Tourism Management 40 (2014) 46e58

Contents lists available at SciVerse ScienceDirect

Tourism Management
journal homepage: www.elsevier.com/locate/tourman

Analysis of international tourist arrivals worldwide: The role of world

heritage sites
Yu-Wen Su, Hui-Lin Lin*
Department of Economics, National Taiwan University, 21 Hsu-Chow Road, 10055 Taipei, Taiwan

h i g h l i g h t s

Investigate in the positive inuence of world heritage sites on international tourist arrivals.

Divide the sample into several groups according to the number of WHSs to study effects of WHSs across these groups.

Explore the pooled, xed and random effect models with panel data (66 countries, 2000e2009).

Eliminate the problem of time-invariant variables in panel data model by increasing the number of countries.

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 19 August 2012
Accepted 24 April 2013

This study examines the impact on inbound tourism caused by the presence of world heritage sites. The
statistics are derived from panel data for 66 countries for the period 2006e2009. The results indicate that
there exists a positive relationship between having such heritage sites and tourist numbers, and the
relationship is stronger for natural rather than for cultural heritage sites. The evidence also indicates the
presence of a U-shaped relationship between numbers of world heritage sites in a country and tourist
numbers. These relationships are found to be robust even though differences in patterns are found in
different regions.
2013 Elsevier Ltd. All rights reserved.

JELS:
L83
C23
Keywords:
WHSs
Tourism demand
Cultural sites
Natural sites
Panel data

1. Introduction
Tourism is one of the leading economic sectors in the world, and
represents a major source of income, employment, exports and
taxes. According to the World Travel and Tourism Council (WTTC),
in 2011 the tourism sector (domestic and international) contributed
almost 5992 billion USD to the global economy. With conrmed
strong linkage effects, the tourism industry also provides almost
260 million job opportunities, accounting for nearly 9% of global
employment. In addition, according to the World Bank Carbon
Finance Unit (CFU) the tourism sector is relatively eco-friendly
compared to the manufacturing sector, and has led to more

* Corresponding author. Tel.: 886 2 2321 7730; fax: 886 2 2322 5657.
E-mail addresses: [email protected] (Y.-W. Su), [email protected]
(H.-L. Lin).
0261-5177/$ e see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.tourman.2013.04.005

sustainable development. Therefore, many countries are emphasizing the development of tourism to drive their green economic
growth.
As disposable incomes and the awareness of the importance of
leisure have increased, so too have the numbers of tourists (Lim,
2006). World Tourism Organization (WTO) statistics reveal the
growth of international tourist arrivals (the x-axis on the righthand side) between 1995 and 2011, as shown in Fig. 1. The number of international tourist arrivals increased from 538 million in
1995 to 940 million in 2010, representing growth of 4.7% on average
in each year. Meanwhile, according to the World Heritage Centre of
the United Nations Educational, Scientic and Cultural Organization (UNESCO), the total number of World Heritage Sites (WHSs)
has risen steadily. Fig. 1 also shows that the number of WHSs (the xaxis on the left-hand side) increased from 468 in 1995 to 936 in
2011, or by 6% per year on average. Thus, these growing trends
appear to suggest that, if the positive effect of WHSs on international tourism is proved, having such sites will lead to increases in

1000

1000

900

900

800

800

700

700

600

600

500

Number of world heritage sites

Tourist arrivals

400
1995

1997

1999

2001

2003

2005

2007

2009

500

47

Tourist arrivals (million)

Number of world heritage sites

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

400

2011

Fig. 1. Numbers of world heritage sites and international tourist arrivals. Source: The
World Tourism Organization (The International tourist arrivals), The World Heritage
Centre, UNESCO (The number of WHSs).

international tourist arrivals and consequently tourism expenditure, thereby beneting the economies of the destination countries.
However, there have so far been few studies on this subject.
In this paper, we investigate the relationship between the WHSs
and tourism demand not only for specic countries but also on a
worldwide level. First, by using data on the number of WHSs in 66
countries between 2000 and 2009, we explore the positive inuence of these sites on international tourist arrivals (international
tourism demand). Second, we divide our sample into several
groups according to the number of WHSs and study the different
effects of WHSs on international tourism demand across these
groups. Third, we apply various xed and random effects models to
the panel data. In addition, we increase the number of countries in
the panel data model to eliminate the problem of time-invariant
variables, or rarely changing variables.
The remainder of this paper is organized as follows. In Section 2,
we provide a literature review together with an analytical framework of WHSs and international tourism worldwide. In Section 3,
we introduce the models setting and the methodology of the panel
data. In Section 4, we present the results of the analysis and discuss
the economic implications. In Section 5, we conclude.
2. World heritage sites and international tourists
In this section, we provide a literature review and briey
describe the analytical framework of WHSs and international
tourism worldwide based on the current situation.
2.1. Literature review
As the number of tourists increase, governments and private
enterprises around the world have been eager to expand their
tourism. Many studies have examined the key elements affecting
tourism demand (e.g., Dhariwala, 2005; Dougan, 2007; Dritsakis,
2004; Naude and Saayman, 2005; Patsouratis, Frangouli, &
Anastasopoulos, 2005; Payne and Mervar, 2002; Tan, McCahon, &
Miller, 2002). It has also been found that tourism destinations
with typical cultural or natural elements constitute one of the chief
attractions for international tourists (e.g., Bille and Schulze, 2008;
Bonet, 2003; Cooke and Lazzaretti, 2008; Deng, King, & Bauer,
2002; Dritsakis, 2004).
Since cultural or natural attractions lead to increased tourism
demand, it could be argued that those attractions that are ofcially
authenticated, i.e., inscribed on the list of WHSs by UNESCO, should
be relatively appealing to international tourists. WHSs have been
found to have signicantly positive effects on the promotion of
domestic or foreign tourism in some specic countries, such as
England (e.g., Herbert, 2001; McIntosh and Prentice, 1999), China

Fig. 2. Number of world heritage sites by location in 2009. Source: The World Heritage
Centre, UNESCO.

(e.g., Li, Wu, & Cai, 2008; Yang, Lin, & Han, 2010) and Germany,
Hungary and Romania (Light, 2000). Nevertheless, studies on the
positive effect of WHSs on tourism have been limited to a single
country, and little research has been done to expand this effect to a
worldwide level.
In addition, in terms of the methodology adopted, some studies
employ the panel data model (e.g., Garin-Munoz and Amaral, 2000;
Ledesma-Rodriguez, Navarro-Ibanez, & Perez-Rodriguez, 2001;
Maloney and Montes-Rojas, 2005; Naude and Saayman, 2005; Yang
et al., 2010), because of the availability of the data. However, in the
panel data model, the problem of time-invariant variables, or rarely
changing variables, is widely discussed (Cellini, 2011; Yang and Lin,
2011). In this paper, we also use the panel data model, and the
number of countries is increased to eliminate the problem of timeinvariant variables.

2.2. Research background

With the increasing trend in terms of the number of WHSs (see
Fig. 1), the geographical distribution of heritage sites is relatively
unbalanced. According to data collected by the World Heritage
Centre, UNESCO, Fig. 2 shows the pie chart of WHSs by region in
2009. WHSs are mainly concentrated in Europe, which accounts for
42% of the total amount, followed by the Asia Pacic with 20%, and
the Americas with 17% (the sum of the North and the South
Americas). Other areas, that is, Africa and the countries of the
Middle East, each account for around 10%. Overall, European
countries, which have highly developed tourism, possess rich cultural and historical attractions, including almost half of all WHSs.

