Вы находитесь на странице: 1из 10

Transport Policy 73 (2019) 41–50

Contents lists available at ScienceDirect

Transport Policy
journal homepage: www.elsevier.com/locate/tranpol

The effectiveness of parking policies to reduce parking demand pressure and T


car use
Xiang Yana,∗, Jonathan Levinea, Robert Maransb
a
Taubman College of Architecture and Urban Planning, University of Michigan, 2000 Bonisteel Blvd, Ann Arbor, MI, 48109-2069, USA
b
Institute for Social Research, University of Michigan, 426 Thompson Street, 3330 ISR, Ann Arbor, MI, 48106-1248, USA

A R T I C LE I N FO A B S T R A C T

Keywords: Evaluating the effectiveness of parking policies to relieve parking demand pressure in central areas and to reduce
Parking policy car use requires an investigation of traveler responses to different parking attributes, including the money and
Parking price elasticity time costs associated with parking. Existing parking studies on this topic are inadequate in two ways. First, few
Time cost of parking studies have modeled parking choice and mode choice simultaneously, thus ignoring the interaction between
Car use
these two choice realms. Second, existing studies of travel choice behavior have largely focused on the money
Joint parking and mode choice model
cost of parking while giving less attention to non-price-related variables such as parking search time and egress
time from parking lot to destination. To address these issues, this paper calibrates a joint model of travel mode
and parking location choice, using revealed-preference survey data on commuters to the University of Michigan,
Ann Arbor, a large university campus. Key policy variables examined include parking cost, parking search time,
and egress time. A comparison of elasticity estimates suggested that travelers were very sensitive to changes in
egress time, even more so than parking cost, but they were less sensitive to changes in search time. Travelers
responded to parking policies primarily by shifting parking locations rather than switching travel mode. Finally,
our policy simulation results imply some synergistic effects between policy measures; that is, when pricing and
policy measures that reduce search and egress time are combined, they shape parking demand more than the
sum of their individual effects if implemented in isolation.

1. Introduction likely to lead to biased model results and misleading policy implica-
tions. For example, in cases where increased parking cost mainly trig-
As a major component of travel-demand-management strategies, gers parking relocation instead of modal switch, a mode-choice study
parking policies have been widely used across cities to pursue two may find the parking price elasticity (with respect to car use) to be quite
important goals—relieving parking demand pressure in central areas low and thus may wrongly conclude that parking price schemes are
and reducing car use. The effectiveness of any parking measure in only moderately effective. Moreover, parking strategies can imply
achieving these goals depends on the extent to which it can influence policy trade-offs, as when efforts to enhance access between satellite
individuals' travel decisions, e.g. their parking destination and travel- parking and employment centers relieve parking pressure in central
mode choices. To support policy in this regard, empirical evidence on areas but encourage car use; in this case, a joint modeling of parking
individuals’ behavioral responsiveness to parking strategies needs im- and travel mode choices can help evaluate the nature and magnitude of
provement. these tradeoffs and thus better inform the desirability of each strategy.
This notion is supported by two observations. First, travelers may Second, while numerous empirical studies have shown that the cost
respond to parking policy in various ways, such as change of parking of parking significantly affects individuals’ travel choices, there seems
behavior, change of travel mode, change of trip destination, change of to be a dearth of information on traveler responses to non-price-related
trip schedule, and change or cancel of trip activities, but most parking parking attributes, such as the time costs of parking. After summarizing
studies have only considered one aspect of these choice behaviors, ra- empirical evidence on traveler responses to parking costs, Marsden
ther than tradeoffs among them. Accounting for traveler responses (2006, p.449) concluded that “less evidence is available on observed
beyond one choice realm is necessary because failing to do so would responses to excess-time, particularly the time taken between parking


Corresponding author. Postal address: Taubman College of Architecture and Urban Planning, University of Michigan, 2000 Bonisteel Blvd, Ann Arbor, MI, 48109,
USA.
E-mail addresses: jacobyan@umich.edu (X. Yan), jnthnlvn@umich.edu (J. Levine), marans@umich.edu (R. Marans).

https://doi.org/10.1016/j.tranpol.2018.10.009
Received 30 May 2018; Received in revised form 5 September 2018; Accepted 30 October 2018
Available online 30 October 2018
0967-070X/ © 2018 Elsevier Ltd. All rights reserved.
X. Yan et al. Transport Policy 73 (2019) 41–50

vehicle and the final destination for commute trips.” Madsen et al. 2.2. Parking location choice studies
(2013) and Inci (2015) noted that most parking studies estimated the
price elasticity of demand for parking without accounting for cruising Besides switching travel mode, travelers may also respond to
for parking (i.e. search time), which would lead to a biased price parking-policy measures by changing their parking locational decisions.
elasticity estimate. Beyond the threat of omitted variable bias, ex- For example, travelers may park at a more remote location when faced
amining the impacts of non-price-related parking attributes on travel with increased parking charges. The change in parking locations is also
behavior matters for at least two reasons. First, where political will to a desirable policy goal if it helps relieve demand pressure for parking in
raise parking prices is lacking, supply constraints and the parking-time activity centers and if it reduces drivers’ behavior of cruising for
costs they generate may amount to a second-best policy. Second, where parking (Shoup, 2006; Van Ommeren et al., 2012). Some researchers
policy seeks to reduce car use and/or to relieve parking pressure and have thus calibrated parking location choice models to examine the
strategies that alter the time costs of parking are considered as alter- influences of different parking attributes on parking decisions. Various
natives to pricing schemes, there is a need to understand their relative parking variables, including parking price, walk time between parking
effectiveness. and destination, search time for parking spots, and parking time limits
To explore these issues, this paper builds a joint model, based on were shown to significantly affect parking choices (Gillen, 1978; Van
revealed-preference survey data, of travel-mode and parking-location der Goot, 1982; Axhausen and Polak, 1991; Tsamboulas, 2001; Golias
choice for commuters to the University of Michigan, Ann Arbor campus, et al., 2002; Chaniotakis and Pel, 2015). Regarding the time-cost vari-
which hosts over 70,000 students, faculty, and staff in an area of nearly ables associated with parking, Axhausen and Polak (1991) suggested
13 square miles. The model features three major parking attribute- that drivers value walking time to the destination more than search
s—parking cost, search time for parking, and egress time from parking time for a parking space, which is in turn valued more than other in-
to destination. We found a parking price elasticity higher than those vehicle travel time.
reported by many previous studies. The results also showed that tra- Although parking-choice studies (more than mode-choice studies)
velers were more sensitive to changes in egress time than changes in have often incorporated parking variables beyond parking cost, the
parking cost, and their sensitivity to changes in search time was lower focus of policy analysis remains on parking-charge-related measures.
than the other two variables. In general, parking policies would lead to That is, researchers rarely apply their model results to evaluate the
more incidents of parking relocation than modal switches. Finally, the relative effectiveness of non-pricing strategies in influencing travel
policy simulation results imply some synergistic effects between decisions.1 Accordingly, while dozens of parking management and
parking pricing and parking measures that alter the time costs of supply strategies have been proposed in practice to tackle parking-re-
parking, i.e. their combined effect is larger than the sum of individual lated problems, the academic literature has generated little satisfactory
effects. information regarding their impacts on parking and travel choice be-
havior (Kuzmyak et al., 2003; Litman, 2018). Up until now, parking
pricing schemes appear to be the most popular parking policy, and they
2. Literature review are usually regarded as the most effective policy measure (Higgins,
1992; Golias et al., 2002); however, several studies have found that
2.1. Parking and auto use travelers are more responsive to parking availability and walk time than
to parking cost, particularly in more central locations such as the cen-
Parking-related studies started to emerge around the 1970s as a part tral business district (Gillen, 1978; Chaniotakis and Pel, 2015).
of the booming individual travel-behavior literature (McGillivray,
1970; Gillen, 1977). A major research interest of parking studies has
2.3. The interaction between parking choice and travel-mode choice
been the impact of the monetary cost of parking on travel demand and
travel behavior, especially the problems associated with free and un-
To facilitate the comparison of study results and allow transfer-
derpriced parking (Shoup, 2005). The common finding was that tra-
ability of research findings from one research setting to other contexts,
velers would significantly reduce their car use if parking were more
researchers often summarize demand response to parking attributes in
expensive. For example, Willson (1992) predicted a reduction of 25–34
the form of elasticities. An elasticity measure indicates the percentage
percent car use if free parking were not offered to employees working in
changes in the demand variable (e.g. parking volumes at a specific site)
downtown Los Angeles. Willson and Shoup (1990) summarized em-
in response to a one percent change in the attribute of interest (e.g.
pirical evidence on how employer-paid parking affects employees’
parking price). While a wide range of elasticity estimates—generally
travel choices and concluded that 19 to 81 percent fewer employees
between −0.1 and −0.6 but as high as −1.2 (Albert and Mahalel,
would drive to work alone if free parking at the workplace were
2006)—has been reported in the literature, the most cited value for the
eliminated. Other studies generally reported similar findings, except
price elasticity of parking demand (number of cars parking) was −0.3
that the degree of influence may differ (e.g. Hess, 2001; Tsamboulas,
(Vaca and Kuzmyak, 2005). The parking price elasticity of car use was
2001).
less clearly estimated, since modal responses are highly sensitive to
While researchers generally recognize that time-cost variables as-
local conditions such as the availability and convenience of competing
sociated with parking such as travel time between parking and desti-
travel modes, especially public transit. In general, researchers believe
nation (commonly referred to as egress time) and search time for a
that a shift in parking location is a more probable behavioral response
parking spot can also affect the decision to drive, few mode choice
to parking interventions than a shift in travel mode (Shiftan, 2002;
studies have incorporated these variables; see discussions in review
Marsden, 2006), a finding that would presumably hold more for se-
articles by Feeney (1989), Young et al. (1991), Marsden (2006) and Inci
lective or localized parking-price increases than in pricing-policy shifts
(2015). The failure to account for these time-cost variables represents a
that affect broader swaths of territory.
potential misspecification in mode-choice modeling. For example, the
While early parking studies were mostly aggregate in nature, ex-
parameter estimate of parking cost may have an upward bias since
amining overall travel and parking demand responses to parking pri-
parking cost and search time are often positively correlated. Practically,
cing changes, more recent work has focused on investigating individual
it also means less evidence is available on traveler responses to time
costs of parking, and so policymakers have little knowledge on the ef-
fectiveness of parking measures that affect total travel time in reducing 1
One of such exceptions is Simićević et al. (2013), where the authors con-
car use. cluded that parking prices would affect car usage while parking time limit
would affect the type of parking used (on-street versus off-street).

