Content uploaded by Daejin Kim
Author content
All content in this area was uploaded by Daejin Kim on Sep 28, 2017
Content may be subject to copyright.
1
Employer-Based Travel Demand Management Program: Employer’s Choice
and Effectiveness
Joonho Ko, Ph.D.
Department of Transportation Research
The Seoul Institute
57 Nambusunhwan-ro Seocho-gu, Seoul 06756, South Korea
Tel: 82-2-2149-1127; Fax: 82-2-2149-1120; Email: jko@si.re.kr
Daejin Kim (Corresponding Author)
School of Civil and Environmental Engineering
Georgia Institute of Technology
790 Atlantic Dr NW, Atlanta, GA 30318, United States
Tel: 1-678-707-0472; Email: daejin.kim@gatech.edu
2
Employer-Based Travel Demand Management Program: Employer’s Choice
and Effectiveness
Abstract
The impact and effectiveness of employers’ travel demand management (TDM) programs have often
been investigated, but there is little research on employers’ preference of TDM programs and the factors
which influence their choices. This study aims to examine employers’ choices of TDM program and the
effectiveness based on a Seoul TDM database. Factors affecting employers’ decisions on whether to
participate in TDM programs and the degree of participation are identified based on regression models:
binary logistic, ordinary least squares and Tobit regressions. In addition, an employers’ TDM program
choice model is developed using the approach of the multiple discrete continuous extreme value
(MDCEV) model, as an employer can participate in multiple TDM programs, and their individual
effectiveness is represented by continuous values of traffic impact fee discount rates. The developed
models consistently indicate that facilities’ characteristics such as owner type and facility size, along with
locational characteristics (e.g., land use and accessibility to transit) are important in choosing a TDM
program. In particular, the MDCEV model shows that employers’ preferences for TDM programs vary
significantly by employer and characteristics of the program.
Keywords: Travel demand management, Employer, Worksite, Multiple discrete-continuous extreme
value model, Transport policy
1. Introduction
An important objective of transport policies in a city is to reduce citizens’ car dependency, encouraging
alternative modes of travel such as public transit, walk and bicycle. It has been shown that the mode shift
from car to alternative modes can be achieved to a substantial degree by various travel demand
management (TDM) strategies. Indeed, James and Brög (2001) showed that in South Perth, Australia, a
TravelSmart® individualized marketing strategy to convert car trips to walking trips by inducing
behavioral changes reduced car usage by 14%. Deployment of carsharing and public bicycle share
programs are also considered effective in reducing car trips and car ownership (Cervero and Tsai, 2004;
3
DeMaio, 2009). Above all, measures focusing on employment sites and commute trips are of importance
since they can mitigate peak time traffic congestion, which usually incurs tremendous social costs
resulting from excessive travel time and wasteful fuel consumption. Transportation agencies have
therefore conceived and implemented various worksite-based interventions in the form of incentives,
disincentives and/or marketing (Dill and Wardell, 2007).
The worksite-based intervention, often called employer-based TDM, is an efficient tool in the
effort to reduce car trips as it is relatively easy to implement, control and track performance (Hasnine et
al., 2016). TDM strategies encompass a wide variety of options including alternative work schedules,
guaranteed ride home, shuttle services and vehicle-use restrictions. These strategies are often selectively
implemented according to their suitability to the characteristics of the employer (e.g., municipal/local
governments and business associations) and particular situations (e.g., geographic conditions, problems to
be addressed and the decision-makers involved) (Litman, 2003). Thus, suitability evaluations under
various contexts are inherently important as they can guide employers in the right direction when
selecting strategies and implementation approaches. A number of studies point to this, empirically
evaluating the effectiveness of TDM strategies (Brockman and Fox, 2011; Merom et al., 2005; O’Fallon,
2010). Together with these evaluations, an understanding of employers’ behavior concerning TDM
strategy adoptions can help policy makers identify and develop effective TDM strategies. However, it
appears that the behavior of employers has rarely been investigated.
Seoul, South Korea’s largest city, has also recognized the need for implementing employer-based
TDM programs as traffic congestion becomes a serious social issue, increasing the risk of degrading the
quality of life and the health of citizens. The employer-based TDM in Seoul is unique in that employers
are incentivized by discounting annual traffic impact fees which are charged to facility owners based on
the size (i.e., floor area) and type (e.g., commercial or educational) of facility (Ko, 2013). The size and
type are generally believed to be the major determinants for traffic induced by the facility. Indeed, the
traffic impact fee for a facility is computed by multiplying three factors: total floor area in square meters,
facility type weight and unit fee. The facility type weight ranges between 0.47 (for factories) and 9.83
(for department stores). The traffic impact fees can be reduced, proportional to the amount of reduced
vehicle traffic resulting from TDM program implementation, as a way to encourage employers to actively
support TDM programs. Seoul Metropolitan Government (SMG) considers a set of TDM programs for
the traffic impact fee discount and monitors program-by-program traffic reduction effects for target
facilities. The monitored results are reflected in the amount of annual traffic impact fees determined, and
the results are stored in a TDM database together with pertinent information on the facilities. This
4
database is a valuable resource enabling researchers to investigate the characteristics of employers’ TDM
participation.
Based on the Seoul TDM database, this study aims to examine the characteristics of the
employers’ choice of TDM program and the effectiveness of the various programs. For this, two analyses
are conducted. The first is to investigate factors affecting employers’ decisions on whether or not to
participate in a TDM program, and the extent of their participation. These investigations are implemented
separately after developing three statistical models of binary logistic regression (for the choice of whether
or not to participate), ordinary least square model and Tobit regression (for the degree of participation); a
Tobit model is supplementarily considered, as the dependent variable (degree of program effectiveness) is
non-negative. The second analysis is to identify the factors which affect an employer’s choice of TDM
program in a combined manner. For this, the multiple discrete continuous extreme value (MDCEV)
model, formulated by Bhat (2005) to handle situations of multiple discreteness combined with a
continuous dimension of choice is used, as one employer can participate in multiple TDM programs, and
their individual effectiveness is represented by continuous values (i.e., traffic impact fee discount rates).
The MDCEV models have been employed for modeling household vehicle fleet composition and usage
(Bhat and Sen, 2006; Imani et al., 2014; You et al., 2014) and activity time-use behavior (Eluru et al.,
2010; Pinjari and Bhat, 2010), but it has not yet been applied to TDM analysis. This study shows that a
number of factors, including facility characteristics (e.g., type and size) and locational factors (e.g., land
use and accessibility to transit) can affect the decision of TDM program participation and its effectiveness.
More importantly, the program-by-program variations of the effect are identified by the MDCEV model.
The results of the models developed are expected to enhance the understanding of employers’ TDM
participation decisions and TDM effectiveness, helping transportation agencies develop and implement
effective TDM programs at worksites.
