This item has the following downloads:
1 MODELING TOURIST FLO WS AND ECONOMIC IMPA CTS: A SPATIAL PERSPECTIV E By YANG YANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013
2 2013 Yang Yang
3 To my wife
4 ACKNOWLEDGMENTS I would like to express my most sincere gratitude to my exceptional dissertation committee members. Dr. Timothy Fik, the chair of the committee, guided me with his huge support and great patience throughout the four years. It has been a privilege to have w orked with such a caring and considerate person. He has always been ready and available to help with academic problems or discussions. I am also grateful for the invaluable help and encouragement from Dr. Peter Waylen, Dr. Liang Mao, and Dr. Chunrong Ai. T hey have all provided insightful suggestions and valuable feedback to my dissertation. My graduate years would not have been the same without the support and encouragement of my many colleagues, fellow graduate students and friends in Department of Geogr aphy at University of Florida. Finally, very special thanks go to my wife, parents, and grandparents for their unyielding love, sacrifice and unconditional support throughout my studying. They never ceased to support and always encourage me to do my best.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 LIST OF ABBREVIATIONS ................................ ................................ ............................. 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 11 2 MODELING SEQUENTIAL TOURIST FLOWS: WHERE IS THE NEXT DESTINATION? ................................ ................................ ................................ ...... 18 Chapter Summary ................................ ................................ ................................ ... 18 Background ................................ ................................ ................................ ............. 18 Literature Review ................................ ................................ ................................ .... 21 Destinatio n Choice of Tourists ................................ ................................ .......... 21 Multi destination Tour ................................ ................................ ....................... 22 Spatial Configuration and Tourism Demand ................................ ..................... 24 Model Specification ................................ ................................ ................................ 26 Data Description ................................ ................................ ................................ ..... 29 Results and Discussion ................................ ................................ ........................... 32 Simulation ................................ ................................ ................................ ............... 37 Chapter Conclusion ................................ ................................ ................................ 38 3 SPATIAL EFFECTS IN REGIONAL TOURISM GROWTH ................................ ..... 47 Chapter Summary ................................ ................................ ................................ ... 47 Introduction ................................ ................................ ................................ ............. 48 Literature Revie w ................................ ................................ ................................ .... 49 Regional Tourism Growth Model ................................ ................................ ...... 49 Spatial Effects of Tourism Growth ................................ ................................ .... 52 Research Methods ................................ ................................ ................................ .. 53 Data Descrip tion ................................ ................................ ................................ ..... 59 Results of ESDA ................................ ................................ ................................ ..... 63 Results of Spatial Growth Regression Model ................................ .......................... 65 OLS Estimates ................................ ................................ ................................ 65 SDM Estimates ................................ ................................ ................................ 68 GW SDM Estimates ................................ ................................ ......................... 73 Chapter Conclusion ................................ ................................ ................................ 77
6 4 ANALYSIS OF TOURISM ECONOMIC IMPACTS AND THEIR DETERMINANTS ................................ ................................ ................................ ... 95 Chapter Summary ................................ ................................ ................................ ... 95 Introduction ................................ ................................ ................................ ............. 95 Literature Review ................................ ................................ ................................ .... 98 Economic Impact of Tourism ................................ ................................ ............ 98 Ap plication of Input Output Analysis in Tourism ................................ ............. 100 Industrial Grouping/Complexes Analysis ................................ ........................ 102 Research Methods and Data ................................ ................................ ................ 105 Input Output Analysis ................................ ................................ ..................... 105 Latent Class Model ................................ ................................ ......................... 110 Data Description ................................ ................................ ............................. 112 Results and Discussion ................................ ................................ ......................... 113 Industrial Grouping Analysis ................................ ................................ ........... 113 Total Linkage ................................ ................................ ................................ .. 115 Multiplier Mea sures ................................ ................................ ........................ 117 Latent Class Modeling of Multipliers ................................ ............................... 118 Chapter Conclusion ................................ ................................ .............................. 120 5 CONCLUSION ................................ ................................ ................................ ...... 134 LIST OF REFERE NCES ................................ ................................ ............................. 138 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 154
7 LIST OF TABLES Table page 2 1 Description of subsequent destination alternatives ................................ ................. 41 2 2 ................................ ................................ .... 41 2 3 Estimation results of nested logit models ................................ ................................ 42 2 4 Simulation results of the probability change with the at traction increase ................ 44 3 1 Descriptions of variables used in the tourism growth model ................................ ... 80 3 2 Pearson correlation coefficients between independent variables ............................ 80 3 3 OLS estimates of tourism growth model ................................ ................................ 81 3 4 Estimates of spatial growth regression mod el ................................ ......................... 82 3 5 Decomposition of spatial effects of independent variables ................................ ...... 84 3 6 Estimates of spatial growth regression model with different weighting matrices ..... 85 3 7 GW SDM parameter summary of inbound tourism growth ................................ ...... 87 3 8 GW SDM parameter summary of domestic tourism growth ................................ .... 87 4 1 Descriptive statistics of independent va riables in the regression model ................ 122 4 2 ......................... 123 4 3 ............................ 125 4 4 Tourism economic impact multipliers in different provinces ................................ .. 127 4 5 Estimation res ults of linear regression ................................ ................................ .. 128 4 6 Model fit based on information criteria ................................ ................................ .. 128 4 7 Estimation results of latent class regression on output multiplier .......................... 129 4 8 Estimation results of latent class regression on type I income multiplier ............... 129 4 9 Latent class membership of different provinces ................................ .................... 130
8 LIST OF FIGURES Figure page 1 1 Organizing framework of the dissertation. ................................ ............................... 17 2 1 Location map of Nanjing and subsequent destination alternatives. ........................ 45 2 2 Decision making process and specification of choice sets in the nested logit model ................................ ................................ ................................ .................. 46 3 1 Histogram of lnrate_inb ................................ ................................ ........................... 88 3 2 Hist ogram of lnrate_dom ................................ ................................ ......................... 88 3 3 Local Moran cluster map of the growth rate of inbound tourism revenue ................ 89 3 4 Local G* cluster map of the growth rate of inbound tourism revenue ...................... 89 3 5 Local Moran cluster map of the growth rate of domestic tourism revenue .............. 90 3 6 Local G* cluster map of the growth rate of domestic tourism revenue .................... 90 3 7 Plot of bandwidth against CV score ................................ ................................ ........ 91 3 8 Spatial distribution of the GW SDM spillover coefficient in the inbound tourism model ................................ ................................ ................................ .................. 91 3 9 Spatial distribution of the GW SDM coefficients in the inbound tourism model ....... 92 3 10 Spatial distribution of the GW SDM spillover coefficient in the domestic tourism model ................................ ................................ ................................ .................. 93 3 11 Spatial distribution of the GW SDM coefficients in the domestic tourism model ... 94 4 1 Scree pl ................. 131 4 2 easures .................... 131 4 3 Tourism linkage matrix plot ................................ ................................ ................... 132 4 4 Map of output multipliers in different provinces ................................ ..................... 132 4 5 Map of type I income multipliers in different provinces ................................ .......... 133
9 LIST OF ABBREVIATIONS C GE Computable General E quilibrium C NTA China National Tourism Administration C V Cross Validation ESDA Exploratory Spatial Data Analysis GIS Geographic Information System GW SDM Geographically Weighted Spatial Durbin Model HAC Heteroskedasticity and Autocorrelation C onsistent I I D Independent and Identically D istributed I O Input Output LISA Local Indicators of Spatial A ssociation MLE Maximum Likelihood Estimation OLS Ordinary Least Squares SAM Social Accounting M atrix SAR Spatial A utoregressive SEM Spatial Error M odel SDM Spatial Durbin Model TALC Tourism Area Life C ycle TSA Tourism Satellite A ccount UNWTO United Nations World Tourism Organization
10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy MODELING TOURIST FLO WS AND ECONOMIC IMPA CTS: A SPATIAL PERSPECTIVE By Yang Yang August 2013 Chair: Timothy J Fik Major: Geography Over the past decade the tourism industry has contributed markedly to the global economy I n many less developed regions, tourist led developm ent is considered between developed and under developed areas. Yet within the large body of tourist and the regional study method to gain a deeper understanding of t he industry. The analysis of tourist flows and their economic impact as defined by r igorous economic geography theories and model, is necessary to improve our unders tanding of tourism from a regional perspective Economic geography based analyses will help us gain valuable insights into the spatial process of tourism and tourist led development. The research in this dissertation provides a series of innovative geographical and econometric analys e s to deepen our u nderstand ing of (a) the subsequent decision choices of tourists with respect to sequential tourist destinations ; (b) the spatial effects (spatial dependence and spatial heterogeneity) in regional tourism growth as an outcome of tourist flows ; and c) the related economic impacts and determinants of the associated multiplier effects of th e regional tourism industry.
11 CHAPTER 1 INTRODUCTION With the increase of leisure time and personal disposable income and the relative decline of travel cost, tourism has become increasingly important as an economic activity in a society where recreation and relaxation have become essential to personal development. Although the concept of tourism as an economic activity seems to be intuitive, there is no consensus in the academic world on its definition A frequently used technical definition of tourism comes from United Nations World Tourism Organization (UNWTO), and suggests that t ourism rsons travelling to and staying in places outside their usual environment ... for leisure, business and other ( Medlik, 2003 ) According to this definition, tourism cover s a large spectrum of travel and travel related activities which include tours and t ravel for business, socialization and religious purposes medical care and education. In modern society, tourism is no longer a trivial leisure activity and it exerts significant economic, social and political impacts. First, the tourism industry has be en found to contribute markedly to the global economy. Now t outed as the world s largest industry, tourism account ed for 9% of the G lob al G ross Domestic Product (GG DP ) in 2010. I n many less developed regions touris t led development is now considered as the last straw to grasp in their attempt to create opportunities that will shrink the gap between developed and under developed areas. Within these regions, it is well understood the preliminary stage of tourism growth does not hinge on well developed i nfrastructure or the intens ity of financial investment but rather on creating unique recreational experiences to satisfy potential consumers.
12 Apart from its extraordinary economic contribution, tourism also brings in considerable social, cultural and p olitical benefits. The satisfaction derived from various touris t activit ies provide s a release from the stress of daily life It has been widely accepted that the touris experience and related interpersonal interaction s are important for personal growth and development ( Mannell & Iso Ahola, 1987 ) Moreover, the communication and interaction between the hosts and guests facilitates the social cultural bond between people and places ( origin s and destination s) and provides a vital force for fostering peace ful relations between regions and cultures ( Louis J, 1988 ) Geographers, who are particularly interested in spatial and temporal patterns of activities and flows, have been actively involving in tourism research ( Nepal, 2009 ) As highlighted by Mitch ell and Murphy ( 1991 ) the m ajor contribution of geographic research in tourism lies on the breadth of its perspective, which is found through examining consideration s that are environmental, cultural, social, economic, and geo historical. Moreover, the geographer brings spatial scal e into the analysis, spanning the local, regional, and global aspects of the industry Ge ograph ic theories of spatial flows and interaction s have incorporated notions of distance decay place perception, and place dependence in modeling frameworks that see k to understand the movement of people from a spatial and behavioral perspective Armed with tools such as G eographic Information Systems (GIS) and gravity models geographers have been productive in tourism research ( Nepal, 2009 ) and have gained a reputation ( referred to as the geographic advantage in their ability to solve tourism problems ( Hanson, 2004 ) The study of tourist related movement and the tourism impact on local and regional economies has become a popular topic in geography studies ( Mitchell & Murphy, 1991 )
13 Yet w ithin the large body of geography literature on tourism, few studies have utilized economic geography and the regional study method to gain understanding of th e industry and the implications of regional based tourism for regional economic development Debbage and Ioannides ( 1998, p. 4 ) described this phenomenon as the silence of economic geographers and they ascribed this silence to the non nature of tourism ( Debbage & Ioannides, 1998 ) However, with the boom of t he tourism industry at both the global and regional scales, the analysis of the tourism industry using rigorous economic geography theories and models seems necessary and even urgent to improve our understanding of tourism from a regional perspective. It i s the economic geographer that offers much in the way of facilitating analyses that can help us gain insights into the spatial process of tourism and tourist led development, as well as local and regional tourism demand assessment s and marketing strategies Within the framework of economic geography and regional studies, this research applies various spatial and economic models to investigate three issues in tourist flows and economic impacts. The first two issues focus on the process and mechanism of tourist flows from demand si de and supply side perspectives (Chapters 2 and 3, respectively ) whereas the third issue concerns one of the major impacts that tourist flows exert on the local community, the economic impact (Chapter 4) (Figure 1 1). These thr ee issues are interconnected with each other. First, to gain a comprehensive picture of tourist flows and tourism growth, both demand side and supply side analys e s are indispensable. The demand side study shed s light on the consumer related factors influencing tourist flows, such as socio demographic attributes of individual tourists
14 whereas the supply side study looks into destination related determinants of tourist flows from a regional perspective Second, both demand side and supply side factors would relate to the economic impact of tourism For example, the expenditure pattern of individual tourists and local industrial structure s inevitably influence how the economic benefits are distribute over various economic sectors of the local and region al economy and how much the total benefits could be Chapter 2 focus es on the modeling of multi destination tourist movement from the demand side. Although a large segment of the tourism literature has been devoted to tourism demand analysis ( Baltas, 2006 ; Crouch, 1994a 1994b 1995 ; Ryan, 2003 ; Song & Li, 2008 ; Song & Witt, 2000 ; Song, Witt, & Li, 2003 ) much less attention has been paid to decision making process es as defined in a multi destination tourism context ( Stewa rt, 1997 ; Tideswell & Faulkner, 1999 ; Wu, Zhang, & Fujiwara, 2011b ) The single destination assumption adopted by most destination choice models is problematic and lack s validity given it does not take into account considerations for how the organizational aspects of tourist destinations affect tourist movement and spatial demand. To fill in th is research gap, three level nes ted logit model incorporating spatial configuration fact or s is applied s T he approach taken in this dissertation represents one of the first a ttempts to consider spatial configuration in a micro level oriented multi destination tour ism choice context Chap ter 3 focuses on two spatial effects or outcomes: spatial depen dence and spatial heterogeneity. These effects/outcomes are examined in a supply side regional tourism growth model. The importance of various destination attributes such as infrastructure, ur banization economies and localization economies is also considered
15 and analyzed according to their contri butions to regional tourism growth under the rigorous empirical framework of a spatial growth model ( Capone & Boix, 2008 ; Lazzeretti & Capone, 2009 ) To date no known tourist research has considered the se spatial effects in tourism growth i n an integral way; and hence, this study fills a gap in the literature. More specifically, this study will introduc e a geographica lly weighted spatial Durbin model to account for both spatial dependence and spatial heterogeneity. The modeling approach adopted in this dissertation will provide output that will enable national and regional governments to better propose a nd target specific tourism development plans as they seek to take full advantage of various destination related factors. Chapter 4 focus es on the various economic impacts of tourism and the variability of impact across different regions. In the tourism eco nomic impact literature, a large variation in the calculated multipliers has been observed ( Archer, 1995 ; Archer & Fletcher, 1996 ; Heng & Low, 1990 ; Mbaiwa, 2005 ; Stoeckl, 2007 ) Therefore, it is necessary to recognize efits acr oss regions so that reasonable estimates on multipliers can be found once region specific determinants of economic impact are identified ( Baaijens & Nijkamp, 2000 ; Baaijens, Nijkamp, & Van Montfort, 1998 ; van Leeuwen, Nijkamp, & Rietveld, 2009 ) To fill the research gap an input output (I O) modeling framework will be applied. Relying on inter sector al linkage measures from an I O table encompassing regional economic linkages across 30 provinces in China the impact of tourism will be highlighted. S everal multiva riate statistical analysis techniques will be employed to uncover the pattern of the forward and backward linkage structure s rel tourism industry.
