of a program based on survey data on TDM program implemented by employer. For an existing employer-based TDM programs that both the employer promoti on data and employee travel behavior information are available, such as the Washington State CTR program, the evaluation process generally consists of calculating and comparing the vehicle trip rate or vehicle miles traveled for each employer before and after the implementation of TDM strategies. For most of other employer-based TDM progr ams, where the employee travel behavior information is not available, the program assessment procedure normally includes applying the TDM models to estimate or pred ict the vehicle trip rate change based on employer program implementation data.

PAGE 150

139 139 The models developed in this dissertati on can easily be applied to evaluate the impacts of existing TDM programs. For me tropolitan areas where a comprehensive commute trip reduction program is implemen ted but no detailed information on employee travel behavior available, the models can be applied to estimate the quantitative impacts of TDM strategies on mode share and the num ber of CTR affected employees working on compressed work schedules and telecommuting when employers and employees information on basic variables such as job title are available. Furthermore, the models may be incor porated into the regional transportation model to reflect TDM impacts in the tran sportation planning process. For the area affected by the Washington State CTR program the models can be directly used to predict the percentage of em ployees working on compressed work week schedules and telecommuting at the TAZ level for CTR a ffected employees. For other areas where detailed employer and employee data is not av ailable, the model developed here may be simplified to use aggregate data at the TAZ level to predict the participation rate of compressed work week schedules and telecommuting. For example, we can use the average commute distance, the percentage of employers affected by TDM strategies, and the percentage of employees by job title and business type to estimate CWW and telecommuting participation rate. The proj ected percentage of employees working compressed work weeks and telecommuting then may be applied to adjust the number of home based work trips to reflect its impacts on the transportation system and, at the same time, to improve the accuracy of the regional planning model. Similarly, the nested logit model, that incorporated the TDM strategies can be applied to reflect the impacts of TDM strategies on mode shares.

PAGE 151

140 140 As for the activity based travel demand model, the CWW and telecommuting models can be applied as sub-models to estimate the job related activities. For example, they can be applied to estimate how many workers will actually go to work and how many will work at home or stay home because of telecommuting and CWW. Finally, the effectiveness of the TDM strategies is one of the important factors that determine the prospect of success of an integrated TDM program. This dissertation identifies effective TDM strategies based on the employees job related characteristics and personal information through modeling the TDM impacts on commuting travel behavior. The findings may help policy make rs evaluate the e ffectiveness of TDM strategies and choose the most effi cient ways to reduce trips rates. 6.2 Major Findings Based on the analysis of the impact of Transportation Demand Management programs on commuters Compressed Work Week choice, telecommuting choice, and journey to work mode choices, the major findings are summarized as following: Job characteristics and employers major bus iness type are important factors that affect commuters choice on CWW, telecommuting, and journey to work mode. This finding suggests that when considering eff ective ways to promote TDM programs that help achieve CTR goals, transportation planne rs and CTR coordinator in each worksite should identify industry characteristics and group commuters based on their job characteristics. Certain combinations of TDM strategies should be tailored to reflect the difference in job requirements, which may wo rk better than providing a uniform TDM program for everyone.

PAGE 152

141 141 TDM promotion activities play an importa nt role in changi ng commuters travel behavior. In this thesis, four TDM promotion activities are identified to have positive and significant effects on commuters choices of CWW, telecommuting, and using alternative modes to driving alone to work. CTR coordi nators should review their TDM promotion activities to see whether they can do more or switch to some more effective ways to communicate with CTR affected employees. I find distributing CTR information regularly, sending CTR program summary a nd information to employees by email, publishing CTR articles regularly, and conducti ng transportation event are effective ways to help employees understand the TDM program benefits and employers supportiveness, and take actions accordingly. TDM promotion should be a continuous e ffort to promote CTR goals. When the number of TDM promotion years is used in regression, I find the time effect of TDM promotion is not constant. I al so find that, while the drive al one share declined significant from 1993 to 2001 for the CTR affected empl oyees, it began to rebound after 2001. This finding may suggest that transportation pl anners in government agencies and CTR coordinators may need to adjust their promo tion efforts from time to time based on their survey results. Financial incentives and disincentives are important determinants of commuters journey to mode choices. In particular, subsidies to shar e-ride, transit, and non-motor have significant positive effects on commuter c hoice of using alternative to drive alone. Parking management, including parking ch arge, discount parking charge for HOV parking, and reserved parking spaces for HOV parking has significant positive impacts on reducing driving alone. Services or amenitie s offered by employers such as guaranteed

