Modeling Gender-Based Differences in Mode Choice considering Time-Use Pattern: Analysis of Bicycle,Public Transit,and Car Use in Suzhou,China

Abstract

Activity-travel behavior differs between women and men. Previous researches dealing with gender differences mainly focus on travel in itself rather than the interaction between time-use pattern and travel mode choice. Based on the activity-travel survey data of Suzhou, China, multigroup structural equation modeling is adopted to explore the reason for gender-based differences in mode choice through comparing the interactions among sociodemographics, time-use pattern, and mode choice. The results indicate that gender-based differences do exist in mode choice. Women prefer traveling by bicycle while men prefer traveling by car in Suzhou. And compared to women, men's mode choice is not so easy to be affected by other travel modes. Besides, gender-based differences exist in the magnitude or the sign of the interrelations among socio-demographics, time-use pattern, and mode choice. It is better to explain gender-based differences in mode choice by including time-use pattern endogenously than through socio-demographics alone. Furthermore, the study shows that by examining the direct, indirect, and total effects in the model system simultaneously, we are able to better capture the differences in mode choice across genders and further able to understand the reason for those differences. Finally, some dedicated suggestions are presented for planners and government to ensure a healthy transportation system.

1. Introduction

“Reform and Opening-Up” policy has brought in rapid economic boom as well as continuous improvement of female social status and educational level in China [1]. Consequently, gender differences become more complex and need special attention. Women's contribution to travel demand is anticipated to grow because of the increase in their labor participation rate caused by their increased social status. Besides, women's specific physical and psychological characteristics make women's behavior differ from men's to some extent. Thus, considering humanized and equitable traffic planning and management, it is worth understanding the difference of travel behaviors between women and men. Since the 1960s in western developed countries, researchers have paid attention to women's travel behavior. Studies have identified gender as a significant factor in daily activity-travel behavior and have found gender differences existing in travel pattern, model choice, and activity participation (allocation of activity duration).

Researchers have paid much attention to gender-based travel patterns and found that women are more likely to adopt complex commute and nonwork chains than men [2, 3]. Particularly, they examined the gender differences in commute trip. It is found that women work closer to home, have shorter commute distance, and make fewer trips than men do [4 –6]. Besides, Li found that women make less stops and stop for shorter durations than men do in morning commutes [7]. Some literatures indicated that women are more apt to make trip chain on the way to and from work and also made more serving-passenger trip chains than men [8]. In terms of travel mode, previous researchers have uncovered a number of differences between men and women. For example, distinct gender differences exist in the purpose of bicycle trips, desired amenities, and safety perceptions [9]. Women are more willing than men to reduce car use [2]. Schwanen et al. found that women are inclined to commute more often by public transport, by bike, or on foot, whereas car use tends to be higher among men for work trips [10, 11]. Furthermore, previous researches have demonstrated that factors affecting gender differences in mode choice include mode attributes, household/individual characteristics, psychological factors, and habits [2, 12]. Most previous studies dealing with the genders merely focused on travel itself, while findings on gender-based activity participation are relatively few. Srinivasan and Bhat found that women are more likely to carry a larger share of household maintenance activities [13]. Cao and Chai investigated time allocation of male and female household heads in Shenzhen, China. The results showed clearly that men are dominant in out-of-home activities, but women dominate in-home activities [14].

On the other hand, in addition to the differences of travel behavior and activity participation across genders, previous researchers have found that complex relationships existed among sociodemographics, activity participation, and travel behavior. Lu and Pas explained relationships among those variables and showed that we can explain travel behavior better by including activity participation than through sociodemographics alone [15]. Kuppam and Pendyala also captured relationships between travel and activity participation [16]. Meanwhile, many researchers studied the relationship of activity and travel from a time-use perspective. Golob developed a trip generation model and showed the relationships among activity needs, travel demand, and travel time [17]. Chung and Ahn captured relationships among sociodemographics, time use, and travel behavior for each day during a week in Korea [18].

Here comes up a question that since gender-based differences exist in activity-travel behaviors and meanwhile complex relationships exist among sociodemographics, travel behavior, and activity participation, then are there any differences existing among those relationships across genders? Previous research did not answer this question. On the other hand, factors found to affect travel mode choice mainly include household/individual characteristics, traffic environment and land use. However, another important factor travelers may also take into account is the restriction of time. Travel is a derivative demand of activity participation [19]. Activity and the corresponding travel must take a time period to be completed respectively, but time is finite. As a result, travel time may be restricted by the duration of activity. In order to get to the destination within limited time, travelers should balance travel time and travel distance and then choose the most appropriate mode before making a trip. Mode choice is probably related to time-use pattern. Hence, it is of great value to examine the relationships between time-use pattern (allocation of activity duration and travel time) and mode choice, as well as the gender-based differences among those relationships.

The primary objective of this study is to explore the gender-based differences in the two following aspects: first, analyzing gender differences existing in time-use pattern and travel mode choice and, second, modeling the interrelations among sociodemographics, time-use pattern, and travel model choice of women and men, respectively, then making a gender-based comparison and deep analysis on the reason why gender-based differences exist in mode choice. The study is based on the activity-travel survey data of Suzhou, China.

The rest of this paper is organized as follows. The following section provides overview conceptual framework and methodology. In Section 3, an overall description of the data is given. Section 4 presents the results of the model estimation and discusses the key findings. Finally, conclusions and future directions are discussed.

2. Methodology

2.1. Conceptual Framework

Figure 1 gives an outline of the model developed and estimated in this research. We propose that sociodemographics affect time-use pattern (allocation of activity duration and travel time) and mode choice. Activity duration affect travel time because of time limitation. Beside, activity duration and travel time affects mode choice. These connections are partly adapted from some previous studies [20 –22].

