The spatial heterogeneity and dynamics of tourism-flow spillover effect: The role of high-speed train in China

Abstract

The existing literature has not fully explored the spatial heterogeneity and dynamics of spillover effects of tourism-flow patterns, and this article makes contributions to addressing this issue. Using spatial autoregressive models and comparing the results for the non-high-speed train (HST) era (2002–2006) and those for the HST era (2011–2015) of China, this article finds that in the HST era, the spillover effect of tourism flows from neighboring regions has changed from positive to negative. Moreover, in the HST era, the cities with HST services have significantly positive effects on for a particular city, while those cities without HST have negative effects. These results reveal the spatial heterogeneity and dynamics of the tourism-flow spillover. After evaluating the variation of other factors which potentially contribute to the change of spillover patterns, this study identifies that the HST service plays an important role in such changes. Finally, this article uses geographically weighted regressions to explore the spatial heterogeneity in an elaborative way.

Keywords

dynamics high-speed train spatial autoregressive models spatial heterogeneity the spillover effect of tourism flow

Introduction

The tourism industry is closely rooted in the spatial mobility of tourists, which draws great attention from researchers to study the spillover effect of tourism flows between regions. However, the early literature on this topic fails to provide generalizable results for the spillover effects due to the limitations of research methods (Yang and Wong, 2012). Recently, with the introduction of spatial econometric techniques into tourism domain, stronger empirical results for explaining the nature of spillover effects have been provided in some researches, such as Yang and Wong (2012) and Deng and Hu (2019). It is valuable that these studies present an overall estimation of the spillover effect. Nevertheless, they may oversimplify the situation due to oversight of the spatial heterogeneity, as well as the dynamics of tourism-spillover patterns.

Specific to China, a few papers (Wang et al., 2012; Wang et al., 2014) highlight that the running of high-speed trains (HSTs) will fundamentally change tourism spatial structure within the regions influenced by the HST network. Indeed, with the introduction of HST in China, the spatial heterogeneity of tourism-flow spillover may be enhanced, possibly arising from the proposed asymmetric impacts of the new developed mode of transportation. For instance, according to Masson and Petiot (2009) and Hiramatsu (2018), the impacts of HST on regional tourism development may vary from city to city. The asymmetric impacts may extend to tourism-flow spillover and likely lead to more intensive spatial heterogeneity.

Further, the tourism-flow spillover pattern may have been altered when regions enters from the non-HST era into the HST era. However, the existing research has not captured such underlying dynamics. If the dynamic change is overlooked, biased results may be obtained for the tourism-flow spillover pattern, or the explanation for tourism-flow spillover may only be partly presented. Therefore, it is meaningful to employ advanced econometric models to explore the spillover patterns from a dynamic perspective.

This article aims to fill the abovementioned research gaps. To achieve this goal, we firstly investigate the nature of tourism-flow spillover effects in the HST era of China, represented by years 2011–2015. Specifically, we select a research area in eastern and central China where the geographic distribution of cities which had (or did not have) HST service was comparatively stable during years 2011–2015, and we estimate the spillover effects from the neighboring cities. Then, we divide the research area into two parts, the HST zone and the non-HST zone, and subsequently evaluate the spillover effect of tourism flow from each zone, respectively, whereby the spatial heterogeneity is unveiled. Further, we apply these spatial models to a data set including the same sample regions in the non-HST era, represented by years 2002–2006. Comparing the regression results between two time periods, we fully demonstrate the dynamics of spillover effects, which are possibly attributed to the intensive use of HST service. Furthermore, we adopt geographically weighted regression (GWR) models to capture the spatial heterogeneity in an elaborative way.

To our knowledge, this study represents the first attempt to explore tourism-flow spillover patterns with the consideration of spatial heterogeneity, dynamics, as well as the role of HST, which we believe cannot be ignored when investigating the recent evolution of regional tourism spatial structure in China. Specifically, this contributes to the current literature as follows.

First, this study elaborately confirms the existence of spatial heterogeneity in terms of the tourism-spillover effect. Using an extended spatial autoregressive (SAR) model, this study confirms that there is a negative spillover effect coming from those cities without HSTs, and a positive one coming from those cities with HSTs. Furthermore, the spatial heterogeneity is elaborately revealed through the estimation results of GWR models.

Secondly, the study initially reveals the dynamics of the tourism-flow spillover effects. The existing literature has demonstrated a positive spillover effect of tourism flows from neighboring regions (Yang and Wong, 2012). However, since their sampled areas/regions are not well equipped with intensive HST networks, they fail to reveal the dynamics of spillover patterns under the influence of HST service. Contrarily, this study finds that the spillover effects of tourism flows from proximate regions had changed from positive to negative in the HST era, which is a striking evidence on how the spillover pattern is altered with the introduction of a great transportation network into the investigated regions.

Thirdly, this study reveals that the HST service plays a key role in the dynamics and spatial heterogeneity of spillover effects. After analyzing the variations in other variables, such as other transportation facilities (including train, highway, and air transportation) and tourism attractions, which have the potential to change the spillover pattern, this study proposes that the observed change of tourism-flow spillover patterns may be attributable to the intensive use of HST service. Moreover, in this study, the negative spillover effect from the regions without HST service and the positive effect from those with HST service is termed the polarization effect of HST, and a theoretical explanation is presented for the polarization effect.

Literature review

The spillover effect of tourism flow

When analyzing the spatial structure of tourism, researchers should not bypass the issue of tourism-flow spillover effects since the nature of spillover effects of tourism flows profoundly determines the spatial structure of tourism for a group of regions. According to Capone and Boix (2008) and Capello (2009), spatial spillover effects represent one kind of economic externalities, which produce noncompensated or indirect impacts for a receiver situated nearby. The word ‘spatial’ is used to emphasize that spatial proximity is a primary condition for the economic externality to occur. In tourism, the spatial spillover effect refers to a phenomenon that tourism activities in one region are closely associated with those in the neighboring regions, or an unintentional effect a region’s tourism industry will receive from the tourism activities in the nearby regions (Yang and Wong, 2012; Yang and Fik, 2014; Yang et al., 2017). Actually, there are various kinds of spatial spillover effects that have been investigated in tourism domain, such as the spillover effect of tourism economic growth (Yang and Fik, 2014) and the spillover effect of tourism employment (Capone and Boix, 2008). Considering the research purpose, in this section we focus on literature related to tourism-flow spillover effects.

The spillover effects of tourism flows denote the impact of tourism flows occurring in neighboring regions onto the tourism flow in a particular region (Yang and Fik, 2014; Yang et al., 2017), which to a certain extent reflects spatial interactions of tourism activities among regions. A positive spillover effect reflects that a group of regions support or complement each other to draw tourists or share the same base of tourist resources; meanwhile, a negative effect reflects spatial competition, which means a group of regions may compete directly with one another to attract tourists. The mechanisms for a positive spillover effect, including productivity spillover, market access spillover, multi-destinations travel plans, etc., are depicted by Yang and Fik (2014), and those for a negative spillover effect are discussed by Patuelli et al. (2013) and Zhou et al. (2017).