PAGE 153

142 142 ride home, providing HOV fleet vehicle, complementary faci lities for bicyclers or walkers are also effective to help reduce driving alone. To achieve CTR goals to reduce vehicle tr ips, besides employer-based strategies, coordination and cooperation from other government agencies are also important components of an overall effort. For exam ple, land use design to include certain amenities and land use policies to allow mixed use of employment, business, and residential may also help to achieve the CTR goals. 6.3 Future Research Since TDM is a long term process, it is important to track individual commuters travel behavior changes over time. A pane l study of a large sample of individual commuters on their CWW, telecommuting, and mode choices will be a more effective way to estimate effectiveness of TDM program s. More importantly, this may have very important policy implementation to identify strategies that work better over time. The other limitation of this research is data restriction. A lthough the Washington State Commute Trip Reduction dataset pr ovides a large sample of individual observations, the data do not ha ve commuters personal informa tion, such as gender, age, household size, and household income. To reduce omitted variable bias, I have to introduce some proxy to capture the commuters difference in social status and household characteristics. It will dramatically increas e the accuracy of the m odels should some basic personal information such as commuters age range and household characteristics be available. Most of TDM strategy information, such as the parking charge, alternative modes subsidies, and the implemented TDM promoti on activities, is repo rted by the Employer

PAGE 154

143 143 Transportation Coordinators at employer leve l. If the data can be collected at the employee level, it will signifi cantly increase the accuracy of those data, consequently, the accuracy of the models.

PAGE 155

144 144REFERENCES Adiv, A. Behavioral Determinants of Rapid Transit Patronage: Why Dont More People Ride BART to Work? Ph.D. Dissertation, Department of City and Regional Planning; University of Ca lifornia, Berkeley, 1980. Allen, R.E. and Hawes, D.K. Attitude toward work, leisure and the four day working week. Human Resource Management 18, 1979, pp. 5 -10. Bard, E. Transit and Carpool Commuting and Household Vehicle Trip Making: Panel Data Analysis. Transportation Research Record No. 1598, 1997, pp. 25 31. Barton Aschman and Associates and Richar d H. Pratt and Associates. Traveler Response to Transportation System Changes, Second Edition. Report No. DOTFH-11-9579, U.S. Department of Transportation, 1981. Bento, A., Cropper, M., Mobarak, A., and Vinha, K. The Impact of Urban Spatial Structure on Travel Demand in the Unite d States. Policy research working paper 3007, World Bank, 2003. Ben-Akiva, M. Structure of Passenger Travel Demand Models Ph.D. dissertation. Department of Civil Engineering, MIT, Cambridge, Massachusetts, 1973. Ben-Akiva, M. and Lerman, S. Discrete Choice Analysis The MIT Press, Cambridge, 1985. Berman, W. Travel demand managemen t: Thoughts on the new role for TDM as a management and Operation Strategy. Journey of Institute of Transportation Engineers Vol. 72, No. 9, 2002. Bernardino, A., Ben-Akiva, M., and Salom on, I. Stated preference approach to modeling the adoption of telecommuting. Transportation Research Record No. 1413, 1993, pp. 22 30. Bhat, C. Covariance heterogeneity in nested logit models: econome tric structure and application to intercity travel. Transportation Research. Part B: Methodological Vol. 31, No. 1, 1997, pp. 11 21.

