You are on page 1of 16

Transportation Planning and Technology

ISSN: 0308-1060 (Print) 1029-0354 (Online) Journal homepage: http://www.tandfonline.com/loi/gtpt20

An investigation of tour generation models


combining two waves of household travel surveys
through pooled models

Kollol Shams, Hamidreza Asgari, Md Sakoat Hossan & Xia Jin

To cite this article: Kollol Shams, Hamidreza Asgari, Md Sakoat Hossan & Xia Jin (2018): An
investigation of tour generation models combining two waves of household travel surveys through
pooled models, Transportation Planning and Technology, DOI: 10.1080/03081060.2018.1435415

To link to this article: https://doi.org/10.1080/03081060.2018.1435415

Published online: 12 Feb 2018.

Submit your article to this journal

Article views: 21

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=gtpt20
TRANSPORTATION PLANNING AND TECHNOLOGY, 2018
https://doi.org/10.1080/03081060.2018.1435415

An investigation of tour generation models combining two


waves of household travel surveys through pooled models
Kollol Shams, Hamidreza Asgari, Md Sakoat Hossan and Xia Jin
Department of Civil and Environmental Engineering, Florida International University, Miami, FL, USA

ABSTRACT ARTICLE HISTORY


This paper presents a study that examines two waves of travel Received 25 April 2016
survey data through a pooled model structure. The pooled model Accepted 6 November 2017
structure provides a means to take advantage of multiple data
KEYWORDS
sources which will lead to a better estimate and understanding of Pooled model; temporal
travel behavior. In particular, it accounts for the difference in data transferability; tour
variance and therefore allows for the comparison of the true generation models; regional
impacts of the model parameters on travelers’ tour-making household travel survey;
behavior. Larger variance is found in the 1998 data than in the model equality test statistics
2010 data. Comparison between model parameters reveals
significant behavioral changes among several socio-economic and
demographic groups. In terms of common variables, the
magnitude of the coefficient values has generally decreased,
which conforms to the overall decreasing trend in traveling.
Overall, the model equality tests indicate that the models
developed based on the two data sources do not have equal taste
parameters, thus the transferability hypothesis is rejected. The
results of this study are expected to have implications for the
application of models based on cross-sectional data, especially
over long time periods.

1. Introduction
The application and development of predictive modeling strategies has turned into a fun-
damental step in transportation science. Covering a wide range of transportation engin-
eering fields from demand management strategies (Asgari and Jin 2017; Asgari, Jin, and
Mohseni 2014; Hossan, Asgari, and Jin 2016; Jin, Asgari, and Hossan 2014; Jin et al.
2017; Shabanpour, Golshani et al. 2017; Shabanpour, Javanmardi et al. 2017; Shams,
Asgari, and Jin 2016; Shams et al. 2017) to traffic operation and safety (Hosseinlou,
Balal et al. 2012; Hosseinlou, Massahi, and Aliabadi 2012; Massahi 2017; Massahi et al.
2016, 2017), mathematical models play an important role to provide reliable estimates
of future transportation system conditions.
In particular, travel demand forecasting and policy analysis rely on behavioral models
that capture the relationships between observed travel activities and attributes of potential
contributors (population, land use, transportation system, and economic development,
etc.). The underlying assumption is that those relationships remain stable while the

CONTACT Hamidreza Asgari hasga001@fiu.edu Department of Civil and Environmental Engineering, Florida
International University, 10555 W. Flagler Street, EC 3720, Miami, FL 33174, USA
© 2018 Informa UK Limited, trading as Taylor & Francis Group
2 K. SHAMS ET AL.

variations in the contributing factors result in changes in travel demand. As society


changes almost constantly, with the introduction of new information technologies, evol-
ving economy, lifestyle shifts, etc., travel behavior (referred to as the relationships) may
change as well. Given the limitations of using cross-sectional data, many studies (cf.
Goodwin, Kitamura, and Meurs 1990; Hensher 1985; Kitamura 1990; Raimond and
Hensher 1997) have stressed the importance of panel data to understand the dynamic
aspects of travel behavior.
Many studies have been conducted to investigate the impacts of major factors that
influence travel behavior. Polzin (2006) examined the probable implications of changes
in three major factors on travel behavior: socio-economic factors, land use, and the trans-
portation system. The impacts on travel growth were also investigated using the past
National Household Travel Survey (NHTS)/National Personal Travel Survey (NPTS)
data from 1977–2001. Gärling et al. (2002) proposed a conceptual framework that can
model the impact of travel demand management policies using the knowledge of behav-
ioral change. Noland and Lem (2002) reviewed past studies that investigated the impacts
on travel demand of changes in transportation infrastructures. Kitamura, Yamamoto, and
Fujii (2003) investigated the effects of varying panel survey length on estimating the tran-
sition of behavior change using a Markov process. Dargay and Hanly (2007) examined the
effect of relocation and changing jobs on commuting modes and travel times in the UK
over the years. Kriger et al. (2011) investigated the factors that lead to major travel behav-
ior shifts and proposed implications of these changes from model development and data
collection perspectives.
Another series of studies were found with the primary focus on the stability of models
over time (cf. Agyemang-Duah and Hall 1997; Badoe and Miller 1995; Badoe and Steuart
1997; Elmi, Badoe, and Miller 1999; Karasmaa and Pursula 1997; Valentin, Prashker, and
Shiftan 2003) for which Fox and Hess (2010) provide a comprehensive synthesis. Ma
(1997) investigated the temporal stability of daily activity and travel pattern at person
and household level using cluster analyses and contingency table method. Mannering, Mur-
akami, and Kim (1994) also studied the transferability of an activity type choice model and
home stay duration model and found no stability in the models over the years. More
recently, Fox et al. (2014) performed the temporal transferability tests of a mode-destination
choice model for the Greater Toronto and Hamilton region and found that level of service
variables were more transferrable than cost variables and socio-economic variables.
Other research efforts have been carried out from a modeling structure perspective.
Huntsinger and Rouphail (2013) examined temporal stability of advanced models (logit
models and cumulative logistic regressions) and additional explanatory variables defining
life cycle, area type and accessibility in order to test whether these variables and advanced
model structures would improve transferability. Forsey et al. (2014) tested Heteroskedastic
generalized extreme value (GEV)-class choice models, while Sikder and Pinjari (2013)
explored multiple discrete continuous extreme structures.
Taking advantage of two waves of regional household travel survey (RHTS) datasets,
the goal of our study could be summarized as investigating whether there are major
travel behavior shifts over the past decade, and if so, in what way the changes have
occurred and to what extent are they taking place. Tour generation choice models are
developed based on the two datasets and the influence of various socio-economic and
demographic variables are examined. One of the advantages of employing repeated
TRANSPORTATION PLANNING AND TECHNOLOGY 3

