This action might not be possible to undo. Are you sure you want to continue?
Direct and Indirect Effects on Weekend Travel Behaviors
Tae Youn Jang* and Jee Wook Hwang**
Received June 10, 2008/Revised 1st: October 16, 2008; 2nd: February 1, 2009/Accepted February 3, 2009
The purpose of this study is to determine the causal factors that influence travel behavior. In particular, the joint relationships between trips, activities and activity durations are researched by examining the effects of personal and household attributes. In order to meet this objective, the total, direct and indirect effects of individual attributes on travel behavior are estimated using structural models. The statistical assumption and the interpretations of coefficients in classical causal analysis are similar to those in linear regression models. Specifically, the direct effects, as estimated coefficients, are only offered and interpreted. However, in travel behavior that results from complicated relationships among trips, activities and individual attributes, the indirect effects among factors related to travel behavior cannot be disregarded. The indirect effects through other intervening factors, in addition to the direct effects, may cause the total effect of specified factors on travel behavior. If only the direct effect is considered for analysis, the causal relationships among factors may not be able to be adequately understood. The advantage of using structural models is that they are able to estimate, in addition to direct structural effects, the indirect effects through other intervening factors. The study also assumes that Saturday travel behavior has an effect on Sunday travel behavior. The subdivided indirect effects of the personal and household attributes on trip generation, activity frequency, and activity durations are empirically analyzed in detail. Keywords: causal relationship, direct and indirect effects, activity duration, structural equations model, travel behavior ···································································································································································································································
1.1 Background Activity-based analysis has received a great deal of attention for its ability to effectively estimate various transportation policies and thereby provide a fundamental understanding of travel demand. Jones et al. (1990) stated that in some instances the forecasts of trip-based models have proven to be inaccurate, and this seems to be the result of a misspecification. Specifically, these forecasts were based on an inappropriate representation of travel behavior relationships, often due to a failure to recognize the existence of linkage among and between trips and activity participation. Because activity-based analysis approaches these problems through a deeper understanding of travel behavior, it is important to determine decision factors of travel behavior by using activity-based analysis and to examine causal relationships among various factors. These relationships are considered as the effects on activity participation. Several researchers have successfully determined the causal factors on travel behavior, based on individual activities. The studies are fulfilled for the relationship between the activity pattern in the trip chaining behavior and individual characteristics such as age, sex, working status, income, lifecycle, car ownership, and residential location (Adler and Ben-Akiva, 1979; Kitamura,
1985; Jean-Claude and Isabelle, 1987; Golob and Hensher, 2007; Hensher, 2007). By examining effects in the model system, Lu and Pas (1999) tried to better capture the relationships between socio-demographics, activity participation and travel behavior. However, their research did not consider latent relationships among elements. Lee and McNally (2006) summarized an investigation on the dynamic processes of activity scheduling and trip linkage and used the ordinal regression models to find the personal characteristics such as gender, marital status, and number of children that are pertinent to the scheduling horizon of activity. By using analysis on inter-shopping duration, Bhat et al. (2004) examined the determinants of the regularity and frequency of the shopping activity participation behavior of individuals. In their study, the effect of the interaction between household members on activity behavior was represented in the form of simple factors such as marital status, spouse’s employment, and household structure. The use of the concepts of activity time-use constraints has shed new light on many aspects of activity-based analysis. Knowing how an individual allocates time to different types of activities can identify the activity participation of an individual. An aspect of an activity-based approach is that it includes the concept of activity durations according to activity purpose. One of the advantages of activitybased analysis is that it is capable of explicitly incorporating the
*Member, Professor, The Research Center of Industrial Technology, Dept. of Urban Engineering, Chonbuk National University, Jeonju 561-756, Korea (Corresponding Author, E-mail: firstname.lastname@example.org) **Professor, Dept. of Urban Engineering, Chonbuk National University, Jeonju 561-756, Korea (E-mail: email@example.com) − 169 −
