741997

research-article2017
JTRXXX10.1177/0047287517741997Journal of Travel ResearchYang et al.

Empirical Research Article

Journal of Travel Research

Public Transport Connectivity and

2019, Vol. 58(1) 25­–41
© The Author(s) 2017
Article reuse guidelines:
Intercity Tourist Flows sagepub.com/journals-permissions
DOI: 10.1177/0047287517741997
https://doi.org/10.1177/0047287517741997
journals.sagepub.com/home/jtr

Yang Yang1, Dong Li2, and Xiang (Robert) Li1

Abstract
In this study, we investigate how dyadic air and rail transport connectivity affects domestic tourist flows among 343 Chinese
cities. Using geo-tagged Sina Weibo data to track tourists during China’s National Day Golden Week in 2014, we estimate
several gravity models with a negative binomial distribution. The estimation results suggest that air transport connectivity
generally has a greater influence than rail transport on dyadic tourist flows, while connectivity provided by ordinary trains
(compared to other rail types) is most important in the context of rail transport. Also, we find the effects of transport
connectivity and intermodal transport competition to depend on the origin-to-destination distance. Different types of railway
trains appear to have distinct effective distance ranges: the effect of high-speed rail trains is strongest at travel distances
between 1,800 and 2,000 km, whereas bullet trains’ effect is strongest at distances between 400 and 600 km.

Keywords
Sina Weibo, gravity model, air transport, transport connectivity, high-speed rail, domestic tourist flows

Introduction in 2014 (Qin 2016). Likewise, government-initiated deregu-

lation efforts and airport construction plans in many tier II
In a typical tourist flow system, transport infrastructure plays and tier III cities helped China become the second-largest air
a vital role in connecting origin and destination locations transport market in the world in 2005 (Zhang et al. 2014).
while ensuring safe, comfortable, and efficient transportation The country’s railways have also benefited greatly from con-
for tourists. When evaluating their travel experiences, tour- tinued infrastructural development, including upgrades to
ists consistently cite transport as having great bearing on trip existing networks and the construction of a nationwide high-
satisfaction (Pritchard and Havitz 2006). Efficient transport speed rail (HSR) network (Qin 2016). The Chinese Mid- to
connectivity can significantly reduce travel time and costs Long-Term Railway Network Plan (2016–2025) includes
for tourists (Peng et al. 2015) while improving destination proposals for the construction of eight HSR corridors with
access for those in origin markets (McKercher 1998). four north–south and four east–west HSR trunk lines (Cao
Additionally, a destination’s transport infrastructure, particu- et al. 2013), which will facilitate travel throughout the coun-
larly its level of sophistication, largely dictates the destina- try (D. Wang et al. 2014).
tion’s tourism demand (Khadaroo and Seetanah 2007b, Much of the relevant literature has used destination-
2008) and overall competitiveness compared to similar centric transport provision measures to examine the effects
regions (Dwyer and Kim 2003; Duval and Schiff 2011). of transport factors on tourism demand (Khadaroo and
When a destination is accessible via different modes of trans- Seetanah 2007b, 2008; Y. Yang and Wong 2012). However,
port (e.g., airplane, train, car, and boat), tourists can choose the role of transport in connecting origin cities has been
the option that best suits their budget, length of stay, and largely overlooked. Although many scholars have studied
travel distance (Thrane 2015). intermodal transport competition and investigated the
Unfortunately, inadequate infrastructure plagues destina- competition/substitution pattern across different types of
tions in emerging travel markets (X. Li 2016; Sheth 2011).
Poor transport infrastructure carries significant financial
1
implications for destinations and hinders their tourism indus- Department of Tourism and Hospitality Management, Temple University,
Philadelphia, PA, USA
try growth (Khadaroo and Seetanah 2007b). Many emerg- 2
Beijing Tsinghua Tongheng Urban Planning & Design Institute, Beijing,
ing-market countries (e.g., China) consider improved China
transport infrastructure a national priority. Over the last
Corresponding Author:
decade, the quantity and quality of transport options in China Yang Yang, Department of Tourism and Hospitality Management,
have grown significantly. Thanks to the National Trunk Temple University, 1810 N. 13th Street, Speakman Hall 111, Philadelphia,
Highway Development Program, China’s national express- PA 19122, USA.
way coverage increased from 400 km in 1990 to 111,900 km Email: [email protected]
26 Journal of Travel Research 58(1)

transport (Jiménez and Betancor 2012; Dobruszkes, a local tourism industry’s overall competitiveness.
Dehon, and Givoni 2014), empirical exploration of the McKercher (1998) investigated how tourists’ destination
effect of transport-mode competition on tourist flows is choices are informed by market access, which refers to the
lacking. To fill this gap, we apply a gravity model to assess competitive (dis)advantages of a destination vis-à-vis its
the relationship between intercity tourist flows and air and proximity to major markets. He found that transport largely
rail transport connectivity. determines market access; that is, greater market access
This study aims to estimate (1) the effect of public trans- encourages sustained tourist flows to a destination and cre-
port connectivity on intercity tourist flows and (2) the effect ates opportunities to attract visitors by leveraging pass-
of intermodal transport competition on intercity tourist through traffic (McKercher 1998). In Prideaux’s (2000)
flows. We also intend to investigate how these effects vary proposed resort development model, transport access to a
with distance. We expect to contribute to current knowledge resort is considered a key factor in determining its rate and
of tourism demand and tourism transport in four ways. First, scale of growth because transport affects the resort’s ability
our study represents a pioneering effort to examine the effect to expand to new markets. Prideaux (2000) also developed
of transport network structure on tourist flows: rather than a transport cost model to examine the role of transport in
relying on destination-side transport infrastructure measures, facilitating destination growth and shaping destination
we focus on transport connectivity between origins and des- market portfolios. The model delineates transport access
tinations to better understand how dyadic linkages shape costs into three categories: actual travel fares, time value
tourist flows. Second, we incorporate the intermodal trans- for the journey, and cost of comfort.
port competition effect into our gravity model, which yields Transport has been found to influence tourist flows in a
critical insights into substitution patterns across different variety of destinations. For example, Khadaroo and Seetanah
types of transport. Third, as the first of its kind, our study (2007b) used destination-side transport capital stocks as a
represents an initial attempt to evaluate how the effect of proxy for transport modes (air, land, and sea) to evaluate
transport connectivity varies at different travel distances. infrastructure’s impact on international tourist flows to
Lastly, we use social media data to track aggregate tourist Mauritius. The effect was estimated to be significant for
mobility during China’s National Day Golden Week, thereby tourists from Europe, America, and Asia, in part because
highlighting geo-tagged social media data as an alternative they prefer to maintain a baseline level of transport-related
source of large-scale tourist flow data. comfort when away from home. Khadaroo and Seetanah
This article is organized as follows. In the next section, (2008) investigated how bilateral tourist flows are deter-
we review relevant literature, namely, studies on the effects mined by destination-side transport infrastructure, including
of air and rail transport on tourist flows and intermodal trans- road, airport, and port measures, using a gravity model. They
port competition. Based on previous research, we then spec- found that transport infrastructure is most influential on tour-
ify and elucidate the empirical model and define our ist flows to African and Asian countries, implying that inter-
variables. After presenting and discussing our estimation national tourists value transport infrastructure highly when
results, we conclude by providing policy implications. traveling to unfamiliar destinations. In another study,
Seetanah and Khadaroo (2009) highlighted the positive
short- and long-term effects of transport capital stock on
Transport and Tourism
incoming tourist arrivals.
The success of the tourism industry is closely tied to trans-
port connectivity between origin and destination locations.
Air Transport and Tourism
In the tourism literature, transport systems are defined
broadly as “the operation of, and interaction between, trans- Air transport possesses several advantages over other trans-
port modes, ways and terminals that support tourists into and port modes, namely, greater speed, safety, service quality,
out of destinations and also the provision of transport ser- and reliability. As suggested in previous literature, air
vices within the destination” (Prideaux 2000, p.56). The pri- transport infrastructure plays a significant role in stimulat-
mary function of transport infrastructure is to provide access ing regional tourism growth. Air travel is generally reliable
to specific regions. Compared to underperforming networks, (Mukkala and Tervo 2013), which is particularly important
effective infrastructure minimizes transportation costs and for long-haul travelers and those traveling to remote and
attracts more tourists (Khadaroo and Seetanah 2007b). peripheral areas with inferior land transport (Mukkala and
Service quality of transport is similarly important; it influ- Tervo 2013). For long-haul travel, air transport greatly
ences tourists’ overall assessment of a destination and, by reduces travel time and, in turn, travel-related opportunity
extension, their revisit intention and word-of-mouth (WOM) costs. Air transport infrastructure may also allow destina-
communication with other potential tourists (Ahrholdt, tions to establish and/or strengthen tourism-related eco-
Gudergan, and Ringle 2016). nomic linkages with other areas to create competitive
A destination’s transport infrastructure (i.e., roads, air- advantages. Bieger and Wittmer (2006) proposed a system
ports, railways, and harbors) plays a central role in shaping model to analyze the interconnection between air transport
Yang et al. 27