16 Furthermore a latent class regression model is applied to analyze determinants of the tourism economic multipliers and identify several latent classes of tourism economic impacts based on the sample of
17 Figure 1 1. Organizing framework of the dissertation Demand Side Supply Side Tourist Flows Social Impact Economic Impact Environmental Impact
18 CHAPTER 2 MODELING SEQUENTIAL TOURIST FLOWS: WHERE IS THE NEXT DESTINATION? Chapter Summary This chapter analyzes the subsequent decision choices of tourists with respect to sequential destinations: why they visit a given destination after visiting a previous one and where are they most likely to go. Using a dataset from an on site tourist survey in Nanjing, China, this decision process is partition ed it into three stages and a nested logit model is used to estimate the determinants in each stage. Apart from tourist individual characteristics and destination attribut es, it is fou nd that the spatial destination choice. Finally, a series of simulations are carried out to understand the competition/substitution patterns between subsequent destinations. Background The study of tourists destination choice is an important topic in tourism studies as it shed lights on the nature and patterns of individual tourism demand The focus on the decision making process of individuals provides a platf orm that more accurate ly reflect s demand level s from particular tourist segment s rather than relying on aggregate demand models that tend to overlook the determinants of destination choice ( Baltas, 2007 ) The majority of approaches investigating tourists destination choice have been couched in traditional framework s that highlight origin to destination tourist movement These studies assume that th ere is only one destination in tourists itinerary ( Nicolau & Ms, 2008 ; Santos, Ramos, & Rey Maquieira, 2012 ; Wu, et al., 2011b ) Under this single destination assumption, it is believed that tourists itineraries in light of numerous alternatives are made according to the expected utility associated with v isiting various
19 destination s that are part of typically shown toward destinations that offer the greatest amount of expected utility, in relation to i ndividual characteristics or preferences of the tourists and particular attributes of the destinations in question ( Seddighi & Theocharous, 2002 ; Wu, Zhang, & Fujiwara, 2011a ) Although a large segment of the tourism literature has been devoted to investigating tourists single destination choice, much less attention has been paid to this decision making process in multi destination tour circumstance. Along with a mu lti destination tour, tourists gain extra utilit y from experienc ing a series of different destinations or diversified tourist attractions and activities. In order to maximize potential utility while minimize expense s to a particular region/area, tourists t end to visit to more than one destination in a single tour ; taking advantage of locations or regions that offer bundled tourist sites or activities ( Smith, 1983 ) Therefore, the single destination assum ption adopted by most destinati on choice models is problematic and lacking validity given it does not take into account considerations for how the organizational aspects of tourist destinations affect tourist movement and spatial demand More importantly, according the theoretical microeconomic models by S antos, Ramos and Rey Maquieira ( 2011 ) the determinants of destination choice under a single destination trip framework might be different from those couched under the multi destination hypothesis. In modeling individual tourist destination choice, another category of determinants tha t previous research has ignored is the geographic factor. While f actors like the effects of spatial structure -the organizational layout and attributes of tourist
20 destinations -as well as competition between destination s and the impact of intervention op portunities or substitute locations have been used in aggregate tourism/recreation demand modeling ( Fesenmaier & Lieber, 1985 ; Hanink & Stutts, 2002 ; Kim & Fesenmaier, 1990 ; Mings & Mchugh, 1992 ) none have been included in micro level or individual touris t demand models t o capture or model sequential tourist flows in relat ion to the demand for multiple tourist locations. Spatial configuration is an important concept to consider in th at it is a reflection of a long term process an d the degree to which alternative destinations or activities locate over geographic space ( Bernasco, 2010 ; Elhorst & Oosterhaven, 2006 ; Pellegrini & Fotheringham, 2002 ) A s tourists at given areas are likely t o visit neighboring area s o r proximate secondary destination s the consideration of spatial configuration offers many possibilities to measure the potential attractiveness of a destination based on th e attractiveness of nearby destinations that are part of a spatial bundle Therefore, spatial configuration effect s must be taken into account to measure the extent to which various locations may have augmented attractiveness from multi destination and secondary tourist locations found within or near primary tourist locations. To fill in the research gap discussed above, a nested logit model incorporating a spatial configuration factor to estimate tourists choice of subsequent destination is applied; based on information obtained from an on si te tourist survey in Nanjing. A three level decision making process is considered First, a tourist at a primary tourist location is assumed to first consider whether to move on to a subsequent destination or return home. Second, those opting to continue m ust then select the type of subsequent destination Thirdly, they must finally choose a specific subsequent destination within
21 that particular type from the subset of feasible destinations that fall within that type Th is multi stage destination choice pro cess has been validated in the context of single destination tourism ( Eugenio Martin, 2003 ; Nicolau & Ms, 2005 2008 ; Seddighi & Theocharous, 2002 ) yet the link to how the decision making process is affected by spatial configuration is largely absent in the literature T o the be st of my knowledge, this study represents one of the first app lications to consider spatial configuration in the context of micro level oriented multi destination tour. Literature Review Destination Choice of Tourists R esearch on tourist destination choice has ad o pted random utility theory, a classic micro economic t heory that emphasizes rank order utility maximization among a set of alternative destinations ( Huybers, 2003a 2003b 2005 ; Nicolau & Ms, 2008 ) It states that when tourists make decisions, they compare the utility of each choice for a given destination choice set Though utility is an unobserved quantity, one could observe people s actual choice ; and hence the rank of utility of each choice al ternative can be deduced from observation In the case of tourist destination choice, a tourist is assumed to compare the utility he/she expects to obtain from visiting a particular destination and cho o se s the one with the greatest expected utility. T he us e of discrete choice model s is compatible with the economic principle of utility maximization, where the utility of a given destination is a function of tourists soci o demographic profile s and the attractiveness of destination s as defined by their attribu tes Since tourist destination choice is regarded as a comp licated decision making process with a large number of heterogeneous choices, many studies have suggested that this process involves multiple stages ; where tourists are assumed to make
22 decisions sequential ly ( Eugenio Martin, 2003 ; Nicolau & Ms, 2005 2008 ; Seddighi & Theocharous, 2002 ) Two types of discrete choice models have been applied to validate this multi stage process : the nested logit model ( Huybers, 2003a 2003b 2005 ; Wu, et al., 2011a ) and the error compone nt logit model ( Nicolau & Ms, 2005 2008 ) In the former model a nested structure is specified for individual stage s of the decision making process while in the latter model (which involves a special type of a mixed logit model ) decisions at di fferent stages are captured by specifying random effects to represent the nest ed structure. Although much of the current tourist literature is devoted to analyzing tourist destination choice, nearly all of these studies assume that tourists choose only one destination ( Lue, Crompton, & Stewart, 1996 ; Tideswell & Faulkner, 2003 ) Hence, the choice outcome of the empirical model is a single destination that is typically the primary destination Two exceptions come from Nicolau and Ms ( 2005 ) and Wu, Zhang, and Fujiwara ( 2011b ) Nicolau and Ms ( 2005 ) examine d the decision to take multi or single destination vacations while Wu, Zhang, and Fujiwara ( 2011b ) applied a universal logit model to analyze destination choice with future dependence. As multi destination tour ist floes are very common, it is essential t o f urther our understanding of multiple destination choices as they relate to the attributes and the geography of destinations and tourist related policy implications. Multi destination Tour In tourism research, much attention has been paid to understanding t he factors contributing to destination choice, intention and the extent of multi destination tour This chapter has largely focused on the questions of why tourists are visit ing more than one destination and how many destinations they will stay during a s ingle tour? These efforts
23 generally include both supply side and demand side factors. S upply side factors indicate the opportunity to visit distinct destinations based on location specific attributes, while demand side factors, such as individual tourist characteristics desire to visit more than one destination based on the need to fulfill an overall tourist experience ( Wu & Carson, 2008 ) On the supply side, it has been shown that the comp atibility to a previous destination is an important consideration when tourists are making decisions regarding their next destination choice As argued by Bristow, Lieber and Fesenmaier ( 1995 ) and Jeng and Fesenmaier ( 1998 ) the increase in utility from multi destination tour is conditional on the compatibility of each destination included in the tour. Jeng and Fesenmaier ( 1998 ) used the term similarity to describe the compatibility between destinations and asserted that destinations which are geographically proximate to each other are more likely to be perceived as similar destinations by tourists. However, according to Lue, et al. ( 1996 ) the cumulative attractiveness will be enhanced if the secondary destination is dissimilar fro m the previous one by offer ing a more diverse tourist experience Whether the cumulative attraction derives from similar or complementary (diverse) attractions is a matter of speculation and depends greatly on the context and nature of the tourist experien ce and the needs of the individuals within a group. Nonetheless, Lue et al ( 1993 ) have argued that multi destination choice will be affected by spatial structure as well as the geographic scale at which the research is being conducted From the demand side, a large number of studies have examined the determinants of multi destination tourism concluding that s ingle and multi destination
24 tourists differ in terms of demographic profile s trip types motivations and intentions service quality assessment s and their propensit ies to participat e in cultural activities ( McKercher & Wong, 2004 ) Various social economic characteristics of individual tourist are found to determine tourists inten t ion and extent or duration of multi destination tourism Multi destination tour patterns can vary greatly depending on to urist motivation, travel distance, familiarity of sites based on past visit ation party size and organization, mode(s) of transport, time and cost constraint s and information sources ( Tideswell & Faulkner, 2003 ) Although past research has shed light on the various factors determining inte ntion and the extent of multi destination tour as related to their needs a nd characteristics there remains a significant gap in our understanding of why tourists select a particular destination during a multi destination tour. To investigate these factors, many statistical and econometric methods have been applied such as conj oint analysis ( Lue, et al., 1996 ) linear regression models ( Bristow, et al., 1995 ; Tideswell & Faulkner, 1999 ) and count data models ( Santos, et al., 2012 ) The use of discrete choice model s to study multi destination tourism has been limited to the work of Nicolau and Ms ( 2005 ) and Wu, Zhang, and Fujiwara ( 2011b ) However, u p to now, no known research in this area has employed a discrete choice model to examine the exact subsequent destination choice in situa tion s specifically involving multi destination tour. Spatial Configuration and Tourism Demand As tourists decision making process on destination choice can be regarded as a spatial choice process, it is necessary to review some background theor ies of spatial choice as couched from a geograph ic perspective. Pellegrini and Fotheringham ( 2002 ) have stated that several issues are raised specifically for spatial choice decision and
25 emph asized the importance of examin ing the inherent spatial patterns among observed choices. Because of limitations to people s ability to process large amounts of information, a hierarchical information processing strateg y is likely to be used when the destin ation choice set is large. In order to capture th e decision making process, spatial configuration factors should be included in the spatial choice model ( Fotheringham, 1988 ) To reproduce the real situation of the spatial choice, various spatial configuration factors have been incorporated in discrete choice modeling ( Arentze, Oppewal, & Timmermans, 2005 ; Bernasco, 2010 ; Cascetta, Pagliara, & Papola, 2007 ; Cascetta & Papola, 2008 ; Pellegrini & Fotheringham, 2002 ) A common measurement of th e spatial configu ration effect is the accessibility of an alternative to all other alternatives When used in a spatial interaction context, the estimated coefficient for location accessibility provides an indication of the competition/substitution patterns between differe nt destination alternatives ; where a positive coefficient suggests an agglomeration effect of alternatives while a negative coefficient suggests a competition effect ( Fik, 1988 ; Fik, Amey, & Mulligan, 1992 ; Fotheringham, 1985 ) In aggregate tourism demand research, many research ers have emphasized the importance of spatial configuration effects ( Fesenmaier & Lieber, 1 985 ; Hanink & Stutts, 2002 ; Kim & Fesenmaier, 1990 ) and the need to incorporate the impact of spatial structure in modeling tourist movement ( Koo, Wu, & Dwyer, 2012 ; Lin & Morais, 2008 ; Tideswell & Faulkner, 1999 ) Due to the lack of attent ion paid to spatial structure effects, many destination choice models lack transferability between study areas ( Lue, et al., 1993 ) It is now accepted that multi d estination decision making and behavior is dependent on the spatial configuration of destinations ( Hwang, Gretzel, & Fesenmaier,
26 2006 ) More specifically, measures of centrality, connectedness /connectivity, directionality and network structure have been f ound to influence aggregate multi destination tour patterns of tourists in Midwest ern states in the U.S. ( Hwang & Fesenmaier, 2003 ) However, to date, no known research has incorporated spatial configuration factors to explain the destination choice of individual tourists. In essence, this study will be one of the first attempts to investigate the spatial configuration effect in disaggregate tourism demand research with individual tourist data. M odel Specification In this chapter a three stage decision making process is specified. A fter visiting a particular destination tourists are assumed to make decisions regarding their next stop sequentially. In the first stage, they choose to move on to the next destination or go back home I n the second stage, conditional on visiting subsequent destinations, they are facing a decision on the type of subsequent destinations to incorporate the hierarchical information processing strategy ( Pellegrini & Fotheringham, 2002 ) In the th ird stage, tourists choose the exact destination based on the type they select in the second stage. In order to estimate the three stage subsequent destination choice model, t ourists are assumed to make a rationalized decision and select the destination th at maximizes their individual utilities. A natural and popular estimation procedure that is consistent with this assumption is the nested logit model. Compared to other discrete choice models, like the multinomial and the conditional logit models, it is in line with independence from irrelevant alternatives, by assuming the sequential process of making choices. This suggests that all choices are grouped within particular nests. Along with the decision making process, the individual makes the choice of nests first,
27 and then choose alternatives within that nest. First, I specify the total utility function of final choice t (given first stage choice m and second stage choice r ) as: (2 1) where the overall utility is represented as the sum of two parts: the systematic or deterministic component, and the stochastic component, which is assumed to follow a Gumbel distribution Assume that one can decompose the stochastic component as a sum of utility from different stages and use explanatory variables at different stages to explain the corresponding utility, the systematic component becomes: (2 2) where Z Y and X are the row vector s of variables for the first second and third stage decision making, respectively. Let C 1 C 2 and C 3 represent the choice variables for the first second and third stage, respectively. C onsistent with utili ty maximization assumption, the probability of visiting subsequent destination t ( C 3 = t ) can be written as a product of the conditional probabilities in each stage: and
28 (2 3) w here N r is the number of alternatives in the nest of C 2 = r N m is the number of second stage alternatives in the nest of C 1 = m and N is the number of first stage alternatives. Moreover, and (2 4) Note that , and are called inclusive value reflecting the expected value of utilities derived from all alternatives within the nest. The coefficient on the inclusive value, measures the relevance of the nested structure, and this value should fall between 0 and 1 to be consistent with random utility maximization ( Hensher, Greene, & Rose, 2005 ) T he NLOGIT 5.0 p ackage is utilized to estimate this model via fully information maximized likelihood estimation, which estimate the parameters simultaneously in the likelihood function. The contribution of an observation i to the log likelihood of the model is: (2 5) To estimate the model defined above one must first specify the determinants for decision s made at different stages ; that is, to choose independent variables Z Y and X in the utility function. For Z variables determining the first stage decision making, I include individual tourist s characteristics, like night (nights of stay in Nanjing) pastvisit (number of previous visits to Nanjing, 1=0; 2=1; 3=2 3; 4= 5 and above) age (age of tourist, 1=14 an d below; 2= 15 24; 3=25 44; 4=45 64; 5=65 and above) motivatio n (1=vacation; 2=sightseeing; 3=VFR; 4=others) organization (1=by affiliations; 2=with
29 families and friends ; 3=by travel agencies; 4=alone). For Y variables influencing the second stage decis ion on selecting the destination type, I select another individual tourist characteristic variable, distance1 which measures the distance from tourist s residence to Nanjing (in 100km). Finally, for X variables explaining the final choice of the exact su bsequent destination, a set of destination attributes are chosen. It includes attraction (number of AAAA scenic spots in the destination), distance2 (distance from Nanjing to the subsequent destination alternative, in 100km), distance3 (distance from the s ubsequent destination alternative to residence, in 100km), and CD which is the competi ng destination effect specified as : (2 6) where t indexes a particular subsequent destination alternative, and j indicates other destination alternatives. d tj is the distance between destinations t and j while d t is the distance from Nanjing to Destination t Following hierarchical competition framework ( Fik, 1988 ; Fik, et al., 1 992 ) only destinations shown in Table 2 1 are counted in the calculation of variable CD Data Description The dataset used in this research comes from the provincial domestic tourist on site survey of Jiangsu province in 2007. The Jiangsu domestic to urist survey is conducted by Jiangsu Tourism Administration, and the questionnaires are distributed to domestic tourists in scenic spots and hotels of thirteen cities in Jiangsu. In this survey, various questions cover individual social demographic informa tion, trip characteristics, and trip satisfaction. As far as I am concerned, this survey is one of the most comprehensive domestic tourist surveys in China, considering its sample size, the
30 scope and variety of questions, and the heterogeneity of surveyed tourists. The responders were screened on the basis of two criteria. Only those traveled to a destination at least 10 km away from home and spent at least 6 hours for tourism were retained in the sample. Note that in this study only the data from Nanjing is considered the capital of Jiangsu province. As the overarching research goal, I focus on tourists subsequent destination choice after visiting Nanjing. Table 2 1 presents the top ten tourists choice f or subsequent destinations. They are Beijing, Shanghai, Hangzhou, Suzhou, Huangshan, Nantong, Changzhou, Yangzhou, Wuxi and Zhenjiang As suggested by Table 2 1, the market share of each subsequent destination is positively related to the number of tourist attractions and negatively associated with th e distance from Nanjing. For example, although there are a large number of tourist attractions in Beijing, due to its large distance from Nanjing, only a small proportion of tourists will choose Beijing as a subsequent destination after visiting Nanjing. T able 2 1 also provides the description of these ten subsequent destinations. It shows that they vary considerably in population size the number of tourist attraction s and the distance from Nanjing. The location s of these destinations are show n in Figure 2 1 Located in Eastern China, Nanjing is the capital of Jiangsu Province, an important city on the middle and lower reaches of the Yangtze River. Six of ten subsequent destinations are located in the same province as Nanjing, namely, Yangzhou, Zhenjiang, Changzhou, Wuxi, Suzhou and Nantong. Figure 2 1 demonstrates their locations relative to Nanjing. Eight of ten destinations locate in Yangtze Delta Area. The other two, Huangshan and Beijing, are relatively distant from Nanjing. One in Anhui Province to t he south of Nanjing, and
31 the other is national capital to the north. According to their distance to Nanjing and their attractiveness measured by the number of AAAA scenic spots, I categorize these ten subsequent destinations into two groups ( Table 2 1). Th e first group (Type A and C 2 = 1 ) includes Shanghai ( C 3 = 1) Suzhou ( C 3 = 2) Hangzhou ( C 3 = 3) Huangshan ( C 3 = 4), and Beijing ( C 3 = 5) They are distant from Nanjing, and enjoy the reputation as a national wide destination with ample tourist attractions. The other group (Type B and C 2 = 2 ) includes cities nearby Nanjing with less tourist attractions, like Yangzhou ( C 3 = 6) Changzhou ( C 3 = 7) Zhenjiang ( C 3 = 8) Nantong ( C 3 = 9), and Wuxi ( C 3 = 10) This categorization of destination s was also corroborated by the preliminary estimation results from a degenerate nested logit ( Hensher, et al., 2005, pp. 577 580 ) Figure 2 2 illustrates the decision tree embedded in the three level nested logit model. In the first stage, tourists visited Nanjing choose to move on to the next destination ( C 1 = 1) or go back home ( C 1 = 2). I n the second stage, I consider two types of subsequent destinations : Type A and Type B as previously specified. In the third stage, the alternative destinations include five in Type A and five in Type B. There are totally eleven alternatives for final choice: ten subsequent destinations and home. It should be noted that, tourists may face different alternatives in the choice set. If they are living in one of these ten subsequent destinations, that destination should be excluded from the choice set, leaving ten alternatives for the final choice. According to the report published by Nanjing Tourism Administration, the total number of domestic tourist arrivals to Nanjing was 44.89 million in 2007. Of these tourists, 26.48% came from Jiangsu Province and 9.52% from Zhejiang Province. Their average ex penditure in Nanjing was 1,131 Yuan RMB, with 17.09% spent on
32 accommodation and 14.53% on food and beverage. Table 2 2 presents the descriptive statistics for the dataset used in this chapter In total the dataset consists of 3,055 observations of tourist s in Nanjing. The top panel shows statistics for continuous variables, while the bottom panel shows those for categorical variables. As shown in the table, the tourist in Nanjing covers a broad geographic range, and the travel distance to Nanjing ( distance 1 ) ranges from 44.5 km to 3849.7 km, with an average distance of 538.8 km. In terms of tourists nights of stay, although the average value is 1.823, some tourists tend to stay far longer, such as the maximum value of 14 nights. For categorical variables the table shows nearly half of the tourists in sample are first time tourists to Nanjing, and 58% aged between 25 and 44. In terms of tourists organization patterns, 44.58% of them are traveling with families and friends, while 26% travel alone. As for to urism purposes, 66% of tourists are sightseeing tourists while only 20% are vacationing tourists. Vacation and sightseeing tourists account for 86% of total tourists. The descriptive statistics of destination variables for each destination alternative can be found in Table 2 1. Results and Discussion Table 2 3 presents the estimation results for the specified nested logit model. Column one provides the results for the model using the whole sample. The estimated inclusive values fall between 0 and 1, indicat ing that the specified nest ed structure ( Figure 2 2 ) is appropriate. For the first stage decision making process, a positive and significant estimated coefficient indicates that the variable is positively associated with tourists intention to move onto th e next destination. Among various independent variables, night is estimated to be negative and statistically significant, suggesting that a tourist with a longer duration of stay in the previous destination Nanjing, is less likely to
33 continue their tour t o the next destination. This can be contributed to the time constraint facing by tourists given that tourists time budget is fixed before the visit ( Lew & McKercher, 2006 ) In other words, once a larger p ortion of time has been spent on a particular destination, it is difficult for the tourist to arrange sufficient time to visit other destinations ; therefore, reducing the likelihood of moving on to a subsequent destination. Another variable, pastvisit is also estimated to be negative and significant. It suggests that the more times the tourist has been to Nanjing in past, the less probability he/she will come to a subsequent destination after Nanjing. This result is consistent with the findings from Hwang, et al. ( 2006 ) and Tideswell and Faulkner ( 1999 ) The estimated coefficient of age is pos itive and significant, indicating that older tourists are more likely visit subsequent destinations after leaving Nanjing. With regard to categorical factors, as indicated by the negative and significant coefficients of three motivation variables, sightsee ing tourists, VFR tourists, and other motivation tourists are less likely to visit subsequent destinations than vacation ing tourists. Th e result also shows that only one organization pattern variable is significant, suggesting that tourists traveling with families and friends are more likely to continue their tour after visiting Nanjing than other three types of tourists. This is because during this type of tourism, the schedule accommodates the preference from group members, and as a result of this variety seeking nature, they are more inclined towards multi destination tour ( Tideswell & Faulkner, 1999 ) For the second stage decision making process, a positive and statistically significant c oefficient of distance1 suggests that, the longer the distance from tourist s
34 residence to Nanjing, the more likely he/she will choose Type A destinations, which is characterized by longer distance from Nanjing. This result is consistent with tourists uti lity maximization theory: to make the time and money costs associated long distance to Nanjing worthwhile, they would come to other reputed destinations to maximize the utility within a single tour ( Tideswell & Faulkner, 1999 ) In the third stage, several destination attributes are found to be important in determin ing tourists choice of subsequent destinations. The coefficient of attraction is positive and significant, indicat ing that bundled attractions provide an inducement to attract tourists, and the larger the bundle the greater the inducement. CD is found to be positive and significant, suggesting th at the spatial configuration effect plays a salient role in the tourist dest ination choice process or in other words, that tourist destinations close to other destinations with a large number of attractions are generally more attractive than isolated tourist destinations for multi destination travelers leaving Nanjing The signif icant and positive coefficient CD is consistent with the findings of travel demand ( Newman & Bernardin, 2010 ) but different from the results in crime location choice ( Bernasco, 2010 ) and inter metropolitan migration ( Pellegrini & Fotheringham, 1999 ) A po sitive competition destination effect highlights the agglomeration effect among various destinations and attractions ( Weidenfeld, Butler, & Williams, 2010a 2010b ) One possible explanation is that the cumulative attraction would increase tourist arrivals by bundling with proximate destination s ( Hunt & Crompton, 2008 ) because tourists might visit them as well during a multi destination tour. Finally, distance 2 and distance 3 are estimated to be significant. A negative coefficient of distance2 suggests that the proximate destination to Nanjing are more
35 likely to be chosen as a tourists subsequent destination, while a positive coefficient of distance3 impl ies that tourists are more likely to choose to subsequent destinations which are distant from their residence, supporting Hwang and Fesenmaier ( 2003 ) s finding that the roundtrip distance by multi destination tourists is larger than that of single destination tourists. To further examine the decision making process for different types of tourist s, I split the sample separately into different groups and estimate the model with these sub samples. Columns two t hrough five in Table 2 3 present the estimation results from tourists with distinct motivations. The sample is split into four sub samples ac cording to In column four, since there is only one or two tourists traveling with families and friends or alone, I estimate the model by excluding these samples to avoid computational difficulty. For the first stage decision makin g, night is estimated to have the greatest impact on vacationing tourists. This can be explained by the sensitivity of time constraints of vacationing tourists after their longer stay in a particular destination ( Yang, Wong, & Zhang, 2011 ) The coefficients for pastvisit and age are estimated to be significant for only sightseeing and VFR tourists. Different from what I discovered from the whole sample model, sightseeing and VFR tourists organized by travel agencies are less likely to continue their trips to a subsequent destination. This can be explained by the fact that these tourists are constrained by the pre arranged schedule by travel agencies, and this constraint reduc es the probability of multi destination tour. Furthermore, the result indicates that organization pattern does not influence the first stage decision making for vacationing tourists.