Italy
Spain
China
Germany
France
Mexico
U.K.
India
Russia
U.S.
Greece
Brazil
Australia
Canada
Sweden
Japan
Poland
Portugal
Iran
Czech

Cultural site
Natural site
Mixed site
0

10

20
30
Number of world heritage sites

40

Fig. 3. Top 20 countries in terms of world heritage sites in 2009. Source: The World
Heritage Centre, UNESCO.

48

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

To reveal this phenomenon in better detail, Fig. 3 shows the top

20 countries according to the numbers of WHSs in 2009 provided
by the World Heritage Centre, UNESCO. The total number of WHSs
is the summation of three kinds of sites: cultural, natural and mixed
sites. Half of these countries are located in Europe. Italy, the country
with the most WHSs, possesses 44 WHSs, including 42 cultural
sites and 2 natural sites. The country with the second most WHSs,
Spain, has 41 WHSs (36 cultural, 3 natural and 2 mixed sites), and
China has 38 WHSs (27 cultural, 7 natural and 4 mixed sites). In
addition, a large proportion of WHSs are cultural sites, which account for around 80% of all sites. Some countries even possess only
cultural sites, such as Iran and the Czech Republic, each of which
has 12 cultural sites. However, in a minority of natural resourceabundant countries, such as the United States and Australia, there
are more natural sites. The U.S. has 12 natural sites among 20 WHSs
(60%), and Australia has 11 natural sites among 17 WHSs (65%).
As for the demand for tourism, Fig. 4 shows the top 20 countries
ranked by international tourist arrivals in 2009, according to data
compiled by the World Tourism Organization. France, the most
popular country for tourism, received 76.8 million international
tourists in 2009. Inbound tourist arrivals in the United States, Spain
and China were 54.9 million, 52.2 million and 50.9 million,
respectively, while Italy received 43.2 million inbound tourists, or
about 60% of the number that France received. International tourist
arrivals in other countries were all less than 30 million in 2009.
Among the 20 countries in Fig. 4, a total of 11 countries
possess rich world heritage sites, which are also ranked in the
top 20 countries according to the number of WHSs (see Fig. 3). In
fact, among 194 countries around the world, the other 9 countries in Fig. 4 are also ranked in the top 40 according to the
number of WHSs, apart from Thailand, which is ranked ftieth.
Meanwhile, the simultaneous growth trends in both WHSs and
international tourists are also shown in Fig. 1. Therefore, the data
reveal that under the growing trend of both WHSs and international tourism, a large proportion of countries popular with
tourists are those which are abundant in cultural or natural
world heritage sites.
2.3. Analytical framework
There are at least two possible reasons why being inscribed on
the WHS list would increase the demand for tourism (Yang et al.,
2010). First, the WHSs are widely used to promote or advertise
tourism in destination countries, not only by travel agencies, but
also by governments. Because of the ofcially strict application and

examination processes, being successfully inscribed on the WHS

list increases the global visibility of the destination countries. Since
WHSs attract the attention of international tourists, the demand for
international tourism rises. Second, in regard to conservation,
UNESCO is prepared to assist those developing countries which lack
the resources or ability to repair and maintain their WHSs. For a
destination country, using such aid well will improve tourism
conditions and further attract international tourists.
Moreover, the main purpose behind listing WHSs is to raise
awareness and mobilize sustainable resources for long-term conservation (according to the World Heritage Centre, UNESCO), and is
not the development of tourism. However, if WHSs have positive effects on tourism demand and further on tourism economies, this
additional benet could also raise awareness and help fund the
conservation efforts. Since the government plays an important role in
administering resources, it is essential to study the economic effects
of WHSs. In spite of this, there has been little research that has done
just that. Therefore, under the growing trend of both WHSs and international tourism, the main purpose of our paper is to conrm the
positive effect of WHSs on the demand for international tourism.
Furthermore, we study how this positive effect has changed.
3. Methodology and data
In this section, we introduce the methodology, the panel data
model that is widely used, and briey introduce the data we
employ.
3.1. Modeling the international tourism demand
To investigate the determinants of international tourist arrivals
worldwide, especially the effect of world heritage sites, a tourism
demand function is estimated in this study. The demand model is
specied as

yit f xit ; zit qi 3 it

(1)

where yit is the quantity of tourism demand, and the subscripts i

and t denote the destination country and time period, respectively.
The xit are the main explanatory variables in which we are interested, the zit are control variables which also affect the demand, and
3 it is a normally distributed error term. Meanwhile, f (.) is a function,
which is set to be linear in this paper. Note that qi is the unobserved
country-specic variable that varies across countries but is
invariant within a country over time.

France
U.S.
Spain
China
Italy
U.K.
Turkey
Germany
Mexico
Austria
Canada
Greece
Thailand
Egypt
Poland
Netherlands
Croatia
Hungary
Morocco
Switzerland
0

10

20

30

40
50
Tourist arrivals (million)

60

70

80

Fig. 4. Top 20 countries in terms of international tourist arrivals in 2009. Source: The World Tourism Organization.

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

More specically, the linear international tourism demand

model is specied as

ARRIit a dWHSit b1 GDPit b2 POPit b3 EXit b4 RAILit

6
X
j1

gj AREAjit

10
X

and cannot improve the model, so they are omitted from the model.
More details about our variables are shown in Table 1, which provides the denitions and descriptive statistics of the variables.
3.2. Varying marginal effect of world heritage sites

b5 FREEDOMit b6 HEALTHit b7 EDUit

49

The marginal effect of WHSs is the partial derivative of ARRIit

with respect to WHSit in Eq. (2).

gk YEARkit qi 3 it

k1

(2)
where the dependent variable, ARRI, is the international tourist
arrivals in country i at time t, which is often treated as the tourism
demand in the literature (e.g., Lim, 2006; Song and Li, 2008). WHS
represents the number of world heritage sites, which is the main
explanatory variable (xit) we are interested in. If the sign of its coefcient, d, is positive, we could say that possessing WHSs would
enhance international tourism after controlling other variables. In
addition, we have replaced WHS by CULTURAL and NATURAL, the
numbers of cultural and natural WHSs, to differentiate the effects of
cultural and natural WHSs on international tourism demand.
The other explanatory variables (zit), are regarded as control variables capturing some possible factors which would inuence the demand. The gross domestic product (GDP) variable represents the
income level, which also captures the degree of economic development in the destination country. The population variable (POP) mainly
controls the size of the destination country. That is, after considering
POP, the effect of the GDP and other explanatory variables could be
measured accurately under the same scale of population. For example,
the positive coefcient of GDP means that among countries with the
same population, international tourists prefer to travel to the richer
one with higher income. Moreover, EX denotes the ofcial exchange
rate between the local currency unit (LCU) and the U.S. dollar, which
represents the price factor in the demand function. If EX goes up, the
traveling price (cost) increases, in which case the number of international tourist arrivals would decrease based on the law of demand.
In addition, the total railway lines (RAIL) in terms of kilometers in
destination countries is employed as a proxy variable for the availability of infrastructure. A country that possesses more railway lines is
a country in which it is more convenient to travel, and this will attract
more international tourists. The FREEDOM variable is the index of
political rights and civil liberties, which is measured on a one-toseven scale. Theoretically, a smaller value of FREEDOM represents a
freer political and civil environment that would make international
tourists feel more secure without red tape and increase their willingness to travel. Moreover, the HEALTH variable is the percentage of
health expenditure in GDP, and is used as a proxy variable for the
environmental sanitation in destination countries. If a country spends
more money caring for its residents health, the sanitary condition in
the country will be further improved. To measure the health quality of
residents and the educational environment in destination countries,
the percentage of expenditure on education in GDP (EDU) is also used
as a proxy variable. In addition, to control the time and regional factors, YEAR and AREA are dummy variables denoting the time from
2000 to 2009 and the geographical position of 6 areas, respectively.
However, there is a potential simultaneous relationship between tourist arrivals and some explanatory variables, such as GDP,
and so variables in the form of a lag of one period enter the equation. The results turn out to be quite consistent with those without
the lag term. Thus, to keep the sample size as large as possible
(using the lag term will reduce the sample size by 66 observations),
we choose the original models without lag terms. Other possible
explanatory variables, such as FDI (measuring the openness level),
WHSs in danger, global infectious diseases, or interaction terms,
have also been considered but have turned out to be insignicant