and tourism in which air transport supply improves the tourism products to create higher-value promotions. Lastly,
quantity, quality (spending power), and structure (length of HSR provides superior on-board comfort and similar advan-
stay and travel purpose) of incoming visitors. Based on this tages over other transport modes.
system, they also developed an analytical framework to Given its growing popularity, many scholars have investi-
understand success factors associated with different airline gated the impact of HSR access on local tourism demand,
business models. with Asia being a common region of study. Su and Wall
The accessibility and type of air travel also informs tourist (2009) found that the opening of the Qinghai–Tibet railway
flows. Khadaroo and Seetanah (2007a) looked at the signifi- greatly boosted tourism in Tibet, and travelers identified the
cant impact of airport terminals on inbound tourist flows to railway as a major factor when choosing Tibet as a destina-
26 island destinations. Tveteras and Roll (2014) investigated tion. Yan, Zhang, and Ye (2014) analyzed the impact of the
the effects of nonstop international flights on tourist flows newly constructed Wuhan–Guangzhou HSR on local tourist
and found that the long-run demand elasticity of long-haul receipts and found significant effects in two of the three
direct flights ranges from 0.3 to 0.5 in Peru. However, Duval provinces served by the railway. Chen and Haynes (2015)
and Schiff (2011) found insignificant effects of nonstop air discovered that China’s HSR led to a 29% increase in inter-
services for international visitors from most countries to national tourism demand from different origin countries.
New Zealand, explaining that leisure travelers are willing to It is important to note that the benefits of HSR differ across
take indirect flights to secure lower fares. Y. Yang and Wong destinations. For example, Pagliara et al. (2015) found Spain’s
(2012) estimated the impact of destination-side air transport HSR system to influence tourists’ destination choices of cities
infrastructure on tourist flows to Chinese cities and found the near Madrid but not travelers’ choice of Madrid itself.
impact of air transport to be consistently lower than that of Delaplace et al. (2014) discovered that the French HSR net-
road transport. An exception is western cities in China, where work’s esteemed reputation informed travelers’ decisions to
the air transport effect is largest because of poor land trans- visit Paris. Italy’s HSR did not exert the same impact in Rome,
port accessibility. Lastly, the expansion of low-cost carriers however, because the country’s system is not as popular.
(i.e., airlines that provide flights at lower prices) not only
improves market accessibility for outbound tourists but also
has a positive impact on the number of inbound tourist arriv-
Intermodal Transport Competition
als to destinations (Rey, Myro, and Galera 2011). The presence of HSR does not necessarily increase travel
demand by itself; instead, it can cannibalize demand for other
transport modes, such as air and rail. As mentioned earlier,
Rail Transport and Tourism rail travel (and particularly HSR) is more competitive at short
Technological advancements have positioned rail transport distances because of higher-frequency services, cheaper
as an appealing travel option in terms of safety, convenience, fares, proximity to city centers, and service reliability and
timeliness, flexibility, and affordability (Yan, Zhang, and Ye safety (Taniguchi 1992). Albalate and Fageda (2016) found
2014). Compared to air transport, rail transport may be more that the launch of Spain’s HSR network did not increase tour-
economical for some tourists. Givoni and Dobruszkes (2013) ist arrivals because some tourist segments use HSR as a sub-
noted the importance of “door-to-door” travel time when stitute for air transport. Dobruszkes, Dehon, and Givoni
evaluating transport connectivity. Unlike airports, most train (2014) found that the intermodal competition between air ser-
stations are located closer to city centers and urban activities, vice and HSR is contingent upon HSR’s travel time and dis-
allowing tourists to immerse themselves in their destination tance; shorter distances come with fewer air service options.
nearly immediately (Fu, Zhang, and Lei 2012). Many train When the HSR network opened in Spain, Jiménez and
stations do not require passengers to undergo thorough (and Betancor (2012) found that the number of air transport options
time-consuming) security screenings, further reducing tour- decreased by 17%, and the market share of major airlines
ists’ transport obligations (Pagliara, Vassallo, and Román across different travel modes declined. Based on model esti-
2012). HSR systems in particular have revolutionized tour- mates, Martín and Nombela (2007) posited that trains divert
ism by minimizing travel time and improving passenger travelers from planes and buses for long-distance routes of
comfort (Albalate and Fageda 2016). HSR availability also more than 500 km, whereas most rail demand comes from car
influences tourists’ perceptions of a destination’s accessibil- users traveling shorter routes. Pagliara, Vassallo, and Román
ity and utility (Pagliara et al. 2015) and continues to change (2012) underscored price and service frequency as shaping
the competitive landscape across destinations (Masson and the substitution of HSR for scheduled air service in Spain.
Petiot 2009; X. Wang et al. 2012). Bazin, Beckerich, and Using a supply-side empirical analysis, Albalate, Bel, and
Delaplace (2011) identified several reasons why HSR is pre- Fageda (2015) found that HSR–air competition depends on
ferred by some tourists in Europe. For short-haul urban tour- HSR network design and length as well as HSR station loca-
ists, HSR provides an attractive alternative to driving by tions. When confronted with HSR competition, air services
eliminating problems related to fatigue, traffic congestion, were reduced more at hub airports than at nonhub airports,
and parking. HSR service can also be bundled with other and even more so at airports without an on-site HSR station.
28 Journal of Travel Research 58(1)

Analyzing transport mode choice from the demand-side location information. Many tourism studies have demon-
perspective provides useful information about competition strated the usefulness and accuracy of geo-tagged social
across different modes. Nerhagen (2003) found that for media data in monitoring large-scale tourist flows (Kádár
visitors to a major ski resort in Sweden, those who traveled 2014; Hawelka et al. 2014). Indeed, Weibo data provide an
long distances preferred traveling by train rather than by innovative way to monitor mobility flows on precise tempo-
car. Likewise, van Goeverden (2009) investigated long- ral and spatial scales (W. Wu, Wang, and Dai 2016). D. Li
haul German and Dutch tourists’ decisions to travel by train and Yang (2017) applied Weibo data to model interprovince
versus alternative modes of transport; the likelihood of tourist flows in China, and our study extends their work by
traveling by rail is highest at travel distances between 600 monitoring dyadic tourist flows at a finer geographic scale
and 900 km and lowest between 1,400 and 1,500 km. at the intercity level.
Thrane (2015) analyzed long-haul Dutch tourists’ transport We used the nearby-search function of a public Weibo
choices among private car, air, and other public transit ser- Application Programming Interface to collect geo-tagged
vices; long-distance (>400 km) travelers and those with data by overlapping a fishnet across China (D. Li and Yang
higher household incomes indicated stronger preferences 2017). All geo-tagged Weibo data were cleaned and pro-
for air transport. cessed to obtain destination information at the prefectural
The foregoing review suggests that the impact of trans- city level using spatial joins. We only retained records
port connectivity on tourist flows is underresearched, espe- related to domestic travel within mainland China; all other
cially in dyadic contexts that include origin and destination data were discarded. Origin information was obtained based
information about transport connectivity. Furthermore, the on users’ geo-tagged check-ins during the seven days prior
results of previous studies fail to illuminate how transport to Golden Week. We also collected geo-tagged information
effects vary at different origin-to-destination travel distances from one month before the holiday. If users’ self-reported
and whether a competition/substitution pattern across trans- home cities (as indicated in their Weibo profile) did not
port modes may shape interregional tourist flows. appear in their geo-tagged records, we excluded those obser-
vations. More than 6.67 million geo-tagged Weibo posts
appeared during Golden Week, representing an 18% increase
Research Methods in the daily average number of geo-tagged Weibo posts
Data Collection compared to the week before. Weibo check-in records with
unique user IDs were considered individual travel records if
This study evaluates intercity tourist flows among 343 a geo-tagged destination city differed from a user’s city of
Chinese prefectural cities during the 2014 National Day origin. Because the Chinese government regulates public
Golden Week (October 1–7). First enacted by the Chinese holidays, we assumed there were very few business-related
central government in 1999, the Golden Week vacation pol- intercity trips during Golden Week. We then aggregated
icy aims to stimulate domestic tourism and related consump- individual geo-tagged Weibo posts to generate a city-to-city
tion to maximize tourism’s multiplier effect on the national tourist flow matrix.
economy (B. Wu et al. 2012). Following a 2007 amendment To verify the credibility of Weibo data in monitoring
to the Golden Week policy, the week coinciding with National tourist flows, we compared the total number of geo-tagged
Day (October 1) became the primary time of year when Weibo posts in each destination with (1) the announced
Chinese travel for vacation (Yan and Zhang 2010). Most number of tourist arrivals during the 2014 Golden Week,
employers in China do not offer paid leave; therefore, the and (2) official statistics on tourist arrivals in 2013. As
seven-day holiday presents a valuable opportunity for shown in the log–log scatterplots (Figure 1), a strong linear
Chinese to take extended vacations, especially to long-haul association suggests the geo-tagged social media data we
domestic destinations (D. Li and Yang 2017). According to collected from 2014 mimic recorded tourism flows in
official statistics, 475 million domestic tourists traveled dur- China. Unlike official statistics, however, Weibo data cap-
ing Golden Week in 2014, totaling 245.3 billion RMB in ture origin information and thus depict origin-to-destina-
tourism receipts, which comprised 8.09% of the annual total tion tourist flows dyadically.
(CNTA 2014). Figure 2 presents the flow map of intercity tourist flows
Because of a lack of official statistics on dyadic tourist based on geo-tagged Weibo posts. A diamond shape appears
flows in China’s prefectural cities, we used an alternative in the map, representing Beijing, Shanghai, Guangzhou, and
data source, Weibo check-ins, to construct a 343-by-343 Chengdu as four apexes. The diamond covers key tourist
tourist flow matrix. Sina Weibo, a microblogging website hotspot areas previously identified in China (Y. Yang and
similar to Twitter, is one of the most popular Chinese social Wong 2013). The three largest pairwise flows are from
media platforms; as of December 2014, there were 175.7 Beijing to Tianjin, from Chengdu to Chongqing, and from
million monthly active users and 80.6 million daily active Guangzhou to Shenzhen. In addition, 69.04% of city-pairs
users. Weibo members can post short text messages and have a value of 0, indicating an absence of Weibo posts in
photos, which include timestamps and optional geo-tagged our data.
Yang et al. 29

Figure 1. Log–log scatterplots of aggregate geo-tagged Weibo counts against tourism statistics.