2. Literature review
A number of studies have shown that a variety of employer-based TDM strategies exist which are
intended to stimulate a mode shift for commuting trips (Scheepers et al., 2014; Hasnine et al., 2016;
Litman, 2003). One such strategy is a travel awareness campaign which seeks participants’ behavioral
changes through social marketing and advertising, as shown in “The Walk to Work Day” campaign
implemented in Australia to encourage more walking. The effectiveness of the campaign was evaluated
by Merom et al. (2005) based on pre- and post-campaign telephone surveys. They found that the
participants in the campaign decreased their “car only” use and increased walking combined with public
5
transport. Economic incentives can also act as a stimulus to mode shifts as illustrated by the parking
cash-out program in California, USA. The program required employers of 50 or more persons who
subsidize commuter parking to offer a cash allowance to employees who are not using their parking
spaces due to mode shifts. Shoup (1997) pointed to the effectiveness of the program based on case
studies of eight firms: a decrease in the number of solo drivers and vehicle-miles traveled (VMT) for
commuting by 17 % and 12%, respectively.
Research has empirically proven that combined TDM measures (implementing several measures
together) can lead to increased reductions in car use when compared to individual measures (Eriksson et
al., 2010). In line with this, TDM measures of “carrot and stick” are often simultaneously implemented,
as illustrated in the Transport Plan of the University of Bristol, UK (Brockman and Fox, 2011). TDM
strategies in this plan included limiting parking spaces, increasing parking charges, improving facilities
for walkers and cyclists, and providing carsharing and free university bus services. A before-after survey
of university staff members revealed significant mode shifts after implementation of the measures: an
increase in the proportion of respondents who walk to work from 19% to 30%, and who cycle to work
from 7% to 12% (Brockman and Fox, 2011). Despite the fact that the university setting might produce
improved outcomes due to the greater work flexibility of staff members and their better understanding of
sustainable transportation, the combined TDM measures are advantageous in that they offer more options
to replace the use of a car to get to work. Herzog et al. (2006) revealed that comprehensive employee
commuter benefit packages, composed of financial incentives, services such as guaranteed ride home and
carpool matching, and informational campaigns, resulted in a reduction of trips and VMT by about 15%
for surveyed employees in the metro areas of Denver, Houston, San Francisco and Washington DC.
These impacts were identified by a survey comparing a commuter group receiving the benefits with
another which did not. Such campaign-type approaches are also often combined with physical
intervention tools. For example, the “Bike Now” program implemented in New Zealand deployed
combined TDM measures to overcome general misconceptions about cycling, while at the same time
bicycle facilities were improved, resulting in shifts from car to bicycle by nearly half of the surveyed
participants (based on a before-after survey targeting 27 workplaces) (O’Fallon, 2010).
The effectiveness of employer-based TDM programs can vary depending on the characteristics of
the employer and/or employees. Zhou et al. (2012) found that employees from a large employer (the
University of California, Los Angeles) had more opportunities for carpooling and thus drove alone less
often even after factoring in residential location, annual income and commute time. The study also
pointed out that different employee groups, usually marked by income levels, may favor different TDM
programs. The importance of the income factor was stressed by Loukopoulus et al. (2004) where it was
6
argued that car-user responses to TDM measures were mainly dictated by a cost-minimization principle.
The association between income and TDM effectiveness was also noted in Hendricks (2005). In addition
to the income effect, the study emphasized the importance of management support and proximity to good
transit service for the success of TDM programs. Dill and Wardell (2007) demonstrated through the
development of statistical models that worksite locational factors (street connectivity, whether it is located
downtown, and accessibility to transit) were critical to the worksite’s transit mode share as well as the
effectiveness of TDM programs. The study also found a positive relationship between employer size
(number of employees) and TDM effectiveness even for worksites with no rail access. These studies
suggest that the consideration of the context in which employers are situated is important to understand
TDM program’s effectiveness and employers’ TDM program choices.
With regard to the method of TDM program evaluation, many studies have relied on before-after
surveys, identifying behavioral differences after intervention (Brockman and Fox, 2011; James and Brög,
2001; Merom et al., 2005; O’Fallon, 2010). In other instances, the target groups’ mode choice changes
(or characteristics) were compared with those of control groups (Herzog et al., 2006; Zhou et al., 2012).
In addition to these applications of revealed preference (RP) data, stated preference (SP) surveys have
been conducted for estimating the expected impact of proposed policies, providing opportunities to
quantitatively compare a variety of potential policy options (Eriksson et al., 2010). Potential TDM
strategies can also be evaluated systematically under a modeling framework. As can be seen in Hasnine
et al. (2016), where a joint RP-SP mode choice model was applied, TDM evaluation tools which have the
capability of testing various scenarios can play such a role. Shiftan and Suhrbier (2002) applied activity-
based modeling approaches in an attempt to develop a tool for evaluating TDM strategies in terms of
travel and emission impact. Dill and Wardell (2007) constructed regression models to predict mode
shares after considering the impact of employers’ TDM strategies. However, the capability of these
approaches appears to be rather limited when it comes to revealing the variation of an individual TDM
program’s effectiveness, which usually interacts with an employer’s characteristics. In addition, the
analyses focused mostly on the behavior of travelers, with little consideration of employers’ TDM
program choices.
As a whole, the literature reviewed suggests that the effectiveness of TDM measures adopted by
employers has been studied often, usually demonstrating the positive impacts towards a reduction in car
commutes. The review also implies that a set of measures have commonly been implemented together to
maximize their effectiveness, requiring an evaluation approach that enables researchers to capture the
variation in program-by-program effectiveness. Previous studies also appear to have shown little interest
in an employer’s choice of TDM programs. It is conjectured that data availability can be the defining
7
issue for carrying out choice studies. Therefore, this study, attempting to investigate employers’ TDM
program participation characteristics, is expected to contribute to the literature by adding insights
regarding TDM program choices at the worksite, based on a reliable data source managed by a local
government.
3. Data
3.1 The employer-based TDM database
The employer-based TDM database is updated annually by SMG to appropriately assess the traffic impact
fees to owners of facilities with a floor area of 1,000 square meters or more. This study utilizes the
database built in 2013 which contains information on 15 employee-related TDM programs, as listed
below.
Car use restriction programs based on vehicle plate numbers (license plate rationing programs) or
drivers’ own choice: 1) final digit (one day out of every ten days), 2) last two digits (two days out
of every ten days), 3) odd and even (every other day), 4) weekly no-driving day program
Parking related programs: 1) no parking for employees, 2) eliminating free parking, 3) reducing
parking spaces
Programs encouraging the use of alternative travel modes: 1) transit subsidies, 2) free commuter
buses, 3) employer-provided taxi service, 4) transit day campaigns, 5) encouragement of bicycle
use
Other: 1) flexible work schedules, 2) carpools, 3) site-specific programs developed by the
employer
The weekly no-driving day program is implemented in Seoul and several other cities in South Korea. The
program participants are required to not use their private car at least one day a week; participants can
select the day(s) of the week considering their vehicle needs. Some incentives such as annual vehicle tax
reductions and parking fee discounts are offered by SMG to the participants unless they violate the no-
driving day rule three times or more in a year (Ko and Cho, 2009). The no-parking program for
employees is regarded as implemented when worksite parking is prohibited for at least 80% of the
employees.