Figure 1:

Conceptual framework of this study.

In summary, the hypotheses are as follows.

Hypothesis 1: sociodemographics have influence on individual's time-use pattern and mode choice.

Hypothesis 2: time-use pattern affects travel mode choice.

Hypothesis 3: sociodemographics not only affect travel mode choice directly but also indirectly affect it via time-use pattern.

Hypothesis 4: activity duration not only affects travel mode choice but also indirectly affects it directly via travel time.

Hypothesis 5: relations exist among various time allocations.

Hypothesis 6: relations exist among various travel mode choices.

2.2. Structural Equation Modeling

This paper aims to figure out the reason for gender-based differences in mode choice through analyzing the relationships among sociodemographics, time-use pattern, and mode choice. Due to the complex relationships among these variables, it is impossible to reveal the structural relationships among these variables through the traditional multiple regression or factor analysis. The advantage of structural equation model (SEM) is its capability in simultaneously estimating the causal relationships among a set of observed variables based on a specified model. In addition, SEM can reveal the direct, indirect, and total effects between two variables. A direct effect is the influence of one variable on another that is not mediated by any other variables, while an indirect effect is one that is mediated by at least another variable. The total effect of one variable on another is the sum of the direct and indirect effects [17, 23]. This paper conceives of better explaining relationships among various variables by capturing direct, indirect, and total effects simultaneously. So SEM seems to be an applicable method.

The standard structural equation model (without latent variables) is given by [23]:

y = B y + Γ x + ζ,

(1)

where y is a (p by 1) column vector of endogenous variables, and time-use pattern and mode choice are endogenous variables to the model in this study; x is a (q by 1) column vector of exogenous variables, and household and individual sociodemographics are exogenous variables in this model.

The structural parameters are the elements of the matrices: B is the (p × p) matrix of direct effects between the p endogenous variables; Γ is the (p × q) matrix of the direct regression effects from the q exogenous variables to the p endogenous variables. And ζ is a column vector of the error terms.

For identification of system (1), B must be chosen such that (I – B) is nonsingular, where I denotes the identity matrix of dimension p.

It can easily be shown that the total effects of the endogenous variables on one other are given by

T_{y y} = {(I - B)}^{- 1} - I .

(2)

The total effects of the exogenous variables on the endogenous variables in this structural equations model are given by

T_{x y} = {(I - B)}^{- 1} Γ .

(3)

If the endogenous variables in the structural equation system can all be considered to be continuous and multivariate normally distributed, the estimation of the system can use normal-theory maximum likelihood [23]. Previous research has demonstrated that, while ML potentially yields biased error estimates, the coefficient estimates will be consistent even with censored endogenous variables, and estimates have been shown to be robust under violations of multivariate normality [17].

3. Data Source and Descriptive Statistics

3.1. Data Source

Data was collected from the activity-travel survey of Suzhou, China, in 2009. As one of the key cities in the Yangtze River Delta, Suzhou is a renowned cultural, historic, and tourist center. It is one of the very highly developed cities in China with an urban area of 1650 km² and a population of 2.4 million. In 2009, its GDP reached an outstanding numerical value of 774.02 billion Chinese Yuan with the per capita GDP of 180 thousand Chinese Yuan. With the gradual extension of city outline and the rapid development of economy, residents now have a much longer distance to travel and a stronger desire to go in motorized ways, which leads to a great contradiction between supply and demand of the transportation resources.

Taking a whole household as a unit, a random sampling and face-to-face interview were adopted for the survey on Wednesday, May 27, 2009. The investigators are required to select citizens randomly in different parts of the city in order to guarantee the quality of our sample. Our sample involved a one-day (workday) activity-travel diary, which was designed to record all activities involving travel details such as purpose, mode, travel time, and origin destination of each trip, for all individuals in the household. It also included sociodemographics of both individual and household. Each household and its members were coded. Finally, 2434 valid forms from 2434 households were received.

3.2. Data Processing

In order to enhance the comparability between men and women, we take female heads and male heads from households with the same nuclear family structure into account. Here, nuclear family is composed barely of parents and kids. Owing to China's population policy of “one family one child,” nuclear family in China mainly refers to household of three members with both parents and the child. Nuclear family occupies the majority of the Chinese family population structure. After the processing of selecting nuclear families, excluding children's activity-travel records, eliminating missing data, and performing logical checking, 1500 women and 1500 men valid cases from 1500 households are used for further analysis.

Then it comes to the collection of individual's activity duration, travel time, and mode choice. The travel survey of Suzhou divided the trip purpose into nine species: work, school, bureaucracy, shopping, social recreation, serving passengers, personal business, returning home, and returning to work. According to Pas and Bowman, out-of-home activities include the following three categories [24, 25]: (i) subsistence (work, school, bureaucracy, and returning to work); (ii) maintenance (shopping, serving passengers, and personal business); (iii) leisure (social recreation). In-home activity is also considered; however this activity is not divided into different types due to the limitation of data. Besides, travel time includes the total time (hour) spent on all trips a day. Mode choice is expressed by the allocation rate of each travel mode on the research day. For example, the allocation rates for travel modes are 0.0% of private car, 66.7% of public transit, and 33.3% of bicycle if someone does not use a private car but uses a bus twice and bicycle once for three trips in a day. With the above criteria, types of activity durations, travel time, and the allocation rate of each travel mode were extracted in the original database by C++ programming.