Many empirical examinations have been carried out using simple tourism systems as context by only including a pair of regions in each case. For instances, Gooroochurn and Hanley (2005) examine the spillover effect of tourism flow between the Republic of Ireland and Northern England and find that the tourism demand of long-haul tourists in one region positively influences that of another region. Hoti et al. (2007) find that the tourism growth rate of Cyprus has a positive spillover effect on Malta. Shareef and McAleer (2008) model the international tourism demand of two island countries—Maldives and Seychelles, and conclude that a significant spillover effect of tourist arrivals exists among these island countries. Balli and Tsui (2016) model the volatility of international tourist arrivals for Australia and New Zealand and demonstrate a significant spillover effect of international tourism demand between the two countries.

By contrast, a few examinations consider a generalized tourism system which includes many regions (greatly more than two regions) to unveil tourism-flow spillover effects. Marrocu and Paci (2013) consider Italy, consisting of 107 provinces, as a case, and demonstrate that the tourist flows towards a certain province are enhanced by flows to the neighboring provinces, that is, a positive cross-province spillover exists. Yang and Wong (2012) consider a greatly larger tourism system, including 341 cities in China, and reveal a positive spillover effect of tourist flows for the domestic and inbound tourism markets, respectively.

However, the current research regarding tourism-flow spillovers is confined to global level analyses, hence missing the potential spatial heterogeneity. It should be noted that, while much literature validates the spatial heterogeneity in tourism development (Deller, 2010; Losada et al., 2019; Yang and Fik, 2014), this heterogeneity in extant literature is not related to tourism-flow spillover. Additionally, these studies have not investigated tourism-flow spillover from a dynamic perspective, thereby failing to capture the dynamic changes of spillover patterns. We suggest that this dynamic becomes more notable when an advanced transportation system is intensively put into use within a massive area such as eastern and central China (the sample area of this study).

Therefore, it is of great importance to investigate the dynamic changes and spatial heterogeneity of the spillover patterns as this study does. Without such investigation, we cannot say we have a holistic understanding on tourism-flow spillover phenomenon.

The heterogenous effects of HSTs

There is much literature attempting to quantify the various impacts of HST service, using different countries as research cases (Albalate et al., 2017; Chen and Haynes, 2015; Zhou and Li, 2018b), among which a few researchers have noticed the heterogenous impacts of HST, which we suggest may be a source for the spatial heterogeneity of tourism-flow spillover effects.

Assuming that the accessibility is always synonymous with the intensive spatial competition among different destinations, Masson and Petiot (2009) propose that different destinations may face different outcomes resulting from the running of HSTs. Some literature pays attention to tunnel effect or corridor effect of HST service. A tunnel effect refers to improving the accessibility between the connected cities, while isolating the space between them and letting these spaces compete with each other to draw tourists (Gutiérrez Puebla, 2005). Similarly, a corridor effect denotes that tourism industries in the cities connected by an HST network will boom at the expense of the decline of those in the cities bypassed by the HST network (Shaw et al., 2014; Wang et al., 2018). Recently, Hiramatsu (2018) also suggests the asymmetric or unequal impact of the HST service among different regions linked by HST service.

Indeed, according to the potential asymmetric effects of HSTs on different regions, the tourism-flow patterns in the HST era may be spatially heterogenous. However, there is no study that makes efforts to demonstrate the heterogeneity systematically. For this reason, this study attempts to conduct empirical study through spatial econometric modeling and GWR to offer an in-depth understanding of the spatial variation of tourism-flow patterns.

Methodology

Samples and data

We use administrative cities in eastern and central China as sample cities, considering that eastern and central China boasts the most intensive network of HSTs in the HST era, represented by years 2011–2015 in this study due to data availability. As our analysis must be conducted in a geographical area that should have a stable geographical distribution of the cities which had (or did not have) HST service, we exclude three provinces including Guangdong, Shanxi, and Hunan. We also exclude Fujian province because of missing data. Finally, 10 provinces and 3 municipalities in eastern and central China serve as the research area, including Jilin, Liaoning, Hebei, Shandong, Jiangsu, Zhejiang, Anhui, Jiangxi, Henan, Hubei, Beijing, Tianjin, and Shanghai, as Figure 1 indicates. In total, this research area includes 133 cities as sample cities.

Figure 1.

Research area.

In Figure 2, we divide the research area into two parts. Part 1 is the shadow areas indicating the HST zone, including 82 cities which had HST service most of the years between 2011 and 2015. Specifically, 73 cities had HST service all the time and 9 cities had HST for at least 3 years between 2011 and 2015. Part 2 is the blank areas indicating the non-HST zone, including 51 cities that did not have HST service most of the years during 2011–2015. Among them, 46 cities didn’t have HST service all the years and 5 cities had HST service for only 1 or 2 years.

Figure 2.

The high-speed train zone and non-high-speed train zone.

Spatial autoregressive model

Generally, spatial autoregressive (SAR) models are employed to capture the spillover effects of tourism flows (Anselin, 1988; Yang and Wong, 2012). The traditional SAR model is specified as follows:

Y = ρ W Y + X β + ε,

where Y is the dependent variable. X represents a set of independent variables, W is n by n spatial weighted matrices whose elements measure the spatial relationships among a pair of regions, and $ε$ denotes the error term (Anselin, 1988).

Following this widely acknowledged model specification, we establish a traditional SAR model as equation (2) to capture the spillover effect of tourism flows coming from the neighboring cities that share a common boundary with a particular city.

\begin{array}{l} ln A R R I V A L_{i t} = α_{0} + ρ_{n} W_{n} ln A R R I V A L_{i t} + α_{1} ln T R A I N + α_{2} ln A I R + α_{3} ln H I G H W A Y \\ + α_{4} ln H O T E L + α_{5} ln T A + α_{6} ln G D P + α_{7} ln F D I + D u m m y_{t} + μ_{i} + ε_{i t} . \end{array}

In equation (2), the dependent variable is ARRIVAL, the number of tourists received by a sample city in one year, which measures tourism flows to a certain city. In the matrix W_n, $W_{i j}$ is assigned to be 1 if city j shares a common boundary with city i, otherwise it is 0. Accordingly, $ρ_{n}$ reflects the spillover effect of tourism flows coming from neighboring cities sharing a common boundary with city i.