Figure 2. Spatial pattern of intercity tourist flows based on Weibo data.

In addition to tourist flow information, we also col- Unlike air transport, the train timetable is based on rail-
lected data on intercity air and rail transport connectivity way stations instead of cities. We measured intercity rail
using the number of scheduled passenger air flights and transport connectivity based on (1) principal and class I
trains based on official timetables from the Civil Aviation stations served by the CRC, (2) station names that con-
Administration of China and China Railway Corporation tained the prefectural city’s name, and (3) other stations to
(CRC), respectively. Because some domestic flights are ensure our data included at least one train station in each
not scheduled daily, we measured air transport connectiv- city. Because of data unavailability, we did not include
ity by the number of flights scheduled per week between some temporary trains that operated exclusively during
two cities. We excluded code-share flights to avoid double Golden Week. According to the CRC, trains can be broadly
counting, and we merged air flight counts for multiple air- categorized into three types: (1) HSR trains, with an opera-
ports serving a single destination. We then constructed a tional speed of up to 300 km/h on HSR lines, most of
343-by-343 matrix, WA, to capture air transport connectiv- which were built after 2007; (2) bullet trains, with an oper-
ity among cities. ational speed of up to 200 km/h on HSR and conventional
30 Journal of Travel Research 58(1)

Figure 3. Histogram of scheduled flights and trains for different distances (in km).

rail lines; and (3) ordinary trains, which include all other assumptions, we can derive a gravity model for dyadic
types of trains operating on conventional rail lines. We tourist flows using individual tourists’ aggregate demand.
then constructed a 343-by-343 rail transport connectivity The general gravity model is written as follows (Sen and
matrix, WR. In total, our sample was made up of 11,793 Smith 1995):
scheduled flights and 4,903 scheduled trains, 24.35% of
which were HSR trains and 31.49% of which were bullet yij = Ai B j f ( Dij ), (1)
trains. The Supplementary Material includes the flow map
of air and rail transport connectivity. where yij is the interaction between two entities i and j,
We further investigated the characteristics of the air and Ai and B j are origin- and destination-specific masses,
rail transport connectivity matrices to understand how con- and f ( Dij ) represents the separation between the two
nectivity varies over different distances between cities. The entities, which is usually captured by the distance between
results appear in Figure 3. First, the distribution of air trans- them.
port connectivity has a flatter center and a fatter tail over The dependent variable in our empirical model is the
distance, indicating that compared to rail transport, air trans- number of geo-tagged Weibo check-in posts, denoted by
port connectivity dominates in long-distance city-pairs. nonnegative integers. We used the count data model to
Second, by comparing the distributions of various rail con- establish a gravity model, which mitigates several draw-
nectivity types, we find that HSR trains’ connectivity has a backs of the traditional log-linear ordinary least squares
longer median distance than that of bullet trains. However, model including (1) estimation bias after logarithmic
HSR and bullet trains are rarely used to travel to long-dis- transformation, (2) violation of the assumption of equal
tance destinations ≥1,500 km away. Lastly, we constructed variance of error terms, and (3) sensitivity of the results to
the indirect connectivity matrices such that W*A = WA • WA zero-valued flows (Burger, van Oort, and Linders 2009).
and W*R = WR • WR to allow for one stopover between two The Poisson model, the most popular count data model,
cities. suffers from a limitation on the equidispersion assump-
tion, which requires equality between the mean and vari-
ance of the dependent variable. Burger, van Oort, and
Empirical Model Linders (2009) recommended using a negative binomial
Based on the utility-maximizing framework, we assume (NB) model to alleviate the equidispersion restriction in
each domestic tourist is economically rational and free to gravity modeling; here, the probability that a destination j
travel domestically. By comparing the expected utility of will attract a domestic tourist from origin i depends on a
any set of domestic destinations, a tourist selects the one set of factors xij and the likelihood of observing a count
that appears to have the greatest utility. According to number of yij for the origin–destination pair i and j, rep-
Morley, Rosselló, and Santana-Gallego (2014), under these resented thus:
Yang et al. 31

f ( yij ) =
(
Γ yij + α −1 )  α −1 
 * 
α −1
 λij*
 *
yij

 ,
world heritage sites (Patuelli, Mussoni, and Candela 2013;
C.-H. Yang, Lin, and Han 2010). We also included two
( )
Γ ( yij + 1) Γ α −1
−1 
 λ ij + α 
−1 
 λij + α 
(2)
dummy variables to capture the city hierarchy in the Chinese
where Γ ( ⋅) denotes the gamma function. The conditional political system: MUNICIPALITY_D and PROV_CAPITAL
expectation of yij , λ ij , is specified as a log-linear function _D. Cities in upper levels of the hierarchy receive significant
of a set of explanatory variables xij . resources from the central government, which likely affects
The NB model collapses to the Poisson model if α their infrastructure construction. Lastly, we considered four
becomes 0; this can be estimated using the maximum likeli- geo-spatial variables. Among them, lnDISTANCE, as a clas-
hood method. sical variable in the gravity model, captures the geographical
separation between origins and destinations (Morley,
Rosselló, and Santana-Gallego 2014); it can also be used as
Variable Definition and Description a proxy for personal road transport connectivity (Bilotkach,
As explained earlier, our dependent variable y comes from Fageda, and Flores-Fillol 2010). Another geo-spatial vari-
the matrix of intercity tourist movements. Exclusion of the able, lnCD_D, measures the competing destination (CD)
diagonal elements in the 343-by-343 matrix yielded a total of effect in the gravity model, which represents spatial competi-
117,306 pairwise observations. Because of data unavailabil- tion versus spatial agglomeration (Y. Yang, Fik, and Zhang
ity, we were unable to obtain water transport connectivity 2013). If d denotes the distance between two cities, the CD
data across cities. According to official statistics, less than index of destination j relative to origin i is calculated as
1.2% of total passenger traffic flows in China are associated
with water transport (National Bureau of Statistics of China p
 A5 p + NPp + WHS p 
2015). Table 1 presents descriptions of our control variables CDij = ∑ 
d pj < dij
d pj
.


(3)
and variables of interest. We introduced connectivity mea-
sures of different transport modes to the model successively.
Endogeneity of connectivity measures is a common issue in Additionally, SAME_PROV indicates within-province tour-
econometric modeling; however, it proved trivial in our ism to control for border/boundary effects on tourism (Smith
research. When transport service providers map out service and Xie 2003), and NEIGHBOR represents tourism between
routes, tourism is usually not a major concern. For example, neighboring cities (Fourie and Santana-Gallego 2011).
as shown in some transport connectivity maps (see Table 1 also presents the statistical summary and collin-
Supplementary Material), HSR and bullet train connectivity earity diagnostics of independent variables in the NB gravity
was largely determined by economic conditions, population, model. Large variations are observed for most variables.
and city location. This argument is further supported by the Regarding collinearity diagnostics, the largest variance infla-
results of auxiliary regressions of air/rail connectivity of tion factor (VIF) value is 2.93 across major independent
each city on the city’s level of domestic tourism specializa- variables. The VIF values of all variables are much lower
tion (domestic tourism revenue over gross domestic product than the cutoff value of 10, and their tolerance values are
[GDP]). The coefficients of tourism specification variables greater than 0.2, suggesting the absence of multicollinearity
are estimated to be insignificant, and these results are avail- (Dormann et al. 2013).
able upon request.
As suggested by past studies, we incorporated a large set
of control variables into our empirical gravity model. In the
Empirical Results
model, lnGDP_O/lnGDP_D measures origin- and destina- In Table 2, we present the estimation results of the NB mod-
tion-specific mass (Marrocu and Paci 2013). lnMOBILE_O/ els. In model 1, our benchmark model, we considered two
lnMOBILE_D measures a city’s information and communi- general transport connectivity variables, lnFLIGHTS (air
cation technology (ICT) development, which can influence transport connectivity) and lnTRAINS (rail transport connec-
the penetration rate of social media tools (including Sina tivity), and the control variables. The likelihood ratio test sug-
Weibo) (Liu et al. 2014). lnHOTEL_D captures the extent of gests the NB model is superior to the Poisson model in
tourism infrastructure in a destination (Patuelli, Mussoni, addressing overdispersion issues. The coefficient of a natu-
and Candela 2013). lnHIGHWAY_D reflects the state of a rally logged independent variable can be directly interpreted
destination’s road transport infrastructure (Y. Yang and Wong as its elasticity (Long and Freese 2006). The coefficients of
2012). In terms of destination attractiveness measures, we lnFLIGHTS and lnTRAINS are estimated to be 0.386 and
considered the three most distinguished tourist attraction 0.275, respectively, indicating that for all city-pairs, a 1%
types in China: AAAAA scenic spots, which are designated increase in air connectivity (as measured by the number of
by the country’s National Tourism Administration as the scheduled passenger flights) and rail connectivity (as mea-
highest level of scenic spots in the country (Hong, Ma, and sured by the number of scheduled passenger trains) leads to a
Huan 2015), national parks (Y. Yang and Wong 2012), and 0.386% and 0.275% increase in tourist flows between cities,
32 Journal of Travel Research 58(1)

Table 1. Descriptive Statistics and Collinearity Diagnostics of Independent Variables.