Marginal effect of WHS

vARRIit
d
vWHSit

(3)

That is, for a destination country, possessing one more WHS

would increase its inbound tourists by d visits. This d is the average
effect across all countries with different numbers of WHSs. However, even though the constant marginal effect of WHSs could be
easily concluded by d, this effect may change by the different
number of WHSs. Thus, to reveal more details, we divide our
sample into several equal parts according to the number of WHSs.
The new model is specied as

ARRIit a

S
X

ds WHSit  gsit b1 GDPit b2 POPit b3 EXit

s1

b4 RAILit b5 FREEDOMit b6 HEALTHit b7 EDUit

6
X

gj AREAjit

j1

10
X

gk YEARkit qi 3 it

k1

(4)
where gsit is the dummy variable for the sth group of the WHS. The
data are divided into S equal parts, according to the number of
WHSs. The WHSit multiplied by gits make us focus on the marginal
effect for a specic range of the number of WHSs.
However, when we decide to divide our sample, the rst question
is concerned with how many groups we would obtain. On the one
hand, if S is too large, which means that the number of groups is large,
the small size of the subsample would give rise to highly sensitive
estimated results. On the other hand, if S is too small, say, S 2 (as a
result of dividing the sample into two equal parts), the large subsample would reveal little of the varying marginal effect of the WHSs.
Moreover, we should note that the WHS is a discrete right-skewed
variable, which means that the number of WHSs is an integer and
the data are concentrated in small numbers. In this distribution, it is
impossible to divide the data into too many equal parts.
As a result, we start with three equal parts and extend this to
seven equal parts (S 3, 4 .7). It turns out that when S is equal to 4,
the decrease pattern is the same as the one when S is equal to 3.
When S is bigger than 5, the estimated results of the U-curve effect
would be similar to the case where S 5. Therefore, in this paper, we
present two representative results, S 3 and S 5, because they are
the smallest groups to capture the pattern of varying marginal effects and perform well in dividing the data into several equal parts.
3.3. Methodology
To understand the preliminary sign of each determinant, pooled
ordinary least squares (OLS) regression is employed at rst (e.g.,
Naude and Saayman, 2005; Yang et al., 2010). Thus, the pooled OLS
residual (uit) is the summation of the country-specic unobserved
variable (qi) and the error term with a normal distribution (3 it):

uit qi 3 it

(5)

We run the BreuschePagan test (BeP test) to test for heteroskedasticity (qi s 0) in the pooled OLS models (Breusch & Pagan,

50

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

Table 1
Denitions of variables and basic statistics.
Variable

Description

ARRI
WHS
CULTURAL
NATURAL

International tourist arrivals (1000)

Number of world heritage sites
Number of cultural world heritage sites
Number of natural world heritage sites

GDP
POP
EX
RAIL
FREEDOM
HEALTH
EDU

GDP (billion, constant 2000 USD)

Population (million)
Ofcial exchange rate (LCU per US$, period average)
Rail lines (total route-km)
The index of political rights and civil liberties
Health expenditure (% of GDP)
Education expenditure (% of government expenditure)

AREA

Dummy variable: Africa, Asia Pacic, Middle East,

Europe, N. America, S. America
Dummy variable: 2000e2009

YEAR

Mean

1979). In the pooled OLS estimation with heteroskedasticity,

omitting the unobserved variable, which may be correlated with
other explanatory variables (xit or zit), will cause severe problems of
bias and inconsistency. Fortunately, this problem could be solved in
a panel data model under certain assumptions. Using the xed effects (FE) or the random effects (RE) technique, we could eliminate
the country-specic effect. The xed effects model assumes that
each country has its own qi and estimates the constant term for
each country, while the random effects model assumes that qi
follows a normal distribution, thus estimating one overall constant
term. In this paper, both the xed and random effects models are
estimated, and then the Hausman test is employed to determine
which model is more accurate. Under the null hypothesis (H0), the
RE model performs better, and if the Chi-square statistic is significant (the p-value is small), then H0 should be rejected and the FE
model chosen. We also show the estimated results of the pooled
OLS for reference. For more details about the panel data model and
the Hausman test, the interested reader should refer to
Chamberlain (1984), Hausman (1978) and Wooldridge (2002).
3.4. Data

Max

3.00
0.00
0.00
0.00

809.00
44.00
42.00
12.00

246.21
42.68
659.77
11545.25
5.19
6.35
15.48

1029.56
143.00
2078.67
25988.60
2.18
2.22
5.50

0.25
0.03
0.00
251.00
1.00
0.01
6.20

11670.80
1331.38
17065.08
228999.00
7.00
16.21
71.09

.25

(B) Actual data

Density

.15

.05

.05

.1

.1

.15

.2

(A) Original data

Density

Min

1100.00
7.29
6.33
1.89

RAIL, HEALTH, and EDU) are collected from the World Development
Indicators (WDI) of the World Bank Online Resources. The data for
the freedom index (FREEDOM) come from the annual report of
Freedom House.
When combining these four data sets, we try to collect as many
informative observations as we possibly can. However, there are
missing data, more or less, for each variable, and especially for
some developing countries whose statistical surveys are less
comprehensive. To focus on countries with relatively more information and avoid too much missing data causing a severe problem
of data imbalance in the panel data, countries with too many kinds
of data unavailable are deleted without loss. We originally collected
the data for the WHSs of 148 countries. After combining data sets,
the data actually used consist of 66 countries. The data distributions
of WHSs before and after combining the data are shown in Fig. 5.
The countries deleted are mostly those containing few or no WHSs.
One thing we should mention is that, among these observations,
there is only one, Israel in 2000, that possesses zero WHSs.
Therefore, sifting the observations cannot only simplify the
analysis but also will not critically affect the estimated results. In
this research, the panel data comprise 66 countries over the period
from 2000 to 2009 with 359 observations after deducting the
missing data for each variable. The names of these 66 countries are
listed in Appendix A.
Two things should be mentioned. First, the literature shows that
the characteristics of the source country may somehow affect the
tourism demand. However, what we only know is the annual

.2

.25

In this study, the data on international tourist arrivals (ARRI)

come from the World Tourism Organization (WTO), while the data
on the number of WHSs, including cultural and natural sites, (WHS,
CULTURAL, and NATURAL) come from the World Heritage Centre,
UNESCO. In addition, the other explanatory variables (GDP, POP, EX,

S.D.

5355.10
5.50
4.23
1.09

10

20

30

40

Number of WHSs

10

20

Number of WHSs
Fig. 5. Distribution of WHSs.

