USFDC Home  USF Electronic Theses and Dissertations   RSS 
Material Information
Subjects
Notes
Record Information

Full Text 
xml version 1.0 encoding UTF8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchemainstance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd leader nam Ka controlfield tag 001 001469434 003 fts 006 med 007 cr mnuuuuuu 008 040524s2004 flua sbm s0000 eng d datafield ind1 8 ind2 024 subfield code a E14SFE0000335 035 (OCoLC)55732377 9 AJR1188 b SE SFE0000335 040 FHM c FHM 090 TA145 1 100 Challa, Srikalyan. 2 245 A structural equation analysis of Florida journey to work characteristics using aggregate Census 2000 data h [electronic resource] / by Srikalyan Challa. 260 [Tampa, Fla.] : University of South Florida, 2004. 502 Thesis (M.S.C.E.)University of South Florida, 2004. 504 Includes bibliographical references. 516 Text (Electronic thesis) in PDF format. 538 System requirements: World Wide Web browser and PDF reader. Mode of access: World Wide Web. 500 Title from PDF of title page. Document formatted into pages; contains 98 pages. 520 ABSTRACT: The need for a better understanding of journey to work behavior has never before been so important. Many transportation corridors are functioning at unacceptable levels of service and many at times to their capacity. This phenomenon is more pronounced during peak period when majority of the population is making their work trip. This research builds on the recent developments in structural equations modeling technique for identifying the sociodemographic influences on the commute behavior of the population in Florida. Towards this purpose a series of five structural equations models are estimated using aggregate level data from Census 2000. Each of these models has a set of journey to work characteristics that are observed for their behavior against prevalent sociodemographic characteristics. The journey to work characteristics identified are exhaustively studied for various relationships to the sociodemographic characteristics. The model estimation led to the identification of relations between various journey to work characteristics and the sociodemographic characteristics at the Census Tract level. Some of the results obtained supported other studies performed earlier. It is hoped that the findings of this research would broaden the horizon in understanding journey to work behavior of the population of Florida. 590 Adviser: Pendyala, Ram M. 653 peak period departure. place of work. mode to work. Census tracts. work commute travel time. 0 690 Dissertations, Academic z USF x Civil Engineering Masters. 773 t USF Electronic Theses and Dissertations. 4 856 u http://digital.lib.usf.edu/?e14.335 PAGE 1 A Structural Equation Analysis of Flor ida Journey to Work Characteristics Using Aggregate Census 2000 Data by Srikalyan Challa A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Civil Engineering Department of Civil and Environmental Engineering College of Engineering University of South Florida Major Professor: Ram M. Pendyala, Ph.D. Steven E. Polzin, Ph.D., P.E. Jian J. Lu, Ph.D., P.E. Date of Approval: April 9, 2004 Keywords: census tracts, work commute travel time, place of work, mode to work, peak period departure Copyright 2004 Srikalyan Challa PAGE 2 ACKNOWLEDGMENTS I am greatly delighted at this opportunity to express my gratitude to Dr. Ram M. Pendyala, my advisor, who is an unvarying source of inspiration for me. I thank him for being there to support and encourage me all through the course of my graduate study at the University of South Florida. I thank Dr. Steve Polzin who is an excellent teacher, for his support and encouragement not only through the coursework but also as the Faculty Advisor of the ITE student Chapter. I thank all the faculty, researchers and staff of the Department of Civil and Environmental Engineering and Center for Urban Transportation Research (CUTR) whose commitment and zeal in their own expertise has made the transportation graduate study program an exciting experience for many a graduate student like me. I also thank my colleagues at USF whose helping hand put me in good stead in numerous occasions. I once again thank Dr. Steve Polzin, Dr. Jian J. Lu, for their support and guidance both in coursework and the performance of this thesis. PAGE 3 TABLE OF CONTENTS LIST OF TABLES iv LIST OF FIGURES v ABSTRACT vii CHAPTER 1. INTRODUCTION 1 1.1 Background 1 1.2 Objectives 2 1.3 Methodology 3 1.4 Outline 3 CHAPTER 2. LITERATURE REVIEW 4 2.1 Introduction 4 2.2 Travel time to work 4 2.3 SOV mode to work 5 2.4 Location of employment 7 2.5 Peak period departure to work 7 2.6 Structural equations modeling 8 CHAPTER 3. DESCRIPTION OF DATA 10 3.1 Census survey description 10 3.2 Census tract level data 12 3.3 Florida census tracts 13 3.3.1 Population 13 3.3.2 Urban and rural classification 15 3.4 Person characteristics across census tracts 16 3.4.1 Gender distribution 18 3.4.2 Worker distribution 22 3.4.3 Worker age distribution 23 3.5 Household characteristics across census tracts 24 3.5.1 Number of households 25 3.5.2 Annual household income 26 3.5.3 Household size 27 3.5.4 Vehicle ownership 29 3.6 Journey to work characteristics 31 3.6.1 Travel time to work 31 3.6.2 Place of work 34 3.6.3 SOV Mode to work 39 i PAGE 4 3.6.4 Peak period departure to work 44 CHAPTER 4. MODELING METHODOLOGY 47 4.1 Background 47 4.2 Model structure 48 4.3 Covariancebased structural analysis 50 4.4 Model identification 51 4.5 Maximum likelihood 53 4.6 Model evaluation 54 4.7 Types of relationships 54 CHAPTER 5. MODEL SPECIFICATION AND RESULTS 56 5.1 Background 56 5.2 Path diagrams 57 5.3 Travel time Place of work model 58 5.4 Trip mode Place of work model 60 5.5 Travel time Trip mode model 62 5.6 Trip mode Peak period model 64 5.7 Travel TimeTrip ModePeak PeriodPlace of work model 66 CHAPTER 6. MODEL RESULTS INTERPRETATION 74 6.1 Travel time to work and household income 74 6.2 Travel time to work and household size 75 6.3 Travel time to work and old age population 75 6.4 Travel time to work and vehicles in household 76 6.5 Travel time to work and population density 76 6.6 Travel time to work and percent female population 77 6.7 Travel time to work and working male population 77 6.8 Place of work and household income 78 6.9 Place of work and household size 78 6.10 Place of work and aged population 78 6.11 Place of work and percent female population 78 6.12 Place of work and vehicles in household 79 6.13 Travel mode to work and household income 79 6.14 Travel mode to work and household size 79 6.15 Travel mode to work and aged population 79 6.16 Travel mode to work and vehicles in household 80 6.17 Travel mode to work and percent female population 80 6.18 Travel mode to work and working male population 80 6.19 Travel mode to work and percent population less than twenty five years of age 80 6.20 Peak period departure and household income 81 6.21 Peak period departure and household size 81 6.22 Peak period departure and male population between the ages of twenty five and fifty 81 ii PAGE 5 CHAPTER 7. CONCLUSIONS 82 7.1 Background 82 7.2 Summary of findings 83 7.2.1 Relations between socioeconomic and journey to work characteristics 83 7.2.2 Relations among journey to work characteristics 84 7.3 Omitted variables 85 7.4 Conclusions and future research directions 85 REFERENCES 87 iii PAGE 6 LIST OF TABLES Table 3.1 Land area and population 15 Table 3.2 Percent urban population distribution in census tracts 16 Table 3.3 Person characteristics mean percent across census tracts 17 Table 3.4 Worker distribution across population less than twenty five years of age 24 Table 3.5 Household characteristics mean percent across census tracts 24 Table 3.6 Vehicle distribution by percent of census tracts 30 Table 3.7 Short travel time in three person households 33 Table 3.8 Percent population greater than seventy years of age and place of work 35 Table 3.9 Percent two vehicle households and place of work 36 Table 3.10 SOV mode to work and population greater than seventy years of age 43 Table 3.11 Auto ownership and SOV usage to work 43 Table 3.12 Percent female population and SOV usage to work 43 Table 3.13 High income households and peak period departure 44 Table 5.1 Travel time Place of work model results 68 Table 5.2 Trip mode Place of work model results 69 Table 5.3 Travel time Trip mode model results 70 Table 5.4 Trip mode Peak period model results 71 Table 5.5 Travel time Trip mode Work location Peak period model results 72 iv PAGE 7 LIST OF FIGURES Figure 3.1 Standard hierarchy of census geographic entities 14 Figure 3.2 Total population density 16 Figure 3.3 Percent male population between twenty five and fifty years of age Across census tracts 18 Figure 3.4 Percent males in census tract 19 Figure 3.5 Population distribution in Florida by census tract 20 Figure 3.6 Percent population greater than seventy years of age in census tract 21 Figure 3.7 Percent females in census tract 22 Figure 3.8 Percent workers in census tract 23 Figure 3.9 Number of households in census tract 26 Figure 3.10 Percent households in census tract with income greater than $60,000 27 Figure 3.11 Percent two person households in census tracts 28 Figure 3.12 Percent three person households in census tracts 29 Figure 3.13 Percent households in census tracts with two or more vehicles 31 Figure 3.14 Percent workers whose work travel time is less than ten minutes 33 Figure 3.15 Percent workers whose work travel time is greater than sixty minutes 34 Figure 3.16 Percent workers working in place of residence 36 Figure 3.17 Workers with work travel time less than ten minutes across Florida 37 Figure 3.18 Workers with work travel time greater than sixty minutes across Florida 38 v PAGE 8 Figure 3.19 Percent workers working in place of work across Florida 40 Figure 3.20 Percent workers using SOV mode to work 41 Figure 3.21 Percent workers using SOV mode to work across Florida 42 Figure 3.22 Percent workers departing during peak period 45 Figure 3.23 Percent workers departing during peak period across Florida 46 Figure 4.1 Structural equations modeling framework 48 Figure 5.1 Travel time Place of work model 59 Figure 5.2 Trip mode Place of work model 61 Figure 5.3 Travel time Trip mode model 63 Figure 5.4 Trip mode and Peak period model 65 Figure 5.5 Travel time Trip mode Work location Peak period model 67 vi PAGE 9 A STRUCTURAL EQUATION ANALYSIS OF FLORIDA JOURNEY TO WORK CHARACTERISTICS USING AGGREGATE CENSUS 2000 DATA Srikalyan Challa ABSTRACT The need for a better understanding of journey to work behavior has never before been so important. Many transportation corridors are functioning at unacceptable levels of service and many a times to their capacity. This phenomenon is more pronounced during peak period when majority of the population is making their work trip. This research builds on the recent developments in structural equations modeling technique for identifying the sociodemographic influences on the commute behavior of the population in Florida. Towards this purpose a series of five structural equations models are estimated using aggregate level data from census 2000. Each of these models has a set of journey to work characteristics that are observed for their behavior against prevalent sociodemographic characteristics. The journey to work characteristics identified are exhaustively studied for various relationships to the sociodemographic characteristics. The model estimation led to the identification of relations between various journey to work characteristics and the sociodemographic characteristics at the Census Tract level. Some of the results obtained supported other studies performed earlier. It is hoped that the findings of this research would broaden the horizon in understanding journey to work behavior of the population of Florida. vii PAGE 10 CHAPTER 1 INTRODUCTION 1.1 Background In this era of changing travel behavior and trip making charact eristics, journey to work constitutes a significant activity in the routine of an individual. Work trips contribute substantially to congestion in ur ban areas. For majority of the population, journey to work is the most predictable in te rms of various trip feat ures like trip length, mode used, and time of departure, origin an d destination, and transfer locations. The observable work trip characteristics include tr avel time, travel mode, time of departure, work place location. The understanding of journey to work charact eristics is of utmost importance to the transportation policy makers at various levels in the government. These characteristics need to indicate the behavior across ju risdictions of various sizes than just observations at specific lo cations. One of the signifi cant factors influencing the characteristics of a work trip is the social and economic variability in worker composition at any given location. These characteristics interact