36 In the second stage, the travel distance from tourists residence to Nan jing is found to be significant for sightseeing and VFR tourists, while in the third stage, the attraction and competition destination variables are significant for only vacationing and sightseeing tourists, suggesting that AAAA scenic spots specified in t he model might not be the major attractions for VFR and other motivation tourists. By comparing the magnitudes of estimated coefficients, it is f ou nd that the coefficients of attraction and CD are largest for vacationing tourists and the coefficient of dis tance2 is largest for sightseeing tourists. These results suggest that vacationing tourists are more attraction based while sightseeing tourists are more distance sensitive With regard to tourists with different organization patterns, the estimation r esults are presented in c olumns six through nine of Table 2 3. The sample is split into four sub The sample of VFR tourists is removed in some models due to computational difficulty. Some differences can be observed by comparing the magnitude, sign and significance of the coefficient estimates from these models. In c olumn six the model for tourists organized by affiliations, distance3 is estimated to be negative and significant, suggesting this type of tourist tends to choose a subsequent destination that is proximate to their residence instead of far away. This indicates a smaller geographic scope f or affiliation organized tour s and it may be explained by the relatively fixed budget for this type of tour. Concerning the model for tourists traveling with families and relatives ( c olumn seven), the result is similar to that from the whole sample model. However, VFR motivation is found to be not significant in making the first stage decision.
37 As for the model of tourists organized by travel agencies ( c olumn eight), a noticeable finding is the relatively large estimated coefficient of attraction suggesting AAAA attraction is an important factor for this type of tourist to choose subsequent destinations. This can be explained by the fact that travel agencies can always get exclusive discount from AAAA scenic spots, and therefore, they always include these attractions in the tour itinerary Finally, the competi ng destination effect is found to be statistica lly significant but only for tourists traveling with famil y and friends. One possible explanation is that as this type of tourist is more likely to engage in multi destination tour ( as implied by the results of the whole sample model), the attraction of a particular destination will be augmented by its nearby destinations if tourists choose to visit them together during a single tour. Simulation One of the most important advantages of discrete choice model is its easy use for simulation ( Hensher, et al., 2005 ) In this study with the estimation results from the whole sample model (co lumn one in Table 2 3), a simulation is carried out based on the scenario of an increase of one additional AAAA scenic spot in the subsequent destination. By comparing the predicted probability before and after this change, one can get a clear picture of s patial interaction and substitution patterns of these destination alternatives. It should be noted that, the increase of an additional attraction does not only change the variable attraction but also enters the variable CD It is assumed that, by a particular increase of attractions, the probability of selecting that destination is increasing while the probability of choosing other destinations decreases. However, since neighbor ing destination s are likely to get benefits from agglomeration
38 effects su ggested by the positive and significant coefficient of CD the agglomeration effect might essentially offset th e probability decrease The simulation results are presented in Table 2 4, and they show that as the attraction of a subsequent destination incr eases, the probability of going home decreases The implication here is that spatial bundling is inducing more tourists to u ndertake multi destination tour. As suggested by the bolded number in the diagonal of Table 2 4, the increase of tourist attraction will contribute to more potential tourist arrivals that, of course, depend on previous level s of tourist attraction. For a destination with limited attraction endowments, the probability increase is larger ; as in the cases of Yangzhou, Zhenjiang and Changz hou, with estimated probability increase s of 0.677%, 0.650% and 0.613%, respectively. Finally, and more importantly one can identify competing destinations and possible collaborative destinations from this analysis For instance, one more AAAA scenic spot in Suzhou will also result in a 0.027% probability increase of tourists to Wuxi. Therefore, Wuxi tends to be a potentially collaborative destination with Suzhou. On the other hand, this increase will induce a relatively large decline of tourists probabil ity to Shanghai (by 0.093% ) and Hangzhou (by 0.108% ). As a result, for tourists from Nanjing seeking subsequent destinations, Shanghai and Hangzhou can be regarded as competing destination s to Suzhou. Chapter Conclusion This research proposed a three stage decision making model to investigate tourists choice for a subsequent destination after visiting a previous one. It distinguishes from previous research by looking into destination choice under a multi destination setting and considering the spatial configuration effect. Based on the estimation result s of a nested logit model, it was found that, tourists age, motivation,
39 organization pattern, duration of stay and past visit ation experience for a previous destination significantly influence decision making in the first stage: whether to move onto a subsequent tourist destination or return home. In the second stage, tourists decision regarding the choice of the type of subsequent destination is based on their distance from the origin to the previous d estination, and while in the third stage, destination attributes like the number of attraction s the competition destination effect, the distance from a previous destination, and the distance between the subsequent destination and their place of residence, were found to play significant roles in choosing a subsequent destination. Utilizing the results of the model, a simulation exercise was undertaken to look into the competition/ substitution pattern between different subsequent destinations. As one of the first research endeavors to investigate individual t ourist destination choice under a multi destination tour circumstance, this research contributes to the literature on tourists destination choice and extends the traditional model by considering multiple and sequential destination visits. Moreover, th is chapter provides statistical evidence of the significance of the spatial configuration effect i n disaggregate touris t demand model. The result suggests that spatial configuration helps to explain tourists destination choice because of the multi destination tour tendencies of tourists. From a policy perspective, based on the results from this research, several implications can be drawn in terms of marketing strategy for destinations and the potential bundli ng of proximate and/or competing destinations As shown in the simulation results table, destinations can find their possible cooperator ; and therefore, carry out synergy marketing with them to enhance the tourist experience
40 Furthermore, through the esti mation and modeling of various effects for different types of tourists, distinct destinations are able to position their own products based on their relative advantage and/or relative geographic location The results of this modeling effort also suggest th at generalizations can be made on various tourist types. For example, some tourist s do not consider the endowment of attraction as important in the decision making process. Others may be more sensiti zed to the travel distance and more apt to engage in organized tourism even if it meant visiting sites that were less endowed As a result, destination s with a limited number of tourist attractions but proximate to Nanjing can target this specific market segment. All in all, this t ype of analysis can greatly enhance the understanding of tourists choice decision and contribute to marketing strategy design s aimed at offering possibilities to the multi destination tourists. Some limitations in this research should be highlighted. Firs t, I only look into one subsequent destination in the proposed model. However, tourists are likely to visit multiple subsequent destination s Therefore, further research can make some extens ions to the nested structure specified to incorporate a more gener al situation and consider numerous sequential destinations Second, as individual heterogeneity of tourists has been documented in tourist destination choice research ( Nicolau & Ms, 2006 ) further research can utilize models incorporating individual heterogeneity to analyze tourist behavior d uring multi destination tour. That behavior may be flexible/adaptive or fixed depending on the type of tourist activity and the characteristics of the tourists themselves.
41 Table 2 1. Description of subsequent destination alternatives Name Market share as subsequent destination from Nanjing Population (in 10,000) Distance from Nanjing (in km) Number of AAAA scenic spots CD Type Beijing 1.48% 1,213 884 37 0.001 A Shanghai 21.34% 1,379 267 17 0.052 A Hangzhou 9.60% 672 234 16 0.046 A Suzhou 25.84% 624 190 18 0.078 A Huangshan 4.50% 148 258 11 0.009 A Nantong 0.94% 766 194 2 0.069 B Changzhou 2.01% 357 113 5 0.067 B Yangzhou 17.05% 459 72 4 0.025 B Wuxi 14.63% 462 151 12 0.093 B Zhenjiang 2.62% 269 65 5 0.020 B Notes: all data is as in 2007. CD refers to the competition destination measurement. Table 2 2. Descriptive statistics of tourists profile Continuous data Variable Mean Standard Deviation Minimum Maximum Cases distance1 5.388 4.460 0.445 38.497 3055 night 1.823 1.208 0.000 14.000 3055 Categorical data Variable value=1 value=2 value=3 value=4 value=5 pastvisit 45.696% 25.663% 13.813% 14.828% age 1.408% 15.385% 58.331% 24.255% 0.622% organization 6.743% 58.003% 9.460% 25.794% motivation 19.640% 66.187% 7.692% 6.481%
42 Table 2 3. Estimation results of nested logit models Whole sample (1) Vacation (2) Sightseeing (3) VFR (4) Other motivation (5) By affiliation (6) With families and friends (7) By travel agencies (8) Alone (9) First Stage constant 0.748** 0.805 0.718* 3.069 0.411 3.557** 1.312*** 2.253 0.204 (0.314) (0.805) (0.383) (5.428) (1.452) (1.802) (0.338) (1.522) (0.977) night 0.298*** 0.550*** 0.268*** 0.223* 0.068 0.227 0.317*** 0.329** 0.298*** (0.038) (0.096) (0.049) (0.115) (0.160) (0.165) (0.050) (0.163) (0.072) pastvisit 0.121*** 0.079 0.273*** 0.353** 0.235 0.158* 0.129*** 0.015 0.102*** (0.020) (0.088) (0.049) (0.161) (0.168) (0.081) (0.026) (0.079) (0.038) age 0.282*** 0.019 0.322*** 0.417* 0.360 0.307 0.383*** 0.138 0.100 (0.062) (0.147) (0.077) (0.217) (0.273) (0.280) (0.082) (0.219) (0.118) motivation=2 0.232** 0.199 0.205 0.297 0.339* (0.104) (0.542) (0.134) (0.384) (0.201) motivation=3 0.397** 0.210 0.856*** (0.175) (0.221) (0.313) motivation=4 0.920*** 1.026* 0.615* 0.373 1.149*** (0.202) (0.546) (0.326) (0.774) (0.436) organization=2 0.528** 0.589 0.125 2.081*** (0.226) (0.577) (0.303) (0.807) organization=3 0.269 0.178 0.561** 0.579* 0.275 (0.198) (0.511) (0.276) (0.323) (0.476) organization=4 0.252 0.220 0.587** 0.287 (0.188) (0.490) (0.263) (0.418) Second Stage constant 0.663** 1.198* 0.568* 1.550 0.415 1.476 0.958*** 2.855 1.064 (0.276) (0.684) (0.306) (5.963) (1.741) (7.652) (0.294) (2.049) (0.671) distance1 0.158*** 0.127 0.163*** 0.132* 0.013 0.123 0.197*** 0.083 0.065** (0.037) (0.086) (0.051) (0.068) (0.080) (0.258) (0.066) (0.173) (0.028)
43 Table 2 3. Continued Whole sample (1) Vacation (2) Sightseeing (3) VFR (4) Other motivation (5) By affiliation (6) With families and friends (7) By travel agencies (8) Alone (9) Third Stage attraction 0.879*** 1.001*** 0.846*** 0.557 1.338 1.863 0.579*** 2.022*** 0.988** (0.149) (0.289) (0.184) (0.795) (1.237) (1.271) (0.177) (0.498) (0.384) distance2 0.767*** 0.735*** 0.912*** 0.286*** 0.266 0.493 0.990*** 0.817*** 0.424*** (0.049) (0.100) (0.068) (0.218) (0.309) (0.307) (0.079) (0.185) (0.099) distance3 0.314*** 0.351*** 0.352*** 0.017 0.138* 0.151*** 0.412*** 0.304*** 0.020 (0.027) (0.055) (0.039) (0.068) (0.071) (0.057) (0.051) (0.099) (0.038) CD 0.004** 0.007** 0.004* 0.007 0.001 0.002 0.008*** 0.004 0.000 (0.002) (0.004) (0.002) (0.011) (0.016) (0.015) (0.002) (0.006) (0.005) Inclusive Values Type A 0.362* 0.389 0.413** 1.291 0.960 1.784 0.242 1.046 0.482** (0.190) (0.425) (0.199) (1.669) (1.356) (1.417) (0.177) (0.910) (0.224) Type B 0.607*** 0.665 0.637*** 1.973 1.117 1.873 0.566*** 0.887 0.344** (0.203) (0.480) (0.215) (2.935) (1.467) (1.353) (0.197) (1.167) (0.146) continue 0.385*** 0.545 0.361*** 0.911 0.367 0.413* 0.415*** 0.600 1.973** (0.107) (0.362) (0.103) (1.095) (0.496) (0.373) (0.126) (0.562) (1.003) No.obs. 3055 600 2022 231 198 205 1772 286 788 L.L 4971.15 1029.9 4 3303.37 309.58 262.0 1 316.9 9 2772.7 5 549.67 1225.92 AIC 3.267 3.486 3.280 2.802 2.808 3.239 3.148 3.949 3.152 Pseudo R 2 0.104 0.134 0.106 0.136 0.129 0.159 0.110 0.191 0.104 Notes: standard error in parenthesis, indicates significant at 0.10 level, ** indicates significant at 0.05 level, *** indi cates significant at 0.01 level. Inclusive values for the choice of go home is fixed to be 1. For motivation, 1=vacation, 2=sight seeing, 3=VFR, 4=others. For organization patterns, 1=by affiliation, 2=with families and friends, 3=by travel agencies, 4=alone.
44 Table 2 4 Simulation results of the probability change with the attraction increase Home Beijing Shangha i Hangzho u Suzhou Huangsh an Nantong Changzh ou Yangzho u Wuxi Zhenjian g Beijing 0.001% 0.016% 0.003% 0.003% 0.005% 0.002% 0.000% 0.000% 0.000% 0.001% 0.001% Shanghai 0.026% 0.006% 0.257% 0.076% 0.059% 0.043% 0.002% 0.013% 0.012% 0.007% 0.014% Hangzhou 0.030% 0.008% 0.071% 0.313% 0.116% 0.047% 0.003% 0.013% 0.012% 0.003% 0.015% Suzhou 0.048% 0.011% 0.093% 0.108% 0.369% 0.072% 0.001% 0.004% 0.023% 0.027% 0.029% Huangshan 0.022% 0.007% 0.060% 0.059% 0.111% 0.289% 0.002% 0.008% 0.008% 0.015% 0.010% Nantong 0.058% 0.005% 0.026% 0.041% 0.028% 0.031% 0.426% 0.041% 0.057% 0.068% 0.070% Changzhou 0.013% 0.021% 0.271% 0.275% 1.192% 0.132% 0.009% 0.613% 0.053% 0.021% 0.065% Yangzhou 0.099% 0.008% 0.054% 0.073% 0.084% 0.044% 0.025% 0.087% 0.677% 0.167% 0.038% Wuxi 0.067% 0.007% 0.045% 0.052% 0.006% 0.041% 0.012% 0.026% 0.063% 0.387% 0.078% Zhenjiang 0.099% 0.008% 0.057% 0.063% 0.086% 0.046% 0.025% 0.082% 0.030% 0.163% 0.650% Notes: Each row represents each scenario of the attraction increase of the corresponding destination.
45 Figure 2 1. Location map of Nanjing and subsequent destination alternatives.
46 Figure 2 2 Decision making process and specification of choice sets in the nested logit model Home Subsequent Destination Destination Type A Destination Type B Beijing Shanghai Suzhou Hangzhou Huangshan Nantong Changzhou Wuxi Yang zhou Zhenjiang Tourist 1 st Sta ge 2 nd Sta ge 3 rd Sta ge
47 CHAPTER 3 SPATIAL EFFECTS IN R EGIONAL TOURISM GROW TH Chapter Summary This chapter examines two particular spatial effects in regional tourism growth: spatial dependence and spatial heterogeneity S patial dependence can be explained by spillover effects of regional tourism growth: the indirect or unintentional effects that a industr ies exert on touris t flows to other regions On the other hand, spatial heterogeneity describes different regional tourism growth patterns because of different resource endowments, infrastructure, and market access indicated by core periphery theory in economic geography. With the spatial growth regression model this chapter identifies a set of factors that explain the rate of regional tourism growth across 342 cities in China from 2002 to 2010. Local economic growth has been found to be the most important factor stimulating both inbound and domestic tourism gro wth. Other significant factors include localization economies, tourist resource endowments, and hotel infrastructure. W ith the tourism growth model incorporating spatial dependence, the significant spatial spillover effects a re recognized, and the cross c ity competition effect i s highlighted in terms of tourism resource endowments for inbound tourism and hotel infrastructure for domestic tourism. Moreover, based on geographically weighted spatial Durbin model, s ubstantial geographic variations were identif ied in terms of local tourism growth patterns. For example, the spillover effect of inbound tourism growth was found to be larger in the Northwest and the North China area, while the spillover effect of domestic tourism growth exhibits a clear pattern of s outh north division.
48 Introduction Along with the global boom of the tourism industry, more and more regions in developing countries have realized the significance of tourism to the local economy. In acknowledgement of the positive multiplier effects of to urism that greatly contribute to the local economy and other forward and backward linked industries tourism has recently received much more support from both regional government and society. A notable example comes from China. To expedite local tourism development, 24 out of 31 Chinese provincial level regions officially declared tourism to be the backbone of their industry ( Wu, Zhu, & Xu, 2000 ) resulting in high priority being given to tourism in all levels of regional master plans. T he development of tourism has been suggested to be an important factor fo r economic growth in peripheral areas ( Z urick, 1992 ) In the context of a depressed economy, tourism growth is regarded as a reasonable choice for regional development ( Hohl & Tisdell, 1995 ) because of unique tourism resource endowment s in some marginalized areas. However, the development plan might not be successf ul in these regions ( Krakover, 2004 ) because tourism growth heavily relies on agglomeration economies rather than resource endowment s ( Capone & Boix, 2008 ) Therefore, the understanding of a regional tourism growth model will greatly help in identifying both advantages and disadvantages of touri sm development in the area, and in implement ing more reasonable and rational polices for the industry development. In tourism research, a large body of literature has been devoted to tourism demand analysis, which is used to investigate demand side factors contributing to tourist flows/revenues ( Song & Li, 2008 ) However, few studies focused on tourism development analysis by considering supply side factors like infrastructure, resou rce
49 endowment and market access. Th is supply side insight is important for regional tourism development analysis by unveiling the importance of various destination attributes. Based on the results from this analysis, more specific efforts can be made to s trengthen a destination s competitive advantages to attract more tourist s and generate more substantial tourism revenue. A part from factors incorporated in conventional economic growth models, two types of spatial effect, spatial dependence and spatial het erogeneity are evident in regional tourism growth. S patial dependence can be explained by spillover effects of regional tourism growth: regions might receive beneficial spillovers through the thriv ing tourism industry in their neighbors On the other hand spatial heterogeneity describes different regional tourism growth patterns because of different resource endowments, infrastructure, and market access indicated by core periphery theory in economic geography. The failure to incorporate spatial dependence and spatial heterogeneity might result in unreliable coefficient estimates. Therefore, a more general ized growth model incorporating both spatial dependence and spatial heterogeneity should be proposed to explore regional tourism growth patterns. Literat ure Review Regional Tourism Growth Model To understand the development trajectory of tourism a set of evolutionary models have been discussed from spatial and/or temporal perspectives ( Dredge, 1999 ; Pearce, 1995, pp. 9 17 ) One of the most popular e volutionary models in the literature is the tourist area life cycle (TALC) model by Butler ( 1980 ) which pointed out a six stage evolutionary sequence of local tourism growth including exploration, invol vement, development, consolidation, stagnation, and rejuvenation or decline. In different
50 development phases, different types of facilities are found to support tourism growth. The TALC model has been applied to study the evolution of various types of dest inations all over the world, and the variations in timing and nature of each stage have been reported and particular extensions have been proposed to improve the original model ( Baum, 1998 ; Butler, 2004 ; Cole, 2009 2012 ; Douglas, 1997 ; Garay & Cnoves, 2011 ; Getz, 1992 ) However, instead of discovering the growth mechanism in different phases, past applications of the TALC model are more enthusiastic in fitting the S shaped curve indicated by the model with their data ( da Conceio Gonalves & Roque guas, 1997 ; Karplus & Krakover, 2005 ; Lozano, mez, & Rey Maquieira, 2008 ; Lundtorp & Wanhill, 2001 ; Moore & Whitehall, 2005 ) and their results tend to be of limited value for guiding local tourism development ( Cole, 2009 2012 ) Apart from the TALC mo del, several other evolutionary models aim to explain the spatiotemporal pattern of local tourism development. In Gormsen s seaside tourism development model ( Pearce, 1995, p. 13 ) the pattern change in accommodatio n types and local participation were emphasized over time and space. Furthermore, in Miossec s model of tourist development, the spatial/social structure change in resorts, transport, tourist behavior and attitudes have been identified over different phase s ( Pearce, 1995, pp. 14 15 ) Oppermann ( 1993 ) proposed another evolutionary model to describe the tourism growth pattern in developing countries and pointed out the dominant role that the capital city plays. In a more recent research by Papatheodorou ( 2004 ) his evolutionary model underline d a dualism in market and spatial structures along with regional tourism growth, and emphasized the linkage between core and peripheral regions.