The key element of the database is traffic impact fee discount rates (in percentages) for each
facility and the individual TDM programs implemented. The rate can be considered as a program-by-
8
program car trip reduction since it is designed to be proportional to the amount of reduced car trips by the
individual programs. SMG has developed a manual for the evaluation of the discount rate, specifying
program-by-program definitions, minimum requirements and criteria (e.g., a minimum number of
participating employees), formulae for computing the rate, monitoring methods and required documents
for verification. Facility owners can obtain reduced traffic impact fees based on the discount rates
(maximum 100%) summed over all the implemented TDM programs. The discount rates for a facility are
evaluated based on quarterly reports submitted by the facility owner and field verifications by SMG staff.
The database also contains information on the characteristics of the facility including facility type, size
and address. In this study, the address information was used for augmenting the database by integrating
with geographic information system-based spatial data, providing locational factors for the facility.
3.2 Sample selection
This study used a subset of the database. First, only four facility types—office, commercial, medical and
educational—were considered, since they are the major sectors of employment involving regular
commutes. The four sectors comprise 63% (=6,024/9,553) of the worksites contained in the database, but
the proportion of employment in the sectors is expected to be even larger. (Unfortunately, the
employment size could not be obtained due to data unavailability.) Second, in order to avoid making
invalid conclusions from a small sample size (usually incurring low statistical significance), TDM
programs in which only a small number of facilities (i.e., < 30) participated were screened out of the
sample. As a result, five programs—odd and even number vehicle plate rationing programs, reduced
parking spaces, transit subsidies, flexible work schedules, and carpooling—were not considered for
analysis. The resultant final data set was composed of 5,995 worksites including 5,083 which did not
report their TDM activities to receive traffic impact fee discounts. This study utilizes about 63%
(=5,995/9,553) of the total number of worksites in the database. Figure 1 illustrates the location of the
facilities considered in this study. Note that a number of sampled facilities were identified as being
concentrated in small areas, raising concern about oversampling from those areas. However, there was
little evidence that such oversampling was systematic, rendering an effective elimination of co-location
effects impractical. Future studies are suggested to carefully handle and investigate this co-location effect
since it can distort modeling output under certain conditions.
9
Figure 1. Location of sampled worksites
3.3 Sample characteristics
The characteristics of the selected worksites and their TDM program participation are summarized in
Table 1. The table indicates that the proportion of worksites which participate in at least one TDM
program is 15.2% (=912/5,995). The average number of TDM programs that a worksite chooses to
implement is 1.6 (=1,467/912), suggesting that many worksites implemented multiple TDM programs.
The most common TDM program was found to be the elimination of free parking (a participation rate of
47.1% for the worksites which implemented at least one TDM program), followed by the weekly no-
driving day program (participation rate of 30.6%). Thus, 71.3% of the worksites which submitted a TDM
program participation report stated that they implemented the elimination of free parking and/or the
weekly no-driving day program. With regard to owner characteristics, 71% (=650/912) of the
participating worksites were found to be owned by private entities while the remaining 29% were in the
public sector, which may be expected, since the number of private facilities is much greater than those in
the public sector. It is worthwhile to note that 81% of the public sector facilities in the selected sample
10
implemented at least one TDM program, compared to just 11% for the private sector, showing the active
participation of the public sector. This situation is somewhat expected, as SMG often requires facilities in
the public sector to participate in the TDM programs.
The table also shows that a significant portion (76%) of the participating facilities is office
buildings. Commercial facilities show the highest participation rate of 22% (=107/486), followed by
office buildings at 16%. The participation rates for educational and medical facilities are only 8% and 9%,
respectively. The higher participation rate for commercial facilities may be expected since they are
usually charged much higher traffic impact fees, thus their greater motivation to participate in order to
obtain larger discounts. Given that all other conditions are equal, commercial facilities should pay greater
fees due to a higher weight factor. For example, the weight factor for major commercial facilities is 5.5
times higher than that of office buildings. It is observed that the average traffic impact fee discount rate
varies program-by-program, with employer-provided taxi service having the highest rate of 18.49%. It is
notable that two programs, the transit day campaigns and work site-specific programs developed by the
employer show a significantly lower discount rate of around 3%. The lower rate suggests that these two
programs may not be effective in reducing vehicle trips for some reason. The 912 worksites which
submitted a TDM activity report were able to cut their traffic impact fees by 21.6% on average. In terms
of facility size, worksites with free commuter buses and site-specific TDM measures were found to be
relatively larger, implying that these two programs are likely to be implemented at worksites with a larger
number of employees. Meanwhile, small employers seem to prefer implementing the weekly no-driving
day program.
The locational characteristics of the worksites were identified by a series of geographic
information system-based processes: address matching and buffer analysis. The two processes produced
information on population and employment size (based on the 2010 census data - the latest available) and
on transportation service levels within 500-meter buffers around the facilities. It is of interest that the
participating worksites are located in areas with relatively lower employment density, compared to areas
where the non-participating worksites are sited. (It is suspected that the co-location effect mentioned
earlier may partly be responsible for this, but there is no definite evidence.) Indeed, the average numbers
of employees within the 500-meter buffers are 34,238 and 40,011 for participating and non-participating
worksites, respectively. This situation is rather counter-intuitive as TDM programs are considered to be
more acceptable to worksites located in areas with high employment density (possibly because of more
likelihood of these areas having better transit accessibility). In particular, the TDM measures of the
weekly no-driving day program and transit day campaign appear to have been implemented at worksites
located in areas with more residents than employees; the respective ratios of the number of employees to
11
population are 0.87 and 0.76 for the two programs. Interestingly, the worksites which implemented free
commuter bus operations were found to be those least served by transit, as indicated by the smaller
numbers of stations, bus stops and bus routes served within the buffers. This suggests that commuter bus
operations are preferred as a way to supplement transit services.