among themselves and also with work trip characteristics to various degrees resulting in the observed journey to work behavior. This thesis is an attempt to gain in sights in to the causal relations for work trip characteristics from the prevailing sociodemogr aphic dispersion in the State of Florida. PAGE 11 1.2 Objectives The overall objective of this research is to develop and estimate simultaneous equation models relating to sociodemogra phic characteristics and journey to work behavior to better address the following issues: Journey to work behavior rela tions at an aggregate level Identification of significant relations hips among sociodemographics and journey to work behavior Understanding the direct, indirect, and total effects in the modeling system Recent research in the travel behavior arena has focused on a variety of disaggregate analysis techniques to understand accurately, by clearly identifying any interrelations among the units of analysis. To utilize the findi ngs at the lowest unit level (say, person or household) in pr edicting the behavior of a larg er geographic unit such as a census tract, county or a state, the quantifie d results (at person or household level) need to be aggregated to the higher level (county, state) of interest. Disaggregate methods of analysis usually n eed data at person or household level. These data collection processes are labor intens ive, which constraints such data collection at various geographic locations. Hence a perf ect representa tion of the population can not be obtained for use to aggregate the estimates. In this context, aggregate models play a significant supporting role not only to check for discrepancies from the estimates at disaggregate level obtained by using localized data but also in giving a over all trend about a characteristic of interest. PAGE 12 1.3 Methodology Structural equation models (SEM) were developed to understand the causal relationships between sociodemogr aphic variables and journey to work characteristics like travel time, time of depart ure, and mode of departure. A set of five models was used to observe c onsistency in the behavior of the relationships. The models developed examine the interactions between sociodemographic characteristics and journey to work variables at the level of census tract. Data for this model development was obtained from the Census Summary File 3 (SF3). 1.4 Outline The remainder of the thesis is organized as follows. Next chapter provides an literature review on the journey to work characteristics and usage of structural equation models to understand travel behavior. The th ird chapter constitutes description of the data that was used in this study. The fourth chapter describes the methodology. The fifth chapter specifies the models used to in this research and tabulates their estimation results. The sixth chapter interprets the model estimation results. The seventh chapter draws conclusions and gives directions for future research which is foll owed by the list of references. PAGE 13 CHAPTER 2 LITERATURE REVIEW 2.1 Introduction Many significant research contributions were made in understanding journey to work characteristics of i ndividuals. Transportation engineers, land planners, policy makers are all interested in understanding the significant, predictable trip making behavior of individuals. The transportation sy stem is at its peak usage during the peak period caused by journey to work trips of i ndividuals. Structural equation modeling is used in this thesis to analyze the commute behavior at an aggregate level. The unit of analysis is census tract a nd the socio economic information obtained is a cumulative for the census tract. 2.2 Travel time to work Brownstone, et al. (2003) developed a logit mode l to identify the road users willingness to pay as much as $30 an hour on the I15 in California. Road users perception of congestion and sensitivity to travel time might vary depending on the available alternatives. This study found that commuters, individuals from high income (income greater than 100K) households, women, and individuals between the ages of 35 and 45 are more likely to pay toll to avoid congestion. These observations give us an understanding on the sensitivity levels of these particular groups to travel time. PAGE 14 Schwanen, et al. (2001) used the data from Dutc h national survey and analyzed the relationship between trav el time and work duration. Th ey found that workers are willing to have travel time up to 10 percent of their work time. Though, this study is from Europe, the results give an understanding of the general commuter tolerance to travel time compared against working time. Bhat and Guo (2003) estimated a mixed l ogit model to identif y the factors in residential choice location. It wa s found that commute travel time is a significant factor in the decision making along with sociodemogr aphics, transportation system LOS. Levinson (1998) analyzed the effect of acce ssibility to jobs and houses at both the home and work ends of trips on commu ting duration for respondents to a household travel survey in metropolitan Washington, DC. He found that residences in jobrich areas and workplaces in housingrich areas are associated with shorter commutes. An implication of this study is that, by balanc ing accessibility, the s uburbanization of jobs maintains stability in commuting durations despite rising congestion, increasing trip lengths, and increased work and nonwork trip making. 2.3 SOV mode to work Collia, et al. (2003) studied the travel behavior of older popu lation (with age greater than 65 years of age) as depicted in 2001 NHTS and found that use of alternative transportation is relatively low. Excluding pers onal vehicle and walking, all other means of transportation account for about 2% of da ily travel. Further, of the older population with medical conditions that affect their trav el, only about 12% use special transportation PAGE 15 services such as dialaride. This shows a greater dependence of elderly population on SOV mode. Bhat, et al (1997) formulated an econometric methodology to estimate the component of the analysis framework involv ing the joint modeling of evening commute mode choice, number of evening commute st ops, and number of stops after arriving home from work. The results of this study indicate strong effects of socioeconomic variables, residential/workplace location characteristics, work schedule characteristics and level of service measures on evening commute mode choice. Cervero (2002) studied the effect of built environment on mode choice. His analysis reveals intensities and mixtures of land us e significantly influence decisions to drivealone, share a ride, or patronize transit, while the influences of urban design tend to be more modest. Sermons and Koppelman (2001) develope d multinomial logit models (MNL) of residential location choice for twoworker households in the San Francisco Bay Metropolitan Area to identify household characteristics that account for the relative differences in household sensitivity to female and male commutes when making residential choices. The results reveal that presence of chil dren, occupation of the male worker, and the relative order of the last residential change and the last change in the female worker's workplace are important de terminants of female and male commuting time parameters in household reside ntial location u tility functions. PAGE 16 2.4 Location of employment Analysis by Clark, et al. (2000) utilized descriptive measures of distance and time to work for preand postresidential re locations and develops estimates from a probability model of workpla ce attraction. The findings indicate that both oneand twoworker households with greater separati on between workplace and residence make decreases in distance and time. Overall, as other studies have shown, women commute shorter distances and are more likely to mi nimize commuting after a move than are men. Randall Crane (1994) found that more likel y, the individual value of a given home and the choice of commute length are based not only on the current job site, bu t also on the expectation of where future jobs will be and the likelihood of both job separations and residential moves. 2.5 Peak period departure to work Conquest, et al. (2002) classified users as (a) ro ute changers, willing to change route both on Interstate 5 and before leav ing; (b) nonchangers, unwilling to change departure time, route, or mode of transporta tion; (c) route and time changers, willing to change route and departure time; and (d) pretrip changers, willing to change departure time, route, or mode before departure but unwilling to change en route. Knowledge of such groups and their behavioral characteristics is useful in designing advanced traveler information systems that seek to affect comm uter behavior and increase the efficiency of current transportation facilities Hendrickson, et al. (1984) examined the flexibility of departure times for the journey to work making use of data gather ed in Pittsburgh, Pennsylvania. Measured PAGE 17 travel time peaking is pronounced for trips in to the Pittsburgh Cent ral Business District, although the variation in travel time is low fo r a particular route, mode and departure time. Estimation of a logit model of simu ltaneous mode and departure time interval choice is reported. Departure time decisions are found to be much more flexible (elastic) than are mode choices. 2.6 Structural equations modeling Structural Equations Modeling (SEM) is a statistical methodology that takes a confirmatory hypothesistesting approach to the analysis of a structur al theory regarding phenomenon. Typically, this theory represen ts causal processes that generate observations on multiple variables. The term structural equation modeling conveys two important aspects of the procedure: (1) that the causal processes under study are represented by a series of structural equations and (2) that these st ructural relations can be modeled pictorially to enable a clearer conceptualization of th e theory under study. Golob and Zondag (1984) is another earl y application of SEM. Golob and Van Wissen (1989) attempt an explanation of car ownership and travel distances by mode, but the SEM has just household income as a household characteristic. Applied to data on attitudes, perceptions, stated behavioral in tentions, and actual behavior, SEM can be used to specify and test alternative hypotheses of causality. Tardiff (1976) used path analysis (a simplified appl ication of SEM) to demonstrate empirical evidence that the causal link from choice be havior to attitudes is stronger than the link from attitudes to choice behavior. Subsequent studies using different forms of simultaneous equation modeling showed cons istently that attitudes, especially PAGE 18 perceptions, are conditioned by choices, while at the same time, attitudes affect choices (e.g., Dobson, et al. 1978). Golob and Brownstone (1992) is another early application of SEM in which it is shown that behavior conditions attitudes, while, simultaneously, attitudes have some affect of behavior. Golob, et al. (1997) presented an SEM in which changes in travel times, attitudes toward carpooli ng, mode choice, and use of an exclusive freeway lane for carpools are modeled over time us ing a U.S. panel dataset. Lu and Pas (1999) have analyzed a set of relationships with travel behavior variables as endogenous and sociodemographics and activity particip ation variables as exogenous variables. This study performs the analysis at a disaggregate level. In summary, many studies as examined in this literature review tried to explore the relationship betw een a particular journey to work char acteristic like travel time, travel mode, peak hour departure, place of wo rk and various other land use and sociodemographic characteristics. Each of these va riables was analyzed in isolation from the other journey to work characteristic. Also, mo st of the studies used disaggregate data for their analysis. Hence, based on the available li terature in using structural equations model and the journey to work characteristics, a se t of five models were developed to explain commute characteristics at an ag gregate level of census tracts. The evolution of census tracts with resp ect to socioeconomic and demographic characteristics and its implications for journe y to work characteristics is an important issue that warrants study. Policy makers a nd planners would be able to assess the potential shifts in journey to work ch aracteristics as census tract evolves. PAGE 19 CHAPTER 3 DESCRIPTION OF DATA The amount of information desired from respondents by transportation planners has increased tremendously over recent years (Kalfs and Saris, 1997). Census survey still continues to be the most utilized information source regarding the prevalent demographic and economic conditions. It is planned to use census data in the proposed structural equation modeling framework of relationships between the journey to work variables and the demographic and economic variables. This chapter aims at presenting an overview of the data set used in this study. This chapter elaborates on the depth of information that is available for usage in such an analysis. The next section gives a brief description of the 2000 Census survey that is used to collect the data that is analyzed in this research. 