51 A ccording to the theory of economic driven tourism growth, it is suggested that local economic g rowth plays a key role in facilitating local tourism development. In the process of economic growth and expansion, more business travelers are attracted ( Oh, 2005 ) On the other hand, economic development leads to a boost of international trade. Therefore, the number of international tourist arrivals increases. Economic expansion also improves local tourism related infrastructures and service qualities because of ex ternal economies ( Capone & Boix, 2008 ) As a result, the overall competitiveness of destination s is enhance d. Many empirical studies found a unidirectional causal relationship from local economic development to tourism growth ( Eugenio Martn, Martn Morales, & Scarpa, 2004 ; Narayan, 2004 ; Oh, 2005 ) and some studies even suggested a reciprocal relat ionship between them ( Dritsak is, 2004 ; Durbarry, 2004 ) However, the economic driven tourism growth hypothesis is not always supported, due to different economic conditions of tourism destinations ( Balaguer & Cantavella Jord, 2002 ; Gunduz & Hatemi J, 2005 ; Lee & Chang, 2008 ) Compared to m any studies proposing conceptual models, very f ew analyzed the regional tourism growth pattern through rigorous economic geography modeling To the best of my knowledge, the only studies of this topic come from the research on regional tourism employment growth rate in Italy ( Capone & Boix, 2008 ; Lazzeretti & Capone, 2009 ) Their research pointed out the importance of agglomeration economies as the major source of regional tourism growth. However, these studies fail to control other factors considered in spatial growth model and did not take spatial heterogeneity in to consideration.
52 Spatial Effects of Tourism Growth As a typical spatial effect the spillover effects would be substantial in tourist flows. The spatial spillover effect refer s to the indirect or unintentional effects that a industr ies exert on touris t flows to other regions. A s a result, a region (a positive spillover effect). Yang and Wong ( 2012b ) provided a framework to interpret the spill over effects in the tourism industry from both supply and demand sides and discussed several important c hannels triggering the effects In some empirical research, the spillover effects have been identified in tourist flows ( Drakos & Kutan, 2003 ; Gooroochurn & Hanley, 2005 ; Neumayer, 2004 ; Yang & Wong, 2012b ) and regional tourism growth ( Capone & Boix, 2008 ; Lazzeretti & Capone, 2009 ) Therefore, it is necessary to capture this effect through spatial dependence specified in a spatial growth regression model ( LeSage & Fischer, 2008 ) Another important spatial effect in regional tourism growth is spatial heterogeneity, which underlines different growt h patterns of tourism in different regions. As indicated by some evolutionary models of tourism growth, various regions might adopt different resources and facilities to support tourism growth. In the TALC model, because of different timing in tourism grow th, tourism in different regions might be positioned in different stages, indicating the distinct growth m e ch an ism s. Gormsen s model also highlights regional heterogeneity, suggest ing that guest house s and private rooms are important accommodation s for tou rism in Periphery I and II but not in Periphery III and IV. Spatial heterogeneity in tourism development has also been identified in many other empirical studies. For example, infrastructure factors are found to be particularly
53 important for tourism in le ss developed destinations ( Eugenio Martn, et al., 2004 ) and transporta tion infrastructure is estimated to be more sensitive to tourism growth in African and Asian destinations ( Khadaroo & Seetanah, 2008 ) In Capon and Boix ( 2008 ) s research of regional tourism growth data in Italy they also found different growth patterns for different types of cities and districts. However, all these studies only considered the heterogeneity between certain groups and did not treat spatial heterogeneity in a rigorous way in their empirical models. Research Methods At the outset of empirical research, exploratory spatial data analysis (ESDA) tools will be applied to recognize the spatial pattern of t ourism growth rates across 342 employed to identify the cluster of cities that has experienced intense increase in tourism revenue from 2002 to 2010. The local Moran I statistic is specified as follows ( Anselin, 1995 ) : and ( 3 1 ) where i and j index the cities ( i j is the mean value of which denotes the growth rate of tourism revenue in each city. W is the spatial weighting matrix: the elements w ii on the diagonal are set to zero, and w ij indicates the way city i spatially connects to city j To specify the spatial weighting matrix W ( 2005 ) and Yang and Wong ( 2012a ) the research applies a nearest neighbor spatial weighting matrix, which guarantees the exogeneity of the matrix ( Anselin & Bera, 1998 ) For a k nearest neighbor matrix, w ij sets to be 1 if city j is the one o f the k th nearest cities of city i In this study, I choose k = 5 and compare i t with the results for k = 10 and 1 and 1.
54 A value of 1 indicates perfect positive autocorrelation 0 indicates no autocorrelation, and 1 indicates perfect negative autocorrelation. To further visualize the cluster, the Moran cluster map is utilized. It shows cities with significant local Moran statistics and indicates them by four different types of clustering corresponding to the local spatial association between a city and its neighbors ( Anselin & Bao, 1997 ) : HH (a city with a high value surrounded by cities of high values), LH (a city with a low value surrounded by cities of high values), LL (a city with a low value surrounded by cities of low values), and HL (a city with a high value surrounded by cities of low val ues). Therefore, based on the local Moran cluster map, significant clusters can be visualized and identified. Furthermore, another local spatial statistics, the local G statistic is defined to measure the spatial pattern around each city, which is given a s ( Ord & Getis, 1995 ) : ( 3 2 ) To get a standardized local G* statistic, one can subtract it from its expectation and divide it by the square root of variance ( Lloyd, 2011 ) : ( 3 3 ) where , and After the normalization, the statistic is normall y distributed. If G* is positive and statistically significant, it suggests a cluster of cities with high values, and if G* is negative and statistically significant, it indicates a cluster of cities with small values.
55 To explain the spatial pattern of reg ional tourism growth rates, t his study intends to extend the general spatial growth regression model to incorporate spatial heterogeneity. The general spatial growth regression model proposed by LeSage and Fischer ( 2008 ) suggests that the long term regional economic level is a function of the characteristics of the home reg ion and neighboring ones, spatial connectivity with other regions, and the strength of spatial dependence. To capture the spillover effects in regional growth and omitted variables with spatial autocorrelation, the model follows a spatial Durbin model (SDM) in a form as: ( 3 4 ) where y is n 1 vector of observed regional growth rates, and W is the spatial weighting matrix. X contains a set of explanatory variables, and is the error term which follows an i.i.d N (0, 2 ) Several studies advocated the use of the SDM in regional growth literature from a technical point of view ( Elhorst, 2010 ; LeSag e & Pace, 2009 ) First, the SDM nests most other spatial regression model and is the more generalized specification. Therefore, it provides the unbiased estimates even though the data generation process is misspecified. Second, it alleviates the omit va riable problem by introducing a set of spatially lagged independent variables. Finally, the SDM enable researcher to obtain and interpret direct and indirect effects embedded in the model specification. In the spatial growth regression model, LeSage and Pace (2009 ) proposed a method to obtain the direct, indirect, and total impacts of different variables in the long run steady state equilibrium. Because of spat ial lags of dependent and independent variables in the model, these impacts may include the feedback effects through spatial
56 spillover. To interpret the direct and indirect effects, the SDM can be re written in a reduced form as follows: ( 3 5 ) Thus, the partial derivative of the model with respect to a particular independent variable can be obtained as: ( 3 6 ) where is the k th independent variable in X with a coefficient of This derivative is an N N matrix instead of a scalar in an ordinary regression model. LeSage and Pace (2009 ) calculated the average direct effect as the sum of the trace elements of the N N matrix in Equation 3 6 divided by the total number of observations. This direct effect reflects the impact of change in of observation i on the dependent variable y of observation i On the other hand, the average tota l effect is calculated by dividing the row sums of that N N matrix by the total number of observations, representing the cumulated impact of change in of observation i on the dependent variable y of all observations. Finally, the difference between the total and direct effect is labeled as the indirect effect, which incorporates the feedback effects across different regions. The standard error of these effect estimates can be obtained from a Markov Chain Monte Carlo method with 2,000 simulated draws ( LeSage & Pace, 2009 ) The spatial Durbin model can be estimated by maximum likelihood estimation (MLE), and its log likelihood function is specified as ( Anselin, 1988 ) :
57 ( 3 7 ) For computational convenience, the log likelihood can be concentrated with respect to the coefficients , and and the noise variance parameter 2 ( LeSage & Pace, 2009 ) Based on the estimates from MLE, one can test the significance of spatial coefficients and through Wald tests ( Greene, 2007 ) How to specify the spatial weighting matrix is a key issue in modeling spatial growth as different matrices capture different channels of spillover ( Corrado & Fingleton, forthcoming ; LeSage & Fischer, 2008 ) Basically, there are three ways to accomplish this The first is nearest neighbor matrix discussed in Equation 3 1. The second one, contiguity based weight matrix, is specified as follows: w ij = 1, if region s i and j share a common boundary; and w ij = 0, otherwise. For the third type, distance decay weight matrix it is assumed that the connection between two places declines as the distance between them increases. Thus, ( 3 8 ) where d ij is the geographical distance between the cent roid s of two region s, and refers to as the distance decay parameter To incorporate the spatial heterogeneity of regional tourism growth and obtain localized estimates for each city in the sample, the SDM model can be extended under a framework similar to geographically weighted regression ( Pez, Uchida, & Miyamoto, 2002 ) SDM). To g eneralize the ordinary SDM, refers to the variance covariance matrix of
58 error term in Equation 3 3, which might be not i.i.d. GW SDM tends to capture spatial heterogeneity by assuming dependent on location of each observation as follows: with elemen ts as and ( 3 9 ) where o is the variance covariance matrix used to obtained the localized estimates at the location 2 is the unknown variance parameter, d oi is the distance from observation i to the focal observation o and o is called the kernel bandwidth. In this study, the Gaussian weight is used as the geographic weights in the GW SDM. refers to the probability density function of standardized normal distribution, and refers to the function of standardized deviation. Ho w to get a reasonable kernel bandwidth o is important to obtain reliable local estimates from the GW SDM. A cross validation method is used to pick up the suitable o by minimizing the cross validation (CV) score ( Fotheringham, Brunsdon, & Charlton, 2002 ) which is specified as: ( 3 10 ) where is the fitte d value of with the observations for point i omitted from the estimation. In this study, the minimization algorithm is based on golden section search and parabolic interpolation ( Brent, 1973 ; Forsythe, Malcolm, & Moler, 1977 ) The kernel bandwidth o that minimizes the CV score is retained for further estimation. Finally, one can get the updated log likelihood function for any focal observation o as: ( 3 11 )
59 By using different log likelihood functions for observations in different locations, one can get a set of estimat es for different location s During the estimation, this model As a res ult, the estimated coefficient of variables will vary over space, suggesting s patial heterogeneity in re gional tourism growth patterns. Matlab was then used to estimate th e general spatial growth model with spatial heterogeneity. Data Description Followin g th e general specification of the spatial growth regression model in Equation 3 4 the dependent variable in this research is the log of the regional tourism growth rate over the years 2002 to 2010 There are two types of regional tourism growth rate one can use for the sample of China : (1) the growth rate of the total revenue from inbound tourism (lnrate_inb) and (2) the growth rate of the total revenue from domestic tourism (lnrate_dom) By comparing the results from the models with different growth measures, one can reveal the major distinct factors of tourism growth. Several independent variables are used to explain the source of long term tourism growth lnrev_inb measures the log of the total revenue of inbound tourism (in 10,000 USD) in 2002, while lnrev_dom is the log of the total revenue of domestic tourism (in 100 million RMB Yuan) in 2002. Both of these independent variables reflect the initial level of local tourism development during the first year of the research period, and their coefficients tend to capture the catch up effect of tourism growth. T heir coefficients are expected to be negative, as cities with a less developed tourism industry have experienced more intense touri sm growth to catch up with the leading cities. Moreover, lnpopden is the log of the population density of city (person / km2) in 2002, which is a common variable to measure urbanization economies ( Frenken, Van
60 Oort, & Verburg, 2007 ) ; lntertiary represents the log of the percentage of GDP from tertiary industr ies relative to the overall GDP, which captures the influence of localization economies Since both urbanization and localization economies could be beneficial for local tourism growth ( Capone & Boix, 2008 ; Lazzeretti & Capone, 2009 ) the estimated coefficients lnpopden a nd lntertiary are expected to be positive. To incorporate the influence of tourism resource endowment, a weighted index of the touris m resource s is used. This index is d efined as the sum of the number of world heritage sites (weighted by four) the number of national parks (weighted by two), and the number of AAAA scenic spots in a city in 2002. lnresource is the log of this weighted index. If the original index equals ze ro, 0.5 is assigned to avoid a log of zero. As argued by Ellerbrock and Hite (1980 ) and Yang, Lin, and Han (2010 ) tourism resources serve as the major engine of local tourism growth. Theref ore, the estimated coefficient of lnresource is expected to be positive. Furthermore, lnhotel is the log of the number of star rated hotels per capita in a city in 2002, which is used to capture the influence of the local tourism infrastructure Also, lnGD P_rate is the log of the growth rate of GDP from 2002 to 2010 The coefficients of these two variables are also expected to be positive As suggested by LeSage and Fischer (2008 ) and Cuaresma, Doppelhofer, and Feldkircher (2012 ) most independent variables specified in the spatial growth regression model use the values at the initial year of the research period to explain the variation in tourism growth rates. Therefore, the problem that arises from the simul taneity between dependent and some independent variables can be partly avoided. More importantly, the possible endogeneity problem can be alleviated by using
61 these predetermined regressors. Furthermore, since all dependent and independent variables have be en transformed into a logarithm, the estimated coefficients of different independent variables can be interpreted as elasticities. As a result, these coefficients can be directly used to gauge the magnitudes of various factors determining local tourism gro wth. T he data set covers 34 2 cities in mainland China. Most Chinese city level data can be obtained from the China Statistical Yearbook for Regional Economy (2003 2011) published by the National Statistical Bureau of China. China City Statistical Yearbook and CNKI Statistical Yearbook Database are referred to in order to fill out some of the missing values. The number of national parks for each city is collected from the website of the Ministry of Housing and Urban Rural Development of China ( http://www.mo hurd.gov.cn ). The data of number of AAAA scenic spots for each city is collected from the website of CNTA (http://www.cnta.gov.cn), and that of number of World Heritage Sites is ob tain ed from the website of the United Nations Educational, Scientific, and C ultural Organization ( http://whc.unesco.org ) This study focuses on all cities in mainland China. However, several issues about the sample cities should be highlighted. First, there are some special counties that are governed directly by the province. In this situation, the study combines these neighboring counties together and treats them as one city. For example, neighboring counties that are directly under the jurisdiction of the provinc ial government in Hainan and Hubei ar e each regarded as single cities Second, these special counties may also be isolated from other special counties and surrounded by cities, such as Jiyuan in Henan and Shennongjia in Hubei. The study considers them as individual cities Thirdly,
62 as there i s no prefecture level cit y governed by m unicipalities such as Beijing, Shanghai, Tianjin and Chongqin g these four are treat ed as general cities. Table 3 1 presents the descriptive statistics of variables considered in the spatial growth regression model. Regarding the two dependent variables, the mean value of lnrate_inb is 1.472, which is smaller than that of lnrate_dom, 1.854. This suggests that the average growth rate of domestic tourism is larger than that of inbound tourism over the 342 cities in Chi na from 2002 to 2010. The mean value of lnGDP_rate is merely 1.305, and it is smaller than those of lnrate_inb and lnrate_dom, which demonstrates that both inbound and domestic tourism have grown at a higher speed than the local GDP over the research perio d. Furthermore, the standard deviation of lnrate_inb, 1.566, is nearly twice as large as that of lnrate_dom, 0.814, which highlights a more substantial variety in inbound tourism growth over cities. Moreover, the mean value of lnresource is 0.309, and this small value can be explained by the fact that 147 out of 342 cities did not have any qualified tourism attractions in 2002. Figures 3 1 and 3 2 further present the histograms of lnrate_inb and lnrate_dom, respectively. It is shown that the distributions o f these two variables are uni modally distributed with two thin tails. Table 3 2 presents the Pearson correlation coefficients between independent variables incorporated in the spatial growth regression model. The only coefficient above 0.6 is the one betw een lnrev_dom and lnpopden, which is 0.603. This indicates that, in 2002, the cities with a higher population density were more likely to attract a large number of domestic tourists. More importantly, 15 out of 20 correlation coefficients are below 0.4, su ggesting the absence of the multi collinearity problem in the specified models ( Leeflang, Wittink, Wedel, & Naert, 2000 )
63 Results of ESDA A set of ESDA tools are employed to investigate the spatial pattern of tourism growth rates (lnrate_inb and lnrate_dom). First, the Globa calculated for these two variables. By using a five nearest neighbor spatial weighting lnrate_dom is 0.223. Both of them are statistically significa nt at the 0.01 level after 99999 permutations. The results suggest that the growth of inbound tourism is distributed in a slightly more clustered way than the growth of domestic tourism. to look into the localized pattern of tourism growth rates over 342 cities in China from 2002 to 2010. Figure 3 3 presents the local Moran cluster map of the growth rate of inbound tourism revenue, and Figure 3 4 illustrates the local G* cluster map of the growth rate of inbound tourism revenue. Both maps provide very similar information regarding the clustering of inbound tourism growth. However, the local G* cluster map fails to recognize the negative localized spatial association, such as LH and HL patte rns. A large area covering several cities in Gansu, Qinghai, and Ningxia has been identified as a cluster of slow inbound tourism growth. Five small areas are found as clusters experiencing intense inbound tourism growth. They are a cluster with several ci ties in Liaoning, a cluster including several bordered cities in Xinjiang, a cluster including three cities in Hubei, a cluster containing some cities in Hunan, and a cluster incorporating some cities in Shandong and Hebei. Compared to the results from Yang and Wong (2012a ) which used the growth rate of inbound tourist arrivals from 2002 to 2006, some clusters are also highlighted in this research, such as the one in Liaoning and the one in Hubei.
64 Figure 3 5 shows the local Moran cluster map of the growth rate of domestic tourism revenue, and Figure 3 6 demonstrates the local G* cluster map of the growth rate of domestic tourism revenue. It is found that the two maps also demonstrate a very similar pattern of the clustering of domestic tourism growth. Two areas are characterized by slow growth of domestic tourism, and they include one in Xinjiang and the other in Guangdong, Hainan, and Fujian. However, different factors can be used to explain the slo w growth in these two areas. For the one in Xinjiang, the remoteness to major domestic tourist market in the Center and East becomes a substantial disadvantage for the local domestic tourism development. With regard to the one in Guangdong, Hainan, and Fuj ian, the relatively saturated domestic tourism demand there impedes the quick growth. On the other hand, three areas are shown with a high growth rate of domestic tourism revenue. The most significant one is located in Ningxia and the southern part of Gans u, while the other two are found in the Southwest (some cities in Guangxi and Guizhou) and the Northeast (several cities in Heilongjiang). These results are partly consistent with those from Yang and Wong (2012a ) which also demonstrated a LL cluster in Guangdong and Guangxi and a HH cluster in the southwest. As the choice of spatial weighting matrices is arbitrary in ESDA, varying spatial weighting matrices should be compared to check the robustness of results. The results are considered to be robust if a city with the significant LISA statistic remains in the same LISA category no matter which spatial weighting matrix is selected ( Le Gallo & Ertur, 2003 ) When comparing the results with the five nearest neighbor matrix with those with other matrices, such a s the ten nearest neighbor matrix, the fifteen nearest
65 neighbor matrix, the contiguity based matrix, and the distance decay matrix, most cities remain within the same LISA category, and cities in each LISA category have little variation. More importantly, no city is seen shifting from one significant LISA category to another. Therefore, the robustness of LISA results is confirmed Results o f Spatial Growth Regression Model OLS Estimates OLS is first used to fit the tourism growth model without taking any s patial effects into consideration. The model includes all specified independent variables. Table 3 3 presents these OLS estimates of various tourism growth models. Model 3.1 (in column 1 of Table 3 3) fits the growth model of inbound tourism revenue and in cludes all 342 cities in China. lnrev_inb is estimated to be negative and statistically significant, highlighting a substantial catch up effect in inbound tourism growth across Chinese cities. lnpopden is positive and statistically significant, suggesting that the advantage stemming from urbanization economies plays an important role in stimulating inbound tourism growth. However, lntertiary, although positive, is not statistically significant, which provides little support for the impact of localization ec onomies on inbound tourism development. For other independent variables, lnresource, lnhotel, and lnGDP_rate are estimated to be statistically significant and positive, indicating that tourism resource endowment, tourism infrastructure, and local economic growth are important factors explaining the growth rate of inbound tourism revenue. S ince all variables are in a logarithm form, one can compare the estimated magnitude s of different variable s and their relative contributions to regional tourism growth ca n be outlined. By comparing the magnitudes of different coefficients, the coefficient of lnGDP_rate, which is 0.685, is found to be the largest. This suggests the substantial importance of economic growth.