cross sectional survey data is that this approach can capture the unobserved heterogeneity
phenomena over time (Davies and Pickles 1985).
This paper contributes to the literature by providing a closer look at travel behavior
shifts over time at a disaggregate level through the employment of a pooled modeling
structure based on two waves of large scale RHTS data collected in 1998 and 2010. This
study focuses on tour generation models for one mandatory purpose – the commute –
and one non-mandatory purpose – shopping – to examine whether the trip-making
behavior for one purpose exhibits more stability than another over time.

2. Data and statistics


This study used two datasets from two waves of the RHTS for the New York metropolitan
region, conducted in 1997–1998 and 2010–2011. The surveys collected detailed personal
and household characteristics, as well as travel diary information from households across
the 28 counties in the region. The 1998 survey captured nearly 91,000 trips taken by about
11,000 households (Parsons Brinckerhoff Quade and Douglas et al. 2000, 2005). The
2010–2011 survey recorded almost 144,000 trips from about 19,000 households and
more than 43,000 individuals.
Consistency of the two datasets was checked. Extensive data processing was performed
to consolidate the data variable categories to ensure compatibility between the two sources.
Table 1 shows a comparison between the two samples for selected socioeconomic and
demographic variables after applying the corresponding expansion factors respectively

Table 1. Sample composition for 1998 and 2010 datasets.


Statistics
2010 Sample
Variables 1998 Sample Adjusted for inflation (with CPI)
Household income High (>US$75k) 20.1% 21.8% 21.8%
Medium (US$30k-75k) 41.4% 29.3% 41.9%
low (<US$30k) 38.5% 48.9% 36.3%
Employed 64.0% 60.0%
Household size (avg.) 2.61 2.53
No. of vehicles in household (avg.) 1.43 1.43
Gender – Male 48.3% 46.2%
Driving license – Yes 79.1% 81.4%
Age <16 25.8% 21.1%
17–25 10.3% 7.5%
26–35 15.5% 12.1%
36–50 25.7% 22.5%
51–65 13.2% 23.6%
66+ 9.5% 13.1%
Household role Male & No Child 23.2% 25.1%
Male & 0–4 Child 10.9% 12.2%
Male & 5–15 Child 15.1% 8.9%
Female & No Child 25.1% 30.4%
Female & 0–4 Child 10.1% 13.1%
Female & 5–15 Child 15.5% 10.1%
Person type Full-time Worker 40.2% 36.4%
Part-time Worker 5.6% 7.6%
Retired 3.1% 3.6%
Non-worker 24.6% 30.0%
University Student 10.7% 9.7%
School Student 15.8% 12.7%
4 K. SHAMS ET AL.