3.1 Census survey description Census 2000 was the largest peacetime effort in the history of United States. Information about 115.9 million housing units and 281.4 million people across the United States was collected. A limited number of questions were asked of every person and housing unit in the United States. This is called the census short form. The questions asked include Household relationship, age, sex, Hispanic or Latino origin, race tenure (home is owned or rented). More detailed information was asked of a sample (1 in every PAGE 20 6) persons or housing units. This is called the long form survey. The long form survey included sections in population and housing. The journeytowork items were provided in the population section. The identification and location of an estimated 118 million housing units in the nation was performed by census bureau by developing and maintaining the Master Address File (MAF). The United States Postal Service (USPS) played a vital role in contributing to the MAF. The census questionnaire and related materials delivered to individual addresses carried the same themes and messages as the overall campaign. The Census Bureau used public meetings and the news media to inform the public about the value of the census and to encourage response. In 2000, in addition to mailing the census questionnaires, the Census Bureau made the forms available in stores and malls, in schools, and in other public locations. A tollfree telephone number was available for those who wished to respond to the census by telephone. People also had the option to respond to the short form through Internet. In Census 2000, the questionnaire mailout/mailback system was the primary means of censustaking, as it has been since 1970. The short form was delivered to approximately 83 percent of all housing units. The short form asked only the basic population and housing questions, while the long form included additional questions on the characteristic of each person and of the housing unit. The long form was delivered to a sample of approximately 17 percent of all housing units. The Census Bureau adopted a ten part, integrated data enumeration strategy to ensure that completed questionnaires were obtained for every household possible. Special populations (American Indians, Alaskan Natives etc) were identified. PAGE 21 3.2 Census tract level data Census tracts are subcounty geographic entities that are viewed as reasonably permanent and are typically identified by state and local participants in the Census Bureaus Participant programs. The definition of census tracts is to generate and maintain longterm statistical units. The census tracts are intended to be maintained over many decades so that longitudinal comparisons can be made over various long form questions. The target population of a census tract is around 4000. Census tracts might be changed with local developments such as new subdivisions, highway construction, etc. In addition, census tracts occasionally are split due to population growth or combined as a result of substantial population decline. Census 2000 is the first census for which the entire United States was covered by census tracts. Census Tracts are numbered by a four system of numbers ranging from 0001 to 9999. Some of the numbers are reserved for certain categories of population. The four digit number is followed by a two digit suffix which indicates the year in which the census tract was identified. The definitional criteria advertised by the Census Bureau contribute to a reasonable amount of uniformity, especially for tracts. Major geographic features, transportation routes, and boundaries of political entities constitute relatively stable universal standards for establishing tracts. The suggested population parameters for tracts and block groups are another source of uniformity. Local and state personnel involved in block group decisions may change them over time to suit changing local conditions. In contrast, the emphasis on preserving longitudinal comparability of tracts contributes to uniformity across time. Especially in the cores of longstanding metropolitan areas, census tracts provide a sound basis for longitudinal analysis. PAGE 22 Due to substantial and systematic variation in the characteristics of counties, they cannot be used as spatial units of analysis. Of the 3142 counties in United States, 25 percent occupy less than 450 square miles and another 25 percent cover at least 900 square miles. Those in the western United States tend to encompass much more territory than do counties in the east. In terms of population variability, 25 percent of U.S. counties have fewer than 11,000 residents while another 25 percent have populations in excess of 60,000. Also, there are statewide idiosyncrasies such as Louisianas Parishes and Virginias treatment of independent cities as county equivalents. Due to these reasons the census tract forms a much more consistent spatial unit of analysis, and hence is used in the thesis. 3.3 Florida census tracts At the geographic level census tracts add up to form counties. The hierarchical chart depicts the various geographic levels identified in the Census enumeration. Florida constitutes 3154 census tracts. These census tracts vary in land area, population, socioeconomic and journey to work characteristics. Census tract is the unit of modeling adopted in this research work. 3.3.1 Population The average land area for the census tracts was found to be 17.1 square miles with a population average of 5067. PAGE 23 Figure 3.1 Standard hierarchy of census geographic entities Source: Census 2000 summary file 3 technical documentation PAGE 24 The population density average of 3320 is far greater number than the density obtained from the average of total population to total land area. This indicates that large areas exist with much lesser population than the average and that population concentrations exist. This nonuniform population distribution can be seen in Figure 3.3 and in Figure 3.2 the distribution of population density across census tracts. Table 3.1 Land area and population N Minimum Maximum Mean Land Area (in Sq. Miles) 3154 .03 1154.74 17.1 Total Population 3154 .00 24506.00 5067 Total Population Density 3154 0 38851 3320 3.3.2 Urban and rural classification The Census Bureau classifies as urban all territory, population housing units located with in urbanized areas (UAs) and urban clusters (UCs). UAs and UCs boundaries constitute densely settled territory which consists of: A cluster of one or more block groups or census blocks each of which has a population density of at least 1000 people per square mile at the time Surrounding block groups and census tracts each of which has a population density of at least 500 people per square mile at a time Less densely settled blocks that form enclaves or indentations, or are used to connect discontiguous areas with qualifying densities All areas that are not urban are classified as rural. Table 3.2 depicts the distribution of urban population in terms of percentage of total population for the entire state PAGE 25 3.4 Person characteristics across census tracts In order to better understand the over all behavior, the census tract composition in terms of persons living in a single census tract is to be clearly identified. These person characteristics are detailed in the next few sections Figure 3.2 Total population density 0100002000030000Total Population Density (pop/sq mi) 02004006008001,000Number of Census Tracts 051015202530Percent Census Tracts Table 3.2 Percent urban population distribution in census tracts Percent urban population in census tract Frequency Percent of census tracts 0 20 195 6.2 20.01 40 81 2.6 40.01 60 94 3.0 60.01 80 136 4.3 80.01 100 2648 84.0 Total 3154 100.0 PAGE 26 It is observed that 84% of the census tracts have 80% or more of their population as urban population. Less than 10 % of the Census tracts have predominantly rural population. Table 3.3 Person characteristics mean percent across census tracts Total number of census tracts 3154 Male age (< 15) years 19.5% Male age (1525) years 12.4% Male age (2550) years 35.7% Male age (5070) years 20.3% Male age ( 70) years 12.1% Female age (< 15) years 17.7% Female age (1525) years 11.4% Female age (2550) years 34.4% Female age (5070) years 21.3% Female age ( 70) years 15.2% Workers 42% Table 3.3 details the mean values of the percent male and female population by age group. The mean value of percent workers among the population in a census tract is found to be 42 percent. Figure 3.3 gives the distribution of percent of males in the 25 to 50 age group across the census tracts. PAGE 27 Figure 3.3 Percent male population between twenty five and fifty years of age across census tracts 0.0020.0040.0060.0080.00100.00Percent Male Population between 25 and 50 years of age 0100200300400500Number of Census Tracts 0246810121416Percent of Census Tracts 3.4.1 Gender distribution The distribution of male population across the census tracts is described in Figure 3.4 PAGE 28 Figure 3.4 Percent males in census tracts 0.0020.0040.0060.0080.00100.00Percent Male Population 02004006008001,0001,2001,400Number of Census Tracts 010203040Percent of Census Tracts PAGE 29 Figure 3.5 Population distribution in Florida by census tract PAGE 30 It can be observed that mean male population in a census tract is around 50 percent. Similarly it was observed that the female average close to 51 %. Figure 3.6 depicts the distribution of population above 70 years of age across census tracts. Figure 3.6 Percent population greater than seventy years of age in census tract 0.0020.0040.0060.0080.00Percent Population greater than 70 Years of Age 0100200300400500600Number of Census Tracts 05101520Percent of Census Tracts PAGE 31 Figure 3.7 Percent females in census tract 0.0010.0020.0030.0040.0050.0060.0070.00Percent Females in a Census Tract 02004006008001,000Number of Census Tracts 051015202530Percent of Census Tracts 3.4.2 Worker distribution The data on weeks worked in 1999 were derived from answers to longform questionnaire Item 30b, which was asked of people 15 years old and over who indicated in longform questionnaire Item 30a that they worked in 1999. The data were tabulated for people 16 years old and over and pertain to the number of weeks during 1999 in which a person did any work for pay or profit (or took paid vacation or paid sick leave) or worked without pay on a family farm or in a family business. Total workers per census tract are seen to be close to 42 percent. PAGE 32 Figure 3.8 Percent workers in census tract 20.0040.0060.0080.00100.00Percent Workers in a Census Tract 0100200300400Number of Census Tracts 024681012Percent of Census Tracts 3.4.3 Worker age distribution The table 3.5 below describes variation in percent workers by varying percent of population less than 25 years of age. Large number of census tracts fall in the category having 20 to 40 percent of the population being less than 25 years of age. Among the census tracts with 20 to 40 percent of population less than 25 years of age, majority of the census tracts seem to have 40 to 60 percent workers in them. As expected, greater number of census tracts with less than 20 percent of population as workers seem to have less than 20 percent of their populations with age less than 25 years of age. The reason PAGE 33 for this might be due to a greater old age population in the census tracts which results in both lesser workers as well as lesser percent of population less than 25 years of age. Table 3.4 Worker distribution across population less than twenty five years of age Census tracts with Percent population less than 25 years of age Total Percent population as Workers 020 2040 4060 6080 80100 020 57 9 5 0 3 74 2040 280 524 199 5 7 1015 4060 132 1690 184 8 1 2015 6080 10 34 3 2 0 49 80100 1 0 0 0 0 1 Total 480 2257 391 15 11 3154 3.5 Household characteristics across census tracts Household characteristics like the person characteristics influence the journey to work characteristics of the residents of a census tract. The following sections describe some of the household characteristics and their relationships. Table 3.6 gives the description on household size, household income and household auto ownership based on as mean values across census tracts. Table 3.5 Household characteristics mean percent across census tracts Total Number of Census tracts 3154 One person household 26.2% Two person household 36.2% Three person household 15.6% Four or more person household 22.0% Low Income (< 30,000$) 38.6% Medium Income (30,000$60,000$) 32.8% High Income ( 60,000$) 28.6% Zero car household 8.6% One car household 40.6% Two car household 38.2% Three or more car household 12.6% PAGE 34 It could be observed from Table 3.6 on an average in all the census tracts two person households seem to form the highest percent of all households at 36.2% which is a little over one third of all households in a typical census tract. Single person households are also as common as two person households, by taking almost one third of all households. Four and more person households are about 20 percent of all the households. Car ownership numbers from table 3.6 show that the largest share is taken by one and two car households, together forming 80 percent of all households. Zero car households form close to 10% of all households in the census tract. Interestingly, contrary to the income figures, we can see that percent of three or more care households are greater than percent of zero car households in a typical census tract. 