<%BANNER%>

Intra-Household Interactions in Social-Recreational Activities and Travel

Permanent Link: http://ufdc.ufl.edu/UFE0044974/00001

Material Information

Title: Intra-Household Interactions in Social-Recreational Activities and Travel
Physical Description: 1 online resource (171 p.)
Language: english
Creator: Lim, Kwangkyun
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: activity-based -- interaction -- intra-household
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre: Civil Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This research explains interactions among household membersduring daily activity- and travel-related decision-making process with focus onsocial-recreational activities.  Withinthe broad spectrum of modeling social-recreational activities and travel, thisstudy focuses on three major aspects: Modeling the Generation ofSocial-Recreational Patterns, Modeling the Choice of Household Vehicle for Social-RecreationalTours, and Modeling the Choice of Time of Day for Joint Social-RecreationalTours. This study uses data from the 2009 National Household Travel Surveyconducted in the United States for the estimation of all models. The empiricalscope of this work is largely restricted to the analysis of the behavior ofcouple households (two-adults comprising a male-female couple with or withoutchildren). In the context of the generation of social-recreationalactivity patterns, interpersonal interactions within a household are largely inthe form of trade-offs between solo and joint activity participation decisions.This research compares fourmethods, depending on different decision-making units, that are capable ofcapturing such interactions between household heads: multinomial logit models,multilinear logit models, parallel choice constrained logit models, andtri-variate binary probit models. We compared the models in terms of predictiveabilities and sensitivities to specific explanatory factors. The next aspect of the research is the modeling of household vehicle choice. Two modelcomponents are developed that can be applied within any operationalactivity-based modeling framework. One allocates each vehicle to a primarydriver in the household (long-term, household-level model). The second modelallocates a vehicle for the joint tours (short-term, tour-level model). Both modelswere estimated using the unlabeled binary-logit approach. This study is furtherlimited to households with two-cars. The final component of this research is on the choice of time-of-day for joint tours. Jointtravel requires the temporal synchronization of travel of all members of thetravel party. This study describes how the time constraints of multiple personscan be effectively accommodated into determining the choice set and,subsequently, the choice of timing of joint tours. In this study, an alternatetwo-step approach is proposed. The first model predicts the time-window chosenfor pursuing the joint discretionary tour and the second model locates the tourwithin the time window by determining the start- and end-times of the toursimultaneously on a continuous scale. Overallthis researchcontributes to understanding of the social-recreational activity-travelpatterns and presents methods to effectively represent these choices within aconventional activity-based modeling framework.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Kwangkyun Lim.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: Srinivasan, Sivaramakrishnan.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0044974:00001

Permanent Link: http://ufdc.ufl.edu/UFE0044974/00001

Material Information

Title: Intra-Household Interactions in Social-Recreational Activities and Travel
Physical Description: 1 online resource (171 p.)
Language: english
Creator: Lim, Kwangkyun
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: activity-based -- interaction -- intra-household
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre: Civil Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This research explains interactions among household membersduring daily activity- and travel-related decision-making process with focus onsocial-recreational activities.  Withinthe broad spectrum of modeling social-recreational activities and travel, thisstudy focuses on three major aspects: Modeling the Generation ofSocial-Recreational Patterns, Modeling the Choice of Household Vehicle for Social-RecreationalTours, and Modeling the Choice of Time of Day for Joint Social-RecreationalTours. This study uses data from the 2009 National Household Travel Surveyconducted in the United States for the estimation of all models. The empiricalscope of this work is largely restricted to the analysis of the behavior ofcouple households (two-adults comprising a male-female couple with or withoutchildren). In the context of the generation of social-recreationalactivity patterns, interpersonal interactions within a household are largely inthe form of trade-offs between solo and joint activity participation decisions.This research compares fourmethods, depending on different decision-making units, that are capable ofcapturing such interactions between household heads: multinomial logit models,multilinear logit models, parallel choice constrained logit models, andtri-variate binary probit models. We compared the models in terms of predictiveabilities and sensitivities to specific explanatory factors. The next aspect of the research is the modeling of household vehicle choice. Two modelcomponents are developed that can be applied within any operationalactivity-based modeling framework. One allocates each vehicle to a primarydriver in the household (long-term, household-level model). The second modelallocates a vehicle for the joint tours (short-term, tour-level model). Both modelswere estimated using the unlabeled binary-logit approach. This study is furtherlimited to households with two-cars. The final component of this research is on the choice of time-of-day for joint tours. Jointtravel requires the temporal synchronization of travel of all members of thetravel party. This study describes how the time constraints of multiple personscan be effectively accommodated into determining the choice set and,subsequently, the choice of timing of joint tours. In this study, an alternatetwo-step approach is proposed. The first model predicts the time-window chosenfor pursuing the joint discretionary tour and the second model locates the tourwithin the time window by determining the start- and end-times of the toursimultaneously on a continuous scale. Overallthis researchcontributes to understanding of the social-recreational activity-travelpatterns and presents methods to effectively represent these choices within aconventional activity-based modeling framework.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Kwangkyun Lim.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: Srinivasan, Sivaramakrishnan.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0044974:00001


This item has the following downloads:


Full Text

PAGE 1

1 INTRA HOUSEHOLD INTERACTIONS IN SOCIAL RECREATIONAL ACTIVITIES AND TRAVEL By KWANGKYUN LIM A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012

PAGE 2

2 2012 K wangkyun Lim

PAGE 3

3 To my l ovely w ife Soonmi

PAGE 4

4 ACKNOWLEDGMENTS This work would not have been possible without my long suffering wife, Soon m i Roh for being a loving, steady supports for me I would also like to acknowledge my grandmother in law, Hyun s oo Han, who devoted her life for tak ing care of my kids for last 7 years The same appreciation also goes to my mom for her continuous love and supports through my life. My three lovely kid s S iwon, Chown, and Soup April were another influence of making this work accomplishable I am most indebted to the patient mentoring and supporting of my advisor Dr. Sivaramakrishnan Srinivasan, Associate Professor, Department of Civil Engineering He has been always a great source of exploring intellectual knowledge throughout the period of my study at The University of Florida and I would not have done the accomplishment without him. I would also like to thank my dissertation committee members, Dr. Lily Elefteriadou, Dr. Scott Washburn, Dr. Yafeng Yin, and Dr. Ruth Steiner for their critical comments to enrich this dissertation. Another big appreciation goes to Dr. Sigon Kim, Professor at Seoul National University of Technology, for his continuous parent al guidance in exploring right directions. Finally, I wish to thank my friends and fellow students at UF for making school life very memorable I would represent my big thanks to them: George Debra, Nagendra Darhka, Vipul Modi, Ashish Kulshrestha, Roosbeh Nowrouzian, Mahmood Zangui, Dimitra Michalaka, Philip Haas, Miguel Lugo, Benjamin Reibach Ruoying Xu Hyoseok Chang and Seckin Ozkul

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURE S ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 1.1 Background and Motivation ................................ ................................ .............. 13 1.2 Research Objectives ................................ ................................ ......................... 14 1.3 Research Organization ................................ ................................ ..................... 15 2 REVIEW OF LITERATURE ................................ ................................ .................... 16 2.1 Introduction ................................ ................................ ................................ ....... 16 2.2 Modeling Generation of Social Recreational Activities and Joint Travel ........... 17 2.3 Vehicle Type Choice for Specific Trips and Tours ................................ ............ 29 2.4 Time of day Choice for Specific Trips and Tours ................................ .............. 33 2.5 Summary ................................ ................................ ................................ .......... 38 3 MODELING THE GENERATION OF SOCIAL RECREATIONAL ACTIVITY PATTERNS ................................ ................................ ................................ ............. 41 3.1 Introduction ................................ ................................ ................................ ....... 41 3.2 Data Description ................................ ................................ ............................... 42 3.3 Model Structures ................................ ................................ ............................... 52 3.3.1 Household Based Models with Household level Utility Functions ............ 53 3.3.2 Household Based Models with Individual level Utility Functions ............. 54 3.3.3 Individual level Models ensuring consistency of joint choices ................. 57 3.3.4 Multi level Models with Error Correlations ................................ ............... 61 3.4 Empirical Results for Worker Households ................................ ......................... 63 3.4.1 Goodness of Fit, Weight, and Interaction Effects ................................ .... 64 3.4.2 Impacts of Explanatory Variables ................................ ............................ 66 3.4.3 Pr edictive Assessments ................................ ................................ .......... 74 3.5 Empirical Results for Non Worker Households ................................ ................. 82 3.5.1 Goodness of Fit, Weight, and Interaction Effects ................................ .... 82 3.5.2 Impacts of Explanatory Variables ................................ ............................ 83 3.5.3 Predictive Assessments ................................ ................................ .......... 89 3.6 Summary ................................ ................................ ................................ .......... 95

PAGE 6

6 4 MODELING THE CHOICE OF HOUSEHOLD VEHICLE FOR SOCIAL RECREATIONAL TOURS ................................ ................................ ...................... 99 4.1 Introduction ................................ ................................ ................................ ....... 99 4.2 Data Des cription ................................ ................................ ............................. 100 4.3 Model Structure ................................ ................................ ............................... 109 4.4 Empirical Results ................................ ................................ ............................ 111 4.4.1 Primary Driver Allocation at Household Level ................................ ....... 111 Impacts of Explanatory Variables ................................ ............................. 112 Predictive Assessments ................................ ................................ ........... 115 4.4.2 Vehicle Type Choice at Joint Tours ................................ ....................... 117 Impacts of Explanatory Variables ................................ ............................. 119 Predictive Assessments ................................ ................................ ........... 122 4.5 Summary ................................ ................................ ................................ ........ 124 5 MODELING THE C HOICE OF TIME OF DAY OF TRAVEL FOR JOINT SOCIAL RECREATIONAL TOURS ................................ ................................ ...... 126 5.1 Introduction ................................ ................................ ................................ ..... 126 5.2 Data Assembly ................................ ................................ ................................ 127 5.3 Model Structures ................................ ................................ ............................. 132 5.4 Empirical Results ................................ ................................ ............................ 135 5.4.1 Time Window Choice ................................ ................................ ............. 135 Time window attributes ................................ ................................ ............ 138 Socio economic and Tour attributes ................................ ........................ 139 Predictive Assessments ................................ ................................ ........... 141 5.4.2 Time of Day Choice Conditional on Time Window ................................ 145 Time window attributes ................................ ................................ ............ 147 Socio economic and Tour attributes ................................ ........................ 148 Predictive Assessments ................................ ................................ ........... 151 5.5 Summary ................................ ................................ ................................ ........ 156 6 CONCLUSIONS ................................ ................................ ................................ ... 158 6.1 Introduction ................................ ................................ ................................ ..... 158 6.2 Contributions ................................ ................................ ................................ ... 158 6.2.1 Mode ling the Generation of Social Recreational Patterns ..................... 159 6.2.2 Modeling the Choice of Household Vehicle for Social Recreational Tou rs ................................ ................................ ................................ ........... 161 6.2.3 Modeling the Choice of Time of Day for Joint Social Recreational Tours ................................ ................................ ................................ ........... 161 6.3 Further Research ................................ ................................ ............................ 162 LIST OF REFERENCES ................................ ................................ ............................. 164 BIOGRAPH ICAL SKETCH ................................ ................................ .......................... 171

PAGE 7

7 LIST OF TABLES Table page 3 1 Detailed social recreational activity patterns with number of tours ...................... 45 3 2 Sample distribution by activity participation pattern at household level ............... 47 3 3 Sample distribution by activity participation pattern at individual level ................ 48 3 4 Summary statistics for explanatory variables ................................ ...................... 51 3 5 Model results of standard MNL model for worker households ............................ 70 3 6 Model results o f MLL model for worker households ................................ ............ 71 3 7 Model results of PCL model for worker households ................................ ............ 72 3 8 Model results of TBP model for worker households ................................ ............ 73 3 9 Model results of standard MNL model for nonworker households ...................... 85 3 10 Model results of MLL model for nonworker households ................................ .... 86 3 11 Model results of PCL model for nonworker households ................................ .... 87 3 12 Model results of TBP model for nonworker households ................................ .... 88 3 13 Comparisons of model attributes ................................ ................................ ....... 98 4 1 Summary statistics for explanatory variables for primary driver allocation ........ 112 4 2 Model estimation result for primary driver allocation ................................ ......... 114 4 3 Summary statistics for explanatory variables for vehicle type selection ............ 118 4 4 Model estimation result for vehicle type selection ................................ ............. 121 5 1 Number of time windows by worker composition ................................ .............. 131 5 2 Average start and end times of time windows ................................ ................... 131 5 3 Summary statistics for explanatory variables for time window selection ........... 137 5 4 Model estimation result for time window selection ................................ ............ 140 5 5 Summary statistics of the time windows chosen (Fractional time spent) ........... 146

PAGE 8

8 5 6 Model estimation result for fractional time spent ................................ ............... 150

PAGE 9

9 LIST OF FIGURES Figure page 3 1 patterns ................................ ................................ ................................ .............. 58 3 2 Observed and pr edicted distributions of worker households ............................. 77 3 3 Sensitivity of maintenance activity engagement by female (worker households) ................................ ................................ ................................ ........ 79 3 4 Sensitivity of mandatory activity engagement by male (worker households) ...... 81 3 5 Observed and predicted distributions of nonworker households ....................... 90 3 6 Sensitivity of maintenance activity engagement by female (nonworker households) ................................ ................................ ................................ ........ 92 3 7 Sensitivity of maintenance activity engagement by male (nonworker households) ................................ ................................ ................................ ........ 94 4 1 Vehicle allocation patterns by primary user ................................ ....................... 104 4 2 A conventional ABM framework with two modules developed .......................... 108 4 3 Observed and predicted primary driver allocation patterns .............................. 116 4 4 Observed and predicted vehicle type choic e patterns ................................ ..... 123 5 1 Departure and arrival time distributions for joint tours (worker households) ... 12 8 5 2 Departure and arrival time distributions for joint tours (nonworker households) ................................ ................................ ................................ ...... 128 5 3 A conceptual illustration for determination of time windows .............................. 130 5 4 Observed and predicted distributions of time window (start time) .................... 142 5 5 Observed and predicted distributions of time window (end time) ..................... 143 5 6 Observed and predicted distributions of t ime window (temporal location) ....... 144 5 7 Observed and predicted distributions of departure time from home ................ 153 5 8 Observed and predicted distributions of arrival time at home .......................... 154 5 9 Observed and predicted distributions of tour duration ................................ ..... 155

PAGE 10

10 6 1 An overview of research contributions in an ABM framework ........................... 160

PAGE 11

11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy INTRA HOUSEHOLD INTERACTIONS IN SOCIAL RECREATIONAL ACTIVITIES AND TRAVEL By Kwangkyun Lim December 2012 Chair: Sivaramakrishnan Srinivasan Major: Civil Engineering This research explain s interactions among household members during daily activity and travel related decision making process with focus on social recreational activities. Within the broad spectrum of modeling social recreational activities and travel, this study focuses on th ree major aspects: Modeling the Generation of Social Recreational Patterns, Modeling the Choice of Household Vehicle for Social Recreational Tours, and Modeling the Choice of Time of Day for Joint Social Recreational Tours. This study uses data from the 20 09 National Household Travel Survey conducted in the United States for the estimation of all models. The empirical scope of this work is largely restricted to the analysis of the behavior of couple households (two adults comprising a male female couple wit h or without children). In the context of the generation of social recreational activity patterns, interpersonal interactions within a household are largely in the form of trade offs between solo and joint activity participation decisions. This research c ompares four methods depending on different decision making units that are capable of capturing such interactions between household heads: multinomial logit models, multilinear logit models, parallel choice constrained logit models, and tri variate binar y probit models.

PAGE 12

12 W e compared the models in terms of predictive abilities and sensitivities to specific explanatory factors. The next aspect of the research is the modeling of household vehicle choice. Two model components are developed that can be applied within any operational activity based modeling framework. One allocates each vehicle to a primary driver in the household (long term, household level model). The second model allocates a vehicle for the joint tours (short term, tour level model). Both mode ls were estimated using the unlabeled binary logit approach. This study is further limited to households with two cars. The final component of this research is on the choice of time of day for joint tours. Joint travel requires the temporal synchronization of travel of all members of the travel party. This study describes how the time constraints of multiple persons can be effectively accommodated into determining the choice set and, subsequently, the choice of timing of joint tours. In this study, an alter nate two step approach is proposed. The first model predicts the time window chosen for pursuing the joint discretionary tour and the second model locates the tour within the time window by determining the start and end times of the tour simultaneously on a continuous scale. Overall this research contributes to understanding of the social recreational activity travel patterns and presents methods to effectively represent these choices within a conventional activity based modeling framework.

PAGE 13

13 CHAPTER 1 I NTRODUCTION 1.1 Background and Motivation The past three decades ha ve seen substantial developments, refinements, and implementations of the activity based methods employed for travel modeling. This has been motivated by the need to make the models sensiti ve to a wide range of Travel Demand Management and other policy actions (Bhat and Koppelman, 2003). In this context, m any effort s have been directed towards the modeling of mandatory (work/school) travel. The efforts in the context of discretionary (recrea tional) travel are relatively less ( Mokhtarian et al., 2006) Further, many models use individuals as decision making units. Yet, the role of household interactions in activity travel decision making has also been well recognized (see for example, Srinivas an, 2004; Zhang and Fujiwara, 2006; Zhang et al., 2009; Anggraini 2009). In general, the intra household interaction behavior can manifest in the form of (1) joint engagement with other household members in activities and travel (synchronization), (2) allo cation of tasks (household chores) for one/more household member to conduct, (3) sharing of household resources, and (4) trade offs between personal serving and household serving travel. In the context of discretionary travel, joint engagement in activiti es and travel and the trade offs people make between solo and joint episodes are perhaps the most dominant forms of household interactions. To be sure, empirical evidence supports the notion that such joint travel by household members represents a substa ntial portion of non work trip purposes (see for example, Steed and Bhat, 2000; Kato and Matsumoto, 2009; Sener et al., 2010).

PAGE 14

14 In light of the discussions thus far, this research contributes to the modeling of household interdependencies in the context of social recreational activities and travel. 1.2 Research Objectives Within the broad spectrum of modeling social recreational activities and travel, this study f ocuses on three major aspects: m odeling the g eneration of s ocial r ecreational p atterns, m odeling the c hoice of household v ehicle for s ocial r ecreational t ours, and m odeling the c hoice of t ime of d ay of t ravel for j oint s ocial r ecreational Tours. Modeling activity or tour generation is generally the first step in any activity based framework f or predicting daily travel patterns. In the context of social recreational activities/travel, it is important to recognize that such activities can be undertaken by individuals both independently and jointly with other household members. Thus, the model ha s to reflect the trade offs made by household members in making such decisions. Further, the model should also ensure consistency in the prediction of joint episodes. Specifically, if one member of the household is predicted to undertake joint social activ ities, there should be another household person also predicted to undertake the same episode. This research compares alternate methods to operationalize such interactions and to ensure consistency in overall predictions. The next two components of the re search focus on joint social recreational tours. Household members choosing to travel jointly to pursue social recreational activities mostly have a choice of household vehicle to be used for the journey. While a larger vehicle may be desirable to accommod ate all passengers, the associated higher costs may also be factored into the decision making. This study builds a model of vehicle choice for joint social recreational travel that considers the needs cost trade off. A n unlabeled choice modeling methodolog y is used flexible to deal with a variety of

PAGE 15

15 household vehicle type attribute s. A good understanding of household vehicle usage patterns can in turn lead to increased accuracy of vehicle emissions and fuel consumption patterns. The final component of t his research is on the choice of time of day for joint tours. Joint travel requires the temporal synchronization of travel of all members of the travel party. This study will examine how the time constraints of multiple persons can be effectively accommoda ted into determining the choice set and, subsequently, the choice of timing of joint tours. This study uses data from the 2009 National Household Travel Survey conducted in the United States for the estimation of all models. The empirical scope of this wo rk is largely restricted to the analysis of the behavior of couple households (two adults comprising a male female couple with or without children). 1.3 Research Organization The rest of this research is organized as follows. Chapter 2 presents a review of the literature and identifies the major contributions of this study. Chapters 3, 4, and 5 include the entire study of social recreational activity participation pattern, household vehicle type choice for social recreation joint travel needs, and time of day choice for social recreation joint travel, respectively. Each chapter explicitly provides data description, modeling framework, discussions of empirical models, and summary at the end. Chapter 6 concludes the research with contributions and identifies areas that need further study.

PAGE 16

16 CHAPTER 2 REVIEW OF LITERATURE 2.1 B a c k ground There is an extensive body of literature on the field of activity based travel demand modeling (see for example Bhat and Koppelman, 2003; Bhat and Singh, 2000; Fujii et al. 1999; Goulias et al., 2011; Jovicic, 2001, Bradley, 2010; PB America, Inc., 2009, Wang, 1997). Within this framework, there is also a growing set of studies on the role of household interactions and associated methods and em pirical models (see for example Gliebe and Koppe l man 2002 and 2005; Srinivasan and Bhat, 2006; Scott and Kanaroglou, 2002; Zhang et al., 2004, 2005, 2006, 2009; Srinivasan and Athuru, 2005; Golob and McNally, 1997; Kato, 2009). To limit the scope of th e literature review and to position this study in the context of the past research, this chapter is organized to correspond to the three major facets of modeling social recreational activities and travel which are the focus of this research The first sect ion examines the literature on the generation of social recreational activities, joint travel, and the methods to operationalize the relevant household interactions. The next section is focused on the allocation of vehicles to household members and to tour s. Finally, research on modeling the choice of time of day is discussed. Prior to proceeding with a detailed discussion of the literature, it is useful to provide ntal focus of this study. Some examples of these activities include: go to gym/exercise/play sports event/go to bar, and visit public places for historical site/museum/park/library. These are also referred to as discretionary or leisure activities in the l iterature and may

PAGE 17

17 be broadly viewed as activities that are undertaken not for reasons of biological sustenance (such as work to earn wages and eating to meet physiological needs) or household/personal maintenance (such as grocery shopping). Individuals hav e flexibility when these activities are undertaken, where, with whom, and when. 2.2 Modeling Generation of Social Recreational Activities and Joint Travel Activity generation and associated behavior are influenced by many factors. In the context of di scretionary activities (i.e., social recreational), we examine the literatures on the factors affecting independent and joint trips making, and behavioral difference between household members based on the travel companions for the joint trips, and investi gate discrete choice modeling methods to operationalize the independent and joint choice outcomes. Chandrasekharan and Goulias (1999) used the Puget Sound Transportation Panel data to study the propensity of people to make solo and joint trips. They starte d with three key motivations: difference between joint trips and solo trips, factors affecting solo and joint trips making, and behavioral difference between the people making only solo trips and their counterparts. They identified different type of joint trips based on the companions of the trips. The dominant share in joint trip making was trips taken with spouses alone (ranged between 35.1 43.0%), children alone (ranged between 23.5 31.2%), and a combination of spouse and children (ranged between 4.3 8. 2%). Formal car pooling arrangements were represented as the third frequent type of joint trip making (ranged between 16.6 18.8%). This clearly shows that joint trips are most likely to be undertaken with family members (approximately 98%). In the overall, increasing number of household sizes with children, a multi adult older group, and a single

PAGE 18

18 household vehicle share were the household demographic characteristics that affected more joint trip arrangements. Similarly, the individuals who have been classif ied as professional and a young group were more likely to undertake joint trips. In contrast, the individuals classified as a driver in the travel diary were less likely to undertake joint trips. Return home, shopping, personal business were the trip purpo ses that were found most likely to be joint trips. In contrast to the previous study, Srinivasan and Bhat (2005) examined companion type for leisure activities pursued by individuals jointly with household and non household members. The 2003 and 2004 Amer ican Time Use Survey data were used to investigate the companion types by segmenting to three types of leisure activities (i.e., in home activity and out of home activity). The overall shares of joint activities of in home episodes were 32.4% during weekda ys and 35.3% during weekends, whereas of out home episodes were 47.6% and 71% during weekends, respectively. Furthermore, the individuals were more likely to purse in home episodes along with household members on the weekdays, while less likely to pursue o ut of home episodes along with three types of out of home leisure activities (i.e., socializing activity, passive activity, and active activity), the longer duration of acti vities, the more likely to be joint with non household members. Similarly, for the socializing and passive activities, weekday episodes were more likely to be pursed with non household members. Whites were also found to less likely undertake solo activitie s for passive/active out of home leisure activities. For the employment status, employed persons were observed to have a higher probability to pursue with other non household members, irrespective of activity

PAGE 19

19 type. In addition, the presence of children aff ected household members were the most favored companions for passive and active leisure. The result indicates that consideration of interdependencies is needed not only in a point of intra household but also in a point of inter household. Both the studies examined thus far were largely exploratory. The next set of models use methods such as Structural Equations Models (SEMs) to examine. Golob and McNally (1997) modeled activity participations and travel duration of the two household heads for three broad a ctivity types: work, maintenance, and discretionary activities. They examined direct linkage effects divided into four types: the travel requirements of out of home activities, within person activity interactions, within person travel interactions, and cro ss person interactions. A two day activity diary data were used from the Portland area. For the within person activity interactions, the negative direct effect was found between work activity participation and other two activity participations, and between maintenance activity participation and discretionary activity participation, especially, maintenance activities of the female head are more sensitive to work activities. As the cross person interaction effect, the more out of home work activities by the m ale head, the more maintenance activity participation and travel durations by the female head, but the less discretionary activity participation. Furthermore, they found that household activity interactions change when children become drivers. Simma and Axhausen (2001) investigated interpersonal and intrapersonal interaction effects with regard to their extent of activity participations (three categories work, maintenance, and leisure) and travel distance using the data from Upper Austria.