30

40

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

number of international tourist arrivals for the destination country,

according to the UNWTO. We are unable to distinguish from which
countries these international tourists come. Thus, we cannot pin
down the data on GDP or POP for the source countries. Fortunately,
the panel data model is employed, so that the effect due to different
source countries can be captured by a country-specic factor, qi.
Moreover, based on the test results, the random effects model
performs better, so we consider that the country-specic factor
would vary in different destination countries.
Second, the effect on income, which usually uses GDP as a proxy
variable, is undoubtedly an important explanatory variable. When
constructing the model, we initially considered the GDP per capita,
which could eliminate the effect of the countrys size because it is
counted based on each person. However, the result turns out to be
insignicant. Therefore, to control for the income effect as well as
the country size, we separate the GDP per capita into two different
variables: the GDP and the population (POP), and the results is
better. Note that, regardless of which control variables are chosen,
the coefcient of WHSs, which is the point of our paper, is almost
the same.
4. Empirical results
Using the panel data, we investigate the effect of WHSs (both
cultural and natural sites) on international tourist arrivals, while
other possible explanatory variables are controlled. We also explore
how this effect changes for different numbers of WHSs.
4.1. Main results
The estimated results for Eq. (2) are shown in Table 2. Models
(1), (2) and (3) use the number of WHSs (WHS) as their explanatory
variable, while Models (4), (5) and (6) separate WHSs into cultural
and natural sites (CULTURAL and NATURAL) to better understand the
inuence of these two kinds of sites. In these models, we estimate
coefcients by pooled OLS regression, and the xed effects and the
random effects models. The latter two models are ideal for dealing
with the country-specic unobserved variables, and could also be
judged by the Hausman test. According to the results of the
BreuschePagan test, the pooled OLS regression with heteroskedasticity is beset by problems resulting in inconsistency and
bias. We could observe that the coefcients are quite different

51

between the pooled OLS regression and the panel data model (both
the xed and random effects models). This result proves that the
country-specic effect should be considered. If we were to just grab
the data and run the regression directly, the estimated results
would be unreliable.
According to the Hausman test, in which the Chi-square statistics are insignicant at the 5% signicance level, the random effects
model, as in Model (3) and Model (6), performs better. This result is
quite reasonable. Because the data comprise a cross section of
countries, in considering the sampling problem, it makes sense to
assume that the omitted variable is distributed randomly. Note that
Model (5), the xed effects model, unexpectedly performs better at
the 10% signicance level. This may be caused by the imprecise
setting of the WHSs, which are assumed to have constant effects.
When the varying effects of WHSs are considered later, all the
random effects models are found to perform better.
In Model (3), the number of WHSs has a signicantly positive
effect on international tourist arrivals. That is, adding one WHS
would on average increase the number of international tourist arrivals by 382,637 in just one year after controlling other variables.
Thus, this positive effect proves that a country possessing more
WHSs would promote international tourism, not only for some
specic countries but for the whole world. Moreover, possessing
more WHSs increases the international tourism demand, which
also brings in relatively more tourist expenditures to the tourismrelated industries, such as accommodation, transportation or
even retail outlets located around the site. These industrial linkages
will generate several times the revenue earned from the visits to
the WHSs themselves.
In Model (6), both the cultural and natural WHSs have significantly positive effects on the number of international tourist arrivals when other variables are controlled. Increasing the number
of cultural sites by one would create an additional 396,659 international tourist arrivals, while adding one more natural site
would increase international tourist arrivals by 418,606, which is
on average 21,947 more tourist arrivals than for an additional
cultural one. To sum up, both the cultural and natural world heritage sites could enhance international tourism, and the effect is
greater for the natural world heritage sites than for the cultural
ones.
Models (3) and (6) assume that the marginal effect of WHSs is
constant, which is quite a simplication. However, the marginal

Table 2
Estimated results of international tourist arrivals (with constant effects of WHSs).

WHS

(1)
Pooled OLS

(2)
Fixed effects

(3)
Random effects

533384.15***
[11.77]

89416.14
[0.64]

382637.04***
[4.69]

CULTURAL
NATURAL
Control var.
Constant

Yes
6461538.15***
[3.54]

BeP test

116.26***
(p-value 0.000)

Hausman test
R-square
Chi-square
Observations

Yes
3396688.79
[1.43]

Yes
3148908.15
[1.41]

(4)
Pooled OLS

(5)
Fixed effects

(6)
Random effects

563357.95***
[10.79]
637057.13***
[4.57]

15548.61
[0.09]
292458.01
[1.07]

396658.60***
[4.16]
418605.71**
[2.10]

Yes
6531921.78***
[3.60]

Yes
3631736.20
[1.53]

Yes
3099144.91
[1.38]

111.78***
(p-value 0.000)

0.815

22.00
(p-value 0.143)
0.539

359

359

1. t statistics are in parentheses.

2. *, ** and *** denote signicance at the 10%, 5% and 1% statistical levels.

0.737
461.239***
359

0.818

25.27*
(p-value 0.089)
0.547

359

359

0.739
460.099***
359

52

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

effect may vary according to the number of WHSs. That is, the effect
of WHSs on international tourist arrivals may differ between
countries with an abundant supply of WHSs and countries with few
WHSs. Thus, to better understand the marginal effect of WHSs for a
specic range of numbers, we have divided our sample into three
and ve equal parts. Still, the pooled OLS, xed effects and random
effects models of Eq. (4) are estimated. The pooled OLS model is
tested for heteroskedasticity by BeP test, and the latter two models
are judged using the Hausman test. According to the test results, all
the random effects models, i.e., Model (9), (12), (15) and (18), are
more accurate (Table 3).
In Model (9), just like in the previous analysis, the marginal
effect of the WHSs is positive. Moreover, this positive effect declines
as the number of WHSs rises. For countries possessing 0e5, 6e11
and more than 12 WHSs, the marginal effects of WHSs are around
692, 495, and 408 thousand, respectively. The positive effect of
WHSs on international tourist arrivals is larger in countries with
fewer WHSs. This result is quite reasonable. For WHS-poor countries, once unknown sites become famous after being included on
the WHS list, this will attract more visits from international tourists. On the contrary, for WHS-abundant countries, which already
possess many attractions and are famous in the global tourism
market, adding one more WHS will result in a smaller increase in
inbound tourists than for WHS-poor countries.
Correspondingly, in Model (12), both cultural and natural sites
have positive effects on international tourist arrivals. These effects
also decrease as the number of cultural and natural WHSs increases. The marginal effects are 562, 490 and 402 thousand for
countries with 0e5, 6e11 and more than 12 cultural WHSs,
respectively. Meanwhile, for countries with 0e3 and more than 4
natural WHSs, the marginal effects are 514 and 419 thousand,
respectively. Note that because the sample size of natural WHSs is
relatively small, it is divided into two equal parts only.

To understand the marginal effects of WHSs in more detail, the

sample is also divided into ve equal parts. In Models (15) and (18)
of Table 4, the effects of WHSs, cultural WHSs and natural WHSs are
still positive. Moreover, the pattern of the decreasing marginal effects as the number of WHSs increases is almost the same. However, after controlling for more WHSs, the marginal effects of WHSs
for WHS-abundant countries (possessing more than 21 WHSs) increase instead. This increase means that when a country possesses
sufcient WHSs, the gearing effect of WHSs will emerge. For
countries possessing 0e3, 4e6, 7e10, 10e20 and more than 21
WHSs, the marginal effects of WHSs decrease from 975, to 580, 498
and 375 thousand, and rise slightly rise to 475 thousand, respectively. Similarly, the marginal effects of cultural WHSs are 776, 361,
373, 285 and 509 thousand, respectively. Meanwhile, the marginal
effects of natural WHSs for countries with 0e3 and more than 4
natural WHSs are 514 and 419 thousand, respectively. In addition,
compared with Models (9), (12), (15) and (18), Model (15) with the
highest R-square of 0.749, is a relatively accurate model. Based on
Models (15) and (18), Fig. 6 shows how these marginal effects of
WHSs vary based on the number of WHSs, and the U-curve of the
effect is quite obvious.
We also considered the quadratic form of the WHSs when a Ucurve resulting in S 5 was observed, but the estimated results of
the quadratic form are not good enough. A possible reason for the
badly-performing quadratic form is that, in Table 4, the coefcients
are not very smooth so that the quadratic form cannot capture the
pattern well. Therefore, dividing the data into 5 groups creates
more exibility to the varying coefcients and ts the model better,
even though the number of coefcients needed to be estimated
increases.
In addition, the other explanatory variables merit discussion.
Among Models (15) and (18), the coefcients of the variables are
quite similar, regardless of the number of WHSs or the cultural and

Table 3
Estimated results of international tourist arrivals (with 3 variant effects of WHSs).