66 Moreover, the coefficient of lnhotel is estimated to be larger than that of lnresource, Models 3.2 to 3.4 (in columns 2 4 of Table 3 3) refer to the inbound tourism models from different regions in China: th e East, the Center, and the West, respectively. By comparing the estimates from these three models, different patterns of inbound tourism growth can be recognized in different regions of China. The estimated coefficient of lnrev_inb is found to be the larg est in the Center, showing that the catch up effect of inbound tourism growth is more substantial among central cities. Moreover, lnresource is estimated to be significant in eastern and central cities, while lnhotel is found to be significant only in west ern cities. This suggests that the tourism resource endowment helps to explain inbound tourism growth in eastern and central cities, and tourism infrastructure dominantly stimulates the growth in western cities. Furthermore, lnGDP_rate is found to be the l argest in eastern cities, which suggest that economic growth is a key factor influencing the growth of inbound tourism revenue there. Model 3.5 (in column 5 of Table 3 3) fits the data of domestic tourism growth over 342 cities in China. All but one variable are estimated to be statistically significant with expected signs. Unlike the estimate in the inbound tourism model (Model 3.1), lnhotel is not significant, and this illustrates that tourism infrastructure is not a major factor contributing to the growth of domestic tourism revenue in China. This is at tributed to the fact that inbound tourists have higher expectation s with regard to infrastructure provision and prefer to use high level infrastructure to maintain the same comfort as they do in home countries ( Khadaroo & Seetanah, 2007 2008 ) lnrev_dom is estimated
67 to be 0.554, and its magnitude is much larger than that of lnrev_inb in Model 3.1 for inbound tourism growth. This result shows that the catch up effect is more pronounced in domestic tourism growth in Chin ese cities Moreover, the estimated coefficient of lntertiary is much larger than that of lnpopden in the model. It suggests that the benefit from localization economies plays a more important role in boosting domestic tourism than that from urbanization economies. In addition, the estimated coefficient of lnresource is larger than its counterpart in Model 3.1, suggesting that the tourism resource endowment is more helpful in explaining domestic tourism growth. This can be explained by the fact that there are more Chinese domest ic tourists interested in sightseeing, and the tourism resource s pecified contains mainly sightseeing attractions. Regarding the regional difference in domestic tourism growth, Models 3.6 to 3.8 are estimated using each respective dataset from the East, t he Center, and the West. The statistically insignificant coefficient of lnrev_dom in Model 3.6 suggests that the catch up effect is not evident among eastern cities. Moreover, lnpopden and lntertiary are estimated to be significant only in Model 3.8, and t his implies that the advantages stemming from urbanization and localization economies are important factors determining domestic tourism growth in only western cities in China. As indicated by the significant coefficients of lnresource in Models 3.7 and 3. 8, the tourism resource endowment is also found to be a major driver of domestic tourism in the Center and the West. Finally, lnGDP_rate is estimated to be the largest in eastern cities. It highlights the paramount role that economic growth plays in stimul ating domestic tourism growth, which is consistent with the finding from the inbound tourism model (Model 3.2).
68 SDM Estimates A further examination of the residuals from Models 3.1 and 3.5 highlights the significant spatial autocorrelation in the predicte residuals is estimated to be 0.290 for Model 3.1 and 0.236 for Model 3.5, both of which are statistically significant at the 0.01 level. The spatial dependence will render possible biased estimates, and therefore, t he spatial effect is necessary to take into consideration in the tourism growth model. Table 3 4 presents the estimation results of spatial growth regression models for both inbound and domestic tourism. Apart from the SDM specified in Equation 3 4, severa l other specifications of spatial regression models are also considered to assess the robustness of model estimates. These alternative models include the spatial autoregressive model (SAR), the spatial error model (SEM), and the heteroskedasticity and auto correlation consistent (HAC) spatial model. The SAR model is a restricted model of the SDM by excluding all spatially lagged independent variables and only includes a spatial lag of dependent variable with a he SEM model specifies a spatially HAC model incorporates spatially autoregressive terms of both the dependent variable and the error term. The first four columns of Table 3 4 give the estimation results of various spatial models of inbound tourism growth. In particular, Model 3.11 refers to the spatial growth regression model with a SDM specification in Equation 3 4. As suggested by the smallest AIC value, Model 3 .11 outperforms the other three models, including the SAR model (Model 3.9), the SEM model (Model 3.10), and the HAC model (Model 3.12). Moreover, the AIC values of these four models are found to be smaller than that of the
69 OLS model (Model 3.1), which is 1218.0 ( Table 3 3). This shows the fact that the inclusion of spatial dependence improves the goodness of fit of the model. This argument is also supported by the significant Wald test statistics on various spatial effects embedded in these four spatial mo dels. With regard to the estimates of independent variables that are not spatially lagged, four models present similar information regarding their significances and signs. However, the estimated magnitudes of some coefficients vary greatly across different models, which highlight the importance of correct model specification. Model 3.11 is of particular interest as it captures additional spatially lagged significant at t he 0.01 level. It highlights the significant spillover effects on inbound tourism growth over cities and suggests that a 1% increase in the growth rate of inbound tourism revenue in a given city will contribute to a 0.532% increase in the growth rate of th e nearby cities. Moreover, lnrev_inb has a significant coefficient of 0.427, which unveils the catch up effect of inbound tourism growth. Other explanatory variables including lntertiary, lnresource, lnhotel, and lnGDP_rate are significant with an expecte d sign. However, lnpopden is found to be insignificant. Although the initial purpose for including the spatially lagged explanatory variables is capturing the unobserved effects ( Elhorst, 2010 ) their estimates also provide some insights on the spillover effects of these variables on regional tourism growth. As shown in Model 3.11, four out of the six spatially lagged explanatory variables are estimated to be statistically significant. Even thou gh lnrev_inb has a negative sign, its spatial lag is estimated to be positive, which suggests that a high
70 value of the initial inbound tourism revenue of nearby regions promotes the local inbound tourism growth. The estimation results also show that the gr owth rate of regional inbound tourism is positively associated with the urbanization economies (lnpopden) and the tourism infrastructure (lnhotel) of the neighboring regions, while it is also negatively associated with resource endowments (lnresource). The positive association depicts an agglomeration effect, while the negative association implies a competition effect. Therefore, the highly significant and negative coefficient of the spatial lag of lnresource indicates that more tourism resource endowments in the neighboring regions will hinder the local inbound tourism growth because of this competition effect across nearby cities. Since the SDM model incorporates the spatial lags of both dependent and independent variables, its estimates are not directly c omparable to their counterparts in other models, such as the OLS model, the SAR model, and the SEM model. The decomposition of spatial effects of independent variables is used to look into the direct, indirect, and total effects of independent variables. T hese results are presented in Table 3 5. It should be noted that the indirect effect captures the cumulative spillover effects over the whole sample of Chinese cities in this study. As shown in Table 3 5, urbanization economies have a positive and indirect effect on the local inbound tourism growth, while localization economies have a positive and direct effect. This is consistent with the estimates from Model 3.12 in Table 3 4. Moreover, tourism resource endowments have a positive direct effect and a negat ive indirect effect, which indicates once again the resource oriented competition effect in regional inbound tourism growth. The total effects summarize the overall impact of various factors in a spatial framework
71 and represent the long run effects in a ne w steady state situation ( Del Bo & Florio, 2012 ) The total effects of tourism infrastructure (lnhotel) and e conomic growth (lnGDP_rate) are found to be significant and substantial. For example, the total effect of tourism infrastructure is 1.063, which can be interpreted as elasticity: a 1% increase in the hotel infrastructure will correspond to a significance i ncrease in regional inbound tourism revenue of 1.063%, accounting for both direct, own region effects, and the positive spillover effects to other regions. It is interesting to point out the insignificant total effect of tourism resource endowments (lnreso urce). This is because the positive direct effect tends to decrease due to the competitive force which pulls tourism revenue and tourist arrivals away. Concerning the estimates of the spatial growth regression model of domestic tourism revenue (Model 3.16) and positive, and it is smaller than its counterpart in Model 3.12. This results suggests that the spillover effect in inbound tourism growth is more substantial than that in domestic tourism growth, which corroborates the findings from Yang and Wong ( 2012b ) Furthermore, domestic tourism growth is found to be positively associated with the local localization economies (lntertiary), resource endowments (lnresource), and economic growth (lnGDP_rate), while it is negatively associated with the initial level of domestic tourism revenue (lnrev_dom). With regard to the spatially lagged independent variables, only two of them are found to be statistically significant, and they are the spatial lag of lnrev_dom and lnhotel. The initial level of domestic tourism revenue in neighboring regions is positively associated with domestic tourism growth, which is in line with the previous findings. Moreover, the tourism infrastructure of neighboring regio ns adversely
72 impacts the local domestic tourism growth, which can be recognized as the competition effects across regions. Looking at the decomposition of the impacts of the variables in Table 3 5, it is shown that the direct effect of lntertiary is the la rgest, which highlights the substantial impact of localization economies on domestic tourism growth across Chinese cities. Furthermore, the total effect of lnGDP_rate is lower than its counterpart in the inbound tourism growth model, and this result shows the fact that local economic growth plays a more important role in stimulating inbound tourism than domestic tourism. In summary, all of these results support the conjecture that t he determinants of international and domestic tourism demand/flows would be different ( Bigano, Hamilton, Lau, Tol, & Zhou, 2007 ; Salman, Shukur, & von Bergmann Winberg, 2007 ; Yang & Wong, 2012b ) Different spatial weighting matrices t end to yield different estimates from the spatial growth regression model. To check the robustness of the aforementioned results, the spatial growth regression model will be fit with different spatial weighting matrices. Table 3 6 presents these results with three other weighting matrices: the distance decay matrix W dist (with distance decay parameter = 2), the ten nearest neighbor matrix W 10nn and the fifteen nearest neighbor matrix W 15nn It should be noted that the contiguit y based matrix is not appropriate to model city level data in China. This is because some island cities make the weighting matrix singular and render computational difficulty in MLE. According to the results shown in Table 3 6, there are only minor changes in the estimates with different spatial weighting matrices. They identify a similar set of significant variables with the same sign.
73 GW SDM Estimates S patial heterogeneity is incorporated in the GW SDM model The simple SDM model assumes a same tourism gr owth mechanism over different cities in China, and the model is likely to mask geographical variations of tourism growth mechanisms. For example, when the coefficient of a factor from the global model is statistically insignificant, the factor may facilita te tourism growth in some regions while reducing it in others. To unveil t he possible spatial variation of each regression coefficient in the SDM model ( Leung, Mei, & Zhang, 2000 ; Pez, et al., 2002 ) the GW SDM model will be applied to explore the spatially varying relationships in the data set. A CV procedure is applied to obtain the suitable bandwidth of geographic weights. Note that bandwidth value s less than 0.25 trigger certain singular matrix problems once geographic weighting is applied to the model Therefore, bandwidths of 0.25 to 2 .00 were applied to get the most appropriate bandwidth of o in Model 3.7 of both inbound and domestic models. Figure 3 7 shows the CV scores from 0.25 to 2. 00. The left one illustrates CV results of the inbound tourism model, and the right one demonstrate s those of the domestic tourism model. With a minimization algorithm of golden section search and parabolic interpolation, a bandwidth of 0.99972 is picked up for the inbound tourism growth model, while a bandwidth of 0.92997 is selected for the domestic t ourism growth model. Table 3 7 summarizes the results of the GW SDM model of inbound tourism growth by reporting the minimum, 25% percentile, median, 75% percentile, and maximum coefficient values for each city. The median value can be compared with the es timate from the SDM model (Model 3.12), and the estimates of most independent variables are fairly close. However, it should be noted that the GW SDM estimates of
74 some variables cover a broad range. For example, the finding that the estimated coefficient o f lnpopden ranges from 0.409 to 0.539 points out a substantial spatial variation of the impact of urbanization economies on inbound tourism growth. More importantly, both positive and negative estimates are found for lnpopden. This suggests that urbanizat ion economies might bring in extra benefits for inbound tourism in some areas while exerting negative impacts on others. Of particular interest, the spillover spillover effec ts across cities. In summary, the sample exhibits substantial spatial heterogeneity, and the SDM model is likely to mask this heterogeneity and yield misleading results. As an output from the GW SDM model, one can map the spatial distribution of the estima ted coefficients of the independent variables. Such visualization tools provide a powerful means to better understanding the impact of various factors on local tourism gro wth. Figures 3 8 and 3 9 illustrate the spatial distribution of various coefficients obtained from the GW SDM model, and the spatial heterogeneity of inbound tourism growth can be observed through these maps. In Figure 3 8, there is a clear geographical pattern of the distribution of cross city spillover effects in inbound tourism growth. The spillover coefficient is found to be large in two major areas: the Northwest and the North China (Huabei) area, while it is small in the Northeast. As indicated in Figure 3 9 (a) the distribution of the coefficient lnrev_inb shows a clear variation as well: the catch up effect of inbound tourism growth is most substantial in the Southwest, which covers a large number of cities in Guangxi, Yunnan, and Guizhou. The variables of lnpopden and lntertiary are estimated to be insignificant or
75 weakly significa nt in Model 3.12. A further look at their GW SDM estimates in Figures 3 9(b) and 3 9(c) shows that they are highly significant in certain areas. Positive and significant estimates of lnpopden predominate the northeastern area, which indicates the positive effect of urbanization economies on spurring local inbound tourism. However, negative and significant estimates occur in the northwestern area. One populated area as a p ossible attraction to inbound tourists. Therefore, lower population density is expected to attract more inbound tourist arrivals. For the estimates of lntertiary, the benefits of localization economies are found to be substantial in a large range of wester n cities in facilitating inbound tourism growth. Figures 3 9(d) and 3 9(e) illustrate the spatial distribution of GW SDM estimates from lnresource and lnhotel. The two maps confirm the contribution of tourism resource endowments and hotel infrastructure to cities. The affluent tourism resource endowments are found to be a huge advantage of inbound tourism development in the Southwest (e.g Yunnan, Guizhou, Guangxi, and Sichuan), while hotel infrastructure plays a more substantial role in facilitating inbound tourism growth in the Northwest (e.g Gansu, Ningxia, and Qinghai). Another noticeable finding from Figure 3 9(e) is that hotel infrastructure is not a significant determinant of local inbound tourism in the Sou theast, which is usually referred to as the most developed area in China. Finally, Figure 3 9(f) demonstrates that the influence of local economic growth on local inbound tourism is at its greatest positive extent in the Northeast, but in most cities it is insignificant.
76 Table 3 8 summarizes the results of the GW SDM model of domestic tourism growth and reports the minimum, 25% percentile, median, 75% percentile, and maximum coefficient values for each city. The median values are very close to the estimate from the corresponding SDM model (Model 3.16). Similar to the re sults from the GW SDM estimates of the inbound tourism growth model, the coefficients of some variables cover a broad range, such as lnGDP_rate ( 0.511 to 0.520), W*lntertiary ( 0.567 to 1.789 ), and W*lnresource ( 0.756 to 0.749). Figure 3 10 demonstrates the spatial distribution of spillover coefficient and suggests that the direct spillover effect of domestic tourism growth is stronger in the South, but the positive spillover declines when mo ving north, with negative effects appearing in the Northwest and Northeast. It exhibits a different pattern from the spillover effect of inbound tourism growth ( Figure 3 8). Figure 3 11(a) depicts the spatial distribution of the estimated catch up effect, and the finding that this effect is most sizable in the western region is partly consistent with pattern shown for the inbound tourism growth in Figure 3 9 (a) Even though lnpopden is not statistically significant in the SDM model (Model 3.16), its GW SDM coefficients cover both negative and positive significant values as shown in Figure 3 11(b) Figure 3 11(b) demonstrates a very clear trend from positive values in the Southwest to negative values in the Northeast. The GW SDM estimates of lntertiary are co nsistently positive in Figure 3 11(c) and localization economies play a very trivial role in domestic tourism growth in the East, especially the coastal area. It is also noteworthy that the influence of tourism resource endowments declines from the West t o the East ( Figure 3 11(d) ), and the impact of hotel infrastructure is only significant in Xinjiang and northeastern cities ( Figure 3 11(e) ).
77 Finally, local economic growth is found to be important for local domestic tourism growth in the Northeast and around Guangxi (including Hainan) (Figure 3 11(f)) Chapter Conclusion With the spatial growth regression model, this chapter has identified a set of factors that explain the rate of regional tourism growth across 342 cities in China from 2002 to 2010. Loc al economic growth has been found to be the most important factor stimulating both inbound and domestic tourism growth. Other significant factors include localization economies, tourist resource endowments, and hotel infrastructure. Moreover, with the tour ism growth model incorporating spatial dependence, the significant spatial spillover effects and catch up effects were recognized, and the cross city competition effect was highlighted in terms of tourism resource endowments for inbound tourism and hotel i nfrastructure for domestic tourism. The differences between inbound and domestic tourism growth patterns were identified in the research as well. The spillover effect was found to be more substantial for inbound tourism growth, while the catch up effect wa s more sizable for domestic tourism growth. Moreover, it has been found that the advantage of localization economies was significant to stimulate domestic tourism growth, and the benefits stemming from urbanization economies were positively associated with inbound tourism growth through an indirect effect. The geographically weighted framework adopted in this research provides a reasonable way to obtain the smooth map of parameters that help to recognize localized patterns of tourism growth, which are like ly to be masked in the global regression. Substantial geographic variations have been identified in terms of local tourism growth patterns. For example, the spillover effect of inbound tourism growth
78 was found to be larger in the Northwest and the North Ch ina area, while the spillover effect of domestic tourism growth exhibits a clear pattern of south north division. More importantly, some insignificant variables in the global regression may have both positive and negative significant coefficients in the GW SDM, such as lnpopden in the domestic tourism model. The results of this research provide several important implications for tourism policymakers. First, tourist destinations should take full advantage of positive spatial spillover effects in tourism grow th and internalize these benefits by collaborative marketing efforts with nearby cities. Second, since the catch up effect have been recognized for less developed tourist destinations, additional efforts can be made to support their tourism growth of these destinations at the beginning stage. Third, the cross city completion effects triggered by tourism resources and hotel infrastructure can be detrimental to local tourism growth, and specific strategies should be considered to alleviate this competition, s uch as the differentiation of products and destination image. Finally, the revealed spatial heterogeneity enables local policymakers to understand the localized pattern of inbound and domestic tourism growth and propose appropriate strategic plan s for loca l tourism growth. In this part of the dissertation, even though spatial heterogeneity has been incorporated to explain regional tourism growth, no statistical tests have been conducted to test the significance of this spatial heterogeneity. Moreover, altho ugh some coefficients of the spatial growth regression model might be spatially varying over different cities, some may not. The GW SDM is unable to restrict these coefficients to
79 be spatially stable. Finally, I did not consider the hierarchical structure of the data and did not incorporate the effect of the individual province in which the city belongs.