PAGE 20

20 For the interpe rsonal effects, the more number of female maintenance trips negatively impact the extent of female leisure trips. For the intrapersonal effects, the higher extents c orresponding trips. The worker female more likely increase s the engagement of maintenance trips by male, meaning the male take over some household labor. In contrast, the male Fujii et al. (1999) of its travel pattern (trip frequency and time) and its preference (satisfaction) by explicitly distinguishing activities jointly or independently. The reveled preference (RP) and stated preference (SP) data collected from the Osaka Kobe metropolitan area in Japan were used to construct a structural equations model Sp ecifically, more time is allocated to joint out of home activities with non household members, in contrast, to joint in home activities with household members. In addition, the amounts of time were observed corresponding to activity types classified by com panion and location types: in home joint activities with household members, in home joint activities with non household members, in home solo activities, out of home joint activities with household members, out of home joint activities with non household m embers, and out of home solo activities. In the overall, the impacts of worker status, working hours, person income, and presence of children were compatible with intuitive expectations, but as an interested example, a person who belongs to a large househo ld tended to spend more time with family, but preferred solo activities.

PAGE 21

21 The rest of models to be discussed use varieties of discrete choice modeling methods to model independent and joint choice outcomes both in the context of social recreational activities and other choices. There are three broad approaches to operationalize household interactions based on the wa ys of defining a decision making unit. The first approach treats the entire household as a decision making unit. This approach enumerates all the possible choice combinations from the members of household to maximize their utilities at the household level. The second approach treats the individual as a decision making unit. This approach reduces the number of utility functions, compared to the household level since the proportional share of a given choice that need to be represented at the household level c an be estimated using are simultaneously used as decision making units (e.g., the decision to pursue independent activities as the individual level while the decisi on to pursue joint activities as the household level). H ousehold based decision making unit : the approach that treats the entire household as a decision making unit includes household group utility maximizing mechanisms (Zhang, Timmermans, and Borgers, 20 02, 2004 & 2005; Zhang and Fujiwara 2006; Zhang et al. 2009; Kato and Matsumoto, 2009 ). Zhang, Timmermans, and Borgers function (called g Logit household time use model) for four different type of activities: (1) in home activity, (2) out of home independent activity, (3) allocated activity, and (4) shared activity. To do so, the authors first defined the household utility function in the form of multi linear group utility function, which can represent interaction wi th other

PAGE 22

22 household members and the degree of relative influence within a household joint decision making. This function simultaneously represents interaction between household members and dependencies between activities. The relative influences were turned out as 64% of the households are male dominant in time allocation, 22% of the household are female dominant, and the remainder of the households are equal power, with the average relative influence of 0.53 for male and 0.47 for female across the entire ho observed in other research, the larger number of workers in the household the less the mal es prefer allocated and shared activities. With the same stream of research, Zhang, Timmermans, and Borgers (2004) also developed a household task and time allocation models based on a multi linear group utility function. There were intra household intera ctions in more than 80% of each the allocated activities, but less likely to concern about the in home activities than the male heads. Furthermore, the similar study was undertaken by Zhang, Timmermans, and Borgers (2005). In continuous efforts of activity time allocation modeling, Zhang and Fujiwara (2006) adopted the iso elastic class of social welfare function which is under the household group decision mechanisms. In terestingly, the equation includes a parameter indicating the influence of intra household interactions of household members. To estimate household time allocation, four activity types were first identified as proposed in Zhang et al. (2002) above using a one week activity diary data from Kakeya and Akagi

PAGE 23

23 area in Japan. They concluded that the significant interaction parameter proves there is negotiation frontier while deciding a preferred household outcome after each household member first identifies his/h er own preferred option. In addition, the husbands had more activities became shorter on rainy days. Another contribution by Zhang et al. (2009) has been undertaken in the context of linear models. This study does not include any related impacts of activity participation patterns. Nevertheless, it is useful to refer in this section for the methodological review s with the same streamline of household group decision making mechanisms for this research. They presented two primary types of household level decision making mechanisms: multi linear utility and iso elastic utility (maximum and minimum utility models). Due to the diversity of group decision contexts, they first applied latent class modeling approach to partitioning the households for any desired number of segments. The heterogeneous models, where each latent class corresponds to a particular type of mech anisms, were compared to homogeneous models and found to perform better. The motivation of applying latent class approach is that group decision making mechanisms could vary with households. Interestingly, the intra household interaction was still influent ial to who has a worker (average weight of husband in joint decision is 0.634). All of the research conducted by Zhang et al. was significantly contributed to the applicat ion of household group decision mechanisms.

PAGE 24

24 In addition, household time allocation model similar to Zhang and Fujiwara (2006) had been proposed by Kato and Matsumoto (2009) using empirical data surveyed in Tokyo and Toyama in Japan. One of interested aspec ts was to observe the intra household interactions between a weekday and a weekend day, between the two cities, of home leisure activities were higher on both of home leisure activities were higher only on weekends, and (3) children had higher propensity of solo out of home leisure activities during the weekends. P erson based decision making unit : A second line of research treats the indiv idual as a decision ma king unit A set of choices at the household level can be represented as a combinatorial set of choices between individuals or/and joint activity/travel (Gliebe and Koppelman, 2002 and 2005) in an inter related decision pattern. Gliebe and Koppelman (2002) developed a proportional share model of daily time allocated by two adult household members with the same streamline of individual models. The key ideas of the proportional share model are that (1) the set of choice alternatives should be the same among i ndividuals, and (2) the proportional share of a joint outcome among individuals needs to be identical across the individuals. They have introduced a weight factor to be able to represent a proportional impact on joint activity decision making between two h ousehold heads using employment levels (such as full time job, part time job, and not employed). As already cited in the approach of household group utility maximizing mechanisms by Zhang et al., the weight was estimated using a logit based parametric func tion. However, they only allowed

PAGE 25

25 opposed to the approach of household group utility maximizing mechanisms (it applies the weight across all alternatives). The simultane ous model estimations for the utilities of two household decision makers were conducted using weekday trip diary data surveyed in Seattle area. The data were identified to four broad activity purposes: (1) subsistence, (2) maintenance, (3) leisure, and (4) in home activities. They further partitioned the maintenance and leisure activities into independent and joint activities. In the overall, full time workers tended to have greater impact on household joint activity decision making. Household employment le vels had negative effects on the time for joint maintenance activities. The more autos per person in a household, the more maintenance and leisure activities became independent. The number of children was found to have a negative impact on joint maintenanc e and leisure activities. Along this line, Gliebe and Koppelman (2005) expanded the earlier proportional share model to a structural discrete choice model that predicts the separate, parallel choices of full day tour patterns by both persons (called a para llel choice constrained logit). Unlike the previous version, each decision maker had a separate expected utility for the shared decision by structuring subsets of joint outcome (each subset has alternatives of decision makers in parallel). Parameterization of an importance function each nest of daily activity travel patterns are weighted to represent the total utility to be derived by the two person household. They hav e identified 10 joint outcomes using trip diary data surveyed in Seattle, Washington: (1) one independent daily pattern, (2) two fully joint daily patterns, and (3) seven partial joint daily patterns), including 94

PAGE 26

26 alternatives total. The factors positivel y affected to household joint decision making were worker status and the number of pre school age children. This suggested that the greater emphasis is placed on the work individuals relative to non workers and on the women with very young children relativ e to the others. A retired household for both adults was more likely to undertake fully joint activities. In contrast, a household with more number of children was less likely to undertake fully joint activities. In case of the households with fewer cars t han workers tended to engage in more shared ride arrangements. The longer commute distance, the greater likeliness of undertaking shared ride arrangement and partial joint activities. Overall, this study provides explicit representation of joint activity p atterns by constructing an individual level framework being able to represent intra household interactions. M ulti level decision making unit : Lastly, the activity travel generation patterns can be estimated at the multi level e.g., the decision to pursue independent activities as the individual level while the decision to pursue joint activities as the household level (Scott and Kanaroglou, 2002; Srinivasan and Bhat, 2006; Srinivasan and Athuru, 2005 ). Scott and Kanaroglou (2002) estimated trivariate ord ered probit models for the daily number of non work, out of home activity episodes that adult couple undertakes together. The models were implemented for the coupled heads by three household types using data surveyed in Toronto in Canada: non worker, one w orker, and two worker households. They structured three activity types at multi level decision making units: male solo activities, female solo activities, and household joint activities. To capture interactions between household heads, each random error te rm was correlated by specifying a standard normal trivariate distribution function. For non worker

PAGE 27

27 households, a positive relationship was found between the male and female solo activity participation (e.g., frequent solo activities by the male increase s uch activities by the female). In contrast, a negative relationship between solo activity (either by the male or by the female) and joint activity participation were found, suggesting a substitution effect between them. The female heads with a driver licen se and access to a vehicle were likely to engage in more independent activity participation. Similarly, in one worker households, the output of error correlations was interpretative as same as the case of non worker households. Working heads who are member s of multiple vehicle holds in a relatively higher income household group had undertaken more independent activities. In contrast, aging had a negative impact on the independent activities. However, if only the worker is a licensed driver, the couple is fo und to be more likely to undertake joint non work activities. In the case of dual worker households, the females were more likely to undertake riding activities for their children to/from daycare, as opposed to their male counterparts. Interestingly, femal es who reside in one vehicle households tended to undertake independent activities than are those who live in multiple vehicle households. Likewise, the error correlation between the male independent activities and female independent activities was found t o be positive. Srinivasan and Bhat (2006) developed an alternative approach to simultaneously model activity participation patterns and its time duration using San Francisco travel survey data. They identified five discrete choices within discretionary ac tivities: (1) of of home activity, and (5) household joint out of home activity, and their corresponding time durations. The disc rete components of the choices

PAGE 28

28 were modeled using a binary logit structure (whether or not undertake a given activity), and the continuous components of the choices (i.e., the activity duration) were modeled using a linear regression structure. They report ed there are intra personal trade offs between in home and out of home activity participation, as well as inter personal trade offs between independent and joint activity participation decisions. The male heads with younger children (less than 10 years) le ss undertake solo out of home activities. In contrast, households with more children were less likely to undertake joint out of home activities. Low income household undertook less solo out of home discretionary activities possibly due to budget constraint s. Finally, workers decrease the propensity of undertaking joint discretionary activities. Similarly, individuals who engaged in maintenance activities more undertook independent out of home discretionary activities during the day. In contrast to the above study, Srinivasan and Athuru (2005) included a different approach. Two dimensional inter related decision framework (or called nested mixed logit model) to allocate maintenance activities to household members was proposed using the data from San Francisco area: whether the maintenance activity is performed independently or jointly, and the person who participates in the activity in case independent activity occurred. The empirical model results indicated that: (1) the head of the household tended to undert ake maintenance activities than his/her spouse, (2) there were less joint maintenance activities as increasing number of employed household members, but more joint maintenance activities as decreasing household income, (3) aging were also observed with inc reasing maintenance activity engagement, (4) females were more likely to undertake maintenance activities, and (5) joint maintenance activities

PAGE 29

29 were preferred in the weekends and lower in households with under pre kindergarten aged children. 2.3 Vehicle Ty pe Choice for Specific Trips and Tours This section discusses the following questions: what are the factors that influence household vehicle type choice for individual trips and tours? Specifically, in the multi vehicle households, the vehicle type decisio n is a vital aspect of vehicle demand and usage modeling, and the decisions impact many other facets of individual activity travel patterns. Also, the vehicle type choice decisions are interrelated with specific activity participation decisions including w hether the activity is arranged jointly or independently, time of day decisions and schedule adjustments (Vovsha and Petersen, 2007). The choice of vehicle types for individual trips is conditional on the household vehicle holdings. Since the literature o n this subject (household level car ownership choices) is quite extensive (see for example Cao et al., 2006; Choo and Mokhtarian, 2004; Zhang et al 2009), we focus our discussion on the trip/tour level car allocation choices which is of direct relevance to our study. However it is useful to acknowledge that the factors that play a role in the household vehicle type choice are (see for example, Golob et al., 1996; Petersen and Vovsha, 2006): household related attributes (income, residential location, numb er of vehicles, number of drivers, number of workers and household size by age group); characteristics of principle driver or drivers of the vehicle (age, gender, and employment status); the characteristics of the vehicle itself (vehicle age, operating cos t, passenger and cargo capacity, body style, and value); tour related attributes (purpose, destination, distance, schedule, number of stops, pure automobile tour versus drive to transit tour); joint travel related attributes (travel party

PAGE 30

30 type, travel part y size, fully joint versus partially joint tours); and zonal attributes (area type at the origin and destination). A number of studies evidenced that the type of vehicle used for different trip purposes is not completely random and can be modeled from tho se factors (Hunt and Petersen, 2005; Vance and Iovanna, 2007; Vovsha and Pertersen, 2007, Petersen and Vovsha, 2006). These factors therefore can be answers for that question in the above while making a decision for intra household vehicle type choice. Th ere is some empirical evidence on the overall (measured in terms of VMT) usage patterns of different vehicle types. An early study by Golob et al. (1996) focused sty le, and model year) of vehicle. An analysis of two vehicle households indicated negative relationships between vehicle age and usage and between vehicle operation men wer e estimated to be more likely to drive larger vehicles (trucks, SUVs, etc.). Another research of exploring household vehicle ownership and the VMT by vehicle type has been done by Bhat and Sen (2006). High income households are unlikely to use pickup truck s and vans, while households residing in low density rural areas are more likely to use pickup trucks and SUVs. Households with very small children (less than 4 years of age) ha ve a strong preference for SUVs and minivans. As a vehicle attribute, only vehi cle operating gasoline cost was used and indicated that vehicle types that are less expensive to operate were preferred. Overall, passenger cars have the highest baseline preference compared to other vehicle types.

PAGE 31

31 More recently, Dravitzki et al. (2009) di scussed on how discretionary (i.e., social recreational) travel influences greenhouse gas emissions. Using the 1997 and 1998 New Zealand Household Travel Survey data, they compared various travel characteristics such as: vehicle occupancy by travel purpose total distance walked against the distance travelled by car, total distance travelled by non auto mode, time of day profiles by travel purpose, and average trip distance by engine size and trip purpose. The authors found that social/recreational trips we re a dominant factor in the uptake of private vehicles, which in turn influence on the size of vehicle purchased (i.e., require a larger than regular vehicle). In the context of modeling the vehicle used for individual trips and tours, it is useful to dis tinguish between solo and joint travel. In the former case, the vehicle used is most substantial fraction of the US households has at least as many cars as drivers w ith each driver being the primary user (PU) of one vehicle. Therefore, the vehicle type choice for solo trips can be largely inferred from the household level allocation of members as primary drivers of different vehicles. In the case of joint trips, the h ousehold members have a choice of which household vehicle to use. While a larger vehicle may be desirable to comfortably accommodate all the passengers traveling together, people may also consider the associated higher costs of using the larger vehicle. It is useful to note that both these facets, i.e., household level allocation of persons to cars, and the choice of car for joint tours have received relatively little attention in the literature. In the context of the former, a recent study by Vyas et al. (2012) is relevant. The authors estimated a joint household level model of the number of vehicles owned, the

PAGE 32

32 corresponding vehicle type choice and the annual mileage on each vehicle, as well as the primary driver assigned to the individual. This model has been embedded within SimAGENT (Simulator of Activities, Greenhouse emissions, Energy, Networks, and Travel) activity based travel and emissions forecasting system for the Southern California Association of Governments (SCAG) planning region. The findings i n terms of primary driver decision behavior are as follows: persons of age between 16 to 25 years are less likely to prefer SUVs and vans than do middle aged and elderly groups; women prefer newer vehicles, but not to drive large cars and small SUVs; worke rs have a higher preference for sub compact cars (especially newer cars) relative to unemployed members in the household, but mid sized SUVs are the most preferred if the commute distance is less than 10 miles. For the vehicle type choice decisions, there were nine vehicle types defined as combination of nine body types and vintage at the household level. Anggraini et al. (2008) have also explored car allocation choice behavior between two household heads, but in the case of car deficient households. Work attributes (duration, number of work episodes, part /full time, and non worker) of each individual dominantly affected the activity for which a car allocation decision is made. As an example, in households where male is a full time worker and female is a n on worker, the car is allocated to the male in 43.67 % of the cases. Peterson and Vovsha (2006) present models for tour level car allocation. They modeled a set of integrated models including composition of car ownership by vehicle type, allocation of ava ilable household cars to joint travel arrangements, and car type choice of the tours for which a car has been allocated. The authors observed patterns of

PAGE 33

33 car type choice to be related to person type, sex of driver, travel purpose, person participation, tra vel party size, income, distance to the primary destination, tour origin/destination area type, and number of stops by both car type and car age. In the overall, women were more likely to use vans mostly for joint travel and escorting than men do. Escortin g was logically associated with a relatively high percentage of vans and low percentage of trucks. Large travel parties proved to be correlated with vans and SUVs, whereas smaller travel parties proved to use automobiles more frequently. In contrast to the extent of travel parties, fully joint tours tended to be made in large automobiles, whereas partially joint tours tended to use vans and SUVs. Interestingly, the greater the need to make intermediate stops, the greater is the likelihood of using vans and Vovsha and Petersen (2007) investigated car type preferences and intra household car allocations by incorporating four short term sub models into travel demand models: individual and joint trav el generation, schedule adjustments, mode choice, and car allocation and type choice. Pair wise correlation patterns were detected between each pair of sub models. An interesting finding was that when there is no car allocation conflict, the choice of indi vidual/joint travel arrangements was probably made first and then choice of a suitable car is made. Joint travels had a greater probability of using large cars. 2.4 Time of day Choice for Specific Trips and Tours Modeling time of day decision is the proces s of scheduling travel which has variations by time of day. Such process includes predicting departure time, arrival time, or duration of the travel. Substantial body of time of day studies has primarily focused

PAGE 34

34 on commute timing decision (i.e., home work/ work home departure time) at an individual daily schedule. Typically, the approaches of time of day modeling have been dominated by two methods of defining the time scale: discrete choice methods and continuous choice methods. Time of day choice analysis has been extensively investigated using the discrete choice methods by dividing the continuous time variable into discrete time intervals. Examples of previous work departure time studies within the discrete contexture include A bkowitz (1981), Small (1982), McCafferty and Hall (1982), Hendrickson and Plank (1984), Chin (1990), Saleh and Farrell (2005), Guo et al. (2005), and Hess et al (2005). They investigated departure time choice for home work or work home trips using multino mial logit (MNL) based approaches. All studies listed above commonly reported occupation, income, age, transportation system level of service, and work schedule Bhat ( 2000) estimated departure time decisions for home based recreational and shopping trips by discretizing continuous time into six discrete time periods such as early morning, a.m. peak, a.m. off peak, p.m. off peak, p.m. peak, and evening. Bhat (1998a, 1998 b) formulated an ordered generalized extreme value (OGEV) model for shopping travel departure time choice and a mixed multinomial logit structure, recognizing natural ordering of time, for social recreational trips. In contrast to the substantial number of discrete choice studies, relatively less studies has examined the departure time decision as a continuous time scale. For example, Komma (2008) used a hazard based duration model to predict home to work commute time for the morning period. Bhat and Steed (2002) used a Cox proportional

PAGE 35

35 hazard duration structure to model the departure time of shopping trips of adult individuals. In addition to such hazard duration models, there are also literatures that are not based on hazard duration techniques in represen ting the continuous time. For example, Gadda et al. (2009) proposed Bayesian techniques using accelerated failure time specification to predict departure times for three trip purposes (i.e., home based work, non home based, and home based nonwork). Ettema and Timmermans (2003) introduced marginal utility profiles to consider aspects of departure time choice and time allocation. The marginal utility functions are continuous functions of departure time expressed as a continuous timescale. However, work along these lines still brings some shortcomings. The shortcomings on the discrete choice approach have been discussed by Bhat and Steed (2002): different temporal partitions can lead to different model results, a time point close to a block boundary is likely to be perceived as being similar rather than as being distinct, temporal traffic patterns in an aggregated time interval are not identified, and lastly the temporal partitioning schemes need to be retained in the forecasting as used in estimation. Although the continuous representation of time in the method was proposed in response to the limitations of discrete choice models, the drawbacks of these models from the practical perspective are their relative complexity in the estimation over the discrete choic e methods. Moreover, the tour based model must simultaneously predict when a tour leaves home and when it arrive back home. However, the previous studies in time of day synthesis have been too much populated on the departure time choice decision. There is a study that simultaneously models departure and arrival times of tours for a

PAGE 36

36 complete time of day schedule as well as reflects the positive features from discrete and continuous choice models. Vovsha and Bradley (2004) proposed a hybrid choice duration analytical structure that combines positive features (easy to estimate and apply) of a discrete choice model and a continuous choice model (flexible at any level of temporal resolution). The proposed utility structure is based on continuous shift variable s, which is different from a general way with all alternative specific coefficients and variables. They used a temporal resolution of one hour, which is expressed in 190 (i.e., 19 20/2) hour by hour tour departure from home and arrival back home time combinations as alternatives. Continuously, the same approach has been applied to the development of Atlanta Regional Commission (ARC) ABM structure (PB Americas, Inc., 2009) for indepe ndent and joint tours. However, applying the time of day choice model to joint tours, the tour departure and arrival period combinations are restricted to only those available for each participant on the tour. Similarly, SACSIM, activity based travel forec asting model for Sacramento Area Council of Goverments (SACOG), developed by Bowman and Bradley (2006) used the same approach for arrival and departure time choice at the tour primary destination, but the model uses as alternatives every possible combinati on of 48 half hour periods in the day (48 49/2=1,716 possible alternatives). Alternatively, departure and arrival times of tours were predicted by estimating activity duration and travel time to stops. For example, Comprehensive Econometric Micro simulato r for Daily Activity travel Patterns (CEMDAP) modeling system determines three time scheduling components for joint discretionary tours such as: the departure time from home, the time duration at stop, and travel time to the stop. By

PAGE 37

37 doing so, departure fr om home and arrival back home times can be calculated. Such tour scheduling components were accommodated using a linear regression method at a continuous time scale (Pinjari et al., 2008). In the overall, the literatures on the activity scheduling choice m odels dealt with the time variable either as several time periods by combining tour departure time with tour arrival time to generate possible alternatives or as continuous time by jointly modeling the tour departure time with time duration and travel time at stop. Another body of literatures focuses on the findings from empirical analysis in nonwork trips. As already mentioned, relatively limited studies examined the time of day decision in nonwork trips. For example, Bhat (1998b) and Steed and Bhat (2000) explored departure time choice for home based social/recreational trips in discrete choice representation. Their key observations were reported as the followings: (1) older persons tended to avoid the early morning and late evening periods, (2) individual s whose households have school going children (6 to 15 years of age) were more likely to pursue the activities during the late parts of the day than the individuals whose household have young children (difficult in the early morning and late parts of the d ay), (3) workers who have a substantial work commitment are very unlikely to participate in recreational activities during the mid day periods (a.m. and p.m. off peak times), but likely to participate in during the evening periods, whereas self employed i ndividuals preferred to participate in during the mid day periods over the non self employed individuals, (4) students tend to pursue the recreational activities during the p.m. peak through evening periods, (5) many recreational departure times were picke d before evening periods, and especially in the a.m. peak period, (6) the drive alone mode is

PAGE 38

38 pursued during the earlier periods of the day than is the non drive alone mode (the later periods of the day), and (7) departure times that take shorter travel ti mes were more preferred by individuals. Okola (2003) focused departure time choice for recreational activities by elderly nonworkers using a discrete choice analysis. She reported social activities are less likely to take place in the morning periods on we ekends than on weekdays. The same effect was observed in the travel party composition. For example, there is a negative impact on the selection of the morning parts of the day for joint activities. Females were much less likely than males to pursue early m orning recreational activities, whereas early morning recreational activities are more preferred by the persons aged 65 to 74 years. 2.5 Summary A growing interest has been exposed to the intra household interaction behavior to explicitly capture the inte rdependency among household members in the development of activity based travel demand models. The synthesis of literature cited in this chapter points out intra activities and travel, s pecifically by limiting to the following three decision components: the generation of social recreational activities, allocation of vehicles to household members and to joint tours, and the choice of time of day for joint tours. Many prior studies, at least, accommodated intra household interactions in such decision circumstances by using adequate explanatory variables, as well as by using different operationalized models. The intent of this study is to contribute to the body of research by modeling su ch interactions involving both household heads.