42
X. Yan et al. Transport Policy 73 (2019) 41–50

choice behavior. Researchers generally consider disaggregate studies to


be more methodologically satisfactory than aggregate studies (Feeney,
1989). However, most choice-behavior studies have modeled travel
mode and parking choices separately, an approach that may lead to
specification errors and biased results. Mode-choice studies often found
a lower-than-expectation parking price elasticity of car use, since higher
parking fees trigger not only modal shift but also parking relocation
(Gillen, 1977). Westin and Gillen (1978) provided an early finding on
this idea by modeling parking cost as an endogenous variable in a mode
choice model. Similarly, parking choice studies could underestimate the
effects of parking policies on travel behavior if their impacts on modal
responses are not accounted for. In order to have a more accurate
evaluation of parking policies, therefore, parking studies should ac-
count for parking and mode choice simultaneously.
Our study contributes to the literature by accounting for both
money and time costs of parking in a joint choice model of parking
location and travel mode, which was calibrated on commuting-trip data
collected on a large university campus. Choice models that account for
the interaction between parking location choice and mode choice have
rarely been applied in the existing literature, with a limited few ex-
ceptions such as Westin and Gillen (1978), Hensher and King (2001),
and Simićević et al. (2013). However, the Westin and Gillen study
specified a sophisticated econometric model which is difficult to re-
plicate. The other two papers relied on stated-preference (SP) data,
often extracted from hypothetical choice experiments, instead of re-
vealed-preference (RP) data which reflect actual market behavior. Fig. 1. Location of university parking lots by permit type required.
While SP data have been widely used in transportation research, re-
searchers generally consider that the results obtained from SP data are
less reliable than those from RP data (Hensher et al., 2005). alternatives available; that is, individuals are more likely to choose to
drive if parking is cheap and convenient. Consequently, when faced
3. Modeling framework with parking policy measures (on a specific parking location), motorists
may respond by not only parking at a different destination but also
Our research approach employs the discrete choice modeling tech- switching to a different travel mode. For certain individuals, however,
nique, which is commonly used to examine consumer choice, in which driving may be the only viable travel option and so they can only react
individuals are assumed to maximize their utility stemming from the to parking changes by changing parking locations.
alternative chosen. To account for the unobserved factors influencing
individuals’ choice decisions, economists have specified discrete choice 4. The data
models based on random utility theory (McFadden, 1981; Ben-Akiva
and Lerman, 1985). The utility of alternative i for individual n Uin is The data used to fit the NL model in this study are commuting-trip
assumed to be made up of a systematic component Vin and a random data collected from faculty and staff at the University of Michigan, Ann
component εin . The former is a function of observed attributes, and the Arbor. This campus provides a natural experiment to examine the im-
latter is an unobserved error term. The most commonly used discrete pact of parking charges and constraints on mode choice. The university
choice model is the multinomial logit model (MNL), in which the error issues several types of parking permits to faculty and staff, including
term εin is assumed to be identically and independently distributed Gold, Blue, Yellow, and Orange, each of which has an annual cost of
across alternatives (IIA) and across observations following a type I ex- $1809, $736, $161, and $80 respectively in 2018 (prices were slightly
treme distribution. While MNL has widespread applications, it is often lower in survey years). Parking spaces that are associated with cheaper
inadequate when some of its basic assumptions are violated, among permits are usually either more constrained or located at more distant
which the IIA assumption violation is most common. sites from university buildings (see Fig. 1). Therefore, faculty and staff
A variety of model extensions have been developed to address this who desire to drive can choose to buy either a cheaper permit at the
issue, and the nested logit (NL) model is the simplest and most widely expense of parking at more constrained and distant space or vice versa.
used (Koppelman and Wen, 1998). The NL model is characterized by Or, when neither alternative is desirable, they may choose non-driving
grouping subsets of alternatives that are similar to each into different travel modes.
nests. This modeling structure assumes that decision makers would be The commuting-trip data came from the University of Michigan
more likely to switch alternatives within a nest instead of across nests. Sustainability Cultural Indicators Program (SCIP) 2012–2015.
The NL model is particularly appealing when the choice decision (Callewaert and Marans, 2017). The SCIP is an annual online survey
modeled can be conceptualized as a hierarchical choice process (e.g. administrated to a cross-sectional sample of faculty, staff, and students
choosing a type of neighborhood and a house) or a sequential one (e.g. (students were excluded from the analysis in this study because they
choosing a job location first and then a residence location). have very limited parking options), and it yields responses from more
This study conceives of travelers as simultaneously deciding on a than 1000 faculty and staff members every year. The original purpose
travel mode and parking destination. When choosing a parking desti- of the SCIP surveys was to assess sustainability knowledge, behaviors,
nation, individuals face tradeoffs among parking cost, search time for a and attitudes—on a broad range of topics such as travel and transpor-
parking space, and travel time from parking to the destination. Decision tation, food, waste prevention and climate change—across university
makers also need to evaluate the desirability of travel modes based on populations over time. The survey asked participants to report their
their respective attributes (e.g. money and time cost, comfort, safety) usual travel mode for commuting trips, parking-permit ownership,
and attribute levels. The two choice decisions interact because the de- work address (department buildings), and home address (nearest in-
sirability of the driving mode is affected by the desirability of parking tersection to their residence). Other information extracted from the