According to the gravity model, there are three sets of variables that determine the tourist flow from a tourist-origin toward a destination (Crampon, 1966; Khadaroo and Seetanah, 2007, 2008). The first set of variables comes from the characteristics of the destination. Specifically, the following variables regarding the tourism destination are controlled. HOTEL indicates the capacity of service infrastructure in a city which may be employed in tourism industry (Yang and Wong, 2012). TA (tourism attraction) represents the overall quality of tourism resources and is one key determinant of tourism attractiveness for a city (Ellerbrock and Hite, 1980; Rosentraub and Joo, 2009; Yang and Fik, 2014). GDP reflects the economic resources that a city may exploit to develop its tourism economy (Marrocu and Paci, 2013; Zhang and Jensen, 2007). FDI reflects the connection of a city with foreign countries and indicates a city’s attractiveness to foreign tourists (Khan et al., 2005; Kulendran and Wilson, 2000). Additionally, $D u m m y_{t}$ is a set of dummy variables to control year variation which also has a role on the destination side. The second set of variables is related to distance factors between tourist-origins and the destination, which in this study are presented by three transportation variables (Lee, 2015). TRAIN reflects the accessibility to a city by conventional train transport (Zhou and Li, 2018a, 2018b). AIR measures the accessibility by air transportation (Yang and Wong, 2012). HIGHWAY measures the accessibility to a city by road transport (Khadaroo and Seetanah, 2008). The last set of variables should come from the side of the tourist-origin, which unfortunately cannot be controlled since the data of tourist sources are unavailable in China.

Further, we use an extended SAR model that incorporates two spatially lagged dependent variables to capture the spillover effects from the HST zone and the non-HST zone, respectively. The general form of the extended SAR model is specified as follows:

Y = ρ_{1} W_{1} Y + ρ_{2} W_{2} Y + X β + ε .

Equation (3) can be transformed into the reduced form shown in equation (4).

Y = {(I_{n} - ρ_{1} W_{1} - ρ_{2} W_{2})}^{- 1} (X β + ε) .

Following Le Sage and Pace (2009), this study conducts maximum-likelihood estimation to address the endogeneity problem resulting from the inclusion of two spatially lagged dependent variables, and hence acquires consistent estimates of the regression parameters. According to Lacombe (2004), the likelihood function is presented as equation (5).

L (β, σ, ρ_{1}, ρ_{2}; Y, X) = {(2 π σ^{2})}^{- n/2} | I_{n} - ρ_{1} W_{1} - ρ_{2} W_{2} | exp (- \frac{1}{2 π σ^{2}} e' e) .

In equation (5), e is equal to $(I_{n} - ρ_{1} W_{1} - ρ_{2} W_{2}) Y - X β$ .

Specifically, the extended model of this study is established as equation (6).

\begin{array}{l} ln A R R I V A L_{i t} = α_{0} + ρ_{1} W_{1} ln A R R I V A L_{i t} + ρ_{2} W_{2} ln A R R I V A L_{i t} + α_{1} ln T R A I N \\ + α_{2} ln A I R + α_{3} ln H I G H W A Y + α_{4} ln H O T E L + α_{5} ln T A + α_{6} ln G D P + α_{7} ln F D I \\ + D u m m y_{t} + μ_{i} + ε_{i t} \end{array}

In equation (6), $W_{1} ln A R R I V A L_{i t}$ and $W_{2} ln A R R I V A L_{i t}$ are employed to capture the spillover effects of tourism flows of the cities within the HST zone and those within the non-HST zone, respectively. Specifically, W₁ and W₂ are established as follows: Firstly, we construct a distance-based matrix W where $W_{i j}$ is equal to $1 / d_{i j}^{2}$ (Getis and Aldstadt, 2004; Lee and Jang, 2013). Hereinto, $d_{i j}$ is the arc-distance between city i and j, which is provided by GeoDa 1.12 software. Secondly, we replace the values in column j of the matrix with 0 if city j was included in the non-HST zone, and hence we get the matrix W₁. Similarly, we replace the value in column j of matrix W with 0 if city j was included in the HST zone, and then we obtain the matrix W₂. Clearly, $ρ_{1}, ρ_{2}$ reflect the spillover effect of tourism flows from the HST zone and from the non-HST zone, respectively. As a standard operation, all matrices are row-normalized prior to estimation (Lee, 2014).

All the variables, except $D u m m y_{t}$ , are in logarithmic form. To avoid a log of zero, it is set as 0.5 when the value of a variable is zero. Table 1 provides detailed definitions and data sources for all variables used in this study.

Table 1.

The detailed definitions of variables.

Variables	Definitions	Data sources
ARRIVAL_it	The number of tourist arrivals to city i at year t (in 1000s).	China Economic and Industrial Database (CEIC)
TRAIN_it	The number of conventional trains running per day in city i at year t.	National Railway Timetables, published by the Ministry of Railways of China annually
AIR_it	The number of flights available in city i at year t.	China Economic and Industrial Database (CEIC)
HIGHWAY_it	The total length of highway in city i at year t (in km).	China Economic and Industrial Database (CEIC)
TA_it	The number of tourist attractions in city i at year t. (In database for 2011–2015, tourist attractions include 5A and 4A scenic spots. In the database for 2002–2006, 4A scenic spots and national parks are included as tourist attractions since 5A attractions were not inscribed in this time period.)	The official website of the China National Tourism Administration and Ministry of Housing and Urban-Rural Development.
HOTEL_it	The number of starred hotels in city i at year t.	China Economic and Industrial Database (CEIC)
GDP_it	The gross domestic product (in 1,000,000,000 CNY) in city i at year t.	China Economic and Industrial Database (CEIC)
FDI_it	The foreign direct investment (in 1,000,000 USD) of city i at year t.	China Economic and Industrial Database (CEIC)

The descriptive statistics for these variables are shown in Table 2. Please take note that in this study, the HST era is represented by years 2011–2015 and the non-HST era is represented by years 2002–2006.

Table 2.

The descriptive statistics for variables.

Variable	Obs.	Mean	Standard deviation	Min.	Max.
In the HST era (2011–2015)
ARRIVAL	665	33,727.25	37,752.05	1104.15	283,695.8
TRAIN	665	74.22707	55.89317	0	345
HIGHWAY	665	12,654.64	6527.487	1555	38,004.28
AIR	665	25,354.59	82,055.52	0	705,774
HOTEL	665	50.57444	61.69733	1	614
TA	665	8.309774	9.163852	0	79
GDP	665	276.356	326.0968	37.25	2512.345
FDI	665	1272.62	2577.618	13.87	21134.44
In the non-HST era (2002–2006)
ARRIVAL	665	8110.884	14,393.83	353	135,902.9
TRAIN	665	62.18496	57.99961	0	361
HIGHWAY	665	6217.345	4297.336	647	36,166
AIR	665	10,235.42	41,596.98	0	409,620
HOTEL	665	41.3203	66.75123	2	700
TA	665	2.840602	4.079242	0	39
GDP	665	70.25787	92.12775	6.480074	952.9193
FDI	665	310.6098	838.8504	1.169672	6258.055

GWR model

According to Brunsdon et al. (1996), a general GWR model is established as equation (7), where $X_{i k}$ is the k-th independent variable. In contrast to a traditional regression model, the parameters which will be estimated in a GWR model are the function of a location denoted by the longitude ( $u_{i}$ ) and latitude ( $v_{i}$ ) of a space sample i. Hence, the parameters estimated by GWR models are localized for each region, which in traditional models are assumed to be constants among all regions.