PAGE 39

39 Firstly, we examined the literature on the generation of discretionary activities (i.e., social recreational), factors affecting independent and joint trips making, and behavioral difference between househ old members based on the travel companions for joint trips, and discrete choice modeling methods to operationalize the independent and joint choice outcomes. There is no study that presents a systematic comparison of alternate methods for capturing the tra de offs between solo and joint activity decisions. This research compares four methods that are capable of capturing such interactions between household heads: multinomial logit models (household based model with a fully enumerated set of choice alternativ es at the household level), multilinear logit models (household based model that considers individual level utilities and group decision making process), parallel choice constrained logit models (person based model ensuring consistency in the prediction of joint participation choices across decision makers), and tri variate binary probit models (multilevel model with inter dependencies demographic characteristics and other factors (such as the weath er on the travel day) are considered as explanatory variables. In the end, we proceed to comparing the models in terms of predictive abilities and sensitivities to specific explanatory factors. The second part of the literature synthesis was focused on all ocation of vehicles to household members and to social recreational joint tours. Empirical research on the allocation of vehicles to household members and to tours is relatively limited. We develop a household level vehicle allocation (primary driver alloc ation) module that can be added to the suite of long term choices model in operational ABM frameworks. In addition, a tour level vehicle allocation (vehicle type choice) module for making joint

PAGE 40

40 tours involving both household heads is developed, which can b e applied subsequently to the conventional tour mode choice models within any operational ABM framework. Several vehicle attributes (such as size/body type, fuel efficiency, age, and operating costs) and socio economic variables (age, and presence of child ren) are included as predictor variables. A good understanding of the choice of household vehicle to be used for the travel can in turn lead to increased accuracy of predicting the air quality from emissions, fuel consumption patterns, and to assess the im pacts of policies based on vehicle type. Finally, the literature on the modeling of the choice of time of day was synthesized. A significant body of literature on time of day modeling is in the context of work tours which are often the first tours to be scheduled. Social/recreational tours are, however, scheduled within specific time windows of the day after scheduling other constrained tours. Many previous studies on the time of day choice modeling have also been dominated by discretizing the continuous time to several time blocks. In this study, an alternate two step approach is proposed. The first model predicts the time window chosen for pursuing the joint discretionary tour and the second model locates the to ur within the time window by determining the start and end times of the tour simultaneously on a continuous scale. This approach leads to a parsimonious specification that is shown to have very good predictive performance. Overall this research contribut es to an improved understanding of the social recreational activity travel patterns and presents methods to effectively represent these choices within a conventional activity based modeling framework.

PAGE 41

41 CHAPTER 3 MODELING THE GENERAT ION OF SOCIAL RECREATI ONAL ACTIVITY PATTER NS 3.1 Background H ousehold members often interact with each other during their daily activity and travel related decision making process. In the context of social and recreational activities, these interactions lead to decisions abo ut pursuing such episodes either independently or jointly with other household members. There are multiple approaches to capture such trade offs within a discrete choice based demand forecasting framework. Four such methods are examined and compared in th is study. These methods are: (1) Household level models with household level utility functions, (2) Household level models with individual level utility functions, (3) Individual level models with constraints to account for consistency in decisions about j oint activities, and (4) Multi level models with correlation patterns to capture trade offs. The intent of this chapter is to present a systematic comparison of these approaches for analyzing the trade offs between solo and joint social/recreational tra vel choices in couple adult households. Data from the 2009 National Household Travel Surveys are used in the analysis. The rest of this chapter is organized as follows. First, the data used in this study is described Next, the model specifications are described The model estimation results, predictive validations and sensitivity analysis are then discussed. Lastly, a summary of this chapter is presented

PAGE 42

42 3.2 Data Description This research utilizes the Florida add on samples from the 2009 National Household Travel Survey (NHTS). This survey collected detailed information on all activity and travel episodes (such as timing, mode, and purpose) for a one day period from the members of 14,327 households residing in Florida. The data have been geo code d to obtain the latitudes and longitudes of all the trip end locations. The survey also demographic characteristics. The raw data was subject to substantial screening and consistency checks to arrive at the final analysis sample. Only households in which all members responded to the survey were considered. Households with missing information on attributes such as locatio n and time of day of trips, and other key individual and household demographic characteristics were excluded from the sample. These checks ensure that we are able to identify the activity participation choices of all household members and also the inter de pendencies in the choices (such as joint travel). Finally, only households in which all persons began and ended their day at home were considered so as to ensure information for complete tours. On the completion of the above described screening process 9, 324 (65.1% of raw sample) households remain. The household sizes of these households are distributed as follows: 2,634 (28.2%) single person, 5,167 (55.4%) two person, 958 (10.3%) three person, and 565 (6.0%) four or more persons. Since the intent of this study is to examine household interdependencies, single person households were excluded. Among the multi person households, we focus on the most dominant category of two person households comprising a male adult and a female adult. Thus, the final analysis

PAGE 43

43 represents 92.6% of all two person households and 72% of all multi adult households. The empirical analysis of household interactions in the social activity behavior of othe r types of multi person households is identified as an area of future research. We have further divided the sample of couple households into two subgroups based on the presence of workers: households with at least a worker 2,570 (53.7%) and households wit hout any worker 2,215 (46.3%). This is because social/recreational activity participation patterns and the intra household interaction mechanisms are likely to be strongly affected by employment related attributes. In order to describe the social/recreati on activity participation choices of household members, the various trips undertaken by the persons were classified into home based tours. A tour is defined as a sequence of trips, with the first trip originating at home and the last trip ending at home wi th the intermediate trips having non home destinations. A few instances of tours without any intermediate stops were also observed (such patterns have been referred to as pure recreational trips in the literature), but these households were excluded from t he current analysis. A tour was classified as a social/recreational (in the survey social/recreational stop was defined as: go to gym/exercise/play sports, rest or relaxation/ vacation, visit friends/relatives, go out/hang out for e ntertainment/theater/sports event/go to bar, and visit public places for historical site/museum/park/library). In many cases 1,808 (65.2%) tours were single or multi stop tours where the only purpose was social recreational. In other cases, 963 (34.8%) tours

PAGE 44

44 were social/recreational stops chained with other purposes such as shopping, personal business or meal. Next, each tour was classified as either solo or joint by matching its attributes (start an d end times, mode, purpose, locations, and etc ) to tours undertaken by the other household member. The joint tour is defined as one in which the two household adults leave home together, engage in the social/recreational activity jointly, and then return home together possibly after engaging in other activities together as well. In contrast, a solo tour is defined as one in which all segments of the tour are undertaken independently by the other household member. Given the above, the detailed choice alter natives are presented in Table 3 1 with the explicit expressions by episode type (solo and joint activity) and gender at the first column. The expression of detailed choice alternatives in the first column is explained as the number of solo social recreati onal tours of male at the most left digit, the number of solo social recreational tours of female at the middle digit, and the number of joint social recreational tours at the most tour by the male and t he female and another two joint tours together (i.e., 4 social recreational tours in the household). Another way of defining choice alternatives can be represented at either household or individual level without using the explicit number of tours by type. The notations used in the table are explained below. It is noted that only less than 2% of samples have been found to involve three+ tours in either solo or joint case such that we are treated them as 2 in the table below. In summary, a substantial propo rtion of persons undertook only one social/recreational tour (solo or joint) on any day : 83.7% (921 out of 1,101 households) for the working

PAGE 45

45 households and 80.9% (905 out of 1,119 households) for the non working households Thus, we chose to focus on activ ity participation pattern decisions as opposed to explicitly modeling the number of tours. Table 3 1. Detailed s ocial r ecreational a ctivity p atterns with n umber of t ours Choice alternatives Working households Non working households Details Household level Individual level Households % Households % Male Female 000 N N N 1 469 57.2 1,096 49.5 001 J J J 105 4.1 164 7.4 002 J J J 10 0.4 13 0.6 010 F N I 312 12.1 236 10.7 011 FJ J IJ 15 0.6 14 0.6 012 FJ J IJ 0 0.0 2 0.1 020 F N I 46 1.8 32 1.4 021 FJ J IJ 1 0.0 4 0.2 100 M I N 304 11.8 254 11.5 101 MJ IJ J 16 0.6 26 1.2 102 MJ IJ J 2 0.1 2 0.1 110 MF I I 158 6.1 196 8.8 111 MFJ IJ IJ 11 0.4 15 0.7 112 MFJ IJ IJ 0 0.0 1 0.0 120 MF I I 27 1.1 43 1.9 121 MFJ IJ IJ 2 0.1 1 0.0 200 M I N 49 1.9 59 2.7 201 MJ IJ J 0 0.0 5 0.2 210 MF I I 31 1.2 28 1.3 211 MFJ IJ IJ 1 0.0 3 0.1 220 MF I I 10 0.4 19 0.9 221 MFJ IJ IJ 1 0.0 1 0.0 222 MFJ IJ IJ 0 0.0 1 0.0 Total households 2,570 100 2,215 100 Based on the social/recreational tour participation decisions of each of the household members the following eight household level outcomes are possible:

PAGE 46

46 1. No one engages in social/recreational activities (N) 2. Only male engages in solo social/recreational activity (M) 3. Only female engages in solo social/recreational activity (F) 4. Both male and female jointly engage in social/recreational activity (J) 5. Both male and female engage in social/recreational activity, but independently (MF) 6. Male engages in solo social/recreationa l activity and both male and female jointly engage in social/recreational activity (MJ) 7. Female engages in solo social/recreational activity and both male and female jointly engage in social/recreational activity (FJ) 8. Both male and female engage in social/r ecreational activity both independently and jointly (MFJ) Table 3 2 shows the distribution of household level activity participation pattern for the two subgroups (working and non working households). For example, the proportion worker households (57.2% versus 49.5%). In contrast, the worker households (4.5% versus 8.0%, and 8.8% versus 12.9%). Since the proportions of the last three alternatives (MJ, FJ, and MFJ) are significantly smaller, we combine these into an

PAGE 47

47 Table 3 2. Sample d istribution by a ctivity p articipation p attern at h ousehold l evel Activity Participation pattern N M F J MF MJ FJ MFJ Worker households (2,570) Number of households 1,469 (57.2%) 353 (13.7%) 358 (13.9%) 115 (4.5%) 226 (8.8%) 18 (0.7%) 16 (0.6%) 15 (0.6%) Non worker households (2,215) Number of households 1,096 (49.5%) 313 (14.1%) 268 (12.1%) 177 (8.0%) 286 (12.9%) 33 (1.5%) 20 (0.9%) 22 (1.0%) At the individual level, four outcomes are possible (both male and female) : 1. Not engage d in social/recreational activities (N) 2. E ngage d in only solo social/recreational activity ( I ) 3. E ngage d in only joint social/recreational activity (J) 4. E ngage d in both solo and joint social/recreational activity ( I J) Table 3 3 shows the distribution of individual level activity participation pattern by gender, for the two subgroups (working and non working households). In the case of worker households, the overall shares of alternatives are similar between the couples across the alternatives. In contrast, the male head tends to undertake more social ndently (27.0% working households.

PAGE 48

48 Table 3 3. Sample d istribution by a ctivity p articipation p attern at i ndividual l evel Activity Participation pattern N I J IJ Worker households (2,570) Male head 1,827 (71.1%) 579 (22.5%) 131 (5.1%) 33 (1.3%) Female head 1,822 (70.9%) 584 (22.7%) 133 (5.2%) 31 (1.2%) Non worker households (2,215) Male head 1, 364 ( 61.6 %) 599 ( 27.0 %) 197 ( 8.9 %) 55 ( 2.5 %) Female head 1, 409 ( 63.6 %) 554 ( 25.0 %) 210 ( 9.5 %) 42 ( 1.9 %) The final part of the data assembly process involved the determination of the explanatory factors to be used in the models for social recreational activity participation patterns. Table 3 4 presents the socio economic variables available from the travel s urveys used in modeling and presents the shares from the worker households and non worker households samples (by household). The values in the table present the proportion of household belonging to the category indicated in the row and these values sum to 1 across all categories for each variable. The substantial differences in summary statistics from the two sample groups are observed. For example, non worker households tend to be elderly persons, and their income is reduced. In contrast to the above trend the daily durations of hours for maintenance activities and discretionary activities for both household heads are higher in the households without a worker, and the likeliness of engagement for those activities are also higher than the households with wo rkers. The data from the travel surveys were supplemented with weather data obtained from the National Climatic Data Center (NCDC). The NCDC provides weather phenomena every one hour time period at 71 different weather stations across Florida

PAGE 49

49 region. Each household was matched to the nearest weather station (0.162 miles in average straight line distance) using the ArcGIS tool. For each household, data on rainfall were collected both for 24 hours and for the period 7 AM 8 PM on the three continuous days ( i.e., the travel day, the day before, and the day after) since recreational episodes are likely affected by rainfall either on the travel day or on the day before and after. Approximately 36% households (averaged across the sample) for the 24 hours and 20% households for the day time were surveyed on a rainy day. Among them, only 16% of households had been recorded more than or equal to 10 millimeters in rainfall during the entire survey day, and these proportions are turned out to be identical over the thr been taken into account in current activity based models. The decision to undertake social recreational activities can also be influenced by the land use characteristics around the responden three accessibility measures were constructed. The first is the total number of recreational acres)/distance]. maximum distance considered here was 5 mile in straight line distance from a household to the center of corresponding parcel. Data o n parcel level land use required for the construction of the above measures were obtained from the Florida Geographic Data Library (FGDL). The worker households tend to live relatively in high residential density area (68,481.9 residences in average) than the non worker households (56,864.4 residences). While the social/recreational accessibility of worker households

PAGE 50

50 (376.7) was lower than the non worker households (498.4), perhaps indicating they live closer to higher density center of the region. However, we did not find a big difference in the magnitude of retail accessibility across the two groups (203.7 for worker households and 195.6 for non worker households).

PAGE 51

51 Table 3 4 Summary s tatistics for e xplanatory v ariables Variables Worker Households Non worker households Mean Mean Household Characteristics Household income (1 18, $ 5K interval ) 13.04 (4.88) 10.08 (4.99) Presence of children 0.09/0.91 0.00/1.00 Vehicle share (vehicle/driver) 0.08/0.92 0.29/0.71 Household tenure Owned/Rented 0.94/0.06 0.95/0.05 Residential area Urban/Rural 0.79/0.21 0.77/0.23 Week (Monday Friday) / Weekend 0.74/0.26 0.71/0.29 Ethnicity White/Black/Hispanic/Other 0.91/0.03/0.01/0.05 0.94/0.03/0.00/0.03 Household worker composition Both/Male only/Female only 0.46/0.32/0.22 N/A Household full time job status Both/Male only/Female only/None 0.30/0.31/0.20/0.19 N/A Rainfall for 24 hours 0/<10/+10 millimeters on the travel day 0.63/0.21/0.16 0.65/0.20/0.15 0/<10/+10 millimeters on the day before 0.63/0.22/0.15 0.64/0.20/0.16 0/< 10/+10 millimeters on the day after 0.63/0.21/0.16 0.64/0.20/0.15 8pm) 0/<10/+10 millimeters on the travel day 0.78/0.14/0.08 0.80/0.14/0.06 0/<10/+10 millimeters on the day before 0.79/0.14/0.07 0.80/0.13/0.07 0/<10/+10 millimeters on the day after 0.79/0.13/0.08 0.80/0.12/0.08 Residential density (residences) 68,481.9 (81,897.3) 56,864.4 (67,813.6) Social Recreational accessibility 376.7 (567.7) 498.4 (719.5) Retail accessibility 203.7 (100.2) 195.6 (92.7) Note: parenthesis is a standard deviation.

PAGE 52

52 Table 3 4 Continued Variables Worker Households Non worker households Mean Mean Member Characteristics Age Male 56.60(13.18) 72.60(9.97) Female 55.23(12.45) 70.13(9.74) Daily duration of hours for mandatory activities Male 3.37(4.26) 0.07(0.61) Female 3.01(4.07) 0.03(0.40) Daily duration of hours for maintenance activities Male 0.62(1.21) 0.91(1.27) Female 0.81(1.34) 0.98(1.30) Daily duration of hours for discretionary activities Male 1.07(1.90) 1.29(1.89) Female 0.99(1.73) 1.29(1.84) Education level Under university/Above university (male) 0.57/0.43 0.62/0.38 Under university/Above university (female) 0.61/0.39 0.75/0.25 Job status Full time/Part time/Not worker (male) 0.60/0.18/0.22 N/A Full time/Part time/Not worker (female) 0.50/0.18/0.32 N/A Work schedule flexibility (Flexible/Not flexible) Flexible (male) 0.32/0.68 N/A Flexible (female) 0.24/0.76 N/A Whether engaged in mandatory activities Yes/No (male) 0.45/0.55 0.02/0.98 Yes/No (female) 0.40/0.60 0.01/0.99 Whether engaged in maintenance activities Yes/No (male) 0.53/0.47 0.69/0.31 Yes/No (female) 0.60/0.40 0.68/0.32 Number of observation (households) 2,570 2,215 Note: parenthesis is a standard deviation. 3.3 Model Structures This section outlines four methods to capture trade offs between solo and joint activity participation decisions. These methods are: (1) Household level models with household level utility functions, (2) Household level models with individual level utilit y

PAGE 53

53 functions, (3) Individual level models with constraints to account for consistency in decisions about joint activities, and (4) Multi level models with correlation patterns to capture trade offs. The methods are presented in the context of couple adult h ouseholds. These households have two adults (one male and the other female) with or without the presence of children. The two household heads are assumed to be the fundamental decision makers. 3.3.1 Household Based Models with Household level Utility Func tions The household based approach models a set of alternatives at the household level (See Table 3 2) and maximizes the utility of entire household. The simplest approach is to set up household level utility functions for each alternative as a function of the relevant explanatory factors including the characteristics of each of the household members. The utility of household h is defined as: ( 3 1 ) where, i is alternatives representing the household level activity participation patterns ( such as none, male only, female only, joint only, all solo, and others), is the alternative specific constant for the alternative i in household h and ik s, ik s and ik s are coefficients to be estimated corresponding to the (kth attribute of household level socio economic characteristics, e.g., household income and day of traveled), (kth attribute of male head characteristics, e.g., age, activity engagement) and (kth attribute of female head characteristics) for alternative i respectively. is the residual error term corresponding to the alternative i for household h

PAGE 54

54 Assuming that the error terms are independently and identically Gumbel distributed acr oss alternatives in household h the probability of choosing activity participation pattern i is determined from the multinomial logit formula : ( 3 2 ) The model parameters and coefficients are then estimated by maximum likelihood estimates (MLE) It is useful to note that nested structures can also be specified and these can be specified in very many ways, but we choose the simpler MNL formulation in this study. 3.3.2 Household Based Models with Individual level Utility Functions In this approach, the alternatives are enumerated at the household level. However, for each option, the utilities for the individual members are first constructed and then aggregated to create the household utility In the case of couple households, one The corresponding individual (Systematic) utility functions are: ( 3 3 ) ( 3 4 ) where, mik fik s, mik s and fik s are coefficients to be estimated for the male and female head utility function s and other notations are the same as above. Note that the alternatives, i, are still the household level outcomes: none male only, female only, joint only, all solo and others)

PAGE 55

55 Unlike in the first formulation with a direct household level utility function in the second formulation with individual level utility functions, the explanatory factors are allowed to have differential effects on the utilities of each decision maker. For instance, same variable to be different across the male and female utility functions. The utility of household h is related to the utilities of the household members in a multiplicative form (Zhang et al., 2009) (th is is a simplification for a two decision maker case from the general, multi person formulation) : ( 3 5 ) where, is the alternative specific constant of alternative i for household h. is the weight of male in household h that effectively rescales the utilities of male relative to those of the female. In essence, the weight reflects the influence/power of the male relative to the female in the decision process. The fe is simply type parametric function as proposed by Zhang et al. (2009) and Gliebe et al. (2005): ( 3 6 ) where Z is a vector of socio economic factors that determine the relative influence is the l th attribute of male in household h is the parameter to estimate the weight of

PAGE 56

56 male. If s are turned out statistically insignificant, the wei ghts for male and female are both equal to 0.5. This formulation allows the influence of the males to be different across households depending on socio economic factors. Finally, it is useful to note that we assume the weights w to be same across all alter natives (i.e., decision power is the same regardless of different alternatives within a household). is an intra household interaction parameter representing a concern for the equity of utility distribution between household members for any alternative. I t is useful to mention that, in earlier deterministic formulations (see Keeney, 1972, for detail), the interaction parameter was constraints as > 1 by scaling deterministic utilities and from zero to one. However, in stochastic utility ma ximization models, the interaction 6 for details). When = 0, the equation above reduces to the simple additive form proposed by Harsanyi (1955) and also collapses to the standard MNL model. That is, the MLL model is a special case of the MNL model. This is explicitly proved using the multiplicative form define d in E quation 3 5 as: ( 3 7 ) As shown in that equation, MLL parameters become identical to the MNL parameters with the product of weights. For the case of equivalent weights for two adult (i.e., weight is 0.5), the MLL parameters in terms of household level characteristics are

PAGE 57

57 equival ent with the corresponding MNL parameters as: 0.5( mik fik ik likewise the MNL parameter reflecting individual level characteristics are the same as the product of 0.5 and the corresponding MLL parameters. As assumed in the MNL model, the error ter ms of the MLL models are also independently and identically Gumbel distributed across alternatives in household h the probability of choosing activity participation pattern i is the same as th e MNL model ( E quation 3 2). 3.3.3 Individual level Models ensuring consistency of joint choices In contrast to the household based approach, the person based approach seeks to maximize the utility of each household member but in an interdependent fashion. As already shown (Table 3 3), there are four individual l evel choice outcomes (none, solo adults in the social recreational tour). The Parallel Constrained Choice Logit (PCL) model ( Gliebe, 2004) can be used to ensure that the individual level model captures such interdependencies and does not lead to unrealistic outcomes at the household leve l. Conceptually each alternative is constrained to be within an appropriate nest to ensure that the choice combinations from each individual are also feasible at the household level. Generally, the alternatives (for both heads) involving joint activities a re included in one nest and those (again for both heads) not involving joint activities are in another nest. The conceptual choice structure is proposed in Figure 3 1.