Variable Definition Mean SD VIF Tolerance

lnFLIGHTS Log of one plus the number of scheduled passenger flights 0.074 0.460 1.19 0.840
between origin and destination each week
lnTRAINS Log of one plus the number of scheduled passenger trains 0.296 0.685 1.43 0.698
between origin and destination each day
lnFLIGHTS_INDIRECT Log of one plus the number of indirect scheduled passenger 1.471 2.886
flights (with one stopover) between origin and destination
each week
lnTRAINS_INDIRECT Log of one plus the number of indirect scheduled passenger 3.860 3.174
trains (with one stopover) between origin and destination
each day
lnTRAINS_EN_ROUTE Log of one plus the number of scheduled en route passenger 0.255 0.635
trains between origin and destination each day
lnTRAINS_OT Log of one plus the number of scheduled passenger trains 0.068 0.321
between origin and destination with the origin as the
originating station and/or the destination as the terminal stop
each day
lnTRAINS_ORDINARY Log of one plus the number of scheduled ordinary passenger 0.260 0.606
trains between origin and destination each day
lnTRAINS_HSR Log of one plus the number of scheduled HSR passenger trains 0.051 0.329
between origin and destination each day
lnTRAINS_BULLET Log of one plus the number of scheduled bullet passenger trains 0.035 0.263
between origin and destination each day
lnGDP_O Log of origin city GDP (in 10,000 RMB) in 2013 6.982 1.090 1.38 0.726
lnMOBILE_O Log number of registered mobile phone users relative to the −0.185 0.491 1.29 0.774
total population in origin in 2013
lnGDP_D Log of destination city GDP (in 10,000 RMB) in 2013 6.982 1.090 2.93 0.341
lnMOBILE_D Log number of registered mobile phone users relative to the −0.185 0.491 1.52 0.657
total population in destination in 2013
lnHIGHWAY_D Log of a destination’s highway density (km per km2) in 2013 −0.452 0.910 2.79 0.358
lnA5_D Log of one plus a destination’s number of AAAAA scenic spots 0.307 0.450 1.68 0.596
in 2014
lnNP_D Log of one plus the destination’s number of national parks in 0.387 0.474 1.36 0.736
2014
lnWHS_D Log of one plus the destination’s number of world heritage sites 0.118 0.280 1.32 0.759
in 2014
lnHOTEL_D Log of a destination’s number of star rated hotels in 2013 3.327 0.924 2.31 0.433
MUNICIPALITY_D Dummy indicator of four direct-controlled municipalities in 0.012 0.107 1.25 0.798
China as destinations; 1 = destination cities Beijing, Shanghai,
Tianjin, or Chongqing, and 0 = otherwise
PROV_CAPITAL_D Dummy indicator of provincial capital cities as destinations; 0.079 0.269 1.34 0.747
1 = capital city of a province as destination, and 0 = otherwise;
lnCD_D Destination’s CD index in 2014 −1.104 0.840 2.78 0.360
lnDISTANCE Log of the geographic distance (in km) between origin and 7.049 0.701 2.40 0.417
destination
SAME_PROV Dummy indicator for travel within a single province; 1 = origin 0.038 0.190 1.49 0.671
and destination cities within the same province, and 0 = otherwise
NEIGHBOR Dummy indicator for travel to a neighboring destination; 0.015 0.123 1.44 0.695
1 = neighboring origin and destination cities sharing a land border,
and 0 = otherwise

Note: GDP, gross domestic product; HSR, high-speed rail; VIF, variance inflation factor.

respectively. Therefore, in general, the data demonstrate that For our control variables, two origin-specific variables,
air transport is more important than rail transport in generat- lnGDP_O (origin GDP) and lnMOBILE_O (mobile user per
ing intercity tourist flows. capita in origin), are estimated to be positive and significant.
Yang et al. 33

Table 2. Estimation Results of NB Models with Basic Specifications.

Variable Model 1 Model 2 Model 3 Model 4 Model 5

lnFLIGHTS 0.386*** 0.215*** 0.486*** 0.353*** 0.386***

(0.014) (0.015) (0.018) (0.014) (0.014)
lnTRAINS 0.275*** 0.254*** 0.295***
(0.010) (0.013) (0.010)
lnFLIGHTS_INDIRECT 0.0749***
(0.003)
lnTRAINS_INDIRECT 0.0187***
(0.004)
lnFLIGHTS*lnTRAINS −0.104***
(0.010)
lnTRAINS_EN_ROUTE 0.114***
(0.011)
lnTRAINS_OT 0.557***
(0.019)
lnTRAINS_ORDINARY 0.285***
(0.012)
lnTRAINS_HSR −0.0115
(0.020)
lnTRAINS_BULLET 0.139***
(0.023)
lnGDP_O 0.490*** 0.462*** 0.490*** 0.480*** 0.493***
(0.015) (0.015) (0.015) (0.015) (0.015)
lnMOBILE_O 0.992*** 0.942*** 0.988*** 0.942*** 0.998***
(0.022) (0.023) (0.022) (0.022) (0.022)
lnGDP_D 0.403*** 0.361*** 0.402*** 0.402*** 0.406***
(0.015) (0.017) (0.015) (0.015) (0.015)
lnMOBILE_D 0.226*** 0.242*** 0.224*** 0.229*** 0.233***
(0.023) (0.023) (0.023) (0.023) (0.023)
lnHIGHWAY_D −0.0812*** −0.0317* −0.0825*** −0.0828*** −0.0816***
(0.019) (0.019) (0.019) (0.019) (0.019)
lnA5_D 0.176*** 0.157*** 0.177*** 0.181*** 0.177***
(0.028) (0.031) (0.029) (0.028) (0.028)
lnNP_D 0.270*** 0.226*** 0.268*** 0.275*** 0.272***
(0.022) (0.021) (0.022) (0.022) (0.022)
lnWHS_D 0.235*** 0.280*** 0.237*** 0.234*** 0.230***
(0.035) (0.035) (0.035) (0.035) (0.035)
lnHOTEL_D 0.241*** 0.203*** 0.241*** 0.242*** 0.240***
(0.016) (0.016) (0.016) (0.016) (0.016)
MUNICIPALITY_D 0.142* 0.177** 0.136 0.167** 0.141
(0.086) (0.090) (0.086) (0.084) (0.086)
PROV_CAPITAL_D 0.789*** 0.691*** 0.790*** 0.814*** 0.786***
(0.035) (0.035) (0.035) (0.034) (0.035)
lnCD_D −0.190*** −0.181*** −0.188*** −0.193*** −0.188***
(0.020) (0.019) (0.020) (0.020) (0.020)
lnDISTANCE −0.866*** −0.873*** −0.863*** −0.883*** −0.872***
(0.020) (0.020) (0.020) (0.020) (0.020)
SAME_PROV 1.808*** 1.817*** 1.809*** 1.763*** 1.809***
(0.041) (0.041) (0.041) (0.040) (0.041)
NEIGHBOR 1.122*** 1.065*** 1.120*** 1.101*** 1.122***
(0.139) (0.126) (0.140) (0.136) (0.140)
CONSTANT −1.234*** −0.701*** −1.260*** −1.050*** −1.223***
(0.211) (0.207) (0.212) (0.209) (0.212)
Lnα 1.604*** 1.588*** 1.602*** 1.598*** 1.604***
(0.010) (0.010) (0.010) (0.010) (0.010)
N 117,306 117,306 117,306 117,306 117,306
Pseudo-R2 0.137 0.139 0.137 0.138 0.137
AIC 345013.4 344265.5 344956.9 344742.7 345031.2
BIC 345197.2 344468.6 345150.4 344936.1 345234.3

Note: AIC = Akaike’s information criterion; BIC = Bayesian information criterion; NB = negative binomial. Asterisks indicate significance at ***0.01, **0.05, and *0.1. Robust
standard errors are presented in parentheses.
34 Journal of Travel Research 58(1)