activities and activity durations are analyzed. activities and activity durations that influence travel behavior. The study empirically emphasizes the total. he/she will participate in that activity the next day. direct and indirect effects of personal and household attributes on travel behavior. The procedure is estimated by using a covariance structure model. 1. to causal relationship among travel mode. The statistical assumption and the interpretations of coefficients in classical causal analysis are similar to those in linear regression models (Mueller. Thirdly. Research Methodology 2. Activity timing and duration are considered to be important factors of activity-based travel modeling systems (Steed & Bhat 2000. The advantage of using structural models is that they are able to estimate. If the indirect effects are disregarded. The study considers a number of trips. The SEM technique has widely used for travel behavior research because of its flexibility in finding relationship among the observed and latent variables. Firstly. 1. 2. there is a difficulty in finding complex relationships between individual attributes and travel behaviors. the task is to explain the causal relationships between the observed variables. The degree of causal relationship is considered as the direct and indirect effects. The development of the LISREL program (Joreskog and Sorbom. and Ruiz and Timmermans (2007) analyzed activity duration to resolve scheduling conflicts. If the specified activity in which the individual plans to participate cannot be performed because he/she spends more time on other prior activities. direct and indirect effects of personal and household attributes on travel behavior in weekend. The SEM technique is applied to panel trip diary data (Golob 1990). previous studies are based on the direct effects of personal and household attributes on travel behavior. the number of trips may not increase a great deal.1 Travel Behavior Model The purpose of constructing the covariance structure model is to determine the degree of effect that personal and household attributes have on travel behavior. which may have closer relationships to each other in terms of how they affect travel behavior. As a result of the use of activity-based analysis. Recently.2 Objectives Fig. activity and travel patterns (Jang 2003). the indirect effects from one or more intervening attributes. 2. Goulias (2002) mentioned that time was a finite and critical resource that was consumed in the engagement of activity and travel. and activity durations as the elements of travel behavior. 1996). Especially. The coefficients as direct effects of variables are only offered and interpreted in those models. 2002) has facilitated the objectives for this study. 2000. Roorda and Ruiz (2008) use a structural equations model technique with latent variables to explore short. 1 shows the concept and procedure used for this study. 2004). Trips and activities depend on the activity duration times. activities. 1996). Secondly. 1989). That is. There are three endogenous variables and two exogenous variables. an analysis is carried out of the effects of Saturday travel behavior on Sunday travel behavior. the effects of individual attributes on trips. Individuals perform mainly discretionary activities during the weekend. Y is a column vector of endogenous variables. Study Concept In equation. A number of trips may also be dependent on activity frequency. It could be assumed that activity durations affect activity frequency during the day. as indicated by the covariance among these variables. (1) shows the structural equations based on Fig. which allows a simultaneous estimation of a series of specified relationships.. the relationships between travel behavior and personal and household attributes have been widely analyzed. Y = BY + ΓX + ξ (1) ζ1 y1 0 β12 β13 y1 γ11 γ12 χ1 + ζ2 y2 = 0 0 β23 y2 + γ21 γ22 χ2 y3 ζ3 0 0 0 y3 γ31 γ32 Fig. Eq. X is a column vector of exogenous variables measured as deviations − 170 − KSCE Journal of Civil Engineering . and to the impacts of personal characteristics and the spatial structure on mode choice (Simma et al.Tae Youn Jang and Jee Wook Hwang factor of time into the travel modeling process (Pas & Harvey 1997). Statistically. the causal relationships between the elements influencing travel behavior are also studied. the causal relationships influencing travel behaviors may not be able to be adequately understood. the study examines the total. Kitamura et al. to trip generation (Golob. (2007) captured tour mode and activity choice interdependences by logit model. Travel behavior can be directly influenced by individual attributes and can also be indirectly influenced through other intervening attributes. Specifically. even if activity frequency increases.and long-term dynamics in activity scheduling. 2001). Krygsman et al. 2002. However. according to the trip linkage. in addition to direct structural effects. effects of Saturday travel behavior on Sunday travel behavior is examined by a structural equations model (SEM) technique. Pendyala et al. It has also been suggested that activity duration is a significant factor in the analysis of travel behavior. When using these models. to vehicle usage and driver allocation (Golob et al.. It is assumed that activity durations influence trip generation and activity frequency. This present study offers an analysis of the joint relationships between trips.
marriage. Fig. (2). Among them. Exogenous variables are individual and household attributes. considering only structural coefficients.Direct and Indirect Effects on Weekend Travel Behaviors Table 1. y3 (dirct effect) and y3 (indirect effect) as shown in Eq.000 persons are surveyed. In describing the definition of effects in structural models. Since the correlation matrix provides adequate results that assist the understanding of the relationships among variables. 2. was introduced in Korea on July 1. Golob. y3 (total effect) = y3 (dirct effect) + y3 (indirect effect) (2) y3 (total effect) (3) = ( γ31 + γ32 ) + ( γ11 β21 β32 + γ21 β32 + γ11 β31 + γ12 β21 β32 + γ32 β32 + γ12 β31 ) 2. sport. the direct effect of an endogenous (or an exogenous) variable on another endogenous variable is defined as the structural coefficient (metric) linking the two variables (Mueller. Table 3 shows a basic analysis of variables in a survey of 568 males and 116 females. 3 / May 2009 indirect effect. family members. a number of activities. Data on their trips. on Sunday females participated in more activities than males. A particular indirect effect of one variable on another variable through one or more intervening variables is defined by the product of associated structural coefficients that link the variables in the particular structural chain. while total durations for sport and leisure between the two days may be similar. age. y2 and y3 are the observed endogenous variables and x1 and x2 are the observed exogenous variables. the effects of x1 and x2 on y3 are composed of y3 (total effect). Originally. A policy. education level and whether there are children under the age of 12. B is a matrix of structural coefficients from endogenous to other endogenous variables. which meets the objectives of the study. (3). 