for both datasets. The inflation concept is also accounted for based on the Consumer Price
Index (CPI) obtained from the U.S Bureau of Labor statistics.
The table shows that household shares by income categories are generally consistent,
with a slight increase in the share of the medium-income group. The employment rate
showed a slight decrease (about 1.8%) from 1998. Population percentages by age group
indicate an aging population in the region, with decreasing shares in all age groups
below 50, and significant increases in the shares for those above 50. These observations
are generally consistent with the nationwide trend. From 1998 to 2010, the number of
household vehicles remains the same, while household size decreased slightly. In terms
of household structure, there seems to be a lower ratio of households with youngest chil-
dren between 5 and 16 years old. Comparing with 1998, there seems to be fewer students
and full-time workers, while more part-time workers, non-workers and retirees are
observed.
This study also incorporated the area type variables that were defined in the New York
Best Practices Model (Parsons Brinckerhoff Quade and Douglas et al. 2000). There are 11
area types defined by both employment and population density, which are calculated
based on a 0.75 mile buffer area from the center of each zone. From Area Type 1
through 11, it represents a general decreasing trend in employment and housing densities,
in other words, from the high density urban core to rural areas.
The two datasets were processed, and tour files were generated based on travel infor-
mation from the survey data. Figure 1 presents the daily average tour rates by various per-
sonal and household variables for comparison between the 1998 and 2010 samples.
In general the figure indicates decreasing tour rates from 1998 to 2010 across all cat-
egories, except for commuting and school tours. The general patterns remain consistent
between the years, e.g. females and licensed drivers took more tours than males and
non-drivers respectively, high-income groups and bigger households showed higher
tour rates than others, and people aged between 35 and 50 had the highest tour rates.
ANOVA tests were also performed to examine whether the differences in tour rates are
statistically significant. The t-statistics show the significant differences in tour rates
between the two survey years for all categories, except for households having more than
five members, people aged between 17 and 25, and those aged between 35 and 50.
In summary, the travel patterns remain consistent at the aggregate level with a general
decreasing trend in terms of trip-making between 1998 and 2010. Further analysis at a dis-
aggregate level is carried out through econometric modeling.

3. Methodology
First, tour generation models by purpose (commuting and shopping) are developed for
individual data years (1998 and 2010). Multinomial logit models are employed for this
study. For a given individual n, the probability of making zero, one, or more tours is
obtained through:
emUi,n
Pi,n =  mUi,n , Uin = bi Xin + ein (1)
e
where, Pin = choice probability of alternative i for individual n; Ui,n = utility for individual
TRANSPORTATION PLANNING AND TECHNOLOGY 5

Figure 1. Comparison of average tour rates by person in 1998 and 2010.

n with alternative i; βi = vectors of estimable taste parameters associated with alternative i;


Xin = vectors of independent variables; €in = random error associated with the utility of
alternative i and individual n, which follows an Independent and Identically Gumbel dis-
tribution; µ = scale parameter, which is inversely related to the variance of the error term.
It should be noted that the scale parameter cannot be determined for a single dataset as
it remains confounded with the coefficients in the utility functions. Thus, in practice, it is
assumed that the variance of the unobserved factors is the same for all decision makers, so
the scale parameter for a single dataset becomes unity, or µ = 1. In this case, the estimated
taste parameters actually represent the true effects of the variables relative to the variance
of the unobserved factors. As a result, if the variance of the unobserved factors differs
between two datasets, the estimated model coefficients could be different even if the
taste parameters have the same effect on utility.
Therefore, for the purpose of this study, the scale effect of the unobserved factors in the
1998 and 2010 data needs to be accounted for, in order to compare the models developed
based on the two datasets. Scale parameters can be estimated using joint likelihood func-
tions and the sequential estimation method (Ben-Akiva and Lerman 1985; Ben-Akiva and
Morikawa 1990; Koppelman and Wilmot 1982; Swait and Louviere 1993). When pooled
models are developed combining both datasets, by allowing the scale of unobserved factors
6 K. SHAMS ET AL.

to vary, a ratio of the two scale parameters can be estimated. For this study, the 1998
dataset is normalized with respect to the 2010 dataset. The scale parameter for 2010 is
set to 1, while the estimated scale parameter µ from the pooled model reflects the
inverse variance ratios relative to the 2010 data source. The joint log-likelihood can be cal-
culated as:
N 
 J
LL(m, b) = dni ln (Pni ) (2)
n=1 i=1

The original parameters estimated based on the 1998 data (denoted as β1998) can then
be scaled to adjust for the variance of unknown factors relative to the 2010 data as
b∗ = b1998 /m, where b∗ is freed from the variance of unknown factors associated with
the datasets and is therefore comparable with the parameters obtained from the 2010 data.
The Model Equality Test Statistic (METS) is also computed to examine whether the
model parameters developed from the 1998 data are able to describe the travel pattern
observed in the 2010 data. The null hypothesis is that the taste parameters from the
two data sources are equal but the variances between the two datasets are unequal. The
METS can be calculated as:
METS = −2[LL(bJ )–LL(b98 ) − LL(b10 )] (3)
where, LL (βJ) = Log-likelihood of the joint model using both 1998 and 2010 datasets; LL
(β98) = Log-likelihood of the 1998 model, and LL (β10) = Log-likelihood of the 2010 model.
This statistic follows a chi-square distribution with degrees of freedom equal to the
difference between the summation of the number of estimates in individual years and
the number of estimates in the joint model (K98 + K10 – KJ).

4. Analysis of model results


Tour generation models are developed using data from the 1998 and 2010 surveys. Tables
2 and 3 respectively present the model estimation results for commuting and shopping
purposes. Each table presents the specifications for three models: the model based on
1998 data, the model based on the 2010 data, and finally the pooled model based on
both datasets. Only significant variables (at 95% confidence interval) are shown in the
table.
For both commuting and shopping models, the estimated scale parameters are
smaller than one, indicating that the 1998 data has a larger variance in the error
terms than the 2010 data. The variance for commute tour-making is larger than that
for shopping tours.