PAGE 58

58 Household Social/Recreational Activity Participation Patterns Figure 3 1. A m odel f ramework for t wo p s ocial/ r ecreational a ctivity p articipation p atterns The PCL models first define the utilities of each household member. Using the same notations as defined in the MLL model above, the corresponding utility functions are: ( 3 8 ) ( 3 9 ) where, is the is k th attributes describing both entire household and personal socio demographic characteristics, and is the residual error term corresponding to the alternative of male m in household The other notations are the same as above. Note that the alternatives are the person level outcomes: none (N), independent only (I), joint only (J), and both independent and joint activities (IJ) )

PAGE 59

59 The probability of choosing alternative for the decision maker p in household can be expressed as the prod uct of probability of choosing nest (at the upper level) and the probability of choosing alternative under the nest. The expression can be represented as: ( 3 10 ) That is, the left term is the conditional probability of person p choosing alternative given the choice, and the right term is the marginal probability of the joint outcome Assuming the error terms of PCL model are also independently and identically Gumbel distributed across alternatives in household and individual p the marginal can then be formulated as: ( 3 11 ) in which and and are importance weights on the expected utilities of the household joint outcomes (i.e., nests) for male m and female f and represent the m and female f respectively. Similarly, and m and female f respectively. They are defined as follows: ( 3 12) ( 3 13)

PAGE 60

60 ( 3 14) ( 3 15) Consequently, the total utility for the marginal share is represented as each decision Substituting the marginal share in E quation 3 10 into the E quation 3 1 1, the probability of choosing alternative i for the male m is obtained. As examples under the nest of joint activity (T), the probability of choosing a joint activity (J) by male is: ( 3 16) The probability of choosing a joint activity ( ) by female is also expressed as: ( 3 17) Likewise, under the nest of independent activity (S), the probability of choosing N by male is: ( 3 18)

PAGE 61

61 The probability of choosing by female is also expressed as: ( 3 19) The equations explained in the above have been also derived by formulating a generalized extreme value (GEV) function by Gliebe (2004), with the cons istent form discrete choice models. The complete likelihood function for a coupled adult in the household is defined as: ( 3 20) where and are one if alternative and are selected by male and female, respectively, otherwise zero. This allows error correlation between the male and the female, across subsets of social/recreational activity participation patterns. T he importance weights are formulated as same as the E quation 3 6 used in the MLL models. 3.3.4 Multi level Models with Error Correlations The fourth approach is to assume that individual max imize their own utilities in making decisions about solo activities but a joint household utility is maximized for decisions about joint activities. The three (binary outcome) decisions (one for each of the male and female adults about solo activities and one for the household about the

PAGE 62

62 joint activity) are however assumed to be correlated. This leads to the trivariate binary probits (TBP) structure. The TBP model defines the utilities for each of the three binary choices. Using the same notations as defin ed in the MLL model above, the corresponding utility functions are: ( 3 21 ) ( 3 22 ) ( 3 23 ) where, and are the alternative specific constants for alternative and respectively. and are the residual errors corresponding to the alternatives in the household The other notations are the same as in the PCC model It is noted that the u tility of not participating in any activities is fixed to zero. The residual s has multivariate normal distribution with mean 0 and variance covariance matrix across the three decisions for each household That is, they are not mutually exclusive alternatives are interrelated for the household joint outcome, and the interrelation can be captured by correlating the random error terms (Scott and Kanaroglou, 2002; Dhakar 2009). Given the abo ve, the joint probability of choosing not to participate in any activities (i.e., no one engages in social/recreational activity) is defined as:

PAGE 63

63 ( 3 24) The joint probabilities of choosing to participate in any combinations of alternatives can then be expressed as: ( 3 25 ) in which: and If alternative is chosen, otherwise. The parameters are estimated by maximizing the likelihood function. This is done using Monte Carlo simulation (See further details on Genz, 1992; Hajivassiliou and McFadden, 1998). The TBP model enables to capture complementary relationship s (e.g., if one activities increase) and substitution relationship s (e.g., if one increases independent between the three episode types. 3.4 Empirical Results for Worker Households This section presents the four (MNL, MLL PCC, and TBP) empirical model results estimated for the worker households (Models for the non worker households are presented in Section 3.5) We apply 80% of the samples for model estimations and 20%

PAGE 64

64 for predictive assessments and sensitive tests All models were estimated using maximum likelihood estimates (MLE) in SAS version 9.2. The MNL MLL PCL, and TBP model s are provided, respectively, in Table s 3 5 to 3 8 T a base category for the MNL and MLL models (Tables 3 5 and 3 6) This alternative includes the joint participation of household members along with independent activities by one or both members. In the case of PC L (both independent and joint by the member) has been referenced as the base categor ies on the male and female head For the TBP model, the base alternatives are not participating in solo/joint activities The utilities of al l base alternatives are then fixed to zero across all models 3.4.1 Goodness of F it, W eight, and I nteraction E ffects The goodness of fit measures such as the rho squared and the Akike information Criterion (AIC) values are fairly similar in the three models (MNL model (AIC 5214.1, 2 0.039), MLL model (5244.8, 0.04), and TBP model (5229, 0.037)) suggesting that one of them is not clearly statistically superior to the other. While such measures in the PCL model (AIC 6165.5 2 0.05 4 ) turns out to be better, the measures are not still readily discernible to be statistically superior to the other. Further, the small rho squared value s indicate that the explanatory factors included explain only a small percentage of the behavior over a simple, constants only model. At the same time, the weight and interaction parameters are found to be statistical ly significant in the MLL model Therefore, the MNL and the MLL models are structurally different for worker households. In the case of the relative weights, a logistic functional form is specified for the males. Since the sum of the weights for male and

PAGE 65

65 female is equal to 1, the weight of the female is implicitly determined. Several explanatory variables were considered in this logistic weight functio n and the employment status of the male head was estimated to be the only statistically significant M en with full time jobs have a greater influence (weight for male= exp( 0 .868) / (1 + exp( 0 .868) ) = 0.704; weight for female = 0.296) in social activity participation decision making In households in which the male does not have a full time job, the two decision makers are equally influential (weight = 0.5 each) in the social activity decision making. The average weight of males across all worker households in the data sample is 0.621. It is useful to mention that the previous research report s the greater influence of males on the household joint decision making. For example, Zhang et al. ( 2009 ) reported that the male head with jobs and average relative weights for male head and femal e head were found to be 0.526 and 0.474, respectively Zhang et al. ( 2002) The intra household interaction parameter (see Table 3 6 for the MLL model) turned out to be negatively and statistically significant for the worker household group. This means tha t a factor that increases (decreases) the utility of an option for an individual does not imply a corresponding increase (decrease) in the overall utility of the household for that option. That is, the interaction parameter shows a strong behavior on the g interpretation of the negative sign is not readily apparent; however, it is interesting to

PAGE 66

66 note that Zhang et al. (2009 ) also report a negative interaction parameter in the c ase of automobile purchase decisions. Similarly a positive sign on the variable of employment status (by male head) is turned out to be significant in the PCL model, leading to a greater influence on the (weight for male= exp(1.256) / (1 + exp (1.256) ) = 0.778; weight for female = 0.222) The average weight of males across all worker households in the data sample is 0.684. Consequently, the PCL and MLL models prove the decision power exist between household heads, leading a greater weight to th e male head with a full time job. T he three error terms in the TBP model (see Table 3 8) more clearly present interaction behavior between three episode settings (male solo, female solo, and joint). The joint engagement and the solo engagement by either ma le or female heads are negatively correlated. This describes a substitution effect between the two engagement types. That is, if one increases solo episodes episodes decrease This is opposed between the solo engagements by ma les and females. 3.4.2 Impacts of Explanatory Variables Turning to the explanatory variables on the final models, the variables are behaved to have differential effects by different model specification s and by different decision maker as mentioned in the section of model structures. The results indicate that socio economic factors, activity engagement patterns, and travel day characteristics have strong impacts on the choice of social recreational activity participations. The detailed effects of influentia l explanatory variables are discussed in the rest of this section. Socio economic factors: Household income and age are the two socio economic factors found to influence the social activity participation decisions. The former is

PAGE 67

67 statistically significant except in the MLL model only and the latter is significant except in the M N L model only. Households with increasing income are more likely to pursue some kind of social recreational activity (except by female only, see negative in Table 3 Table 3 8 ). A positive relationship between income and increased pursuit of leisure (including joint leisure) appears int uitively reasonable and consistent with past findings (see for example Schlich et al., 2004 ). According to age effects with the increase in age of the m ale household head, the propensity of not pursuing social recreational activities is increased. The s ame pattern is explored with the increase in age of the female head (see a positive coefficient on 7). It does appear intuitively reasonable that younger couples will have a more active social life. Scott and Kanaroglou (2002) also report that age has a negative impact on pursuit of non work out of home episodes in general. Activity engagement patterns: This study assumes that the leisure activity decisions are constrained by mandatory and maintenance activity decisions. All models common ly the household head leads to reduce d joint leisure pursuits. In the MNL model, increased mandatory activity duration undertaken by either the male or the female decreases the probability of joint social activities. The MLL model also indicates that increased work duration (for either male or female) on any day leads to an increased preference for no social activity or solo episodes rather than joint participation in leisure. The PCL model and TBP model support such activity

PAGE 68

68 participation patterns to be ascertained as the longer work duration by one or both head tends to reduce solo and joint leisure pursuits. In addition, the PCL model explains one increases the work duration solo social episodes likely increase. These results are intuitively expected since more time constraints for workers mean less free time for entertainment and pleasures, which would decrease social recreational activity participati ons in favor of independent activities. The influence of maintenance activity durations is interesting. The duration of time invested by males in out of home household maintenance has no significant effect on patterns However, with increasing time spent by the female for maintenance on any day, the male head is found to be less likely to pursue solo discretionary activity on the same day. While this cannot be explained purely by time budget effects; perhaps t here are social norms at play here. The models also suggest reduced chances of joint leisure engagement with increasing maintenance responsibilities of the female head during the day. Travel day characteristics: All mode l s indicate that the weekend days are more conducive for joint leisure activities. On examining the effect of weather, the MNL and PCL models indicate a clear reduction on social activity participation patterns on rainy days (precipitated more than or equa l to 10 millimeters), while the increase of male solo male head in Table 3 7). The TBP model suggests that the female head is found to be less likely to pursue solo and j oint discre tionary activity on the same day. Past research such as the one by Cools et al. (2008) suggests that diverse weather conditions

PAGE 69

69 (snowfall, rainfall, wind speed and temperature) can reduce travel demand on transportation networks.

PAGE 70

70 Table 3 5. Mo del r esults of s tandard MNL m odel for w orker h ouseholds N (none) M (male only) F (female only) J (joint only) MF (both indep.) Variable Param. t stat Param. t stat Param. t stat Param. t stat Param. t stat Constant Socio economic factors HH income A ge of male Age of female 3.116 0.052 12.64 4.75 1.540 6.51 1.428 0.026 4.71 1.69 1.336 5.55 1.053 4.56 Activit y engagement patterns Mandatory activity of male Mandatory activity of female Maintenance activity of male Maintenance activity of female Travel day On a weekday 1.066 0.637 0.674 4.35 5.29 3.22 0.506 0.665 0.598 0.737 1.87 3.98 4.52 2.96 1.402 0.705 5.21 2.86 0.820 0.377 2.43 1.72 0.448 0.696 1.65 2.69 Rain (rainfall 10mm) 0.212 1.66 Number of observations (HHs) Log likelihood estimate using constant only 2,056 2655.92 Log likelihood estimate of final specification 2552.07 squared 0.039 Akaike information criterion (AIC) 5214.1

PAGE 71

71 Table 3 6. Model r esults of MLL m odel for w orker h ouseholds Variable N (none) M (male only) F (female only) J (joint only) MF (both indep.) Param. t stat Param. t stat Param. t stat Param. t stat Param. t stat Constant 2.172 7.09 1.951 8.86 1.452 7.55 1.682 6.69 1.266 6.28 Male head Socio economic factors HH income Age 0.019 2.96 Activity engagement patterns Mandatory activity Maintenance activity Travel day On a weekday 2.107 5.49 1.138 2.98 2.480 6.09 1.085 2.54 Rain (rainfall 10mm) Female head Socio economic factors HH income Age Activity engagement patterns Mandatory activity 3.396 5.51 2.625 5.46 1.271 2.37 Maintenance activity 0.764 1.92 2.072 4.98 1.320 2.31 Travel day On a weekday 1.936 3.38 Rain (rainfall 10mm) Variable for weight parameter of member in joint decision time) Intra household interaction 0.868 0.276 3.13 3.25 Number of observations (HHs) Log likelihood estimate using constant only 2,056 2655.92 Log likelihood estimate of final specification 2550.39 squared 0.040 Akaike information criterion (AIC) 5244.8

PAGE 72

72 T able 3 7 Model r esults of PC L m odel for w orker h ouseholds Variable N (none) I (solo only) J (joint only) Param. t stat Param. t stat Param. t stat Male head Constant 2. 698 5 37 2. 142 3.89 1.2 82 4. 67 Socio economic factors HH income 0.037 3.2 1 Age of male 0.012 2. 82 Activity engagement patterns Mandatory activity of male 0.7 21 5.9 4 Mandatory activity of female 3.029 3.9 0 3 025 3. 87 Maintenance activity of male Maintenance activity of female 0.24 3 2. 19 Travel day On a weekday 2. 3 1 6 3.94 2. 362 3 97 Rain (fainfall 10mm) 1.4 4 7 2 05 1.4 30 2.01 Female head Constant 3.743 7 08 3.213 6.62 2.1 77 6 05 Socio economic factors HH income 0.08 0 3. 29 0.06 5 2. 55 Age of female 0.014 3.4 0 Activity engagement patterns Mandatory activity of male 1. 868 7. 24 1. 915 7. 12 Mandatory activity of female 0.6 1 9 5. 31 0. 823 1 7 3 Maintenance activity of male Maintenance activity of female 0.3 15 2 88 0. 755 2.15 On a weekday Rain (fainfall 10mm) 0.24 7 1.6 7 W eight parameter time) To the fem ale under solo outcome nest To the male under joint outcome nest 1.256 1.256 3 .84 3.84 Number of observations (HHs) Log likelihood estimate using constant only Log likelihood estimate of final specification squared Akaike information criterion (AIC) 2,056 3229.7 305 4 7 0.05 4 61 65 5

PAGE 73

73 Table 3 8. Model r esults of TBP m odel for w orker h ouseholds M (male only) F (female only) J (joint only) Variable Param. t stat Param. t stat Param. t stat Constant 0.481 2.82 0.722 12.33 1.441 9.35 Socio economic factors HH income 0.019 2.96 0.035 3.50 A ge of male 0.004 1.91 Age of female Activity engagement patterns Mandatory activity of male 0.380 6.08 0.458 4.17 Mandatory activity of female 0.284 4.50 0.519 4.45 Maintenance activity of male Maintenance activity of female 0.128 2.07 0.217 3.41 Travel day On a weekday 0.333 3.24 Rain (fainfall 10mm) 0.175 2.03 0.233 1.64 ? Correlation coefficients MF (male and female) 0.368 9.81 MJ (male and joint) 0.144 2.27 FJ (female and joint) 0.145 2.27 ? Number of observations (HHs) 2,056 Log likelihood estimate using constant only 2698.3 Log likelihood estimate of final specification 2597.5 squared 0.037 Akaike information criterion (AIC) 5229

PAGE 74

74 3.4.3 Predictive Assessments The discussions thus far indicate that although using different decision making units makes the model specifications structurally different (statistically significant weight and interaction terms a lso make the MNL and the MLL models structurally different) the overall finesses of the four models are similar. Therefore, we proceed to comparing the models in terms of predictive abilities and sensitivities to specific explanatory factors. To do so, 20 % of the data (514 worker households) used for model estimations were randomly drawn for the use of predictive assessments Three types of assessments are performed: (1) prediction of household level outcomes, (2) prediction of individual level outcomes, and (3) sensitivities to specific factors. Household level : In performing the prediction of household level outcomes, t he estimated models are applied to determine the probabilities for each alternative for each of the 514 households. MNL and MLL models directly predict the household level outcomes. The probabilities from the PCL and TBP models were suitably aggregated to determine the corresponding household level outcomes. The probabilities are then averaged across all households for each alternative to determine the predictive sample share for each alternative (i.e., aggregated shares). Figure 3 2 presents the (aggregate) observed and predicted social/recreational activity participation patterns for worker households both at the hou sehold level (chart A) and at the person level (chart B and C) The lines connecting points on the figure s do not have any meaning and are simply to provide a trend. Overall, the observed and predicted patterns are fairly similar between the MNL and the ML L models (although

PAGE 75

75 they slightly over predicts the proportion of alternative N M, and J while under predicting the proportion of F MF, and O) and between the PCL and TBP models (over predicts the proportion of alternative M, F while under predicting the proportion of N and MF). The extent of difference between the observed and predicted distributions can be quantified using the mean absolute error (MAE) measure. For each alternative, the absolute value of the difference in the number of observed and pred icted households is first calculated, and then the values are averaged across all social/recreational activity participation patterns. The MAE s are 4.32 for MNL model 5.01 for MLL model 12.01 for PCL model, and 9.82 for TBP model Next, we compare the m odels in terms of their accuracy in predicting activity patterns on each household (i.e., disaggregate analysis). For the disaggregate validations, the log likelihood (LL) estimates for the validation sample are calculated as LL= i is 1 if alternative i is chosen in household h otherwise 0. The LL estimate s are 659.6 for MNL model 665.1 for MLL model 669.2 for PCL model, and 665.8 for TBP model Both these measures indicate that the MNL model performs the best, and th e PCL model the worst. Individual level : In performing the prediction of individual level outcomes, a similar procedure to the one previously described is used. However, in this case, the PCL and TBP modes give the individual level outcomes directly. The h ousehold level predictions from the MNL and MLL models were converted into person level predictions to facilitate comparisons.

PAGE 76

76 Charts B and C of Figure 3 2 present the observed and predicted shares represented at the person level (male heads and female he ads) for worker households. Across all models, the predicted patterns in the male heads (left side) are relatively similar to the observation than in the female heads (right side). In common, the proportion of alternative N and Other (IJ & J) is over predi cted, while alternative I (solo only) is under predicted. In contrast to the MAE measure at the household level, the smaller MAE measures are observed in the TBP model ( 2.22 from male heads and 3 .63 from female heads), while the larger MAE measures are observed in the MLL model ( 3.04 from male heads and 6.72 from female heads). Those MAE measures are comparable with the MNL ( 3 .0 2 from male heads and 5.61 from female heads) and PCL ( 1.55 from male hea ds and 5.53 from female heads) models. For the disaggregate validations, the log likelihood (LL) estimates are 753.2 for MNL model ( 376.2 for males and 377.2 for females) as the best fit, and 762.7 for the PCL model ( 382.0 for males and 380.7 for fem ales) as the worst fit.

PAGE 77

77 Figure 3 2. Observed and p redicted d istributions of w orker h ouseholds 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 N (None) M (Male only) F (Female only) J (Joint only) MF (Male &Female) O (Other) Proportion of Worker Households A. Household level Observed MNL MLL PCL TBP 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 N (None) I (Solo) J&IJ (Other) Proportion of Worker Households B. Person level (Male head) N (None) I (Solo) J&IJ (Other) C. Person level (Female head)

PAGE 78

78 Sensitivity tests: Next we examine the differences in sensitivities to specific The sensitivities are also tested both at the househ old level and at the person level. In the first case, the women who were not participating in maintenance activities are now assumed to do so (the maintenance activity variable is changed from 0 to 1 for 46% of the females in the sample). The probabilitie s are re calculated across the four models Finally, the aggregate probabilities from the new case are subtracted from the original (base) case to determine the aggregate impact of the change in the maintenance activity engagement behavior. Chart A of Figu re 3 3 shows t hese changes at the household level All model s solo and joint social activity participation (a larger reduction on alternative N in the MNL model) and (MF), and other social activity participation (a larger increase on alternative F in the PCL model). However, the predictions on alternative N in the MNL and MLL models are opposed to the predictions in the PCL and TBP models (2.61% in the MNL and 1.29% in the MLL model vs. 0.40% in the PCL and 0.78% in the TBP model). The rest charts show at the person level changes. The chart B presents the probability changes for the male heads and the bottom chart C for the female heads. The direction of change on alt ernative N and alternative I is the same across gender and models, whereas the direction on the same alternative is opposite by gender. F or the male heads, the higher variation on the two alternatives (N and I) is observed in the MNL model ( 2.56 % on N vs. 2.80% on I), and higher variation in the PCL model for female heads ( 2. 47 % on N vs. 2. 84 % on I).

PAGE 79

79 Figure 3 3 Sensitivity of m aintenance a ctivity e ngagement by f emale (worker households) -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) M (Male only) F (Female only) J (Joint only) MF (Male &Female) O (Other) % Change A. Household level MNL MLL PCL TBP -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J&IJ (Other) % Change B. Male head -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J&IJ (Other) % Change C. Female head

PAGE 80

80 In the second case, the men who were not participating in m a ndatory activities are now assumed to do so (the mandatory activity variable is changed from 0 to 1 for 52% of the males in the sample). Chart A of Figure 3 4 presents the changes in the aggregate social engagement patterns across the sample at the househo ld level All models predict a substantial increase in the proportion of households not pursuing social activities and those in which only the female undertakes social activity. All other patterns have a reduction in probability. The magnitude of the chang es also appears to be comparable, although the PCL model predicts a higher proportion of households to stop pursuing social recreational activities ( 6.03 %). Correspondingly, the reduction in the male only pattern is also higher in the case of the PC L model ( 4.08 %). Chart B and C of Figure 3 4 show at the person level changes. An increase in the proportion of households not pursuing social activities is also observed across all models. However, the amounts of change are more remarkable in the male heads (7. 15% to 8. 59 %) than in the female heads ( 2.18 % to 2.64 %). The probability of solo episodes by males is dramatically dropped when the men are assumed to engage in the mandatory activities ( 3.84% to 5. 46 %), whereas the proportion of females pursuing solo ep isodes is slightly increased ( 1.26 % to 1.45%).