WHS (0e5)
WHS (6e11)
WHS (12 up)

(7)
Pooled OLS

(8)
Fixed effects

(9)
Random effects

649675.36***
[3.42]
524222.32***
[6.37]
552693.31***
[11.03]

471627.62***
[2.34]
261853.32*
[1.74]
148088.03
[1.07]

691823.08***
[4.30]
494946.73***
[5.13]
408442.34***
[4.91]

CULTURAL (0e5)
CULTURAL (6e11)
CULTURAL (12 up)
NATURAL (0e3)
NATURAL (4 up)
Control var.
Constant

Yes
6738359.51***
[3.50]

BeP test

132.97***
(p-value 0.000)

Hausman test
R-square
Chi-square
Observations

Yes
1631441.45
[0.67]

Yes
4313752.97*
[1.87]

(10)
Pooled OLS

(11)
Fixed effects

(12)
Random effects

343297.35
[1.60]
415442.66***
[4.64]
574952.78***
[10.33]

328116.38
[1.44]
149397.07
[0.78]
45192.87
[0.26]

561704.25***
[2.96]
489724.36***
[4.35]
402313.64***
[4.14]

1245048.05***
[4.24]
525079.02***
[3.59]

431978.19
[1.39]
319811.62
[0.85]

514498.16*
[1.88]
419242.31*
[1.82]

Yes
6414358.53***
[3.40]

Yes
2276009.67
[0.93]

Yes
3789343.37
[1.63]

165.27***
(p-value 0.000)

0.816

22.37
(p-value 0.216)
0.537

359

359

1. t statistics are in parentheses.

2. *, ** and *** denote signicance at the 10%, 5% and 1% statistical levels.

0.733
466.364***
359

0.823

23.03
(p-value 0.288)
0.537

359

359

0.734
455.597***
359

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

53

Table 4
Estimated results of international tourist arrivals (with 5 variant effects of WHSs).

WHS (0e3)
WHS (4e6)
WHS (7e10)
WHS (11e20)
WHS (21 up)

(13)
Pooled OLS

(14)
Fixed effects

(15)
Random effects

717354.26***
[2.38]
585643.22***
[3.95]
531494.95***
[5.83]
299724.16***
[4.54]
716652.88***
[13.62]

759195.41***
[2.86]
346695.08*
[1.91]
254907.28*
[1.64]
107881.51
[0.76]
184221.66
[1.27]

975394.75***
[4.15]
580418.01***
[4.06]
498169.26***
[4.61]
356668.67***
[4.01]
474846.97***
[5.42]

CULTURAL (0e3)
CULTURAL (4e6)
CULTURAL (7e10)
CULTURAL (11e20)
CULTURAL (21 up)
NATURAL (0e3)
NATURAL (4 up)
GDP

(16)
Pooled OLS

(17)
Fixed effects

(18)
Random effects

214136.49
[0.86]
389665.43***
[2.75]
289594.82***
[3.21]
354390.56***
[5.20]
913255.68***
[12.23]

446149.55
[1.59]
33589.64
[0.16]
45539.87
[0.23]
129725.75
[0.73]
131702.65
[0.79]

776329.93***
[3.29]
360513.45***
[2.34]
373257.65***
[3.08]
284574.55***
[2.75]
508761.04***
[5.16]

1259993.21***
[4.52]
439219.79***
[3.26]

599675.81**
[2.13]
158265.28
[0.44]

750817.04***
[2.93]
298612.05
[1.34]

1011.06***
[4.09]
22989.06***
[7.83]
392.67***
[2.96]
168.28***
[10.78]
167820.66
[1.18]
200791.39
[1.48]
354455.52***
[7.00]

3912.38***
[2.73]
25653.01
[0.76]
158.20
[1.02]
47.84*
[1.83]
460859.81***
[2.37]
516818.85***
[3.00]
51271.76
[1.17]

1191.96***
[2.36]
14762.50***
[3.31]
102.65
[0.81]
102.34***
[6.07]
217766.63*
[1.69]
308048.78**
[2.01]
1071.33
[0.03]

351.46
[1.29]
18369.94***
[6.30]
514.67***
[3.85]
206.62***
[11.49]
258111.45*
[1.80]
157775.79
[1.16]
332053.13***
[6.65]

3096.00**
[2.17]
22490.11
[0.69]
149.48
[1.00]
61.58***
[2.39]
518235.42***
[2.69]
359633.58**
[2.14]
48415.31
[1.14]

1026.69**
[1.99]
11419.79***
[2.46]
115.16
[0.92]
109.33***
[6.56]
273159.23*
[1.64]
231209.95
[1.52]
7087.38
[0.17]

AREA
YEAR
Constant

Yes
Yes
6039499.56***
[3.12]

Yes
Yes
1659178.51
[0.68]

Yes
Yes
4300985.68*
[1.85]

Yes
Yes
6256850.52***
[3.46]

Yes
Yes
4221109.45*
[1.75]

Yes
Yes
2887056.58
[1.23]

BeP test

181.48***
(p-value 0.000)

POP
EX
RAIL
FREEDOM
HEALTH
EDU

Hausman test
R-square
Chi-square
Observations

223.73***
(p-value 0.000)

0.841

19.79
(p-value 0.471)
0.558

359

359

0.749
486.3***
359

0.843

11.37
(p-value 0.955)
0.486

359

359

0.739
506.274***
359

1. t statistics are in parentheses.

2. *, ** and *** denote signicance at the 10%, 5% and 1% statistical levels.