3.3. Descriptive Statistics

3.3.1. Household Characteristics

Descriptive statistics of household characteristics are shown in Table 1. It is noteworthy that household bicycle ownership (including electric bicycle) is far more than car ownership in Suzhou, which is a typical characteristic of Chinese vehicle ownership. Besides, due to Suzhou's traffic demand management policy of limiting motorcycle use, motorcycle ownership is relatively low (only 8.5 percent of households have motorcycle). Household car ownership is relatively high compared to other Chinese cities, which may be attributed to Suzhou's rapidly economic development and motorization.

Table 1:

Descriptive statistics of household characteristics and individual characteristics.

Variables	Characteristics	Men		Women
Variables	Characteristics	Frequency	%	Frequency	%

NW	Number of workers
	0	3	0.2	3	0.2
	1	120	8.0	120	8.0
	2	1377	91.8	1377	91.8

CA	Children's age (years)
	<6	77	5.1	77	5.1
	6–14	952	63.5	952	63.5
	14–19	471	31.4	471	31.4

AHI	Annual household income (10,000 Yuan/year)
	<4	429	27.2	429	27.2
	4–8	670	42.5	670	42.5
	>8	479	30.4	479	30.4

BCO	Bicycle ownership (quantity of vehicles)
	0	78	5.2	78	5.2
	1	279	18.6	279	18.6
	2	637	42.4	637	42.4
	>2	506	33.7	506	33.7

MCO	Motorcycle ownership (quantity of vehicles)
	0	1373	91.5	1373	91.5
	1	123	8.2	123	8.2
	>1	4	0.3	4	0.3

CAO	Car ownership (quantity of vehicles)
	0	734	48.9	734	48.9
	1	633	42.2	633	42.2
	>1	133	8.9	133	8.9

JOB	Employment status
	Nonworker	12	0.8	136	9.1
	Worker	1488	99.2	1364	90.9

AGE	Age (years)
	≤40	714	47.6	1002	66.8
	>40	786	52.4	498	33.2

EDU	Educational level
	Low	269	17.9	369	24.6
	Middle	529	35.3	549	36.6
	High	702	46.8	582	38.8

3.3.2. Gender Differences in Individual Characteristics

Table 1 also shows the descriptive statistics of individual characteristics. It reveals that the average age of the women is less than that of the men, which has a strong relation with the Chinese traditional marriage view that it is better for the husband to be older than his wife. Although men's employment rate is higher than women's (99.2% versus 90.9%), women's employment rate is high in itself, indicating the enhancement of Chinese women's social status. In addition, generally, women's educational level is lower than men's.

3.3.3. Gender Differences in Time-Use Pattern

From Table 2, both men's and women's average subsistence activity durations are approximately 8 hours a day, which is consistent with the reality. The descriptive statistics indicate that women tend to take part in more in-home activities than men, while men participate in more out-of-home activities, which is similar to Cao and Chai's findings [14]. Besides, men's out-of-home subsistence activity duration and leisure activity duration are both longer than women's while women's maintenance activities duration is longer than men's. The reason for this may lie in the fact that women take a bigger share of household and family responsibilities, while men mainly play the role of breadwinner and are more social. Further, the average travel time of men is longer than that of women. The possible explanation may be that men take part in more social activities and thus generate more travel.

Table 2:

Descriptive statistics of time-use pattern and mode choice.

Variables	Characteristics	Men		Women
Variables	Characteristics	Mean	SD	Mean	SD

DSU	Duration of subsistence activity (hour)	8.00	3.11	7.35	3.54
DMA	Duration of maintenance activity (hour)	0.35	1.52	0.46	1.13
DLE	Duration of leisure activity (hour)	0.25	1.19	0.16	0.84
DIN	Duration of in-home activity (hour)	14.28	2.28	15.05	3.02
DTR	Travel time (hour)	1.12	0.84	0.98	0.68
RWA	Allocation rate of walk (%)	20.6	21.4	21.8	26.7
RCY	Allocation rate of bicycle (%)	35.7	45.4	44.3	48.2
RPU	Allocation rate of public transit (%)	9.8	28	10.6	27.3
RMO	Allocation rate of motorcycle (%)	4.5	23.4	3.5	21.4
RCA	Allocation rate of car (%)	23.6	48.2	16.6	39.4
ROT	Allocation rate of other modes (%)	5.8	22.9	3.2	12.5

3.3.4. Gender Differences in Mode Choice

For both genders, the allocation rate of bicycle (including electric bicycle) accounts for the largest proportion among various modes, which is reasonable in China. Additionally, women's allocation rate of bicycle is much larger than men's (44.8% versus 35.7%), while men's allocation rate of car is much larger than women's (25.6% versus 17.6%). Table 2 indicates that women are more inclined to travel by bike than men, while men are more likely to travel by car in Suzhou. Besides, the low motorcycle rate is consistent with its low ownership. Furthermore, it is noteworthy that the allocation rate of public transit is relatively low (about 10 percent) among all travel modes for both men and women. In the context that Chinese government encourages citizens to travel by public transit to relieve traffic congestions and to protect the environment, the low public transit rate necessitates us to pay close attention to it and take measures to improve public transit rate in midi-distance trip.

Generally, the statistical characteristics are consistent with the reality, indicating samples with rationality and representation.

3.4. Exogenous and Endogenous Variables

Based on the above descriptive statistical analysis and extensive prior research, exogenous and endogenous variables are selected to conduct structural equation modeling analysis. Sociodemographic characteristics are the exogenous variables, containing two groups: household characteristics and individual characteristics. Household characteristics include the number of household workers, children's age, household annual income, bicycle ownership, and car ownership. Individual characteristics include age, employment status, and educational level of both men and women. As discussed above, the sample consists of both household heads from nuclear family. Therefore, family structure is not included as exogenous variables.