43
X. Yan et al. Transport Policy 73 (2019) 41–50

SCIP dataset includes respondents’ demographic and socioeconomic 4.2. Measuring parking-related variables
characteristics, how important they valued walkability, bikeability, and
transit availability when they decided where to live, and their en- Besides the access time from parking lots to workplaces estimated
vironmental consciousness (how much they are concerned about the above, other parking-related variables include parking cost (permit
impact of auto travel on the environment). cost) and search time for parking. Since this study models mode choice
Although the survey asked participants to choose a usual commute in a one-way commute trip, parking cost was calculated by dividing
travel mode from a set of eight options (drive a car, walk, bike, ride the annual permit cost by 500 (assuming there 250 workdays and each one-
bus, ride the bus and bike, rideshare, park and ride, and other), we have way trip shares half of the parking cost). Therefore, the parking costs
aggregated them into four composite alternatives—drive a car, ride the for driving with Gold, Blue, Yellow, and Orange permits are $3.62,
bus, walk, and bike—for analysis. The driving mode was then inter- $1.47, $0.32, $0.16, respectively.
acted with the permit type variable to generate four driving and permit Following a procedure described below, we estimated the search
type combinations. After a data cleaning process, a total of 2861 cases time for finding a parking spot from data on parking lots’ capacity and
were included in subsequent analysis. Among these respondents, shares vacancy rates. These data were obtained from the University of
for each alternative to be modeled are: drive with a Gold Michigan Logistics, Transportation & Parking, which sends out staff
permit—14.1%, drive with a Blue permit—42.4%, drive with a Yellow each year to record the number of vacant spaces in each parking lot,
permit—13.4%, drive with an Orange permit—6.4%, walk—8.4%, every morning and afternoon on five consecutive weekdays. We ag-
bike—6.4%, and ride the bus—8.8%. gregated these data into a single indicator of parking constraint for each
parking lot, which is the average vacancy rate in the mornings.3 While
to our knowledge there have been no publications that estimate search
4.1. Measuring level of service attributes for each travel mode time for parking space from vacancy rates of parking structures/sur-
faces, empirical work on on-street parking shows that search time only
The most essential variables for a mode choice model are the level- starts to rise above zero minutes when vacancy rate is below 15%, and
of-service attributes for each travel mode. To obtain these measures, we it rises exponentially when the vacancy rate approaches zero (Shoup,
first geocoded the home and work address information of respondents 2005; Levy et al., 2013; Inci et al., 2017). Apart from vacancy rate,
into geographic coordinates. Then, we applied the Google Maps search time is also likely to be affected by parking lot capacity, espe-
Distance Matrix API to obtain home-to-work travel time estimates for cially when the vacancy rate is zero, since individuals are likely to
three travel modes: walking, biking, and transit (driving alternatives spend more time searching in larger lots.
will be discussed separately in the next paragraph). Compared to a We converted parking capacity and vacancy rates into search time
traditional network-analysis procedure, this technique is much more following a conceptual approach as follows. Let us denote search time
efficient and can potentially increase the accuracy of travel time esti- as t, parking lot capacity as C, vacancy rate as V, expected number of
mates (Wang and Xu, 2011).2 There is only a single estimate for the parking space examined until a vacant space is located as N, vehicle
walking and biking mode at any time. For the transit mode, however, travel speed when searching for parking as v (set as eight miles per
the outputs depend on the specifications of departure/arrival time, hour), and width of a parking space as l (set as 10 feet). For on-street
which accounts for transit schedules, and transit-routing preferences, parking, researchers often assume that vacancy parking spaces are
which includes two options—less walking and fewer transfers. Different randomly distributed (see Inci et al., 2017). Therefore, the expected
specifications could lead to somewhat different travel time estimates. number of spaces examined N = 1/V, and so search time t = N × l/v =
To account for such variability, the transit travel time estimates used in l/(V × v). However, in parking structures, parking spaces are usually
this study are the average of eight estimates resulting from the com- filled up sequentially and so drivers may have examined all occupied
binations of four arrival-time points on a weekday (8:00 a.m., 9:00 space before locating a vacancy one. It thus follows that the expected
a.m., 10:00 a.m., 10:30 a.m.) and two transit-routing-preference op- number of spaces examined N = (1-V) × C, and so search time t = N ×
tions (less walking and fewer transfers). In addition, since wait time for l/v = (1-V) × C × l/v. For small lots, the two approaches would result
the first leg of the transit trip is not accounted for by the Google Maps in similar search time estimates; for large lots, however, the first ap-
API (wait time for transfer stops is included in the estimate), we have proach is likely to produce a search time estimate that is much smaller
added 10 min of wait time to all transit travel time estimates. than the second approach. In this study, we took the middle ground and
For each driving-plus-permit alternative, we assumed that in- specified search time t = 0.5 × l/(V × v)+0.5 × (1-V) × C × l/v.
dividuals will park at the nearest parking lot that accepts this type of As discussed in Inci et al. (2017), this specification will be proble-
permit. Then, again using the Google Maps API, we obtained travel time matic when a parking lot reaches full occupancy. The expected number
estimates from home to parking lots for these alternatives. Similar to of parking spaces examined will be infinite for formula N = 1/V, which
travel time estimates for transit, these estimates are subject to specifi- in turn produces an infinitely large search-time estimate. To address
cations of the traffic model and departure/arrival time. We selected the this issue, a constraint was imposed on search time. Adopting the ap-
“best guess” option for the traffic model. Also, to account for travel- proach of Inci et al. (2017), we set the expected number of parking
time differences in peak hours versus nonpeak hours, we took the spaces examined in a parking lot as five times of its capacity when it
average of travel-time estimates in four arrival-time points (8:00 a.m., reaches full capacity. That is to say, the vacancy rate of any parking lot
9:00 a.m., 10:00 a.m., 10:30 a.m.) on a typical weekday. Finally, we was constrained to be no less than 1/(5 × C).
obtained the travel time estimates from parking lots to work destina- A final important step of data processing is to delete alternatives
tions by walking and by taking the bus. After adding 5 min of waiting to within each individual's choice set that are either unavailable or un-
the transit mode, the smaller value of these two is taken as the egress- reasonable. In the current study, we removed the alternatives from an
time estimate from a parking lot to someone's workplace. individual's choice set under the following conditions:

1) For walking and biking, estimated travel time is larger than 90 min;
2 2) For transit, Google Maps returned an output of “not applicable,” or
One limitation of Google Maps API is that it only generates travel time es-
timates based on the most updated road network and transit schedules. Our
data are from 2012 to 2015 and we obtain the home-to-work travel time esti-
3
mates in 2017. Since the years are pretty close and the transport systems in our To reduce measurement error, some small parking lots (i.e. capacity is less
study area did not experience significant changes during those gap years, we than 50) were grouped with their nearby lots to calculate an “areawide”
consider these travel time estimates to be reliable. average vacancy rate.