Y_{i} = α (u_{i}, v_{i}) + \sum_{k} β_{k} (u_{i}, v_{i}) X_{i k} + ε_{i} .

Equation (8) reveals the estimation of the local parameters $\hat{β} (u_{i}, v_{i})$ , where $W (u_{i}, v_{i})$ is an n × n spatial weighted matrix, which plays a crucial role in the process of parameter estimation.

\hat{β} (u_{i}, v_{i}) = {[X^{T} W (u_{i}, v_{i}) X]}^{- 1} X^{T} W (u_{i}, v_{i}) Y .

The most popular kernel technique to define $W (u_{i}, v_{i})$ is the Gaussian distance decay-based function, which is shown in equation (9) (Fotheringham et al., 2003).

W (u_{i}, v_{i}) = exp (- \frac{d_{i j}^{2}}{h_{}^{2}}) .

Therein, $d_{i j}$ is the Euclidean distance of location i and j, meanwhile h is the bandwidth to calibrate the influence of distance-decay. Accordingly, the cross-validation (CV) procedure is conducted to find out an optimal bandwidth by minimizing the CV value with the golden-section search technique (Fotheringham et al., 2003). The CV value is specified as equation (10) where ${\hat{y}}_{\neq i} (h)$ denotes the fitted value of y with ${\hat{y}}_{i} (h)$ excluded.

C V = \sum_{i = 1}^{n} (y_{i} - {\hat{y}}_{\neq i} (h))^{2} .

Employing different log-likelihood functions for observations in different locations, we can obtain a set of estimates in the GWR model for different regions.

Results

Main results for the HST era

Table 3 reports the study results based on the data of sample cities in the HST era (2011–2015). According to the result of Hausman test, all estimation results in this study are estimated by the fixed effect technique (Hausman, 1978). Model (1) reveals the results of equation (2); model (2) and model (3) include the spatially lagged dependent variables ( $W_{1} ln A R R I V A L_{i t}$ and $W_{2} ln A R R I V A L_{i t}$ ), respectively, and model (4) includes both as equation (6) does.

The coefficient of the spatial lag term ( $ρ_{n}$ ) in model (1) is statistically significant, which suggests a negative spatial spillover effect of tourism flows from neighboring cities. This result presents a tourism-flow pattern that differs from Yang and Wong’s (2012) study. Switching attention to the other models, we can find that the coefficients of $W_{1} ln A R R I V A L_{i t}$ ( $ρ_{1}$ ) in model (2) and model (4) are significantly positive, however, those of $W_{2} ln A R R I V A L_{i t}$ ( $ρ_{2}$ ) in model (3) and (4) are significantly negative, which reveal that in the HST era, the HST zone and non-HST zone exert totally different spatial spillover effects for a particular city whereby the spatial heterogeneity emerges. Specifically, this demonstrates that the cities with high-speed services still exert positive effects to a given city, which is consistent with the past study, while those cities without HSTs induce negative effects, which is revealed for the first time by this study. Logically, we deduce that the global negative spillover impacts observed in model (1) mainly come from the non-HST zone, which confirms the value of the heterogeneity analysis.

According to Akaike’s information criterion (AIC), Bayesian information criterion (BIC), and the maximum likelihood value, model (4) shows the best model fit among models (2)–(4) (Akaike, 1974; Burnham and Anderson, 2004; Raftery, 1995). According to model (4), the coefficient of $W_{1} ln A R R I V A L_{i t}$ implies that a 100% increase in the arrival of the cities located in the HST zone will stimulate a 26.6% increase of tourism flow for a particular city. To contrast, the coefficient of $W_{2} ln A R R I V A L_{i t}$ suggests that a 100% increase in cities located in the non-HST zone will lead to a 48.7% decline of the tourism flow. This result indicates that there exists heterogenous spillover effects coming from these two zones. The work of Gooroochurn and Hanley (2005) and Hoti et al. (2007) demonstrate asymmetric effects of tourism-flow spillover, that is, the spillover effect from region A to region B is stronger than that from B to A. This study also demonstrates stronger heterogeneity by revealing a different sign of spillover effects coming from different regions.

Except the variables on highway and traditional train transportation infrastructure, all controlled variables exert significantly positive effects on regional tourism flows as expected. In detail, the coefficients of lnHOTEL and lnTA reveal a significant contribution of service infrastructure and tourism resources on regional tourism flow, in line with Ellerbrock and Hite (1980), Rosentraub and Joo (2009), Yang and Wong (2012), and Yang and Fik (2014). Moreover, the positive and significant impacts of lnGDP and lnFDI are also revealed, consistent with the findings of Kulendran and Wilson (2000), Khan et al. (2005), Zhang et al. (2011), and Marrocu and Paci (2013). Additionally, AIR also has a positive and significant effect on tourism flow, which corroborates the findings of Yang and Wong (2012).

Table 3.

The results for the high-speed train era (2011–2015).

	(1)	(2)	(3)	(4)
Variables
lnTRAIN	−0.00327	−0.00435	0.00688	0.00956
lnAIR	0.0122	0.0151*	0.0121	0.0125*
lnHIGHWAY	−0.139	−0.131	−0.133	−0.132
lnHOTEL	0.212***	0.186***	0.221***	0.213***
lnTA	0.0583***	0.0502**	0.0443**	0.0407*
lnGDP	0.404***	0.280*	0.411***	0.333**
lnFDI	0.0904***	0.0693***	0.0958***	0.0845***
$D u m m y_{2012}$	0.126***	0.0812***	0.157***	0.129***
$D u m m y_{2013}$	0.263***	0.172***	0.342***	0.287***
$D u m m y_{2014}$	0.387***	0.258***	0.531***	0.459***
$D u m m y_{2015}$	0.532***	0.363***	0.706***	0.614***
Spatial coefficients
ρ_n	−0.217***
ρ₁		0.170*		0.266***
ρ₂			−0.460***	−0.487***
N of Obs.	665	665	665	665
AIC	−457.2	−447.5	−473.5	−478.4
BIC	−398.7	−389.0	−415.0	−415.4
Maximum likelihood value	241.6	236.8	249.8	253.2
Hausman test	23.28**	21.13**	23.34**	23.27**

Note: AIC: Akaike’s information criterion; BIC: Bayesian information criterion; N of Obs.: number of observations.

***, ** and * represent the significance at the 1%, 5%, and 10% statistical levels, respectively.

Main results for the non-HST era

With the purpose of revealing the dynamics of the spillover effect, we apply the models in Table 3 to the data in the non-HST era (years 2002–2006), and Table 4 reports the results. Models in Table 4 clearly present a distinctly different story regarding the spillover effects of tourism flow. The positive coefficients of $ρ_{n}$ (0.391) indicate a positive spillover effect from neighboring cities, while the coefficients of $W_{1} ln A R R I V A L_{i t}$ and $W_{2} ln A R R I V A L_{i t}$ are consistently positive, suggesting that as for tourism-flow spillover effects, the marked HST zone and the marked non-HST zone in the time before the availability of HST service are homogeneous.