Endogenous variables contain time-use pattern and travel mode choice of both men and women. Time-use pattern is captured by various activity durations (subsistence, maintenance, leisure, and in-home activities) and travel time. Due to the limitation of data in Suzhou, three types of travel modes, bicycle, public transit, and car, are adopted to conduct the following analysis on gender difference.

4. Results and Analysis

4.1. Model Development and Final Model Estimated Results

4.1.1. Model Development

SEM normally involves six steps: model specification, implied covariance matrix determination, model identification, model estimation, model evaluation, and model respecification [23]. Due to large amount of variables and their complex relationships, significance tests were conducted to help us guide the model specification portion of this study. The sociodemographics can be considered as discrete variables while time-use pattern and mode choice are continuous variables. Thus, analysis of variance was adopted to test the significance between discrete variables and continuous variables. Meanwhile, correlation analysis was applied to test the significance of those continuous ones.

Based on previous researches, hypothetical relations, and significance test results, the causal structures of two baseline models, one for the women group and the other for the men group, were specified, respectively. Due to endogenous variables that are continuous and multivariate normally distributed, models were estimated with maximum likelihood (ML) method using LISREL/PRELIS 8.7. The baseline models were then validated using a cross-validation strategy to address the sample sensitivity issue in post hoc model fitting. Since we do not have a second sample for the study, we cross-validate the baseline model for the women group with two subsamples randomly selected from the original sample. About half of the cases in the original sample were randomly selected as the calibration sample, while the other half of cases were the validation sample used to test invariance across the calibration and validation samples. And it is same as men group.

4.1.2. Goodness of Fit and Estimated Results

The chi-square and probability values are calculated based on normality. If the chi-square value is big, the model has a tendency not to fit the data. If the sample size is over 200 and the value of the goodness of fit index (GFI) is at least 0.9, it is thought that the model has no problem. The required accepted adjusting goodness of fit index (AGFI) is 0.9 or more. The normalized fit index (NFI) frequently used to test fitness should be between 0.0 and 1.0, and the required accepted value is more than 0.9. There is no standard value for the root mean square residual (RMR). However, if fitness is good, it has a value close to 0, while if the fitness is bad, the value becomes bigger [26].

The final models are shown in Tables 3 and 4. The goodness of fit statistics of the two baseline models indicate that both models fit the data well (women: chi-square = 40.50, degree of freedom (df) = 58, GFI = 1.00, AGFI = 0.99, NFI = 1.00, and RMR = 0.011; men: chi-square = 37.89, df = 57, GFI = 1.00, AGFI = 0.99, NFI = 1.00, and RMR = 0.011).

Table 3:

Total, direct, and indirect effects of sociodemographics on time-use pattern and mode choice (men and women).

	NW		CA		AHI		BCO		CAO		JOB		AGE		EDU
	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women

DSU
DE	0.64**			0.57*							4.34***	10.56 ***
IE			0.19*	−0.23*					−0.11*	0.16**	0.66**	−4.33***	0.09*
TE	0.62**		0.19*	0.34*					−0.11*	0.16**	5.01***	6.23***	0.09*
DMA
DE			−0.16**							0.09**
IE				−0.08*					0.07**	−0.04**	−0.41*	−1.42***	−0.06**
TE			−0.12**	−0.08*					0.07**	0.05***	−0.41*	−1.42***	−0.06**
DLE
DE	−0.21*	−0.24**					0.06*		0.14***		−0.98**				−0.07*
IE	−0.09**		−0.03*	−0.02*					0.02*	−0.01**	−0.76***	−0.36***
TE	−0.3**	−0.24**	−0.03*	−0.02*			0.06*		0.16***	−0.01**	−1.74***	−0.36***			−0.08*
DIN
DE								0.08**			−0.63**	−0.82***
IE	−0.39*	0.17**		−0.24*	−0.01**	0.02***	−0.03*	−0.03***		−0.18***	−3.23***	−4.48***		−0.01*		−0.02**
TE	−0.39*	0.17**		−0.24*	−0.01*	0.02***	−0.03*			−0.18***	−3.86***	−5.3***		−0.01*		−0.02**
DTR
DE									−0.18***	−0.1***	1.24 ***	0.61 ***	0.14***		0.05*	0.08***
IE	−0.05**	−0.03**	−0.08**	−0.03*		0*	0.01*		0.06**	0.01*	−0.62***	−0.58***	−0.04*		−0.02*	0*
TE	−0.05**	−0.03**	−0.08**	−0.03*		0*	0.01*		−0.12***	−0.09***	0.62**		0.11***			0.08**
RCY
DE					−0.09***	−0.05***	0.06***	0.08***	−0.35***	−0.03**				−0.04**		−0.06***
IE			0.01*			0.03***		0.1***	0.02***		−0.15***	−0.11***	−0.02***	−0.01**	−0.01*	−0.05***
TE			0.01*		−0.09***	−0.02***	0.06***	0.18***	−0.33***	−0.04*	−0.15***	−0.11***	−0.02***	−0.06**	−0.01*	−0.11***
RPU
DE									−0.01*			0.03***
IE	0*					−0.04***	−0.03***	−0.04***			0.03*		0.01***	0.03**	−0.03**	0.03***
TE	0*			0*		−0.04***	−0.03***	−0.04***	−0.03*		0.03*		0.01***	0.03**	−0.03**	0.03***
RCA
DE					0.01*	0.02*	0.01*	0.04***	0.04**	0.03**					0.02*	0.03*
IE			−0.01*		0.08***	0.04***	−0.04***	−0.08***	0.31 ***	0.02 *	0.1 ***	0.1 ***	0.01***	0.03**	0.02***	0.05***
TE			−0.01*		0.09***	0.06***	−0.03**	−0.04***	0.35***	0.05***	0.1***	0.1***	0.01***	0.03**	0.04***	0.08***

DE: direct effect; IE: indirect effect; TE: total effect. *Significant at the 0.05 level. **Significant at the 0.01 level. ***Significant at the 0.001 level.