44
X. Yan et al. Transport Policy 73 (2019) 41–50

Table 1 Table 2
A descriptive profile of the level of service variables. Nested logit model outputs.
Attribute Alternative Unit Total Sample Sample choosing this Alternatives coefficients t-stat
alternative
Constants
Mean Standard Mean Standard Walk Walk 1.191 8.38
deviation deviation Transit Transit 4.586 6.95
DriveGold DriveGold 9.996 15.09
Travel time1 DriveGold Minute 17.57 13.24 17.85 11.00 DriveBlue DriveBlue 9.951 15.16
DriveBlue 17.99 13.47 20.75 14.38 DriveYellow DriveYellow 9.305 15.00
DriveYellow 16.90 12.98 24.35 14.49 DriveOrange DriveOrange 9.180 14.89
DriveOrange 17.15 12.70 22.86 13.98
Walk 34.71 14.48 21.58 11.08 LOS variables
Bike 20.3 13.35 13.07 8.85 Travel time All alternatives −0.048 −8.88
Transit 49.62 19.55 44.66 13.65 Search time Driving alternatives −0.042 −6.15
Search time DriveGold Minute 0.53 0.72 1.20 0.71 Egress time Driving alternatives −0.102 −6.90
DriveBlue 7.71 7.65 4.99 5.69 Travel cost by household income Driving alternatives 0.126 1.60
DriveYellow 2.73 1.58 2.33 1.52 Parking cost by household income Driving alternatives −0.148 −6.75
DriveOrange 3.61 1.95 3.93 1.66
Egress time DriveGold Minute 4.09 3.17 4.90 2.68 Socioeconomic and demographic variables
DriveBlue 4.09 3.18 3.12 2.22 Household income Driving alternatives 0.117 2.10
DriveYellow 9.44 3.58 5.85 2.84 Vehicle per capita Driving alternatives 0.822 4.47
DriveOrange 11.21 2.81 10.02 3.41 Male Walk, Bike 0.410 2.67
Travel cost2 DriveGold Dollar 2.37 3.11 2.12 2.43
DriveBlue 2.41 3.14 3.00 3.62 Behavioral preference variables
DriveYellow 2.37 2.95 4.01 3.63 Bike- and walk-ability Walk, Bike 1.815 21.98
DriveOrange 2.39 2.93 3.58 3.41 Transit availability Transit 1.364 19.72
Parking cost DriveGold Dollar 3.62 0.00 3.62 0.00 Environmental consciousness Walk, Bike, Transit 0.306 2.55
DriveBlue 1.47 0.00 1.47 0.00
DriveYellow 0.32 0.00 0.32 0.00 IV parameters (RU2)
DriveOrange 0.16 0.00 0.16 0.00 Driving alternatives 0.376 6.76
Transit 1.000
Notes: 1. The travel time for non-driving alternatives is the one-way home-to- Walk, Bike 0.594 5.83
destination total trip time, and the travel time for driving alternatives excludes
search time and egress time. 2. Since the University provides free transit pass to Sample size 2861
Log likelihood (null model) −4289.95
all faculty and staff, the travel costs for Walk, Bike, and Transit are all equal to
Log likelihood −2912.53
zero. The travel costs for driving alternative are the product of the travel dis-
Adjusted McFadden psuedo R2 0.316
tance between an individual's home and the parking lots and the variable cost of Chi-squared vs. MNL model 95.290
driving, which is assumed to be 30 cents per mile. The fixed cost of car own- Rejection significance 0.000
ership is not accounted for here.

the distance between home and the nearest bus stop is larger than
one mile, or the travel time estimate is the same as walking.
3) For driving alternatives, the given individual owns no personal ve-
hicle. Also, a given permit type option is not available if the esti-
mated egress time is larger than 20 min or if the sum of search and
egress time is larger than that for a cheaper permit type.

After this procedure, among the 2861 individuals in our sample,


driving with Gold, Blue, Yellow, or Orange permit, walking, biking, and
Fig. 2. Two-level nest structure for the NL model.
riding the bus are available in 2784, 2808, 2117, 1886, 1245, 2260, and
1966 choice sets respectively; and the number of individuals choosing
each alternative is 438, 1078, 384, 168, 287, 222, and 284 respectively. non-motorized alternatives (i.e. walking and biking) were 0.372 and
The mean and standard deviation of the level-of-service attributes in- 0.604 respectively, both within the theoretical range of 0–1. A logsum
fluencing travel choices are shown in Table 1. parameter value closer to zero indicates increased substitution among
alternatives in that nest, and so there is more substitution among
driving alternatives than that between walking and biking.
5. Empirical analysis Building on prior research on travel behavior, we included five
groups of independent variables into the joint mode and parking choice
5.1. Model specification model, including alternative-specific constants (ASCs), level of service
(LOS) attributes associated with each alternative, socioeconomic and
Table 2 presents the results of the two-level nested model, estimated demographic attributes associated with each individual, built-environ-
by normalizing the level-2 scale parameter (i.e. RU2). The adjusted ment factors, and behavioral-preference-related variables. The meaning
McFadden's pseudo R-square for this model was 0.317, which indicates of most variables should be self-explanatory and thus no detailed de-
satisfactory model fit. While not presented here, an MNL was specified finitions are presented. We computed the variance inflation factor (VIF)
to be compared with the NL model, but it resulted in much worse model value for all independent variables and found that all the variables have
fit, and a Hausman test showed that the independence of irrelevant a VIF value of smaller than 5, which indicates little multicollinearity in
alternative assumption was violated. The nesting structure of the NL the data. In specifying the model, a major decision involves de-
model is shown in Fig. 2. We tested several other groupings of alter- termining how each independent variable will enter the systematic
natives—for example, combining all non-driving alternatives—but they utility functions of alternatives. Once this decision is made, one should
resulted in worse model fit. The logsum parameters (often referred to as decide if a variable will have the same or different coefficients for each
inclusive value parameters) for the driving alternative alternatives and