4.1. Commute tours


For commute tours, the 1998 model and the 2010 model show consistent results in general
as indicated in Table 2, except for part-time workers. The 1998 model indicates that part-
time workers are more likely to make one commute tour and less likely to make two or
more commute tours than other people, while the 2010 model shows negative impacts
on both alternatives. After adjusting for the scale effect, a comparison of the coefficient
TRANSPORTATION PLANNING AND TECHNOLOGY 7

Table 2. Summary of pooled tour generation model by commute purpose.


Commute tour frequency (Base: 0 Tour)
1998 2010 Pooled Model
Variables 1 tour 2+ tours 1 tour 2+ tours 1 tour 2+ tours
Constant 1998 −0.50 −3.57 0.30 −5.2
(−7.12) (−10.04) (3.0) (−11.4)
2010 1.57 −1.20 1.7 −1.74
(22.2) (−12.0) (16.4) (−7.4)
Income High Income −0.42 −0.44
(−6.39) (−4.0)
Low Income 0.11
(2.08)
Age group Age 17–25 0.18 0.31 0.25
(2.13) (3.7) (3.11)
Age 26–35 0.24 0.19
(4.5) (3.78)
Age 51–65 0.50 0.33 0.34
(3.37) (5.3) (4.65)
Land Use Area type 1
Area type 3 0.17 0.22
(2.18) (2.54)
Area type 4 0.12 0.12 0.25
(1.5) (1.47) (2.92)
Area type 5 0.15
(5.19)
Area type 6 0.25 0.15 0.29
(3.26) (2.8) (4.34)
Area type 7 0.54 0.57 0.68
(2.27) (3.18) (2.07)
Area type 8 0.17 0.41 0.15 0.58 0.26 0.64
(2.61) (2.4) (2.9) (6.57) (4.33) (5.88)
Area type 9 0.19 0.25
(2.77) (2.52)
Household Role Female & −0.30 −0.21 −0.19 −0.29 −0.30
0–4 Child (−6.87) (−5.7) (−2.5) (−.0) (−.75)
Female & −0.15 −0.14 −0.23 −0.19
5–15 Child (−2.08) (−2.5) (−3.86) (−1.65)
Male &
0–4 Child
Male & 0.23 0.42
5–15 Child (3.19) (2.0)
Person type Part-time worker 0.58 −0.40 −1.16 −0.88 −1.13 −0.74
(8.16) (−2.47) (−28.7) (−11.96) (−27.98) (−9.78)
Household Vehicle Total Number 0.09 0.06 0.17 0.07 0.14
(1.83) (3.43) (5.8) (4.05) (4.45)
Scale Parameter 1998 0.609
2010 1
Adjusted R 2 0.28 0.376 0.347
Log Likelihood (Initial) −9478 −23434 −32913
Log Likelihood (Final) −6779 −14619 −21431

values reveals some interesting results. The t-statistics, that test for the significance level of
differences between coefficients, show significant changes in a few parameters.
While income has no significant impact on commute tours in the 1998 model, results
suggest a negative impact on the high income category of multiple commute tours in 2010.
In other words, high income individuals are more likely to make one single work tour, if
any. This may be related to their high status and related responsibilities. With improve-
ments in computing and telecommunications technology, managers and supervisors
who are usually located at higher levels of organizational hierarchy, are less likely to
8 K. SHAMS ET AL.

Table 3. Summary of pooled tour generation models by shopping purpose.


Shopping purpose tour model (base: 0 tour)
1998 2010 Pooled Model
Variables 1 tour 2+ tours 1 tour 2+ tours 1 tour 2+ tours
Constant 1998 −2.30 −4.45 −2.76 −6.32
(−14.5) (−12.0) (−13.9) (−13.78)
2010 −3.02 −5.92 −2.81 −6.09
(−36.2) (−26.7) (−45.4) (−32.7)
Income High Income −0.14
(−4.02)
Low Income 0.17
(4.04)
Age group Age <17 −1.23 −0.93 −1.43
(−5.58) (−5.02) (−1.97)
Age 17–25 −0.56 −0.65 −0.66
(−3.98) (−7.64) (−8.73)
Age 26–35 −0.38 −0.52 −0.16 −0.30
(−3.89) (−2.03) (−2.43) (−5.26)
Age 51–65 0.17 0.30 0.14 0.32
(4.49) (3.05) (4.06) (3.67)
Land Use Area type 1 1.09 0.91
(2.54) (2.41)
Area type 4 −0.18 −0.12
(−2.36) (−1.78)
Area type 8 0.19 0.33 0.16
(1.99) (2.48) (1.42)
Area type 9 0.28 0.09
(3.52) (2.87)
Household Role Female & 0.36 0.13 0.09
0–4 Child (2.37) (3.33) (1.5)
Female & 0.48 0.25 0.32 0.12
5–15 Child (3.78) (3.86) (2.01) (2.41)
Female & No child 0.26 0.09
(3.17) (2.38)
License status Licensed 0.49 0.78 0.63 0.94 0.59 0.87
(4.56) (2.6) (11.94) (6.15) (12.08) (6.04)
Person Type Part time worker 1.12 0.89 1.19 0.33 1.03
(4.2) (14.8) (6.5) (6.01) (6.29)
Retired 0.77 1.54 0.62 0.16 1.05
(8.14) (7.3) (5.56) (2.04) (4.95)
Non-worker 0.74 1.59 1.46 1.99 1.07 1.88
(7.53) (7.09) (33.3) (15.9) (29.41) (16.09)
School Student −1.57 −3.96
(−14.87) (−10.87)
Scale 1998 0.726
2010 1
Adjusted R 2
0.516 0.594 0.579
Log Likelihood (Initial) −7722 −35199 −44389
Log Likelihood (Final) −3705 −14265 −18638