Table 4:

Total, direct, and indirect effects among time-use pattern and mode choice (men and women).

	DSU		DMA		DLE		DIN		DTR		RCY		RPU		RCA
	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women	Men	Women

DSU
DE			−1.22***
IE		−0.41***		−0.25***	0.07*
TE		−0.41***	−1.22***	−0.25***	0.07*				0.63**
DMA
DE	−0.02*	−0.23***							−0.51**
IE	0.02**	0.09***	−0.24*	−0.41***	−0.04**
TE		−0.13***	−0.24*	−0.41***	−0.04**				−0.39***
DLE
DE	−0.15***	−0.06***		0.12*
IE		0.03***	0.19***	−0.1***	−0.01*				−0.1*
TE	−0.15***	−0.03***	0.19***	0.02***	−0.01*				−0.1*
DIN
DE	−0.89***	−0.92***	−0.6***	−0.73***	−0.56***	−0.71***						0.16***			−0.1**
IE	0.09**	0.5***	1.12***	−1.28***						−0.03***	0.08**	0.04***				−0.09***
TE	−0.8***	−0.42***	0.52*	−2.01***	−0.59***	−0.71***		0*		−0.03***	0.08**	0.21***			−0.22**	−0.09***
DTR
DE	−0.02*	0.05*	0.63 ***	0.29 ***	0.13***	0.17***	0.04*	0.08**
IE	−0.04**	−0.1***	−0.11**	−0.21***	−0.05**	−0.05**			−0.27**	0*	0*	0.02*				−0.01*
TE	−0.06**	−0.05***	0.52***	0.09***	0.08***	0.11***	0.03**	0.08**	−0.27**	0*	0*	0.02*				−0.01*
RCY
DE			0.04**	0.03 ***	0.02 **	0.02**			−0.14***	−0.16***						−0.45***
IE	0*		−0.08***	−0.04***	−0.01***	−0.02***	0*	−0.02**		−0.04***		0.26***				−0.12***
TE	0*		−0.04***		0.01*		0*	−0.02**	−0.12***	−0.21***		0.26***				−0.57***
RPU
DE			0.01***	0.02***	0***	0.01***			0.01***	0.01***	−0.95***	−0.97***			−0.94***	−0.96***
IE	0***	−0.01***		−0.02***			0*	0.01**	0.04***	0.11***	0.7 ***	0.31 ***	1.13*		−1.06*	0.3***
TE	0***	−0.01***					0*	0.01**	0.04***	0.12***	−0.25***	−0.66***	1.13*		−1.06*	−0.66***
RCA
DE		0.01 ***	0.02***						0.08 ***		−0.89***	−0.46***		−0.57***
IE		0***	0.03**	0.02***	−0.01*		0*	0.01**	0.07***	0.1***	0.14***	−0.12***		−0.64*	1.13*	0.26***
TE		0**	0.04***	0.02***	−0.01*		0*	0.01**	0.15***	0.1***	−0.74***	−0.58***		−1.21*	1.13*	0.26***

DE: direct effect; IE: indirect effect; TE: total effect. *Significant at the 0.05 level. **Significant at the 0.01 level. ***Significant at the 0.001 level.

In addition to these two final baseline models, a multigroup analysis was conducted to test invariance of causal structures across genders. Theoretically, group comparison is performed hierarchically with increasing constraints imposed on the model, varying from constraints only on the same model structure across groups to constraints on the same model structure, parameters, residuals, and variance-covariance [23]. Basically, three different assumptions of group equity are tested in the case that the baseline model does not have a measurement component. These three assumptions are equal structural weights, equal structural covariance, and equal structural residuals across groups. Multigroup analysis for gender comparison was performed against the two baseline models. No matter which baseline model was used, the chi-square value is significant when structural weights are set to be equal across genders (P = 0.005 for women's baseline model and P = 0.010 for men's baseline model). This suggests that the null hypothesis that the women group and the men group can be modeled by the same causal structure should be rejected. Significant variant model structures are found across genders, which indicates that the interactions among sociodemographic, time-use pattern, and mode choice are significantly different between women and men.

4.2. Gender-Based Results Analysis

4.2.1. Gender-Based Differences in the Relationships among Sociodemographics, Time-Use Pattern, and Bicycle Usage

Table 3 indicates that household characteristics that directly influence both genders' bicycle usage significantly include annual household income, household bicycle ownership, and car ownership, but there are some gender differences in the magnitude of these influences. When bicycle ownership increases, the probability of bicycle usage increases, and this influence of bicycle ownership on bicycle usage is larger for women than for men. On the other hand, when annual household income and car ownership increase, the probability of bicycle usage decreases, and the influence is smaller for women than for men. These direct effects suggest that women are more likely to travel by bicycle than men. Besides, as women get older, they tend to reduce bicycle usage. One possible explanation could be that old women have weaker physical strength; thus they have to use nonphysical travel mode instead of such manual mode as bicycle. We also find that women with higher educational level tend to decrease their bicycle usage. However, men's allocation rate of bicycle is generally not so closely directly related to their age and educational level.