3.5.1 Number of households The number of households in a census tract averages at 2000 as shown in Figure 3.9. The size of the household varies from single person household to households with more than 7 persons. Census defines households as including all the people who occupy a housing unit. (People not living in households are classified as living in group quarters.) A housing unit is a house, an apartment, a mobile home, a group of rooms, or a single room occupied (or if vacant, intended for occupancy) as separate living quarters. Separate living quarters are those in which the occupants live separately from any other people in the building and that have direct access from the outside of the building or through a common hall. The occupants may be a single family, one person living alone, two or more families living together, or any other group of related or unrelated people who share living quarters. PAGE 35 Figure 3.9 Number of households in census tracts 0.002000.004000.006000.008000.00Total Number of Households 0100200300Number of Census Tracts 0246810Percent of Census Tracts 3.5.2 Annual household income The mean distribution of income of income across census tracts shows that the low income households defined as households with income less than 30,000$ per annum, form the largest share of households in a typical census tract. On average 32.8% of all households in a census tract are middle income households, with income in the range 30K60K per annum. An income variable used in this study is percent households with income greater than 60,000$ per annum, which averages at about a third of the household in a typical census tract. PAGE 36 Figure 3.10 Percent households in the census tracts with income greater than $60,000 0.0020.0040.0060.0080.00100.00Percent Households with Income greater than 60 K 050100150200250300Number of Census Tracts 0246810Percent of Census Tracts 3.5.3 Household size It can be observed that on an average 40 percent of the population lives in one or two person households. Another 45 percent of a typical census tract lives in 3, 4 or 5 person households. Only about 15 percent of the population constitutes those living in large households averaging sizes above 6 persons and also those living in group quarters. Figures 3.11 and 3.12 depict the distribution of two and three person households across the census tracts. PAGE 37 Figure 3.11 Percent two person households in census tracts 0.0020.0040.0060.0080.00100.00Percent Two Person Households 0100200300400500Number of Census Tracts 0246810121416Percent of Census Tracts PAGE 38 Figure 3.12 Percent three person households in census tracts 0.0010.0020.0030.00Percent Three Person Households 050100150200250300Number of Census Tracts 0246810Percent of Census Tracts 3.5.4 Vehicle ownership The table 3.7 explains the household vehicle ownership distribution by percent of total households in a census tract. From the table below it can be observed that for 91 percent of census tracts, the proportion of zero car households is less than 20 percent. Majority of the census tracts have one or two car households forming 20 to 60 percent of the total households. The distribution between the two levels (20 percent to 40 percent and 40 percent to 80 percent) seems to be equally distributed in almost half of the census Tracts. PAGE 39 Table 3.6 Vehicle distribution by percent of census tracts % Census tracts with Household car ownership % of Total households 0 1 2 3+ 0 20 91.0 4.8 6.6 86.1 20.01 40 7.2 43.3 47.8 13.9 40.01 60 1.5 46.8 43.2 0.1 60.01 80 0.3 4.8 2.3 0.0 80.01 100 0.0 0.2 0.2 0.0 Three car households have a similar distribution among the census tracts as the zero car households. 86 percent of the census tracts have three or more car households forming less than 20 percent of the entire households. Unlike the zero car households, a significant percent of census tracts (14 percent) seem to fall in the category of having 20 to 40 percent of their households with three or more cars. PAGE 40 Figure 3.13 Percent households in census tracts with two or more vehicles 0.0020.0040.0060.0080.00100.00Percent Households with Two or more Vehicles 050100150200Number of Census Tracts 0123456Percent of Census Tracts 3.6 Journey to work characteristics 3.6.1 Travel time to work Travel time is the time required to traverse between two points in space. Travel time is commonly perceived measure of understanding the ease of navigating through a roadway network. For a greater population travel time to work is of greatest consideration. Job and residential location choice are impacted by a large variety of PAGE 41 factors, many of which are related with a worker's orientation towards his/her household and leisure. Some workers show inflexibility in their travel times. Women especially seem to use commuting as a buffer between the different roles of being an employee and taking care of household members. Figure 3.7 depicts the distribution of short work trip across various census tracts in Florida. Short work trips are defined as the trip which has less than 10 minutes of travel time. It is interesting to note that the percentage of short trips is in the lowest bracket (less than 10 percent) in the census tracts surrounding urbanized areas. These might be the suburban areas from where people commute to the urbanized areas to work. Figure 3.8 shows the spatial distribution of long trips, defined as work trips with travel time more than 60 minutes, across the census tracts. It can be seen that low percent(less than 5%) of long trips exists in the census tracts very adjacent to an urbanized area. Census tracts with high percent (greater than 20%) of these trips lies in census tracts which are farthest from any urbanized area or in the farther suburban areas. Many earlier studies have shown that income effects sensitivity to travel time for work trips. The high income population has high opportunity cost of time and hence high commuting cost per mile. The sensitivity of high income population to travel time tends to be high. Shown below is the histogram for percent of workers having travel time lesser than 10 minutes and also for percent of census tracts having travel time greater than 60 minutes. PAGE 42 Figure 3.14 Percent workers whose work travel time to work is less than ten minutes 0.0020.0040.0060.0080.00100.00Percent Workers whose Travel time to work is less than 10 minutes 0100200300400500Number of Census Tracts 0246810121416Percent of Census Tracts Table 3.7 Short travel time in three person households Census tracts with Percent three person households Percent workers with travel time less than 10 minutes 020 2040 4060 6080 80100 Total 020 90 49 14 4 2 159 2040 390 209 24 3 0 626 4060 1501 348 20 2 0 1871 6080 482 10 4 1 0 497 80100 0 1 0 0 0 1 Total 2463 617 62 10 2 3154 The Table 3.9 shows that greater number of census tracts fall in the category with less than 20 percent three person households and 40 to 60 percent workers with travel PAGE 43 time less than 10 minutes. Higher the three person households lesser the population with travel time less than 10 minutes. Figure 3.15 Percent workers whose work travel time to work is greater than sixty minutes 0.0020.0040.0060.0080.00Percent workers whose Travel time to work is greater than 60 minutes 0200400600800Number of Census Tracts 0510152025Percent of Census Tracts 3.6.2 Place of work Place of work is one of the important characteristic of the journey to work at both individual level as well as at the level of census tracts. Places, for the reporting of decennial census data, include census designated places, consolidated cities, and incorporated places. PAGE 44 Each place is assigned a fivedigit Federal Information Processing Standards (FIPS) code, based on the alphabetical order of the place name within each state. If place names are duplicated within a state and they represent distinctly different areas, a separate code is assigned to each place name alphabetically by primary county in which each place is located, or if both places are in the same county, alphabetically by their legal description (for example, city before village). Percent workers working in their place of work is of prime interest in this chapter. The geographic distribution can be seen in figure 3.10. The percent averages at about 25% of the workers in a census tract. When comparing the old age population and percent workers working in place of residence, it can be observed from table 3.10 that the greater number of census tracts with higher percent of old age population have lesser percent of workers working in place. Table 3.8 Percent population greater than seventy years of age and place of work Census tracts with Percent population greater than 70 years of age Percent workers working in place 020 2040 4060 6080 80100 Total 020 1439 279 53 11 0 1782 2040 496 125 32 5 3 661 4060 311 68 11 2 0 392 6080 122 14 2 0 1 139 80100 176 3 1 0 0 180 Total 2544 489 99 18 4 3154 It can also be observed from the table 3.11 that greater share of the census tracts with higher percent of two vehicle households have lesser percent of workers working in place of residence. PAGE 45 Figure 3.16 Percent workers working in their place of residence 0.0020.0040.0060.0080.00100.00Percent Workers working in their Place of Residence 0100200300400500600Number of Census Tracts 05101520Percent of Census Tracts Table 3.9 Percent two vehicle households and place of work Census tracts with Percent two vehicle households Percent workers working in place 020 2040 4060 6080 80100 Total 020 50 758 920 55 0 1783 2040 47 349 254 11 0 661 4060 65 237 89 2 0 393 6080 25 79 31 1 2 138 80100 21 85 70 3 0 179 Total 208 1508 1364 72 2 3154 PAGE 46 Figure 3.17 Workers with work travel time less than ten minutes across Florida PAGE 47 Figure 3.18 Workers with work travel time greater than sixty minutes across Florida PAGE 48 3.6.3 SOV mode to work Single Occupancy Vehicles (SOV) continues to be the single largest mode choice for a work trip. The changes that took place in the last few decades led to a greater increase in the SOV usage. Expanding role of women in the paid labor force, reduction in family size, greater proportion of old age population are accustomed drivers. Many such factors contributed to greater auto usage in general and also SOV usage to work. It can be observed from the histogram below that greater number of census tracts have more than 60 percent of their workers using SOV mode to work. The percent averages at about 80. As it can be seen from the spatial distribution of percent workers in a census tract using SOV mode to work, the census tracts in the urban areas have a greater percent of their workers using SOV mode to work. The percent falls down a little bit in the census tracts which form the suburban areas, as people might have longer work trips and other household activities like dropping kids at school can be performed in the process. Though this proportion might be lesser than what one can find in the urban areas but still it is a very sizable proportion of the workers in the census tract (65 75 percent). PAGE 49 Figure 3.19 Percent workers working in place of residence across Florida PAGE 50 Figure 3.20 Percent workers using SOV mode to work 20.0040.0060.0080.00100.00Percent workers using SOV to work 0100200300400500Number of Census Tracts 0246810121416Percent of Census Tracts It is interesting to observe in table 3.12 that greater number of census tracts with higher percent of aged population has greater percent of SOV usage to work. PAGE 51 Figure 3.21 Percent workers using SOV mode to work across Florida PAGE 52 Table 3.10 SOV mode to work and population greater than seventy years of age Census tracts