2 shows the example of a possible path diagram used to calculate the effect values. 1. which contains the subdivided indirect effects. and ζ is a column vector of error terms of endogenous variables. The correlation matrix (the simple standardized variance/covariance) is calculated by dividing the variance/covariance by the standard error. The structural model differs from other multivariate analysis in the use of data of variance/ covariance or a correlation matrix. However. This policy has now been firmly established. the sum of all particular indirect effects is defined as the total indirect effect. 1996. Process for Subdivided Indirect Effects on y3 Path x1 → y1 → y2 → y3 x1 → y2 → y3 x1 → y1 → y3 x2 → y1 → y2 → y3 x2 → y2 → y3 Fig. Endogenous variables include a number of trips generated. This policy therefore influences weekend travel behavior. Definition of Effects between Variables x2 → y1 → y3 Subdivided indirect effects γ 11β 21β 32 γ 21β 32 γ 11β 31 γ 12β 21β 32 γ 22β 32 γ 12β 31 from their means. Information about all activities was collected during a designated 48-hour period in a weekend. 2004.33 km² in 2006. 13. and leisure.05. activities and activity durations during the weekend were surveyed by mail and interviews in June and July of 2006. Table 1 shows the process used to calculate the subdivided indirect effects on y3. Jeonju City has a population of approximately 620. and activity durations according to purposes on Saturday and Sunday respectively. The computational definitions of the direct and indirect effects of an exogenous variable on an endogenous variable in a structural equation model are based on the fact that the covariance between two variables can be completely decomposed and written as functions of the model-implied parameters. Korea. trips. y1. 2003). For example. The first parenthesis represents the total direct effects and the second parenthesis represents the total Vol. 436 workers invest non-zero time for all weekend activities this study classifies. Activity durations are the total time that an individual spends for shopping. that enforces the week to consist of 5 working days and 2 non-working days. Exogenous variables are gender. Covariance provides sound results for reciprocal comparison among variables because it is influenced by the estimated dimension. which is statistically significant at a level of p=<0. the sum of the direct effect and the indirect effect is defined as the total effect of an exogenous variable on an endogenous variable.3 Basic Analysis on Variables The study is based on an examination of weekend travel behaviors of 436 workers who live in Jeonju City. is as shown in Eq. There are no statistical differences between males and females in Saturday trips and activities and in Sunday trips. Males spend more time − 171 − . It is assumed that weekend discretionary activities have closer relationships to each other than weekday obligatory ones. 2. The total effect on y3. In general. Saturday shopping durations were longer than those on Sunday. Finally. Γ is a matrix of structural coefficients from exogenous to endogenous variables. γ s are the direct effects of exogenous variables on endogenous variables and β s are the direct effects of endogenous variables on endogenous variables. activities and total duration times are classified into endogenous variables. it is utilized as the input data format. Activities and activity durations may be the intervening variables which cause indirect effects. No. As shown in Table 2.000 and an area of 206.2 Definition of Effect Analysis The study applies the structural modeling process in order to determine the direct and indirect effects among variables. income.
4 36.2 44. The GFI (goodness of fit index) is not influenced by the change of sample size and the violation of multivariate normal distribution. graduate school=3) Number of household members Total household income (Korea currency :10. the model has a tendency to not fit the data. 0-200=1.2 28.5 182.0c 25.8a 133. (min.40 3.1 023.1 133.25a 1.46 1.0b 25.9 208.0 147.1c 38.9 157.03ª 2.44 1.3 243.55 3.90a 1.8 26.00ª 2.71 2.0 195.9 244.25 1.28 3.9.33 3.2 28.7ª 191.9 100. fitness indices are used in order to appraise whether the data fits.34 1. single=0) Age of individual Education level (under high school=1.25 1.8 260.3 052.9b 199.59 1.0 26. Youngest people have a tendency to take more trips and to participate in more activities. female=0 Marital status (married=1.50 1.49 2. cycle.6ª 181.7 187.85 1.77 2.0 076.7 31.8a 157.90 2.06a 3.6 28.2 299.7 215.3a 138. As the size of the family increases. Mean Values for Variables Saturday Variable Male Female Married Single 0-35 36-45 46+ 0-3 4 5+ 0-200 201-300 300+ Freq.) duration (min.8ª 194. hiking.3 224.01 2.8 32.2 188. college=2.0 231.25c 1.) 57.) 058.) 34. tennis. etc.3 20.16b 2.6 035.98 3.15ª 1.0 Leisure (min.1 Assessing Model Fit When a covariance structure model is constructed for providing data.7b 043.4 195.07 1. etc.3 104.75a 2.9 049.4 41.) 159.0 178.18 2.000 won.8 046.12a 3.09 3.2 38.4 209.67 2. 1.26 3.88 1.) 185.04 1. the model is able to find a fit for the data.4 156.23ª 1.9ª 142.5a 17.0 167.29ª 2.7 Sport (min.0 215.2 222. tea time.1 28.53 1.7 43. movie. There are some exceptions but these variables also show the statistical significances in most activity durations.55 2.05 3.9a 088.) Male=1. If the sample size is over 200 and the value of the GFI is at least 0.7a 222.6 135.12 3.4 172.54 Act.63 Sunday Shop.1 shopping during the weekend because they may have less chance to shop during weekdays due to their obligatory working activities. 300+=3) Existence of children under 12 in the household (1.83 2. 3.93 3.6 25.8 26. Results 3.3 177. education and whether a family has children show statistical significances in testing differences among groups for trips and activities.1 47. Married people make more trips and participate in more activities than single people during the weekend.13 2.53 2.31 1. Household income.2 186.24ª 2.5 Endogenous variables Exogenous variables Gender Marital Status Age Family Income HighEducation College Graduate Child a Yes No Significant at p=<0.) Leisure activity (church. If the chi-square value is high.38 3.5 23.1ª 186.2 Trips 2. meeting friends.2 33.2 037.62a 3.3a 147.2 31.60 2.8 074.3 168.14a 1.5a 206.01. shopping durations also increase.73 1.27b 2. 