PAGE 81

81 Figure 3 4 Sensitivity of m andatory a ctivity e ngagement by m ale ( w orker h ouseholds) -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) M (Male only) F (Female only) J (Joint only) MF (Male &Female) O (Other) % Change A. Household level MNL MLL PCL TBP -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J&IJ (Other) % Change B. Male head -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J&IJ (Other) % Change C. Female head

PAGE 82

82 3.5 Empirical Results for Non Worker Households The MNL, MLL PCC, and TBP model are provided, respectively, in Table s 3 9 to 3 12 for non working households. The base categories through the model estimations across the four models have been identically defined as the case of working households. 3.5.1 Goodness of F it, W eight, a nd I nteraction E ffects In the methodological stand point of the MLL model since the interaction and the weight parameters were estimated to be statistically insignificant the MLL model collapses into the MNL model. The reader will note that the model par ameters of MLL model in Table 3 10 are empirically identical to the corresponding parameters of MNL model in Table 3 9 As examples, for the household variables, the coefficient in the MNL model is equal to the summation of two decision s multiplied by weight in the MLL model (e.g., household income: 0.071 in the MNL = 0.5( 0.07 1 0.07 1 ) in the MLL). For the personal variables, the coefficient in the MNL model is empirically equal to the coefficient multiplied by weight in the MLL model maintenance activity: 1.2 29 in the MLL). The goodness of fit measures such as the rho squared and the Akike information Criterion (AIC) values for the MNL and the MLL models are also exactly the same as well as the y are fairly similar as the TBP model In the PCL model, such measures are better than the other models. However, the measures over the four models do not appear readily discernible to be statistically superior to the other. As in the case of working house holds, the small rho squared value s indicate that only a small percentage of the behavior ( over a constants only model ) is explained by the expl anatory factors included.

PAGE 83

83 3.5.2 Impacts of Explanatory Variables The results indicate that socio economic factor s, activity engagement patterns, and travel day characteristics have strong impacts on the choice of social recreational activity participations. The detailed effects of influential explanatory variables are discussed in the rest of this section. I nterpret ations from the family of household based models are provided here for the MNL model as it has already been established that the MNL and the MLL models are empirically identical. Socio economic factors: Household income and age are the two socio economic f actors found to influence the social activity participation decisions. Higher income households are found more likely to undertake social recreational activities, but the propensity of joint pursuits is decreased. This generic relationship between increased income and increase leisure pursuits is intuitively reasonable and consistent with the empirical literature (also observed in the case of the working households in this study). According to age effects from the P CL and TBP models the same pattern is also explored as in the working households: with the increase in age of any household head, the propensity of not pursuing social recreational activities is increased. Activity engagement patterns: All models suggest that e ngagement of maintenance activities by male negatively affects the probability of social recreational activity participations ( see a positive sign on alternative N and PCL model s ), while the engagement of maintenance activities by female reduces the propensity of male solo activity participation (M) (the latter observation also made in the case of worker households). The PCL model explains the female increases the maintenance duratio n then social

PAGE 84

84 episodes males from the MNL and TBP models) or female solo participations. There are no mandatory activities defined for non worker households. Travel day characte ristics: O n examining the impacts of travel day, weekdays are estimated to be more conducive for solo discretionary activities even in non working households across all models However, one different finding is explored in the PCC model, with the increased which is opposed to the observations on the MNL and PCL models from working households (see a positive sign on N in Tables 3 5 and 3 7). This perhaps is explained that nonworking households ar e less restrictive in pursuing leisure pursuits during a weekday than are working households. Weather is found to be a statistically significant predictor in the PCL and TBP models. The former indicate a clear reduction on social activity participation pat terns on a rainy day. The latter is found to be less likely to pursue joint discretionary activity on the same day.

PAGE 85

85 Table 3 9. Model r esults of s tandard MNL m odel for n on w orker h ouseholds N (none) M (male only) F (female only) J (joint only) MF (both indep.) Variable Param. t stat Param. t stat Param. t stat Param. t stat Param. t stat Constant Socio economic factors HH income Age of male Age of female 3.118 0.071 15.58 6.30 1.635 7.99 1.595 0.043 6.02 2.61 1.573 0.053 6.23 2.80 1.108 5.49 Activit y engagement patterns Maintenance activity of male Maintenance activity of female 0.609 5.72 0.601 4.25 Travel day On a weekday 0.433 2.65 0.375 2.17 0.471 2.72 Rain (fainfall 10mm) 0.290 2.16 Number of observations (HHs) Log likelihood estimate using constant only 1,772 2554.85 Log likelihood estimate of final specification 2501.17 squared 0.020 Akaike information criterion (AIC) 5062.3

PAGE 86

86 Table 3 10. Mod el r esults of MLL m odel for n onw orker h ouseholds Variable N (none) M (male only) F (female only) J (joint only) MF (both indep.) Param t stat Param. t stat Param. t stat Param. t stat Param. t stat Constant 3.110 15.55 1.654 8.10 1.592 6.01 1.568 6.21 1.109 5.50 Male head Socio economic factors HH income Age 0.071 6.27 0.043 2.60 0.053 2.78 Activity engagement patterns Maintenance activity Travel day On a weekday 1.223 5.74 0.419 2.57 0.373 2.16 0.469 2.70 0.290 2.16 Female head Socio economic factors HH income 0.071 6.27 0.043 2.60 0.053 2.78 Age Activity engagement patterns Maintenance activity 1.229 4.34 Travel day On a weekday 0.419 2.57 0.373 2.16 0.469 2.70 Rain (rainfall 10mm) 0.290 2.16 Variable for weight parameter of member in joint decision Maintenance activity of male Intra household interaction Number of observations (HHs) Log likelihood estimate using constant only 1,772 2554.85 Log likelihood estimate of final specification 2501.16 squared 0.020 Akaike information criterion (AIC) 5092.3

PAGE 87

87 Table 3 11 Model r esults of PC L m odel for n on w orker h ouseholds Variable N (none) I (solo only) J (joint only) Param. t stat Param. t stat Param. t stat Male head Constant 1. 224 2. 64 2.130 10 56 0. 203 0 30 Socio economic factors HH income 0.0 57 5. 20 0.03 3 1 9 1 Age 0.02 8 5 .0 7 0.0 20 2 37 Activity engagement patterns Maintenance activity of male 0.6 3 3 5. 23 Maintenance activity of female 0.319 2.69 Travel day On a weekday 0. 5 63 4. 3 7 Rain (fainfall 10mm) 0. 2 91 2. 20 Female head Constant 2.864 3.54 3.2 03 4 54 2.877 5. 1 6 Socio economic factors HH income 0.1 71 3 20 0.1 40 2. 60 0.1 09 2. 55 Age 0.0 20 2.37 Activity engagement patterns Maintenance activity of male 0.518 4.27 Maintenance activity of female 3. 4 07 7. 62 3. 3 9 5 7 53 Travel day On a weekday 0. 3 20 2. 48 Rain (fainfall 10mm) 0.2 33 1. 6 4 Variable for weight parameter of member in joint decision Maintenance activity of male 5.259 3 87 Number of observations (HHs) Log likelihood estimate using constant only Log likelihood estimate of final specification squared Akaike information criterion (AIC) 1,772 3354.8 3 2 40 8 0.03 4 6529.7

PAGE 88

88 Table 3 12. Model r esults of TBP m odel for n on w orker h ouseholds M (male only) F (female only) J (joint only) Variable Param. t stat Param. t stat Param. t stat Constant 0.267 1.05 0.155 0.61 1.091 15.11 Socio economic factors HH income 0.035 5.50 0.026 3.99 A ge of male 0.014 4.46 Age of female 0.014 4.36 Activity engagement patterns Maintenance activity of male 0.309 4.47 0.302 4.38 Maintenance activity of female 0.156 2.37 Travel day On a weekday 0.183 2.52 0.201 2.73 0.151 1.78 Rain (fainfall 10mm) 0.211 1.82 ? Correlation coefficients MF (male and female) 0.450 12.53 MJ (male and joint) 0.213 4.01 FJ (female and joint) 0.220 4.01 ? Number of observations (HHs) 1,772 Log likelihood estimate using constant only 2613.3 Log likelihood estimate of final specification 2552.4 squared 0.023 Akaike information criterion (AIC) 5139

PAGE 89

89 3.5.3 Predictive Assessments 443 nonworking households are utilized for predictive assessments and sensitivity comparisons. It is noted that s ince the MLL model is identical to the MNL model in the non worker case, predictive analysis and sensitivity comparisons are presented as the same (in significant weight and interaction terms ). Figure 3 5 presents the (aggregate) observed and predicted social/recreational activity participation patterns for non worker hous eholds both at the household level and at the person level Household level: While the proportion of households is relatively over predicted on alternative MF ( male and female solo ), the overall fit of the MNL model is still comparable to the PCL and TBP models (see chart A of Figure 3 5) with the smaller MAE measure ( 3.09 ) in the MNL model ( 1 6 13 for the PCL model and 16.66 for the TBP model). For the disaggregate validations, the smaller LL estimate is 670.6 in the MNL model. Based on those measures th e MNL model still shows better predictive assessment for the sample as in the case of working households. Person level: Charts B and C of Figure 3 5 present proportions of alternatives represented at the person level. The predictive abilities of all model s are fairly similar across gender MAE estimates are found to be the smaller in the TBP model as 5.90 16.08 for the PCL model). The LL estimates are almost identical across all model s ( 815.3 for the PCL model to 817.6 for the TBP model). All models appear equally good in terms of predictive ability

PAGE 90

90 Figure 3 5 Observed and p redicted d istributions of n onw orker h ouseholds 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 N (None) M (Male only) F (Female only) J (Joint only) MF (Male &Female) O (Other) Proportion of Worker Households A. Household level Observed MNL&MLL PCL TBP 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 N (None) I (Solo) J&IJ (Other) Proportion of Worker Households B. Person level (Male head) N (None) I (Solo) J&IJ (Other) C. Person level (Female head)

PAGE 91

91 Sensitivity tests: On the examination of differences in sensitivities, the following changes in two variables are applied to the models: participation. The sensitivities are also tested both at the household level and at the person level. In the first case, the women who were not participating in maintenance activities are now assumed to do so. At the household based alternatives (see Figure 3 6 A ), all models predict in creases the proportion of nonworker households not participating in social recreational activities. Besides the alternative J (joint) and O (other) the directions of change on other alternatives are same across the models: the greater reduction is found o n alternative M (male only) from the two model ( MNL 2 .44 %, TBP 1. 78 ) and on alternative J from the PCL model ( 1.40%) The rest charts of Figure 3 6 shows at the person level changes. In the case of males ( chart B ), increasing proportions on alternative s N and I are the same across the models, whereas the proportion of undertaking joint activities is decreased only in the PCL model : the higher proportional reduction is observed on alternative N in the PC L model (3 .13 %) and a smaller reduction on alternati ve I i n the MN L model ( 2. 03 %). The proportional changes on alternatives N and I are 1.63% and 1.52%, respectively in the TBP model. In the case of females (chart C ), the directions of change across alternative s of both MNL and TBP models are the same but the trends on alternatives N and J are different in the PCL model.

PAGE 92

92 Figure 3 6 Sensitivity of m aintenance a ctivity e ngagement by f emale ( n on w orker h ouseholds ) -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) M (Male only) F (Female only) J (Joint only) MF (Male &Female) O (Other) % Change A. Household level MNL&MLL PCL TBP -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J (Joint) % Change B. Person level (Male head) -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J (Joint) % Change C. Person level (Female head)

PAGE 93

93 In the second case, the men who were not participating in maintenance activities are now assumed to do so. Figure 3 7 presents the changes in the aggregate social engagement patterns at two decision making units All models substantially increase the proportion of nonworker households not pursuing social activities by forcing alternatives M, F, MF, and O identically decreased (see chart A) At the person level changes the directio n and the amount of changes more appear unique across all models and genders ( see chart s B and C)

PAGE 94

94 Figure 3 7 Sensitivity of m aintenance a ctivity e ngagement by m ale ( n onworker h ouseholds) -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) M (Male only) F (Female only) J (Joint only) MF (Male &Female) O (Other) % Change A. Household level MNL&MLL PCL TBP -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J (Joint) % Change B. Person level (Male head) -6% -4% -2% 0% 2% 4% 6% 8% 10% N (None) I (Solo) J (Joint) % Change C. Person level (Female head)

PAGE 95

95 3. 6 Summary This chapter contributes towards the understanding of a recreational activity participation pattern decision by incorporating household interaction behavior. Interpersonal interactions within a household lead to recognizing such activity patterns undertaken by individuals both independently and jointly with other household members. In the last few years there has been a growing interest to explicitly accommoda te such inter personal interactions. This research proposes four alternative methods that are capable of capturing such interactions between household heads (they are categorized into three based on decision making units ) First, a m ultinomial logit (MNL) structure is specified for the household utility as a function of the relevant explanatory factors including the characteristics of each of the household members. Another approach, within the same household based decision making units, is a multilin ear logit (MLL) structure which is based on theories of group decision making behavior as a weighted sum of the wise products (i.e., a weighted multiplicative form). In contrast to the household based approach, a paral lel constrained choice logit (PCL) model is structured at person based decision making units. The PCL specification seeks to maximize the utility of each household member in an interdependent fashion. To avoid infeasible choice outcomes, a constrained choi ce framework at person level is constructed. Lastly, a tri variate binary probit (TBP) model is specified at multi level decision making units (i.e., pursue independent activities at the person level while the decision to pursue joint activities at the hou sehold level). To capture interactions between household heads each random error term i s correlated by specifying a

PAGE 96

96 standard normal trivariate distribution function. The sets of choices at non household based decision making units (i.e., PCL and TBP struc tures) must be replaced as the set of choices at the household This research utilizes the Florida add on samples from the 2009 National female person households and 72% of all multi adult households. We have further divided the sample of couple households into two subgroups based on the presence of workers: households with at least a wo rker 2,570 (53.7%) and households without any worker 2,215 (46.3%). This is because social/recreational activity participation patterns and the intra household interaction mechanisms are likely to be strongly affected by employment related attributes. Bas ed on the social recreational participation decisions of each of the household members, the following eight household level outcomes are possible: no one engages only male engages independently, only female engages independently, both male and female enga ge jointly, both male and female engage independently, male engages independently and both male and female jointly engage, female engages independently and both male and female engage jointly, and both male and female engage both independently and jointly. Similarly, there are four person based outcomes possible: no engage ment in social/recreational activities solo engagements, joint engagements, and e ngage d in both solo and joint social/recreational activity. All models are estimated for each of the two population segments leading to a total of eight models. The results indicate that household income, age of male, engagement of mandatory activities, engagement of maintenance activities, day of the week, and

PAGE 97

97 whether or not it is rain ing have strong impacts on the choice of social recreational activity participations. In the case of a working household, a weight parameter both in the MLL and the PCL models was turned out statistically to be significant While their weight of males was less on the solo cases (PCL model) than the females (this is d ue to fundamentally different functional forms of likelihood maximizations ) the observation that male heads with full recreational activity participation pattern is reasonable from the MLL model (i.e. the average weight of males across all worker households is 0.634) as well as from the PCL model (i.e., the average weight of males only for joint cases is 0.778) In addition, the TBP model convinces that there is a sub stitution effect between male /female solo engagements and joint engagements ( if one increases solo episodes episodes decrease ). This is opposed between the male solo engagements and the female solo engagements. In the case o f the nonworker household group, the MLL model collapses to the standard MNL model since the effects of household interactions and the weights were observed statistically in significant. The model parameters were also examined to establish equivalence betwe en the MNL and the MLL models The same interaction effects were also observed in the TBP model as in the case of working households. The model superiority tests are conducted both at an aggregate level (i.e., means absolute errors) and at a disaggregate l evel (i.e., log likelihood estimates for the validation sample) across the two population segments All model s are almost equally good to explain the validation sample. In addition, we provide the results of a sensitivity

PAGE 98

98 analysis with respect to the impac ts of mandatory and maintenance activity participation decisions on the social/recreational choices both at the household based alternatives and at the person based alternatives All models indicate different trends in changes in behavior and in magnitude s of the estimated changes. Overall, the MNL model was relatively straightforward to estimate, interpret, and predict the household activity generation patterns, but missing behavioral insights such as decision powers, and relationship among alternatives. Consequently, Table 6 1 can modeling perspectives. Table 3 1 3 Comparisons of model attributes Criteria MNL model MLL model PCL model TBP model Estimation Easy Difficult Difficult Difficult Interpretation Easy Difficult Difficult Easy Behavioral insights None Decision power, interaction Decision power, Similarity between alternatives Correlation between alternatives Predictive accuracy Household level Higher Higher Lower Lower Person level Lower Lower Higher Higher Ease of prediction Closed form formula Closed form formula Closed form formula Require integration process

PAGE 99

99 CHAPTER 4 MODELING THE CHOICE OF HOUSEHOLD VEHICLE FOR SOCIAL RECREATIONAL TOURS 4.1 Background The activity based modeling (ABM) approach has significantly improved the forecasts of various aspects of travel patterns. However, these models are still limited in predicting the specific household vehicle used for any trip Accurate quantification of emissions and air quality requires detailed information (s uch as size, body type, fuel type, and age ) about the vehicles being used for each trip. Thus, modeling the type of vehicle used for each trip can significantly impr ove air quality modeling practice as well as the quantification of fuel consumption. In the context of modeling the vehicle used for individual trips, it is useful to distinguish between solo and joint trips. In the former case, the vehicle used is most likely to be the vehicle for which the traveler is the primary driver. This is because a substantial fraction of the US households has at least as many cars as drivers with each driver being the primary user of one vehicle Therefore, the vehicle use patte rns can be largely inferred from the household level car allocation models (i.e., primary user allocation to one vehicle). The car allocation model can be a subsequence of conventional car ownership models which have been an overwhelmed area In the case o f joint trips, the household members have a choice of which household vehicle to use. The choice is conditional on the existing mode choice models if a household vehicle is chosen. While a larger vehicle may be desirable to comfortably accommodate all the passengers traveling together, people may also consider the associated higher costs of

PAGE 100

100 using the larger vehicle. As already highlighted in Chapter 2, the literatures in these areas are extremely limited. In light of these points we focus on modeling the a llocation of household vehicles to members and the choice of household vehicle for joint travel. The empirical scope of this work limited to couple adult households with multiple vehicles. The rest of this chapter is organized as follows. A description of the data used for analysis is presented first. This is followed by a description of the model structure (unlabeled discrete choice models). The empirical results are then presented and discussed. 4.2 Data Description The study on vehicle type choices uses data from the national samples of the 2009 National Household Travel Survey. This is unlike the study on activity generation discussed in Chapter 3 which used only the Florida samples of the same survey. The use of the national sample provides us a signif icantly larger share of social recreational tours. The NHTS data include detailed information on all activity and travel episodes (such as timing, mode, and purpose, travel party, etc.) for a one day period from the member of 136,140 households over the e ntire states. The survey also collected demographic characteristics. The raw data was subject to substantial screening and consistency checks to arrive at the final analysis sample. Only households in which all members responded to the surveyed were considered. Households with missing information on attributes such as trip purpose and time of day of trips, and other key individual and household demographic characteristics were excluded from the sample. These che cks ensure that we are able to identify the inter dependencies (such as solo and joint travel) of all

PAGE 101

101 household members that affect household vehicle type choice decisions. Finally, only households in which all persons began and ended their day at home were considered so as to ensure information for complete tours. Within the category of two adult households, we focus on couples (since the intra household interactions may be different in other two adult household structures such as roommates or the prese nce of unrated individuals). Since the survey does not explicitly identify the relationship of the adults, couple households were defined as two adult households with one male adult and one female adult with an age difference of <= 18 years. Such household s include two household heads (one male and one female adult) with or without the presence of children (children are defined as persons aged 17 years or less). W e further focus on a subset of 22,566 households ( irrespective of whether they undertook social activities or not ) or pick up truck) in which one vehicle is allocated to the male whereas another to the households with th ree or more vehicles were relatively fewer (along couple adult households). The next step in the overall data assembly procedure focuses on processing the travel diaries of households to identify tours. The process of constructing a tour is similar to that employed for the Florida samples in C hapter 3 to a large extent. A tour was defined as a sequence of trips, with the first trip originating at home and the last trip ending at home with the intermediate trips having non home destinations. A few instances of tours without any intermediate stops were also observed (such patterns have been referred to as pure recreational trips in the literature), but these households

PAGE 102

102 were excluded from the current analysis. The purpose of a tour was defined to be such as go to gym/exercise/play sports, rest or relaxation/vacation, visit friends/r elatives, go out/hang out for entertainment/theater/sports event/go to bar, and visit public places (historical site/museum/park/library). Only 11,516 couple adult, two vehicle households had social recreational tours Next, each tour was classified as ei ther solo or joint by matching its attributes (start and end times, mode, and purpose at every stop) to that of those of tours undertaken by the other household member (adult or child). The joint tour is defined as one in which the multi household member s leave home together, engage in the social/recreational activity jointly, and then return home together possibly after engaging in other activities together as well. It is useful to recognize that the national samples of the NHTS do not include detailed s patial information on the trip ends. Therefore it was not possible to perform the spatial matching of trips by the different household members. We have conducted a consistency check using the 2009 Florida add on samples which include the spatial informatio n on each trip end. With the same social recreational tours, 1,522 tours (this number was counted at the person level) were pulled out as joint cases with the additional attribute of trip end locations, as opposed to 1,572 joint tours without the location information, which in turn concludes that the definition of a joint tour without the spatial information is still enough to be used. A solo tour is defined as one in which all segments of the tour are undertaken independently of the other household member. The very few cases of partially joint tours

PAGE 103

103 (the two household adults traveling together for part of the tour) were simply re classified as solo tours (such as riding purpose travel). For all the social recreational tours, the mode of travel was identifie d (based on the mode used to access the first social recreational stop within the tour), and if a household vehicle was used, the attributes of this vehicle was determined as well.

PAGE 104

104 11,516 HHs with Social recreational tour(s) 16,065 solo social recreational tours 3,357 joint social recreational tours 3,799 tours with time overlaps 7,276 tours without time overlaps 731 tours with 1 adult & children 2,626 Tours with 2 adults & 11,075 tours by HH vehicle 4,990 tours by non auto or non HH vehicle 623 tours by HH vehicle 108 tours by non auto or non HH vehicle 2,310 tours by HH vehicle 316 tours by non auto or non HH vehicle 312 tours with time overlaps 311 tours without time overlaps No time overlaps Driver=PU? Yes (89.0%) No (11.0%) Driver=PU? Yes (79.0%) No (21.0%) Driver=PU? Yes (84.9%) No (15.1%) Driver=PU? Yes (67.5%) No (32.5%) Driver=PU? Yes (56.5%) No (43.5%) F igure 4 1 Vehicle allocation patterns by primary user 22,566 Couple adult Households (HH) with 2 vehicle s 11,050 HHs without Social recreational tour(s)

PAGE 105

105 Figure 4 1 presents the overall vehicle allocation patterns for all the solo and joint tours undertaken by the households in the analysis sample. The 11,516 households generated a total of 16,065 solo tours and 3,357 joint tours (all these are social recreational tours). Among the solo tours, about 70% were undertaken using a hous ehold vehicle and the rest were undertaken using other modes (transit, walk, bike, or non household vehicle are possible options but a bout 77.3 % of these tours were made using the walk/bike mode and a substantial fraction of these tours were less than 4 mi les in length ) The focus of this study is on the 11,075 tours made using a household vehicle. These are further classified based on whether any time overlaps exist with other tours generated in the same household. The existence of time overlap with a solo social recreational tour indicates that there is at least one other previous tour in time order (of any purpose and using any household vehicle) that was undertaken by some other household member that started or ended within the time duration of the socia l tour under consideration. It is useful to make this distinction as the presence of an overlapping tour may limit the availability of specific vehicles for the solo social tour. In our sample, 34% of the solo social tours (made using a household vehicle) had time overlaps and the rest did not. The vehicle allocations are examined for each case in terms of whether the adult used his/her primary vehicle. Not surprisingly, in over 79% of the cases, the vehicle used on the solo social tour is the primary vehic le of the driver making the tour (or equivalently, the adult making the tour is the primary user (PU) of the vehicle used for the tour). The share is higher (89% versus 79%) in the case of tours with time overlaps, as would be expected. Broadly, the descri variable is perhaps the most dominant predictor of vehicles used for solo social travel.