with Percent population greater than 70 years of age Percent workers using SOV mode to work 020 2040 4060 6080 80100 Total 020 13 0 1 0 0 13 2040 21 2 2 0 0 25 4060 122 23 3 0 3 151 6080 985 313 84 21 5 1408 80100 1009 421 96 25 7 1558 Total 2148 759 186 46 15 3154 Table 3.10 shows a higher SOV mode usage in census tracts with greater percent of two or more vehicle households. Table 3.11 Auto ownership and SOV usage to work Census tracts with Percent households with two or more vehicles Percent workers using SOV mode to work 020 2040 4060 6080 80100 Total 020 12 0 0 1 1 14 2040 15 6 2 0 0 23 4060 78 61 10 0 0 149 6080 78 918 405 7 0 1408 80100 25 523 947 64 1 1560 Total 208 1508 1364 72 2 3154 Table 3.12 shows that higher is the percent of females in a census tract greater is the chance for it to have a higher SOV usage to work. Table 3.12 Percent female population and SOV usage to work Census tracts with Percent females Percent workers using SOV mode to work 020 2040 4060 6080 80100 Total 020 7 1 3 0 0 11 2040 2 1 22 0 0 25 4060 0 14 131 6 0 151 6080 0 22 1375 10 0 1407 80100 1 12 1541 6 0 1560 Total 10 50 3072 22 0 3154 PAGE 53 3.6.4 Peak period departure to work Peak period travel constitutes the most demand on the capacity of the regional transportation system. Though the percent of daily traffic traveling during peak hours decreased over the past few decades, the actual volume using the roadway network during peak period increased tremendously and in many corridors the system operates at its peak capacity. It can be seen from the table 3.15 that as percent households with high income increases the number of census tracts with higher percentage of workers departing during peak period increases. Table 3.13 High income households and peak period departure Census tract with Percent households with income greater than $60,000 Percent workers departing during peak period 020 2040 4060 6080 80100 Total 020 15 1 0 1 1 18 2040 87 20 11 8 1 127 4060 1015 1134 409 119 5 2682 6080 41 127 103 51 1 323 80100 3 1 0 0 0 4 Total 1161 1283 523 179 8 3154 PAGE 54 Figure 3.22 Percent workers departing during peak period 0.0020.0040.0060.0080.00100.00Percent workers departing to work during Peak period 0100200300400500600Number of Census Tracts 05101520Percent of Census Tracts For the purpose of this thesis, peak period is identified as the two hours having the highest departures to work. In this case it is between 6 AM to 8 AM. The distribution of percent workers departing during peak hour is shown in the histogram above. The average percent of workers departing during peak period is 50 percent. Also the spatial distribution of the percentage departures is shown in the figure 3.14. PAGE 55 Figure 3.14 Percent workers departing during peak period across Florida PAGE 56 CHAPTER 4 MODELING METHODOLOGY 4.1 Background Many a researchers used structural equati ons model to analyze individuals travel behavior. These models were useful for the analysis of structural relations among the variables in a model. In a structural analysis approach, also know as causal analysis, path analysis, or simply simultaneous equations, the phenomenon under study is specified in terms of causeandeffect relationships (Golob and Meurs, 1987). The relationships are either unidirectional, that is, they each postulate that one variable influences another, or reciprocal where relati onships are specified in both directions. In this way, many structural equation models incorporate both direct and feedback influences. This chapter attempts a review of the best current practice in specifying and estimating such sophisticated models. The j ourney to work variables are the endogenous variables and the socioeconomic variables form the exogenous variables. This chapter estimates the state of the art methods in specifying and estimating such sophisticated models. AMOS 4.01 was used to estimate the models. PAGE 57 4.2 Model structure A structural equations model structure was proposed to take in to consideration the effects of socioeconomic variables on the journey to work characteristics at the level of census tracts for the state of Florida. The structure is shown in figure 4.1 Figure 4.1 Structural equations modeling framework The model system can be specified in the structural equations framework as shown: Y 1 = Y 1 11 + Y 2 21 + ..+ Y J J1 + X 1 11 + ..+ X K K1 + 1 Y 2 = Y 1 12 + Y 2 22 + ..+ Y J J2 + X 1 12 + ..+ X K K2 + 2 . . . . . . . . . . . . . Endogenous variables Endogenous (JTW) variable 3 Endogenous (JTW) variable 2 Endogenous (JTW) variable 1 Socioeconomic characteristics of census tracts Exogenous variables Y G = Y 1 1J + Y 2 2J + ..+ Y J JJ + X 1 1J + ..+ X K KG + G Where, Y = { Y 1 Y 2 .., Y J }= Limited dependent variables like travel time to work, mode used in the journey to work, peak time travel to work and place of work PAGE 58 X = {X 1 X 2 X 3 .., X K } = Socioeconomic factors such as household size, household income, vehicle ownership, age distribution, gender etc. A = { 11 , JJ }= Matrix of parameters associated with endogenous variables B = { 11 , KJ }= Matrix of parameters associated with exogenous variables = { 1, .., J }= Matrix of unobservable values of the random error components A typical structural equations model (with J endogenous variables) is defined by a matrix equation system as shown in Equation 4.1. J11...BAXY...JYY This equation can be rewritten as BXAYY (or) )(BXA)(IY1 Where Y is a column vector of endogenous variables, A is a matrix of parameters associated with endogenous variables, X is a column vector of exogenous variables, B is a matrix of parameters associated with exogenous variables, and is a column vector of error terms associated with the endogenous variables. Estimation procedures for a set of structural equations could be performed one equation at a time or the entire equation set together. The estimation where the equations are estimated one at a time when the equations are identified is called limited information estimation. One of the most common estimation is the ordinary least squares (OLS) PAGE 59 estimation. If the estimation is performed for all the structural equations together, it is called full information maximum likelihood estimation. Since the full information maximum likelihood methods consider the entire set of equations at a time for estimation, the resulting estimates are more precise. However, the estimation of full information maximum likelihood methods are computationally burdensome to estimate. 4.3 Covariancebased structural analysis In the covariance based structural analysis approach, the estimation procedure minimizes the difference between the sample covariance and covariance predicted by the model. The fundamental hypothesis for the covariancebased estimation procedures is that the covariance matrix of the observed variables is a function of a set of parameters as shown: = () where, is the population covariance matrix of observed variables, is a vector that contains the model parameters, and () is the covariance matrix written as a function of The relation of to () is basic to an understanding of identification, estimation, and assessments of model fit. The matrix () has three components, namely, the covariance matrix of Y, the covariance matrix of X with Y, and the covariance matrix of X. The implied covariance matrix of Y can be derived as: YY () = E(YY) = E['] ))(BXA))((I(BXA)(I11 PAGE 60 = (E(BXXB) + E(BX) + E(XB) + E()) 1A)I( '1A)(I = ( BB + ) 1A)I( '1A)I( Let = covariance matrix of X and = covariance matrix of The implied covariance matrix of X, XX () = E(XX) = and XY () = E(XY) = E[X 1A)(I )(BX ] = 1A)(IB' Then, it can be shown that (Bollen 1989): A)(IB'BA)(IA))(I(BBA)(I)(1111 4.4 Model identification Model identification in simultaneous structural equations systems is a mathematical problem concerned with the ability to obtain unique estimates of the structural parameters. It is associated with the question of the possibility or impossibility of obtaining meaningful estimates of the structural parameters. If an estimate of a structural parameter is in fact an estimate of that parameter and not an estimate of something else, then the parameter is said to be identified. The identification problem is typically resolved by applying restrictions on model parameters. The restrictions usually employed are zero restrictions where certain endogenous variables and certain exogenous variables do not appear in certain equations. PAGE 61 Various rules such as tRule, Null B Rule, and Recursive Rule are used for verifying identification of the whole structural model. A model is over identified when each parameter is identified and at least one parameter is overidentified. A model is exactly identified if each parameter is identified but none is overidentified. The model estimation is based on the relation of covariance matrix of observed variables and to that one with structural parameters. In case of perfect specification = (). If we consider a simple structural equation where the parameter associated with exogenous variable is set to one: y 1 = x 1 + 1 where y 1 = an endogenous variable in the first structural equation of the model x 1 = an exogenous variable in the first structural equation of the model and 1 = random disturbance associated with the first equation of the model. The covariance matrix of y 1 and x 1 is )VAR(x)y,COV(x)x,COV(y)VAR(y111111 The matrix in terms of the structural parameters is 1111111111)( At this stage of estimation we do not know either the covariances and variances or the parameters. Hence we need to arrive at the sample estimates of unknown parameters based on sample estimates of the covariance matrix. The sample covariance matrix is given by )VAR(x)y,COV(x)x,COV(y)VAR(yS111111 PAGE 62 This is made equal to the implied covariance matrix, 11 11 11 111111 and are chosen such that 11 is closest to S. When equations with more complexities exist, similar process adopted for estimating unknown parameters in A, B, and In order to achieve this objective a fitting function F(S,)( ) is defined which is minimized. The fitting function has following properties: F(S, ) is a scalar () F(S, ) >=0 () F(S, ) = 0 =S () () F(S, ) is continuous in S and () () 4.5 Maximum likelihood (ML) The fitting function that is minimized in the maximum likelihood method of estimation of structural parameters is (Bollen, 1989): FML = log  + tr (S ) log  S  (G + K) ()( 1 Where, G = Number of excluded endogenous variables on RHS of the model, and K = Number of included exogenous variables on RHS of the model. The asymptotic covariance matrix for the ML estimator is given by, 1 ML 2F E 1N 2 PAGE 63 When is substituted for an estimated asymptotic covariance matrix that allows tests of statistical significance on parameters of is obtained. 4.6 Model evaluation The F ML estimators provide test for overall model fit for overidentified models. Researchers used several ways to calibrate the match of S and Joreskog and Sorbom (1986) proposed a goodness of fit Index (GFI) and adjusted goodness of fit Index for models fitted with F ML GFI ML = ]S)tr[(]I)Str[(12121 AGFI ML = ]GFI1[2df1)K(K1ML Brown and Cudeck,1993 proposed root mean square error of approximation (RMSEA) which is a measure of compensation for the effect of model complexity. A value of the RMSEA of about 0.05 or less would indicate a close fit of the model in relation to the degrees of freedom. 