The magnitudes of these two coefficients are larger than

those of their destination-specific counterparts, lnGDP_D
(destination GDP) and lnMOBILE_D (mobile user per cap-
ita in destination), both of which are also statistically signifi-
cant and positive. Interestingly, the coefficient of
lnMOBILE_O is close to 1. This result can be partially
attributed to the social media data we used to construct our
intercity tourist movement matrix; the ICT penetration rate
on the origin side, proxied by lnMOBILE_O, is highly asso-
ciated with the popularity of social media use among origin
residents. With regard to other destination-side variables, the
significant coefficient of lnHOTELS_D (number of star-
rated hotels) indicates that hotel infrastructure plays a sig-
nificant role in explaining dyadic tourist flows.
All our attraction variables—lnA5_D (AAAAA attrac-
tions), lnNP_D (national parks), and lnWHS_D (world heri- Figure 4. Effects of different types of public transport
connectivity on domestic tourist flows.
tage sites)—are estimated to be significant and positive,
suggesting that domestic tourists travel to certain destina-
tions to visit these attractions. Based on the magnitudes of significant; however, that of lnTRAINS_OT is almost five
their coefficients, we find that national parks and world heri- times larger than that of lnTRAINS_EN_ROUTE, demon-
tage sites (lnWHS_D) are more attractive than AAAAA sce- strating that for each city-pair, trains originating from the ori-
nic spots (lnA5_D) to domestic tourists. For the two city gin and/or terminating at the destination boost tourist flows
hierarchical indicators, MUNICIPALITY_D is positive and much more effectively than en route service between two cit-
moderately significant at the 0.10 level, whereas PROV_ ies. One reason for this is that CRC usually assigns many
CAPITAL_D is positive and significant at the 0.01 level. more seats at originating and terminating stations than at oth-
These results show that provincial capitals tend to attract ers, therefore limiting en route stations’ transport capacity.
twice (exp(0.789)) as many tourists as other cities, all other In model 5, we compare the effects of three major types of
variables remaining constant. Moreover, the significant and trains: ordinary passenger trains (lnTRAINS_ORDINARY),
negative coefficient of lnCD_D highlights cross-destination HSR trains (lnTRAINS_HSR), and bullet trains (lnTRAINS_
competition losses rather than agglomeration gains in inter- BULLET). The coefficient of lnTRAINS_ORDINARY is
city tourist flows. The coefficient of lnDISTANCE, the tradi- the largest, indicating that compared to HSR and bullet
tional distance decay coefficient in the gravity model, is trains, the connectivity of ordinary passenger trains domi-
estimated to be −0.866 and statistically significant. The polit- nately shape intercity tourist flows in China. The coefficient
ical and physical linkage variables, WITHIN_PROV and of lnTRAINS_BULLET is also positive and significant.
NEIGHBOR, are both positive and significant. Compared to However, that of lnTRAINS_HSR is insignificant, which
other city-pairs, tourist flows are six times higher (exp(1.808)) may lead to questions regarding the effectiveness of the
between cities within the same province and three times national HSR network in boosting tourist flows. This result is
higher (exp(1.122)) between bordering cities, all other vari- consistent with Albalate and Fageda’s (2016) finding about
ables held constant. the limited impact of HSR on tourism demand in Spain.
Model 2 includes two additional variables, lnFLIGHTS_ Lastly, the signs, magnitudes, and significance levels of the
INDIRECT and lnTRAINS_INDIRECT, which allow for one estimated coefficients are largely consistent across control
stopover connection between two cities via air and rail, variables in models 1–5, suggesting robustness of the results
respectively. Both variables are estimated to be positive and with regard to model specifications.
significant. However, their magnitudes are considerably Figure 4 summarizes the overall effects of different
smaller than their direct connectivity counterparts. In model types of transport connectivity (estimated coefficients of
3, we include an interaction term, lnFLIGHTS*lnTRAINS, to measures and their corresponding 95% confidence inter-
capture the intermodal competition effect between air and rail vals) on intercity tourist flows as depicted in models 1–5.
transport. Its coefficient is estimated to be negative and sig- All effects except for those of HSR trains are statistically
nificant, highlighting the fierce competition between air and significant and positive. By comparing their magnitudes,
rail transport in shaping domestic tourist flows in China. In we find that (1) the effect from indirect transport connec-
model 4, we distinguish between two different types of rail tivity is consistently small for air flights and trains, and (2)
transport connectivity: through en route trains (lnTRAINS_ connectivity between originating/terminating trains has
EN_ROUTE) and through originating and/or terminating the largest effect in spurring domestic tourist movement
trains (lnTRAINS_OT). Both coefficients are positive and between cities.
Yang et al. 35

Figure 5. Effect of different transport connectivity measures by

distance. Figure 6. Effect of different intermodal transport competition
measures by distance.
As noted in previous literature, transport connectivity
effects depend heavily on travel distance and time (Givoni and distances between 1,400 and 1,600 km, whereas the most
Dobruszkes 2013; Bilotkach, Fageda, and Flores-Fillol 2010). intense air–bullet train competition (lnFLIGHTS*lnTRAINS_
Therefore, we estimated the distance-based effects of different BULLET) occurs for distances between 1,200 and 1,400 km.
transport connectivity measures using their interactions with a
set of distance–interval dummy variables. Because of space
Conclusion
limitations, we present these results in the appendix. Figure 5
demonstrates how the effects vary over distance. The distance- In this study, we used geo-tagged Sina Weibo posts to track
based coefficient of lnFLIGHTS generally increases over dis- intercity domestic tourist flows between Chinese cities dur-
tance, suggesting that air transport connectivity becomes more ing the 2014 National Day Golden Week. Using a gravity
important as travel distance increases, especially for long-haul model, we investigated how air and rail transport connectiv-
travel. Similarly, Figure 5 shows that the effect of rail transport ity shapes intercity tourist flows. Our results suggested that
connectivity (lnTRAINS) increases over distances smaller both types of transport play a significant role in strengthen-
than 2,600 km, and its magnitude is close to that of air trans- ing linkages between cities in terms of tourist flows, with
port connectivity (lnFLIGHTS). However, for travel distances significant indirect connectivity effects that are smaller in
more than 2,800 km, the effect of rail transport declines magnitude. Among different connectivity measures by train
sharply, implying that air transport is particularly important for type, we found that originating and/or terminating trains are
long-haul domestic tourists. Regarding the distance-based much more influential than en route trains in boosting tourist
effects of HSR and bullet trains, Figure 5 indicates that the flows, and ordinary trains facilitate intercity tourist flows
effect of HSR trains is strongest for travel distances between more than bullet and HSR trains. The impacts of transport
1,800 and 2,000 km, while that of bullet trains is strongest for connectivity and air–rail competition vary based on origin-
distances between 400 and 600 km. These results corroborate to-destination distance. Although the effect of air transport
Fu, Zhang, and Lei’s (2012) argument that Chinese HSR ser- connectivity strengthens as city-pair distance increases, the
vices are competitive in city-pairs of short to medium dis- effects of HSR and bullet trains are significant only for city-
tances. Results for the distance-based effect of ordinary trains pairs with short and medium distances between them. Lastly,
are similar to those for all types of trains (lnTRAINS), and the results show that the air–rail intermodal competition
they are available on request. effect generally becomes more intense as distance increases.
We estimated additional models to understand how intermo- Across different rail transport connectivity measures, we
dal air–rail competition varies over different distance intervals; found that connectivity by ordinary trains had the largest
these results are presented in the appendix. Specifically, we impacts on intercity tourist flows. This result can be explained
examined the competition effect between transport modes via by the substantial scope and size of the ordinary train network
interactions of intertransport competition measures and dis- in China. Only when a transport network achieves a certain
tance–interval dummy variables. Figure 6 shows that competi- size can it benefit the local tourism industry. As shown in the
tion effects change over distance as suggested by our estimation Supplementary Material, China’s HSR and bullet train net-
results. Notably, air–rail competition (lnFLIGHTS*lnTRAINS) works remained relatively underdeveloped in 2014, covering
generally increases over distance. The most intense air–HSR a small number of cities. Our results indicated that the impact
competition (lnFLIGHTS*lnTRAINS_HSR) exists for of HSR connectivity is insignificant overall and only
36 Journal of Travel Research 58(1)