Such positive spillover effects are consistent with the findings of Yang and Wong (2012). It is notable that during the research period in Yang and Wong (2012), the HST network was not developed for the sample regions. In other words, the spatial heterogeneity observed in Table 3 can be detected after HST transportation was intensively put into use in the research area. The comparison between the results in Table 3 and those in Table 4 reveals well the dynamics of the spillover effect.

Furthermore, it seems slightly counterintuitive that when the HST network was intensely put into operation in the research area, the positive impacts of HIGHWAY become insignificant. A possible reason for this may be that highway transport has been developed sufficiently in years 2011–2015, and hence it no longer is a positive determinant of regional tourism arrivals in this time period. The insignificant impact of the conventional train is validated both for the HST era and for the non-HST era, which is consistent with Zhou and Li (2018a, 2018b).

Table 4.

The results for the non-high-speed train era (2002–2006).

	(5)	(6)	(7)	(8)
Variables
lnTRAIN	0.00123	−0.00230	0.00266	−0.000221
lnAIR	0.0235***	0.0237***	0.0218**	0.0231***
lnHIGHWAY	0.102**	0.104**	0.120**	0.110**
lnHOTEL	0.0832**	0.0908**	0.0964***	0.0875**
lnTA	0.0274**	0.0256**	0.0222*	0.0245*
lnGDP	0.204**	0.216**	0.252***	0.221**
lnFDI	−0.0160	−0.0171	−0.0188	−0.0174
$D u m m y_{2003}$	−0.0193	−0.0382	−0.0328	−0.0387
$D u m m y_{2004}$	0.0849**	0.0194	0.0243	−0.0128
$D u m m y_{2005}$	0.148***	0.0427	0.0702	−0.00291
$D u m m y_{2006}$	0.164**	0.0311	0.0193	−0.0558
Spatial coefficients
ρ_n	0.391***
ρ₁		0.555***		0.408***
ρ₂			0.504***	0.256***
N of Obs.	665	665	665	665
AIC	−512.1	−501.1	−488.1	−506.7
BIC	−453.6	−442.6	−429.6	−443.7
Maximum likelihood value	269.1	263.5	257.1	267.3
Hausman test	29.21***	32.87***	28.22***	29.30***

Note: AIC: Akaike’s information criterion; BIC: Bayesian information criterion; N of Obs.: number of observations.

***, ** and * represent the significance at the 1%, 5% and 10% statistical levels, respectively.

Examining the variation of other variables potentially to change spillover pattern

The change of spillover patterns from the non-HST era to the HST era, indicated by Table 3 and 4, may be attributed to the variation of other factors. Therefore, a careful check on the variation of other variables, which have the potential to change the spillover patterns from a theoretical perspective, is necessary. Considering that a traffic interconnection is the prerequisite for all economic externalities among regions to come out, we first check the variation of transportation infrastructure, including traditional train, highway, and air transportation. Additionally, according to Zhou et al. (2017), tourism attractions, a key variable reflecting the tourism development level or the quality of tourism products for a region, have a significant role in determining the cross-regional tourism-spillover in China, thus we examine the variation of tourism attractions.

Specifically, to compare the differences between the HST zone and non-HST zone, t-tests are employed. We first calculate the variation amplitude of one variable between two time periods for the HST zone and that for the non-HST zone, respectively. Then, we use t-tests to investigate whether the variation amplitude of one variable in the HST zone and non-HST zone are significantly different from one other. If it is not significant, we suggest that the variable is not a factor that contributes to the change of the spillover pattern.

The detailed results are reported in Table 5, and the p-value suggests that there is no significant difference for the development of traditional train, highway transportation, and tourism attractions. By contrast, the increase of flights between these two zones is significantly different, and the HST zone appears to display a greater increase in air flights. However, this does not mean that air traffic plays an important role in changing the spillover patterns. Compared to the HST transportation, the transportation capacity of air traffic is extremely limited. Taking 2015 as an example, the total passenger throughput of airports in our research area was 488 million. Comparing with the enormous figure of tourism arrivals (4485.7 million) in 2015, the tourists travelling by air only account for 10.88% collectively. Furthermore, considering that a part of these 488 million air travelers may travel to tourism destinations outside the research area and a part of them may not travel for the purpose of tourism, the true proportion of tourists who travel by air transportation is expected to be smaller than 10.88%. Therefore, we cannot argue that the unbalanced development of air transportation capacity between the HST zone and the non-HST zone may induce the dramatic change of spillover patterns.

Table 5.

Variation of other variables during the non-high-speed train era and high-speed train era.

Variation	(1) For samples in HST zone	(2) For samples in non-HST zone	(1) – (2) difference	p-Value
The increase of mean number of traditional trains	12.6	11.1	1.5	0.647
The increase of mean number of flights	21,499.0	4861.4	16,637.6	0.028**
The increase of mean highway length (in kilometers)	6557.4	6244.2	313.2	0.657
The increase of mean number of tourism attractions	4.7	4.5	0.2	0.803

Note: We calculate a 5-year average value for each investigated variable in a time period, and then obtain the variation of this variable between the two time-periods.

** Significance at the 5% statistical level.

Addition to air flight number, we also consider the spatial distributions of new airports. There are a total of 15 cities among the 133 sample cities where new airports were put into operation after 2006. According to Figure 3, there are 5 airports added to the HST zone and 10 airports added to the non-HST zone, which also indicates an unbalanced development of air transportation. However, the transportation capability of these new airports is not relevant. On average, the number of flights operating in these new airports accounts for merely 3.3% of the total flights number in the entire research area during the time span of 2011–2015.

Figure 3.

The distribution of airports in the research area.

In summary, we find that for HST zone and non-HST zone, there is no significant difference regarding the development of traditional train, road transportation infrastructures, and tourism attractions between two time periods. Additionally, although the variation of air traffic is statistically significant, it may not affect the spillover pattern due to the limited transportation capacity. So we reject the possibility that there are other factors that may lead to the changes of tourism-flow patterns that this study has unveiled. Since the HST service is the most significant change between the two time periods and it undoubtedly altered the basic conditions for regional tourism flows, we propose that the change of spillover patterns has a root in the availability of HST service. To solidify this proposition, we further present a theoretical explanation in section “A theoretical explanation for the role of HSTs.”