PAGE 156

145 145 Bhat, C. and Sardesai, R. The Impact of Stop-Making and Travel Time Reliability on Commute Mode Choice. Transportation Research. Part B: Methodological Vol. 40, No. 9, 2006, pp. 709 730. Bhattacharjee, D., Haider, S., Tanaboriboon, Y., and Sinha, K. Commuters attitude towards travel demand management in Bangkok. Transport Policy 2, 1997, pp. 61 170. Brownstone, D. and Golob, T. The Eff ectiveness of Ridesharing Incentives: Discrete-Choice Models of Commuting in Southern California. University of California Transportation Center, Univ ersity of California, Irvine, 1991. Center for Urban Transportation Research, Un iversity of South Florida. Worksite Trip Reduction Model (Accessed July 1, 2007). Cervero, R. Mixed Land-Uses and Commuting: Evidence From the American Housing Survey. Transportation Research. Part A: Policy and Practice Vol. 30, No. 5, 1991, pp. 361 377. Choo, S., Mokhtarian, P. L., and Salomon, I. Does telecommuting reduce vehicle-miles traveled? An aggregate time series analysis for the U.S. Transportation: Planning, Policy, Research, Practice Vol. 32, No. 1, 2005. Drucker, J. and Khattak, A. J. Propensity to work from home: Modeling results from the 1995 National Personal Transportation Survey. Transportation Research Record No. 1706, 2000, pp. 108 117. Duany, A., Plater-Zyberk, E., and Speck, J. Suburb Nation: The Rise of Urban Sprawl and the Decline of the American Dream North Point Press, 2000, pp. 88 94. U.S. Environmental Protection Agency. COMMUTER model. (Accessed June 1, 2006) Ferguson, E. Privatization as choice proba bility, policy process and program outcome: The case of transportation management associations. Transportation Research, Part A: Policy and Practice, Vol. 31, No. 5, 1997. FHWA TDM Evaluation Model. Estimation the Effect of Alternative Work Schedules on Travel Acti vity and Emission, 1993. Francis, W. and Groninga, C. The Effects of the subsidization of employee parking human behavior. Unpublished research pape r, School of Public Administration, University of Southern California, 1969.

PAGE 157

146 Fulton, L.M., Noland, R. B., Meszler, D. J ., and Thomas, J.V. A statistical analysis of induced travel effects in the U.S. mid-A tlantic region. Journal of Transportation and Statistics, Vol. 3, No. 1, 2000, pp. 1 14. Government Accounting Office. Trans portation Infrastructure: States' Implementation of Transportation Management Systems, January, 1997, GAO/RCED 97 32. Greene, W. Econometric Analysis ( 4th edition ). Prentice Hall, 2000, New York. Greene, W. NLOGIT Version 3.0 Reference Guide, 2002 Giuliano, G. and Golob, T.F. Staggered work hours for traffic management: a case study. Transport Research Record 1280, 1990, pp. 46 58. Goulias, K., Pendyala, R., and Kitamura, R. A practical method for the estimation of trip generation and trip chaining. Transportation Research Record 1285, 1990, pp. 47 56. Hamer, R., Kroes, E., and Van Ooststroom, H. Teleworking in the Netherlands: An evaluation of changes in travel behavlour. Transportation Research Record No. 1357, 1992. Hendricks, J. S. Commuter Choice Pr ogram Case Study Development and Analysis. National Center for Transit Res earch, University of South Florida, 2004 . Higgins, T. J. Parking Management a nd Traffic Mitigation in Six Cities: Implications for Local Policy. K.T. Analytics, Inc., paper presented to the Transportation Research Boar d, Washington D.C., 1989. Ho, A. and Stewart, J. Case study on impact of 4/40 compressed workweek program on trip reduction. Transportation Research Record No. 1346, 1992, pp. 25 32 Hung, R. Use compressed work weeks to reduce work commuting, Transport Research Part A Vol. 30, No. 1, 1996, pp. 11 19. Institute of Transportation Engineers. GAO releases management systems report, ITEJournal April, 1997 . Johnson, M. A. Psychological Variables and Ch oices Between Auto and Transit Travel: A Critical Research Review, Working Pape r No. 7509, Institute of Transportation and Traffic Engineering, Universi ty of California, Berkeley, 1975.