take part in multiple daily commutes; instead, they rather control everything from their
office (or home) or shift the supervision responsibilities to other personnel.
A brief review of age categories reveals a few noteworthy issues. For instance, there is no
significant shift in the commute behavior of 17–25 age groups. The probability of individ-
uals aged 26–35 making one single commute tour has increased while the contribution of
age category 51–65 to multiple commute tours has decreased. Considering the fact that
individuals between 17–25 years old are mostly students and have more constrained
time budgets may well explain why they are now less likely to make multiple commutes.
TRANSPORTATION PLANNING AND TECHNOLOGY 9

On the other hand, as individuals grow older, they are less likely to make multiple tours
and utilize any other work arrangement opportunity such as telecommuting.
Taking household roles into account, it looks as if a more balanced distribution of work
among household members is observed in 2010 compared to 1998. In view of that, single
commute tours in 1998 are positively affected by certain male household members and
negatively impacted by certain female roles. This shows a traditional structure where
males are more responsible for out-of-home work than females. However, the observed
increase in female coefficients (followed by a decrease in male influence on the model)
bodes for a more equitable responsibility structure where females are as much involved
in occupational activities as men.
The shift observed in part time workers’ behavior is also interesting. Accordingly, part
time workers showed a high positive impact on single commutes back in 1998, while this
turned into a significant negative coefficient in 2010. Considering their negative effect on
multiple commutes, it looks as if part time workers are also more interested in reducing
their daily commutes through other work arrangements such as telecommuting.
In terms of land use, one may easily observe that more area types are involved in commute
trips in 2010. While area types 4, 6, and 8 reflect significant positive impacts on the model in
1998, the 2010 model adds three more area types (including 3, 7, and 9) which in turn results in
a relative reduction of other area types compared to 1998. This might suggest a more balanced
geographical distribution of commute trips in 2010 which could stem from urban develop-
ment between the time period. The higher employment densities of area types 3 and 7, for
example, provide solid evidence to explain their positive contribution to the model in 2010.
It is also interesting to see that area type 9 with relatively low population and employment
density has a positive contribution on multiple commute trips compared to 1998.
The contrast between model coefficients can be visualized in Figure 2. It graphs the
1998 model coefficients after applying the scale parameter (represented by the horizontal
axis) against the 2010 model coefficients (represented by the vertical axis). A solid line

Figure 2. Parameter comparison of 1998 and 2010 commute tour generation model.
10 K. SHAMS ET AL.

with 45 degree angle (slope equals 1) is also provided as a reference, which indicates a
perfect match between the two models. The closer the dots are to the reference line, the
smaller the differences between the two coefficient values. Again, part-time workers
(having different signs in the two datasets), females with small children, people aged
between 51 and 65 and area type showed significant deviations from the reference line.

4.2. Shopping tours


For shopping tours, the 1998 model and 2010 model show consistent signs for all expla-
natory variables. This corroborates the larger scale parameter in the shopping model,
which indicates that there is less noise in the 1998 data in terms of trip-making behavior
for shopping purposes than for commute purposes. Comparing the coefficient values
across the data sources reveals significantly higher impacts on shopping tour-making in
1998 for several factors. In particular, people aged between 26 and 35 showed higher nega-
tive propensity in making shopping tours, while females (regardless of whether they have
children or the age of their children) and retired persons were significantly more likely to
make shopping tours in 1998 than in the 2010 data.
Results indicate that income has no impact on any of the models, neither in 1998 nor in
2010. Though a positive correlation between income and shopping activities (trips) is gen-
erally expected, one reason could be the fact that shopping activities in this study included
a wide range of activities such as routine daily retail shopping to specific shopping pur-
poses with lower frequencies (such as clothes, furniture, etc.). While the latter is certainly
affected significantly by income, the former is regarded as a casual (and necessary) activity
on a daily basis among all income categories and therefore may not be significantly influ-
enced by low or high incomes.
In view of age groups, minor individuals have no impact on the 1998 model. In 2010,
however, they tend to reduce the probability of one single shopping tour per day. In other
words, if they want to make shopping tours, they are more likely to have more than one
tour per day in 2010 compared to 1998. This may stem from higher mobility and acces-
sibility opportunities or may simply reflect changes in minor people lifestyle. Furthermore,
people aged 26–35 were less likely to participate in shopping tours back in 1998. In
addition to mobility and accessibility constraints, the fact that such the age group consists
mostly of workers can shed light on the situation. Considering the improvements in terms
of technology and telecommunications, workers have more freedom these days and are
more likely to participate in non-mandatory activities compared to stricter spatial con-
straints 12 years before. It is also interesting to see that people aged 51–65 are much
more likely to accomplish shopping trips in 2010.
In terms of family roles, the impact of females on single shopping tours has significantly
decreased from 1998 to 2010. Accordingly, this might bode for a more balanced distri-
bution of maintenance duties nowadays as males show similar participation in shopping
activities as females. However, in the presence of children aged 5–15, the probability of
more than one shopping tour per day increases, which may signify how household struc-
ture can impose more responsibilities on female members.
Licensed drivers do not show significant behavioral changes in the study period. Results
indicate that both in 1998 and 2010, holding a driver’s license will significantly increase the
probability of single or multiple shopping tours per day.
TRANSPORTATION PLANNING AND TECHNOLOGY 11

Figure 3. Parameter comparison of 1998 and 2010 shopping tour generation model.