Robust testing

To test whether the nature of spillover patterns depends on how the matrices are constructed, we establish a new matrix W_n as one robust test for equation (2), where $W_{i j}$ is assigned to be 1 if city j is one of the 5 nearest neighboring cities of i, otherwise, it is 0 (Yang and Wong, 2012). Also, we construct different spatially weighted matrices for $W_{1}, W_{2}$ in equation (6), where $W_{i j}$ is assigned to be $1 / d_{i j}$ (Getis and Aldstadt, 2004). Intrinsically, in the new matrices $W_{1}, W_{2}$ , the rate of the distance decay becomes slow or a city located far away has a stronger spatial relationship with a given city.

The results of the robust tests for the HST era are shown in Table 6.

Table 6.

The results of robust test for the high-speed train era (2011–2015).

	(9)	(10)	(11)	(12)
Variables
lnTRAIN	−0.00603	−0.00520	0.00649	0.01000
lnAIR	0.0122	0.0149*	0.0118	0.0119
lnHIGHWAY	−0.118	−0.137	−0.126	−0.141
lnHOTEL	0.207***	0.187***	0.225***	0.215***
lnTA	0.0546***	0.0499**	0.0445**	0.0359*
lnGDP	0.446***	0.275*	0.436***	0.280*
lnFDI	0.0961***	0.0715***	0.0969***	0.0835***
$D u m m y_{2012}$	0.136***	0.0593	0.253***	0.155***
$D u m m y_{2013}$	0.280***	0.131**	0.555***	0.380***
$D u m m y_{2014}$	0.416***	0.207**	0.876***	0.668***
$D u m m y_{2015}$	0.567***	0.300***	1.139***	0.876***
Spatial coefficients
ρ_n	−0.308***
ρ₁		0.293		0.879***
ρ₂			−1.139***	−1.356***
N of Obs.	665	665	665	665
AIC	−465	−446.5	−470.6	−481.1
BIC	−406.5	−388.0	−412.1	−418.1
Maximum likelihood value	245.5	236.3	248.3	254.5
Hausman test	23.55**	21.57**	20.27**	21.41**

Note: AIC: Akaike’s information criterion; BIC: Bayesian information criterion; N of Obs.: number of observations.

***, **, and * represent the significance at the 1%, 5%, and 10% statistical levels, respectively.

Similarly, the results of the robust tests for the non-HST era are shown in Table 7.

Table 7.

The results of robust test for non-high-speed train era (2002–2006).

	(13)	(14)	(15)	(16)
Variables
lnTRAIN	0.00167	−0.00195	0.000795	−0.00154
lnAIR	0.0221**	0.0233**	0.0232**	0.0233**
lnHIGHWAY	0.103**	0.107**	0.114**	0.108**
lnHOTEL	0.0885**	0.100***	0.105***	0.100***
lnTA	0.0246*	0.0248*	0.0235*	0.0246*
lnGDP	0.200**	0.249***	0.273***	0.252***
lnFDI	−0.0182	−0.0183	−0.0186	−0.0183
$D u m m y_{2003}$	−0.0271	−0.0457*	−0.0357	−0.0447*
$D u m m y_{2004}$	0.0768**	−0.0404	−0.0144	−0.0428
$D u m m y_{2005}$	0.142***	−0.0572	−0.00516	−0.0601
$D u m m y_{2006}$	0.154**	−0.103	−0.0885	−0.116
Spatial coefficients
ρ_n	0.393***
ρ₁		0.729***		0.630***
ρ₂			0.663***	0.112
N of Obs.	665	665	665	665
AIC	−509	−475.9	−468.5	−474.1
BIC	−450.5	−417.4	−410.0	−411.1
Maximum likelihood value	267.5	250.9	247.3	251.1
Hausman test	28.99***	27.91***	30.62***	25.35**

Note: AIC: Akaike’s information criterion; BIC: Bayesian information criterion; N of Obs.: number of observations.

***, ** and * represent the significance at the 1%, 5%, and 10% statistical levels, respectively.

According to the results of Table 6, the change of the neighboring matrix does not result in a significant change to the nature of the tourism-flow spillover effects from the neighbors, as model (9) shows. Additionally, when we set a slow rate for distance decay, the values of spatial coefficients $ρ_{1}, ρ_{2}$ become larger, as models (10)–(12) indicate. According to the values of AIC, BIC, and maximum likelihood values, models (10)–(12) seem to be more appropriate than models (2)–(4), which implies that when we construct a spatial matrix for the HST era, setting a slow rate on distance decay is better than setting a quick one; or a city located far away shows a stronger spatial association with a given city in the HST era. In other words, the availability of HST service strengthens the spatial interactions among regions.

Comparing the results reported in Table 7 with those in Table 4, we also conclude that the change of spatial matrices does not induce a difference on the nature of spillover patterns in the non-HST era, except for the coefficient $ρ_{2}$ in model (16). However, an insignificantly positive coefficient $ρ_{2}$ in model (16) does not undermine the core findings of this study because model (8) is clearly superior to model (16) by the values of AIC, BIC, and maximum likelihood. Contrarily, setting a quick rate on distance decay is better than setting a slow one in a “slow” era when HST service is unavailable for all sample cities.

A theoretical explanation for the role of HSTs

The estimation results indicate that for a given city in the HST era, the cities on the HST corridor are inclined to share the sources of tourists with it, pointing to the positive spillover effect from the HST zone. Meanwhile, those outside the corridor are inclined to compete against it in drawing tourists, pointing to the negative one from the non-HST zone. This spillover pattern is termed the polarization effect of HSTs in this study.

Herein, we present explanations for the positive spillover effect coming from the HST zone as follows. Firstly, we assume that a representative city R (the investigated object) is in the network of HST transportation. It is easy for R to benefit from tourism flows circulating in areas linked by the HST network, due to a significant space compression effect, or the integrated transportation between city R and other cities in the HST network (Givoni, 2006; Sun and Lin, 2018; Wang et al., 2014; Wang et al., 2018). Secondly, we assume that the representative city R is in the outer edges of the HST network. City R may also get a positive externality from tourist flows in the HST zone, because there is a large volume of tourists agglomerating on the HST zone, and these tourists can move quickly through the HST corridors and extend their travel to city R. Therefore, the cities in the HST network exert a positive spillover effect for city R, whether R is in the HST zone or not.

As for the negative spillover effect from the non-HST zone, we have a similar process of reasoning. We assume that city R is a node of the HST transportation network. The HST service is unavailable in the non-HST zone and hence the space compression effect cannot apply to this zone. Moreover, the volume of tourists visiting the non-HST zone is comparatively small, meanwhile the mobility of these tourists is also constrained. Hence, there is less possibility for a positive spillover to emerge from the non-HST zone for city R.

Then, we assume that city R is in the outer edges of the HST network. Usually, the main tourists are concentrated in the HST zone. When a tourist who stays on the network of the HST extends his/her travel to city R, it is highly possible that he or she has to cancel a visit to other cities in the non-HST zone due to time constraints (Zhou et al., 2017). As a result, the cities in the non-HST zone have to compete against each other to draw the tourists clustering on the HST network, and hence a negative spatial interdependency comes out. Additionally, when a corridor of HSTs is extended along a certain zone, the previously integrated space is actually divided and the cities in the non-HST zone are isolated from each other. Consequently, it becomes difficult that tourists may spillover from one city in the non-HST zone to other cities also in the non-HST zone.