4.7 Types of relationships Three different types of relationships were studied from the structural equations modeling procedures utilized in this study. They are direct effects, indirect effects and total effects. A direct effect is one in which a variable has a direct relationship or link to PAGE 64 another variable where as an indirect effect is one in wh ich a variable effects another variable through an intermediate variable. The total effect of one variable on another is the sum of its direct and indirect effects. PAGE 65 CHAPTER 5 MODEL SPECIFICATION AND RESULTS 5.1 Background The objective of this study is to explor e relationships among work travel and socioeconomic characteristics for a given fixed geographic s cale and utilize the observed relations in better understanding journey to work patterns in Florida. Towards this goal, and to appreciate the complicat ed interrelations th at exist among the variables of interest, a single model specification would not suffice. Hence, a set of five models were developed to better comprehend and compar e the interactions among the endogenous and exogenous variables. The level of geography us ed to implement these models suits the data requirements for structural equation model estimation. As mentioned previously, endogenous variab les used were travel time to work (for travel time less than ten minutes and greater than sixty mi nutes), work location, travel mode used to reach work, and peak hour proportion. Socioeconomic variables include household size, car ownership, age, gender, income, and population density taken per unit land area. This chapter describes the structural relations among the journey to work variables and socio economic va riables through a series of structural equations models. All five models developed in this st udy are explained in this section. PAGE 66 5.2 Path diagrams A path diagram is a pictorial representation of a system of simultaneous equations. Path Diagram pictures the relationships among the variable of interest. The symbols to understand path diagrams are as follows. The observed variables are enclosed in boxes. Observed Variable The error terms are enclosed in a circle n A straight arrow signifies assumption that variable at base of the arrow causes variable at the head of the arrow. A double headed arrow represents an unanalyzed association between the two variables. The n which forms subscript to the error term represents its relation to a particular endogenous variable. PAGE 67 5.3 Travel time Place of work model Travel Time Place of Work Model is specified to identify the relationships among the two journey to work characteristics, travel time (percent workers whose travel time is less than ten minutes, percent worker s whose travel time is greater than sixty minutes) and percent workers working in same place as residence. This model also explores the relation between socioeconomic characteristics of the geographic unit, which in this case is the census tract, and two journey to work variables. This structural equation system has thr ee equations. The socioeconomic variables included are population density in the censu s tract, percent female population, percent male population between the ages twenty to fifty, population with age greater than seventy, percent households w ith two and three persons, pe rcent households with annual income greater than sixty thousands and vehicle ownership variable used is percent households with two or more vehicles. 1 , 2 , and 4 are the error terms associated with percent workers with travel time less th an ten minutes, percent workers with travel time greater than sixty minutes and percent wo rkers working in the place of residence. A causal structure was prespecified based on the observed descriptive analysis of the data from the census tracts in Florida. This basic model structure was adjusted depending on 95 percent statis tical significance of the causal relations previously assumed. New relations were tried till a perfec t fit was obtained. This model is depicted in figure 5.1. PAGE 68 Figure 5.1 Travel time Place of work model Percent workers working in place Percent Workers with Travel time less than 10 min 1 2 3 Percent Workers with Travel time greater than 60 min Percent two or more vehicles in households Percent Households with Income greater than 60 K Percent three person households Percent Males between ages 25 and 50 y ears Percent two person households Percent Population with age greater than 70 years Percent Females Total Population Density PAGE 69 5.4 Trip mode Place of work model The journey to work characteristics whose relationships were explored with socioeconomic variables are percent workers using single occupanc y vehicles (SOV) to work and percent workers working in the same place as their residence. This model is graphically depicted in the Figure 5.2 This two structural equation system has as its exogenous variables, average population density by area of census tract (sq. m ile), gender distribution of females, and percent population with age less than twenty five years, percent households with size two, percent households with annual income gr eater than hundred thousands and percent households with two or more vehicles. 1 and 2 are the error terms corresponding to percent workers using SOV to go to work and percent people working in place of their residence respectively. PAGE 70 Figure 5.2 Trip mode Place of work model Percent Workers using SOV mode to wor k Percent two or more vehicles in households Percent Households with Income greater than 100 K Percent three person households Total Population Density Percent population with less than 25 y ears of a g e Percent two person households Percent Females Percent workers working in place 2 1 PAGE 71 5.5 Travel time Trip mode model This model specifies statistically signi ficant relations between two journey to work variables representing travel time and mode of travel for work trips. Travel time is represented in the model framework by percen t workers having travel time less than ten minutes. Travel mode is represented by the pe rcent workers using SOV for work trip in the census tract. The exogenous variables used in this two equation structural equation model include all the variables that are used in Travel Time Place of Work Model (i.e.) population density, female population percenta ge, percent of male population between ages twenty five and fifty, percent popula tion above the age of seventy, percent two person and three person households, percent hous eholds with income greater than sixty thousand, percent households w ith two or more vehicles. The error terms for endogenous variables, percent workers whose travel time to work is less than ten minutes and per cent workers who use SOV to work are 1 , and 2 respectively. This model is depicted in figure 5.3. PAGE 72 Figure 5.3 Travel time Trip mode model Percent workers using SOV mode to work Percent Workers with Travel time less than 10 min 1 2 Percent two or more vehicles in households Percent Households with Income greater than 60 K Percent three person households Percent two person households Percent Population with age greater than 70 years Total Population Density Percent Males between ages 25 and 50 y ears Percent Females PAGE 73 5.6 Trip mode Peak period model This model was estimated with its endoge nous variables as pe rcent workers using SOV to work and percent workers traveling during peak period to work. This model is like previous models A, Trip Mode Place of Work Model, Travel Time Trip Mode Model is a two equation structural model. All the socioeconomic variables used in Travel Time Trip Mode Model are used as the exogenous variables in this model. This is the last of the two equation structural model in the set. Work travel mode is depicted by the one of the endogenous variables, percent of workers using SOV mode to wor k. Peak period travel is highes t percent of travel in a two hour period. This period is identified as between 7:00 AM and 9:00 AM. So the percent of workers starting their work trip between 7:00 AM and 9:00 AM constitute the peak hour travel. Same set of socioeconomic variables that are used in Trip Mode Place of Work Model and Travel Time Trip Mode Model were used as exogenous variables. This model is depicted in Figure 5.4 PAGE 74 Figure 5.4 Trip mode and Peak period model Total Population Density Percent Females Percent Population with age greater than 70 years Percent two person households Percent Males between ages 25 and 50 y ears Percent workers departing during peak period Percent three person households Percent Households with Income greater than 60 K Percent two or more vehicles in households 2 1 Percent Workers using SOV mode to wor k PAGE 75 5.7 Travel time Trip mode Peak period Work location model In order to explain the re lation between work location and peak period travel and work trip travel time and peak period travel two models were used. Travel Time Trip Mode Work Location Peak Period Model has percent workers whose work trip travel time is less than ten minutes, percent workers w hose work trip travel time is greater than sixty minutes, percent workers using SOV to work, percent working in location of their residence and percent with peak period travel. The exogenous variables used in this mode l are same as the ones used in Travel Time Trip Mode Model and Trip Mode a nd Peak Period Model. Travel Time Trip Mode Work Location Peak Period Model is a five structural equation system. Travel Time Trip Mode Work Location Peak Period Model is depicted in Figure 5.5. PAGE 76 Figure 5.5 Travel time Trip mode Work location Peak period model Percent workers using SOV mode to work Percent Workers with Travel time less than 10 min 1 2 Percent Workers with Travel time greater than 60 min Percent Males between ages 25 and 50 y ears Percent workers working in place Percent Females Percent Population with age greater than 70 years Percent two or more vehicles in households Percent Households with Income greater than 60 K Percent three person households Percent two person households Percent workers departing during peak period 3 4 5 PAGE 77 Table 5.1 Travel time Place of work model results Endogenous Variable Intercept Effect HH with two or more vehicles Two person HH Population with age > 70 years HH with income > 60,000 per annum Three person HH Females Population density Workers working in place Male population with 25 < age < 50 years Workers 45.341 Total 0.856 0.148 0.709 0.392 0.310 0.638 0.000 0.000 0.000 working in Direct 0.856 0.148 0.709 0.392 0.310 0.638 0.000 0.000 0.000 Place Indirect 0.000 0. 000 0.000 0. 000 0.000 0. 000 0.000 0. 000 0.000 Travel time 38.635 Total 0.033 0.089 0.000 0.053 0.012 0.245 0.000 0.038 0.112 greater than 60 Direct 0.000 0.095 0.027 0.037 0.000 0.221 0.000 0.038 0.112 minutes Indirect 0.033 0.006 0.027 0.015 0.012 0.025 0.000 0.000 0.000 Travel time 29.196 Total 0.099 0.017 0. 082 0.003 0.652 0.204 0.001 0.115 0.000 less than Direct 0.000 0.000 0.000 0.048 0.616 0.278 0.000 0.115 0.000 10 minutes Indirect 0.099 0.017 0.082 0.045 0.036 0.073 0.000 0.000 0.000 Note: N = 3154 ChiSquared = 15.813 with df =11; pvalue = 0.148; CFI = 1; RMSEA = 0.012 All Variables Significant at 95% level All Variables are in Per centage excluding population density PAGE 78 Table 5.2 Trip mode Place of work model results Endogenous Variable Intercept Effect Households with two or more vehicles Households with income greater than 100 K per annum Three person households Two person households Population density Females Workers working in place Population with age less than 25 years Workers 62.115 Total 0.645 0.480 0.376 0.405 0.000 0.000 0.000 0.000 working in Direct 0.645 0.480 0.376 0.405 0.000 0.000 0.000 0.000 Place Indirect 0.000 0.000 0.000 0.000 0.000 0.00 0 0.000 0.000 Workers 33.209 Total 0.434 0.153 0.131 0.055 0.000 0.642 0.030 0.862 using SOV Direct 0.414 0.139 0.143 0.043 0.000 0.642 0.030 0.862 mode to work Indirect 0.020 0.015 0.011 0.012 0.000 0.000 0.000 0.000 Note: N = 3154; ChiSquared = 3.171 with df = 4; pvalue = 0.530; CFI = 1; RMSEA = 0.000 All Variables Significant at 95% level All Variables are in Percenta ge excluding population density PAGE 79 Table 5.3 Travel time Trip mode model results Endogenous Variable Intercept Effect HH with two or more vehicles HH with income > 60,000 per annum Two person HH Population with age greater than 70 years Male population between ages 25 and 50 years Three person HH Females Population density Workers using SOV mode to work Workers 18.040 Total 0.410 0.063 0.278 0.405 0.494 0.201 0.790 0.000 0.000 using SOV Direct 0.410 0.063 0.278 0.405 0.494 0.201 0.790 0.000 0.000 mode to work Indirect 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Travel time 49.988 Total 0.099 0.015 0.067 0.098 0.119 0.658 0.296 0.001 0.241 less than Direct 0.000 0.000 0.000 0.000 0.000 0.609 0.106 0.001 0.241 10 minutes Indirect 0.099 0.015 0.067 0.098 0.119 0.048 0.190 0.000 0.000 Note: N = 3154; ChiSquared = 11.855 with df = 7; pvalue = 0.105; CFI = 1; RMSEA = 0.015 All Variables Significant at 95% level All Variables are in Percenta ge excluding population density PAGE 80 Table 5.4 Trip mode Peak period model results Endogenous Variable Intercept Effect Females HH with two or more vehicles HH with income > 60000 per annum Three person HH Two person HH Population age > 70 years Male population between ages 25 and 50 years Population density Workers using SOV mode to work Workers 18.200 Total 0.788 0.408 0.063 0.209 0.280 0.406 0.499 0.000 0.000 using SOV Direct 0.788 0.408 0.063 0.209 0.280 0.406 0.499 0.000 0.000 mode to work Indirect 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Workers 18.780 Total 0.225 0.059 0.072 0.327 0.080 0.227 0.314 0.000 0.285 departing in Direct 0.000 0.057 0.090 0.268 0.000 0.111 0.171 0.000 0.285 Peak period Indirect 0.225 0.116 0.018 0.060 0.080 0.116 0.142 0.000 0.000 Note: N = 3154; ChiSquared = 7.045 with df = 5; pvalue = 0.217; CFI = 1; RMSEA = 0.011 All Variables Significant at 95% level All Variables are in Per centage excluding population density PAGE 81 Table 5.5 Travel time Trip mode Work location Peak period model results Endogenous Variable Intercept Effect HH with 2 or more vehicles Two person HH Population age > 70 years HH with income > 60 K Three person HH Females Population density Workers working in place Male 