12
Table 1. Selected Sample Characteristics
Factors
Facilities participating in at least one TDM program
License plate rationing programs
No employee parking
Free parking
eliminated
Bicycle use
encouraged
Final digit
Last two digits
Weekly no-driving
day program
Number of facilities sampled
69
102
279
86
430
202
Owner
Private
64
59
81
77
380
111
Public
5
43
198
9
50
91
Located in CBD
3
4
6
19
28
2
Facility type
Commercial
2
0
3
56
25
36
Educational
9
11
17
0
20
11
Medical
7
5
11
10
26
10
Office
51
86
248
20
359
145
Average traffic impact fee
discount rate (%)
8.46
15.62
17.31
10.29
16.16
7.72
Average facility floor area (m2)
10,453.60
16,051.98
7,635.38
39,797.80
15,270.03
17,240.57
Average number of residents1
23,927.96
20,891.01
26,014.77
20,207.59
22,841.73
24,891.94
Average number of employees1
35,207.65
35,899.79
22,628.20
54,924.12
37,928.42
24,996.50
Average number of subway
stations1
1.26
1.39
0.89
1.31
1.37
0.90
Average number of bus stops1
7.96
7.80
7.49
9.87
8.03
8.20
Average number of subway
lines1
1.20
1.29
0.84
1.19
1.28
0.87
Average number of bus routes1
18.80
19.67
15.23
26.98
20.76
16.13
Average distance to the nearest
subway station (meters) 2
362.96
394.19
488.09
349.74
336.64
453.86
Average distance to the nearest
bus stop (meters) 2
113.04
129.95
134.73
90.86
106.34
128.20
Average number of intersections1
232.59
193.84
248.16
190.70
245.14
207.28
Average total road length
(kilometers) 3
102.21
96.80
98.94
101.44
106.00
95.80
13
Table 1 (continued)
Factors
Facilities participating in at least one TDM program
Facilities not participating
in any TDM program
Free commuter
buses
Employer-provided
taxi service
Transit day
campaign
Site-specific
programs developed
by the employer
Overall
Number of facilities sampled
70
128
63
38
912
5,083
Owner
Private
42
95
21
6
650
5,021
Public
28
33
42
2
262
62
Located in CBD
3
9
1
5
53
344
Facility type
Commercial
10
44
12
27
107
379
Educational
3
7
4
1
53
585
Medical
9
16
7
1
62
610
Office
48
61
40
9
690
3,509
Average traffic impact fee discount
rate (%)
12.84
18.49
2.96
3.47
21.60
0.00
Average facility floor area (m2)
46,756.13
37,373.77
13,610.20
64,886.71
16,599.12
4,168.73
Average number of residents1
19,135.86
21,047.73
26,913.06
19,364.58
23,375.15
23,350.74
Average number of employees1
34,254.13
41,858.94
20,362.35
54,284.39
34,238.02
40,011.25
Average number of subway
stations1
0.76
1.09
1.05
1.26
1.19
1.00
Average number of bus stops1
7.51
8.01
7.70
10.39
7.99
8.35
Average number of subway lines1
0.71
1.02
0.95
1.16
1.12
0.94
Average number of bus routes1
19.56
21.37
15.13
28.45
19.15
17.95
Average distance to the nearest
subway station (meters) 2
517.30
402.01
404.18
310.92
396.87
433.36
Average distance to the nearest bus
stop (meters) 2
129.50
123.67
121.80
94.89
117.75
105.16
Average number of intersections1
153.66
171.08
232.16
170.84
233.71
227.55
Average total road length
(kilometers) 3
83.68
96.62
96.77
96.01
102.07
107.53
1. Measured within a 500-meter buffer around the facility; 2. Measured based on the Euclidean distance; 3. Measured within a one-kilometer buffer around the facility
14
4. Methodology
4.1 Models for decisions on whether to participate in TDM programs and effectiveness of the
programs
This study develops statistical models to explore worksite TDM program participation characteristics by
utilizing the sample characteristics in Table 1 as explanatory variables. First, a binary logistic regression
model is constructed in order to investigate which factors are associated with the participation decisions
by assigning a value of one for worksites for which the traffic impact fee discount rate is larger than zero
and a value of zero for all others. The motivation of developing this model is that it is necessary to
clearly identify which worksites are not participating in TDM programs. Although this logistic regression
model does not fully utilize the information provided in the data set (by simply converting non-zero
values into ones), the result would guide policy makers to conceive adequate strategies to encourage the
participation of TDM programs under the condition that only a small portion of worksites (15.2% for this
study’s sample) are participating in the programs. In addition, an ordinary least square regression model
is developed using a dependent variable of the summed traffic impact fee discount rate over the individual
TDM programs. To this end, a Tobit model is also considered, as the dependent variable cannot have
negative values, as specified below (Tobin, 1958).
In this equation, wi and ti are observed independent and non-negative dependent variables for worksite i
and ui follows iid N(0, σ2). In addition, θ is the parameter to be estimated. The Tobit model is described
in terms of a latent variable t*, supposing that ti* is observed if ti* > 0 and is not observed if ti* ≤ 0. The
result of these models is expected to reveal the relationship between the magnitude of the combined TDM
impacts and the relevant factors.
4.2 Joint model of TDM program participation and effectiveness
Under the context of this study, a worksite can opt to implement multiple TDM programs with
different degrees of car trip reduction. This situation is modeled using the MDCEV model. The MDCEV
model is an excellent approach which can handle behavioral phenomena characterized by multiple
discrete choices and a joint continuous choice dimension. The MDCEV model can also deal with
diminishing marginal returns or satiation effects which are not reflected in standard discrete choice
15
models (Bhat, 2005; Bhat, 2008; Bhat and Sen, 2006). The functional form of utility for the MDCEV
model is (Bhat, 2008):
,
where U(x) is a quasi-concave, increasing and continuously differentiable function with respect to
consumption (in this study, car trip reduction effects represented by traffic impact fee discount rates)
quantity (Kⅹ1)-vector x (xk ≥ 0) for all k, and ψk, γk and αk are parameters associated with TDM program
k of the K alternative programs. The parameters, ψk, γk and αk represent baseline marginal utility,
translation parameters and satiation parameters, respectively. The translation parameter γk enables corner
solutions and also acts as a linear satiation parameter; the higher the value of γk, the less the satiation
effect in the consumption of alternative k. This study particularly applies the γ-profile approach with no
outside good (i.e., a TDM program chosen to be implemented by all employers) for fitting a model. In
the approach, the satiation parameter αk is set to zero and thus the utility becomes a simple linear system
(Bhat, 2008):
.
The baseline utility is parameterized in the following functional form to introduce the impact of observed
and unobserved alternative attributes of baseline utility:
,
where zk is a set of attributes describing the TDM program k and the worksite and εk represents
unobserved characteristics that impact the baseline utility for program k. From the analyst’s perspective,
the employer is maximizing the random utility subject to a binding linear budget constraint. By assuming
an extreme value distribution for εk and independent relationships between εk and zk (k = 1, 2, …, K), the
MDCEV formulates the probability that an employer will allocate the available budget to multiple
alternatives (for more details, Baht, 2005; Baht, 2008). In this study, the MDCEV model estimation was
conducted using the program developed based on R codes (MARG, 2016).