This study concludes that as an overall estimation, the spillover effects of tourism flows from neighboring regions changes from positive to negative in the HST era. This conclusion is also related to the polarization effect. Generally, some neighbors of city R are in the HST network and some in the non-HST network. To city R, the neighboring cities in the HST network exert a positive spillover effect and those outside the HST network produce a negative one. These two effects counteract each other, and ultimately a negative outcome emerges as this study demonstrates.

The heterogeneity of tourism-spillover effects

To further identify the heterogeneity of spillover effects, the GWR model is introduced. Given that a GWR model is unable to control the significant time variation which has been demonstrated by SAR models in Table (3), we shorten the research time period and only use the 2014–2015 subset data to estimate the GWR model. The aforementioned CV technique is employed to obtain an optimal bandwidth for the construction of spatial weights as equation (9) indicates. The results show a very strong model fit with an R² value of 0.89.

Figures 4 to 6 map the coefficients of $ρ_{1}, ρ_{2}, ρ_{n}$ , respectively, enabling us to visualize the heterogeneity of spillover effects on a local level. A sample city is filled in with white when its coefficient is statistically insignificant.

According to Figure 4, 105 cities in the research area receive significant spillover effects from the HST zone. Among them, most of the cities (88 cities) receive positive spillover effects from the HST zone.

Figure 4.

The spatial distribution of the coefficients of $ρ_{1}$ .

According to Figure 5, there are 114 cities which receive significant spillover effects from the non-HST zone. Specifically, most of the regions located in the North China Plain area receive significantly negative spillover effects from the non-HST zone.

Figure 5.

The spatial distribution of the coefficients of $ρ_{2}$ .

As Figure 6 indicates, the significantly negative spillover effects from neighboring cities occur in northwestern Shandong province, northern Henan province, most of Hebei province, Beijing, and Tianjin city. To the contrary, the sample cities in western Jilin, eastern Liaoning, and western Hubei province receive great positive spillover effects from their neighbors.

Figure 6.

The spatial distribution of the coefficients of ρ_n.

Logically, the value of $ρ_{1} - ρ_{2}$ may reflect the level of the polarization effect. Noting that when one value of $ρ_{1}$ and $ρ_{2}$ of a particular city is statistically insignificant, we consider that the polarization effect received by this city is statistically insignificant. Figure 7 maps the spatial distribution of the value $ρ_{1} - ρ_{2}$ , visually revealing the heterogeneity of the polarization effect of HSTs proposed by this study. According to Figure 7, the sample cities in most of Shandong province, Hebei province, Henan province, northeastern Jiangsu province, southern Jiangxi province, southern Zhejiang province, northern Hubei province, and western Jilin province receive greater polarization effects relative to other sample cities.

Figure 7.

The spatial distribution of the values of $ρ_{1} - ρ_{2}$ .

Conclusions

Employing the two time periods (the non-HST era represented by years 2002–2006 and the HST era represented by years 2011–2015) and city-level panel data of eastern and central China and spatial modeling techniques, this study initially reveals the spatial heterogeneity and dynamics of tourism-flow spillover effects with the consideration of the role of HST service, which we believe fills an important research gap in current literature. The main results are concluded as follows.

Firstly, for a given city, tourism flows of the cities in the HST network contribute positive spillover effects, while those of cities outside the HST network induce the negative ones, which demonstrates the spatial heterogeneity of spillover effects. Moreover, this different nature of spillover effects (coming from HST zone and from non-HST zone) is termed the polarization effect of HST service, and this study presents a theoretical explanation to solidify this effect. The spatial heterogeneity of spillover effects and the polarization effect is further unveiled in detail through GWR models.

Secondly, by comparing the results for the non-HST era and those for the HST era, this study demonstrates the dynamics of spillover effects. As a distinct example, the tourism-flow spillover effect from the neighboring cities is positive in the non-HST era, but it changes to be negative in the HST era. After examining the variation of other variables during the two time periods, this study eliminates other potential explanatory factors and suggests that it is the intensive use of HST service that changes the spillover patterns, whereby the role of HST service is validated.

In the light of these findings, there are several valuable practical implications for tourism policy makers and tourism operators in China.

First, given the significant polarization effect of HST service, it is expected that in the long-term, tourism activities of China will agglomerate on the HST zone, and the non-HST zone will be gradually marginalized. For a given region, the trend of tourism agglomeration may be more obvious if the polarization effect is stronger. Naturally, the polarization effect of HST service rationalizes the concentration of tourism investment on the HST zone.

Second, effectively taking advantage of the HST network is very relevant for tourism development of the regions involved. Specifically, policy makers and tourism operators in cities outside the HST network should enhance cooperation with their counterparts in the HST zone. The policy makers should improve the transportation connectivity with the HST zone, facilitating the city to benefit from the positive spillover effect of tourism-flow clustering along the HST network. Operating a specialized public transportation to connect tourism attractions with HST stations is highly recommended. As for tourism operators, it is worth initiating cooperation among tourism attractions in a given city and those in cities connected in the HST network when recommending travel routes for tourists. Another suggestion for tourism operators is to advertise in the HST zone to increase the visibility of their tourism products. The effective sites for such advertisement may be on HSTs, in HST stations, and tourism attractions located in the HST zone. The advertisement helps to guide tourism flows of the HST zone into a given city which is excluded from the HST network. Additionally, a particular city should adopt a differentiation strategy to avoid the competition from the cities located outside the HST network, so that the potential “threats” brought about by the operation of HST service may be converted to the real “opportunities” for regional tourism.

Finally, when policy makers or tourism operators of a particular city design their specific strategies to develop the tourism industry or businesses in the HST era, the spatial heterogeneity of spillover effects, as well as that of polarization effects, should be taken into account, aside from the general strategies aforementioned. Only when a policy or strategy is compatible with the geographic characteristics, linked by the economic and cultural characteristics, of a specific region, these policies or strategies can generate the expected outcomes for a region in the HST era.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Natural Science Foundation of China (NSFC) under the project numbers: 71773101 and 41801138.

ORCID iD

Ian Sutherland

Seul Ki Lee

References

Akaike

(1974) A new look at the statistical model identification. In: Selected Papers of Hirotugu Akaike. New York, NY: Springer, pp. 215–222.

Albalate

Campos

Jiménez

(2017) Tourism and high speed rail in Spain: Does the AVE increase local visitors? Annals of Tourism Research 65: 71–82.

Anselin

(1988) Spatial Econometrics: Methods and Models. Dordrecht: Kluwer Academic Publisher.

Balli

Tsui

WHK

(2016) Tourism demand spillovers between Australia and New Zealand: evidence from the partner countries. Journal of Travel Research 55(6): 804–812.