80 Table 3 1 Descriptions of variables used in the tourism growth model Variable obs. mean Std. Dev. Min Max lnrate_inb 342 1.472 1.566 6.359 6.765 lnrate_dom 342 1.854 0.814 0.315 6.063 lnrev_inb 342 5.750 2.532 1.347 12.648 lnrev_dom 342 2.069 1.631 4.305 6.902 lnpopden 342 5.238 1.485 1.349 7.904 lntertiary 342 1.022 0.214 2.220 0.395 lnresource 342 0.309 1.061 0.693 3.892 lnhotel 342 2.850 1.049 5.691 0.766 lnGDP_rate 342 1.305 0.275 0.495 2.758 Table 3 2. Pearson correlation coefficients between independent variables Variable ln rev_inb ln rev_dom ln popden ln tertiary ln resource ln hotel lnpopden 0.264 0.603 lntertiary 0.357 0.220 0.008 lnresource 0.590 0.546 0.158 0.294 lnhotel 0.519 0.308 0.150 0.297 0.454 lnGDP_rate 0.077 0.049 0.120 0.014 0.036 0.089
81 Table 3 3. OLS estimates of tourism growth model Variable Model 3.1 Model 3.2 Model 3.3 Model 3.4 Model 3.5 Model 3.6 Model 3.7 Model 3.8 Inbound Inbound Inbound Inbound Domestic Domestic Domestic Domestic lnrev_inb 0.345*** 0.307*** 0.585*** 0.329*** (0.055) (0.063) (0.084) (0.090) lnrev_dom 0.435*** 0.0602 0.424*** 0.581*** (0.055) (0.067) (0.088) (0.061) lnpopden 0.269*** 0.000590 0.214 0.0779 0.145*** 0.124 0.100 0.251*** (0.068) (0.170) (0.208) (0.085) (0.041) (0.100) (0.086) (0.050) lntertiary 0.236 0.516 0.779 0.688 0.378** 0.291 0.328 0.594** (0.419) (0.556) (0.584) (0.599) (0.186) (0.334) (0.325) (0.262) lnresource 0.154* 0.166* 0.352** 0.151 0.164*** 0.0467 0.179** 0.283*** (0.087) (0.090) (0.136) (0.167) (0.045) (0.062) (0.075) (0.059) lnhotel 0.390*** 0.0115 0.253 0.549*** 0.0401 0.0582 0.0249 0.0456 (0.122) (0.120) (0.213) (0.190) (0.036) (0.055) (0.101) (0.058) lnGDP_rate 0.685* 1.188*** 0.858* 1.001 0.375*** 0.658*** 0.537* 0.241 (0.377) (0.335) (0.512) (0.619) (0.139) (0.190) (0.272) (0.202) constant 2.454*** 1.845 4.104*** 3.204** 1.724*** 1.284 2.869*** 1.834*** (0.833) (1.420) (1.206) (1.251) (0.303) (0.780) (0.705) (0.375) N 342 102 108 132 342 102 108 132 Sample All East Center West All East Center West R square 0.190 0.466 0.459 0.189 0.432 0.253 0.509 0.565 AIC 1218.0 247.3 346.5 517.8 649.6 145.2 184.5 267.3 BIC 1244.8 265.6 365.3 537.9 676.4 163.6 203.3 287.5 Notes: *** indicates significance at the 0.01 level, ** indicates significance at the 0.05 level, and indicates significanc e at the 0.10 level. Robust standard error presented in parenthesis.
82 Table 3 4. Estimates of spatial growth regression model Variable Model 3.9 Model 3.10 Model 3.11 Model 3.12 Model 3.13 Model 3.14 Model 3.15 Model 3.16 Inbound Inbound Inbound Inbound Domestic Domestic Domestic Domestic Model SAR SEM HAC SDM SAR SEM HAC SDM Spatial weight W 5nn W 5nn W 5nn W 5nn W 5nn W 5nn W 5nn W 5nn lnrev_inb 0.339*** 0.410*** 0.394*** 0.427*** (0.039) (0.040) (0.039) (0.039) lnrev_dom 0.416*** 0.472*** 0.466*** 0.474*** (0.032) (0.033) (0.032) (0.033) lnpopden 0.180*** 0.0134 0.0429 0.104 0.153*** 0.117*** 0.0812** 0.0547 (0.052) (0.079) (0.079) (0.084) (0.031) (0.035) (0.037) (0.041) lntertiary 0.318 0.420 0.418 0.630* 0.265 0.264* 0.318** 0.404*** (0.349) (0.324) (0.304) (0.330) (0.164) (0.153) (0.145) (0.155) lnresource 0.174** 0.297*** 0.291*** 0.281*** 0.147*** 0.161*** 0.152*** 0.163*** (0.082) (0.082) (0.078) (0.081) (0.040) (0.039) (0.037) (0.038) lnhotel 0.339*** 0.207** 0.175** 0.219** 0.00511 0.00602 0.0122 0.0114 (0.085) (0.089) (0.085) (0.087) (0.039) (0.040) (0.039) (0.040) lnGDP_rate 0.473* 0.460* 0.431 0.456* 0.330*** 0.256** 0.222* 0.218* (0.254) (0.277) (0.267) (0.275) (0.119) (0.127) (0.125) (0.129) W*lnrev_inb 0.237*** (0.073) W*lnrev_dom 0.316*** (0.063) W*lnpopden 0.411*** 0.0319 (0.115) (0.067) W*lntertiary 0.118 0.385 (0.740) (0.340) W*lnresource 0.437*** 0.0398 (0.156) (0.074)
83 Table 3 4. Continued Variable Model 3.9 Model 3.10 Model 3.11 Model 3.12 Model 3.13 Model 3.14 Model 3.15 Model 3.16 W*lnhotel 0.274* 0.181** (0.157) (0.071) W*lnGDP_rate 0.298 0.216 (0.489) (0.228) constant 2.350*** 4.069*** 4.692*** 1.348 1.232*** 2.122*** 3.038*** 1.032** (0.651) (0.785) (0.798) (1.219) (0.294) (0.311) (0.437) (0.465) 0.502*** 0.349*** 0.532*** 0.251*** 0.340*** 0.402*** (0.060) (0.131) (0.059) (0.063) (0.121) (0.067) 0.630*** 0.771*** 0.477*** 0.660*** (0.058) (0.055) (0.067) (0.071) 1.264*** 1.209*** 1.152*** 1.179*** 0.596*** 0.565*** 0.540*** 0.554*** (0.049) (0.048) (0.050) (0.046) (0.023) (0.022) (0.023) (0.021) N 342 342 342 342 342 342 342 342 Wald1 70.169*** 119.947*** 284.619*** 79.900*** 15.971*** 50.798*** 117.099*** 35.542*** Wald2 48.264*** 47.491*** AIC 1165.9 1147.6 1143.6 1133.0 638.5 613.3 607.6 606.9 BIC 1200.5 1182.1 1181.9 1190.5 673.0 647.8 646.0 664.4 Notes: *** indicates significance at the 0.01 level, ** indicates significance at the 0.05 level, and indicates significance at the 0.10 level. Standard error presented in parenthesis. Wald1 refers to the Wald test on rho and (or) lambda, and Wald2 refers to t he Wald test on spati al lag of independent variables.
84 Table 3 5. Decomposition of spatial effects of independent variables Inbound model (Model 3.12) Domestic model (Model 3.16) Direct effect Indirect effect Total effect Direct effect Indirect effect Total effect lnrev_inb 0.425*** 0.020 0.404*** lnrev_dom 0.462*** 0.198** 0.264** lnpopden 0.053 0.707*** 0.654*** 0.055 0.019 0.036 lntertiary 0.685* 0.862 1.548 0.446*** 0.907 1.354** lnresource 0.245*** 0.577* 0.332 0.164*** 0.042 0.206 lnhotel 0.266*** 0.797** 1.063*** 0.002 0.289*** 0.291** lnGDP_rate 0.543** 1.137 1.680* 0.240** 0.497 0.737** Notes: *** indicates significance at the 0.01 level, ** indicates significance at the 0.05 level, and indicates significance at the 0.10 level.
85 Table 3 6. Estimates of spatial growth regression model with different weighting matrices Variable Model 3.17 Model 3.18 Model 3.19 Model 3.20 Model 3.21 Model 3.22 Inbound Inbound Inbound Domestic Domestic Domestic Model SDM SDM SDM SDM SDM SDM Spatial weight W dist W 10nn W 15nn W dist W 10nn W 15nn lnrev_inb 0.443*** 0.431*** 0.423*** (0.040) (0.039) (0.040) lnrev_dom 0.468*** 0.487*** 0.492*** (0.032) (0.033) (0.032) lnpopden 0.0131 0.0900 0.0475 0.0964*** 0.0659* 0.0477 (0.074) (0.082) (0.082) (0.037) (0.039) (0.039) lntertiary 0.559* 0.569* 0.568* 0.388** 0.461*** 0.448*** (0.336) (0.329) (0.336) (0.153) (0.150) (0.150) lnresource 0.279*** 0.245*** 0.252*** 0.136*** 0.160*** 0.158*** (0.083) (0.081) (0.083) (0.039) (0.038) (0.038) lnhotel 0.243*** 0.225** 0.249*** 0.0346 0.0200 0.0311 (0.090) (0.088) (0.088) (0.041) (0.040) (0.039) lnGDP_rate 0.537** 0.374 0.568** 0.172 0.148 0.130 (0.271) (0.281) (0.278) (0.123) (0.129) (0.126) W*lnrev_inb 0.425*** 0.271*** 0.206* (0.140) (0.100) (0.120) W*lnrev_dom 0.528*** 0.465*** 0.483*** (0.113) (0.080) (0.095) W*lnpopden 0.420** 0.412*** 0.392** 0.124 0.122 0.0666 (0.207) (0.142) (0.163) (0.108) (0.080) (0.091) W*lntertiary 1.276 0.702 0.973 0.219 0.425 0.193 (1.413) (1.099) (1.300) (0.567) (0.488) (0.596) W*lnresource 0.553* 0.435** 0.470* 0.0537 0.0679 0.0987 (0.310) (0.208) (0.247) (0.148) (0.099) (0.117) W*lnhotel 0.317 0.338 0.464* 0.310*** 0.252*** 0.221** (0.254) (0.211) (0.271) (0.107) (0.092) (0.109) W*lnGDP_rate 1.010 0.935 0.419 0.672 0.352 0.602* (1.066) (0.617) (0.756) (0.476) (0.287) (0.345) constant 2.899 0.363 0.384 0.678 0.793 0.00522 (2.427) (1.660) (1.943) (0.771) (0.598) (0.702) 0.774*** 0.587*** 0.622*** 0.664*** 0.515*** 0.586*** (0.091) (0.070) (0.077) (0.114) (0.082) (0.095) 1.206*** 1.181*** 1.200*** 0.551*** 0.542*** 0.544*** (0.047) (0.046) (0.046) (0.021) (0.021) (0.021) N 342 342 342 342 342 342
86 Table 3 6. Continued Variable Model 3.17 Model 3.18 Model 3.19 Model 3.20 Model 3.21 Model 3.22 Wald1 73.021*** 70.469*** 66.072*** 33.768*** 39.282*** 38.236*** Wald2 45.930*** 47.174*** 32.214*** 46.852*** 61.262*** 62.122*** AIC 1142.6 1128.1 1136.0 602.7 591.0 593.4 BIC 1200.2 1185.6 1193.5 660.2 648.5 650.9 Notes: *** indicates significance at the 0.01 level, ** indicates significance at the 0.05 level, and indicates significance at the 0.10 level. Standard error presented in parenthesis. Wald1 refers to the Wald test on rho and (or) lambda, and Wald2 refers to the Wald test on spati al lag of independent variables.
87 Table 3 7. GW SDM parameter summary of inbound tourism growth Minimum 25% percentile Median 75% percentile Maximum ln rev_inb 0.512 0.484 0.456 0.427 0.249 ln popden 0.409 0.018 0.036 0.062 0.539 ln tertiary 0.317 0.324 0.555 0.717 1.747 ln resource 0.021 0.265 0.322 0.374 0.443 ln hotel 0.016 0.158 0.217 0.346 0.640 ln GDP_rate 0.040 0.369 0.503 0.572 1.253 W*ln rev_inb 0.389 0.098 0.178 0.199 0.590 W*ln popden 0.300 0.368 0.448 0.573 1.058 W*ln tertiary 0.338 0.742 0.914 1.065 1.840 W*ln resource 1.496 0.672 0.464 0.271 0.064 W*ln hotel 0.207 0.305 0.432 0.580 1.242 W*ln GDP_rate 0.451 0.296 0.034 0.272 1.590 constant 1.338 2.262 2.551 3.094 5.728 0.064 0.446 0.476 0.497 0.594 Table 3 8. GW SDM parameter summary of domestic tourism growth Minimum 25% percentile Median 75% percentile Maximum ln rev_dom 0.706 0.559 0.431 0.337 0.209 ln popden 0.326 0.009 0.087 0.164 0.249 ln tertiary 0.197 0.351 0.426 0.549 1.112 ln resource 0.009 0.049 0.117 0.243 0.573 ln hotel 0.073 0.045 0.021 0.023 0.393 ln GDP_rate 0.511 0.096 0.149 0.237 0.520 W*ln rev_dom 0.596 0.098 0.281 0.413 0.568 W*ln popden 0.391 0.303 0.127 0.139 0.855 W*ln tertiary 0.567 0.210 0.539 0.718 1.789 W*ln resource 0.756 0.148 0.103 0.052 0.749 W*ln hotel 0.384 0.131 0.056 0.020 0.672 W*ln GDP_rate 0.614 0.058 0.235 0.282 1.434 constant 0.948 1.739 2.298 2.744 7.108 1.000 0.190 0.339 0.448 0.579
88 Figure 3 1 Histogram of lnrate_inb Figure 3 2 Histogram of lnrate_dom
89 Figure 3 3 Local Moran cluster map of the growth rate of inbound tourism revenue Figure 3 4 Local G* cluster map of the growth rate of inbound tourism revenue
90 Figure 3 5 Local Moran cluster map of the growth rate of domestic tourism revenue Figure 3 6 Local G* cluster map of the growth rate of domestic tourism revenue
91 Figure 3 7 Plot of bandwidth against CV score Figure 3 8 Spatial distribution of the GW SDM spillover coefficient in the inbound tourism model
92 (a) (b) ( c ) ( d ) ( e ) ( f ) Figure 3 9. Spatial distribution of the GW SDM coefficient s in the inbound tourism model
93 Figure 3 10 Spatial distribution of the GW SDM spillover coefficient in the domestic tourism model
94 (a) (b) ( c ) ( d ) ( e ) ( f ) Figure 3 11. Spatial distribution of the GW SDM coefficient s in the domestic tourism model
95 CHAPTER 4 ANALYSIS OF TOURISM ECONOMIC IMPACTS AND THEIR DETERMINANTS Chapter Summary To understand the economic contributions made by the tourism industry, this chapter applie s a set of tools to understand tourism's linkages and multipl iers from the input output table of 30 provinces in China at the year of 2002. After obtaining inter sector linkage measures, factor analysis and cluster analysis are utilized to recognize the pattern of the forward and backward linkage structure of the to urism industry and group micro level industries into s I t i s also found that tourism is weakly backward linked to other industries i Finally, a latent class regression model is applied to analyze determinants of the tourism economic multipliers and identify several latent classes of tourism economic impacts In general, the profit of tourism firms within the province is found to be an important factor determining the output multiplier and the type I income multiplier of tourism Introduction The tourism industry enjoys a reputation for substantial multiplier effects on local economies and its striking economic importance has been emphasized in many studies ( Archer, 1995 ; Croes & Vanegas, 2008 ; Vaughan, Farr, & Slee, 2000 ) Ryan ( 2003 ) sug gested that the general economic benefits of tourism includ e earning foreign exchange and taxation revenue, generating employment, redistributing income, etc. To understand the economic contributions made by the tourism industry, various multipliers have b een use d to measure the economic importance of tourism to the local economy such as output multiplier, income multiplier, employment multiplier, and g overnment
96 revenue multiplier ( Gasparino, Bellini, Del Corpo, & Malizia, 2009 ) Several methods have been applied to estimate these impacts like the economic base model ( Hefner, 1990 ) the input output (I O) model ( West & Gamage, 2001 ) the social accounting matrix (SAM) model ( Daniels, Norman, & Henry, 2004 ) and the computable general equilibrium (CGE) model ( Sugiyarto, Blake, & Sinclair, 2003 ) In past literature, t he I O model has been extensively used to calibrate the linkage of tourism to other industries and the total economic effects that tourism triggers ( Song, Dwyer, Li, & Cao, 2012 ) The I O table records the transactions between large numbers of industries and provides valuable information to trace the economic impact of the tourism industry upon other related industries and the whole economy. There fore, it provides a reduced form picture that is tractable and representative of the overall economic structure. The analysis of the I O table dates back to early works in both economics and geography ( Hewings, 1985 ) In economics, the theoretical background is rooted in the concept of Marxian and Walrasian general equilibrium, suggesting a stable state of transactions between industries in an economic system ( Miller & Blair, 1985 ) On the other hand, the linkage to geography can be found in several early works on urban economic base and city classification ( Hewings, 1985 ) In the 1980s, I O analysis attracted sizable research efforts from economic geographers and regional scientist to understand the regional economic structure and appraise the e conomic impact. In the tourism economic impact literature, a large variation in the calculated multipliers has been observed. For example, while Archer ( 1995 ) and Heng and Low ( 1990 ) highlighted the great economic importance of tourism to the local economy,
97 Mbaiwa ( 2005 ) and Stoeckl ( 2007 ) found that tourism brought very limited contribution s to t he local economy in their study areas. By using I O analysis, Heng and Low (1990 ) found that the tourism income multiplier in Singappre is 0.98, while Lee (1996 ) calibrated the income multiplier of tourism in South Korea to be 0.49. Other tourism income multiplier results include 0.65 for metropolitan Victoria, Canada ( Liu & Var, 1983 ) ; 0.80 for Hawaii, U.S ( Liu, 1986 ) ; and 0.40 for Tanzania ( Kweka, Morrissey, & Blake, 2003 ) To unveil factors contributing to this variation, some studies investigated its determinants through past published articles and identified such factors as publication date and the scale of study region ( Baaijens, et al., 1998 ; van Leeuwen, Nijkamp, & Rietveld, 2006 ; van Leeuwen, et al., 2009 ) However, it will be much more meaningful to investigate the variation of tourism economic impacts by a cross sectional sample from similar areas, and shed light on the essential factors explaining tourism s economic benefits. Another weakness of the p ast literature lies i n the lack of comparison and integration of tourism impact in multiple areas, since m ost studies focus on the impact analysis within a particular area or region. The in depth analysis on multiple regions could provide more information on the heterogeneity of tourism linkage and the industrial structure which delineate s the forward and back ward linkage between tourism and other industries. To the best of my knowledge, this research represents one of the first efforts in tourism studies to highlight industrial clustering in terms of their linkages to the tourism industry and the regional hete T o fill the research gap s discussed above I O analysis is applied in this chapter to calculate the economic impact of the tourism industry on the local economy of 30
98 provinces in China at the year of 2002. After ob taining inter sector linkage measures from the I O table, some multivariate statistical analysis tools are utilized to recognize the pattern of the forward and backward linkage structure of the tourism industry. By doing this, I identify similar micro leve l industries and group them into a smaller set of industries, which can be recognized as industrial cluster regional economics Finally, a latent class regression model is applied to analyze determinants of the tourism econom ic multiplier s calibrated from I O analysis and identify several latent class es of tourism economic impacts Literature Review Economic Impact of Tourism Economic impact analysis is an important element of a toolbox that can be adopted by tourism planners to look into the economic benefits generated by tourism ( Tyrrell & Johnston, 2006 ) To estimate the economic impacts of the tourism industry, several methods have become fair l y popular. Ryan ( 2003 ) identified four methods : multiplier studies, input output measure, satellite accounts, and local impact studies with ad hoc measures. More specifically, frequently utilized tourism economic impact models include the economic base model ( Hefner, 1990 ) the input output (I O) model ( West & Gamage, 2001 ) the social accounting matrix (SAM) model ( Daniels, et al., 2004 ; Polo & Valle, 2009 ) the computable general equilibrium (CGE) model ( Sugiyarto, et al., 2003 ) and the tourism satellite account (TSA) model ( Frechtling, 2010 ) The I O model has been extensively used in examini ng the economic impact of tourism, especially in early years. It enables researchers to trace flows of goods and services through different sectors and estimate the direct and indirect impacts of tourism on the local economy ( Klijs, Heijman, Maris, & Bryon, 2012 ; Song, et al., 2012 )
99 Regarded as an extended I O model, the CGE model includes a set of equations capturing the major relationships within the economy. The model treats the economy as an integrated system with feedback mechanisms and considers disequilibrium and m arket imperfections within the local economy ( Klijs, et al., 2012 ; Song, et al., 2 012 ) Furthermore, another tourism economic impact model, the TSA model, becomes more and more popular in both academia and industry. It refers to a standard method recommended by UNWTO for measuring tourism economic contr ibutions, which rests upon a statistical accounting framework Based on various tourism economic impact models, t he variation in the magnitu d e of tourism economic impact has be en observed across different regions. By reviewing past estimates of tourism mu ltipliers, Lee (1996 ) suggested that their sizes depend on the strength of inter sectoral linkages, the degree of lea kages outside the economy, and the general size of the economy. Pratt (2011 ) also found that economies with a larger population have a larger tourism output/income multiplier. According to the Huse, Gustavsen, and Almedal (1998 ) discovered that this impact hinges on local infrastructure local tourism age, and attributes of tourism industry. Ryan ( 2003 ) explained several major determinants of tourism economic impacts, which include the level of economic development, the nature of tourist at tractions and facilities, ownership of tourism business, the employment of labor, government provision of infrastructure tourist type, and the linkage to other economic sectors. From the I O analysis results of 114 regions in five states of the U.S, Chang ( 2001 ) found that tourism economic multiplier s can be explained by the regional population size and the classification of regions as
100 s Furthermore, using a survey data from 429 different tourism enterprises in Northern Australia Stoeckl ( 2007 ) showed that business level multipliers of tourism industry is negatively correlated with the remoteness of l ocation, and he argued that this may be explained by the fact that tourism is weak ly link ed to the major industry, agriculture, in the study areas. To synchronize the results from previous studies and provide a systematic review of tourism economic impa ct meta regression has been used to explore possible factors influencing the results of tourism multipliers obtained from past studies ( Baaijens, et al., 1998 ; van Leeuwen, et al., 2006 2009 ) Van Leeuwen, Nijkamp, & Rietveld ( 2006 ) suggested that the size of the tourism multiplier is associated with visitor numbers and expenditures, natural and cultural attractions in the area, and admin i strative level s. In th eir additional research, they found that the size of the economic base and total tourism demand also help ed to explain the variation of tourism multipliers ( van Leeuwen, et al., 2009 ) Application of I nput O utput Analysis in Tourism T here has been a lon g tradition in tourism studies of us ing I O models to evaluate the regional economic impact brought about by the tourism industry. This can be explained by the several advantages I O analysis offers First, regional I O tables are effortless to obtain in m ost regions and provide precise data sources for I O analysis. Therefore, it is not necessary to conduct a large scale survey to get the data. Second, compared to other complicated models, it is fairly easy to compute and comprehend the results from an I O model regarding the inter sector al linkage and the overall economic importance to the local economy. Other noticeable advantages of I O analysis include the flexibility in research scales and the ability to study tourists with distinct expenditure
101 pattern s ( Sun, 2005 ) However, several major drawbacks of I O model s have been noted, such as its unrealistic assumptions and the absence of income feedbac ks, resource limitations, and price adjustments ( Polo & Valle, 2009 ) The purchase and sale c oefficients matrices derived from I O tables describe the relationship between different industries in the economy and provide sufficient information about the industrial structure of the region/nation. Based on 52 potential criteria, Klijs, et al. (2012 ) evaluated the performance of five economic impact models of tourism, including the export base model, the Keynesian model, the ad hoc model, the I O model, and the CGE model. Their results suggested I O models as the compromise after taking several essential criteria into consideration, and recommended I O models to analyze tourism impa cts when the research model is not too complex. Moreover, some previous studies highlighted that the I O model yields similar results as those from the CGE model ( McGregor, Swales, & Yin, 1996 ) and I O analysis is particularly suitable for small and open economies ( West, 1995 ) The validity of I O analysis hinges heavily on several important assumptions embedded in the I O table. First, each industry in the table should be exclusively defined without overlaps between industries, and households are assumed to purchase goods from the same division of industries. Second, on the supply side, the linearity assumption suggests a linear relationship between industrial inputs and outputs while the homogeneity assumption refers to the identity of goods that are produced by ever y industry. Third, the demand supply equilibrium should be achieved and the price is expected to be exogenous to the I O table. In I O applications on tourism impact analysis, several extensions have been made to relax some of these assumptions, such
102 as th e non linear I O model ( West & Gamage, 2001 ) the capacity constrain model ( Fletcher, 1989 ) and the occupation base model ( Daniels, et al., 2004 ) Apart from the traditional use of I O analysis in the d irect assessment of tourism economic multipliers, another important I O application in tourism studies is the analysis of income/employment red istribution Heng and Low (1990 ) found that the estimated tourism economic impacts are relatively uniform across tourists from different origins and with different purposes. Archer and Fletcher (1996 ) and Sun (2005 ) compared the economic impact of different inbound tourist segments according to their expenditure data and I O analysis and picked up certain tourist segments that are more influential to local economies. The inter industrial linkage between tourism and other industries can also be examined by I O models. As noted by Lee (1996 ) it is important to understand who gains and who loses from tourism growth, and it is also important to recognize the industries that are substantially linked to tourism. Beynon, Jones, and Munday (2009 ) applied linkage analysis to study the intensity o f transaction linkage s of tourism related sectors in Wale s ; to analyze how these linkages contribute to other parts of the economy. However, it appears that no known study has adopted multivariate methods to study the industrial clustering according to their linkages to tourism, which is important for recognizing the inter industrial connection between tourism and other industries in the economy. Industrial Grouping/Complexes A nalysis Since an I O table incorporates the inter sector al purchase and sale flow information of a large number of industries the table is too complex to be used to