16
5. Results
5.1 Models for decisions on whether to participate in TDM programs and the effectiveness of the
programs
As shown in Table 2, three regression models were developed by utilizing the factors (except the number
of facilities sampled) in Table 1 as explanatory variables. Some variables, including floor area,
population and employment size, and the number of intersections, were transformed by taking logarithms
for fitting better models. All three models which capture only significant factors at a level of 0.10 appear
to be statistically satisfactory in terms of their explanatory power, as suggested by R2 values, F-statistics
and Chi-square statistics. Overall, the directions and relative magnitudes of the estimated parameters are
found to be consistent across the three models, suggesting any types of the model may be appropriate to
explain the characteristics of the TDM program participations. However, it should be noted that the
magnitude of the parameters in the Tobit model are substantially larger than those in the ordinary
regression model, and the variable of the number of intersections is not significant for the regression
model. The greater parameters in Tobit models can often be observed as a scale factor between zero and
one should be multiplied with the estimated Tobit parameters so as to make the Tobit parameters
comparable to those of ordinary regression models (Wooldridge, 2009). Indeed, the marginal effect of the
log of facility floor area (the estimated parameter = 15.989) at a median value is estimated to be 2.66 by
the Tobit model, which is comparable to 3.238 in the ordinary regression model. Although a detailed
investigation concerning these rather different results is not within the scope of this study, care should be
taken when predicting the effectiveness of the TDM programs using the models.
The models clearly indicate that employers in the private sector are less likely to participate in
TDM programs; in fact the binary logistic regression model suggests that private sector employers are 93%
less likely to implement TDM programs. Regarding the facility type, medical facilities show the least
tendency to participate in the programs. It is conjectured that the high income and rather irregular
commutes of employees in the medical service area might negatively influence program implementation.
The models show that office and commercial facilities are more likely to participate. The size of the
facility is found to have a positive association with TDM participation; supposing that the facility size is
proportional to the number of employees in the facility, the finding suggests that larger employers may
have more options for the deployment of TDM programs, which is consistent with the finding in Zhou et
al. (2012).
The models also show the importance of the locational characteristics of a worksite. Firstly,
population size is found to exert a negative impact on implementing TDM programs. This situation may
17
be expected, since more population means that the area is likely to be a residential area, where the level of
public transit service is usually lower than that of business areas. The lower transit service level can be a
hurdle for actively implementing car trip reduction measures. However, the population size variable is
significant only at a level of 0.10 in the logistic regression model, and other models did not reveal its
significance. Interestingly, the models suggest that the number of employees in the surrounding area is
negatively associated with TDM program implementations. This finding is rather counter-intuitive since
areas with high employment density are less likely to induce commutes by car (Chatman, 2003). A
potential explanation is that it might not be necessary to implement TDM programs for those worksites
located within high employment areas, since employee car dependency is likely already sufficiently low.
The low TDM program participation rate (15.2%) mentioned in the previous section may partly reflect
this situation. However, care should be taken in interpreting this, since employers may have different
preferences for TDM programs depending on their situation. Program-wise analyses to be presented in
the next section are expected to offer better explanations concerning this issue. (The MDCEV model in
the next section shows that the employment size variable is only significant for the weekly no-driving day
program and operation of free commuter buses.)
Accessibility to the subway is also found to play a critical role in the decision concerning TDM
participation; better accessibility (more subway lines and closeness to stations) means more chances of
implementing TDM programs. In contrast, it is found that good accessibility to bus stops negatively
affects participation in a TDM program. This situation is consistent with the fact that the average distance
to the nearest bus stop is 117.8 and 105.2 meters for TDM participating and non-participating worksites,
respectively. It is not clear why worksites closer to bus stops are less likely to participate in the programs.
It is just conjectured that facilities with poor accessibility to transit might actively implement TDM
programs providing transportation services such as free commuter bus operation and employer-provided
taxi service to mitigate the poor accessibility. Thus, good (or poor) accessibility to transit does not
always mean higher (or lower) likelihood of implementing TDM programs, suggesting the importance of
TDM program-wise analyses. (The MDCEV model in the next section reveals that the variable of the
distance to the nearest bus stop is not significant for any TDM programs.) The number of intersections,
which may be closely related to street connectivity, is found to have a positive relationship with the
chance of implementing TDM programs and in increasing their effectiveness. This finding is expected,
since worksites located in areas with more intersections are likely to be more accessible by walking, thus
increasing the likelihood that restrictions on car use will be put in place.
18
Table 2. Binary Logistic Regression, Multiple Regression and Tobit Models for Worksite Participation in TDM programs
Variables
Binary logistic regression
Multiple regression
Tobit regression
B
SE
Exp(B)
B
SE
B
SE
Constants
-2.217
1.392
0.103
3.910 ***
1.375
-62.363 ***
9.590
Facility characteristics
Owned by private entities
-3.701 ***
0.166
0.027
-17.027 ***
0.462
-43.517 ***
1.878
Facility type (reference = office facility)
Educational facility
-0.611 ***
0.181
0.552
-1.225 ***
0.340
-8.892 ***
2.234
Medical facility
-0.787 ***
0.174
0.466
-1.304 ***
0.333
-10.611 ***
2.169
Log of facility floor area (square meters)
1.253 ***
0.047
3.382
3.238 ***
0.104
15.989 ***
0.622
Locational factors
Log of population size within a 500 meter buffer
-0.186 *
0.095
0.842
-
-
-
-
Log of employment size within a 500 meter buffer
-0.628 ***
0.062
0.547
-1.091 ***
0.119
-7.29 ***
0.690
Number of subway lines within a 500 meter buffer
0.183 ***
0.066
1.174
0.442 ***
0.120
3.529 ***
0.840
Distance to the nearest subway station (kilometers)
-0.468 **
0.216
0.608
-
-
-
-
Distance to the nearest bus stop (kilometers)
1.720 ***
0.588
4.363
4.688 ***
1.376
17.457 **
2.664
Log of the number of intersections within a 500
meter buffer
0.289 ***
0.066
1.315
-
-
2.469 ***
7.263
1Nagelkerke R2 = 0.438
2Hosmer and Lemeshow statistic = 18.513 ***
-2 Log likelihood = 3,388.564
Adjusted R2 = 0.328
F-statistics = 418.0 ***
3L(c) = -6,213.559
4L(β) = -5,315.772
-2 [L(c)- L(β)] = 1,795.573
>=
= 26.1
Number of observations (n) = 5,995
1Nagelkerke R2 is one of the goodness-of-fit indices defined by
/( (Field, 2009), where n represents sample size. For L(c) and L(β), see below.
2Hosmer and Lemeshow statistic is a chi-square statistic comparing the observed frequencies with those expected under an estimated model. In general, the smaller Hosmer-
Lemeshow statistic or the larger corresponding p-value indicates that the assumption of the logistic response function is more appropriate for the estimated model.
3 L(c) is log likelihood of the constant-only model.
4 L(β) is log likelihood of the final model.
Note: * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
19
5.2 Joint model of TDM program choice and effectiveness
Table 3 illustrates the estimated MDCEV model for the employers’ TDM program choices and their
individual effectiveness measured by the traffic impact fee discount rates. For the final model, numerous
model specifications were tested and only significant variables at a level of 0.1 were retained. (An
inclusion of all the variables in the MDCEV model failed to properly produce standard errors of the
parameters, potentially due to too many variables and a complex mechanism of the model estimation
procedure. In addition, the MDCEV model with all the variables, regardless of their significance levels,
is likely to be too complex, as the corresponding parameters would need to be estimated for all 9 TDM
programs.) In the model estimation, site-specific TDM programs developed by the employer served as
the baseline alternative. Note that the MDCEV model is based on only 916 worksites which implemented
at least one TDM program while the regression models in the previous section utilized all the worksites,
including the ones which did not participate in any TDM program.