PAGE 158

147 Johnston, R.A. and Rodier, C.J. Critique of metropolitan planning organizations capabilities for modeling transportati on control measures in California. Transportation Research Record, Vol. 1452, 1994, pp. 18 26 King County Department of Assessment, 2005. http://www.metrokc.gov/assessor/download/download.asp (Accessed August 5, 2007) Kuppam, A., Pendyala, R., and Rahman, S. Anal ysis of the Role of Traveler Attitudes and Perceptions in Explaining Mode-Choice Behavior. Transportation Research Record No. 1676, 1999. pp. 68 76. Louviere, J. The Development and Test of Math ematical Models of Traveler Perceptions and Decisions. Final Report 27, The Institute of Urban and Regional Research, University of Iowa Iowa City, 1981. Mannering, J.S. and Mokhtarian, P.L. Modeling the choice of telecommuting frequency in California: An exploratory analysis. Technological Forecasting and Social Change Vol. 49, 1995, pp 49 73. McFadden, D. Conditional Logit Analysis of Qualitative Choice Behavior Zarembka(ed), Frontiers in Econometrics Academic Press, New York, 1973, pp. 105 142. McFadden, D. Economic choice s. American Economic Review, Vol. 91, No. 3, 2001, pp. 351 378. Mehranian, M., Wachs, M., Shoup, D. a nd Platkin, R. Parking Cost and Mode Choices among Downtown Workers: A Case Study. Transportation Research Record No. 1130, 1987, pp. 1 5. Meyer, M. A Toolbox for Alleviating Traffic C ongestion and Enhancing Mobility. Institute of Transportation Engineers (I TE) and Federal High way Administration < http://ntl.bts.gov/lib/8000/8700/8780/toolbox.pdf >. Mokhtarian, P. Telecommuting and Travel: St ate of the Practice, State of the Art. Transportation, Vol. 18, 1991. Mokhtarian, P. and Solomon, I. Modeling the choice of telecommuting: setting the context. Environmental Planning Vol. 26, 1994, pp. 749 766. Mokhtarian, P. and Salomon, I. Modeling the Choice of Telecommuting: Identifying the Choice Set and Estimati ng Binary Models for Technology-Based Alternatives Choice. UCTC working paper, No. 264, 1995.

PAGE 159

148 Mokhtarian, P. and Salomon, I. Modeling th e desire to telecommute: The importance of attitudinal factors in behavioral models. Transportation Research A Vol. 31, No. 1, 1997, pp. 35 50. Nakamura, K. and Kockelman, K.M. Conge stion pricing and road space rationing: An application to the San Fr ancisco Bay Bridge corridor. Transportation Research Part A Vol. 36, No. 5, 2002, pp. 403 417. Nilles, J. Traffic reduction by telecommuting: A status review and selected bibliography. Transportation Research A, Vol. 22, No. 4, 1988, pp. 301 317. Nollen, D. The compressed working week: Is it worth the effort? Industrial Engineering Vol. 13, 1981, pp. 58 64. Nozick, L., Borderas, H., and Meyburg, A. Evaluation of travel demand measures and programs: a data envelopment analysis approach. Transportation Research Part A Vol. 32, No. 5, 1998, pp. 331 343. Ory, D. and Mokhtarian, P. Dont Work, Work at Home, or Commute? Discrete Choice Models of the Decision for San Francisco Bay Area Residents. The University of California, Davis, 2005. (7 July 2007) Parkhurst, G. Park and ride: Could it lead to an increase in car traffic? Transport Policy Vol. 2, No. 1, 1995, pp. 15 23. Parkhurst, G. Influence of bus-based park a nd ride facilities on users car traffic. Transport Policy Vol. 7, No. 2, 2000, pp. 159 172. Pendyala, R. M., Goulias, K. S., and Kita mura, R. 1991. Impact of telecommuting on spatial and temporal patt erns of household travel. Transportation (Netherlands) Vol. 18, No. 4, 1991. Peng, Z., Dueker, K., and Strathman, J. Residential Location, Employment Location, and Commuter Responses to Parking Charges. Transportation Research Record No. 1556, 1996, pp. 109 118. Popuri, Y. D. and Bhat, C. R. On modeli ng choice and frequency of home-based telecommuting. Transportation Research Record No. 1858, 2003, pp. 55 60. Replogle, M. Transportation conformity a nd demand management: vital strategies for clean air attainment. E nvironmental Defense Fund, 1993 .