PAGE 106

106 The joint tours are subdivided into two groups based on the number of adults in their travel party composition. Abou t 22% (731 tours) of the joint tours involve one adult and one or more children and the rest involve both household adults (with or without the presence of other children in the tour). Among the joint tours involving one adult and children, 623 (85.2%) we re undertaken using household owned vehicles and 108 (14.8%) joint tours by a non auto mode or non household owned vehicles On further classifying the 623 tours by time overlaps and examining the vehicle used, we still find that the driver (adult) of the joint tour is very likely to use his/her primary vehicle for joint tours with children. As would be expected, the likelihoo d using the primary vehicle is higher (85% versus 67.5%) when is still the most dominant predictor of vehicles used for joint travel of an adult with children for so cial tours. Finally, we examine the remaining 2,626 (78.2% of all joint tours) tours created by both adults with or without children in their travel party composition. Among them 316 (12.0%) joint tours were not undertaken by household owned vehicles (ag ain, about 79.5% of these tours were made using the walk/bike mode and a substantial fraction of these tours were less than 4 miles in length ). In this case, the issue of time overlaps does not arise as both adults are undertaking the joint social tour und er consideration (solo tours of children are not considered in this analysis). Thus, both vehicles are always available for use in the joint tour. In this case, it is interesting to note that the vehicle used for the joint tour is not the same (43.5%) as t he primary vehicle of the driver of the joint social tour.

PAGE 107

107 Overall, the exploratory analysis indicates that knowing the primary driver of each vehicle is important from the standpoint of allocating vehicles to solo tours and joint tours made by one adult with children. While current activity based models do predict household level car ownership (such as the prediction of whether household has fewer is limited This chapter contributes towards developing a household level vehicle allocation (primary driver allocation) module that can be added to the suite of long term choices model in operational ABM frameworks (See Figure 4 2). In the case of joint tours involving both household heads, an allocation model is developed in this study to predict the choice of household vehicle (vehicle type choice) to be used for making the tour (conditional on choosing a household vehicle to travel). This model can be applied subsequent to the conventional tour mode choice models within any operational ABM framework (Figure 4 2).

PAGE 108

108 F igure 4 2. A c onventional ABM f ramework with t wo m odules d eveloped Employment/School Location Car Ownership Primary driver Allocation Daily Activity Patterns Tour Primary Destination Tour Primary Mode Tour Time of day Trip level Generation/Location/ Mode/Time of day Joint Tours: Vehicle type Choice Solo Tours: Allocate to Primary Vehicle

PAGE 109

109 4.3 Model Structure As discussed previously, two models are developed in this chapter: a household level driver allocation model and a tour level vehicle allocation model. The former allocates one of the household vehicles to the male head (and implicitly the other vehicle is allocated to the female) and the latter allocates one of the vehicles to the joint tour undertaken by both household heads. In essence both are binary choice models given that there are exactly two alternatives (vehicles). However, there is no common label that can be applied to describe the two alternatives. For instance both vehicles owned by a household could be sedans or SUVs; alternatively the household could own one sedan and one SUV. In some cases, the sedan could be the newer vehicle whereas in other cases, the SUV could be the newer one. Therefore, the unlabeled approach is used in these models. Specifically, the two characterized by several attributes such as make, age, and f uel efficiency. Most o f vehicle type and mode choice analysis use the labeled approach and the constant terms associated with each label can be interpreted as capturing the mean effects of the However, for example, in c hoosing a car, the decision maker may not always perceive it to be a choice among (Hensher et al., 2005). The unlabeled approach is more appropriate in such situations and is routinely used in destination choice models.

PAGE 110

110 ( 4 1) where, an index representing the vehicle ( =1 represen ts the first vehicle and =2 represents the second vehicle in the household; the ordering is arbitrary). In the case of the household level primary driver model, refers to the utility obtained by allocating vehicle to the male head of the household. In the case of the tour level vehicle choice models, refers to the utility obtained by allocating vehicle to the joint tour. s are coefficients to be estimated corresponding to the ( th vehicle attrib ute for alternative ; examples include body type and age ). Since the two alternatives are arbitrary, nonvehicle related attributes (such as socio economic characteristics) have to be included in the utility function by interacting these with vehicle att ributes. In our models, the vehicle body type is used as the interacting factor. is a binary value to indicate the vehicle (body) type (e.g., if =auto then =1; otherwise 0), which use to create interaction terms for generic variables in order for the model to be correctly specified with the vehicle type attribute. There are four vehicle types, auto, van SUV, and pick up and the last type is used as the reference category. s are coefficients to be estimated corresponding to the ( t h attribute of household characteristics for the primary driver allocation model; and th attribute of household and tour characteristics for the vehicle type choice model) for vehicle type The unlabeled choice approaches exclude constant terms from a ll alternatives in the utility functions. is the residual error term for alternative

PAGE 111

111 Assuming that the error terms are independently and identically Gumbel distributed across alternatives, the probability of choosing activity participation patter n is defined as: ( 4 2 ) The model parameters and coefficients are then estimated by maximum likelihood estimates (MLE ). 4.4 Empirical Results This section presents empirical results estimated for the household level primary driver allocation model and tour level vehicle type choice model. We commonly apply 80% of the final samples for the model estimations and 20% for predictive assessments. All variables identified in Tables 4 1 and 4 2 have been considered as potential explanatory variables and the final specification includes only statistically significant effects The statistical software package SAS version 9.2 i s used to estimate model s Th e detailed effects of influential explanatory variables are discussed in the rest of this section. 4.4.1 Primary Driver Allocation at Household Level Table 4 1 present s explanatory variables used for the vehicle primary driver allocation. The values in the table indicate the proportion of categorical variables (such as presence of children, household tenure, ethnicity and etc.), and these values sum to 1. The vehicle attributes were obtained by comparing miles per gallon (reported by Energy Information Admi nistration (EIA) by vehicle types and models), model years, and

PAGE 112

112 vehicle annual miles to explain which vehicle is more efficient, older, and frequently used between two household vehicles. T able 4 1. S ummary s tatistics for e xplanatory v ariables for primary driver allocation Variables Mean Household Attributes Household income (1 18, $ 5K interval ) 13.75 Average age of adults 55.8 (Male) and 53.5 (Female) 0.32 (Yes) and 0.68 (No) Household tenure 0.95 (Owned) and 0.05 (Rented), Residential area 0.64 (Urban) and 0.36 (Rural) Ethnicity 0.91 (White) and 0.09 (Other) Household worker composition 0.47 (Both), 0.21 (Male only), 0.11 (Female only), and 0.21 (None) status 1.00 (Both) and 0.00 (Male only, Female only, and None) Household vehicle holding types 0.48 (Auto), 0.09 (Van), 0.24 (SUV), and 0.19 (Pick up) Vehicle Attributes Extent of vehicle efficiency 0.5 (More efficient and Less efficient) Extent of vehicle oldness 0.5 (Older and Newer) Extent of vehicle usage 0.5 (More frequent and Less frequent) Number of households 22,566 The unlabeled binary logit (BL) model for the primary driver allocation decision is provided in Table 4 2 T he choice is modeled in terms of which vehicle is allocated to the male head of the household. The other vehicle is implicitly allocated to the female. Impacts of Explanatory Variables Vehicle attributes : The primary driver allocation for the couple adult households is affected by four vehicle attributes: fuel efficiency extent of vehicle oldness and usage and type of vehicle body The vehicle with higher miles per gallon (greater efficiency) is more likely to be assigned to the female adult (i.e., a neg ative sign on the male side).

PAGE 113

113 The male head has a greater probability of being the primary driver of the older vehicle. The vehicle used more extensively (larger annual mileage) is more likely to be allocated to the male. In terms of vehicle type, the male is less likely to be the principal driver of a SUV. The body type is also the vehicle attribute that is interacted with the socio economic variables, and these are discussed next. Socio economic factors : Four socio economic factors are turned out to be statistically significant, and they are interacted with four vehicle body types by households, the males is more likely to be the principal driver of autos, while in the female only wo rker households the males is more likely to be the principal driver of Vans. This might be explained by the fact that the non worker males would take more household responsibilities (such as home chores, and escorting children, etc.) than do the worker fem ales. Consequently, those responsibilities will need a large car. Similarly, females with children tend to be the primary driver of larger vehicles (Vans and SUVs). On the age effect, we found that older males are more likely to be the primary d river of pi ck ups. Finally, whit e males do no t prefer to be the primary driver of vans.

PAGE 114

114 Table 4 2. Model e stimation r esult for p rimary d river a llocation Vehicle primary driver allocation to the male adults Variables Coeff. t stat. Vehicle Attributes Extent of vehicle efficiency More efficient 0.224 9.25 Extent of vehicle oldness Older 0.224 11.73 Extent of vehicle usage More frequent 0.175 9.25 Type of vehicle Auto Van SUV 0.696 3.18 Socio economic Factors Worker composition (either part or full time) Male only worker (Auto) 0.436 6.80 Male only worker (Van) Male only worker (SUV) Female only worker (Auto) Female only worker (Van) 0.576 3.45 Female only worker (SUV) Age (Auto) 0.054 43.79 Age (Van) 0.027 8.91 Age (SUV) 0.031 9.74 Ethnicity White (Auto) White (Van) 0.415 2.78 White (SUV) Presence of children Child (Auto) Child (Van) 1.175 11.17 Child (SUV) 0.506 6.32 Observations 18,053 HHs Log likelihood at zero coefficients 12513.4 Log likelihood at convergence 17099.6 Rho squared 0.268

PAGE 115

115 Predictive Assessments Since an unlabeled approach is employed for model development, it is difficult to directly picture the vehicle allocation patterns (at both the household and tour levels) in terms of commonly use labels for vehicle types. For instance, it is not directly evident what types (body type) of vehicles are allocated more to men than women o r how much more likely are men than women to driver less efficient cars. To examine these, predictive assessments are made using the remaining 20% of the analysis sample. The estimated models were applied to the validation datasets to determine the probabi lities probabilities are aggregated by attributes values on the vehicles to determine aggregate shares by different labels. Figure 4 3 presents the results of this an alysis for the primary driver allocation model. The top portion of the figure presents the allocation by body type across the gender categories. Since autos dominate the body type among the four types considered in this analysis, a high proportion of both men and women are allocated to the auto. However, women are more likely to be allocated the auto than men and a similar trend is observed for both SUVs and Vans. However, men are much more likely to be allocated a pick up than women. The bottom charts in t he figure present the results based on vehicle age, usage and efficiency. Clearly men are more likely to use older and less fuel efficient vehicles. However, the differences based on overall usage (VMT) are not very pronounced. All charts in Figure 4 3 pre sent both the predicted (darker and thinner bars) and observed (lighter and thicker bars) patterns and it is evident that the observed patterns are fairly closely replicated by the model.

PAGE 116

116 Figure 4 3 Observed and p redicted p rimary d river a llocation p atterns 0 10 20 30 40 50 60 Auto(F) Auto(M) Van(F) Van(M) SUV(F) SUV(M) PickUp(F) PickUp(M) % Households Vehicle Allocation to Female (F)/Male (M) Observed Predicted 0 10 20 30 40 50 60 70 Older Newer Less frequent More frequent Less efficient More efficient % Households Vehicle Allocation with Variable Attributes Observed Predicted

PAGE 117

117 4.4.2 Vehicle Type Choice at Joint Tour s Table 4 3 present s explanatory variables used for the vehicle type choice modeling 2,310 social recreational joint tours conducted by 2,219 households are utilized for the estimation. Similarly, t he value s of categorical variables indicate the proportion s summing to one. The fuel cost is calculated as: (Tour distance Gasoline price on the traveled date) / ( EIA derived mile per gallon Number of passengers) and following the equation 2.19 U.S. dollars w ere spent per passenger (including driver) for the joint tour engagements. Among the vehicle types, possession type (49%). For the identification of primary user of the vehicle, 6% of vehicles were not specified from the survey in terms of who the primary user is. In this case, we assumed that the vehicles could be used by anyone in the household. Thus, this share is remained as the same in the variables of work status and gender of primary user (i.e., NA=0.06). Overall, the pri mary users of the vehicles are likely to be either non workers (51%) or male adults (52%). Also, non gasoline (such as hybrid and electricity) vehicles are much less populated than the gasoline vehicles. Lastly, vehicle oldness and usage are provided to ex plain which vehicle is older and frequently used between two household vehicles.

PAGE 118

118 T able 4 3. S ummary s tatistics for e xplanatory v ariables for vehicle type selection Variables Mean Household Attributes Household income (1 18, $ 5K interval ) 12.81 Average age of couple 59.25 Ethnicity 0.92 (White), 0.08 (Other) Household tenure 0.94 (Owned), 0.06 (Rented) Residential area 0.71 (Urban), 0.29 (Rural) Number of households 2,219 Tour Attributes Entire tour distance (miles) 35.99 Number of passengers 0.82 (Two or three), 0.12 (Four), 0.06 (Five+) Number of stops 2.88 (including the return home stop) 0.11 (Yes), 0.89 (No) Social recreational trip purpose 0.49 (Recreational), 0.03 (Vacation), 0.48 (Visit friends or site) Primary tour purpose (the stop purpose with the longest dwell time) 0.00 (Mandatory), 0.30 (Maintenance), 0.70 (Discretionary) Departure time 0.38 (Peak, 7 9:30 a.m. or 3:30 6 p.m.), 0.62 (Off peak) Travel day 0.39 (Mon Thu), 0.61 (Fri Sun) Number of joint tours 2,310 Vehicle Attributes Fuel cost per passenger ($) 2.19 Vehicle type 0.49 (Auto), 0.09 (Van), 0.22 (SUV), 0.20 (Pick up) Work status of primary user 0.43 (Worker), 0.51 (Non worker), 0.06 (NA) Gender of primary user 0.52 (Male adult), 0.42 (Female adult), 0.06 (NA) Fuel type 0.97 (Gasoline), 0.03 (Other) Extent of vehicle oldness 0.50 (Older, Newer) Extent of vehicle usage 0.50 (More frequent, Less frequent)

PAGE 119

119 Impacts of Explanatory Variables Vehicle attributes : The vehicle type choice for social recreational joint tours was found to be affected by five vehicle attributes (see Table 4 4) : fuel cost per passenger, work status of primary user, vehicle fuel type, extent of vehicle oldness and usage. The vehicle consuming less fuel costs per passenger is more likely to be selected for the joint social recreational travel. There is also a preference for alternative fuel vehicles (such as hybrid and electricity cars). Both these results suggest a preference for cost reductions in making choices about joint social recreational travel. Older vehicles are preferred less for joint social tours and those t hat are used more in the overall (greater annual mileage) are also used for joint tours. The car which is the primary vehicle of the worker in the household is less likely to be used for joint social recreational tours. This is probably because the vehicle used for daily commute is generally the smaller vehicle in the household fleet. Tour attributes : The decisions on vehicle type choice for social recreational joint travels found to be affected by four different tour attributes: number of passengers, pres ence of children, detailed tour purpose, and travel day. The size of the travel party clearly affects the choice of the vehicle to use used with larger travel parties being more likely to travel in larger vehicles. For travel party size of three or less, t he SUV has a strong preference. However, for travel parties of size four or more, the van is preferred if one is available. In terms of the composition of travel party, if there is at least a child on the joint travel, automobiles are found less likely to be used for that travel which seems reasonable since the larger vehicles are generally perceived to be safer for transporting

PAGE 120

120 children Travel for vacation and travel during week end days indicate a preference for larger cars, especially SUVs.

PAGE 121

121 Table 4 4. Model e stimation r esult for v ehicle t ype s election Vehicle type selection for joint tours Variables Coeff. t stat. Vehicle Attributes Fuel type Gasoline 0.680 2.79 Extent of vehicle oldness Older 0.495 8.72 Extent of vehicle usage More frequent 0.549 9.48 Fuel cost per passenger ($) 0.040 1.64 Work status of main user Worker 0.235 2.19 Tour Attributes Number of passengers 0.318 2.43 0.382 2.31 4 passengers (Auto) 0.565 1.75 4 passengers (Van) 1.778 3.85 4 passengers (SUV) 1.480 3.69 1.186 2.37 0.735 1.80 Detailed social recreational tour purposes Vacation (Auto) Vacation (Van) Vacation (SUV) 0.789 1.67 Presence of children Child (Auto) 0.626 2.17 Child (Van) Child (SUV) Travel day Fri Sun (Auto) 0.417 2.39 Fri Sun (Van) 0.720 3.09 Fri Sun (SUV) 0.499 2.27 Observations 1,848 joint tours Log likelihood at zero coefficients 1280.9 Log likelihood at convergence 1051.8 Rho squared 0.179

PAGE 122

122 Predictive Assessments Figure 4 4 presents similar analysis for the tour level vehicle allocation models. Vans and SUVs are more likely than not to be allocated as the vehicle for joint tours. In contrast pick up trucks are generally not likely to be allocated as the vehicle for the joint tours. Newer vehicles, vehicles that are used more in the overall (annual miles) and the non workers primary vehicle are more likely to be allocated to the joint tour. The effect of the newer vehicle is particularly striking.

PAGE 123

123 Figure 4 4 Observed and p redicted v ehicle t ype c hoice p atterns 0 10 20 30 40 50 60 Auto(C) Auto(N) Van(C) Van(N) SUV(C) SUV(N) PickUp(C) PickUp(N) % Households Vehicle Type Choice for Chosen(C)/Notchosen(N) Observed Predicted 0 10 20 30 40 50 60 70 80 Older Newer Less frequent More frequent Worker Nonworker % Households Vehicle Type Choice with Variable Attributes Observed Predicted

PAGE 124

124 4.5 Summary Predicting the vehicle used for individual trips and tours can help improve the quantification of energy consumption, quality of emissions forecasts, and assess impacts of policies that var y by vehicle type. However, this aspect has not been extensively incorporated into activity based modeling frameworks. The focus of this chapter is to contribute towards that end. Data from the 2009 National Household Travel Survey are used in this analysi s. The empirical scope of primary driver allocation analysis is limited to two adult two car households where one vehicle is allocated to the male whereas another to the female and further limited to the context of social recreational tours for the vehicle type choice analysis; however, this methodology can be directly extended to other cases as well. strongest predictor of the vehicle allocated to independent tours and tou rs made by adults with children. This indicates that the vehicle choice for independent travel is largely a long the member. In the case of joint tours, there is clearly a choice of vehicle to be made. Households may make trade offs between the need for a larger vehicle to comfortably accommodate the entire travel party and the higher costs associated with using th e larger vehicles. Following the exploratory analysis, two models were developed. One allocates each vehicle to a primary driver in the household (long term, household level model). The second allocates a vehicle for the joint tours (short term, tour leve l model). Both models were estimated using the unlabeled binary logit methodology. Several vehicle

PAGE 125

125 attributes (such as size/body type, fuel efficiency, age, and operating costs) and socio economic variables (age, and presence of children) were estimated to be statistically significant predicators of the vehicle allocation patterns. In the case of vehicle choice for joint tours, the day of the week and the travel party composition of the tour also affected the vehicle chosen. Predictive analyses clearly demo nstrate the overall intuitive reasonableness of the estimated models. The chapter also identifies how the estimated models from this study can be added to operational ABM frameworks currently being implemented to enhance the model predictions to include the explicit identification of specific vehicles used for solo and joint tours.

PAGE 126

126 CHAPTER 5 MODELING THE CHOICE OF TIME OF DAY OF TR AVEL FOR JOINT SOCIA L RECREATIONAL TOURS 5.1 Ba ckground This chapter presents a model for the time of day choice for joint social recreational tours. Although there is a substantial body of literature o n modeling time of day of activities and travel, m any of such studies focus on work trips. These trips generally do not involve joint travel and are often the fi r st trips to be scheduled in a day J oint travel on the other hand, requires the temporal synchronization of travel of all members of the travel party. Further, these trips are often scheduled within specific time windows of the day after other constrained travel such as commute has been sch eduled. This chapter describes how the time constraints of multiple persons can be effectively accommodated into determining the choice set and, subsequently, the choice of timing of joint tours. A two step modeling approach is proposed. The first model pr edicts the time window chosen for pursuing the joint discretionary tour and the second model locates the tour within the time window by determining the start and end times of the tour simultaneously on a continuous scale. The empirical scope of this work is limited to joint tours untaken by couple households with or without the presence of children. The rest of this chapter is organized as follows. A description of the data used for analysis is presented first. This is followed by a description of the mod el structure. The empirical results are then presented and discussed.

PAGE 127

12 7 5.2 Data Assembly The data used in this analysis comprise 2,609 joint tours undertaken by couple adult households obtained from the 2009 US National Household Travel Survey (the procedur e for assembling this sample was discussed in detail on Chapter 4 ) These tours involved the participation of both household adults (male and female). Children, if present in the household, could have also been a member of the tour party. These tours were undertaken either in a household vehicle, a non household vehicle, or by non motorized modes (walk or bike). Figure 5 1 and 5 2 show the observed time of day distributions ( departure times from home and arrival times at home for the joint tours ) from worker and nonworker households, respectively using 1 hour time period. The departure times have the highest peak at around 5 7 PM on the worker households, whereas at around 10 11 AM on the nonworker households : the earliest departure time was ob served at 4:16 AM and the latest departure time at 11 PM over the entire samples Similarly, the different time of day pattern was found in the arrival time observation based on the household work status. For example, t he arrival times have the highest pea k at around 8 10 PM for the worker households whereas the peak early starts from around 4 PM. Still, 3.4 % ( 88 out of 2, 609 tours) tours arrived at home after the midnight.

PAGE 128

128 Figure 5 1. Departure and a rrival t ime d istributions for j oint t ours ( w orker h ouseholds) Figure 5 2. Departure and a rrival t ime d istributions for j oint t ours ( n onworker households)

PAGE 129

129 The first step in the overall data assembly procedure was the identification of the available time windows for the joint discretionary tour. The proc edure is conceptually illustrated in Figure 5 3. In this example, the male undertook a mandatory tour from 7:20 AM to 5 PM and no maintenance tours. Therefore, two time window are available for scheduling the social recreational activities: the first time window from 4:00 AM (start of the survey day) to 7:20 AM and a second time window from 5 PM to 4 AM of the next day (end of survey day). The female, who undertakes two maintenance tours, has three windows potentially available for social recreational parti cipations. Thus, two time windows are commonly available for joint travel by these two persons: the first time window starts 4 a.m. and continues until 7:20 a.m. and the second window starts at 7:30 p.m. and continues until 4 a.m. on the following day. Ove rall, the process of determining time windows involves blocking out times of the day in which either adult undertook mandatory or maintenance tours. Independent discretionary tours are assumed to be scheduled after scheduling joint discretionary tours. It is also useful to note that time windows of less than five minutes were disregarded.

PAGE 130

130 Figure 5 3. A c onceptual i llustration for d etermination of t ime w indows On completion of the above analysis for all 2,609 joint tours, 855 (32.8%) had one time window These represent the cases in which both household adults did not undertake either mandatory or maintenance tours during the day and, hence, had the full day available for scheduling the joint social tours. About 44.6% (1,164 out of 2,609) of the househol ds has two time windows available, 17% has three, 4.4% had four, and about 1.2% had five or more. The cases with five or more time windows were excluded from further analysis. Table 5 1 shows the distribution of number of time windows by household type (n umber of workers in the household). We can see that non worker households are more likely to have one time window whereas the worker households are more likely to have two time windows.