Similar to the finding by the previous regression models, the resultant MDCEV model indicates
that the owner type of facility is critical for deciding which TDM programs to implement and their
effectiveness. Indeed, the facility owner variable was found to be significant for most TDM programs,
suggesting that facilities owned by private entities are less likely to implement TDM programs. As
mentioned previously, this is ascribed to the fact that employers in the public sector are often required to
implement the programs. However, the program-by-program significances of the variable imply that
three programs—license plate rationing according to the final digit, no parking for employees, and
eliminating free parking—may be equally important for both private and public entities, as the variable is
not significant for the three programs. The facility size is another important factor for most TDM
programs except for the license plate rationing program according to the last two digits and the no parking
for employees program. In particular, the model reveals that free commuter buses and employer-provided
taxi service are the most preferred programs by larger employers, as suggested by the positive signs and
magnitudes of the estimated parameters for the variable. Meanwhile, final digit-license plate rationing
and weekly no-driving day programs seem to be preferred by smaller employers. Interestingly, this
situation infers that costly TDM programs such as free commuter buses and employer-provided taxi
service may be viable only for larger employers.
The MDCEV model suggests that educational and medical facilities, which were previously
found to be less likely to participate in the TDM programs, may have different levels of preference for
each program. More specifically, educational facilities appear more likely to participate in license plate
20
rationing programs and employer-provided taxi service, while medical facilities are found to prefer
implementing no parking for employees, employer-provided taxi services and transit day campaigns.
Commercial facilities show a weaker tendency toward the weekly no-driving day program, eliminating
free parking and free commuter buses, compared to office facilities (the reference facility type). These
detailed alternative-by-alternative preference identifications seem to be the important benefit of the
MDCEV model.
Whether the facility is located within a central business district (CBD) is also an important factor,
which was not identified as significant in the previous regression models. The MDCEV model indicates
that facilities located in the CBD are more likely to participate in the weekly no-driving day program (at a
rather lower significance level, 0.10) and no parking for employees. The finding that no parking for
employees, one of the strictest car-use control programs, is preferred by the facilities located in the CBD
is plausible, since employee resistance to such control measures can be minimized only when transit
service is well-provided. The transit service level in the CBD is significantly higher than in other areas in
Seoul (Ko et al., 2011). Meanwhile, encouraging bicycle use is found to be the least preferred program in
the CBD, which may be due to the lack of bicycle-related facilities such as bicycle rental systems, bicycle
lanes and bicycle parking in the CBD. From a policy perspective, this finding suggests that an adequate
level of built environment may be required for implementing various TDM measures.
Accessibility to transit service is also a critical factor for TDM program choices and their
effectiveness, as already shown in the previous regression models. In particular, the number of subway
lines appears to positively influence the likelihood that TDM programs such as license plate rationing and
elimination of free parking are in place. The distance to the nearest subway station is found to be another
significant variable, indicating that programs encouraging use of bicycles and free commuter buses are
more likely to be implemented and effective with an increase in distance. This situation seems to be
counter-intuitive, but suggests that non-car commuters’ inconvenient access to subway stations located
away from worksites can be mitigated by providing or encouraging alternative travel modes. This finding
is notable since it infers that the general belief that TDM effectiveness has a positive relationship with
transit service levels cannot always be applied. The number of intersections also seems to positively
influence the implementation of TDM measures, in particular for the weekly no-driving day program and
eliminating free parking. Considering that more intersections means a higher degree of street
connectivity, which promotes walking, this finding suggests that favorable environments for walking may
be necessary for effective car use reductions.
21
Table 3. MDCEV Model for TDM Program Participation Effectiveness: Baseline Utility
Variables
Coefficient (t-Statistic)
License plate rationing program (final digit)
Log of facility floor area (square meters)
-0.759 (-5.240) ***
Educational facility (reference = office)
1.202 (3.158) ***
Number of subway lines within a 500-meter buffer
0.296 (1.933) *
License plate rationing program (last two digits)
Owned by private entity
-2.471 (-9.871) ***
Number of subway lines within a 500-meter buffer
0.444 (3.917) ***
Educational facility (reference = office)
1.498 (4.188) ***
Weekly no-driving day program
Owned by private entity
-3.073 (-14.329) ***
Log of facility floor area (square meters)
-0.637 (-5.777) ***
Commercial facility (reference = office)
-2.749 (-4.273) ***
Log of employment size within a 500-meter buffer
-0.321 (-3.788) ***
Located in CBD area
0.941 (1.756) *
Log of the number of intersections within a 500-meter buffer
0.272 (2.706) ***
No parking for employees
Medical facility (reference = office)
1.068 (2.808) ***
Located in CBD area
1.384 (3.550) ***
Eliminating free parking
Log of facility floor area (square meters)
-0.290 (-3.203) ***
Commercial facility (reference = office)
-2.331 (-9.977) ***
Number of subway lines within a 500-meter buffer
0.272 (3.192) ***
Log of the number of intersections within a 500-meter buffer
0.384 (4.617) ***
Encouraging bicycle use
Owned by private entity
-2.169 (-9.854) ***
Log of facility floor area (square meters)
-0.476 (-4.721) ***
Log of distance to the nearest subway station (meters)
0.24 (1.957) **
Located in CBD area
-1.526 (-2.049) **
Free commuter buses
Owned by private entity
-2.344 (-5.746) ***
Log of facility floor area (square meters)
1.244 (6.346) ***
Commercial facility (reference = office)
-1.974 (-5.151) ***
Log of employment size within a 500-meter buffer
-0.412 (-2.625) ***
Log of distance to the nearest subway station (meters)
0.789 (3.087) ***
22
Log of the number of intersections within a 500-meter buffer
0.374 (2.376) **
Employer-provided taxi service
Owned by private entity
-1.853 (-6.459) ***
Log of facility floor area (square meters)
0.294 (2.320) **
Educational facility (reference = office)
0.899 (2.027) **
Medical facility
1.169 (2.987) ***
Transit day campaign
Owned by private entity
-3.259 (-7.708) ***
Log of facility floor area (square meters)
-0.538 (-3.540) ***
Medical facility (reference = office)
1.550 (3.346) ***
Note: df = degrees of freedom. Goodness of fit: log likelihood of the base model at convergence (df = 19) = -4,667.30, log
likelihood of the final model at convergence (df = 54) = -4,175.42, likelihood ratio = 983.76,
= 66.6, Number of
observations = 912.
* Significant at 10%; ** Significant at 5%; *** Significant at 1%.