0) Table 3.0 36.2 Leisure (min.84 1.) duration (min.34b 1.1 017.9 39.4 067.9 293.05.1c 051.5ª 193.53ª 2.3 Sport (min.7 20.29 3.2c 211. The chi-square value and the probability value are calculated based on normality. c Significant at p=<0.81 Act.1ª 022.38 2.7a 083.02 2.5a 189.0 42.54a 1.7 15.27 1.) Sport activity (walking. 1.1 234.6a 020.1 169.0 22.75a 2.12 Shop. b Significant at p=<0.31 2. those indices with a strong point and those with a weak point on the hypothesized model should be used for hypothesis tests. The family is a significant factor influencing Sunday trips and activities. 568 116 412 272 280 236 168 248 300 136 348 212 124 164 460 060 352 332 Trips 3.Tae Youn Jang and Jee Wook Hwang Table 2.2 179. Variable Definition Variable Trips Activities Shopping Sport Leisure Gender Marriage Age Education Family Income Child Definition Saturday & Sunday Number of trips Number of activities Shopping activity duration (min.19 2.66ª 2. (min.) 239.5 212.28a 1. restaurant.95ª 3. The KSCE Journal of Civil Engineering − 172 − .3 179.5ª 103.54 1. golf. Since indices need to be specified.4 140.3 17.9a 12.0 24.9b 41. 201-300=2. Married people spend more time shopping and less time on leisure activities.
920 0.005 -0.053 -0.021 0.022 -0.193 -0.0 but rarely may be greater than 1.164 -0. The initial model is based on the concept of the study.015 0.952 0. If CN (critical N) is generally over 200. The values for the RMR range from 0 to 1.05 (Byrne. indices of comparative fit typically choose a baseline model that is known to previously provide a poor fit to the data. If fitness is good.053 -0. Because males and married people show a positive sign in the direct effects on trip generation.059 -0. A hypothesis test is performed based on the inverse relationship between the fit and simplicity of the model in order to find an improved model.000 GFI 0.0 to 1.056 -0.258 0.171 Education 0.9 or greater.076 -0.035 0.123 Family -0. The required acceptable AGFI is 0.013 0.005 -0. The NFI (normed fit index).008 0.9 means that the model has a 90% better fit than the null model (Kelloway.048 - CN 27. Many of the indices are based on choosing a model as a baseline and comparing the fit of theoretically derived models to the baseline model.112 -0. 1998).038 Age -0.6 340.05 155.175 Marriage 0. null 0. they tend to generate more trips during the weekend.029 0.72 292. income and education are positively associated with Saturday activities but others are negatively associated with them. income. the causal relationship between variables may not be able to be adequately understood. 2nd value : direct effect.082 -0.000 0.061 0.2 Effects on Trips and Activities Generally.055 0.194 0. 3 / May 2009 − 173 − . family. A null hypothesis shows that the increased fit is not significant at the expense of simplicity. family. An alternative hypothesis is that the increased fit is significant at the expense of simplicity. In activities.010 0. All values have statistical significance at a level of p<0. 3rd value : indirect effect Vol. The difference between the two models is 2 2 tested with χ 2 difference=χ A-χ B and ∆df = dfA – df B . and education are factors with a higher effect on Saturday trips and Table 4.017 0.016 -0.9.193 -0.054 0. The NNFI (non-normed fit index) is in the range from 0. Therefore.101 -0.041 -0.7 hypothesis is rejected and the modified model is chosen.029 0.070 -0.841 0.0.141 -0. The null model is the simplest model that specifies no relationships between the variables composing the model. The RMR (root mean square residual) indicates the average discrepancy between the elements in the sample and hypothesized covariance matrices. the null.065 0. Older aged people tend to generate more Saturday trips and fewer Sunday trips. children.000 0.597 0.091 -0. should be between 0. 1989).050 0. If the direct effects are only considered for analysis as they are in classical models. The most common baseline model is the null model. the coefficients from modeling represent the direct effects.017 0. age. An improved model makes progress by improving the fit at the expense of simplicity.043 0. Age.790 0.026 -0.227 -0.547 0.029 -0.057 0.61 - 3.098 -0.920 - RMR 0.203 0.0 and the required accepted value should be greater than 0.111 -0.31 and is greater than the critical chi-squared statistic (χ 2 =18. An NFI of 0. frequently used to test fitness.48).104 0.109 0.23 117. education and whether a family has children.0 and 1.193 -0.1.980 - AGFI 0. Total.016 Trips Sunday Activities 1st value : total effect.070 0. In Table 4.000 0. then it has a value close to zero.027 0.064 0. it is accepted that the model satisfactorily reflects the observed data. the 2 value of χ difference with ∆df of 7 is 174.107 0.065 0.023 0.103 0.916. 3.022 -0. In addition to the direct effects. Rather than comparing against a model that provides a perfect fit to the data. Assessing Fits Model Null Initial(A) Modified(B) A:B χ 2 df 136 46 39 7 p-value 0. Specifically.31 Table 5.122 -0.005 0. However.92 174. The modified model has 340. Several attributes show greater positive or negative total effects.066 -0.051 -0.197 0.056 0. 13.126 -0. it should be less than 0.074 -0.021 -0.101 -0.120 0.968 - NNFI 0.013 -0. The modified model has reasonable indices. income and education are positively associated with Sunday activities.096 Child -0.010 Income 0.003 0. In Table 4.Direct and Indirect Effects on Weekend Travel Behaviors AGFI (adjust goodness of fit index) is the modified GFI. The modified model is the final model built by adding free parameters in order to improve the fit without changing the concept of the study. No.022 0. The difference between models may be tested with the χ 2 test.61.069 -0.039 -0.022 -0.019 0. Table 5 demonstrates the total and indirect effects of individual attributes on trips and activities.025 -0. Direct and Indirect Effects of Attributes on Weekend Trips and Activities Variable Trips Saturday Activities Gender 0.045 -0. have negative effects on trips.112 0.922 - NFI 0.186 -0.002 -0.025 -0. initial and modified models are compared in order to find an improved model.00.01.