In view of person type, the probability that a part time worker makes a single shopping
tour has increased in 2010. In contrast, retired individuals are less likely to make single or
multiple shopping tours in 2010 compared to 1998.
A comparison of the two models with emphasis on land use reveals higher preferences
for single shopping tours in areas with low population and employment density in 1998. In
2010, area type 1 with the highest employment density shows the highest positive impact
on multiple shopping tours. In addition, in the case of area type 8, there is a shift from
single shopping tours to multiple tours.
Similarly, a plot is created that graphs the 1998 model coefficients after applying the
scale parameter against the 2010 model coefficients, as shown in Figure 3. Parameters
that show large deviation from the reference line are labeled.

4.3. Model equality tests


As mentioned in the methodology section, a METS test is carried out to examine the
hypothesis of equal taste parameters and unequal variances between the two data
sources. The METS for the commute model equals −2 [(−21431) – (−6779) –
(−14619)] = 66, which is greater than the critical value for chi-square of 40.1 with 27
degrees of freedom. The METS for the shopping model equals −2 [(−18638) – (−3705)
– (−14265)] = 1336, which is greater than the critical value for chi-square of 45 with 31
degrees of freedom. Both tests reject the null hypothesis, indicating that the parameters
from the two data sources are not equal.

5. Conclusions
The major objective of this paper has been to explore how travel behavior changes through
time and what have been the major contributing factors. In this respect, this research work
12 K. SHAMS ET AL.

examined two waves of travel survey data (1998 and 2010) through a pooled model struc-
ture. By accounting for the difference in data variances, the pooled model estimated the
scale factor, providing the analyst with the capability of comparing the true impacts of
the model parameters on travelers’ tour-making behavior. Two major tour purposes
were investigated, namely commuting and shopping. The 1998 data reflected larger var-
iance magnitudes compared to the 2010 data. Furthermore, results indicated that travel
behavior is exposed to significant behavioral changes as time passes. Such variations
stem from several socio-economic and demographic factors to land use and built environ-
ment issues.
In terms of commute tours, for instance, a more balanced distribution between males
and females was observed in 2010, which may suggest a more modern family structure
where females are as responsible as men in out-of-home work activities. Part time
workers reflected higher negative impacts on daily commutes (either single or multiple)
which may stem from more sophisticated work arrangements such as the development
of telecommuting opportunities which did not exist a decade before. In addition, more
area types were involved in commute generation in recent years which indicates the
impact of urban sprawl in mandatory trip generation.
When it comes to shopping, the impact of females on single shopping tours has signifi-
cantly decreased from 1998 to 2010. Accordingly, this might suggest a more balanced dis-
tribution of maintenance duties in recent times as males show similar participation in
shopping activities as females. However, in the presence of children aged 5–15, the prob-
ability of more than one shopping tours per day increased, which indicates how household
structure can impose more responsibilities on female members. Focusing on person type,
part time workers were more prone to making a single shopping tour in 2010. In contrast,
retired individuals were less likely to make single or multiple shopping tours in 2010 com-
pared to 1998. Comparing the two models with emphasis on land use, higher preferences
for single shopping tours were observed in areas with low population and employment
density back in 1998. In 2010, high employment density (area type 1) showed the
highest positive impact on multiple shopping tours.
In general the magnitudes of coefficients have decreased, which is consistent with the
overall decreasing trend in traveling. Overall, the model equality tests indicated that the
models developed based on the two data sources do not have equal taste parameters,
thus are not transferable. This may have implications on applying models based on
cross-sectional data, especially over long time periods. The pooled model structure pro-
vides a means to take advantage of multiple data sources which will lead to better estimates
and understanding of travel behavior. Future work will be carried out expanding the scope
to other travel purposes and other choice behavior.

Disclosure statement
No potential conflict of interest was reported by the authors.

References
Agyemang-Duah, Kwaku, and Fred L. Hall. 1997. “Spatial Transferability of an Ordered Response
Model of Trip Generation.” Transportation Research Part A: Policy and Practice 31 (5): 389–402.
TRANSPORTATION PLANNING AND TECHNOLOGY 13