1600

Tourist arrivals(1000)

WHSs

Cultural WHSs

Natural WHSs

1200

800

400

0-3

4-6

7-10
11-20
Number of WHSs

20 up

Fig. 6. Marginal effects of WHSs, cultural WHSs and natural WHSs.

natural sites that are included in the model. In Model (15), for
example, the marginal effect of GDP is positive, which means that
tourism demand benets from economic development. When the
GDP of the destination country (GDP) increases by 1 billion USD,
international tourist arrivals will increase by 1192. Meanwhile, the
signicantly negative effect of the population (POP) indicates that,
when other things are equal, international tourists would like to
travel to destinations with fewer people or smaller country size.
For instance, among countries with the same numbers of WHSs,
economic achievements, and infrastructure, etc., international
tourists would be more likely to choose those smaller in size or
that are less crowded. This kind of destination may be easier to
travel around or more suitable for short vacation arrangements. In
addition, railway lines (RAIL) have a positive effect on international tourist arrivals, with the number of international tourist
arrivals increasing by 102 for each extra kilometer that the

54

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

railroad route is extended. It thus makes sense that the destination will be more attractive to international tourists when the
basic transportation is more convenient. The political and civil
freedom (FREEDOM) variable negatively affects the tourism demand at the 10% signicance level. When the index of freedom,
measured on a one-to-seven scale, increases by one (becoming
less free), the number of international tourist arrivals declines by
217,767. That is, freer countries attract more international tourists.
In addition, the health expenditure share of GDP (HEALTH) has a
positive inuence. When a country spends more money on health,
say, 1% of GDP, it will improve the sanitary conditions and increase
its inbound tourist visits by 308,048.
The effects of other controlled variables, namely, the exchange
rate (EX) and the expenditure proportion of education (EDU), are
insignicant in Model (15). However, it should be noted that the
sign of the exchange rate is negative, which means that
increasing the relative price will make the number of international tourist arrivals drop. Thus, the price effect of tourism exists, even though the coefcient is insignicant at the 10%
signicance level.
4.2. Comparison of regions and time periods
The behavior of tourists may vary in different destinations, and
some effects may also change over time. Therefore, based on Model
(15), we separate our observations according to the region and the
time period to reveal more details. This further research may also
be seen as a robustness check of our model, especially for WHS
variables. The estimated results of eight models are classied

according to four regions in Table 5: Africa, Asia, Europe and

America, and four time periods in Table 6: 2000e2001, 2003e2003,
2004e2006 and 2007e2009.
Three things should be noted. First, to avoid the small sample
problem, we combine the Asia Pacic with the Middle East as the
Asia group, and combine North America with South America as the
America group. Second, because the numbers of WHSs of African
countries are all below ten, the coefcients of WHS (11e20) and
WHS (21 up) are eliminated in Model (19). Third, in mid-2003, the
SARS epidemic occurred, especially in China, Singapore and Canada. Later, in 2009, the H1N1 epidemic also occurred in several
countries. These two dummy variables of SARS (for countries
whose conrmed cases were over 200 in 2003) and H1N1 (for
countries whose conrmed cases were over 5000 in 2009) enter
our models to control for the effect of these diseases in the relatively small sample.
Even though the coefcients of the WHSs are not quite constant
between Models (19) and (26), their marginal effects are all positive. After controlling other explanatory variables, the estimated
results show that the positive effect of WHSs is quite robust so that
the sign would not change for the different subgroups. In Section
4.1, we know that as the number of WHSs increases, the marginal
effect declines, and then rises after a country possesses sufcient
WHSs. However, in Models (19)e(22), this U-curve is not obvious in
each region. For each region, the marginal effects of the WHSs
exhibit different patterns. In Africa, where the numbers of WHSs
are all below 10, the marginal effect is less for WHS-poor countries
than for WHS-rich countries. In Asia, international tourists are
mainly attracted by countries possessing fewer than 10 WHSs. On

Table 5
Estimated results of international tourist arrivals (by region).
(19)
Africa

(20)
Asia

(21)
Europe

(22)
America

WHS (11e20)

311945.43
[1.48]
700653.37***
[5.26]
668600.19***
[8.44]
e

WHS (21 up)

569240.22
[1.21]
306748.19
[1.18]
578652.66***
[2.35]
27387.03
[0.18]
52326.23
[0.31]

1556751.62***
[3.48]
895027.86***
[4.09]
633983.70***
[4.54]
428506.56***
[3.38]
465289.26
[1.41]

1491710.19
[0.59]
368894.77
[0.31]
601731.02
[1.50]
1309071.96***
[3.57]
1133299.43***
[5.19]

GDP

6097.11
[0.30]
21793.65
[0.57]
102.11
[0.52]
226.39***
[2.61]
88542.01
[0.80]
119010.42
[0.60]
71233.96
[1.53]

350.46
[0.63]
15756.36***
[4.51]
134.88
[1.06]
245.68***
[4.93]
11224.13
[0.05]
271410.15
[1.10]
78745.46
[1.02]

4338.24
[0.86]
455643.96***
[6.01]
1130.06
[1.00]
278.66*
[1.72]
1270677.58***
[3.07]
450324.73**
[2.04]
107376.01
[0.79]

4629.76***
[2.73]
108256.42***
[2.46]
1709.17
[0.43]
43.47
[0.58]
1696149.2
[0.54]
689745.98
[0.95]
522070.2
[1.34]

AREA
YEAR
Constant

No
Yes
2510878.64
[1.56]

No
Yes
92106.25
[0.03]

No
Yes
2528566.8
[1.11]

No
Yes
19603323.34
[1.14]

R-square
Chi-square
Observations

0.956
719.79***
53

0.542
109.96***
126

0.891
1003.18***
145

0.997
5060.39***
35

WHS (0e3)
WHS (4e6)
WHS (7e10)

POP
EX
RAIL
FREEDOM
HEALTH
EDU

1. t statistics are in parentheses.

2. *, ** and *** denote signicance at the 10%, 5% and 1% statistical levels.

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

55

Table 6
Estimated results of international tourist arrivals (by time period).

WHS (0e3)
WHS (4e6)
WHS (7e10)
WHS (10e20)
WHS (21 up)
GDP
POP
EX
RAIL
FREEDOM
HEALTH
EDU

(23)
2000e2001

(24)
2002e2003

(25)
2004e2006

(26)
2007e2009

479370.55
[0.08]
411201.49*
[1.82]
343859.20**
[2.09]
360265.29***
[2.54]
376666.37***
[2.77]

403720.74
[1.05]
394058.86
[1.32]
307086.84**
[2.08]
245966.27*
[1.88]
542954.80**
[4.32]

613022.27*
[1.65]
442751.21*
[1.92]
510984.50***
[2.64]
271689.13*
[1.70]
834250.42***
[5.95]

1227144.45***
[4.47]
309651.41
[1.50]
284435.48*
[1.70]
224635.71
[1.48]
909319.29***
[4.93]

685.57
[1.04]
18545.22***
[3.36]
1047.28
[0.90]
179.00***
[3.73]
256215.71
[0.63]
311163.79
[0.78]
24163.51
[0.32]

828.67
[1.47]
21720.38***
[4.94]
475.25
[1.38]
162.64***
[3.94]
527021.99
[1.43]
340883.73
[1.36]
106243.45*
[1.91]

853.48
[0.91]
4842.39
[0.27]
344.05
[0.81]
52.80***
[3.69]
13912.68
[0.04]
2511.2
[0.01]
64066.16
[1.11]

711.92
[0.64]
175.16
[0.01]
193.14
[0.75]
318.41***
[4.22]
123137.59
[0.65]
82209.48
[0.62]
50353.33
[1.04]

2306389.69***
[3.72]

SARS

359981.67
[0.57]

H1N1
AREA
YEAR
Constant

Yes
No
3827725.47
[0.58]

Yes
No
2939585.09
[0.81]

Yes
No
2011503.17
[0.57]

Yes
No
2696260.86
[0.86]

R-square
Chi-square
Observations

0.825
170.96
72

0.889
262.29
70

0.839
207.1
120

0.782
188.65
97

1. t statistics are in parentheses.

2. *, ** and *** denote signicance at the 10%, 5% and 1% statistical levels.

the contrary, in Europe and America, the marginal effects of WHSs

are found to be U-shaped, but the turning points are different. In
Europe, the turning point occurs when the number of WHSs ranges
between 11 and 20, while in America, the turning point is located in
the 4e6 WHSs group. Even though the individual pattern in each
region is different, the overall U-shaped pattern can be observed in
Fig. 7(A).