037 0.029 0.144 0.111 -0.016 -0.002 through sport and leisure durations.071 0. with the exception of Saturday shopping.012 0.020 through activities.038 -0. Gender has a positive subdivided indirect effect of 0.062 0. Family.009 0.188 -0.000 -0.033 through shopping duration while it has negative subdivided indirect effects of -0.153 0.058 -0.260 0. 2nd value : Direct effect. respectively.000 Income -0. For example.055 0.157 0.119 -0. Family has the same patterns in Saturday activity durations as marriage.137 0.067 0. Furthermore. On Sunday.157 -0.120 0. as shown in Table 5.104 0.105 0. 3.084 0. Direct and Indirect Effects of Attributes on Weekend Activity Durations Variable Shopping duration Saturday Sport duration Leisure duration Shopping duration Sunday Sport duration Leisure duration Gender 0.057 0.011 0. based on Fig.001 0.001 0. with the exception of Sunday shopping duration.071 0.035 0. age has a negative total effect of -0.064 0.017 0.076 0. and -0.004 -0.003 and 0.000 -0.139 0.041 -0. Marriage is the major factor where more time is spent on most activities but less time for Saturday leisure and Sunday sport.160 0.129 0.000 -0.002 -0. Table 7 shows the subdivided indirect effects of attributes on Saturday trips through activity durations and activity frequency. income and education for trips and age for Sunday activities also have the same tendency.007 -0.040 0.029 0.077 -0.000 0.077 0. males negatively and indirectly influence trips through activities on Saturday.055 -0.4 Subdivided Indirect Effects on Trips and Activities through Durations The subdivided indirect effects on weekend trips and activities through activity durations are manually calculated.044 -0. while income and children have a higher effect on Sunday activities.029 -0.124 0. the total subdivided indirect effects by gender on Saturday trips are 0.052 0. Some attributes show sign differences between the direct effects and the total effects. Several variables have larger indirect effects than direct effects on activity durations.015 -0.004 -0.008 0. Furthermore. family has a positive effect on Sunday shopping duration but has negative signs for Sunday sport and leisure durations. There are no subdivided indirect effects on activity durations for Saturday sport and Sunday leisure.000 Child -0. Specifically. The factor of whether a family has children also has negative effects on all activity durations during the weekend.000 -0.042 0. The factors of Table 6. 2 and Table 3.000 Marriage 0.040 -0. These indirect effects are composed of the subdivided indirect effects.077 0. Total.000 Family 0.000 Age -0. indirect effects can not help being considered in modeling. Males have a positive subdivided indirect effect on Saturday trips through shopping duration.071 -0.030 0.008 0.069 0. age influences Saturday trips more indirectly than directly.037 0. Table 5 demonstrated the indirect effects of individual attributes on trips and activities.001 0.025 -0.138 0. 3rd value : Indirect effect − 174 − KSCE Journal of Civil Engineering . It is assumed that as age tend to make people spend more time for Sunday sport time.008. These sign differences result from the indirect effects of the exogenous variables on the endogenous variables through other intervening endogenous (or exogenous) variable(s) such as activity durations. such as family on shopping duration and age on sport duration.023 0. 3. but are negatively associated with the subdivided indirect effect through sport and leisure durations.071 -0.103 -0. age has a marginal positive direct effect of 0.153 -0.138 0.3 Effects on Activity Durations In Table 6.Tae Youn Jang and Jee Wook Hwang activities.007 0. Ultimately.109 on Saturday trips.000 -0.000 Education -0.007 0. the total and direct effects on each activity duration shows the same signs over all attributes even if some indirect effects have different signs from some attributes.003 0.146 -0.052 -0.028 0.055 0.094 -0.185 -0.011 -0.000 1st value : Total effect.039 -0.020 0.109 -0.025 0. Generally.087 0.049 -0. So.062 0.020 0.101 -0. the positive effect of age on sport duration is mainly shown as an indirect effect rather than a direct effect.002 on Saturday trips but it has a larger amount of negative indirect effect of -0.220 0.022 -0.000 0.240 -0.015 -0.077 -0.007 -0.056 0.111 on Saturday trips through other variables such as activity durations.023 0.029 0.005 0. age has negative effects on durations of all activities except for Sunday sport duration. which is equal to the indirect effect of gender on Saturday trips.013 -0. income and education for trips and gender for activities on Saturday.006 0.026 -0.004 -0.012 0.020 0. The higher income person and the higher educated person are more likely to spend more time for all activities.057 -0. Males may spend a longer time shopping during the weekend and spend more time duration on Sunday sport and leisure than they do on Saturday.028 -0.037 -0. These are age.057 0.260 -0.057 0.113 0.044 0. Therefore.009 0.