45
X. Yan et al. Transport Policy 73 (2019) 41–50

utility function it enters. We constrained all coefficients to be the same indicator, instead of that on the travel-cost indicator, was used here
across alternatives, which means that the influence of each attribute on because the latter was less reliable. Assuming that each individual earns
the utility of each alternative is assumed to be the same. half of his/her annual household income and that the work hours for a
We tested a variety of specifications before deciding on the final year is 2000 h, we found that individuals’ values of travel time, search
form. The money-cost variables (travel and parking costs) were divided time, and egress time were 0.39, 0.34, and 0.83 times, respectively, of
by household income in order to account for the fact that lower-income their wage rate. In other words, for someone with an annual income of
individuals tend to be sensitive to changes in them. The second column $50,000 (wage rate is $25 per hour), our model showed that his/her
of Table 2 lists the alternatives that each independent variable is as- values of travel time, search time, and egress time were $9.73 per hour,
sociated with. Travel time and travel cost were specified as generic $8.51 per hour, and $20.68 per hour respectively.
variables across all alternatives. We attempted to model in-vehicle time Individuals' socioeconomic status and demographic characteristics
(IVT) and out-of-vehicle time (OVT) separately but obtained un- were found to correlate with their travel choices. Our model showed
satisfactory results (parameter estimate on IVT was insignificant and that commuters who had a higher household income and better access
positive). Parking-related variables, including search time, egress time to personal vehicles were more likely to drive to work. In addition,
(from parking lot to workplace), and parking cost, only entered the males were more inclined to use non-motorized modes than females.
utility functions of the driving alternatives. They are the key variables While not shown here, we also tested the association between age and
of interest in this study. travel-choice probabilities but found no statistically significant re-
lationships. Our study also showed that individuals’ residential pre-
5.2. Coefficient estimates ferences may shape their travel behavior. Those who preferred to live in
walkable/bikeable communities were more likely to travel on foot or
In general, the coefficient estimates carried expected signs, and their with their bicycles, and those who valued transit access at home were
magnitudes were in reasonable range. We set the biking mode as the more prone to take the bus. Finally, environmental consciousness, as
reference alternative (i.e. the ASC for biking was set to zero), and so the measured by how much people are concerned about auto and airplane
ASCs were relative in magnitude to biking. All ASCs were positive and travel, was negatively associated with car use.
significant at a 95% of confidence level (except that the ASC for transit
is insignificant), which suggests that everything being equal, in- 5.3. Elasticities
dividuals in our survey prefer other travel modes over biking.
Consistent with theoretical prediction, coefficients on all LOS var- To assess the sensitivity of outcome variables in response to changes
iables—except travel cost (by household income)—were significant and in each independent variable, it is common practice to convert the
negative. While parking cost (by household income) was found to ne- coefficients obtained from discrete-choice models into elasticities. An
gatively affect the choice probability of driving alternatives, the coef- elasticity measure shows the percentage changes in the choice prob-
ficient on travel cost was positive but not statistically significant. A ability of an alternative in response to a one percent change in a vari-
primary reason for this unexpected positive sign may be that our model able. The measure is termed direct elasticity when evaluating how an
did not consider carpooling as an alternative. Statistically, to obtain a alternative's probability responds to a one percent change in its own
negative coefficient on travel cost requires the choice probability of attribute, and is termed cross elasticity when evaluating responses to a
certain alternatives to go down when they have become more expensive one percent change in other alternatives' attribute. Given that the pri-
than their competing alternatives. In the context of our data, we would mary focus of this study is to examine the impacts of parking policies,
need to observe the choice probability of driving declining as in- we calculated elasticities with respect to following variables: parking
dividuals’ travel distance from home to work increases, because all non- cost, search time, and egress time. Results are given in Table 3. Note
driving alternatives had a travel cost of zero. Yet the opposite was that the elasticities were obtained using the probability-weighted
observed because driving became the only viable travel mode as travel sample enumeration technique rather than by averaging individual
distance increased. Introducing carpooling, a cheaper travel alternative elasticities directly (Louviere et al., 2000).
(compared to driving alone) whose choice probability is likely to in- Note: Numbers in bold fond are direct elasticities and other numbers
crease as travel distance increases, into the model may reverse the sign, are cross-elasticities. The elasticities are read by row, for example, the
but our data did not allow it. first row for parking price (i.e. price change in Blue permit) shows that
Comparing the magnitudes of time variable coefficients generates a 1% increase in parking cost will lead to a 0.61% reduction in the
several important insights. First of all, we found that individuals value sample share choosing to drive with a Blue permit, a 0.61% increase in
egress time more than travel time. Parking studies have consistently choosing to drive with a Gold permit, …and a 0.16% increase in
reported that drivers have a strong inclination to park at places very choosing to take the bus.
close to their destinations (Inci, 2015), and so it is reasonable for this The parking price elasticity rises with parking charges and ranged
preference to partially translate into the valuation of egress time. The from −0.21 to −1.89, and the choice probability-weighted average
value of egress time to travel time ratio was 2.13, which is close to the was −0.78. Overall, these estimates are larger in magnitude than the
values reported by early studies (Axhausen and Polak, 1991; Young typical range of −0.1 to −0.6 reported in early parking studies (Vaca
et al., 1991). Second, somewhat unexpectedly, our model implies that and Kuzmyak, 2005) but are close to those reported by Hensher and
the value of search time is less than that of travel time, resulting in a King (2001). Both the Hensher and King paper and our study calibrated
value of time ratio of 0.88. This ratio is smaller than 1.38, the mean a joint mode and parking choice model, and most early parking studies
value of search time to IVT ratio in the ten studies reviewed by only modeled parking choices. Therefore, early parking studies may
Wardman (2001). Besides a simple measurement error explanation, i.e. have underestimated demand elasticity with respect to parking charges
search time in our study was not observed but estimated from data on due to the failure to account for modal responses. Another important
parking lot vacancy and capacity, it may also because our study com- reason for the larger price elasticity estimates obtained here is that our
bined IVT and OVT into a single travel time variable. Since OVT is in study examines parking and mode choice behavior of long-term com-
general valued more than both IVT and search time (Wardman, 2001; muters, while other studies often include other types of travelers such
Abrantes and Wardman, 2011), it is reasonable for the value of search as short-term visitors. Studies have shown that individuals who park
time to be smaller than the value of travel time. more frequently and at a longer duration at a place tend to be more
Based on the coefficient estimates on the cost and time variables, we sensitive to price changes (Kelly and Clinch, 2006; Brooke et al., 2014).
calculated the implied willingness-to-pay for time reductions among the Furthermore, existing parking studies often model parking locations at
study population. The coefficient estimate on the parking-cost a zonal level, such as groups of blocks (Van der Goot, 1982) and inside

46
X. Yan et al. Transport Policy 73 (2019) 41–50

Table 3
Direct and cross elasticities.
DriveGold DriveBlue DriveYellow DriveOrange Walk Bike Bus

Parking cost
DriveGold −1.89 0.57 0.30 0.29 0.06 0.08 0.07
DriveBlue 0.61 −0.61 0.50 0.62 0.08 0.13 0.16
DriveYellow 0.04 0.02 −0.26 0.24 0.01 0.01 0.02
DriveOrange 0.00 0.01 0.05 −0.21 0.00 0.00 0.01
Search time
DriveGold −0.05 0.01 0.01 0.01 0.00 0.00 0.00
DriveBlue 0.41 −0.26 0.01 0.14 0.03 0.04 0.05
DriveYellow 0.03 0.02 −0.15 0.10 0.00 0.01 0.01
DriveOrange 0.02 0.02 0.06 −0.30 0.00 0.00 0.01
Egress time
DriveGold −0.61 0.16 0.15 0.16 0.02 0.03 0.02
DriveBlue 0.39 −0.33 0.23 0.29 0.04 0.06 0.06
DriveYellow 0.19 0.17 −1.09 0.64 0.03 0.04 0.06
DriveOrange 0.12 0.12 0.38 −2.07 0.02 0.03 0.05