moderately significant at a few middle-distance intervals. not least, our study represents the very first research effort
One possible explanation is that the HSR system was designed investigating interregional/intercity domestic tourist flows
as a node-to-node transport mode to serve prime urban areas within China. With the estimates of origin- and destination-
and may not fully address tourists’ transportation needs specific independent variables in gravity models, we are able
(Albalate and Fageda 2016), thereby limiting its impact. Also, to better understand the underlying factors shaping the
high prices may hinder HSR’s recognition as a competitive domestic tourist flow pattern. Because of the statistical inter-
transport option because mass tourists tend to be more sensi- linkage between count data models (macrolevel) and discrete
tive to price than to time (Albalate and Fageda 2016), espe- choice models (microlevel) (Schmidheiny and Brülhart
cially in emerging markets. 2011), our estimates can also be used as tuning parameters
Our study made several theoretical contributions to the cur- for tourist flow simulation at a micro level based on indi-
rent knowledge of tourist flows. First and foremost, we inves- vidual tourists (Nicholls, Amelung, and Student 2017).
tigated the transport–tourism relationship from a dyadic To fully capitalize on the positive effects of transport con-
perspective. Unlike previous studies capturing transport nectivity, local tourism administrative units should work
effects by measuring destination-specific transport infrastruc- closely with public transport service providers to encourage
ture investment/stock (e.g., Khadaroo and Seetanah 2007b, and fulfill tourism demand. For instance, tourism administra-
2008), we looked into the transport connectivity between pair- tors and stakeholders are recommended to participate
wise origin and destination. It has been known that the invest- actively at various levels of transport planning projects to
ment and improvement of transport infrastructure per se do not improve overall transport connectivity to origin destinations,
necessarily improve the public transport connectivity. For particularly those with significant opportunities to expand
example, airlines or rail service providers may arrange fewer the market share. Instead of normal en route trains, tourism
scheduled services because of regional transport regulations, administrators of a destination should schedule more trains
logistical issues, and considerations on network optimization on which the destination is the originating/terminating stop.
with origins (Starkie 2012) despite superior destination-spe- Second, some tourism marketing campaigns should be
cific transport infrastructure. In many cases, it is the transport directed specifically to public transport passengers, such as
connectivity that imposes a more direct and instant effect on by playing destination advertisements on trains and planes,
tourist flows. Second, we proposed and empirically confirmed providing tourism-themed food and beverage, and naming
the intermode competition between air and rail transport, trains or planes after major destination attractions.
which was largely untapped in the previous tourism literature. In a similar vein, our results on intermodal transport com-
This competition effect highlights a substitution pattern petition and the distance-based effects of transport connec-
between these two public transport modes from a demand per- tivity have several important implications for local tourism
spective, and an overinvestment on transport may lead to a authorities when it comes to collaborating with transport ser-
decreasing return of benefits. Third, our study is among the vice providers to improve destination competitiveness. First,
very few empirical studies scrutinizing the distance-based given the competition between air and rail transport, redun-
transport effects: how the effect of transport connectivity var- dant investments should be avoided in long-term strategic
ies at different travel distances of tourists. Although these plans related to transport infrastructure development.
effects have been recognized in various conceptual studies Second, for major markets at different distance intervals,
(Prideaux 2000), none has empirically confirmed them. Our diverse transport modes should be promoted to maximize
results on these distance-based effects would help academics benefits. Our findings also provide important insights for
and practitioners better understand the nature and geographic destinations, railway companies, and airlines in identifying
scope of transport effect on tourist flows. A one-size-fits-all their target markets and allocating marketing resources based
specification of transport effect in empirical models can be on geographic locations and distance. For example, markets
problematic, and a spatial domain should be incorporated 1,600 to 2,000 km away from a destination, which are also
when evaluating the effects of different transport modes. accessible via HSR, should be prioritized when setting strat-
Another important contribution of this study is that it egies. Further, the results of this study suggest that Chinese
demonstrated social media’s usefulness in tracking domestic destinations and tour operators should not underestimate the
tourism travel patterns. Compared to traditional tourist flow role of ordinary trains in encouraging domestic tourism.
data gathered from aggregated, static official statistics and For planning purposes, these findings can be used in sce-
small-scale surveys (Hawelka et al. 2014), social media data nario analyses when transport connectivity changes. Local
provide more precise and dynamic information (W. Wu, tourism authorities may use our estimated parameters to iden-
Wang, and Dai 2016). To date, many big data analytic appli- tify the competitive advantages and disadvantages of new
cations in tourism are exploratory by nature (Pan and Yang transport plans and to project future tourist flows in the short
2016). In this study, however, we used big data to conduct and long term. For example, several new HSR routes were pro-
explanatory research with sound theoretical underpinnings. posed in the Chinese Mid- to Long-Term Railway Network
Using social media data, especially geo-tagged posts, as a Plan (2016–2025). Destinations can use our model estimates to
digital footprint measure, future applications of this data at evaluate the potential benefits associated with HSR expansion
various geographic scales could be quite promising. Last but to other markets. Then, marketing campaigns can be tailored to
Yang et al. 37

maximize said benefits. Our results can also be used to conduct and content design. Moreover, our results suggested that
a cost–benefit analysis of airport construction plans from a national parks and world heritage sites are more attractive
tourism perspective to determine appropriate capacity for a than AAAAA scenic spots to domestic tourists. Therefore,
new airport. Our findings provide several important parameters effectively embracing positive spillovers from these two
for simulation analysis, such as those related to distance decay types of attractions is essential to strengthening an area’s
and the competing destination effect. It is therefore possible to overall competitiveness as a tourist destination.
construct a gravity-type accessibility measure for geo-market- Our study did have some limitations. Weibo data con-
ing (Borodako and Rudnicki 2014) and to use agent-based straints prevented us from tracking within-city domestic flows.
modeling to monitor an individual tourist’s behavior as an This issue may be particularly detrimental for municipalities
agent (Nicholls, Amelung, and Student 2017). like Beijing and Shanghai, which have significantly larger
Lastly, the estimation results of some control variables land areas and populations than other cities. Also, because of
have useful implications for local tourism administrative data unavailability, we were unable to conduct a longitudinal
units and marketing organizations. Our results highlight analysis to examine the effects of transport connectivity over
strong territory effects associated with provincial borders time. Moreover, because tourist flows may also partially deter-
that impede domestic tourist flows. Minimizing the disad- mine transport connectivity, endogeneity problems may be a
vantages stemming from these territory effects requires tac- concern in our empirical model. Lastly, because of spillover
tical collaboration between destinations in different effects, tourist flows to a city can be influenced by tourist
provinces. Moreover, for destinations and attractions flows to neighboring cities (Marrocu and Paci 2013); our grav-
employing geo-fencing strategies to lure out-of-province ity model is unable to incorporate this type of spatial depen-
tourists (i.e., delivering targeted messages or advertisements dence. Therefore, in the future, researchers should collect
to mobile users who enter or leave a location), this finding panel data sets and apply spatial econometric models to better
could be instrumental in radius setting, audience targeting, investigate tourist flows as a type of spatial interaction.

Appendix
Table A1 . Estimation Results of NB Models with Distance-Based Effects.
Model A1 Model A2 Model A3 Model A4 Model A5 Model A6 Model A7

lnFLIGHTS 0.375*** 0.380*** 0.384*** 0.493*** 0.403*** 0.400***

(0.014) (0.014) (0.014) (0.018) (0.014) (0.014)
lnTRAINS 0.277*** 0.295*** 0.281*** 0.279***
(0.010) (0.010) (0.010) (0.010)
lnTRAINS_ORDINARY 0.286*** 0.282***
(0.012) (0.012)
lnTRAINS_HSR −0.00202
(0.020)
lnTRAINS_BULLET 0.184***
(0.024)
lnGDP_O 0.490*** 0.489*** 0.491*** 0.493*** 0.491*** 0.490*** 0.490***
(0.015) (0.015) (0.015) (0.015) (0.015) (0.015) (0.015)
lnMOBILE_O 0.992*** 0.996*** 0.998*** 0.998*** 0.988*** 0.991*** 0.992***
(0.022) (0.022) (0.022) (0.022) (0.022) (0.022) (0.022)
lnGDP_D 0.403*** 0.401*** 0.404*** 0.406*** 0.402*** 0.403*** 0.403***
(0.015) (0.015) (0.015) (0.015) (0.015) (0.015) (0.015)
lnMOBILE_D 0.226*** 0.231*** 0.232*** 0.233*** 0.224*** 0.225*** 0.227***
(0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023)
lnHIGHWAY_DD −0.0830*** −0.0743*** −0.0797*** −0.0794*** −0.0824*** −0.0815*** −0.0812***
(0.019) (0.019) (0.019) (0.019) (0.019) (0.019) (0.019)
lnA5_D 0.176*** 0.175*** 0.175*** 0.174*** 0.177*** 0.177*** 0.176***
(0.029) (0.029) (0.029) (0.029) (0.029) (0.029) (0.028)
lnNP_D 0.272*** 0.269*** 0.271*** 0.272*** 0.269*** 0.269*** 0.271***
(0.022) (0.022) (0.022) (0.022) (0.022) (0.022) (0.022)
lnWHS_D 0.234*** 0.234*** 0.232*** 0.232*** 0.237*** 0.234*** 0.235***
(0.035) (0.035) (0.035) (0.035) (0.036) (0.035) (0.035)
lnHOTEL_D 0.240*** 0.245*** 0.242*** 0.241*** 0.241*** 0.241*** 0.241***
(0.016) (0.016) (0.016) (0.016) (0.016) (0.016) (0.016)
MUNICIPALITY_D 0.138 0.144* 0.142* 0.142 0.136 0.138 0.144*
(0.086) (0.087) (0.086) (0.086) (0.086) (0.086) (0.086)
PROV_CAPITAL_D 0.787*** 0.783*** 0.787*** 0.786*** 0.792*** 0.790*** 0.787***
(0.035) (0.035) (0.035) (0.035) (0.035) (0.035) (0.035)

(continued)
38 Journal of Travel Research 58(1)

Table A1. (continued)