0.017 0.064 -0. Family and education also have a posi- tive indirect effect on Sunday trip generation. Specifically.0. Subdivided Indirect Effects of Attributes on Saturday Travel Behavior Variable Shopping duration Sport duration Saturday trips Leisure duration No.007 0.015 0.062 -0.015 -0.164 Marriage 0.186 Education 0.056 Education -0. income and education shows the positive value.0.015 0.045 0.006 -0.112 -0.000 .034 -0.005 -0.0.038 0.119 0.005 0.017 income and education have opposite results to that of gender.000 .001 -0.042 .0. Income and education are positively and indirectly associated with activities through sport and leisure durations.0.002 0.077 0.002 -0.0.002 -0.003 0.004 0.131 .003 -0. Table 8. with the exception of gender. Males influence positively and indirectly Saturday activities through shopping duration.002 0.001 .090 -0.092 0.001 -0.002 0. family and children have negative indirect effects on Sunday activities.066 0.004 0. In total.001 . No.005 .0.0.0.016 0. of Activities Sunday total Shopping duration Sport duration From Saturday Leisure duration No. and education show positive subdivided indirect effects on Sunday activities.006 0.006 .0. Education shows higher subdivided indirect effects on Sunday activities through Saturday travel behavior than Sunday.017 Age -0.002 0.021 . In summary.007 .022 -0.002 0.0.055 Vol. The attribute that generally has the highest positive or negative indirect effects on trips is that of leisure duration.052 -0.0.0.173 0.124 0.000 -0.021 Family 0. gender.0.001 0.112 Income 0.042 -0. family.0. Generally.050 0.033 -0.025 0.165 0.0. 3 / May 2009 − 175 − .003 -0.057 Income -0.018 0. Saturday travel behavior may not significantly influence Sunday activity generation.020 0.053 Age .000 .010 0. Sunday trips are highly and positively influenced through Sunday durations and activities by income which shows the Sunday total indirect effect of 0. In Table 9.087 . 13.0.086 . while age.001 0. of Activities Saturday total Total Gender 0.008 0. marriage and child.011 0.013 0.025 .013 0.014 0. most attributes show positive subdivided indirect effects on Sunday trip generation through activity generation frequency. In the analysis of Sunday total. gender is the most negative and indirect factor on Sunday trips through Sunday activity frequency.001 .111 -0.0.012 -0.0.000 0.001 0.009 .007 -0.015 .013 0.002 .036 .010 .018 .010 .026 .007 0. gender. most indirect effects on Sunday activities are through Sunday activity durations. but negatively influence Saturday activities through sport and leisure durations.249 . the subdivided indirect effects of personal and household attributes on Sunday activity frequency through Sunday activity durations are highly positive by gender and negative by child in Sunday total.012 0.027 .0.056 -0.005 0. of Activities Total Shopping duration Saturday activities Sport duration Leisure duration Total Gender 0.031 0. Age is a negative indirect factor on Saturday activities.000 0.0. marriage.000 0.069 .0.004 -0.019 . In analysis of effects of personal and household attributes on Sunday trip generation through Saturday travel behavior.004 -0.004 .002 -0.0. Marriage and family have positive subdivided indirect effects on activities through shopping and sport durations but have negative indirect effects on activities through leisure duration.015 0.0.0.0.167 .012 -0.0. Subdivided Indirect Effects of Attributes on Sunday Trips Variable Shopping duration Sport duration From Sunday Leisure duration No.249. Age and children show negative indirect effects on trips through all activity durations.0.000 . income.259 Child 0.Direct and Indirect Effects on Weekend Travel Behaviors Table 7.103 -0.027 0.031 0.002 0.0. Table 8 shows the subdivided indirect effects of attributes on Sunday trips through Saturday and Sunday activities and durations.002 -0. Especially.0.0. income and education positively influence Sunday trip generation.036 -0.027 Marriage 0.318 .004 .039 Family 0.011 0.022 0.045 0.000 .028 0.003 .0.012 0.004 0.002 0.029 -0. In analysis of a number of Sunday activities.031 0. but are negatively associated with activities through shopping duration on Saturday.015 Child -0.015 0.037 .0.020 0.031 -0.0.0.