or outside of a cordon zone (Azari et al., 2013), which can also lead to distant parking lots and destinations are likely to significantly boost the
underestimated price elasticity because these models are unable to chances of these lots being utilized. Indeed, studies often report that
capture within-zone parking location change. Finally, while the daily convenient transit/shuttle services between park-and-ride facilities and
parking permit cost is low compared to urban downtown parking rates, popular destinations are essential gradients for the success of park-and-
the purchase of a university parking permit is a lump-sum payment in ride schemes (e.g. Bos et al., 2004). Finally, the cross-elasticities of non-
hundreds of dollars; individuals tend to be more sensitive to larger driving modes with respect to time variables were small, which in-
expenses. dicates that non-pricing parking measures would primarily influence
The cross elasticities with respect to parking price changes for the choice of parking destination rather than travel mode.
driving alternatives were significantly larger than those for non-driving
modes. For example, a 1 percent increase in the cost of a Blue permit 6. Policy simulations
would be expected to yield a 0.62 percent increase in the use of range
lots, but only a 0.16 percent increase in the use of the bus. This suggests Finally, we applied the model results to simulate market shares of
that motorists respond to parking charges by primarily changing their each alternative under different parking policy schemes in order to
parking choices instead of switching to an alternative travel mode. To evaluate their respective effectiveness in achieving different policy
achieve the goal of reducing car use, increasing the price of Blue per- goals. The effectiveness of different policy measures depends on not
mits (the most popular and second-most expensive type) would be more only the sensitivity of traveler responses to each policy variable (i.e.
effective than charging more for other permits. In fact, these results demand elasticity with respect to a variable), but also the practical
suggest that charging more for the cheaper permits (i.e. Yellow and amount of change that each policy can effect on these variables. For
Orange) would create very little incentive for drivers to give up their example, although parking demand elasticity with respect to egress
personal vehicles. Two plausible explanations may account for these time may be much larger than with respect to parking cost, pricing
low cross-elasticity values. First, the parking permits are relatively schemes (e.g. doubling the parking cost) are often more effective than
cheap compared to the convenience and travel-time savings offered by measures that reduce egress time (e.g. providing more frequently
personal vehicles against competing modes. University employees tend shuttle services between parking lots and destinations) because there is
to have a higher value of time than the general public, and driving is less room for improvements on egress time.
mostly preferred. Second, in the study area, households with higher The most important goals of contemporary parking policies are two-
income tend to live at places closer to campus where there is decent fold—relieving parking demand in central locations and reducing car
transit service, whereas lower-income households tend to live further use. We thus considered three types of policy measures to achieve these
away from where the only viable travel option is driving. Therefore, if goals—higher pricing in central areas (i.e. more expensive Blue and
additional parking charges were introduced, higher-income individuals Gold permits), increasing capacity of popular distant parking lots (in
will keep driving; on the other hand, the lower-income individuals who order to reduce search time in Yellow and Orange lots), and enhancing
are sensitive to parking pricing can only drive to a cheaper parking lot transit connections between parking lots and destinations (to reduce
instead of shifting to transit. This reasoning shows that parking pricing egress time from Yellow and Orange lots to destinations). The first
schemes would have major equity implications (Chatman and Manville, policy effects changes on the driving with Blue and Gold permit alter-
2018). natives, and the other two influence driving with Yellow and Blue
The elasticity estimates with respect to search time were in general permit alternatives. Implementing the first two policies is straightfor-
much smaller than those with respect to parking costs, which suggests ward. As for the third policy, the university may consider running on-
that travelers were much more sensitive to parking price than search demand ridesourcing shuttles instead of traditional buses in order to
time. Those who spent more time on searching (mostly Blue permit maximize operating efficiency (Yan et al., 2018). Based on realistic
holders), however, were more responsive to additional search time. On considerations of the local context, we assumed the following for each
the other hand, individuals’ parking choice was found to be profoundly policy:
affected by changes in egress time. The effect rose with distance, that is,
the elasticity estimate was higher for more remote lots (Yellow and 1) Policy 1: Increase the cost of Blue and Gold permits by 50%;
Orange) than nearby lots (Gold and Blue). The direct egress time 2) Policy 2: Increase the capacity of Yellow and Orange lots to reduce
elasticity for drivers who parked at Orange lots was −2.07, which the maximum search time for these lots to 3 min, i.e. those who were
means that a 1% decrease in the egress time from an Orange lot to estimated to spend more than 3 min in searching for a Yellow or
destination would lead to a 2.07% increase in parking volumes at an Orange parking space would only spend 3 min. This amounts to an
Orange lot. This implies that measures to enhance connections between overall 20% reduction in search time for Yellow lots and an overall

47
X. Yan et al. Transport Policy 73 (2019) 41–50

Table 4
Predicted demand change under different policy measures.
Drive Drive Drive Drive Walk Bike Transit Change of parking demand in Change in non-driving
Gold Blue Yellow Orange central areas modes

Current market share 15.31% 37.68% 13.42% 5.87% 10.03% 7.76% 9.93%

predicted market share 15.36% 37.65% 13.41% 5.86% 9.76% 8.03% 9.93%

Market share change


Policy 1: increase price of Blue and Gold −7.50% −2.12% 4.93% 2.52% 0.56% 0.70% 0.91% −9.62% 2.17%
permits by 50%
Policy 2: reduce search time in Yellow −0.19% −0.34% 0.13% 0.48% −0.02% −0.02% −0.04% −0.53% −0.07%
and Orange lots
Policy 3: reduce egress time from Yellow −0.69% −1.67% −0.04% 2.76% −0.08% −0.10% −0.19% −2.36% −0.37%
and Orange lots
Combining policy 1 and 2 −7.64% −2.60% 5.02% 3.14% 0.54% 0.66% 0.86% −10.23% 2.07%
Combining policy 1 and 3 −8.00% −4.50% 4.93% 5.93% 0.44% 0.55% 0.64% −12.50% 0.44%
Combining policy 1, 2, and 3 −8.54% −6.97% 6.59% 7.83% 0.32% 0.38% 0.39% −15.51% 1.09%

34% reduction in search time for Orange lots. and more convenient satellite parking create for car use. Moreover,
3) Policy 3: Enhance transit connections between Yellow and Orange there appear to be significant synergies between policies. When in-
parking lots and destinations to improve the egress time for these troduced together, these parking policies would produce a greater effect
alternatives as follows. If the estimated egress time by walking was on reducing the demand pressure in Gold and Blue lots than the sum of
no more than 5 min, individuals were assumed to walk and so the their individual effects. Nonetheless, this synergistic effect would un-
egress time would be equal to walk time; otherwise, travelers were dermine the incentive that higher pricing creates for non-driving al-
assumed to take the transit instead and their egress time would be ternatives, understandably so because these measures have jointly
3 min of wait time plus the in-vehicle travel time. Based on the as- made driving to Yellow and Orange lots much more attractive. If re-
sumption that ridesourcing shuttles instead of traditional buses ducing car use is prioritized over parking relocation as a policy goal,
would be operated to serve these connections, we set the in-vehicle policymakers may consider measures that more directly affect mode
travel time to be the average of current car travel time and current choice, such as expanding carpooling incentives and enhancing public-
transit travel time. This amounts to an overall 13% reduction in transit services. However, these alternatives may face political chal-
egress time for Yellow lots and an overall 20% reduction in egress lenges as they amount to diverting parking revenue to uses that neither
time for Orange lots. improve nor connect to parking facilities (Rye and Ison, 2005).

Table 4 presents the current market share, model-predicted market


share, and market-share changes under different policy measures for 7. Conclusion
the alternatives modeled in this paper. The last two columns present the
change of parking volume in central areas (i.e. the sum of changes in This paper developed a joint model of parking location and travel
driving with Blue and Gold permits) and the change in non-driving mode choice based on faculty and staff commuters to the University of
modes (i.e. the sum of changes in walk, bike and transit), which are the Michigan, Ann Arbor. Key policy variables studied here include parking
indicators of the major policy goals. A negative sign indicates a desir- cost, parking search time (cruising time), and egress time (the time
able change for the former and the opposite is true for the latter. required to reach one's destination from a parking spot). The parking
These results generate several important insights. First, they show price elasticity estimated in this study ranged from −0.21 to −1.89
that while parking pricing can relieve parking pressure and reduce car with a choice-probability-weighted average of −0.78, which is larger
use simultaneously, policies that reduce search time and egress time in than the value of −0.3 often reported in parking studies. Two major
satellite lots imply tradeoffs between the two goals: greater attractive- sources of discrepancy are the failure to account for modal responses in
ness of the remote lots can attract parkers from central areas, but it can early parking models and this study's focus on long-term commuting
also encourage car use. Second, for all three policies, the magnitude of trips, which are more sensitive to parking price changes than less-fre-
change in parking volume in central areas was much larger than that in quent and shorter-duration trips. Of the time-cost variables, we found
the use of non-driving modes, suggesting that the primary effect of that travelers were more sensitive to egress time than search time. With
parking policies is triggering parking relocation instead of modal the advent of intelligent parking services, which provide parking
switch. Third, even though policy 2 assumed a higher percentage availability information to travelers through internet and smartphone
change than policy 3 in their corresponding parking attributes, the apps, parking search time is likely to exert even less of an influence on
number of changes in parking and modal choice caused by policy 2 was travelers' parking and mode choice decisions (Cao and Menendez,
much more modest than that by policy 3. That is to say, enhancing the 2018). The parking demand elasticity with respect to egress time esti-
connectivity between parking facilities and destinations would be more mated here was even larger than that to parking price, which suggests
effective than expanding the capacity of remote parking lots in that there may be great potential to encourage remote parking by im-
achieving policy goals; and so policymakers should prioritize parking proving transit and shuttle connections with final destinations. The
management strategies over increasing supply. advent of on-demand transportation ordered via smart devices—and
We also tested the effects of several policy combinations, and the potentially self-driving vehicles—can expand policymakers' options in
results are shown in the last three rows of Table 4. As expected, com- improving this kind of access.
bining pricing with measures that enhance the attractiveness of cheaper Our study shows that travelers tend to respond to parking policies
and more distant parking facilities would lead to a greater impact on by relocating to a different parking lot instead of switching to an al-
parking demand reduction in central areas. In addition, the net effect on ternative travel mode. This finding holds despite the fact that the uni-
travelers’ use of non-driving modes was positive, which indicates that versity provides free transit passes to all faculty and staff. The in-
higher parking fees can effectively counter the incentives that easier effectiveness of parking strategies to promote transit use results from
two reasons. First, in the United States where both gas and parking