The estimation results also indicate that the total effects of sociodemographics on allocation rate of bicycle can be quite different from the direct effects, due to the indirect effects. For instance, although women's employment status does not directly influence their bicycle usage, it has a remarkable negative indirect effect on bicycle. As a result, the total effect is significant. Through tracing the path diagram of the model, we find that the negative indirect impact of employment status on allocation rate of bicycle comes from several sources: travel time, subsistence activity duration, maintenance activity duration, leisure activity duration and in-home duration. Travel time especially plays the most important role. As shown in Table 3, employed women tend to have more travel time, and Table 4 indicates that when travel time increases, the allocation rate of bicycle decreases; thus employed women have a smaller bicycle rate. The negative effect of travel time on bicycle usage is consistent with the existing findings that bicycling is mainly considered when travel distance is less than 6 kilometers or travel duration is expected to be within 30 minutes. When compared to women, employed men decrease their allocation rate of bicycle to a greater extent, mainly because men's employment affects travel time to a greater extent than that of women.

Besides, Table 4 illustrates that women's allocation rate of bicycle increases as the amount of time spent on their maintenance and leisure activities increases. Similar tendency happens to men. On the other hand, for both genders, maintenance activity duration has negative indirect influence on bicycle usage, but there exist some gender-based differences in the magnitude of the influences. Our further analysis shows that these indirect effects mainly come from travel time. More maintenance activity duration induces more travel time directly, and more travel time decreases bicycle rate. The reason why the indirect influence of maintenance on bicycle usage is larger for men than for women is that the direct influence of maintenance on travel time is larger for men than for women. Additionally, in-home activity duration does not directly influence bicycle usage for both men and women as we would expect. However, there exists a significant indirect influence from women's in-home time duration to their allocation rate of bicycle. Consequently, the total effect reveals that women who spent more time in home use less bicycle.

Moreover, women who travel by more cars tend to decrease their bicycle usage remarkably, while men do not have this property. This may be due to the fact that men use bicycle quite less than women, so men's allocation rate of bicycle is not so easy to be affected by other travel modes as women's.

4.2.2. Gender-Based Differences in the Relationships among Sociodemographics, Time-Use Pattern, and Public Transit Usage

Table 3 shows that household car ownership has significant negative influence on men's public transit usage, indicating that men coming from household with cars are likely to reduce public transit usage, while women do not have this property. Sociaodemographics variable that directly affects women's public transit is women's employment status. Employed women tend to travel by more public transit, and this may further suggest that women probably choose public transit mode in their commuting trips. But men's public transit usage is not closely related to employment.

Although other sociaodemographics except variables for household car ownership and women's employment status are not directly associated with public transit, they have remarkable indirect effects on it. These indirect effects also show some gender differences. For example, household annual income has indirect effect on women's public transit rate, indicating that women living in an affluent household tend to use less public transit potentially. But men's public transit usage is not related to household annual income. In terms of the indirect educational level on public transit, it is positive for women but negative for men. Namely, higher educational level induces women to use public transit but reduces men's public transit usage potentially. Moreover, people tend to use more public transit as they get older for both genders, but the magnitude of the positive effects is larger for women than for men. These may further suggest that women are more likely to use public transit than to men.

In the context of employed women tending to use more public transit, their public transit usage is not directly related to subsistence activity duration, which runs counter to our expectation. However, women's subsistence activity duration has negative indirect influence on their public transit usage. The likely explanation for this may be that more subsistence activity duration will directly lead to more car usage as shown in Table 4, and more car usage brings about less public transit usage. This phenomenon reveals that as, more time is spent on subsistence activity, car is more competitive than public transit for women.

Interestingly, Table 4 indicates that the positive direct effect of women's maintenance activity duration on public transit usage is offset by the indirect effect, so the total effect is insignificant. This indirect effect is mainly mediated by bicycle usage, and it mainly comes from substitution effects among travel modes. Women spending more time on maintenance tend to use more bicycle, and more bicycle usage results in less public transit usage. Additionally, contrary to the negative effects travel time has on bicycle usage, travel time positively affects public transit usage for both genders, and it affects women's public transit usage to a greater extent.

Further, Table 4 indicates that, for both genders, the negative direct effect of bicycle usage on public transit usage is offset by a positive indirect effect, so that the total effect is negative but smaller than the direct one. Further analysis shows that the indirect effect is mainly caused by car usage. More allocation rate of bicycle will decrease the allocation rate of car, and less car usage will induce more public transit usage. Although the negative direct effects for both genders are similar in the magnitude, the positive indirect effect is larger for men than for women, so the total effects indicate that bicycle usage will decrease public transit usage to a greater extent for women.

4.2.3. Gender-Based Differences in the Relationships among Sociodemographics, Time-Use Pattern, and Car Usage

The results show that household income and car ownership positively affect both men's and women's car usage as we would expect, and the influence on men is much larger than that on women. Though car ownership especially has similar direct effect on both genders' car usage, the magnitude of the total effect for men is 7 times that for the women, caused by the remarkable larger indirect effect for men. Through tracing the source of this indirect effect for men, it turns out that the indirect effect coming from bicycle usage plays the most important role. When car ownership increases, men reduce more bicycle usage than women, which in turn induces more car usage. These findings suggest that men are more likely to travel by car than women.

Interestingly, we find that, for both genders, people with the availability of cars tend to reduce bicycle usage, while people with the availability of bicycles prefer traveling by car as well. Besides, educational level has significant positive influence on the allocation rate of car, and it affects women's car usage to a greater extent. A possible explanation could be that people with higher educational background may be more likely to have high-wage jobs and can afford private vehicles relatively. In terms of age, it has positive influence on car usage, which is similar to its influence on public transit.