001 0.007 . However.006 . multiplied by the direct effect (-0.035 .086 .0.0.038 Age .0.175 Marriage 0.016 3.009 0. All direct effects between activity durations show negative signs.0.013 0. of Activities Saturday total Total Gender 0. some links are not significant statistically but they may be still needed to calculate indirect effects between variables.090 .000 0.000 0.002 0.010 Income 0.000 .0.0.0.028 0.001 0. It is assumed that the person that participates more in Saturday activities is also active on Sunday.0.037 on shopping duration.035 0. a number of trips decrease as leisure duration increases.Tae Youn Jang and Jee Wook Hwang Table 9.023 .146 . Saturday activities have a major indirect effect on Sunday trip generation with the value Fig.040 0.0. Durations for shopping and sport positively and directly influence Saturday trips. The direct effect from Saturday activities to Sunday activities has a positive sign of 0.011 0.078 0.004 .0. Subdivided Indirect Effects of Attributes on Sunday Activities Variable Shopping duration From Sunday Sport duration Leisure duration Sunday total Shopping duration Sport duration From Saturday Leisure duration No. The explanation of the model is done by effects of coefficients rather than by statistical significance.011 .0. 3.009 .0.0.005 0.002 0. which may mean that an individual participates more in activities of a shorter time.159 .5 Effects among Trips. activity frequency and leisure duration have no statistical significance in direct effects on trips at p=<0.001 0.123 Family 0.001 .320) of leisure on shopping.009 .004 .096 Child 0.0. This indirect effect (0.005 . on Saturday. In particular.112 .0.003 0. leisure duration has a statistical significance in total positive effect on trips. all activity durations show negative direct effects on trips but the total effects become positive due to higher positive indirect effects. However. It has been previously mentioned that Sunday trips and activities may be indirectly influenced by Saturday travel behavior.05.002 0.002 .001 0.0. Sunday trips are highly and indirectly influenced by Saturday leisure duration as the value of 0.0.51.001 0.000 . On Saturday.0.002 0.000 .016 0.006 0.0.0.003 0. That is. the duration increase of one activity causes the decrease of other activity durations. Trips have positive total effects from activity frequency and durations.014 . sport duration has a positive indirect effect of 0.0.025 .002 . On Sunday.0.0.0.0.6 Subdivided Indirect Effects of Saturday Travel Behavior on Sunday Travel Behavior Sunday trips are positively and indirectly influenced by all activity durations of Saturday in Table 10. On Sunday.000 .001 0.003 0.037) of sport on shopping is calculated by the direct effect (-0.006 . 3.001 0.0.012 .174 0.0. which indicates that sport indirectly influences shopping through leisure in duration.014 .091.016 0.0.000 .0.037 0. Direct and Indirect Effects among Travel Behavior Variables − 176 − KSCE Journal of Civil Engineering . Total. 3.061 0. Specifically.171 Education 0.037 0.112) of sport on leisure. However.000 .0. Activities and Activity Durations The relationships between Endogenous variables are shown in Fig.043 0.000 0.007 0.
2 and Table 1. and Axhausen. Schonfelder. I. Acknowledgements This work was supported by The Research Center of Industrial Technology at CBNU. 1. of Sustainable Transportation. Golob. Golob. “Intershopping duration: an analysis using multiweek data.” Geography Analysis.” Transportation Research B. 1. pp.. F. (1990). UK. Furthermore. age has a minimal positive direct effect on Saturday trips. D.007 0. 343-370. 4. D. W. and Hensher. pp. Vol. of Transportation Engineering.” New Developments in Dynamic and Activity-based Approaches to Travel Analysis. F. P. “Causal relationships among travel mode. No. 1. ed. T. K.” Intl J. Vol. Conclusions This study analyzed the total. CRC Press. F. 16-22.” J.. Those people who participate in more Saturday activities are assumed to be active on Sunday because Saturday activity frequency positively influences Sunday activities. while. 23. 243-257.. (1989). 13. V. “Structural equation modeling for travel behavior research. “Some insights into the key influences on tripchaining activity and public transport use of seniors and the elderly.” J. A primer of LISREL. Joreskog. No. activity.294 of 0. and Ben-Akiva. Bhat. Sunday sport and leisure durations show negative direct effects on trips but the total effects are positive. F. 38. the trip generations are positively and directly influenced by activity frequency and activity durations. T. B.020 0.147 Total 0.032 0. D. Vol. A. No. 30. Hensher. Byrne. Golob. 1. New York Berlin Heidelberg. Y. 298-312. age has more effects on a number of trips indirectly than it does directly.032 0. Based on the model results. (2002). “Activity analysis: State-of-the-art and future direction. 1. P.237 0.083 Activities 0... (1979). (2007). T. M. 3 / May 2009 − 177 − . T..012 Leisure duration 0. S. “The trip chaining activity of Sydney residents: A cross-section assessment by age group with a focus on seniors. Jean-Claude.” Transportation Research A. 1. The structural model used in the study has a reasonable fit and satisfactorily reflects the observed data. C.147. pp. 103-118. (1996). SpringerVerlag.. 