PAGE 131

131 Table 5 1. Number of t ime w indows by w orker c omposition Worker composition 1 Time Window 2 Time Window s 3 Time Windows 4 Time Window s Dual worker Single worker Non worker 186 (23.3%) 200 ( 24.9 %) 414 (5 1 8 %) 383 (34. 3 %) 3 26 ( 29.2 %) 4 07 (36. 5 %) 1 28 (30. 0 %) 117 ( 27.5 %) 18 1 (42. 5 %) 31 (2 8 4 %) 38 ( 34.9 %) 4 0 (3 6 7 %) Num. of households (2,451) 800 1,1 16 4 26 1 09 Table 5 2 presents the mean starting and ending time for each time window across the corresponding joint tours. Note that, by definition, the start time for the first time window is 4 AM (start time of the survey day) and the end time of the last time window is 4 AM of the next day (end time of the survey day). Table 5 2. Average s tart and e nd t imes of t ime w indow s Number of time windows available Time window 1 Time window 2 Time window 3 Time window 4 Start End Start End Start End Start End 2 Time windows (1,164 tours) 4:00 am 10:51 am 2:35 pm 4:00 am 3 Time windows (442 tours) 4:00 am 9:38 am 12:04 pm 3:12 pm 4:49 pm 4:00 am 4 Time windows (115 tours) 4:00 am 8:51 am 10:27 am 12:02 pm 1:32 pm 4:19 pm 5:38 pm 4:00 am T he chosen time window is identified from the available time windows by comparing the start and end times of the tour with those of the windows. The temporal position of the tour within the chosen time window is determined in terms of the duration of time

PAGE 132

132 tour t o end of time window). All these durations are converted into proportions of the total time window durations. On an average, 26.6% of the time window was pre tour, 26.5% was tour based (this is the fractional time consumed for the joint tour given the time windows chosen, and the average duration of tour engagements is 3.65 hours ), and 46.9% was post tour. It is useful to note that it is possible for the tour to depart at the start of the time window (pre tour proportion = 0) and/or end at the end time of t he time window (post tour proportion = 0). 5.3 Model Structures This section of the chapter describes the modeling approach. The time of day choice for the joint social recreational tour (start and end times) is modeled using a two step approach. First, the choice of time window is modeled using an unlabeled discrete choice modeling approach. Next, the start and end times of the joint tour within the chosen time window is modeled on a continuous time scale using a fractional split mode l. Each of these is discussed in detail next. In modeling the time window selection, the unlabeled discrete choice approach is used. This is similar to the approach used for vehicle type choice in Chapter 4. This allows the treatment of alternatives as hav windows. As would be expected, for households which have the entire day available for making joint tours (i.e., only one time wi ndow) are not included in the model for time window choice. The utility function is expressed as:

PAGE 133

133 ( 5 1) where, an index representing the time windows ( =1 represents the first time window and =2 represents the second time window in the household; they are chronologically ordered). refers to the utility obtained by choosing time window for the joint tour. s are coefficients to be estimated corresponding to the ( th time window attribute for alternative ) Examples of time window specific attributes include duration of time window and temporal location The temporal location of the time window is defined based on the start and end times of the time window and include the following six options morning morning (MM), morning noon (MN), morning evening (ME), noon noon (NN), noon evening (NE), and evening eve ning (EE). These variables are defined in further detail later on. Since non time window related attributes (such as tour attributes and socio economic characteristics) are generic, those variables are treated as the alternative specific variables in the utility function by interacting with time window attributes. Throughout the modeling, the temporal location of time window is used as the interaction factor. Therefore, a binary value to indicate the temporal location t (MM, MN, ME, NN, NE, and EE: e.g., if t = MM then =1; otherwise 0), is used to create interaction terms for the generic variables The last temporal location (i.e., EE) is used as the reference category. s are coefficients to be estimated corresponding to the ( th attri bute of tour and household characteristics for the vehicle type choice model) for temporal location The unlabeled choice approaches exclude constant terms from all alternatives in the utility functions. is the residual error term for alternative

PAGE 134

134 Assuming that the error terms are independently and identically Gumbel distributed across alternatives, the probability of choosing the time window is defined as: ( 5 2 ) Next, we use fractional split logit model to the exact start and end times of the joint tour conditional on having chosen a time window. The temporal position of the joint tour within the chosen time window effectively splits the time window into three periods: window). Thus the start and end times of the tour can be modeled by determining the proportional al location of the time window duration to the three periods. Let be the fraction of the total time window duration allocated to period ( =pre tour, tour based, and post tour) such that: ( 5 3 ) where, is the set of three periods available to each person. is then specified as: ( 5 4 ) where, represents a vector of explanatory variables corresponding to time window and alternative is a coefficient to be estimated. is the residual error term without substantive behavioral interpretation

PAGE 135

135 The parameters of the above model are estimated using a quasi likelihood estimation approach by maximizing the foll owing log likelihood function: ( 5 5 ) 5.4 Empirical Results This section presents the empirical model results. The model for the choice of time window is presented first followed by the model for the time of day of tours within the chosen time window. We commonly apply 80% of the final samples for the model estimat ions and 20% for predictive assessments. The statistical software package SAS version 9.2 i s used to estimate the model s Several factors were considered as potential explanatory variables but the final specification includes only statistically significant effects 5.4.1 Time Window Choice Table 5 3 present descriptive statistics for the explanatory variables considered in the time window choice modeling. The values in the table indicate the proportion of categorical variables (such as presence of children household tenure, ethnicity and etc.), and these values sum to 1. Activity types undertook before and after the time window have been identified by gender. As an example, there are 11 percent of time and 10 percent come before time zones for the time window. To do so, 24 hours were discretized into three time zones: 4 AM to 11 AM as Morning (M), 11 AM to 4 PM a s before/after noon (N), and 4 PM to 4 AM of next day as evening/night (E). Two sets of three time zones from the

PAGE 136

136 starting and ending time of each time window produce six different combinations: MM, MN, ME, NN, NE, and EE (i.e., ME represents the time win dow starts in the morning and ends at night ). In the other attributes, there are 10 percent of tours that traveled with children. M any of the samples are represented as vehicle tours: only 14 percent conducted by walking or bicycling. One third (33%) of sa mple was comprised from the multi worker household, whereas another one third (38%) from the nonworker households.

PAGE 137

137 T able 5 3. S ummary s tatistics for e xplanatory v ariables for time window selection Variables Mean Time Window Attributes Length of time window (hours) 8.42 (5.01) Tour engagement types before the time window Mandatory tour by male 0.11 Mandatory tour by female 0.07 Maintenance tour by male 0.32 Maintenance tour by female 0.30 Tour engagement types after the time window Mandatory tour by male 0.10 Mandatory tour by female 0.07 Maintenance tour by male 0.33 Maintenance tour by female 0.30 Temporal zone of time window MM/MN/ME/NN/NE/EE 0.31/0.13/0.09/0.05/0.22/0.20 Tour Attributes Presence of children No/Yes 0.90/0.10 Mode used for the joint tour Walk or bike/Auto 0.14/0.86 Social recreational trip purpose Recreational/Vacation/Visit friends or site 0.56/0.03/0.41 Traveled on Mon Thu/Fri Sun 0.40/0.60 Number of intermediate stops 1/2/3/4+ 0.66/0.17/0.08/0.09 Household Attributes Household income (1 18, $ 5K interval) 13.26 (4.62) Presence of children 0.10/0.90 Average age of couple 56.1 (18.61) Owned household/Rented 0.95/0.05 Residential area Urban/Rural 0.75/0.25 Ethnicity White/Other 0.92/0.08 Household worker composition Dual/Single/None 0.33/0.29/0.38 Number of joint tours 1,721 Note: parenthesis is a standard deviation

PAGE 138

138 The unlabeled multinomial logit ( MNL ) model for the time window allocation decision is provided in Table 5 4 T he choice is modeled in terms of wh ich time window is allocated to the social recreational joint tour that is undertaken within the time window Time window attributes The time window allocation for the social recreational joint tours is affected by four time window attributes: length of time block temporal zones of time window and tour engagement types before and after the time window The effect of the length of time wind indicates that social recreational joint tours are more likely to take place within time windows of longer duration. The temporal zone variables suggest that social recreational joint tours are more li kely to be pursued during the midday and evening periods, while less likely to be participated in during the morning periods. In terms of tour engagement types before and after the time windows, the time window is more likely to be allocated to the join to ur when there are maintenance tours (by either a male or a female) were conducted prior to the time window. Conversely, the time window is less likely to be selected for the joint tour when followed by a mandatory tour, regardless of gender. Overall, the impacts of the tour engagement types on the time window allocation suggest that social recreational joint tours are more likely to take place after the engagements of maintenance and mandatory tours. This might be a plausible reason that the decision of so cial recreational activity engagements in general come after the decisions of mandatory and maintenance activities in the existing activity based model frameworks. The temporal zone variables are used in interacting with social economic variables through t he modeling, and these are discussed next.

PAGE 139

139 Socio economic and Tour attributes Three socio economic factors and one tour attribute appear in the final model specification: presence of children, household work status, average age of the adults, and transpor tation mode used for the social recreational joint tour. T hey are interacted with temporal zone variables as a base category. Joint tours with children tend to be pursued during the period of between the morn ing and evening. This may be related to the need of sharing more time with their children for recreational pleasures. Couple worker households are unlikely to participate in social recreational joint tours during the periods of between the morning and midd ay. This is a reasonable result since such individuals are at work during these times. On the age effect, aging showed a positive sign on the morning periods. This suggests that the older couples are more likely to take a part of morning periods for their social recreational activities. Lastly, negative impacts on the time periods of ME (morning and evening) and NE (noon and evening) were explored in the case of joint tours conducted by walk or bike modes. This is quite intuitive because the tours conducted by non auto modes would take short time duration rather than do the auto travel.

PAGE 140

140 Table 5 4. M odel e stimation r esult for time window selection Variables Coeff. t stat. Time window Attributes Length of time window available (hours) 0.165 10.48 Temporal zone for time window available MM (Ranged between morning and morning) 0.744 3.02 MN (Ranged between morning and noon) ME (Ranged between morning and evening/night) NN (Ranged between noon and noon) NE (Ranged between noon and evening/night) 0.791 5.64 EE (Ranged between evening/night and evening/night) Base Base Tour engagement types before the time window Mandatory tour by male Mandatory tour by female Maintenance tour by male 0.277 2.23 Maintenance tour by female 0.305 2.55 Tour engagement types after the time window Mandatory tour by male 0.801 3.63 Mandatory tour by female 0.780 5.15 Maintenance tour by male Maintenance tour by female Socio economic & Tour Attributes Presence of children Child (ME) 1.791 4.97 Child (EE) Base Base Household work status Dual worker (MN) 0.517 2.63 Dual worker (EE) Base Base Average age of couple adult Average age (MM) 0.007 1.75 Average age (MN) 0.004 1.97 Average age (EE) Base Base Transportation mode used Non auto mode (ME) 1.412 2.17 Non auto mode (NE) 0.549 2.29 Non auto mode (EE) Base Base Observations 1,377 joint tours Log likelihood at zero coefficients 1800.1 Log likelihood at convergence 1283.2 Rho squared 0.287

PAGE 141

141 Predictive Assessments Remaining 20% of the analysis sample (344 joint tours) is used for predictive assessments. The estimated model was applied to the validation datasets to determine and not have a ny specific meanings, beside that they were assigned chronologically. The probabilities are estimated using the significant attributes on the joint tours to determine aggregate shares by different labels. Figures 5 4 to 5 6 present the results of this anal ysis for the time window allocation model. The first figure presents the observed and predicted patterns in terms of the start time of time windows with one hour time periods. 4 AM (24%) is dominated as the start time of chosen time windows from the observ ation. The model under predicts the proportion at 4 AM, while over predicting the proportions mostly at 11 AM to 2 PM. It is evident that the time windows started at around 11 AM to 6 PM (64%) are more likely to be chosen for the joint tours. Figure 5 5 pr esents similar patterns in terms of the end time of time windows. 4 AM of next day (67%) is observed as the highest proportion from the observation, and the model over predicts at this time. Temporal locations of the chosen time windows (i.e., combinations of start/end time of time window) are presented in Figure 5 6 with 6 labeled locations. 39% of chosen time windows spanned from N (11 AM to 6 PM) to E (4 PM to 4 AM) are preferred to take pl ace joint discretionary tours. There are minor over prediction on the prediction on the proportion

PAGE 142

142 Figur e 5 4. Observed and p redicted d istributions of t ime w indow ( s tart t ime) 0 5 10 15 20 25 30 4am 5am 6am 7am 8am 9am 10m 11am 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm 12am 1am % Joint tours Time of Day Observed Predicted

PAGE 143

143 Figure 5 5. Observed and p redicted d istributions of t ime w indow ( e nd t ime) 0 10 20 30 40 50 60 70 80 90 7am 8am 9am 10m 11am 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm 12am 1am 2am 3am 4am % Joint tours Time of Day Observed Predicted

PAGE 144

144 Figure 5 6. Observed and p redicted d istributions of t ime w indow ( t emporal l ocation) 0 5 10 15 20 25 30 35 40 45 M to M M to N M to E N to N N to E E to E % Joint tours Temporal Location of Time Window Observed Predicted

PAGE 145

145 5.4.2 Time of Day Choice Conditional on Time Window Table 5 5 shows the time window attributes used for the fractional time split modeling. The length of time window for the chosen was increased to 12.36 hours compared to the one presented in the Table 5 3. The identification of activity type engagements undertook before and after the time window explains more insights of the time window selection behavior. That is, social/recreational activities are more likely to be engaged in after the mandatory activity engagements regardless of gender (19% and 12% before the ti me window vs. 3% and 2% after the time window). Similarly, such pattern is explored in the case of maintenance tour engagements. However, the threshold of engaging in social/recreational joint tours seem not clear whether the engagements come after the mai ntenance tours (20% and 17%), as the mandatory tours. On the location of temporal zones, we further segmented the time zones into five (these were three zones in the time window selection): 4 AM to 9 AM as early Morning, 9 AM to 12 PM as late Morning, 12 P M to 4 PM as midday, 4 PM to 8 PM as evening, and 8 PM to 4 AM of next day Lastly, the tour attributes and household characteristics were also accommodated in the model estimation process.

PAGE 146

146 T able 5 5. S ummary s tatistics of the t ime w indows c hosen (Fractional time spent) Variables Mean Time Window Attributes for the chosen Length of chosen time window (hours) 12.36 (3.88) Tour engagement types before the chosen time window Mandatory tour by male 0.19 Mandatory tour by female 0.12 Maintenance tour by male 0.37 Maintenance tour by female 0.37 Tour engagement types after the chosen time window Mandatory tour by male 0.03 Mandatory tour by female 0.02 Maintenance tour by male 0.20 Maintenance tour by female 0.17 Temporal zone of start time for the chosen time window Early morning/Late morning/Midday/Evening/Night 0.26/0.20/0.34/0.20/0.00 Temporal zone of end time for the chosen time window Early morning/Late morning/Midday/Evening/Night 0.01/0.04/0.10/0.14/0.70 Tour Attributes Presence of children No/Yes 0.90/0.10 Mode used for the joint tour Walk or bike/Auto 0.14/0.86 Social recreational trip purpose Recreational/Vacation/Visit friends or site 0.56/0.03/0.41 Traveled on Mon Thu/Fri Sun 0.40/0.60 Number of intermediate stops 1/2/3/4+ 0.66/0.19/0.08/0.07 Household Attributes Household income (1 18, $ 5K interval) 13.26 (4.62) Presence of children 0.11/0.89 Average age of couple 56.78 (18.79) Owned household/Rented 0.90/0.10 Residential area Urban/Rural 0.75/0.25 Ethnicity White/Other 0.92/0.08 Household worker composition Dual/Single/None 0.33/0.29/0.38 Number of joint tours 1,721 Note: parenthesis is a standard deviation

PAGE 147

147 The fractional time allocation is estimated using a fractional split logit model for the cases with more than or equal to two time windows available, which is consistent with the samples used for the time window selection estimation. As mentioned earlier, t he time fraction is modeled in terms of how much time is consumed for the joint tour given a time window Table 5 6 presents the model output with three alternatives converted into temporal proportions of the tour within the chosen time window ( Time window attributes The time window attributes in the final model specification include number of time windows available, the length of chosen time window, temporal zone of start/end time for the chosen time w indow, and tour engagement types before/after the time window The joint tours with more number of available time windows tend to increase the the length of each time window available, each length will be shorter in the case with more number of time windows than the case with less number. Intuitively, the time window with less time period is more likely to be fully used for the tour engagement than the time window with more time period, whic increase of length of time window. In terms of temporal zone of start time for the time window, the duration the length of duration is relatively higher than the la ter part of the day. Similar patterns

PAGE 148

148 are also T our joint engagement is more likely to be taken place in the later part of the day (we observe the departure times have the highest peak at around 4 8 PM in Figure 5 7), but the joint tours that undertake in earlier time tend to have relatively longer tour durations. In contrast, the length which explains the tour arrival time is get ting close to the end time of time window. Lastly, tour engagement types before the time window were applied only to the such variables could be more directive factors to the nearby alternative, rather than the alternative on far side. By doing so, mandatory and maintenance tour engagements (regardless of gender) before the time windo decreased. Similarly, the maintenance tour engagements (regardless of gender) after recreational joint tours preferably start right after the mandatory and maintenance tour shorter). Socio economic and Tour attributes One soc io economic factor and two tour attributes were turned out statistically significant in the final model specification: household work status, transportation mode used, and number of intermediate stops in the social recreational joint tour. The fractional t ime after the joint tour tends to be decreased (i.e., Post tour) in the dual

PAGE 149

149 worker households. This explains the dual worker households have more likelihood to end the joint engagement as close to the end time of given time window as possible. Clear evide investigated on the variable of non auto mode. This suggests that social recreational joint tours conducted by walking or bicycling tend to less spend times than do by auto modes. Lastly, the time duration for the joint tours is likely increased with the need of more intermediate stops.

PAGE 150

150 Table 5 6 M odel e stimation r esult for fractional time spent Variables Pre tour Tour based Post tour Coeff. t sta. Coeff. t sta. Coeff. t sta. Constant 0.319 1.85 0.906 5.38 Time window Attributes Number of time windows available 0.105 1.67 Length of time block (hours) 0.105 3.82 0.091 4.14 Temporal zone of start time for the time window Early morning (4 am 9 am) 3.322 7.44 0.982 2.96 Late morning (9 am 12 pm) 1.807 4.41 0.529 1.90 Midday (12 pm 4 pm) 1.117 3.28 Evening (4 pm 8 pm) Night (8 pm 4 am) Base Base Base Base Temporal zone of end time for the time window Early morning (4 am 9 am) Late morning (9 am 12 pm) Midday (12 pm 4 pm) Evening (4 pm 8 pm) 0.635 1.86 Night (8 pm 4 am) Base Base Base Base Tour engagement types before the time window Mandatory tour by male 0.741 2.26 Mandatory tour by female 0.392 1.71 Maintenance4 tour by male 0.805 2.00 Maintenance tour by female 0.404 1.77 Tour engagement types after the time window Mandatory tour by male Mandatory tour by female Maintenance tour by male 0.988 3.16 Maintenance tour by female 0.762 2.38 Socio economic & Tour Attributes Household work status Dual worker 0.369 2.05 Transportation mode used Non auto mode 1.584 4.42 1.645 4.95 Number of intermediate stops 0.428 4.18 0.282 3.20 Observations 1,377 joint tours Log likelihood at zero coefficients 4525.2 Log likelihood at convergence 4104.2 Rho squared 0.093

PAGE 151

151 Predictive Assessments The same validation sample used for the predictive assessments of time window selection is utilized for the fractional time spent model. Since the fractional model was estimated based on the time windows chosen from the observation, the aggregate shares by three alternatives are first determined using the final model specification presented in the Table 5 6 above. The observed proportional shares on the three alternatives are: 25.8% (Pre tour), 28.2% (Tour based), and 46% (Post tour). The predicted proporti small sample size for the validation, the overall predictions are not well fit with the observe d shares. For example, the predicted shares appear better fit with the observed shares using the entire sample (i.e., Pre tour (29.1%), Tour based (24.1%), and Post tour (46.7%)). The predictive assessments thus far were provided independently from the two separated models (time window selection and fractional time split). Next predictive assessments focus on the representations of predicted departure/arrival times by inter relating the calculated probabilities from the two separated models. To do so, we first generate a random number ranged 0 to 1 to identify which time window is chosen based on calculated probabilities on each sample. Second, the calculated fractional times are app lied to the time window selected from the previous step to calculate the departure and arrival times. Then the single iteration is tabularized in one hour time segments for a full day to represent time of day distributions. Through the several iterations o f such

PAGE 152

152 processes, we take the averaged frequencies at each time of day to present the tour departure/arrival time distributions. Figures 5 7 to 5 9 present the results of this analysis. There are three distributions across the figures: a dotted red bar rep resents the observed time of day distribution, a horizontally lined blue based on the time window (TW) that actually chosen from the observation (this is not related with the results of time window allocation model such that it purely shows predictive asse ssments for the fractional time split model only), and a solid blue bar based on the TW model. In common, the two predicted time of day distributions are relatively similar, while more discrepancies are observed when comparing to the observed time of day d istribution. This issue is caused by the facts that the fractional split model is not much accurate to replicate the true observed time fractions, and the sample size for predictive assessments is relatively small. The proportions at around 11 AM to 4 PM and at 7 PM for departure time of day distribution (Figure 5 7) are too much under estimated, whereas over estimated at 9 PM to 3 AM. Similarly, the proportions of arrival time of day (Figure 5 8) at around 12 AM and 11 PM to 3 AM are over estimated, wher eas under estimated around at 3 PM to 7PM. Figure 5 9 presents the durations of time calculated using the tour departure time from home and the tour arrival time at home. There are under estimations on the time durations less than 1.5 hour and greater than 5 hour. The over estimations are also observed in the range of between 2 hour and 5 hour.

PAGE 153

153 Figure 5 7. Observed and p redicted d istributions of d eparture t ime from h ome 0 2 4 6 8 10 12 14 16 5am 6am 7am 8am 9am 10m 11am 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm 12am 1am 2am 3am % Joint tours Time of Day Observed Predicted (based on chosen TW) Predicted (based on TW model)

PAGE 154

154 Figure 5 8. Observed and p redicted d istributions of a rrival t ime at h ome 0 2 4 6 8 10 12 14 16 5am 6am 7am 8am 9am 10m 11am 12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm 12am 1am 2am 3am 4am % Joint tours Time of Day Observed Predicted (based on chosen TW) Predicted (based on TW model)

PAGE 155

155 Figure 5 9 Observed and p redicted d istributions of t our d uration 0 2 4 6 8 10 12 14 16 18 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 12 12.5 % Joint tours Tour Duration (hrs) Observed Predicted (based on chosen TW) Predicted (based on TW model)

PAGE 156

156 5.5 Summary This chapter presents a systematic procedure to predict the time of day choice for joint tours. The choice set is generated considering the schedule constraints of both adults. The overall choice process is broken down into two steps: the choice of time window and the choice of tour start and end times within the chosen time window. The data used in this analysis comprise 2,609 joint tours undertaken by couple adult households obtained from the 2009 US National Household Travel Survey. The first step in the overall data assembly procedure was the identification of the available time windows for the joint discretionary tour. Then, the temporal position of the tour within the chosen time window is determined in terms of the duration of time before departure for the tour, the tour duration (from start of tour to end of tour), and durati on of time after the end of tour. All these durations are converted into proportions of the total time window durations. Two models are structured separately to predict the time of day schedule: one estimates in terms of wh ich time window is allocated to t he social recreational joint tour using an unlabeled multinomial logit model, another estimates the time fraction consumed for the joint tour given the chosen time window using a fractional split logit model. A wide set of explanatory variables (i.e., time window attributes, socio economic characteristics, and tour attributes) were considered in the model estimations. Several time window attributes (such as length of time block, temporal zones of time window, and tour engagement types before and after the time window) and socio economic and tour variables ( presence of children, household work status, average age of a couple adult, and transportation mode used ) were estimated to be statistically

PAGE 157

157 significant predicators of the time window allocation patterns. In the case of fractional time spent for joint tours, number of time windows available, the length of time window chosen, temporal zone of start/end time for the time window chosen, tour engag ement types before/after the time window, household work status, transportation mode used, and number of intermediate stops also affected the time fraction Predictive analyses also provide the overall fitness of the estimated models.