Table 3 indicates that the indirect effects of employment status on car usage for men and women have the same coefficient of 0.1. Our further analysis shows that the same patterns across genders seem to have been generated by different processes. The indirect effect of employment status on car usage for men is largely caused by travel time. Employed men spend more time on travel than men who are not employed as shown in Table 3, and more travel time results in more car usage as shown in Table 4, while for women the indirect effect mainly comes from subsistence activity duration. Employed women spend more time on subsistence activity than those who are not employed, and women with more subsistence activity duration are more likely to travel by car.

With respect to the influence of activity participation duration on car usage, it is not so significant than that on bicycle and public transit usage. Even so, some remarkable gender differences are also found. For instance, women who spend more time on subsistence activity duration tend to use more car, while men's car usage is not affected by subsistence activity duration. Meanwhile, Table 4 indicates that maintenance activity duration has positive influence on car usage, and the influence on men's car usage is larger than that on women's. Additionally, men with more leisure activity duration tend to reduce their car usage due to the indirect effect. Through tracing the path diagram of the model, we find that the negative indirect impact is mainly from the substitution effects among travel mode. Men who spend more time on leisure activity will increase their bicycle usage, while more bicycle usage leads to less car usage. It is probably that men prefer bicycle when they go to participate in leisure activity, while for women the influence of leisure activity duration on car usage is not significant.

Table 4 also indicates that men's travel time is directly related to their car usage significantly. More travel time leads to more allocation rate of car. The explanation is that car is mainly considered to have advantage over long distance and time-consuming trip. On the other hand, travel time does not directly affect women's car usage but has positive indirect effect on it. Moreover, women's car usage will decrease remarkably if the allocation rate of public transit increases. But for men, their car usage is not affected by public transit. This further indicates that men are not so easy to change their car usage.

5. Conclusions and Future Research

This study seeks to explore the reason for gender-based differences in mode choice through examining the complex relationships among sociodemographics, time-use pattern, and mode choice with a focus on gender differences. Based on the activity-travel data of Suzhou, data processing, descriptive analysis, and rigorous significance tests have been conducted. Then structure equation models developed for women and men are compared. The results prove that our hypotheses proposed are reasonable. Major findings are presented as follows.

(i) Gender-based differences do exist in mode choice. Women are more likely to travel by bicycle than men while men use more cars than women in Suzhou. The allocation rate of public transit is relatively low when compared to that of bicycle and that of car, but women are more likely to use public transit than men. Meanwhile, substitution effects exist among various travel modes. For a specific mode, men's mode choice is not so easy to be affected by other travel modes as women's.

(ii) Interrelations exist among sociodemographics, time-use pattern, and mode choice, and gender-based differences exist in the magnitude or the sign of these interrelations. For example, household income and car ownership increase men's car usage to a greater extent than women's. Travel time affects public transit usage for both genders, but the magnitude of the positive effects is larger for women than for men. Maintenance activity duration has positive influence on car usage, and the influence on men is larger than that on women. Besides, higher educated level induces women to use public transit but reduces men's public transit usage potentially. Additionally, age and educational level negatively affect women's bicycle usage, but they do not affect men's significantly.

(ii) We can explain gender-based differences in mode choice better by including time-use pattern endogenously in the model. Time-use pattern has influences on mode choice, and gender differences exist in these influences. For instance, women who spend more time on subsistence activity duration tend to use more car, while this does not happen to men. Men who spend more time on leisure activity will increase their bicycle usage and thus reduce their car usage, but women do not have this property. Activity durations especially will indirectly influence mode choice mediated by travel time. Further, time-use pattern is the main intermediary that causes the indirect effects that sociodemographics have on mode choice. For instance, the negative indirect effect that women's employment status has on bicycle usage comes from travel time and various activity durations, and travel time plays the most important role.

(iv) By examining the direct, indirect, and total effects in the model system simultaneously, we are able to better capture and understand the differences in mode choice across genders as well as the reason for these differences. For example, although the negative direct effects of bicycle usage on public transit usage for both genders are similar in the magnitude, the positive indirect effect offsetting the direct effect to some extent is larger for men than for women, so the total effects indicate that bicycle usage decreases public transit usage to a greater extent for women. Moreover, through effects analysis, we find that some same patterns across genders seem to have been generated by different processes. The indirect effect of employment status on car usage for men is largely caused by travel time, while the same magnitude of indirect effect for women mainly comes from subsistence activity duration.

These results will lead to a better understanding of the gender-based differences in mode choice and highlight that the effects of time-use pattern on mode choice are different across genders. According to the results, there are some suggestions for planners and government to ensure a healthy transportation system. Nowadays, Chinese government has given priority to the development of public transit to relieve traffic congestions and to protect environment. Due to the fact that women are more likely to change their mode choice than men and women are more likely to travel by public transit than men, it might be easier to persuade women to shift to buses, so policy makers should be more oriented towards women to some extent. Companies can operate more low-underpan vehicles to make women wearing high-heel shoes get on/off the bus easily. Besides, they should improve the vehicle's inner traveling environment, such as broadcasting TV program in line with women's taste, to attract more female passengers. On the other hand, it is necessary to explore how women's public transit usage can induce men to use more public transit as well. In terms of the phenomenon that employed people and people with high educational background prefer car rather than bicycle, it is necessary to take some measures to prevent bicycle usage turning into car usage. It may be better for urban planners to design high-density mixed land use to reduce commuter's travel distance. Additionally, educational institutions should instill the concept that bicycle and public transit are more environment friendly than car to their students.

However, this study still has several limitations. Firstly, due to the limitation of data, electric bicycle is merged into bicycle in this study. However, there exist some different characteristics between bicycle and electric bicycle. This paper does not study motorcycle usage either. Based on other data sources, future research should pay attention to these travel modes in addition to modes considered in this study. Secondly, for travel mode choice, this study just considers the allocation rate of each travel mode in a day but ignores the interactions among mode choice in each trip. Future research can focus on mode choice in the trip chaining to deeply explore the gender-based differences. Finally, in addition to the differences, the interactions between men and women in one family, such as the relationship of activity-travel pattern and mode choice, need to be further studied.