1600

1600
Africa

Asia

Europe

America

(B)
Tourist arrivals(1000)

(A)
Tourist arrivals(1000)

Moreover, in Models (23)e(26), which are classied by four time

periods, the U-shaped features of the marginal effects of the WHSs
are obvious. The marginal effects are particularly large in WHSpoor and WHS-abundant countries, and they are small between
these two groups. This pattern is also displayed in Fig. 7(B). Based
on the time periods, these sub-samples reveal that our model is
quite robust.

1200

2000-2001

2002-2003

2004-2006

2007-2009

7-10

11-20

20 up

1200

800

400

800

400

0
0-3

4-6

7-10

11-20

20 up

0-3

Number of WHSs
Fig. 7. Marginal effects of WHSs (by region and time period).

4-6

Number of WHSs

56

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

Table 7
Contribution of newly-inscribed WHSs in 2011.
Newly-inscribed WHSs in 2011

In-sample countries

Petroglyphic Complexes of the Mongolian Altai

Coffee Cultural Landscape of Colombia

Selimiye Mosque and its Social Complex

The Persian Garden

Hiraizumi (Temples, Gardens and Archaeological
Sites Representing the Buddhist Pure Land)

Ogasawara Islands

Ningaloo Coast

Fagus Factory in Alfeld

The Causses and the Cvennes, Mediterranean
agro-pastoral Cultural Landscape

West Lake Cultural Landscape of Hangzhou

Cultural Landscape of the Serra de Tramuntana

Longobards in Italy. Places of the Power (568e774 A.D.)

Country

# of
WHSs

Marginal effect of
WHSs (1000)
(A)

Average
receipt (USD)
(B)

C
C
C
C
C

Mongolia
Colombia
Turkey
Iran
Japan

3
7
10
13
16

975
498
498
357
357

616
1244
965
1136
1846

600
620
480
405
659

N
N
C
C

Japan
Australia
Germany
France

16
19
36
37

357
357
475
475

1846
4990
1959
763

659
1781
931
362

C
C
C

China
Spain
Italy

41
43
47

475
475
475

838
1141
970

398
542
461

United Arab
Emirates
Barbados
Sudan
Nicaragua
Jordan
Ukraine
Syrian Arab
Republic
Kenya
Kenya
Senegal
Vietnam
Ethiopia

975

1032

1006

1
2
2
4
5
6

975
975
975
580
580
580

2162
712
358
916
209
621

2108
694
349
531
121
360

6
6
6
7
9

580
580
580
498
498

807
807
542
814
3391

468
468
314
405
1689

Out-of-sample countries

Cultural Sites of Al Ain (Hat, Hili, Bidaa
Bint Saud and Oases Areas)

Historic Bridgetown and its Garrison

Archaeological Sites of the Island of Meroe

Len Cathedral

Wadi Rum Protected Area

Residence of Bukovinian and Dalmatian Metropolitans

Ancient Villages of Northern Syria

C
C
C
M
C
C

C
N
C
C
C

Fort Jesus, Mombasa

Kenya Lake System in the Great Rift Valley
Saloum Delta
Citadel of the Ho Dynasty
Konso Cultural Landscape

Contribution of
WHSs (million USD)
(A)  (B)

Type

1. C: cultural site, N: natural site, M: mixed site.

2. Missing data of tourist arrivals in 2009: Iran, Ethiopia (replaced by 2008 data), Senegal (replaced by 2007 data), United Arab Emirates (replaced by 2005 data).

In addition, as for the control variables, the signicant effect

indicates that regional differences exist in Tables 5 and 6. The
effect of population (POP) in Table 5, for example, changes from
negative to positive for Europe. The source of this negative effect
of POP may be the Asian countries. The larger countries in Asia,
such as China or Russia, are relatively poor in terms of security and
sanitation. Thus, international tourists may become hesitant to
select them as their rst choice. On the contrary, Europe is the
most frequently visited region in the world, according to the
UNWTO. The wealth of European cultures, the variety of its
landscapes and the exceptional quality of its tourist infrastructure
are likely to be among the reasons why tourists choose to take
their holidays in Europe. In 2009, ve of the top ten countries for
tourists in the world were in Europe, namely, France, Spain, Italy,
the United Kingdom and Germany, which are the relatively large
and more populated countries in Europe. Thus, the common factor
of these countries, which is not captured by any other control
variables, is captured by the variable POP, which causes the coefcient to be positive.
Moreover, the two global epidemics, SARS and H1N1, have
indeed had negative impacts on international tourism worldwide
in the last ten years. The SARS epidemic resulted in a signicant
reduction of around 2.3 million international tourists in 2003,
while the H1N1 outbreak was insignicant, with the number of
international tourists being reduced by about 359,982 in 2009. In
addition, it should be noted that the size of the subsample
used to perform the robustness check is relatively small so that
the coefcients would become more sensitive than those in
Model (15).

4.3. Evaluation of the economic contribution of WHSs

In this section, we employ our model to calculate the contribution of newly-inscribed WHSs to destination countries. According to
the World Heritage Center, the latest list of newly-inscribed WHSs
reects the 2011 vision. Table 7 lists these newly-inscribed WHSs
and their related economic contributions. In Table 7, the marginal
effect comes from our model, while the average receipt is calculated
by dividing the total tourism receipts in 2009 by the international
tourist arrivals in 2009, based on data provided by the World
Development Indicators. Note that according to the number of
WHSs in each country, our model proves that the marginal effect of
WHS on international tourist arrivals differs from country to country. In addition, these 25 countries are divided into two groups. The
rst is the in-sample country, which is included in the 66 countries
of our panel data, while the other is the out-of-sample country.
The contribution of WHSs is obtained by multiplying the marginal
effect of WHSs by the average receipts (expenditure) of inbound
tourists in destination countries. Because the buying power and travel
costs are different, the average receipts vary across countries, and so
do the contributions of WHSs. Australia, for example, possesses 19
WHSs, and the marginal effect of WHSs on international tourist arrivals is around 375,000. However, the average receipt from international tourists is quite high, amounting to 4990 USD per person.
The forecasted contribution of this newly-inscribed WHS is about
1781 million USD. Comparatively speaking, even though the marginal
effect of WHSs is higher in China than in Australia, the contribution of
WHSs is lower, or around 398 million USD, because the average
receipt is much lower than for other countries. In addition, the