103 generalize about the industrial structure and policy implications. As a result, it is more preferable for economists planners and political decision makers to understand a smaller number of industri al clusters rather than the scatter of heterogeneous industries presented in the original I O table ( Czamanski & Ablas, 1979 ) Several endeavors have been made by economic geogra phers and regional scientists to simplify the I O table for further analysis. For example, various dimension reduction methods have been introduced to identify the group of industries or industry complexes to depict a more concise and simplified structure of local economic sectors ( hUallachin, 1984 1985 ; Roepke, Adams, & Wiseman, 1974 ) The terminology industrial cluster and industrial complex have been heavily used in early literature to present a more concise understanding of the structure of I O tables. According to Czamanski and Ablas (1979 ) an industrial cluster is a subset of indus tries that are connected by flows of goods and services that are stronger tha n those linking them to other industries. T he industries within the same industrial cluster are believed to be more homogeneous. Moreover, industrial complex places emphasis on the spatial concentration of industries in addition to general connections linked by flows of goods and services However, in many papers, the industrial complex was used interchangeably with the term industrial cluster. Finally, it should be noted that th e analysis of industrial cluster/complexes can be regard ed as a method of aggregating the I O table, which is another important stratum of I O analysis ( Oksanen & Williams, 1992 ) To group different industries according to their interconnectedness embedded in the I O table, Streit (1969 ) designed a coefficient to measure the intensity of inter
104 industrial flows and applied it to recognize the grouping of industries in France and Germany Czamanski and Ablas (1979 ) compared several methods including several multivariate analysis tools and triangularisation/graph theory methods t o identify industrial clu sters and complexes of 78 industr ies. They highlighted significant differences between the results of different methods. hUallachin (1984 ) utilized principle component analysis (PCA) aided by eigenvector rotation to identify regional industrial complexes in the state of Washington based on 49 49 purchase and sale coefficient matrices, and 15 complexes were finally identified. He argued that the PCA results provide more information on the industrial structure than those from simple eigenvector analysis and graph theo ry method. Factor analysis has been frequently used in I O analysis to understand the latent industrial clusters. With a 44 44 I O table, Roepke, et al (1974 ) applied factor analysis to look into the industrial complexes in Ontario, Canada. According to different matrices specified for factor analysis, three types of complexes were identified, and t hey found that different methods generate very simil ar results on the grouping of industries. Oksanen and Williams (1992 ) furthermore, employed factor anal ysis to aggregate an I O table of 66 industries and advocated factor analysis on the direct plus indirect matrix to keep a reasonable number of industrial clusters. Moreover, Loviscek (1982 ) highlight ed the needs to consider both backward and forward linkages in identifying industrial clusters through I O analysis By looking into the clustering results from purchase and sale coefficient matrices, he found striking differences in terms of inter industry relationship. Since the single backward or forward measure could be misleading, he proposed a weighted average measure of the two to address demand
105 and supply constra ints simultaneously. Apart from first order factor analysis used in most literature, hUallachin (1985 ) applied higher order factor analysis to examine the hierarchical industrial structure of regional econo mies through requirement/sales correlation matrices from the I O table The aforementioned methods are capable of identifying the group of industries based on a single I O table of all national/local economic sectors. In this study, the linkages between to urism and other relevant industries are of particular interest As a result, to identify the group of industr ies in terms of their inter connectedness with tourism in multiple I O tables, the method adopted in this research should be a little different fro m past literature It appears that even though I O analysis has been popular in tourism research over decades no known study has looked into the grouping of tourism related industries. This study is expected to contribute to the current body of literatur e by understanding the structure of industries that are highly linked to tourism, either backward linked or forward linked. Research Methods and Data Input Output Analysis A typical I O table assumes that the economic activity of a region can be partitione d to a number of industries, and each industry produces a single and unique product. I O analysis relies on the I O account of a particular region, and the I O account captures the interconnectedness of various industries by recoding the transaction betwee n different industries as either suppliers or demanders. Apart from the information between industries, the I O account also lists the values of sales of each industry to final consumers, including households, exports, investments, and government units. Mo reover, the additional purchase information outside all industries
106 of the region is recorded, such as the payments for value added items and net imports. A typical I O table is specified as follows ( Isard, et al., 1998, p. 45 ) To From Intermediate use Final use Total 1 2 . j . n Consum ption Invest ment Govern ment Exports 1 x 11 x 12 . x 1j . x 1n c 1 i 1 g 1 e 1 X 1 2 x 21 x 22 . x 2j . x 2n c 2 i 2 g 2 e 2 X 2 . . . . . . . i x i1 x i2 . x ij . x in c i i i g i e i X i . . . . . . n x n1 x n2 . x nj . x nn c n i n g n e n X n Value added v 1 v 2 . v j . v n V Imports m 1 m 2 . m j . m n M Total X 1 X 2 X j X n C I G E where x ij = the monetary value of sales from industry i to industry j ( i j n ); X i = the total value of goods produced by industry i in the region; v i = industry i v 1 + v 2 +.. + v n ; m i = industry i m 1 + m 2 +.. + m n ; c i = household consumption on industry i c 1 + c 2 +.. + c n ; i i = purchases of goods from industry i as investments, and I = i 1 + i 2 +.. + i n ; g i = government purchases of goods from industry i and G = g 1 + g 2 +.. + g n ; e i = exports of goods of industry i and E = e 1 + e 2 +.. + e n To complete X i = x 1i + x 2i +.. + x ni + v i + m i = x i1 + x i2 +.. + x in + c i + i i + g i + e i In the t able each of the first n how many has been consumed by intermediate use of other industries and how many has been sold to the final consumers. On the other hand, each of the first n columns presents the distribution of an indus important information considered in this study comes from the upper left block of the I O
107 table, which indicates the economic transactions between industries. This is known as an I O transaction matrix, and de noted as Z which is an n n square matrix. Column vectors in this matrix represent direct transactions as inputs to various industries, while row vectors denote direct transactions as outputs of different industries. Before calculating certain measures o n industrial linkages and multipliers, several new matrices have to be defined. First, an n n technical coefficients matrix A which is also known as a direct input coefficients matrix or direct requirement matrix, is defined as: (4 1) where a ij indicates the proportion of input from industry i used to produce output of industry j One can also obtain an n n direct output coefficient matrix B which is also known as a sales coefficient matr ix, as follows: (4 2) where b ij indicates the proportion of industry i j as input for production. In matrices A and B only the direct interconnectedness between industries is captured, and the indirect effect coming from the loops is overlooked. To look into both direct and indirect effects, two matrices A and B are specified as follows: (4 3) (4 4)
108 where I is an n n identity matrix. Matrix A is known as a Leontief Inverse or total requirement matrix, while matrix B is known as a Ghosh Inverse matrix. After obtaining these two new matrices, one can calculate the total forward and backward linkage of each industry. The simple backward linkage of a particular industry is defined as the corresponding column sums of A *, while the simple forward linkage is d efined as the corresponding row sums of B *. To facilitate the comparison across different provinces, the direct linkage measures are defined as the standardized linkage in each province: (4 5) (4 6) where TBL j denotes the total backward linkage of industry j and TFL i denotes the total forward linkage of industry i is the i,j th element in matrix A *, and it reflects the value of outputs in industry i that are required for a unit of new final demand for output in industry j ( Miller & Blair, 1985 ) is the i,j th element in matrix B *. The I O table also enables us to calculate a set of multiplier measures to reflect the impact of tourism on the overall economy. The first multiplier measure of industry j is the output multiplier, O j which is the specified as the unstandardized version of the total backward linkage: (4 7)
109 Output multiplier of an industry represents how much each unit of final demand for a particular industry is blown up to the output requirement of all industries in the economy ( Isard, et al., 1998 ) Therefore, the output multiplier is able to m easure the degree of total interconnectedness between tourism and other industries. Generally, researchers are more likely to be concerned with the economic impact of final demand measured by household income, employment, tax revenue, etc., rather than the gross output of each industry. Therefore, a more informative multiplier measure that can be obtained from the I O table is the income multiplier. There are two basic types of income multipliers that can be calibrated through an I O table: type I income mu ltiplier and type II income multiplier ( Miller & Blair, 1985 ) For type I income multiplier, it is specified as: (4 8) where Y j denotes the type I income multiplier of industry j l i indicates the labor input coefficient of industry i indicating household income received per unit of output. This multiplier represents direct and indirect effects of a change in output by industry j on total household income in the region. In this chap ter I O models are applied to calculate various multipliers of the tourism industry and forward and backward linkages to other industries The reason to choose the I O model is twofold. First, the I O model has been extensively applied in touris m research and it is particular suitable for the analysis of inter sector al relations ( Cai, Leung, & Mak, 2006 ; Fletcher, 1989 ; Frechtling & Horvth, 1999 ; West & Gamage, 2001 ) Compared to results obtained from some more sophisticated models, like the CGE model the I O model yields similar results regarding tourism i mpacts ( Zhou, Yanagida, Chakravorty, & Leung, 1997 ) Second, d ue to the unavailability of
110 social accounting matrix data, further analysis on the SAM and CGE models requires tremendous efforts for multiple regions. Therefore, several major ind ex es will be calculated from the I O model, including the total forward link age index, the total backward linkage index, the output multiplier, and the type I income multiplier ( Isard, et al., 1998 ) After basic I O analysis factor analysis will be applied to extract latent factors of backward and forward linked industries of tourism based on the data of b ackward linkage elements (in A *) and forward linkage elements (in B *) of different sectors in 30 provinces of China. As suggested by hUallachin (1985 ) factor analysis can efficiently reduce the d imension and reproduce the inter industry proportion of the I O table by providing a parsimonious accounting of regional industrial structures. It is of particular interest to obtain the industrial clustering results from backward linkage indexes because t he r equirement similarity in A is important to combine industries that are homogeneous as suppliers to the tourism industry. Latent Class Model In Chapter 3 spatial heterogeneity was incorporated by using a generalized geographically weighted regression framework, and the localized coefficients can be obtained within this framework In this part, a latent class model is introduced to unveil the potential heterogeneity in factors determining tourism economic multipliers. The regression coefficients in this model are specified to be the same for observations within the same latent class while varying across difference classes. The l atent class regression (or finite mixture regression) model specifies the density of a dependent variable, y a s a linear combin ation of J different densities, which is specified as:
111 ( 4 9 ) where i indexes the observation and j indexes the latent class, j J is the density of j th class(component) and is the probability of being j th class Since is varying across different classes and it is assumed that and for the j th c lass regression, the factors contribute to multiplier effects for different classes in a different way. Finally, a constant only multinomial logit model can be assume for ( Cameron & Trivedi, 2009 ) It is specified as: ( 4 10 ) where the J th coefficient of is set to be zero for identification purposes. By taking Equation 4 10 into Equation 4 9, the parameters , and can be obtained by the maximization of the log likelihood function. To determine the empirical optimal value of J (the total number of cla sses) in the latent class model, a common way is to compare the information criteria (e.g. AIC and BIC) associated with different J values from maximum likelihood estimation ( Bhatnagar & Ghose, 2004 ; Clark, Etil, Postel Vinay, Senik, & Van der Straeten, 2005 ) In general, lower values of information criteria measures are preferred. After the estimation of parameters, the posterior probability of class membership for each observation i can be derived using Bayes theorem, and it is specified as:
112 ( 4 11 ) where refers to the probability densit y function of the standardized normal distribution. The class membership is predicted with the highest posterior probability In the empirical latent class model of tourism multiplier s the dependent variable, y is the tourism economic multiplier (outpu t or income multiplier) of each province obtained from the I O model. The independent variables x will be defined in the later sections This type of model can be estimated by LIMDEP 9.0 with built in syntax for latent class regression models ( Greene, 2007, pp. 558 567 ) Data Description T he I O table s with 4 2 sectors ( industries) of 30 of Chin ese provinces in 2002 were obtained from C hina National Statistics Bureau Due to the unavailability of I O table data, Tibet, which is also a provincial level unit in China, is excluded from the research sample. For brevity, a sim ple label is assigned to each industry. The tourism industry is listed as a separate industry in these tables, which is defined as the industry that provides particular services to both packaged and non packaged tourists and business travelers. These servi ces include, but not limit to, traveling consultation, schedule arrangement, tour guide, food and accommodation service, and transportation during their tour. Compared to the I O tables used in previous research, the data adopted in this dissertation posse sses several advantages. First, the tourism industry
113 has been identified as a unique and integrated sector in the national economic account, and therefore, it inherently avoids possible aggregation bias if the tourism industry has to be aggregate d from mor e than one conventional industries in the I O table ( Briassoulis, 1991 ) Second, the data provides valuable information on regional heterogeneity in terms of economic structure. Since all 30 I O tables were calibrated under the supervision o the industrial linkage and multiplier measures can be directly comparable across different provinces. Dependent variables in the latent class regression model come from the results of I O analysi s, and the output multiplier and the type I income multiplier will be used. Regarding independent variables x, two variables will be considered, and they include ln GDP (the logarithm of total GDP of the province in 2002, in 100 billion RMB Yuan) and profit (the profits of tourism firms per employee in 2002, in 10,000 RMB Yuan). The data for these two variables was obtained from China Statistical Yearbook 2003 and China Tourism Statistical Yearbook (Supplementary) 2003, respectively. Descriptive statistics o f these two variables are presented in Table 4 1 The average value of lnGDP in all provinces is 7.975 with a standard deviation of 0.893, showing a substantial difference across different provinces. Moreover, the average profit per employee in tourism fir ms is 810 RMB Yuan, suggesting that a large number of tourism firms in China were suffering from negative profits. Results and Discussion Industr ial Grouping Analysis In this section t forward and backward linkage measures to different industries will be calculated for 30 provinces in China. After that, factor analysis is
114 performed based on these measures of different provinces to identify the grouping of tourism linked industries so that all other sectors are classified into a small num ber of groups according to linkage patterns to t he m. Since the original I O tables comprise a large set of industries, factor analysis is conducted to present a dissection of inter industry structure and group industries ackward and forward linkages to them. In the dataset for factor each other industry, and the row denotes the linkage measures of each province. After factor analysis of PCA w ith a varimax rotation, Figure 4 1 presents the scree plot of highlighted with a large than one eigenvalue, and a total of 86.93% variance has been explained by these 13 factors ( Table 4 2 ). As shown in Table 4 2 all 41 industries are classified into 13 groups according to according to their roles as suppliers to the tourism industry. Facto r 1 explains several manufacture industries and Factor 2 explains major petroleum industries. Furthermore, highlight the food service to tourists, while Factor 4 describes five business service and are grouped by Factor 6. As a caveat, when explaining the factor analysis results, the
115 grouping of these industries is not based on their inherent characteristics; instead, it s. The s ame factor analysis procedure is applied to group industries according Figure 4 2, only eight factors whose eigenvalues are large than one are found, and they explain a total of 92.98% variation. According to the results of factor analysis presented in Table 4 3 two dominant factors explain 66.59% of variance. Compared to the grouping results based on interpret. One possible re ason to explain this difficulty is that tourism usually weakly connects to other industries as a supplier, and the latent structure of forward linkage is hard to depict. Total Linkage Total backward and forward linkages (Equations 4 5 and 4 6, respectivel y) help to understand the importance of tourism in the local economy in relative to other industries. Recall that the total backward linkage reflects the relative importance of tourism as a purchaser to all other industries in the economy, while the total forward linkage represents the relative importance of tourism as a supplier to other industries ( Cai, et al., 2006 ) Total linkage measures indicate how the tourism industry is linked to other industries, and therefore can be used to determine the key industry in the local economy. To compare the regional difference on these two linkage measures of the tourism industry in different provinces, a two by two matrix plot is applied to show the values of these linkages based on the total forward and backward linkage coordinates. In Figure 4 3, the y axis denotes the total forward linkage while the x axis denotes the backwar d one. Each data point represents the total forward and backward
116 linkage measures of each province. Data points in the top left quadrant of the plot indicate provinces with above average forward linkages of tourism and below average backward linkages of to urism, while those in the bottom right quadrant includes provinces with below average forward linkages of tourism and above average backward economy can be categoriz ed into four different types ( Beynon, et al., 2009 ) : Key industry of the economy, if TBL > 1 and TFL > 1; Backward linkage oriented industry, if TBL > 1 and TFL < 1; Forward linkage oriented industry, if TBL < 1 and TFL > 1; Weak oriented industry, if TBL < 1 and TFL < 1. A general pattern observed from Figure 4 3 is that even though the total backward linkage of tourism is larger than one in 15 out of 30 provinces, it s total forward linkage is consistently less than one. Only Hebei has a total forward linkage index of more than one, and in particular, a large number of provinces have a total forward linkage index less than 0.2. These results are consistent with the fin dings from Cai, et al. (2006 ) and Pratt (2011 ) which showed that tourism oriented sectors are strongly backward linked to other sectors while weakly forward linked to others. This can be explained by the fact that tourism sectors provide goods and services directly to end users instead of intermediate users, and therefore, there are virtually l imited linkages from tourism to the downstream industries. Fifteen provinces have a total backward linkage index larger than one, and these provinces are characterized by an above average backward linkage of tourism compared to other industries in the prov ince. In particular, the tourism industry in
117 Gansu, Anhui, and Zhejiang has the largest total backward linkage index, showing the significant importance of tourism in these provinces. On the other hand, the total backward linkage index is found to be less than 1 in the other 15 provinces. Hainan, Hebei, and Tianjin are found to have the smallest values of this index, suggesting a fairy limited economic effect that tourism can generate to other industries as a seller. Multiplier Measures The multiplier measu res obtained from I O analysis serve as important indicators to assess the performance of tourism in the local economy, and a high tourism multiplier measure indicates large economic benefits that tourism contributes to the economy. Table 4 4 presents the results of two economic multiplier measures of the tourism Recall that an output multiplier of tourism measures the total value of production in the economy that is necessary to satisfy a unit of output for the tourism ind ustry. The output multiplier of Beijing is estimated to be 2.156, suggesting that additional outputs of 2.156 RMB Yuan from all sectors in the economy are required for 1 RMB Yuan new final demand for the output of the tourism industry in Beijing. Zhejiang, Guangdong, and Anhui are found to have the largest output multiplier of tourism in all of provinces, indicating that the tourism industry in these three provinces is highly connected with other industries as a purchaser. On the other hand, Hebe i, Hainan, and Qinghai are characterized by the smallest output multiplier of tourism, showing the limited importance of tourism as a purchaser in the local economy. The type I income multiplier is found to be highly correlated with the output multiplier, with a Pearson correlation coefficient of 0.702. According to Table 4 4 the type I income multiplier is estimated to be the largest in Gansu, Anhui, and Shanxi, indicating the substantial
118 economic impact of tourism on local household income. On the other hand, this multiplier is found to the smallest in Beijing, Hebei, and Hainan. Furthermore, the maps of output multipliers and type I income multipl iers are presented in Figures 4 4 and 4 5, respectively. Unlike the spatial distribution of tourist flows ( Yang & Wong, 2012a ) and hotel location ( Luo & Yang, 2012 ) in China, there is no observable spatial clustering of tourism multipliers across different provinces. A further dependence of these multipliers. Using a five nearest neighbor spatial weighting matrix, the global 0.034 with a pseudo p value of 0.529, and that of type I income multipliers is 0.102 with a pseudo p v alue of 0.223. Latent Class Modeling of Multipliers As discovered in the last section, a substantial variation has been identified in the estimated tourism economic multipliers across 30 of Chinese provinces. To unveil factors explaining tourism economic multipliers, two simple linear regression models are fitted on both the output multiplier and the type I income multiplier of tourism in 30 provinces of China, respectively. Due to the limited size of samples, only two independent variables are considered in the model. They are lnGDP and profit. The estimation results of these regression models are presented in Table 4 5 In Model 4.1 of the output multiplier, profit is estimated to be statistically significant and positive, suggesting that the profit of to urism firms per employee is a major determinant of output multipliers. On the other hand, although lnGDP is estimated to be positive, it is not statistically significant. In Model 4.2 of the type I income multiplier, the estimated coefficient of profit is positive albeit insignificant at 0.10 level, while lnGDP is also not statistically significant.