Asgari, Hamidreza, and Xia Jin. 2017. “An Evaluation of Part-Day Telecommute Impacts on Work
Trip Departure Times.” Travel Behavior and Society. doi:10.1016/j.tbs.2017.04.002.
Asgari, Hamidreza, Xia Jin, and Ali Mohseni. 2014. “Choice, Frequency, and Engagement – A
Framework for Telecommuting Behavior Analysis and Modeling.” Transportation Research
Record: Journal of the Transportation Research Board 2413: 101–109.
Badoe, Daniel A., and Eric J. Miller. 1995. “Analysis of the Temporal Transferability of Disaggregate
Work Trip Mode Choice Models.” Transportation Research Record 1493: 1–11.
Badoe, Daniel A., and Gerald N. Steuart. 1997. “Urban and Travel Changes in the Greater Toronto
Area and the Transferability of Trip-Generation Models.” Transportation Planning and
Technology 20 (4): 267–290.
Ben-Akiva, Moshe E., and Steven R. Lerman. 1985. Discrete Choice Analysis: Theory and
Application to Travel Demand. Vol. 9. Cambridge, MA: MIT Press.
Ben-Akiva, Moshe, and Takayuki Morikawa. 1990. “Estimation of Travel Demand Models from
Multiple Data Sources.” International symposium on transportation and traffic theory, 11th,
1990, Yokohama.
Dargay, Joyce, and Mark Hanly. 2007. “Volatility of Car Ownership, Commuting Mode and Time
in the UK.” Transportation Research Part A: Policy and Practice 41 (10): 934–948.
Davies, Robert B., and Andrew R. Pickles. 1985. “Longitudinal Versus Cross-Sectional Methods for
Behavioural Research: A First-Round Knockout.” Environment and Planning A 17 (10): 1315–
1329.
Elmi, Abdishakor, Daniel Badoe, and Eric Miller. 1999. “Transferability Analysis of Work-Trip-
Distribution Models.” Transportation Research Record: Journal of the Transportation Research
Board 1676: 169–176.
Forsey, David, Khandker Nurul Habib, Eric J. Miller, and Amer Shalaby. 2014. “Temporal
Transferability of Work Trip Mode Choice Models in an Expanding Suburban Area: The
Case of York Region, Ontario.” Transportmetrica A: Transport Science 10 (6): 469–482.
Fox, James, Andrew Daly, Stephane Hess, and Eric Miller. 2014. “Temporal Transferability of
Models of Mode-Destination Choice for the Greater Toronto and Hamilton Area.” Journal of
Transport and Land Use 7 (2): 41–62.
Fox, James, and Stephane Hess. 2010. “Review of Evidence for Temporal Transferability of Mode-
Destination Models.” Transportation Research Record: Journal of the Transportation Research
Board 2175: 74–83.
Gärling, Tommy, Daniel Eek, Peter Loukopoulos, Satoshi Fujii, Olof Johansson-Stenman, Ryuichi
Kitamura, Ram Pendyala, and Bertil Vilhelmson. 2002. “A Conceptual Analysis of the Impact of
Travel Demand Management on Private Car Use.” Transport Policy 9 (1): 59–70.
Goodwin, P., R. Kitamura, and H. J. Meurs. 1990. “Some Principles of Dynamic Analysis of Travel
Behaviour.” In New Developments in Dynamic and Activity-Based Approaches to Travel Analysis,
edited by P. Jones, 56–72. Aldershot: Gower.
Hensher, David A. 1985. “Longitudinal Surveys in Transport: An Assessment.” New Survey Methods
in Transport. Utrecht: VNU Science Press.
Hossan, Md Sakoat, Hamidreza Asgari, and Xia Jin. 2016. “Investigating Preference Heterogeneity
in Value of Time (VOT) and Value of Reliability (VOR) Estimation for Managed Lanes.”
Transportation Research Part A: Policy and Practice 94: 638–649.
Hosseinlou, Mansour H., Esmaeil Balal, Aidin Massahi, and Iman Ghiasi. 2012. “Developing
Optimal Zones for Urban Parking Spaces by Arc GIS and AHP.” Indian Journal of Science
and Technology 5 (11): 3618–3622.
Hosseinlou, Mansour H., Aidin Massahi, and Mehdi V. Aliabadi. 2012. “Developing Tehran
Vehicles’ Air Pollution Macroscopic Models and Presenting the Method for Estimating
Pollutants’ Emission Rates in Urban Networks.” Science Series Data Report 4 (4): 73–92.
Huntsinger, Leta, and Nagui Rouphail. 2013. “Temporal Stability of Generation Choice Models.”
Transportation Research Record: Journal of the Transportation Research Board 2344: 79–87.
Jin, Xia, Hamidreza Asgari, and Md Sakoat Hossan. 2014. “Understanding Trip Misreporting in
Household Travel Surveys by Comparing GPS-Assisted and Diary-Based Samples.” CICTP
2014: 3401–3412.
14 K. SHAMS ET AL.