PAGE 160

149 Rose, G. Providing premium carpool parki ng using a low-tech ITS initiative. Journey of Institute of Transportation Engineers Vol. 72, No. 7, 2002, pp. 32 36. Ronen, W. and Primps, S.B. The compresse d work week as organizational change: Behavior and attitudinal outcomes. Academic Management Review Vol. 6, 1981, pp. 61 74. Salomon, I. Telecommunications and trav el: Substitution or Modified Mobility? Journal of Transport Economics and Policy, Vol. 19, No. 3, 1985, pp. 219 235. Sampath, S., Saxena, S., and Mokhtarian, P. L. The effectiveness of telecommuting as a transportation control measure. Institute of Transportation Studies, University of California, Berkeley, Research report, UCD-ITS-RR-91-10, 1991. Schrank, D. and Lomax, T. Urban Roadway Congestion Annual Report 1998 Texas Transportation Institute, College Statation, Texas, 1998. Schrank, D. and Lomax, T. The 2005 Urban Mobility Report Texas Transportation Institute, Colle ge Statation, Texas, 2005. Shafizadeh, K. R., Mokhatarian, L. P., Niemei er, D. A., and Salomon, I. The costs and benefits of telecommuting : A review a nd evaluation of micro-scale studies and promotional literature. PATH research report UCB-ITS-PRR-2000-13, 2000 Shafizadeh, K. R., Niemeier, D. A., Mokht arian, P., and Salomon, I. Costs and Benefits of Home-Based Telecommuti ng: A Monte Carlo Simulation Model Incorporating Telecommuter, Employer, and Public Sector Perspectives. Journal of Infrastructure Systems, Vol. 13, No. 1, 2007 Shoup, D. and Breinholt, M. Employer-pai d parking: A nationwide survey of employers parking subsidies policy. The Full Social Costs and Benefits of Transportation. Greene, E., Jones, D., and Delucchi, M. Springer-Verlag, Berlin, 1997, pp. 371 385. Shoup, D. Evaluating the Effects of Ca lifornias Parking Cash-out Law: Eight Case Studies. Transport Policy Vol. 4, No. 4, 1997, pp. 201 216. Shoup, D. and Pickrell, D. Free Parking as a Transportation Problem. U.S. Department of Transportation, Washington D.C., 1980. O'Sullivan, A. Urban Economics Chicago: Irwin, 2003.

PAGE 161

150 Sullivan, M. A., Mahmassani, H. S., and Yen, J. Choice model of employee participation in telecommuting under a cost-neutral scenario. Transportation Research Record No. 1413, 1994, pp. 42 48. Sundo, M. and Fujii, S. The effects of a compressed working week on commuters daily activity patterns. Transportation Research, Part A Vol. 39, 2005, pp. 835 848. Surber, M., Shoup, D., and Wachs, M. Eff ects of Ending Employer-Paid Parking for Solo Drivers. Transportation Research Record No. 957, 1984, pp 67 71. Tanaboriboon, Y. Demand managementa n alternative approach to relieve traffic congestion in the developing count ries: Asian metropoliss context. In: Proceedings of the Japan society of civil engineers. Doboku Gakki Ronbun Heokokusheu, Vol. 488, 1994, pp. 11 19. Tanadtang, P., Park, D., and Hanaoka, S. Incorporating Uncertain and Incomplete Subjective Judgments into the Evaluati on Procedure of Tr ansportation Demand Management Alternatives. Transportation Vol. 32, No. 6, 2005, pp. 603 626. Taylor, C. J., Nozick, L. K., Meyburg, A. H. Selection and evaluation of travel management measures. Transportation Research Record, Vol. 1598, 1997, pp. 49. Thorpe, N., Hills, P., and Jaensirisak, S. Public attitudes to TDM measures: A comparative study. Transport Policy Vol. 2, No. 4, 2000, pp. 243 257. Train, K. A structured logit model of auto ownership and mode choice. Review of economic studies Vol. 47, No. 147, 1980. Transport Research Laboratory. The Demand for Public Transit: A Practical Guide Transportation Research Labor atory, Report TRL 593, 2004 (www.trl.co.uk ), . Transportation Research Ci rcular 433: TDM Innovation and Research Symposium: Setting a Strategic Agenda for the Future. TRB, National Research Council, Washington, D.C., Oct. 1994. United States Environmental Protection Agency. Commuter model v2.0 user Manual, 2005 Accessed June 2007 at .