Footnotes

Acknowledgments

This research is supported by the National Key Basic Research Program of China (2012CB725400) and Natural Science Foundation of China (51378120, 51338003, and 51178109). Fundamental Research Funds for the Central Universities and Foundation for Young Key Teachers of Southeast University are also appreciated.

References

Higgins

L. T.

Zheng

Liu

, and Sun

C. H.

, “Attitudes to marriage and sexual behaviors: a survey of gender and culture differences in China and United Kingdom,” Sex Roles, vol. 46, no. 3–4, pp. 75–89, 2002.

Matthies

Kuhn

, and Klöckner

C. A.

, “Travel mode choice of women: the result of limitation, ecological norm, or weak habit?” Environment and Behavior, vol. 34, no. 2, pp. 163–177, 2002.

McGuckin

, “Differences in trip chaining by men and women: research on woman's issues in transportation,” in Proceedings of the Transportation Research Board of the National Academies, pp. 49–55, Chicago, Ill, USA, November 2004.

McLafferty

, “Gender, race, and the determinants of commuting: New York in 1990,” Urban Geography, vol. 18, no. 3, pp. 192–212, 1997.

Turner

and Niemeier

, “Travel to work and household responsibility: new evidence,” Transportation, vol. 24, no. 4, pp. 397–419, 1997.

Sarmiento

, “Household, gender and travel,” in Women's Travel Issues: Proceedings from the Second National Conference, October 1996, FHWA, U.S. Department of Transportation, 2003.

, “Comparing women's and men's morning commute trip chaining in Atlanta, Georgia: research on women's issues in transportation,” in Proceedings of the Transportation Research Board of the National Academies, pp. 14–20, Chicago, Ill, USA, November 2004.

Strathman

and Dueker

, “Understanding trip chaining,” in 1990 NPTS Special Reports on Trip and Vehicle Attributes, pp. 1–27, FHWA, U.S. Department of Transportation, 1995.

Krizek

K. J.

and Johnson

P. J.

, “Gender differences in bicycling behavior and facility preferences: research on women's issues in transportation,” in Proceedings of the Transportation Research Board of the National Academies, pp. 31–40, Chicago, Ill, USA, November 2004.

10.

Schwanen

Dijst

, and Dieleman

F. M.

, “A microlevel analysis of residential context and travel time,” Environment and Planning A, vol. 34, no. 8, pp. 1487–1507, 2002.

11.

Schwanen

Dieleman

F. M.

, and Dijst

, “The impact of metropolitan structure on commute behavior in the Netherlands: a multilevel approach,” Growth and Change, vol. 35, no. 2, pp. 304–333, 2004.

12.

Collins

C. M.

and Chambers

S. M.

, “Psychological and situational influences on commuter-transport-mode choice,” Environment and Behavior, vol. 37, no. 5, pp. 640–661, 2005.

13.

Srinivasan

and Bhat

C. R.

, “Modeling household interactions in daily in-home and out-of-home maintenance activity participation,” Transportation, vol. 32, no. 5, pp. 523–544, 2005.

14.

Cao

and Chai

, “Gender role-based differences in time allocation: case study of Shenzhen, China,” Transportation Research Record, vol. 2014, pp. 58–66, 2007.

15.

and Pas

E. I.

, “Socio-demographics, activity participation and travel behavior,” Transportation Research A, vol. 33, no. 1, pp. 1–18, 1999.

16.

Kuppam

A. R.

and Pendyala

R. M.

, “A structural equations analysis of commuters' activity and travel patterns,” Transportation, vol. 28, no. 1, pp. 33–54, 2001.

17.

Golob

T. F.

, “Structural equation modeling for travel behavior research,” Transportation Research B, vol. 37, no. 1, pp. 1–25, 2003.

18.

Chung

J.-H.

and Ahn

, “Structural equation models of day-to-day activity participation and travel behavior in a developing country,” Transportation Research Record, no. 1807, pp. 109–118, 2002.

19.

Timmermans

Arentze

, and Joh

C.-H.

, “Analysing spacetime behaviour: new approaches to old problems,” Progress in Human Geography, vol. 26, no. 2, pp. 175–190, 2002.

20.

Lee

Hickman

, and Washington

, “Household type and structure, time-use pattern, and trip-chaining behavior,” Transportation Research A, vol. 41, no. 10, pp. 1004–1020, 2007.

21.

Zhang

Timmermans

H. J. P.

, and Borgers

, “A model of household task allocation and time use,” Transportation Research B, vol. 39, no. 1, pp. 81–95, 2005.

22.

Yang

Wang

Ren

Fan

, and Chen

, “Structural equation model to analyze sociodemographics, activity participation, and trip chaining between household heads: survey of Shangyu, China,” Transportation Research Record, no. 2157, pp. 38–45, 2010.

23.

Bollen

K. A.

, Structural Equations with Latent Variables, John Wiley & Sons, New York, NY, USA, 1989.

24.

Pas

E. I.

, “Analytically derived classifications of daily travel-activity behavior: description, evaluation, and interpretation,” Transportation Research Board, vol. 879, pp. 9–15, 1982.

25.

Bowman

J. L.

, The Day Activity Schedule Approach to Travel Demand Analysis, Massachusetts Institute of Technology, 1998.

26.

Jang

T. Y.

, “Causal relationship among travel mode, activity, and travel patterns,” Journal of Transportation Engineering, vol. 129, no. 1, pp. 16–22, 2003.