PAGE 158

158 CHAPTER 6 CONCLUSIONS 6.1 Background This research endeavors to explain household interactions among household members during daily activity and travel related decision making process However, the interactions have not been fully taken into account in the contex t of discretionary (social recreational) travel than mandatory and maintenance travel. We found that social recreational purpose leads to an increase in interactions among household members and a dominant form of interactions is a trade off between solo an d joint activity participation. Within the broad spectrum of modeling social recreational activities and travel, this research focuses on three major aspects: (1) Modeling activity p atterns, (2) Modeling the c hoice of h ousehold v ehicle, and (3) Modeling th e c hoice of t ime of d ay for j oint t ours. This study uses data from the 2009 National Household Travel Survey conducted in the United States for the estimation of all models. The empirical scope of this work is largely restricted to the analysis of the behavior of couple households (two adults comprising a male female couple with or without children). The rest of this chapter provides an overview of contributions and followed by directions for further research. 6.2 Contributions This research stands on t he operational developments that capture household interactions throughout the household decision making process in pursuing social recreational episodes, rather than developing a new methodology. O verall contributions of this research are indicated in Fig ure 6 1 (gray color) within a conventional ABM

PAGE 159

159 framework This research contributes toward the understanding intra household interactions and developing operational models that c an be applied subsequent to any operational ABM framework 6.2.1 Modeling the Generation of Social Recreational Patterns The generation of social recreational activity patterns are based on four different decision making units and methods: multinomial logit models (household based model with a fully enumerated set of choice altern atives at the household level), multilinear logit models (household based model that considers individual level utilities and group decision making process), parallel choice constrained logit models (person based model ensuring consistency of joint partici pation choices across decision makers), and tri variate binary probit models (multilevel model with inter dependencies captures via correlations). All models are able to effectively capture interactions between household heads for the generation of social recreational activity patterns. We observed a negative Rainfall wa s a clear predict or less pursuing social recreational activities In terms of weight powers, employment status of male heads likely affects joint decisions in social recreational activity participation patterns in working households. In the end, we compared the models in terms of predictive abilities and sensitivities to specific explanatory factors. T he models are empirically similar but differ in terms of the behavioral insights they provide and the effort needed t o build and use them. Such an extensive comparison has not been presented before and the results from this study will serve to inform making trade offs between modeling simplicity and theoretical rigor.

PAGE 160

160 Figure 6 1 An overview of research contributions in an ABM f ramework Employment/School Location Car Ownership Daily Activity Patterns Tour Primary Destination Tour Primary Mode Tour Time of day Trip level Generation/Location/ Mode/Time of day Short term Choices Long term Choices Employment/School Location Car Ownership Primary driver Allocation Daily Activity Patterns Tour Primary Destination Tour Primary Mode Joint Tour: Tour Time of day Trip level Generation/Location/ Mode/Time of day Joint Tours: Vehicle type Choice Short term Choices Long term Choices Solo Tours: Allocate to Primary Vehicle Solo Tour: Tour Time of day

PAGE 161

161 6.2.2 Modeling the Choice of Household Vehicle for Social Recreational Tours Modeling the choice of household vehicle contributes towards understandin g household level vehicle allocat ion pattern s for the choice of vehicle for social recreational tours. T wo models explore household ve hicle allocations. One allocates each vehicle to a primary driver in the household (long term, household level model) so that the right vehicle may be allocated to the solo tours undertaken by individuals. The second model allocates a vehicle for the joint tours (short term, tour level model). Two models are developed for households with two vehicles, making a binary choice. An u nlabeled binary logit approach is proposed for the two models since t here are no common labels that can be applied t o describe the two alternatives: i n some cases, the decision maker may not always perceive the body types to be a choice; it is possible that the options are er er more efficient less tours (with or without children). The more efficient (greater miles per gallon), the newer (newer model year), and the less used ve hicle (larger annual mileages) tends to be allocated to the female head. If a vehicle is newer used more often, and has more fuel efficiency then it is more likely chosen for making joint social recreational activities. 6.2.3 Modeling the Choice of Time of Day for Joint Social Recreational Tours In modeling time of day choice, constructing choice alternatives in the case of joint tours requires the temporal synchronization of travel of all members of the travel party This study describes how the time c onstraints of multiple persons can be effectively accommodated into determining the choice set and, subsequently, the choice of timing

PAGE 162

162 of joint tours. Many previous studies on time of day choice modeling have been dominated by discretizing the continuous t ime in to several segmented time blocks. In this study, an alternate two step approach is proposed. The first model predicts the time window chosen for pursuing the joint discretionary tour and the second model locates the tour within the time window by det ermining the start and end times of the tour simultaneously on a continuous scale. We observed j oint social recreational tours are more likely to take place within the longer time segment available, particularly during the midday and evening periods, and after the engagement of maintenance and mandatory purposes. The time period between the morning and midday is less preferred for the joint tours in dual worker households, whereas morning times are more preferred by older couples. Overall, t he method is sh own to be effective from the standpoint of predictive accuracy. 6. 3 Further Research At the outset of this research it is necessary to address directions for further research in order to become a more useful component of activity based travel demand model system. First, the research scope was limited to couple adult household interactions. In exploring the household interactions, we assumed that children are not decision makers; instead, they are just considered as travel companions. This is critical fo r describing the realistic representations of interactions that take place in various household types. Obviously, the interactions need to be extended to households with more than two adults, including households with single adult and children. For example the pursuits of leisure activities in the households with children would be motivated more frequently by their children. In such cases, the interactions between adults and children

PAGE 163

163 could not be examined in this study. Another aspect of interactions is to explore intra personal interactions among activity engagement types undertaken in a full day. Since discretionary activity participation patterns are perhaps particularly influenced by necessary activities, an explicit model to investigate such interactio ns among activity engagement types in a correlated fashion would further enhance research conclusions On the estimation of decision powers employment status of males governs interactions with dissimilar proportions between male heads and female heads. However, due to the different functional forms of likelihood specifications, the interpretations we re not consistent between the household heads over the MLL and PCL models. This issue may require the development of appropriate modeling methods to ensure consistent interpretations.

PAGE 164

164 LIST OF REFERENCES Abkowitz, M.D. (1981) An analysis of the commuter d eparture time decision. Transportation 10 (3), 283 297. Algers, S., Daly, A., Widlert, S. ( 1997 ) Modeling travel behaviour to support policy m aking. In: Stopher, P., Lee Gosselin, M. (Eds.), Understanding Travel Behaviour in an Era of Change. Stockholm, Pe rgamon, pp. 547 569. Anggraini, R. (2009) Household Activity Travel Behavior: Implementation of Within Household Interactions. Ph.D Disseratation, Technische Universiteit Eindhoven. Bhat, C. R. (1998a) Analysis of Travel Mode and Departure Time Choice for Urban Shopping Trips. Transportation Research B, Vol. 32, No. 6, pp. 361 371. Bhat, C. R. (1998b) Accommodating flexible substitution patterns in multi dimensional choice modeling: formulation and application to travel mode and departure time choice Transportation Research B, Vol. 32, No. 7, pp. 455 466. Bhat, C. R. Frusti, T., Zhao, H., Scho¨nfelder, S., Axhausen, K.W. (2004) Inte rshopping duration: an analysis using multiweek data. Transportation Research Part B 38 (1), 39 60. Bha t, C. R. and Koppelman, F. S. (2003) Activity Based Modeling of Travel Demand. International Series in Operations Research & Management Science. Vol. 56. pg. 39 65. Bhat, C.R., and S. Sen (2006), Household Vehicle Type Holdings and Usage: An Application of the Multiple Discrete Continuous Extreme Value (MDCEV) Model, Transportation Research Part B, Vol. 40, No. 1 pp. 35 53. Bhat, C. R. and S ingh S K. (2000) A Comprehensive Daily Activity Travel Generation Model S ystem for Workers. Transportation Research A, Vol. 34 : 1 22 Bhat, C. R. and Steed, J.L. (2002) A Continuous Time Model of Departure Time Choice for Urban Shopping Trips, Transportation Research Part B, Vol. 36, pp. 207 224. Black, C. S. and Schreffler, E. N., (2010) Understanding Transport Demand Management and Its Role in Delivery of Sustainable Urban Transport. Transportation Research Record 2163: 81 88. Blair, S. L. and Lichter, D. T. (1991) Measuring the Division of Household Labor: Gender Segregation of Housework among American Couples. Journal of Family Iss ues Vol. 12, No 1. Bowman, J. L. and Bradley, M. (2006) SACSIM/05 Activity Based Travel Forecasting Model for SACOG Freaturing DaySIM the Person Day Activity and Travel Simulator, Technical Memo Number 7, Time of Day/Activity Scheduling Models, Draft 2

PAGE 165

165 Bra dley, M Bowman J. L. and Griesenbeck B. (2010) SACSIM: An applied activity based model system with fine level spatial and temporal resolution, Journal of Choice Modeling (publication pending, http://jbowman.net/refereed/2010.Bradley_Bowman_Griesenbeck.SacSim.pdf ) Brayfield A. A. ( 1992 ) Employment resources and housework in Canada. Journal of Marriage and Family 54: 19 30 Brines J. (1993) The exchange value of housewo rk. Ration. Soc. 5: 302 340. Cao, X., Mokhtarian, P.L. and Handy, S.L. (2006) Neighborhood Design and Vehicle Type Choice: Evidence from Northern California, Transportation Research D, 11: 133 145. Chandrasekhara n, B. and Goulias, K. G. (1999) Exploratory Longitudinal Analysis of Solo and Joint Trip Making Using the Puget Sound Transportation Panel. Trans portation Research Record 1676: 77 85. Choo, S. and Mokhtarian, P. L. (2004) What type of vehicle do people drive? The role of attitude and lifestyle in in fluencing vehicle type choice. Transportation Research Part A, Vol. 38: 201 222. Cools, M., Moons, E., Wets, G. (2010) Assessing the impact of weather on traffic intensity. Weather Clim. Soc. 2,60 68 Corfman, K. P., Gupta, S., (1992) Mathematical models o f group choice and negotiation. Working paper No. 67. Dhaka, N. S. (2009) Modeling Face to Face and Internet Based Social Activity Participation Decisions. Master Thesis, University of Florida. Dravitzki, V., Lester, T., & Walton, D. (2 009) Social/recreat ional travel and its influence on transport's greenhouse gas emissions. Proceedings of the 32nd Australasian transport research forum (ATRF), Ettema, D. and Timmermans, H. (2003) Modeling Departure Time Choice in the Context of Activity Scheduling Behavio r, Transportation Research Record, 1831. No. 03 2425, pp. 39 46. Engagement Using a Model System of Activity Travel Behavior and Time Use. Transportation Research Rec ord 1676: 11 19. Models: A 24. Genz, A. (1992) Numerical Computation of Multivariate Normal Probabilities Journal o f Computational and Graphical Statistics, 1, 141 150

PAGE 166

166 Gliebe, J. P., and Koppelman, F. S. (2002) A model of joint activity participation between household members. Transportation (29): 49 72. Gliebe, J. P., and Koppelman, F. S. (2005) Modeling household act ivity travel interactions as parallel constrained choices. Transportation (32): 449 471. Golob, T. F., Kim, S. and Ren, W. (1996) How Households Use Different Types of Vehicles: A Structural Driver Allocation and Usage Model. Transportation Research A, Vo l. 30, No. 2: 103 118 Golob, T. F. and McNally, M. G. (1997) A model of activity participation and travel interactions between household heads. Transportation Research Part B 31 (3): 177 194 Journal of Family and Economic Issues. Vol. 12, No. 3: 253 294. Goulias, K., Chen, Y., Bhat, C., Paleti, R., Pendyala, R., and Konduri, K. (2011) Activity Based Travel De mand Model SimAGENT Progress Report to the Modeling Task Force, Southern California, Association of Government, http://www.scag.ca.gov/modeling/mtf/prese ntations/032311/mtf032311_2012rtp SimAgent.pdf Guo, J.Y., Srinivasan, S., Eluru, N., Pinjari, A., Copperman, R., and Bhat, C. (2005) "Activity Based Travel Demand Analysis for Metropolitan Areas in Texas: CEMSELTS Model Estimations and Prediction Procedures 4874 Zone System CEMDAP Model Estimations and Procedures, and the SPG Software Details," Report 4080 7, prepared for the Texas Department of Transportation. Hajivassiliou, V. A., an d McFadden, D. (1998) The Method of Simulated Scores for the Estimation of LDV Models, Econometrica, Vol. 66, No. 4: 863 896. Harsanyi, J. C., (1955) Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility. Journal of Political Economy. Vol. 63, No. 4: 309 321. Hensher, D. A., Rose, J. M., Greene, W. H. (2005) Applied Choice Analysis A Primer. Cambridge University Press. Hess, S., Polak, J.W., Bierlaire, M. (2005) Functional approximation to alternative specific constants in time period choice modeli ng, CTS Working paper, Centre for Transport Studies, Imperial College London. H unt, K. and Petersen, E. (2005) The Role of Gender, Work Status and Income in Auto Allocation Decisions, Proceedings of the Annual TRB Meeting, Washington, D.C. Jovicic, G., (20 01) Activity Based Travel Demand Modelling A Literature Study, Danmarks TransportForskning.

PAGE 167

167 Kamo Y. ( 1994 ) Division of household work in the United States and Japan. Journal of Family Issues Vol. 15: 348 78 Kato, H. and Matsumoto, M. (2009) Intra househol d interaction in a nuclear family: a utility maximizing approach. Transportation Research Part B 43 (2): 191 203. Keeney, R. L., (1972) Utility functions for multiattributed consequences. Management Science. Vol. 18, No. 5: 276 287. Kitamura, R. (1983) Ser ve Passenger Trips as a Determinant of Travel Behavior. Recent Advances in Travel Demand Analysis edited by S. Carpenter and P. Jones, pp. 137 162. Kitamura, R. (1996) Application of Models of Activity Behavior for Activity Based Demand Forecasting. Activ ity Based Travel Forecasting Conference. New Orleans, 1996. Pg. 119 150. Komma, A. (2008) Continuous Departure Time Choice Models for Home to Work Commute, University of Florid, Master Thesis. Lee, B., and Timmermans, H. (2007) A Latent Class Accelerated Hazard Model of Activity Episode Durations. Transportation Research Part B, Vol. 41, pp. 426 447. McCafferty, D., Hall, F.L. (1982) The use of multinomial logit analysis to model the choice of time to travel. Economic Geography 36 (3), 236 246. McFadden, D (1978) Modeling the Choice of Residential Location. In Spatial Interaction Theory and Residential Location, edited by A. Karlquist et al., 75 98. Nothr Holland: Amsterdam. Meka, S., Pendyala, R., and Kumara, W. (2002) A structural equation analysis of wi thin household activity and time allocation between two adults. Paper presented at the 81 st Annual Meeting of the Transportation Research Board, Washington, D.C. Mokhtarian, P., I. Salomon, and S. Handy. ( 2006 ) The impacts of ICT on leisure activities and travel: A conceptual exploration. Transportation 33:263{289. Townsend, T. A. (1987) Estimation of Relationships for Limited Dependent Variables. Econometrica 26, 24 36. Nonworker 93. PB Americas, Inc. (2009) Activity Based Travel Model Specifications: Coordinated Travel Regional Activity Based Modeling Platform (CT RAMP) for the Atlanta Region Atlanta Regional Commission Petersen, E. and Vovsha, P. (2006) Intrahousehold Car Type Choice for Different Travel Needs, Transportation Research Record, Vol 1985: 207 219

PAGE 168

168 Pinjari, A., N. Eluru, S. Srinivasan, J.Y. Guo, R.B. Copperman, I.N. Sener, and C.R. Bhat. (2008) CEMDAP: Mode ling and Microsimulation Frameworks, Software Development, and Verification, TRB 87th Annual M eeting CD ROM, Washington, D.C. Presser H B. ( 1994 ) Employment schedules among dual earner spouses and the division of household labor by gender. American Socio logical Review Vol. 59 No. 3 : 348 64 Rich, J. H. ( 2001 ) Modeling two worker househo lds. Paper presented at the 9 th I nternational Association for Travel Behavior Conference, Gold Coast, Queensland, Australia, 2 7 July. Reichman, S. (1976) Travel Adjustmen ts and Life Styles: a Behavioral Approach. In Stopher, P & Meyburg, A (eds) Behavioral Travel Demand Models. Lexington MA, Lexington Books, D. C. Health. Ross, CE. (1987) The division of labor at home. Soc. Forces 65: 816 833. Sale h, W., and Farrell, S. (2005) Implications of congestion charging for departure time choice: work and non work schedule flexibility. Transportation Research A, Vol. 39, pp. 773 791. Schlich, R., Schonfelder, S., Hanson, S., and Axhausen, K. W. (2004) Structures of Leisure Travel : Temporal and Spatial Variability. Transport Reviews. Vol. 24 No. 2, pp. 219 237 Scott, D. M. and Kanaroglou P. S. (2002) An activity episode generation model that captures interaction between household heads: development and empirical analysis. Transpor tation Research Part B 36 (10): 875 896. Sener, I. N., Eluru, N., and Bhat, C. R. (2010) On Jointly Analyzing the Physical Activity Participation Levles of Individuals in a Family Unit Using a Multivariate Copula Framework. University of Texas at Austin. T echnical Report 161020 1. Shelton B. A. (1992) Women, Men, and Time: Gender Differences in Paid Work, House work, and Leisure. Westport, CT: Greenwood. Shelton B. A. and John, D. (1996) The Division of Household Labor. Annu. Rev. Socio. 22: 299 322. Shifta n, Y., Ben Akiva, M., Proussaloglou, K., Jong, G., Popuri, Y., Kasturirangan, K. and Bekhor, S., (2003) Activity Based Modeling as a Tool for Better Understanding Travel Behaviour. TATBR Conference paper. Lucerne, 2003. Simma, A. and Axhausen K. W. (2001 ) Within Household Allocation of Travel: Case of Upper Austria. Transportation Research Record. Vol. 1752: 69 75.

PAGE 169

169 Sivakumar, A. and Bhat, C. (2002) A Fractional Split Distribution Model for Statewide Commodity Flow Analysis", Transportation Research Record Vol. 1790, pp. 80 88. Small, K.A. (1982) The scheduling of consumer activities: Work trips. American Economic Review 72 (3), 467 479. Srinivasan, K., and Athuru, S. R., (2005) Analysis of within household effects and between household differences in main tenance activity allocation. Transportation (32): 495 521. Srinivasan, S. (2004) Modeling Household Interactions in Daily Activity Generation. Ph.D Dissertation. University of Texas, Austin. Srinivasan, S. (2007) Fractional Split Models of the Time Use Patterns of Elderly Americans. For presentation at the International Association for Time Use Research Conference, October 17 19, Washi ngton D.C., U.S.A. Srinivasan, S. and Bhat, C. R. (200 5) Companionship for leisure activities: an empirical analysis using the American time use survey. Innovations in Travel Demand Modeling Conference. Austin, Texas. Sriniv asan, S. and Bhat, C. R. (2006) A multiple discrete continuous model for independent and joint discretionary ac t i v i t y participation decisions. Transportation ( 33 ): 49 7 515 Steed, J. and Bhat, C. (2000) On Modeling Departure Time Choice for Home based Social/Recreational and Shopping Trips. Transportation Research Record, Vol. 1706, pp. 152 159. Timmermans, H., Arentze, T. and Joh, C. (2002) Analyzing Space Time Behavior: New Approaches to Old Problems. Progress in Human Geography 26, 2. Pg. 175 190. Vance, C. and Iovanna, R. (2007) Gender and the Automobile: An Analysis of Nonwork Service Trips. Transportation Research Record, Vol. 2013: 54 61 Vovsha P. and Bradley, M. (2004) Hybrid Discrete Choice Departure Time and Duration Model for Scheduling Travel Tours. Transportation Research Record, Vol. 1894: 46 56. Vovsha, P. and Pertersen, E. (2007) Incorporating Car Type Preferences and Intrahousehold Car Allocation into Travel Demand Models, TRB 86th Annual Meeting Compendium of Papers CD ROM Vovsha P ., Petersen E & Donnelly E (2003) Explicit modeling of joint travel by household members: st atistical evidence and applied approach. Transportation Research Record 1831: 1 10.

PAGE 170

170 Vuong, Q. (1989) Likelihood Ratio Tests for Model Selection and Non nested Hypotheses Econometrica, Vol. 57, No. 2, pp. 307 333. Vyas, G., Paleti, R., Bhat, C., Goulias, K, Pendyala, R., Hu, H. Adler, T., and Bahreinian, A. (2012). A Joint Vehicle Holdings (Type and Vintage) and Primary Driver Assignment Model with an Application for California. TRB 91st Annual Meeting Compendium of Papers DVD. Wang, R. (1997) An Activity Based Trip Generation M odel. Ph.D Dissertation, University of California Irvine. Wen, C. H., and F. S. Koppelman (1999) Integrated Model System of Stop Generation and Tour Formation for the Analysis of Activity and Travel Patterns. Transportation Research Record 1676: 136 144. Y e, X. and Pendyala, R.M. (2005) A Model of Daily Time Use Allocation Using Fractional Logit Methodology. In H.S. Mahmassani (ed.) Transportation and Traffic Theory: Flow, Dynamics, and Human Interaction. Pergamon, Elsevier Science Ltd., Oxford, UK, pp. 507 524. Zhang, J., Timmermans, H. J.P. and Borgers, A. (2002) Utility maximizing model of household time use for independent, shared, and allocated activities incorporating group decision mechanisms. Transportation Research Record 1807: 1 8. Zhang, J., Timmermans, H. J.P. and Borgers, A. (2004) Model Structure Kernel for Household Task and Time Allocation Incorporating Household Interaction and Inter Activity Dependency. Paper presented at the 83rd Annual Meeting of Transportation Research Board, Washi ngton, DC. Zhang, J., Timmermans, H. J.P. and Borgers, A. (2005) A model of household task allocation and time use. Transportation Research Part B 39 (1): 81 95. Zhang, J., Fujiwara, A. (2006) Representing household time allocation behavior by endogenously incorporating diverse intra household interactions: A case study in the context of elderly couples. Transportation Research Part B 40(1): 54 74. Zhang, J., Kuwano, M., Lee, B. and Fujiwara, A. (2009) Modeling household discrete choice behavior incorporat ing heterogeneous group decision making mechanisms. Transportation Research Part B (43): 230 250. ( 2011 ).

PAGE 171

171 BIOGRAPHICAL SKETCH Kwangkyun Lim received his Ph.D from the University of Florida in the fall of 2012. He D epartment of S tructural E ngineering at the Seoul National University of Technology in 2002. A D epartment of R ailway M anagement & P olicy at the same school was accomplished in 2006. He entered Texas A&M University, College Station, in August 2007 and graduated with a degree of master in Transportation Group in December 2008 He began his doctoral study in Civil Engineering at the University of Florida since Aug ust 2009.