48
X. Yan et al. Transport Policy 73 (2019) 41–50

prices are low (Shoup, 2005), driving is still much more attractive than suitable to inform parking policies that are intended to affect long-term
taking the bus given the travel-time savings, convenience, and flex- and long-duration parking, such as workplace and residential parking
ibility that it provides. Second, in the study area, higher-income policies. The results are less transferable to retail parking and down-
households tend to live close to campus, where better transit services town parking that attracts travelers with various trip purposes, various
are provided, whereas lower-income households tend to live further parking frequencies and duration, and less familiarity with local
away where driving is the often only viable travel option. Therefore, parking conditions. More broadly, our study results have several im-
lower-income households who are more likely to take advantage of the portant policy implications that are transferable to a wide range of
free transit pass end up not being able to do so but can only relocate to contexts. First, pricing remains the single most effective parking policy
more remote parking lots when parking charges are increased. Our to alter travel behavior. This is not only because individuals are highly
findings thus provide important lessons for cities that are considering sensitive to parking cost, but also because other parking attributes have
the promotion of non-driving travel modes by implementing parking- much less room for change compared to parking price. Second, the
related strategies. importance of egress time implies that strategies that alter egress time
The parking environment of the University of Michigan differs from can be effective policy tools when the political will for increased
those of cities in major respects, notably in the parking charges that parking charges is lacking. Third, when parking experiences localized
commuters face. These differences present both a challenge and an overcrowding despite adequate region-wide supply, parking demand
opportunity. The challenge is that the model results are not directly management strategies can be more effective (while in general more
transferable. Parameter estimates will surely vary between campus and desirable) than increasing parking supply. The fact that enhancing ac-
city—just as they would vary between cities with their variations in cess between remote parking lots and workplace has a greater impact
regionwide parking conditions, parking enforcement, availability and than increasing the capacity of crowded remote parking lots supports
convenience of alternative travel mode, and local population char- this notion. Finally, considering the significant synergies between pri-
acteristics (Galbraith and Hensher, 1982). On the other hand, while cing and policies that reduce search and egress time, parking strategies
urban parkers often see themselves as tied to their specific employer's are best implemented as a policy bundle. A “carrot and stick” im-
site for parking, the integrated campus parking system described here plementation may also result in greater acceptance of the parking
invites users to use lots at different convenience and price points policy by local constituents.
throughout the 13-square mile campus area, giving transportation re-
searchers a nearly ideal opportunity to study parking choice behavior Declarations of interest
and its price and non-price determinants. In addition, the campus offers
a useful laboratory for modeling parking and travel-mode choices si- None.
multaneously because cycling, walking, and transit use are realistic
options for many commuters. Therefore, research presented here can Acknowledgements
help identify relevant dimensions of parking and travel-mode decisions
and their interplay, which can advance parking-behavior research and This study is a part of the "Reinventing Transportation and Urban
policies it supports in other locales. Mobility" project, funded by the Michigan Institute for Data Science. We
Results of this study are based on commuting trips of faculty and thank Lisa Solomon at the University of Michigan Logistics,
staff to a university campus. The findings presented here are most Transportation, and Parking for her staff support for this project.

Appendix. Robustness checks

A number of assumptions were made in order to estimate the parking search time in this study, which means that the search time variable may be
subject to large measurement errors. To check if our results are robust to alternative measurements of this variable, we re-fit the nested logit after
applying different sets of assumptions. Selected model outputs (coefficient estimates on level-of-service variables only) and the direct-elasticity
estimates on parking-related variables were presented in Table 5. In each column, only one assumption is different from the ones described in the
main text. Instead of specifying search time t = 0.5 × l/(V × v) + 0.5 × (1-V) × C × l/v, Column (1) and (2) specify t = l/(V × v) and t = (1-V)
× C × l/v respectively. Column (3) and (4) sets the cruising speed to be 5 mile/hour and 10 mile/hour respectively. Column (5) and (6) constraint
the vacancy rate of any parking lot to be no less than 1/(2 × C) and 1/(10 × C) respectively. The model outputs and elasticity estimates are close in
all six columns and so our results are robust to alternative measures of the parking search time variable.

Table 5
Selected model outputs and elasticity estimates.

(1) (2) (3) (4) (5) (6)

coefficient (t-stat) coefficient (t-stat) coefficient (t-stat) coefficient (t-stat) coefficient (t-stat) coefficient (t-stat)

Travel time −0.051** (−7.11) −0.056** (−10.12) −0.051** (−8.73) −0.049** (−9.25) −0.046** (−8.79) −0.051** (−9.39)
Search time −0.037** (−3.90) −0.024** (−3.91) −0.023** (−6.37) −0.042** (−5.95) −0.041** (−5.56) −0.030** (−6.04)
Egress time −0.052** (−3.72) −0.077** (−6.47) −0.098** (−6.89) −0.101** (−6.89) −0.100** (−6.70) −0.103** (−6.80)
Travel cost 0.103 (1.14) 0.066 (1.33) 0.163* (2.23) 0.081 (1.24) 0.131* (2.10) 0.075 (1.13)
Parking cost −0.142** (−4.11) −0.093** (−6.07) −0.164** (−6.89) −0.137** (−6.62) −0.134** (−6.46) −0.149** (−6.63)

Alternative Elasticity Elasticity Elasticity Elasticity Elasticity Elasticity

Parking cost DriveGold −1.67 −1.93 −1.84 −1.94 −1.95 −1.92


DriveBlue −0.57 −0.47 −0.65 −0.59 −0.58 −0.62
DriveYellow −0.34 −0.23 −0.27 −0.26 −0.26 −0.26
DriveOrange −0.19 −0.19 −0.21 −0.21 −0.21 −0.21
(continued on next page)

49
X. Yan et al. Transport Policy 73 (2019) 41–50

Table 5 (continued)

(1) (2) (3) (4) (5) (6)

coefficient (t-stat) coefficient (t-stat) coefficient (t-stat) coefficient (t-stat) coefficient (t-stat) coefficient (t-stat)

Search time DriveGold −0.04 −0.05 −0.05 −0.04 −0.05 −0.03


DriveBlue −0.29 −0.10 −0.28 −0.23 −0.23 −0.21
DriveYellow −0.26 −0.16 −0.16 −0.14 −0.15 −0.11
DriveOrange −0.36 −0.41 −0.32 −0.27 −0.32 −0.22
Egress time DriveGold −0.31 −0.75 −0.52 −0.69 −0.65 −0.62
DriveBlue −0.20 −0.32 −0.30 −0.34 −0.32 −0.34
DriveYellow −1.17 −1.19 −0.99 −1.14 −1.17 −1.10
DriveOrange −1.57 −2.22 −1.82 −2.22 −2.20 −2.09