Brunsdon

Fotheringham

Charlton

(1996) Geographically weighted regression: a method for exploring spatial nonstationarity. Geographical Analysis 28(4): 281–298.

Burnham

Anderson

(2004) Multimodel inference: understanding AIC and BIC in model selection. Sociological Methods & Research 33(2): 261–304.

Capello

(2009) Spatial spillovers and regional growth: a cognitive approach. European Planning Studies 17(5): 639–658.

Capone

Boix

(2008) Sources of growth and competitiveness of local tourist production systems: an application to Italy (1991–2001). The Annals of Regional Science 42(1): 209–224.

Chen

Haynes

(2015) Impact of high-speed rail on international tourism demand in China. Applied Economics Letters 22(1): 57–60.

10.

Crampon

(1966) A new technique to analyze tourist markets. Journal of Marketing 30(2): 27–31.

11.

Deller

(2010) Rural poverty, tourism and spatial heterogeneity. Annals of Tourism Research 37(1): 180–205.

12.

Deng

(2019) Modelling China’s outbound tourist flow to the ‘Silk Road’: a spatial econometric approach. Tourism Economics 25(8): 1167–1181.

13.

Ellerbrock

Hite

(1980) Factors affecting regional employment in tourism in the United States. Journal of Travel Research 18(3): 26–32.

14.

Fotheringham

Brunsdon

Charlton

(2003) Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. Hoboken, NJ: John Wiley & Sons.

15.

Getis

Aldstadt

(2004) Constructing the spatial weights matrix using a local statistic. Geographical Analysis 36(2): 90–104.

16.

Givoni

(2006) Development and impact of the modern high-speed train: a review. Transport Reviews 26(5): 593–611.

17.

Gooroochurn

Hanley

(2005) Spillover effects in long-haul visitors between two regions. Regional Studies 39(6): 727–738.

18.

Gutiérrez Puebla

(2005) El tren de alta velocidad y sus efectose spaciales. Investigaciones Regionales 5: 199–221.

19.

Hausman

(1978) Specification tests in econometrics. Econometrica 46: 1251–1271.

20.

Hiramatsu

(2018) Unequal regional impacts of high speed rail on the tourism industry: a simulation analysis of the effects of Kyushu Shinkansen. Transportation 45: 677–701.

21.

Hoti

McAleer

Shareef

(2007) Modelling international tourism and country risk spillovers for Cyprus and Malta. Tourism Management 28(6): 1472–1484.

22.

Khadaroo

Seetanah

(2007) Transport infrastructure and tourism development. Annals of Tourism Research 34(4): 1021–1032.

23.

Khadaroo

Seetanah

(2008) The role of transport infrastructure in international tourism development: a gravity model approach. Tourism Management 29(5): 831–840.

24.

Khan

Toh

Chua

(2005) Tourism and trade: cointegration and Granger causality tests. Journal of Travel Research 44(2): 171–176.

25.

Kulendran

Wilson

(2000) Is there a relationship between international trade and international travel? Applied Economics 32(8): 1001–1009.

26.

Lacombe

(2004) Does econometric methodology matter? An analysis of public policy using spatial econometric techniques. Geographical Analysis 36(2): 105–118.

27.

Le Sage

Pace

(2009) Introduction to Spatial Econometrics. Boca Raton, FL: Chapman and Hall/CRC.

28.

Lee

(2014) Revisiting the impact of bed tax with spatial panel approach. International Journal of Hospitality Management 41: 49–55.

29.

Lee

(2015) Quality differentiation and conditional spatial price competition among hotels. Tourism Management 46: 114–122.

30.

Lee

Jang

(2013) Asymmetry of price competition in the lodging market. Journal of Travel Research 52(1): 56–67.

31.

Losada

Alen

Cotos-Yanez

, et al. (2019) Spatial heterogeneity in Spain for senior travel behavior. Tourism Management 70: 444–452.

32.

Marrocu

Paci

(2013) Different tourists to different destinations. Evidence from spatial interaction models. Tourism Management 39: 71–83.

33.

Masson

Petiot

(2009) Can the high speed rail reinforce tourism attractiveness? The case of the high speed rail between Perpignan (France) and Barcelona (Spain). Technovation 29(9): 611–617.

34.

Patuelli

Mussoni

Candela

(2013) The effects of World Heritage Sites on domestic tourism: a spatial interaction model for Italy. Journal of Geographical Systems 15(3): 369–402.

35.

Raftery

(1995) Bayesian model selection in social research. Sociological Methodology 25: 111–164.

36.

Rosentraub

Joo

(2009) Tourism and economic development: Which investments produce gains for regions? Tourism Management 30(5): 759–770.

37.

Shareef

McAleer

(2008) Modelling international tourism demand and uncertainty in Maldives and Seychelles: a portfolio approach. Mathematics and Computers in Simulation 78(2-3): 459–468.

38.

Shaw

Fang

, et al. (2014) Impacts of high speed rail on railroad network accessibility in China. Journal of Transport Geography 40: 112–122.

39.

Sun

Lin

(2018) Move fast, travel slow: the influence of high-speed rail on tourism in Taiwan. Journal of Sustainable Tourism 26(3): 433–450.

40.

Wang

Qian

Chen

, et al. (2014) Influence of the high-speed rail on the spatial pattern of regional tourism–taken Beijing–Shanghai high-speed rail of China as example. Asia Pacific Journal of Tourism Research 19(8): 890–912.

41.

Wang

Niu

Qian

(2018) Evolution and optimization of China’s urban tourism spatial structure: a high speed rail perspective. Tourism Management 64: 218–232.

42.

Wang

Huang

Zou

, et al. (2012) Effects of the high speed rail network on China’s regional tourism development. Tourism Management Perspectives 1: 34–38.

43.

Yang

Fik

(2014) Spatial effects in regional tourism growth. Annals of Tourism Research 46: 144–162.

44.

Yang

Wong

(2012) A spatial econometric approach to model spillover effects in tourism flows. Journal of Travel Research 51(6): 768–778.

45.

Yang

Fik

Zhang

(2017) Designing a tourism spillover index based on multidestination travel: a two-stage distance-based modeling approach. Journal of Travel Research 56(3): 317–333.

46.

Zhang

Jensen

(2007) Comparative advantage: explaining tourism flows. Annals of Tourism Research 34(1): 223–243.

47.

Zhang

Guo

(2011) A study of the agglomeration of China’s convention industry: an economic and neo-economic geography framework approach. Tourism Economics 17(2): 305–319.

48.

Zhou

(2018a) High-Speed Trains and Regional Tourism Economies of China: An Empirical Study. Xiamen: Xiamen University Press.

49.

Zhou

(2018b) The impact of high-speed trains on regional tourism economies: empirical evidence from China. Tourism Economics 24(2): 187–203.

50.

Zhou

Yang

(2017) The spillover effect of attractions: evidence from Eastern China. Tourism Economics 23(4): 731–743.