PAGE 162

151 Vaca, E., and Kuzmyak, J. R. Traveler response to transportation system changes. Chapter 13 parking pricing and fees. TCRP Report No. 95, 2005. (Accessed January 2, 2007) Victoria Transport Policy Institute. TDM Encyclopedia . Viegas, J. Making urban road pricing accep table and effective: searching for quality and equity in urban mobility. Transport Policy Vol. 8, 2001. Wache, M. Anticipated attitudinal responses to dual-mode transit systems and their effects on mode choice Transportation Research Board Special Report, No. 170, 1976. Wachs, M. Transportation demand management: policy implications of recent behavior research. UCTC No. 23, 1990. Transportation Center, The University of California, Berkeley (R eprinted from Journal of Planning Literature). Wambalaba, F., Concoc, S., Chavarria, M. Price Elasticity of Rideshare: Commuter Fringe Benefits forVanpools. National Center for Transportation Research, Center for Urban Transportation Research < www.nctr.usf.edu/pdf/527-14.pdf >. Washbrook, K., Haider, W., and Jaccard, M. Estimating Commuter Mode Choice: Discrete Choice Analysis of the Impact of Road Pricing and Parking Charges. Transportation: Planning, Policy, Research, Practice Vol. 33, No. 6, 2006, pp. 621 639. Washington State Department of Trans portation. Commute Trip Reduction Program < http://www.wsdot.wa.gov/TDM/CT R/default.htm> (Accessed April 5, 2006). Washington State Departme nt of Transportation. CTR Task Force 2005 Report to the Legislature < http://www.wsdot.wa.gov/TDM/CTR/libary.htm> (Accessed May 5, 2006). Washington State Legislature. WAC 468-63-020, Definitions. < http://apps.leg.wa.gov/WAC/default.as px?cite=468-63-020> ( Accessed May 1, 2006). Washington State Departme nt of Transportation. TDM Effectiveness Evaluation Model (TEEM) (Accessed August 15, 2006). Williams, R. Generalized Ordered Logit/ Partial Proportional Odds Models for Ordinal Dependent Variables. The Stata Journal Vol. 6, No. 1, 2006, pp. 58 82.

PAGE 163

152 Wilson, W. R. Estimating the travel a nd parking demand effects of employerpaid parking. Regional Science and Ur ban Economics, Vol. 22, 1992, pp. 133 145. Wilson, W.R., and Shoup, D. Parking subs idies and travel c hoices: Assessing the evidence. Transportation, Vol. 17, 1990, pp. 141 157. Winters, P. Transportation Demand Mana gement. Transportation in the New Millennium. Washington DC: TRB. A5010: Committee on Transportation Demand Management, 2000 . York, B. and Fabricatore, D. Puget S ound Vanpool Market Assessment, Office of Urban Mobility, WSDOT, 2001 www.wsdot.wa.gov/mobility/TDM/studyvpmrkt.html

PAGE 164

ABOUT THE AUTHOR Liren Zhou has been working as a gra duate assistant in the Travel Demand Management program at the Center for Urba n Transportation Research (CUTR) at the University of South Florida while working to ward his Ph.D. in Civil Engineering with a major field in transportation modeling and pla nning. Before he started his Ph.D. studies, he worked as a transportation planner/en gineer from 2003 to 2004 at Gannet Fleming, Inc. He also worked as a mechanical engi neer and a project manager when he was in China, where he obtained his Bachelors degree in Mechanical Engineering. Liren has co-authored two papers published in the Journal of Public Transportation and Compendium of Technical Papers ITE 2007 Annual Meeting and Exhibit. Another paper based on this disse rtation research, entitled An Empirical Analysis of Compressed Work Week Choices Based on Washington State Commute Trip Reduction Data, has been accepted for publication in Transportation Research Record: Journey of the Transportation Research Board.