The baseline constants shown in Table 4 provide an indication of the inherent preference for
various TDM programs and marginal utility at zero consumption for the different alternatives. The
resultant baseline constants suggest that the campaign-type programs of the weekly no-driving day and
transit day campaign are most preferred. Conversely, the costly programs of free commuter buses and
employer-provided taxi service have the smallest baseline preference. (Note that employer-provided taxi
service does not have a significantly different baseline utility from that of the baseline alternative, site-
specific programs developed by the employer.) This trend clearly indicates that employers are more
willing to choose TDM programs that demand fewer financial resources to operate.
A higher translation parameter for an alternative indicates that employers are less satiated with
the consumption of that alternative and are therefore more likely to actively participate in the program,
thereby promoting greater effectiveness. Table 4 indicates that a license plate rationing program
according to the last two digits does not have a statistically significant satiation effect, suggesting that this
program is commonly implemented together with other programs and therefore has a higher effectiveness.
A similar interpretation may be applied to the program of eliminating free parking with a higher
translation parameter. Site-specific programs developed by the employer have the greatest satiation effect,
suggesting that these programs are customized for worksites with peculiar characteristics, for which
conventional TDM programs may not be effective in reducing car use.
23
Table 4. MDCEV Model for TDM Program Participation: Baseline Constants and Translation Parameters
TDM Program
Coefficient (t-Statistic)
Baseline constants
License plate rationing program (final digit)
7.159 (5.350) ***
License plate rationing program (last two digits)
2.223 (6.700) ***
Weekly no-driving day program
11.797 (9.599) ***
No parking for employees
0.541 (1.881) *
Eliminating free parking
3.392 (3.145) ***
Encouraging bicycle use
6.616 (5.183) ***
Free commuter buses
-12.052 (-3.925) ***
Employer-provided taxi service
-0.179 (-0.136)
Transit day campaign
7.563 (5.657) ***
Translation parameters
Site-specific programs developed by employer
1.284 (3.052) ***
License plate rationing program (final digit)
28.308 (1.990) **
License plate rationing program (last two digits)
98.668 (1.483)
License plate rationing program (weekly no-driving day program)
23.160 (4.031) ***
No parking for employees
17.181 (2.305) **
Eliminating free parking
48.683 (3.398) ***
Encouraging bicycle use
6.661 (3.913) ***
Free commuter buses
9.154 (2.323) **
Employer-provided taxi service
19.787 (3.445) ***
Transit day campaign
2.740 (2.069) **
* Significant at 10%; ** Significant at 5%; *** Significant at 1%.
6. Conclusions
This study attempts to explore employer TDM program participation characteristics using a TDM
database containing information on the individual participating employer’s TDM program(s) and their
effectiveness, represented by the traffic impact fee discount rates. For this, factors affecting employers’
decisions on whether to participate in TDM programs and the degree of their participation were identified
based on regression models. In addition, an employer’s TDM program choice model was developed
using the MDCEV model approach, as an employer can participate in multiple TDM programs and their
individual effectiveness is represented by continuous values.
24
The developed regression models suggested that employer TDM program participation can be
affected by various factors including employer type, facility size and accessibility to transit. Private and
smaller facilities were found less likely to participate in the TDM programs. Compared to offices and
commercial facilities, educational and medical facilities appeared to be less likely to participate in TDM
programs. Better subway accessibility tended to increase TDM participation and its effectiveness, as
expected. In contrast to our expectations, employment density around worksites (measured within a 500-
meter buffer) was found to negatively affect TDM implementation. Although this situation can be partly
explained by the low TDM participation rate (about 15%), more in-depth analysis may be required for
fuller clarification. The developed regression models revealed another counter-intuitive finding, in that
better accessibility to bus stops has a negative association with TDM implementation. It was conjectured
that facilities with poor accessibility to transit might actively implement TDM programs of providing
transportation services such as free commuter buses and employer-provided taxi services to mitigate the
poor accessibility. Thus, good (or poor) accessibility to transit does not always mean more (or less)
chances of implementing TDM programs, suggesting a need for TDM program-wise analysis.
A TDM program-wise analysis was conducted based on the MDCEV model. Overall, the
developed MDCEV model revealed similar patterns of the role of similar factors; for example, private
facilities showed a weaker tendency to participate in the TDM programs. However, the model indicated
that worksites might have different preferences for implementing TDM programs, depending on worksite
characteristics. High-cost programs (i.e., free commuter buses and employer-provided taxi services) were
found to be preferred by larger employers while low-cost campaign-type programs (e.g., final digit
license plate rationing and weekly no-driving day program) were preferred by smaller employers. The
baseline constants of the MDCEV model also indicated that campaign-type programs were preferred by
most employers while the high-cost programs were least preferred. This suggests that policy makers need
to consider providing financial support for employers to increase their TDM options.
The models suggested that built environment such as transit accessibility and street connectivity
is crucial for improved implementation of TDM programs. In particular, built environment favorable to
transit and walking was found to positively influence the implementation of car use restriction programs.
As Scheepers et al. (2014) have illustrated, this finding proves that built environment is a crucial element
for a shift from cars to alternative modes. Interestingly, worksites with poor accessibility to transit were
found more likely to implement TDM programs of providing or encouraging alternative transportation
options such as commuter buses or bicycles. This implies that worksite characteristics should be taken
into account when choosing TDM programs.
25
This study is meaningful in that it revealed employers’ behavioral response regarding TDM program
choices, which is a significant addition to the current literature as these issues have rarely been
investigated. From a methodological perspective, the application of the MDCEV model to the TDM
study is also unique. In spite of its contributions, this study is expected to be further improved by
considering employee characteristics such as income level and commute distance, which can dictate
preferences for TDM programs (Zhou et al., 2012). Supplementary surveys collecting such information
may enable researchers to capture the impact of more diverse factors. In addition, it is worthwhile to
examine the differing responses of both private and public sector employers in more detail, as the models
developed pointed to this aspect consistently. Approaches of market segmentation (private vs. public)
and/or different modeling frameworks (e.g., nested structure) would be helpful in the investigation.
Applicability of this study is expected to be enhanced by adding policy analyses which can be conducted
by establishing adequate scenarios. Unfortunately, the calculations of marginal effects and elasticity for
variables in the MDCEV model are not as straightforward as ordinary regression models due to the non-
linearity between explanatory and response variables, requiring complex computation processes and/or
simulation techniques (Bhat and Sen, 2006; Imani et al., 2014; Pinjari and Bhat, 2010). Thus, additional
studies may be needed to expand the current study’s applicability for capturing the impact of various
policies. It should be noted that the findings identified can be applied only to the context of this study,
which targets worksites in a large city with a population density of about 17,000 people per square
kilometer. Travel patterns in urban settings are significantly different from those of suburban areas where
ample free parking exists, few or no alternatives to driving are provided, and car use is seldom restricted
(Ko, 2013). Therefore, some degree of caution needs to be exercised when transferring the findings in
this study to other geographical areas. As this study has demonstrated the potential of the TDM database
for identifying employers’ TDM program choices and implementation characteristics, and the database
has been built up on a yearly basis, future studies are encouraged to conduct longitudinal analyses that can
reveal time variations in employer TDM program choice behavior.