References Koppelman, F.S., Wen, C.H., 1998. Alternative nested logit models: structure, properties
and estimation. Transp. Res. Part B Methodol. 32 (5), 289–298.
Kuzmyak, R.J., Weinberger, R., Pratt, R.H., Levinson, H., 2003. Parks. Transportation
Abrantes, P.A., Wardman, M.R., 2011. Meta-analysis of UK values of travel time: an Research Board of the National Academies, Washington, DC (Chapter 18).
update. Transport. Res. Pol. Pract. 45 (1), 1–17. Levy, N., Martens, K., Benenson, I., 2013. Exploring cruising using agent-based and
Albert, G., Mahalel, D., 2006. Congestion tolls and parking fees: a comparison of the analytical models of parking. Transportmetrica: Transport. Sci. 9 (9), 773–797.
potential effect on travel behavior. Transport Pol. 13 (6), 496–502. Litman, T., 2018. Parking Management Best Practices. Routledge.
Axhausen, K.W., Polak, J.W., 1991. Choice of parking: stated preference approach. Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods: Analysis and
Transportation 18 (1), 59–81. Applications. Cambridge university press.
Azari, K.A., Arintono, S., Hamid, H., Rahmat, R.A.O., 2013. Modelling demand under Madsen, E., Mulalic, I., Pilegaard, N., 2013. A Model for Estimation of the Demand for On-
parking and cordon pricing policy. Transport Pol. 25, 1–9. street Parking.
Ben-Akiva, M.E., Lerman, S.R., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Marsden, G., 2006. The evidence base for parking policies—a review. Transport Pol. 13
Application to Travel Demand, vol. 9 MIT press. (6), 447–457.
Bos, I.D., Van der Heijden, R.E., Molin, E.J., Timmermans, H.J., 2004. The choice of park McGillivray, R.G., 1970. Demand and choice models of modal split. J. Transport Econ.
and ride facilities: an analysis using a context-dependent hierarchical choice ex- Pol. 192–207.
periment. Environ. Plann. 36 (9), 1673–1686. McFadden, D., 1981. Econometric models of probabilistic choice. In: Manski, C.,
Brooke, S.L., Ison, S.G., Quddus, M.A., 2014. Parking choice. In: In: Ison, S.G., Mulley, C. McFadden, D. (Eds.), Structural Analysis of Discrete Data with Econometric
(Eds.), Parking: Issues and Policies, Transport and Sustainability, vol. 5 Emerald Applications. MIT Press, Cambridge, MA, pp. 198–272.
Group Publishing, Bingley, UK. Rye, T., Ison, S., 2005. Overcoming barriers to the implementation of car parking charges
Callewaert, J., Marans, R.W., 2017. Measuring progress over time: the sustainability at UK workplaces. Transport Pol. 12 (1), 57–64.
cultural indicators program at the University of Michigan. In: Handbook of Theory Shiftan, Y., 2002. The effects of parking pricing and supply on travel patterns to a major
and Practice of Sustainable Development in Higher Education. Springer International business district. In: Travel Behaviour. Edward Elgar Publishing, pp. 37–52.
Publishing, pp. 173–187. Shoup, D.C., 2005. The High Cost of Free Parking. American Planning Association,
Cao, J., Menendez, M., 2018. Quantification of potential cruising time savings through Chicago.
intelligent parking services. Transport. Res. Pol. Pract. 116, 151–165. Shoup, D.C., 2006. Cruising for parking. Transport Pol. 13 (6), 479–486.
Chaniotakis, E., Pel, A.J., 2015. Drivers' parking location choice under uncertain parking Simićević, J., Vukanović, S., Milosavljević, N., 2013. The effect of parking charges and
availability and search times: a stated preference experiment. Transport. Res. Pol. time limit to car usage and parking behaviour. Transport Pol. 30, 125–131.
Pract. 82, 228–239. Tsamboulas, D.A., 2001. Parking fare thresholds: a policy tool. Transport Pol. 8 (2),
Chatman, D.G., Manville, M., 2018. Equity in congestion-priced parking: a study of 115–124.
SFpark, 2011 to 2013. J. Transport Econ. Pol. (JTEP) 52 (3), 239–266. Vaca, E., Kuzmyak, J.R., 2005. TCRP Report 95: Traveler Response to Transportation
Feeney, B.P., 1989. A review of the impact of parking policy measures on travel demand. System Changes: Chapter 13: Parking Prices and Fees. Transportation Research Board
Transport. Plann. Technol. 13 (4), 229–244. of the National Academies, Washington, DC.
Galbraith, R.A., Hensher, D.A., 1982. Intra-metropolitan transferability of mode choice Van der Goot, D., 1982. A model to describe the choice of parking places. Transport. Res.
models. J. Transport Econ. Pol. 7–29. Gen. 16 (2), 109–115.
Gillen, D.W., 1977. Estimation and specification of the effects of parking costs on urban Van Ommeren, J.N., Wentink, D., Rietveld, P., 2012. Empirical evidence on cruising for
transport mode choice. J. Urban Econ. 4 (2), 186–199. parking. Transport. Res. Pol. Pract. 46 (1), 123–130.
Gillen, D.W., 1978. Parking policy, parking location decisions and the distribution of Wang, F., Xu, Y., 2011. Estimating O–D travel time matrix by Google Maps API: im-
congestion. Transportation 7 (1), 69–85. plementation, advantages, and implications. Spatial Sci. 17 (4), 199–209.
Golias, J., Yannis, G., Harvatis, M., 2002. Off-street parking choice sensitivity. Transport. Wardman, M., 2001. A review of British evidence on time and service quality valuations.
Plann. Technol. 25 (4), 333–348. Transport. Res. E Logist. Transport. Rev. 37 (2–3), 107–128.
Hensher, D.A., King, J., 2001. Parking demand and responsiveness to supply, pricing and Westin, R.B., Gillen, D.W., 1978. Parking location and transit demand: a case study of
location in the Sydney central business district. Transport. Res. Pol. Pract. 35 (3), endogenous attributes in disaggregate mode choice models. J. Econom. 8 (1),
177–196. 75–101.
Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice Analysis: a Primer. Willson, R.W., Shoup, D.C., 1990. Parking subsidies and travel choices: assessing the
Cambridge University Press. evidence. Transportation 17 (2), 141–157.
Hess, D., 2001. Effect of free parking on commuter mode choice: evidence from travel Willson, R.W., 1992. Estimating the travel and parking demand effects of employer-paid
diary data. Transport. Res. Rec.: J. Transport. Res. Board (1753), 35–42. parking. Reg. Sci. Urban Econ. 22 (1), 133–145.
Higgins, T.J., 1992. Parking taxes: effectiveness, legality and implementation, some Yan, X., Levine, J., Zhao, X., 2018. Integrating ridesourcing services with public transit:
general considerations. Transportation 19 (3), 221–230. an evaluation of traveler responses combining revealed and stated preference data.
Inci, E., 2015. A review of the economics of parking. Econ. Transport. 4 (1–2), 50–63. Transport. Res. C Emerg. Technol. https://doi.org/10.1016/j.trc.2018.07.029. in
Inci, E., Ommeren, J.V., Kobus, M., 2017. The external cruising costs of parking. J. Econ. press.
Geogr. 17 (6), 1301–1323. Young, W., Thompson, R.G., Taylor, M.A., 1991. A review of urban car parking models.
Kelly, J.A., Clinch, J.P., 2006. Influence of varied parking tariffs on parking occupancy Transport Rev. 11 (1), 63–84.
levels by trip purpose. Transport Pol. 13 (6), 487–495.

50

Вам также может понравиться