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

economic contribution of WHSs can be seen as the lower bound of the

increase in tourism income while other control variables are unchanged. That is, after considering the changes in other variables,
such as economic growth, the improvement of transportation or
increased political liability, the number of inbound tourists will increase further, and bring in more income from tourism. Thus, the
authorities of the destination countries could refer not only to the
result, but also to the method in order to evaluate the economic
contribution of WHSs and to budget for their conservation.
5. Conclusion
In this paper, we investigate the positive relationship between
the WHSs and international tourist arrivals at the worldwide level
using panel data for 66 countries between 2000 and 2009. We also
study the effect of new inscriptions on the World Heritage List, and
observe how this inuence changes over time.
According to the estimated results, a country possessing one more
WHS would increase its annual international tourist arrivals by
382,637. While both cultural and natural WHSs could enhance the
inbound tourism, the effect of natural WHSs is slightly larger than the
effect of cultural ones. The marginal effects of WHSs on international
tourist arrivals are 396,659 million and 418,606 million for cultural
and natural WHSs, respectively. Moreover, in considering that the
marginal effect may vary with the number of WHSs, we divided our
sample into three and ve equal parts to better understand the
marginal effect of WHSs for a specic range of numbers. After dividing
our data into ve equal parts, the positive effect of WHSs was found to
decline while the number of WHSs actually rose. However, when a
country possesses sufcient WHSs, this effect increases slightly. The
effect of WHSs exhibits a U-shaped pattern as the number of WHSs
increases. In addition, for each region, the marginal effects of WHSs
demonstrate different patterns, but our results remain quite robust in
different time periods.
As for WHS-poor countries, once unknown sites become famous
after being inscribed on the WHS list, there will be more visits from
international tourists. On the contrary, for WHS-rich countries,
which already possess many attractions and are famous in the
global tourism market, adding one more WHS would lead to
smaller increases in inbound tourists than in the case of WHS-poor
countries. However, for the WHS-abundant countries (possessing
more than 21 WHSs), the marginal effects of WHSs increase
instead. This increase means that when a country possesses sufcient WHSs, the gearing effect of WHSs will emerge.
To sum up, increasing the number of WHSs will have a signicantly
and robustly positive effect on international tourist arrivals. Therefore,
a country possessing a WHS is in a win-win situation not only for the
sustainable conservation of cultural achievements and natural resources, but also for the development of the tourism economy.
Moreover, we could say that these two purposes are not contradictory,
but rather complementary. It is because conservation is the only way
to maintain sustainable tourism income from WHSs, and this tourism
income is indispensable for the further preservation of WHSs.
More information will be gained after extending the time span or
the cross section of available data associated with WHSs. Furthermore, extending the data will make the regional analysis more
meaningful and reliable. Even though the positive impact of WHSs
on international tourist arrivals does not change across regions or
time periods in our paper, the coefcient itself is different and deserves further study. In addition, we use international tourist arrivals
as the dependent variable, which captures the international tourism
demand. However, a gap between tourist arrivals and tourist incomes may exist, because the consumption behavior of tourists may
differ across countries. Thus, the exchanges between tourist arrivals
and incomes, compared with the costs of maintaining WHSs, should

57

be explored using cost-benet analysis. These topics are thus both

important and interesting for further research.
Appendix A. 66 countries whose data were used (in
alphabetical order)

Argentina
Belgium
Chile

Armenia
Benin
China

Australia
Brazil
Colombia

Austria
Bulgaria
Croatia

Azerbaijan
Cameroon
Czech
Republic
Egypt
Estonia
Finland
France
Georgia
Greece
Hungary
Indonesia Iran
Ireland
Italy
Japan
Korea
Kyrgyzstan Latvia
Madagascar Malaysia
Mali
Mexico
Moldova
Morocco
Netherlands Norway
Pakistan
Peru
Poland
Portugal
Romania Russia
Saudi Arabia
Slovenia
South Africa Spain
Sweden
Switzerland
Tunisia
Turkey
Ukraine
U.K.
U.S.

Bangladesh
Canada
Denmark
Germany
Israel
Lithuania
Mongolia
Philippines
Slovakia
Thailand
Uruguay

References
Bille, T., & Schulze, G. G. (2008). Culture in urban and regional development. In
V. A. Ginsburgh, & D. Throsby (Eds.), Handbook of the economics of art and
culture). Amsterdam: North Holland-Elsevier.
Bonet, L. (2003). Cultural tourism. In R. Towse (Ed.), A handbook of cultural economics). Cheltenham: Edward Elgar.
Breusch, T. S., & Pagan, A. R. (1979). Simple test for heteroscedasticity and random
coefcient variation. Econometrica, 47(5), 1287e1294.
Cellini, R. (2011). Is UNESCO recognition effective in fostering tourism? A comment
on Yang, Lin and Han. Tourism Management, 32, 452e454.
Chamberlain, G. (1984). Panel data. InHandbook of econometrics, Vol. 2, (pp. 1247e
1318). Amsterdam: North Holland.
Cooke, P., & Lazzaretti, L. (2008). Creative cities, cultural clusters and local economic
development. Cheltenham: Edward Elgar.
Deng, J., King, B., & Bauer, T. (2002). Evaluating natural attractions for tourism.
Annals of Tourism Research, 29(2), 422e438.
Dhariwala, R. (2005). Tourist arrivals in India: how important are domestic disorders? Tourism Economics, 11(2), 185e205.
Dougan, J. W. (2007). Analysis of Japanese tourist demand in Guam. Asia Pacic
Journal of Tourism Research, 12(2), 79e88.
Dritsakis, N. (2004). Cointegration analysis of German and British tourism demand
for Greece. Tourism Management, 25(1), 111e119.
Garin-Munoz, T., & Amaral, T. P. (2000). An econometric model for international
tourism ow in Spain. Applied Economics Letters, 7, 525e529.
Hausman, J. A. (1978). Specication test in econometrics. Econometrica, 46, 1251e1271.
Herbert, D. (2001). Literary places, tourism and the heritage experience. Annals of
Tourism Research, 28(2), 312e333.
Ledesma-Rodriguez, F. J., Navarro-Ibanez, M., & Perez-Rodriguez, J. V. (2001). Panel
data and tourism: a case study of Tenerife. Tourism Economics, 7, 75e88.
Li, M., Wu, B., & Cai, L. (2008). Tourism development of world heritage sites in
China: a geographic perspective. Tourism Management, 29, 308e319.
Light, D. (2000). Gazing on communism: heritage tourism and post-communist
identities in Germany, Hungary and Romania. Tourism Geographies, 2(2), 157e176.
Lim, C. (2006). A survey of tourism demand modeling practice: issues and implications. In International handbook on the economics of tourism (pp. 45e72).
Cheltenham: Edward Elgar.
McIntosh, A., & Prentice, R. (1999). Afrming authenticity: consuming cultural
heritage. Annals of Tourism Research, 26, 589e612.
Maloney, W. F., & Montes-Rojas, G. V. (2005). How elastic are sea, sand and sun?
Dynamic panel estimates of the demand for tourism. Applied Economics Letters,
12, 277e280.
Naude, W. A., & Saayman, A. (2005). The determinants of tourist arrivals in Africa: a
panel data regression analysis. Tourism Economics, 11(3), 365e391.
Patsouratis, V., Frangouli, Z., & Anastasopoulos, G. (2005). Competition in tourism
among the Mediterranean countries. Applied Economics, 37, 1865e1870.
Payne, J. E., & Mervar, A. (2002). A note on modeling tourism revenues in Croatia.
Tourism Economics, 8, 103e109.
Song, H., & Li, G. (2008). Tourism demand modeling and forecasting: a review of
recent research. Tourism Management, 29, 203e230.
Tan, A. Y. F., McCahon, C., & Miller, J. (2002). Modeling tourist ows to Indonesia and
Malaysia. Journal of Travel and Tourism Marketing, 13, 61e82.
Wooldridge, J. M. (2002). Basic linear unobserved effects panel data models. In
Econometric analysis of cross section and panel data (pp. 247e291). Cambridge,
MA: MIT Press.
Yang, C.-H., & Lin, H.-L. (2011). Is UNESCO recognition effective in fostering
tourism? A comment on Yang, Lin and Han: reply. Tourism Management, 32,
455e456.
Yang, C.-H., Lin, H.-L., & Han, C.-C. (2010). Analysis of international tourist arrivals in
China: the role of world heritage sites. Tourism Management, 31, 827e837.

58

Y.-W. Su, H.-L. Lin / Tourism Management 40 (2014) 46e58

Yu-Wen Su received Ph.D. from National Taiwan University (NTU). Her research focuses on econometrics, time
series analysis and tourism economics.

Hui-Lin Lin is Professor of Economics, NTU. She received

Ph.D. from Brown University. Her research focuses on
econometrics, industrial economics and tourism economics.