Jin, Xia, Md Sakoat Hossan, Hamidreza Asgari, and Kollol Shams. 2017. “Incorporating Attitudinal
Aspects in Roadway Pricing Analysis.” Transport Policy. doi:10.1016/j.tranpol.2017.03.020.
Karasmaa, Nina, and Matti Pursula. 1997. “Empirical Studies of Transferability of Helsinki
Metropolitan Area Travel Forecasting Models.” Transportation Research Record: Journal of the
Transportation Research Board 1607: 38–44.
Kitamura, Ryuichi. 1990. “Panel Analysis in Transportation Planning: An Overview.”
Transportation Research Part A: General 24 (6): 401–415.
Kitamura, Ryuichi, Toshiyuki Yamamoto, and Satoshi Fujii. 2003. “The Effectiveness of Panels in
Detecting Changes in Discrete Travel Behavior.” Transportation Research Part B: Methodological
37 (2): 191–206.
Koppelman, Frank S., and Chester G. Wilmot. 1982. Transferability Analysis of Disaggregate Choice
Models. HS-035 371.
Kriger, D., R. Wolff, A. Subhani, and D. Cleghorn. 2011. “Trends in Urban Travel: Implications for
Planning and Forecasting.” 2011 conference and exhibition of The transportation association of
Canada. transportation successes: let’s build on them. 2011 congress et exhibition de
l’Association des transports du Canada, Les Succes en Transports: Une Tremplin vers l’Avenir.
Ma, Jun Konstadinos G. 1997. “A Dynamic Analysis of Person and Household Activity and Travel
Patterns Using Data from the First Two Waves in the Puget Sound Transportation Panel.”
Transportation 24 (3): 309–331.
Mannering, Fred, Elaine Murakami, and Soon-Gwan Kim. 1994. “Temporal Stability of Travelers’
Activity Choice and Home-Stay Duration: Some Empirical Evidence.” Transportation 21 (4):
371–392.
Massahi, Aidin. 2017. “Multi-Resolution Modeling of Dynamic Signal Control on Urban Streets.”
FIU Electronic Theses and Dissertations 3349.
Massahi, Aidin, Mohammed Hadi, Maria Adriana Cutillo, and Yan Xiao. 2017. “Estimating the
Effects of Urban Street Incidents on Capacity.” Transportation Research Record: Journal of the
Transportation Research Board 2615: 55–61.
Massahi, Aidin, Mohammed Hadi, Yan Xiao, Tao Wang, and Xuanwu Chen. 2016. “Improved
Model for Estimating Incident Impact on Urban Street Travel Time with Consideration of
Upstream Intersection Capacity Reduction.” Transportation research board 95th annual
meeting.
Noland, Robert B., and Lewison L. Lem. 2002. “A Review of the Evidence for Induced Travel and
Changes in Transportation and Environmental Policy in the US and the UK.” Transportation
Research Part D: Transport and Environment 7 (1): 1–26.
Parsons Brinckerhoff Quade & Douglas, Cambridge Systematics, NuStats International. 2000.
“Regional Travel-Household Interview Survey Methods and Implementation.” Report for the
New York Metropolitan Transportation Council (NYMTC) and the North Jersey
Transportation Planning Authority (NJTPA).
Parsons Brinckerhoff Quade & Douglas, PB Consult, AECOM Consult, Ubitran Associates,
Ubanomics, Alex Anas & Associates, NuStats International, George Hoyt & Associates. 2005.
“Transportation Models and Data Initiative.” General Final of New York Best Practive Model
(NYBPM). Report for the New York Metropolitan Transportation Council (NYMTC).
Polzin, Stephen E. 2006. The Case for Moderate Growth in Vehicle Miles of Travel: A Critical
Juncture in US Travel Behavior Trends. Tampa, FL: University of South Florida, Center for
Urban Transportation Research.
Raimond, Timothy, and David A. Hensher. 1997. “A Review of Empirical Studies and
Applications.” In Panels for Transportation Planning, 15–72. Boston, MA: Springer.
Shabanpour, Ramin, Nima Golshani, Sybil Derrible, Abolfazl Mohammadian, and Mohammad
Miralinaghi. 2017. “Joint Discrete-Continuous Model of Travel Mode and Departure Time
Choices.” Transportation Research Record: Journal of the Transportation Research Board 2669:
41–51.
Shabanpour, Ramin, Mahmoud Javanmardi, Mehran Fasihozaman, Mohammad Miralinaghi, and
Abolfazl Mohammadian. 2017. “Investigating the Applicability of ADAPTS Activity-Based
Model in Air Quality Analysis.” Travel Behavior and Society doi:10.1016/j.tbs.2017.02.004.
TRANSPORTATION PLANNING AND TECHNOLOGY 15

Shams, Kollol, Hamidreza Asgari, Md Sakoat Hossan, and Xia Jin. 2017. “Value of Reliability for
Road Freight Transportation: Evidence from a Stated Preference Survey in Florida.”
Transportation Research Record, Journal of Transportation Research Board. doi:10.3141/2610-05.
Shams, Kollol, Hamidreza Asgari, and Xia Jin. 2016. “Valuation of Travel Time Reliability in
Freight Transportation: A Review and Meta-Analysis of Stated Preference Studies.”
Transportation Research Part A: Policy and Practice 102: 228–243.
Sikder, Sujan, and Abdul Pinjari. 2013. “Spatial Transferability of Person-Level Daily Activity
Generation and Time Use Models: Empirical Assessment.” Transportation Research Record:
Journal of the Transportation Research Board 2343: 95–104.
Swait, Joffre, and Jordan Louviere. 1993. “The Role of the Scale Parameter in the Estimation and
Comparison of Multinomial Logit Models.” Journal of Marketing Research 30: 305–314.
Valentin, Cotrus A., Joseph N. Prashker, and Yoram Shiftan. 2003. “Analysis of Trip Generation
Characteristics in Israel for the Years 1984, 1996/7 and Spatial & Temporal Transferability of
Trip Generation Demand Models.” The 2003 Transportation Research Board Annual
Meeting, Washington, DC, January 2003.

You might also like