34-55. All direct effects between activity durations are negative. A. Avebury. Transportation system planning: Methods and application. K. No.” Transportation Research B. F. (2007). Zhao. this study suggests that indirect effects should be reflected for an adequate understanding of travel behavior models. based on Fig. R. These sign differences are due to the indirect effects of the exogenous variable on the endogenous variable through other possible intervening endogenous (or exogenous) variable(s). pp. pp.. 3. 39-60. Vol. the study tried to determine the causal factors effecting travel behavior. Vol. Aldershot.005 0. (2003). H. Vol. T. T. Vol. E. 4. 53-68. Sunday trip generation is positively and indirectly influenced by all Saturday activity durations.. Aldershot. Kim. Jones. E. P. Classic causal models only consider direct effects. pp. 15. Scientific Software International.52. ASCE. Golob. pp. Inc. (2003). T. Frusti. Lula. the sign differences are found between direct effects and total effects. ed. G. and Ren. Vol. “A theoretical and empirical model of trip chaining behavior. The subdivided indirect effects are also manually calculated. 19. No. 141-162. Vol. 4. of Transport Geography. pp. T. In particular. Because travel behavior includes complicated relationships among factors. pp. No.. In the analysis of effects between travel behaviors. M. For example.Direct and Indirect Effects on Weekend Travel Behaviors Table 10. the structural models determine indirect effects through intervening attributes. P. 1-17. 129. pp. it will be necessary to analyze the effects on travel behavior by using periodical multi-day data for an improved understanding of travel behavior. and Isabelle. basic applications and programming for confirmatory factor analytic models.147 0. Jones. Jones. From this study. Specifically. (2002).” J. (2004). Pas. In the analysis of effects of personal and household attributes. 37.” Transportation Research B.057 0. among Sunday activity durations.051 0. 2. and Sorbom. Gower. Golob. but it has a larger negative indirect effect on them through activity durations.. 13. “The causal influences of effects of income and car ownership on trip generation. and Benson. Subdivided Indirect effects of Saturday Travel Behavior on Sunday Travel Behavior Variable To Sunday Total Trips Activities From Saturday Shopping duration 0. T. pp. T. LISREL 8. References Adler. Vol.052 Sport duration 0. the major indirect effect factor on Sunday trips is Saturday activities. in addition to direct effects. T. (1989). No. No. In the future. K. direct and indirect effects of personal and household characteristics on travel behavior using structural models. K. a dependency is found in the relationships between weekend travel behaviors.000 0. of Transport Economics and Policy. No. “How households use different types of vehicles: a structural driver allocation and usage model. C. W. (1990). No. Jang. “Structural equation modeling of travel choice dynamics. (1987). Goulias. and travel patterns. The effects of Saturday travel behavior on Sunday travel behavior were also analyzed. Sunday activities have the highest indirect effects from leisure.” Developments in Dynamic and Activity-based Approaches to Travel Analysis. “Toward conceptualizing tripchaining behavior: A review. 1-25. Lawton. S.
” Transportation Research Record 1706. and Koppelman. pp. and Bhat. No. 1. (2000). M.” Transportation. Arentze. Pendyala. Oxford. Cambridge. and McNally. and Timmermans. J. “Time use and activity perspectives in travel behavior research. M. F. Vrtic. (1999). Activity Participation and Travel Behavior. E. M. M. 41. Steed. Pas. 347-359. “On modeling the departure time choice for home-based social/recreational and shopping trips. pp. P. “Trip-chaining in a linear city.. “On the structure of weekly activity/travel patterns. W. and Goulias. No. pp. “A conceptual and methodological framework for the generation of activity-travel patterns. “Capturing tour mode and activity choice interdependencies: A co-evolutionary logit modelling approach. Basic Principles of Structural Equation Modeling. accessibility and personal characteristics: The case of the Upper Austria Region. 823-839. (2004). Vol. 4. (1996). “An exploration of the relationship between timing and duration of maintenance activities.. 1. “Micro-simulation of activity-travel patterns for travel demand forecasting. A. S. C. 33. Mueller. J. R.” Transportation Research Part A. 27. No. 25-51.” Transportation Research A. pp. (1997). 27. 31.” Transportation Research A. Vol. 29. Pendyala. Simma. (2002).. S. H. R. “Changing the duration of activities in resolving scheduling conflicts. E. (2003). England. E. 155-167. C. Berlin. Lu. R. Heidelberg. Vol. Krygsman. (2001). 9. T.. 2. Springer-Verlag. pp. 37. (2000). No. R. and Timmermans. S. SAGE Publications. pp. Kitamura. pp. Stopher. − 178 − KSCE Journal of Civil Engineering . T. Vol. A. New York. No. and Narayanan. (2008). M. (1985). C. Vol. M.” Understanding Travel Behavior in Era of Change. 429-456. Vol. and Bhat. Using LISREL for structural equation modeling.” Transportation Research A. R. 1. (1998). P.. 10. K. pp. Wen. Kitamura.” Transportation. 42.. Lee. R. J.Tae Youn Jang and Jee Wook Hwang Kelloway. R. 19. and Harvey. 315-338. No. “Time use research and travel demand analysis and modeling. 5-23. R. No. pp. Ruiz. X. and Axhausen. K. 913-933. G.” The European Transport Conference. No. Pendyala. C.” Transportation. Elsevier.” Transportation Research B. Vol. Chen. 2. “Interactions between travel behavior. 152-159. Vol. H. TRB. ed. 1. pp. Vol.” Transportation Research Part A. (2000).. (2007). 1-4. K. No. O. R. R. P. pp. and Pas. 1-18. “Socio-demographics.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.