25< age < 50 years SOV mode to work Workers 44.895 Total 0.856 0.147 0.709 0.391 0.316 0.645 0.000 0.000 0.000 0.000 working in Direct 0.856 0.147 0.709 0.391 0.316 0.645 0.000 0.000 0.000 0.000 Place Indirect 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Workers 17.149 Total 0.406 0.286 0.403 0.061 0.211 0.792 0.000 0.031 0.498 0.000 using SOV Direct 0.379 0.281 0.381 0.049 0.202 0.812 0.000 0.031 0.498 0.000 mode to work Indirect 0.026 0.005 0.022 0.012 0.010 0.02 0.000 0.000 0.000 0.000 Travel time 42.284 Total 0.033 0.085 0.000 0.055 0.012 0.239 0.000 0.039 0.108 0.000 > 60 Direct 0.000 0.090 0.027 0.040 0.000 0.214 0.000 0.039 0.108 0.000 minutes Indirect 0.033 0.006 0.027 0.015 0.012 0.025 0.000 0.000 0.000 0.000 Travel time 28.607 Total 0.129 0.042 0.113 0.020 0.653 0.224 0.001 0.111 0.045 0.091 less than Direct 0.000 0.000 0.000 0.028 0.600 0.222 0.001 0.108 0.000 0.091 10 minutes Indirect 0.129 0.042 0.113 0.048 0.053 0.002 0.000 0.003 0.045 0.000 Workers 36.183 Total 0.066 0.110 0.214 0.063 0.326 0.291 0.000 0.010 0.329 0.150 departing in Direct 0.000 0.000 0.129 0.043 0.244 0.000 0.000 0.026 0.174 0.143 Peak period Indirect 0.066 0.110 0.085 0.02 0.082 0.291 0.000 0.015 0.154 0.007 Note: N = 3154; ChiSquared = 10.114 with df =13; pvalue = 0685; CFI = 1; RMSEA = 0.000 All Variables Significant at 95% level All Variables are in Percentage except population density PAGE 82 Table 5.5 Continued Endogenous Variable Intercept Effect Travel time > 60 minutes Travel time < 10 minutes Workers 44.895 Total 0.000 0.000 working in Direct 0.000 0.000 Place Indirect 0.000 0.000 Workers 17.149 Total 0.000 0.000 using SOV Direct 0.000 0.000 mode to work Indirect 0.000 0.000 Travel time 42.284 Total 0.000 0.000 greater than Direct 0.000 0.000 60 minutes Indirect 0.000 0.000 Travel time 28.607 Total 0.000 0.000 less than 10 Direct 0.000 0.000 minutes Indirect 0.000 0.000 Workers 36.183 Total 0.737 0.081 departing in Direct 0.737 0.081 Peak period Indirect 0.000 0.000 PAGE 83 CHAPTER 6 MODEL RESULTS INTERPRETATION This Chapter explains the relations obtained by using structural equations analysis using journey to work characteristics as endogenous variables and sociodemographic characteristics as exogenous variables. All these relations are obtained at an aggregate level of census tracts. It should also be noted that the data used through the course of this thesis is primarily from the Census Summary File 3 which does not include information on other traditionally observed significant factors effecting journey to work characteristics. 6.1 Travel time to work and household income In the models developed, variable used to represent high income population in a census tract is the percent households with income greater than sixty thousand per annum. Two variables were used to represent short and long trip travel times. They are percent workers with work trip travel time greater than sixty minutes and percent workers with work trip travel time less than ten minutes. It has been observed from the estimation of models that percent households with annual income greater than sixty thousands has a negative influence on percent workers having sixty minutes or more travel time to work. This implies that high percent of households in a census tract with income greater than sixty thousands cause the census PAGE 84 tract to have lesser percent of workers having their work travel time greater than sixty minutes. As can be observed from the figure 3.15, distribution of percent workers whose travel time to work is less than 60 minutes, the percent of workers with travel time greater than 60 minutes is a small percentage averaging around 8 percent of the total number of workers, across all the census tracts. This consistently low value of the travel time variable itself might draw a negative relation towards percent households with high income. 6.2 Travel time to work and household size Model estimates for household size indicate that percent households with three persons have a positive effect on percent workers who travel more than sixty minutes to work. This means that greater the percent of three person households in a census tract greater is the percent of workers having long commutes to work. It should be noted that it is an indirect positive effect between these two variables. 6.3 Travel time to work and old age population The variable used for representing the old age population proportion in the model was percent population whose age was greater than seventy years of age. Usually, population of this age constitutes retirees, people with health conditions, people who do not have significant contribution in the work force. The results of the models showed that percent population with age greater than 70 had a negative effect on percent workers with travel time less than ten minutes. This implies that census tracts with higher percent of old aged population (age greater than seventy PAGE 85 years) has lesser percent of its worker population having less than ten minutes of travel time to work. The negative effect of percent population with age greater than 70 on percent workers having less than 10 minutes of travel time is an indirect effect. 6.4 Travel time to work and vehicles in household The variable, percent household with two or more vehicles is taken to represent auto ownership. Model estimates show that percent household with two or more vehicles variable has a negative effect on percent workers whose travel time is less than ten minutes. At the same time, percent households with two or more vehicles have a positive effect on the percent workers whose travel time to work is greater than sixty minutes. Both the effects observed are indirect effects. 6.5 Travel time to work and population density Population density is defined as number of people living per unit land area. In this case it has been defined as population per square mile averaged over the census tract. The estimates in the models show that population density has a negative effect on percent workers whose travel time is less than ten minutes. Due to very high values of density in the data set and low values of percent workers whose travel time to work is less than 10 minutes; such pronounced difference might show a negative effect on percent workers with less than 10 minutes travel time. PAGE 86 6.6 Travel time to work and percent female population A number of studies have shown that women have significantly different travel patterns than men. Women tend to have shorter average trip lengths. Women account for roughly twothirds of the new entrants into the labor force in the last twenty years, and rising female labor force participation rates account for a substantial portion of the overall growth in travel and automobile use. Womens householdserving travel patterns appear to be a function of both socialization and the sexual division of household responsibilities. The estimates in the model show that percentage female population in the census tract has a negative effect on percent workers whose work trip travel time is greater than sixty. Interestingly, the estimates also show that greater female proportion also has a negative effect on percent work trips with travel time less than 10 minutes. Again, due to a consistently high value of percent female population and consistently low values for percent workers whose travel timeare less than 10 minutes the negative effect might have shown up. 6.7 Travel time to work and working male population The model estimates indicate that the percent males in the age group 25 to 50 have a negative effect on percent long work trips. The magnitude of this effect perceived is very small compared to magnitudes of other relations. Especially, with percent workers having travel time greater than 60 minutes having very low magnitudes, this relation needs to be considered in comparison with other relations. PAGE 87 6.8 Place of work and household income From the model results it can be understood that percent high income households (defined as households with annual income greater than 60 K) tend to have a positive direct effect on percent of workers working in same place as their residence. Higher the percent of households with income greater than 60 K, greater would be the percent of workers working in their place of residence. 6.9 Place of work and household size The estimation results showed that percent two person households in a census tract has a negative effect on percent workers working in their place of residence. Similar results are shown by three person households. The low magnitude of the endogenous variables must be watched for before concluding the negative effect. 6.10 Place of work and aged population Model estimates show a negative direct effect of percent aged population on the percent workers working in place. Higher percent of aged population means a lower percents of workers working in the place of their residence. 6.11 Place of work and percent female population Higher percent of females in a census tract has a strong positive direct effect on the percent workers working in the place of their residence. Higher the percent females in census tracts, higher are the percent of workers working in the place of their residence. PAGE 88 6.12 Place of work and vehicles in household Percent households with two vehicles have a negative effect on percent workers working in place. This relation is also a negative direct relation. Higher two vehicle households, lesser are the percent workers working in the place of residence. 6.13 Travel mode to work and household income The estimates from the model indicate that percent households with annual income greater than sixty thousands and percent households with income greater than hundred thousands both have a negative effect on percent workers using SOV to work in a census tract. This relation is contradictory to general observation of finding greater percent of SOV users to work, with increased percent of high income households in a region. This relation needs further exploration. 6.14 Travel mode to work and household size The estimation results show that percent two person households have a positive effect on percent workers using SOV mode to work. Though, estimation results for percent households with three person shows a positive effect on percent workers using SOV to work, it has a lesser magnitude than the effect for a two person household 6.15 Travel mode to work and aged population It is observed from the model estimates that percent population above seventy years has a positive effect on the percent of workers using SOV mode to work. This relation has a complete direct effect. PAGE 89 6.16 Travel mode to work and vehicles in household The model estimates show that percent households with two vehicles have positive effect on percent workers using SOV mode to work. This implies that greater the percent of two vehicle households in a census tract, higher are the percent of workers that use SOV mode to work. 6.17 Travel mode to work and percent female population Many studies in the past have shown that women are more likely to travel to work in a SOV than men. It is found in this analysis that percent female population has a positive effect on percent workers using SOV mode to work. This implies that greater the percent of females, greater is the percent workers using SOV mode to work in the census tract. 6.18 Travel mode to work and working male population Working male population is taken as the male population between ages 25 and 50. It was observed that percent male population between the ages 25 and 50 has a positive effect on percent workers using SOV mode to work. 6.19 Travel mode to work and percent population less than twenty five years of age Percent population of age less than 25 years of age has a negative effect on percent workers using SOV mode to work. PAGE 90 6.20 Peak period departure and household income It has been observed that households with income greater than 60 K annually has a positive effect on percent workers departing during peak period to work. 6.21 Peak period departure and household size The estimation results show that as the percent of three person households increases the percent workers departing during the peak period increases. 