Model A1 Model A2 Model A3 Model A4 Model A5 Model A6 Model A7

lnCD_D −0.189*** −0.201*** −0.193*** −0.192*** −0.188*** −0.189*** −0.189***

(0.020) (0.019) (0.020) (0.020) (0.020) (0.020) (0.020)
lnDISTANCE −0.873*** −0.873*** −0.873*** −0.870*** −0.862*** −0.865*** −0.865***
(0.020) (0.021) (0.020) (0.020) (0.020) (0.020) (0.020)
SAME_PROV 1.809*** 1.857*** 1.818*** 1.827*** 1.809*** 1.806*** 1.808***
(0.041) (0.041) (0.041) (0.041) (0.041) (0.041) (0.041)
NEIGHBOR 1.113*** 1.215*** 1.134*** 1.152*** 1.122*** 1.123*** 1.122***
(0.140) (0.134) (0.138) (0.138) (0.140) (0.140) (0.139)
CONSTANT −1.187*** −1.193*** −1.197*** −1.241*** −1.266*** −1.243*** −1.235***
(0.212) (0.214) (0.212) (0.212) (0.212) (0.211) (0.211)
Distance-based variable lnFLIGHTS lnTRAINS lnTRAINS_ lnTRAINS_ lnFLIGHTS* lnFLIGHTS* lnFLIGHTS*
(in km) HSR BULLET lnTRAINS lnTRAINS_HSR lnTRAINS_BULLET
0–200 0.0739 0.0349 −0.298*** −0.142*** −0.219*** −0.156***
(0.132) (0.023) (0.038) (0.043) (0.050) (0.005)
200–400 0.292*** 0.306*** −0.0290 0.194*** −0.0945*** −0.0801*** −0.0597**
(0.036) (0.016) (0.031) (0.034) (0.017) (0.025) (0.027)
400–600 0.373*** 0.320*** 0.0275 0.326*** −0.0863*** −0.0715*** −0.0294
(0.035) (0.017) (0.032) (0.044) (0.017) (0.028) (0.043)
600–800 0.322*** 0.292*** 0.0651 0.252*** −0.117*** −0.104*** −0.107***
(0.034) (0.019) (0.045) (0.060) (0.015) (0.024) (0.022)
800–1,000 0.333*** 0.291*** 0.114* 0.284*** −0.0983*** −0.0451 −0.0700**
(0.031) (0.026) (0.060) (0.097) (0.017) (0.029) (0.035)
1,000–1,200 0.357*** 0.305*** 0.0852 0.0391 −0.103*** −0.0723*** −0.138***
(0.033) (0.032) (0.081) (0.138) (0.018) (0.024) (0.044)
1,200–1,400 0.360*** 0.303*** 0.0396 −0.139 −0.133*** −0.117*** −0.179***
(0.031) (0.038) (0.099) (0.255) (0.021) (0.037) (0.052)
1,400–1,600 0.389*** 0.341*** 0.210 0.109 −0.135*** −0.201*** −0.140*
(0.036) (0.048) (0.201) (0.369) (0.026) (0.071) (0.072)
1,600–1,800 0.413*** 0.340*** 0.575* −0.667 −0.120*** −0.0413 −0.150
(0.039) (0.066) (0.337) (0.647) (0.030) (0.106) (0.123)
1,800–2,000 0.491*** 0.438*** 0.848* −0.133*** −0.00255
(0.057) (0.087) (0.514) (0.041) (0.076)
2,000–2,200 0.523*** 0.574*** 0.0930 −0.0866** −0.0169
(0.063) (0.139) (0.646) (0.044) (0.153)
2,200–2,400 0.397*** 0.419*** −0.185**
(0.060) (0.125) (0.076)
2,400–2,600 0.489*** 0.555*** −0.129*
(0.073) (0.206) (0.070)
2,600–2,800 0.539*** 0.509*** −0.137
(0.123) (0.186) (0.089)
2,800–3,000 0.558*** 0.0617 −0.221
(0.095) (0.278) (0.158)
3,000–3,200 0.454*** 0.321 −0.115
(0.098) (0.411) (0.257)
3,200–3,400 0.561*** 0.481 −0.0944
(0.108) (0.563) (0.174)
3,400–3,600 0.740*** −0.197 −0.356
(0.106) (1.021) (0.369)
3,600–3,800 0.559*
(0.308)
3,800–4,000 0.557**
(0.271)
4,000 0.472*
(0.269)
Lnα 1.603*** 1.601*** 1.603*** 1.603*** 1.602*** 1.603*** 1.603***
(0.010) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010)
N 117,306 117,306 117,306 117,306 117,306 117,306 117,306
Pseudo-R2 0.137 0.137 0.137 0.137 0.137 0.137 0.137
AIC 345027.4 344925.4 344996.9 344991.9 344986.1 345017.5 345015.0
BIC 345404.6 345273.6 345296.7 345272.4 345343.9 345298.0 345276.2

Note: AIC = Akaike’s information criterion; BIC = Bayesian information criterion; NB = negative binomial. Asterisks indicate significance at ***0.01, **0.05, and *0.1. Robust
standard errors are presented in parentheses.
Yang et al. 39

Declaration of Conflicting Interests Dobruszkes, Frédéric, Catherine Dehon, and Moshe Givoni. 2014.
“Does European High-Speed Rail Affect the Current Level of
The author(s) declared no potential conflicts of interest with respect
Air Services? An EU-wide Analysis.” Transportation Research
to the research, authorship, and/or publication of this article.
Part A 69:461–75.
Dormann, Carsten F, Jane Elith, Sven Bacher, Carsten Buchmann,
Funding Gudrun Carl, Gabriel Carré, Jaime R García Marquéz,
The author(s) received no financial support for the research, author- Bernd Gruber, Bruno Lafourcade, and Pedro J Leitão. 2013.
ship, and/or publication of this article. “Collinearity: A Review of Methods to Deal with It and a
Simulation Study Evaluating Their Performance.” Ecography
Supplemental Material 36 (1): 27–46.
Duval, David Timothy, and Aaron Schiff. 2011. “Effect of Air
Supplemental material is available in the online version of this arti-
Services Availability on International Visitors to New Zealand.”
cle and at http://tirs.sagepub.com/supplemental
Journal of Air Transport Management 17 (3): 175–80.
Dwyer, Larry, and Chulwon Kim. 2003. “Destination
References Competitiveness: Determinants and Indicators.” Current
Ahrholdt, Dennis C., Siegfried P. Gudergan, and Christian M. Issues in Tourism 6 (5): 369–414.
Ringle. 2016. “Enhancing Service Loyalty.” Journal of Travel Fourie, Johan, and María Santana-Gallego. 2011. “The Impact of
Research 56 (4): 436–50. Mega-Sport Events on Tourist Arrivals.” Tourism Management
Albalate, Daniel, Germà Bel, and Xavier Fageda. 2015. 32 (6): 1364–70.
“Competition and Cooperation between High-Speed Rail and Fu, Xiaowen, Anming Zhang, and Zheng Lei. 2012. “Will China’s
Air Transportation Services in Europe.” Journal of Transport Airline Industry Survive the Entry of High-Speed Rail?”
Geography 42:166–74. Research in Transportation Economics 35 (1): 13–25.
Albalate, Daniel, and Xavier Fageda. 2016. “High Speed Rail and Givoni, Moshe, and Frédéric Dobruszkes. 2013. “A Review of
Tourism: Empirical Evidence from Spain.” Transportation Ex-Post Evidence for Mode Substitution and Induced Demand
Research Part A 85:174–85. Following the Introduction of High-Speed Rail.” Transport
Bazin, S., C. Beckerich, and M. Delaplace. 2011. “High Speed Reviews 33 (6): 720–42.
Railway, Service Innovations and Urban and Business Hawelka, Bartosz, Izabela Sitko, Euro Beinat, Stanislav Sobolevsky,
Tourisms Development.” In Economics and Management of Pavlos Kazakopoulos, and Carlo Ratti. 2014. “Geo-located
Tourism: Trends and Recent Developments, edited by S. M. Twitter as Proxy for Global Mobility Patterns.” Cartography
and M. A. Lisbon: Collecçao Manuais, Universidade Luisiada and Geographic Information Science 41 (3): 260–71.
Editora. Hong, Tao, Tao Ma, and Tzung-Cheng Huan. 2015. “Network
Bieger, Thomas, and Andreas Wittmer. 2006. “Air Transport and Behavior as Driving Forces for Tourism Flows.” Journal of
Tourism—Perspectives and Challenges for Destinations, Business Research 68 (1): 146–56.
Airlines and Governments.” Journal of Air Transport Jiménez, Juan Luis, and Ofelia Betancor. 2012. “When Trains
Management 12 (1): 40–46. Go Faster Than Planes: The Strategic Reaction of Airlines in
Bilotkach, Volodymyr, Xavier Fageda, and Ricardo Flores-Fillol. Spain.” Transport Policy 23:34–41.
2010. “Scheduled Service versus Personal Transportation: The Kádár, Bálint. 2014. “Measuring Tourist Activities in Cities Using
Role of Distance.” Regional Science and Urban Economics 40 Geotagged Photography.” Tourism Geographies 16 (1): 88–
(1): 60–72. 104.
Borodako, Krzysztof, and Michał Rudnicki. 2014. “Transport Khadaroo, A. J., and B. Seetanah. 2007a. “Research Note:
Accessibility in Business Travel—A Case Study of Central Does Transport Infrastructure Matter in Overall Tourism
and East European Cities.” International Journal of Tourism Development? Evidence from a Sample of Island Economies.”
Research 16 (2): 137–45. Tourism Economics 13 (4): 675–84.
Burger, Martijn, Frank van Oort, and Gert-Jan Linders. 2009. Khadaroo, Jameel, and Boopen Seetanah. 2007b. “Transport
“On the Specification of the Gravity Model of Trade: Zeros, Infrastructure and Tourism Development.” Annals of Tourism
Excess Zeros and Zero-Inflated Estimation.” Spatial Economic Research 34 (4): 1021–32.
Analysis 4 (2): 167–90. Khadaroo, Jameel, and Boopen Seetanah. 2008. “The Role of
Cao, Jing, Xiaoyue Cathy Liu, Yinhai Wang, and Qingquan Li. Transport Infrastructure in International Tourism Development:
2013. “Accessibility Impacts of China’s High-Speed Rail A Gravity Model Approach.” Tourism Management 29 (5):
Network.” Journal of Transport Geography 28:12–21. 831–40.
Chen, Zhenhua, and Kingsley E. Haynes. 2015. “Impact of High- Li, Dong, and Yang Yang. 2017. “GIS Monitoring of Traveler
Speed Rail on International Tourism Demand in China.” Flows Based on Big Data.” In Analytics in Smart Tourism
Applied Economics Letters 22 (1): 57–60. Design: Concepts and Methods, edited by Z. Xiang and D. R.
CNTA (China National Tourism Administration). 2014. Tourism Fesenmaier. Cham: Springer.
Statistical Report of National Day Golden Week, 2014. Beijing: Li, Xiang. 2016. “Emerging-Market Research: New Bottles AND
China National Tourism Administration. New Wine?” Journal of Travel Research 55 (4): 419–26.
Delaplace, Marie, Francesca Pagliara, Julie Perrin, and Samuel Liu, Yu, Zhengwei Sui, Chaogui Kang, and Yong Gao. 2014.
Mermet. 2014. “Can High Speed Rail Foster the Choice of “Uncovering Patterns of Inter-Urban Trip and Spatial
Destination for Tourism Purpose?” Procedia—Social and Interaction from Social Media Check-In Data.” PLoS One 9
Behavioral Sciences 111:166–75. (1): e86026.
40 Journal of Travel Research 58(1)