Acknowledgements
This research did not receive any specific grant from funding agencies in the public, commercial, or
not-for-profit sectors.
26
References
1. Bhat, C.R. 2005. A multiple discrete-continuous extreme value model: formulation and
application to discretionary time-use decisions. Transportation Research B 39 (8), pp. 679-707.
2. Bhat, C.R. 2008. The multiple discrete-continuous extreme value (MDCEV) model: role of
utility function parameters, identification considerations, and model extensions. Transportation
Research B 42, pp. 274-303.
3. Bhat, C.R., Sen, S. 2006. Household vehicle type holdings and usage: an application of the
multiple discrete-continuous extreme value (MDCEV) model. Transportation Research B 40 (1),
pp. 35-53.
4. Brockman, R., Fox, K.R. 2011. Physical activity by stealth? The potential health benefits of a
workplace transport plan. Public Health 125 (4), pp. 210-216.
5. Cervero, R., & Tsai, Y. (2004). City CarShare in San Francisco, California: second-year travel
demand and car ownership impacts. Transport Research Record: Journal of Transportation
Research Board 1887, pp. 117-127.
6. Chatman, D. 2003. How density and mixed uses at the workplace affect personal commercial
travel and commute mode choice, Transport Research Record: Journal of Transportation
Research Board 1831, pp. 193-201.
7. DeMaio, P. 2009. Bike-sharing: history, impacts, models of provision, and future. Journal of
Public Transportation 12(4), pp. 41-56.
8. Dill, J., Wardell, E. 2007. Factors affecting worksite mode choice: findings from Portland,
Oregon. Transport Research Record: Journal of Transportation Research Board 1994, pp. 51-57.
9. Eluru, N., Pinjari, A., Pendyala, R., Bhat, C.R. 2010. An econometric multi-dimensional choice
model of activity-travel behavior, Transportation Letters 2 (4), pp. 217-230.
10. Eriksson, L., Nordlund, A.M., Garvill, J. 2010. Expected car use reduction in response to
structural travel demand management measures. Transportation Research F 13 (5), pp. 329-342.
11. Field, A. 2009. Discovering statistics using SPSS Third Edition. Sage, London, UK.
12. Hasnine, M. S., Weiss, A., Habib, K. N.. 2016. Development of an employer-based
transportation demand management strategy evaluation tool with an advanced discrete choice
model in its core. Transportation Research Record: Journal of the Transportation Research Board
2542, pp. 65-74.
13. Hendricks, S. 2005. Effectiveness of programs for work site trip Reduction: the influence of
organizational culture. Transportation Research Record: Journal of the Transportation Research
27
Board, 1924, pp. 207-214.
14. Herzog, E., Bricka, S., Audette, L., Rockwell, J. 2006. Do employee commuter benefits reduce
vehicle emissions and fuel consumption? Results of the Fall 2004 Surveys of Best Workplaces
for Commuters. Transport Research Record: Journal of Transportation Research Board 1956, pp.
34-41.
15. Imani, A.F., Ghafghazi, G., Eluru, N., Pinjari, A.R. 2014. A multiple-discrete approach for
examining vehicle type use for daily activity participation decisions. Transportation Letters 6 (1),
pp. 1-13.
16. James, B., Brög, W. 2001. Increasing walking trips through TravelSmart® Individualised
Marketing. World Transport Policy & Practice, 7 (4), pp. 61-66.
17. Ko, J. 2013. Vehicle Trip Generation Rates for Office Buildings under Urban Settings, ITE
Journal 83 (2). pp. 41-45.
18. Ko, J. and Cho, Y. 2009. Voluntary Program to Reduce Car Use: Weekly No-Driving Day in
Seoul, South Korea, Transportation Research Record: Journal of the Transportation Research
Board, No. 2118, pp. 1-7.
19. Ko, J., Park, D., Lee, S. 2011. Transit ridership evaluation using aggregate-level causal analyses
of subway mode shares in Seoul, South Korea. Transport Research Record: Journal of
Transportation Research Board 2219, pp. 59-68.
20. Litman, T. 2003. The online TDM encyclopedia: mobility management information gateway.
Transport Policy 10, pp. 245-249.
21. Loukopoulus, P., Jakobsson, C., Gärling, T., Schneider, C.M., Fujii, S. 2004. Car-user responses
to travel demand management measures: goal setting and choice of adaptation alternatives.
Transportation Research D 9 (4), pp. 263-280.
22. MARG. 2016. R Codes for Travel Modeling [online]. Mobility Analytics Research Group.
Available at: http://www.mobilityanalytics.org/r-codes-for-travel-modeling.html, Accessed on
August 2, 2016.
23. Merom, D., Miller, Y., Lymer, S., Bauman, A. 2005. Effect of Australia's Walk to Work Day
campaign on adults' active commuting and physical activity behavior. American Journal of
Health Promotion 19 (3), pp.159-162.
24. O’Fallon, C. 2010. Bike Now: exploring methods of building sustained participation in cycle
commuting in New Zealand. Road Transport Research 19 (2), pp. 77-89.
28
25. Pinjari, A.R., Bhat, C.R. 2010. A multiple discrete-continuous nested extreme value (MDCNEV)
model: formulation and application to non-worker activity time-use and timing behavior on
weekdays. Transportation Research B 44 (4), pp. 562-583.
26. Scheepers, C., Wendel-Vos, G., Broeder, J., van Kempen, E., Wesemael, P., Schuit, A. 2014.
Shifting from car to active transport: a systematic review of the effectiveness of interventions.
Transportation Research A 70, pp. 264-280.
27. Shiftan, Y., Suhrbier, J. (2002). The analysis of travel and emission impacts of travel demand
management strategies using activity-based models. Transportation 29, pp. 145-168.
28. Shoup, D.C. 1997. Evaluating the effects of cashing out employer-paid parking: eight case
studies. Transport Policy 4 (4), pp. 201-216.
29. Tobin, J. 1958, Estimation of relationships for limited dependent variables. Econometrica 26, pp.
24-36.
30. You, D., Garikapati, V., Pendyala, R., Bhat, C., Dubey, S., Jeon, K., Livshits, V. 2014.
Development of vehicle fleet composition model system for implementation in activity-based
travel model. Transportation Research Record: Journal of the Transportation Research Board
2430, pp. 145-154.
31. Wooldridge, J. 2009. Introductory Econometrics Fourth Edition. South-Western Cengage
Learning, Mason, Ohio, U.S.
32. Zhou, J., Wang, Y., Schweitzer, L. 2012. Jobs/housing balance and employer-based travel
demand management program returns to scale: evidence from Los Angeles. Transport Policy 20,
pp. 22-35.