6.22 Peak period departure and male population between the ages twenty five and fifty It is observed from the model estimates that male population between the ages of 25 and 50 years have a positive effect on the percent worker departures during peak period. PAGE 91 CHAPTER 7 CONCLUSIONS 7.1 Background This research builds on the recent developments in the utilization of structural equations modeling to identify effects of so cioeconomic variables on the journey to work characteristics at an aggregate level. The Census 2000 data was utilized at an aggregate level to perform the analysis. Es timation was performed using the 3154 census tracts in Florida as units of analysis. Equa tions were developed using journey to work characteristics as endogenous variables and socioeconomic characteristics as exogenous variables. The journey to work variables incl ude: travel period to work (less than 10 minutes and greater than 60 minut es) percent values for less than 10 minutes and greater than 60 minutes are considered, SOV mode to work percent value for mode chosen for work trip, place of work person working in place of residence, peak period of departure workers departed to work in peak period (two hour period). Causal relations between socioeconomic and endogenous variables and between endogenous variables were studied. This method of analysis of census data at an aggregate level has a significant advantage in terms of convenience of development when compared to other data intensive procedures for describing the same journey to work behavior. Another significant advantage of using aggregate census data is that, it is free from the PAGE 92 complexities of sampling a given region for obs erving travel behavioral characteristics. Since sampling for Census data is done on a national level, data obtained at various geographies can be compared to observe local trends or regional vari ation in any of the observable travel behavior characteristic. 7.2 Summary of findings This research effort aims at a compre hensive understanding of the journey to work characteristics in the st ate of Florida. Structural eq uations modeling seems to be a very effective means for simultaneously representing multiple causal relationships among the endogenous variables in the presence of mu ltiple error covariances. A set of five models were used to arrive at these results. 7.2.1 Relations between socioeconomic a nd journey to work characteristics The estimation results obtained show th e relationships betw een the aggregate socioeconomic variables and journey to work characteristics at an aggregate level. Though there are several effects that were iden tified through this analysis, some of the results need to be further explored or can not be generalized and conclusions drawn based on them. Percent workers having a short or long travel time are very small in magnitude compared to other variables in the analysis and some of the results obtained in this analysis are counter in tuitive. Similarly, the high value of population density compared to all other values in the analys is also cause wrong effects to be found in the analysis. Some of the relations obtained in the analysis between the journey to work variables and socioeconomic characteristics of the census tracts include: percent females PAGE 93 in a census tract seem to have a negative eff ect on the percent work ers whose travel time to work is greater than sixty minutes. Per cent high income households had a positive direct effect on percent worker s working in place of reside nce. Percentages of both two and three person households had a negative e ffect on percent workers working in the place of their residence. Vary low percenta ges of the endogenous variables should be taken in to consideration before any conc lusion can be drawn from the relationships. Percent female population in a census tract ha d positive effect on the workers working in the place of their residence. Percent households with income greater than 60 K per annum seem to have a negative effect on percen t workers using SOV mode to work. This relation needs to be further examined. Such relationships were observed between other endogenous and exogenous variables. 7.2.2 Relations among journey to work characteristics The relationships obtained between various journey to work characteristics were also obtained in the analysis. Percent wo rkers working in place of residence has a negative effect on percent workers with travel time greater than 60 minutes to work. The low values of percent workers having travel time greater than 60 minutes need to be noted. Percent workers working in place also had a negative effect on percent workers using the SOV mode to work and percen t workers departing during peak period. Similarly percent workers who had travel time less than 10 mi nutes had a negative effect on percent workers departing in peak period. It is also observed that percent workers who have their travel time greater than 60 mi nutes had negative effect on percent workers departing during peak period. It is also obs erved that percent workers departing using PAGE 94 SOV mode to work had negative effect on percent workers having their travel time less than 10 minutes and a positive effect on per cent workers using SOV mode to work. Some of the relationships observed are to be furthe r analyzed with a broader range of values and also in the presence of other traditionally observed significant factors relating to journey to work characteristics and to avoid pitfalls in dr awing conclusive relationships. 7.3 Omitted variables In the present analysis of the journey to work characteristics, the exogenous variables considered for the analysis are the prevailing sociodemogra phics in the census tracts. Just the sociodem ographic variables do not encomp ass the complete list of factors that might significantly effect the jour ney to work characteris tics. Incorporation of some other characteristics representative of the prevailing land use would better the estimates. Network mobility and areawid e congestion factors also need to be incorporated in the analysis. Employment locations relative to the resi dential locations is a strong measure in determining journey to work characteristics of a census tract. These variables if suitable used along with the sociodemographic variab les would provide better estimates of the relationships. 7.4 Conclusions and future research directions Structural equations methodology (SEM) has been successfully applied to analyze the journey to work behavior using aggregate Census data. The results of these models are at census tract level wh ich could be aggregated to Counties and States. Unlike PAGE 95 disaggregate level models; the models in this thesis do not need any additional investment in resources for conducting surveys etc. For a general understanding of the journey to work beha vior this research gives very usable and considerably accurate relati ons. These models if applied and tested against other disaggregate modeling techniqu es and various geogra phies, would form a very powerful and quick tool to understand ove rall relationships in journey to work data. Also, incorporation of other important aggregate factors lik e network density, areawide network level of service characteristics, employment totals would definitely improve the estimates. In conclusion, there still exists a great ne ed for models that could serve the users with limited resources. There lies a tremendous scope in refining the models developed in this thesis and bringing out more readily usab le models that could deliver solutions with lesser data input. PAGE 96 REFERENCES Arbuckle, J.L., AMOS Users Guide Version 4.0., 2000 SmallWaters Corporation, Chicago, IL. Becker, G.S., 1965. A Theory of the Allocation of Time. The Economic Journal, Vol. 75, No. 299, 493517. Bhat, C.R., Guo, J.A., 2004. Mixed spatially correlated logit model: formulation and application to residential choice modeling. Transportation Research B, Vol. 38, Issue 4, 147168. Bhat, C.R, Singh, S.K., 2000. A comprehensive daily activitytravel generation model system for workers. Transportation Research A, Vol. 34, Issue 1, 122. Bollen, K.A., 1989. Structural Equations with Latent Variables. Wiley, New York. Browne, M.W., 1974. Generalized least squares estimators in the analysis of covariance structures. South African Statistical Journal, 8: 124. (Reprinted in D.J. Aigner and A.S. Goldberger (eds.), 1977. Latent Variables in SocioEconomic Models. Amsterdam: NorthHolland. Browne, M.W., Cudeck, R., 1993. Alternative way of assessing model fit. In Bollen, K.A. and Long, J.S. Testing structural equation models. Newbury Park, California: Sage, 136162. Brownstone D., Ghosh A., Golob, T.F., Kazim, C. and Amelsfort, D.A., 2003. Drivers willingnesstopay to reduce travel time: evidence from San Diego I15 congestion pricing project. Transportation Research A, Vol. 37, 373387. Cervero, R., 2002. Built environment and mode choice: toward a normative framework. Transportation Research D, Vol. 7, 265284. Clark, W.A.V., Huang, Y. and Withers, S., 2003. Does commuting distance matter? Commuting tolerance and residential change. Regional Science and Urabn Economics, Vol. 33, 199221. Collia, D.V., Sharp, J. and Giesbrecht, L., 2003. The 2001 national household travel survey: a look into the travel patterns of older Americans. Journal of Safety research, Vol. 34, 461470. PAGE 97 Conquest, L., Spyridakis, J., Heselkorn, M. and Barfield, W., 1993. The effect of motorist information on commuter behavior: Classification of drivers in commuter group. Transportation Research C, Vol. 1, 183201. Crane, R., 1996. The influence of uncertain job location on urban form and the journey to work. Journal of Urban Economics, Vol. 39, 342356. Dobson, R. F., Dunbar, C.J., Smith, D., Reibstein and Lovelock. C., 1978. Structural models for the analysis of traveler attitudebehavior relationships. Transportation, 7, 351363. Fujii, S., Kitamura, R., 2000. Evaluation of Tripinducing Effects of New Freeways Using a Structural Equations Model System of Commuters' Time use and Travel, Transportation Research B, Vol. 34, 339354. Golob, T.F., Zondag, E.R., 1984. A causal model of mobility. Presented at the Tenth Transportation Planning Research Colloquium. Zandvoort, the Netherlands. Golob, T.F., Meurs H., 1987. A structural model of temporal changes in multimodal travel demand. Transportation Research, 21A, 391400. Golob, T.F., van Wissen L.J., 1989. A joint household travel distance generation and car ownership model. Transportation Research, 23B, 471491. Golob, T.F., Kitamura R., Lula C., 1994. Modeling the Effects of Commuting Time on Activity Duration and NonWork Travel. Presented at Annual Meeting of Transportation Research Board, Washington, D.C. Golob, T.F., Bunch D.S., Brownstone D., 1997. A vehicle usage forecasting model based on revealed and stated vehicle type choice and utilization data. Journal of Transport Economics and Policy. Hederickson, C., Plans, E., 1984. The flexibility of departure for work trips. Transportation Research A, Vol. 18, Issue 1, 2536. Johnston, J., DiNardo, J., 1997. Econometric Methods. Fourth Edition. Wiley, New York. Kalfs, N., Saris., W.E., 1997. New Data collection methods in Travel Surveys. ActivityBased Approaches to Travel Analysis. Elsevier Science Ltd., U.K. Kline, R.B., 1998. Software programs for structural equation modeling: AMOS, EQS, and LISREL. Journal of Psychoeducational Assessment, 16: 302323. PAGE 98 Kuppam, A.K., 1999. A Limited Dependent Variable Structural Equations Model of Travel Behavior for Transportation Policy Evaluation. PhD Dissertation, Department of Civil and Environmental Engineering, University of South Florida, Tampa, FL. Kuppam, A.R., Pendyala, R.M., 2000. An exploratory analysis of commutersactivity and travel patterns. Presented at the 79h Annual Meeting of the Transportation research Board, Jan. 913, Washington, DC. Levinson, D.M., 1998. Accessibility and journey to work. Journal of Transport Geography, Vol. 6, 1121. Lu, X., E.I. Pas, 1997. A structural equations model of the relationships among sociodemographics, activity participation and travel behavior. Presented at 76th Annual Meeting of the Transportation Research Board, January 1216, Washington, DC. Lu, X., Pas, E.I., 1999. Sociodemographics, Activity Participation and Travel Behaviour, Transportation Research A, Vol. 33, 118. Pas, E.I., 1996. Recent advances in activitybased travel demand modeling. Presented at the TMIP Conference on ActivityBased Travel Forecasting, New Orleans, LA, 25. Pas, E., Harvey. A.S., 1997. Understanding travel behavior in an era of change. (Eds.) Peter R. Stopher.and Martin LeeGosselin. New York: Pergamon Press. RDC, Inc. ActivityBased Modeling System for Travel Demand Forecasting. TMIP1995. Sakano, R. and J. Benjamin, 2001. A structural equation analysis of revealed and stated travel mode and activity choices. Presented at the 80h Annual Meeting of the Transportation research Board, Jan. 711, Washington, DC. Schwanen, T. and Dijst, M., 2002. Traveltime ratios for visits to the workplace: the relationship between commuting time and work duration. Transportation Research A, Vol. 36, 573592. Sermons, M.W. and Koppelman, F.S., 2001. Representing the difference between female and male commute behavior in residential location choice models, Journal of Transport Geography, Vol. 9, 101110. Tardiff, T.J., 1976. Causal inferences involving transportation attitudes and behavior. Transportation Research, 11: 397404. 