Long, J. Scott, and Jeremy Freese. 2006. Regression Models for Schmidheiny, Kurt, and Marius Brülhart. 2011. “On the Equivalence
Categorical and Limited Dependent Variables Using Stata. of Location Choice Models: Conditional Logit, Nested Logit
College Station, TX: Stata Press. and Poisson.” Journal of Urban Economics 69 (2): 214–22.
Marrocu, Emanuela, and Raffaele Paci. 2013. “Different Tourists Seetanah, Boopen, and Jameel Khadaroo. 2009. “An Analysis
to Different Destinations. Evidence from Spatial Interaction of the Relationship between Transport Capital and Tourism
Models.” Tourism Management 39:71–83. Development in a Dynamic Framework.” Tourism Economics
Martín, Juan Carlos, and Gustavo Nombela. 2007. “Microeconomic 15 (4): 785–802.
Impacts of Investments in High Speed Trains in Spain.” The Sen, A., and T. E. Smith. 1995. Gravity Models of Spatial Interaction
Annals of Regional Science 41 (3): 715–33. Behavior. New York: Springer.
Masson, Sophie, and Romain Petiot. 2009. “Can the High Speed Sheth, Jagdish N. 2011. “Impact of Emerging Markets on
Rail Reinforce Tourism Attractiveness? The Case of the Marketing: Rethinking Existing Perspectives and Practices.”
High Speed Rail between Perpignan (France) and Barcelona Journal of Marketing 75 (4): 166–82.
(Spain).” Technovation 29 (9): 611–17. Smith, Stephen L. J., and Philip Feifan Xie. 2003. “Estimating the
McKercher, Bob. 1998. “The Effect of Market Access on Destination Distance Equivalence of the Canada–U.S. Border on U.S.-to-
Choice.” Journal of Travel Research 37 (1): 39–47. Canada Visitor Flows.” Journal of Travel Research 42 (2):
Morley, Clive, Jaume Rosselló, and Maria Santana-Gallego. 2014. 191–94.
“Gravity Models for Tourism Demand: Theory and Use.” Starkie, David. 2012. “European Airports and Airlines: Evolving
Annals of Tourism Research 48:1–10. Relationships and the Regulatory Implications.” Journal of Air
Mukkala, Kirsi, and Hannu Tervo. 2013. “Air Transportation and Transport Management 21:40–49.
Regional Growth: Which Way Does the Causality Run?” Su, Ming Ming, and Geoffrey Wall. 2009. “The Qinghai–Tibet
Environment and Planning A 45 (6): 1508–20. Railway and Tibetan Tourism: Travelers’ Perspectives.”
National Bureau of Statistics of China. 2015. China Statistical Tourism Management 30 (5): 650–57.
Yearbook 2015. Beijing: China Statistics Press. Taniguchi, Mamoru. 1992. “High Speed Rail in Japan: A Review
Nerhagen, Lena. 2003. “Travel Mode Choice: Effects of Previous and Evaluation of the Shinkansen Train.” UCTC No. 103.
Experience on Choice Behaviour and Valuation.” Tourism Berkeley: University of California Transportation Center.
Economics 9 (1): 5–30. Thrane, Christer. 2015. “Examining Tourists’ Long-Distance
Nicholls, Sarah, Bas Amelung, and Jillian Student. 2017. “Agent- Transportation Mode Choices Using a Multinomial Logit
Based Modeling a Powerful Tool for Tourism Researchers.” Regression Model.” Tourism Management Perspectives
Journal of Travel Research 56 (1): 3–15. 15:115–21.
Pagliara, Francesca, Andrea La Pietra, Juan Gomez, and José Tveteras, Sigbjorn, and Kristin H. Roll. 2014. “Non-stop Flights
Manuel Vassallo. 2015. “High Speed Rail and the Tourism and Tourist Arrivals.” Tourism Economics 20 (1): 5–20.
Market: Evidence from the Madrid Case Study.” Transport van Goeverden, Cees D. 2009. “Explaining Factors for Train Use
Policy 37:187–94. in European Long-Distance Travel.” Tourism and Hospitality
Pagliara, Francesca, José Vassallo, and Concepción Román. 2012. Planning & Development 6 (1): 21–37.
“High-Speed Rail versus Air Transportation.” Transportation Wang, Degen, Jia Qian, Tian Chen, Meifeng Zhao, and Yun Zhang.
Research Record 2289:10–17. 2014. “Influence of the High-Speed Rail on the Spatial Pattern
Pan, Bing, and Yang Yang. 2016. “Monitoring and Forecasting of Regional Tourism—Taken Beijing–Shanghai High-Speed
Tourist Activities with Big Data.” In Management Science in Rail of China as Example.” Asia Pacific Journal of Tourism
Hospitality and Tourism, edited by M. Uysal, Z. Schwartz, and Research 19 (8): 890–912.
E. Sirakaya-Turk. Waretown, NJ: Apple Academic Press. Wang, Xin, Songshan Huang, Tongqian Zou, and Hui Yan. 2012.
Patuelli, Roberto, Maurizio Mussoni, and Guido Candela. 2013. “Effects of the High Speed Rail Network on China’s Regional
“The Effects of World Heritage Sites on Domestic Tourism: A Tourism Development.” Tourism Management Perspectives
Spatial Interaction Model for Italy.” Journal of Geographical 1:34–38.
Systems 15 (3): 369–402. Wu, Bihu, Lan Xue, Alastair M. Morrison, and Xi Y. Leung. 2012.
Peng, Bo, Haiyan Song, Geoffrey I. Crouch, and Stephen F. Witt. “Frame Analysis on Golden Week Policy Reform in China.”
2015. “A Meta-analysis of International Tourism Demand Annals of Tourism Research 39 (2): 842–62.
Elasticities.” Journal of Travel Research 54 (5): 611–33. Wu, Wenjie, Jianghao Wang, and Tianshi Dai. 2016. “The
Prideaux, Bruce. 2000. “The Role of the Transport System in Geography of Cultural Ties and Human Mobility: Big Data
Destination Development.” Tourism Management 21 (1): 53– in Urban Contexts.” Annals of the American Association of
63. Geographers 106 (3): 612–30.
Pritchard, Mark P., and Mark E. Havitz. 2006. “Destination Yan, York Qi, and Hanqin Qiu Zhang. 2010. “The Determinants of
Appraisal: An Analysis of Critical Incidents.” Annals of the 1999 and 2007 Chinese Golden Holiday System: A Content
Tourism Research 33 (1): 25–46. Analysis of Official Documentation.” Tourism Management
Qin, Yu. 2016. “China’s Transport Infrastructure Investment: Past, 31 (6): 881–90.
Present, and Future.” Asian Economic Policy Review 11 (2): Yan, York Qi, Hanqin Qiu Zhang, and Ben Haobin Ye. 2014.
199–217. “Assessing the Impacts of the High-Speed Train on Tourism
Rey, Belén, Rafael L. Myro, and Asun Galera. 2011. “Effect of Demand in China.” Tourism Economics 20 (1): 157–69.
Low-Cost Airlines on Tourism in Spain. A Dynamic Panel Yang, Chih-Hai, Hui-Lin Lin, and Chia-Chun Han. 2010. “Analysis
Data Model.” Journal of Air Transport Management 17 (3): of International Tourist Arrivals in China: The Role of World
163–67. Heritage Sites.” Tourism Management 31 (6): 827–37.
Yang et al. 41

Yang, Yang, Timothy Fik, and Jie Zhang. 2013. “Modeling Author Biographies
Sequential Tourist Flows: Where Is the Next Destination?” Yang Yang, PhD, is assistant professor in Department of Tourism
Annals of Tourism Research 43:297–320. and Hospitality Management at Temple University. His areas of
Yang, Yang, and Kevin K. F. Wong. 2012. “A Spatial Econometric research interest include tourism demand analysis and location
Approach to Model Spillover Effects in Tourism Flows.” analysis in the hospitality and tourism industry.
Journal of Travel Research 51 (6): 768–78.
Yang, Yang, and Kevin K. F. Wong. 2013. “Spatial Distribution Dong Li, PhD, is deputy director of the Innovation Center for
of Tourist Flows to China’s Cities.” Tourism Geographies 15 Technology at Beijing Tsinghua Tongheng Urban Planning &
(2): 338–63. Design Institute. His areas of research interest include urban plan-
Zhang, Qiong, Hangjun Yang, Qiang Wang, and Anming ning and spatial data analysis.
Zhang. 2014. “Market Power and Its Determinants in the Xiang (Robert) Li, PhD, is professor in Department of Tourism
Chinese Airline Industry.” Transportation Research Part and Hospitality Management at Temple University. His research
A 64:1–13. focuses on destination marketing and tourist behavior.