119 substantial economic, cultural, and physic differences between them. A latent class regression model is therefore introduced to capture this heterogeneity in tourism multiplier models. A set of latent class regression models are estimated according to the specification and procedure elaborated previously Before estimating the model, it i s necessary to determine the optimum number of latent class, J, in the model. Table 4 6 presents the values of information criteria for different numbers of classes specified in the model. As highlighted by the smallest values of AIC, BIC, and HQIC, a 3 cl ass model is selected for the output multiplier model while a 4 class model is chosen for the type I income multiplier model. Table 4 7 presents the estimation results of the latent class regression model on output multiplier (Model 4.3). In the three late nt classes discovered, lnGDP and profit are estimated to be statistically significant only in class 3. Using Equation 4 9, one can predict the class membership of each province. Table 4 9 presents these results of class membership. Only 5 provinces belong to class 3, and they include Heilongjiang, Shanghai, Henan, Hainan, and Qinghai. Table 4 8 presents the estimation results of the latent class regression model on the type I income multiplier of tourism (Model 4.4). Even though both lnGDP and profit are es timated to be insignificant in the linear regression model (Model 4.2), they are statistically significant in the models of certain classes. For example, lnGDP is estimated to be statistically significant and positive in the model for classes 2 and 3, sugg multipliers in these provinces. This can be explained by the fact that the large economy
120 base provides more diversified economic structure and therefore is more likely to absor b the leakage effect from tourism ( Huse, et al., 1998 ) This diversified economy also enables the tourism industry to find suitable suppliers, therefore contributing to a higher level of intercon nectedness with other industries ( Stoeckl, 2007 ; van Leeuwen, et al., 2009 ) profit is also found to be statistically significant and positive in these two classes. This can be explained by the fact that in a province with more profitable tourism firms, the tourism industry is more likely to connect to other industries through the divers ified operation strategy of the firm. The detailed list of provinces belonging to classes 3 and 4 can be found in Table 4 9 Chapter Conclusion The contribution of this research is to conduct linkage analysis of tourism across different provinces in China and compare the economic multipliers of tourism through latent linkage measures f or the various selected industries, factor analysis is used to find a small number of industrial cl usters. Moreover, according to results from total forward and backward linkage indexes of tourism, it was found that tourism is weakly backward models were used to expl provinces. The results of linear regression suggested that, in general, the profit of tourism firms per employee is positively associated with the outcome multiplier. By contrast, the latent class regression model highlighted the heterogeneity in multiplier models and demonstrated associated with both the outcome multiplier and the type I income multiplier.
121 The results from this res earch can be used to assess the effectiveness of industrial development strategies. To propose strategic plans for the tourism industry, certain efforts should also be allocated on other tourism facing industries as those highlighted in the results of indu strial grouping. Moreover, the comparison of tourism economic multipliers shows where the government spending on tourism would have the greatest impact on the local economy in terms of total value of outputs or household income. According to the results, c entral government should allocate more tourism budgets on provinces with higher tourism economic multipliers and linkages. Finally, since the profits of tourism firms are found to be an important factor determining the output multiplier and the type I inco me multiplier of tourism, extra efforts should be made to improve the performance of tourism business and therefore magnify the possible economic benefits of tourism on the local economy. Several limitations should be stated for this research. First, there are still several other linkage measures that can be used to look into the inter connectedness between industries, such as eigenvector measures ( Beynon, et al., 2009 ; Midmore, Munda y, & Roberts, 2006 ) Second, even though various multiplier s were utilized to investigate the impact of the tourism industry on the local economy, the results were limited to these impacts on output and income only, and did not cover the multipliers of employment and government expense. In future research, longitu dinal study should be conducted to trace the trajectory of inter industrial linkage change according to their linkages to tourism ( Cai, et al., 2006 ; Chen & Var, 2010 ; Henry & Deane, 1997 ) and examine the economic importance of tourism over different stages in the Tourism Area Life Cycle ( Pratt, 2011 )
122 Table 4 1 Descriptive statistics of i ndependent variables in the regression model Variable Defnition Mean Std. Dev. Min Max lnGDP Log of total GDP (in 100 billion RMB) 7.975 0.893 5.831 9.511 profit Log of profits of tourism firms per employee (in 10,000 RMB) 0.081 0.490 0.960 1.320
123 Table 4 Factor Factor loading Eigenvalue Proportion Cumulative Factor 1 10.080 24.59% 24.59% Chemicals 0.672 Metal Products 0.470 Electrical Machinery 0.628 Communication Equipment 0.590 Instruments 0.739 Other Manufacture 0.796 Power Supply 0.585 Technical Services 0.624 Factor 2 4.574 11.16% 35.74% Crude Petroleum 0.859 Petroleum Processing 0.898 Metal Smelting 0.629 General Machinery 0.718 Transport Equipment 0.683 Factor 3 3.271 7.98% 43.72% Agriculture 0.950 Food 0.871 Hotel and Restaurant 0.818 Factor 4 3.147 7.67% 51.40% Post 0.616 Business Services 0.703 Science and Research 0.535 Education 0.808 Culture 0.592 Factor 5 2.528 6.16% 57.56% Textile 0.682 Clothing 0.590 IT 0.868 Factor 6 2.125 5.18% 62.74% Other Mining 0.628 Nonmetallic Minerals 0.833 Construction 0.778 Factor 7 1.809 4.41% 67.16% Paper 0.560 Gas Supply 0.655 Water Supply 0.949
124 Table 4 2. Continued Factor Factor loading Eigenvalue Proportion Cumulative Factor 8 1.694 4.13% 71.29% Mining 0.853 Real Estate 0.494 Factor 9 1.601 3.91% 75.19% Waste 0.767 Other Social Services 0.776 Factor 10 1.441 3.51% 78.71% Sawmills 0.794 Wholesale and Retail 0.837 Factor 11 1.232 3.00% 81.71% Finance 0.640 Social Security 0.839 Factor 12 1.113 2.71% 84.43% Transport 0.458 Public Administration 0.900 Factor 13 1.029 2.51% 86.93% Metal Ore 0.779
125 Table 4 Factor Factor loading Eigenvalue Proportion Cumulative Factor 1 21.190 51.68% 51.68% Crude Petroleum 0.980 Petroleum Processing 0.859 Chemicals 0.873 Metal Smelting 0.959 Metal Products 0.781 General Machinery 0.702 Power Supply 0.898 Post 0.952 IT 0.837 Wholesale and Retail 0.900 Hotel and Restaurant 0.739 Business Services 0.947 Technical Services 0.924 Education 0.692 Social Security 0.881 Public Administration 0.880 Factor 2 6.110 14.90% 66.59% Metal Ore 0.703 Textile 0.859 Clothing 0.954 Sawmills 0.684 Paper 0.858 Electrical Machinery 0.695 Instruments 0.733 Gas Supply 0.631 Transport 0.828 Factor 3 3.413 8.33% 74.91% Agriculture 0.573 Food 0.809 Nonmetallic Minerals 0.911 Transport Equipment 0.975 Water Supply 0.608 Construction 0.699 Factor 4 2.400 5.85% 80.76% Finance 0.952 Real Estate 0.664 Other Social Services 0.671
126 Table 4 3. Continued Factor Factor loading Eigenvalue Proportion Cumulative Factor 5 1.614 3.94% 84.70% Communication Equipment 0.853 Other Manufacture 0.663 Culture 0.657 Factor 6 1.274 3.11% 87.81% Mining 0.837 Science and Research 0.746 Factor 7 1.103 2.69% 90.50% Other Mining 0.816 Factor 8 1.020 2.49% 92.98% Waste 0.965
127 Table 4 4. Tourism economic impact multipliers in different provinces Province Output Multiplier Rank Type I Income Multiplier Rank Beijing 2.156 26 1.477 30 Tianjin 2.168 25 1.830 27 Hebei 1.675 30 1.563 29 Shanxi 2.427 16 13.072 3 Inner Mongolia 3.147 6 5.910 8 Liaoning 2.266 21 2.391 19 Jinlin 3.124 7 5.827 9 Heilongjiang 2.200 24 2.282 20 Shanghai 3.427 4 5.544 10 Jiangsu 2.540 11 2.528 15 Zhejiang 3.739 1 10.545 5 Anhui 3.464 3 13.942 2 Fujian 2.211 22 1.938 24 Jiangxi 2.472 14 2.651 14 Shandong 2.282 20 1.960 23 Henan 2.363 17 2.505 16 Hubei 2.431 15 2.404 18 Hunan 2.345 18 2.681 13 Guangdong 3.536 2 7.102 6 Guangxi 2.645 10 11.700 4 Hainan 1.694 29 1.627 28 Chongqing 2.201 23 2.281 21 Sichuan 2.961 8 5.473 11 Guizhou 2.128 27 2.405 17 Yunnan 2.721 9 4.901 12 Shaanxi 2.513 12 6.380 7 Gansu 3.407 5 14.500 1 Qinghai 2.125 28 1.929 25 Ningxia 2.341 19 1.903 26 Xinjiang 2.482 13 1.984 22
128 Table 4 5. Estimation results of linear regression Model 4.1 Model 4.2 Output Multiplier Type I Income Multiplier lnGDP 0.079 0.061 (0.092) (0.782) profit 0.482*** 1.523 (0.148) (1.010) constant 1.979** 5.383 (0.719) (6.511) observations 30 30 R squared 0.251 0.033 AIC 44.40 172.6 BIC 48.61 176.8 Notes: indicates p< 0.10, ** indicates p<0.05, *** indicates p<0.01. Robust standard error in parenthesis. Table 4 6. Model fit based on information criteria Class AIC BIC HQIC Output multiplier model 2 1.530 1.950 1.664 3 0.786 1.440 0.995 4 1.222 2.110 1.506 Type I income multiplier model 2 4.479 4.899 4.613 3 3.951 4.605 4.161 4 2.736 3.623 3.020
129 Table 4 7. Estimation results of latent class regression on output multiplier Model 4.3 Class 1 Class 2 Class 3 lnGDP 0.049 0.030 0.076*** (0.350) (0.121) (0.001) profit 0.384 0.193 0.686*** (1.253) (0.136) (0.002) constant 2.999 2.626*** 1.867*** (2.747) (0.990) (0.007) sigma 0.126* 0.256*** 0.001 (0.076) (0.052) (0.002) probability 0.198*** 0.638*** 0.164** (0.073) (0.089) (0.068) AIC 0.786 BIC 1.440 Notes: indicates p<0.10, ** indicate s p<0.05, *** indicates p<0.01. Table 4 8. Estimation results of latent class regression on type I income multiplier Model 4.4 Class 1 Class 2 Class 3 Class 4 lnGDP 1.381 0.107*** 0.173*** 4.826 (1.514) (0.036) (0.051) (4.241) profit 4.698 1.480** 0.635*** 1.039 (2.103) (0.599) (0.093) (4.064) constant 23.417* 1.046*** 1.111*** 42.520 (12.375) (0.306) (0.401) (34.272) sigma 1.267 0.000 0.085** 0.439 (1.392) (0.002) (0.035) (0.436) probability 0.228 0.116 0.450*** 0.206** (0.279) (0.335) (0.140) (0.089) AIC 2.736 BIC 3.623 Notes: indicates p<0.10, ** indicate s p<0.05, *** indicates p<0.01.
130 Table 4 9. Latent class membership of different provinces Province Model 4.3 Model 4.4 Beijing 2 4 Tianjin 2 3 Hebei 2 4 Shanxi 2 1 Inner Mongolia 1 4 Liaoning 2 3 Jinlin 1 4 Heilongjiang 3 3 Shanghai 3 1 Jiangsu 2 3 Zhejiang 1 1 Anhui 1 1 Fujian 2 4 Jiangxi 2 3 Shandong 2 2 Henan 3 3 Hubei 2 3 Hunan 2 3 Guangdong 1 1 Guangxi 2 1 Hainan 3 3 Chongqing 2 3 Sichuan 2 1 Guizhou 2 3 Yunnan 2 4 Shaanxi 2 4 Gansu 1 1 Qinghai 3 3 Ningxia 2 2 Xinjiang 2 3
131 Figure 4 Figure 4
132 Figure 4 3. Tourism linkage matrix plot Figure 4 4. Map of output multipliers in different provinces
133 Figure 4 5. Map of type I income multipliers in different provinces
134 CHAPTER 5 CONCLUSION This dissertation presents an important effort towards a better understanding of tourist flows and impacts from spatial perspectives. V arious spatial and regional models were considered to examine tourist flows from both t he demand and supply sides and (Figure 1 1) In particular, this dissertation focused on tourism demand under a multi destination scenario and extend ed the regional growth regression model by considering spatial dependence spatial heterogeneity and supply side factors Moreover, a latent class regression analysis was used to explain regional difference s in tourism related economic multipliers. For the demand side analysis a three stage decision making model was introduced in Chapter 2 This model highlighted of subsequent destination primary destination was visited. The r esults of a nested logit model suggested that al pattern, duration of stay, and past visitation experience at a previous destination significantly influence decision making in the first stage: whether to move onto a subsequent tourist destination or return home. In the was shown to be associated with the distance from the ir origin /residence to the previous destination. Then, in the third stage, destination attributes like the number of a ttraction s the competition destination effect, the distance from a previous destination, and the distance between the subsequent destination s and their place of residence, were found to play significant roles in cho osing a subsequent destination.
135 For the supply side analysis, a set of factors w ere identified in Chapter 3 to explain the rate of regional tourism growth across 342 cities in China over the period 2002 to 2010. Local economic growth is the most important factor stimulating both inbound and dome stic tourism growth Other significant factors include d localization economies, tourist resource endowments, and hotel infrastructure. More important ly the significant degree of spatial dependence in the model suggested the presence of significant spat ial spillover effects and regional catch up effects. The geographically weighted framework adopted in this work provide d a reasonable way to obtain a map of parameter values that account for the spatial heterogeneity of tourism growth and help explain geographic variations in local tourism growth patterns. The modeling effort suggested that the spillover effect of inbound tourism growth was found to be larger in the Northwest and the North China area, wh ereas the spillover effect of domestic tourism gro wth exhibits a clear pattern of a south north division. Differences in tourism spillover patterns suggest that inbound versus domestic tourism are affected by two different and distinct spatial processes. To analyze the economic impact of tourism a factor analysis was employed in Chapter 4 with the extract ion and identification of a small number of industrial clusters according to their linkages to the tourism industry The calibrated total forw ard and backward linkage indices suggested that tourism is we akly backward linked to other industries in nearly all of thirty provinces. S everal regression models were also used to explain the variation of regional economic multipliers of tourism across the thirty provinces. The results of the regression ana lysis suggest that, in general, the profit of touris t related firms per employee is positively associa ted with the outcome multiplier.
136 At the same time the results of the latent class regression model suggested heterogeneity in multiplier models and reve aled profits are positively associated with both the outcome multiplier and the type I income multiplier for provinces that fall into in specific latent classes From a policy perspective, several implications can be drawn from this research F irst, the demand side analysis for multi destination tour ist flows suggests that tourist destinations can enable or facilitate flows. In short, tourist destinations can formulate marketing strategies aimed at exposing agglomeration economies. For example, they can offer or advertise the potential bundling of proximate and/or competing destinations in a given geographic area to attract more tourists Second, the supply side analysis of tourism growth su ggests that tourist destinations should be able to take full advantage of positive spatial spillover effects in tourism growth and internalize these benefits through collaborative marketing efforts with nearby cities. T he revealed spatial heterogeneity and distinction between inbound tourism versus domestic tourism enables local policymakers to understand that localized pattern s of inbound and domestic tourism growth are driven by different forces and factors. As such, propose d strategic plan s aimed at boos ting local tourism growth and development should be sensitized to those differences Finally, the results of the economic impact analysis imply that improving the profitability of touris t related enterprises is an important endeavor towards magnifying the economic impact of the tourism industry at large As shown in Figure 1 1, there are still several potential fields for further research to help understan d tourism from a spatial perspective. First, the demand side and supply side factors may act simultaneo usly to influence touris t flows. A well known and
137 traditional analytic tool, the spatial interaction model, is able to take both origin and destination specific attributes into consideration and scrutinize spatial configuration effects, like the destinat ion competition effect and intervening opportunities. Second, there are various impacts that tourist flows can generate apart from the economic impacts Therefore it is necessary to examine the tourist associated social, cultural, and environmental impacts Finally, it should recognized that impacts have feedbacks on both demand side and supply side factors F eedback could be included in a modeling framework which integrates and analyzes the economic, social, cultural, and environmental spillover ef fects and impacts of tourism, while differentiating between feedbacks for inbound versus domestic touris m The development of such a framework can aid in policy formulation. For example, t o maximize the potential economic multiplier s or to mitigate the det rimental environmental impacts, local tourism authorities might allocate different resources and /or enhance attractions to support tourism on a need specific basis; thereby, change the mix of supply side factors in accord ance with a targeted policy outcome Any tangible impacts in the destination s ; and hence adjust the expectation or perception of future tourists from the demand side. Incorporating supply and demand side considerations in a tourism impact model that is spatial sensitized, includes heterogeneity and assesses the economic, social, cultural, and environmental spillovers and feedback can have tremendous benefits for both policymakers and planners.
138 LIST OF REFERENCES Anselin, L. (1988). Spatial Econometrics: Methods and Models Dorddrecht: Kluwer Academic Publishers. Anselin, L. (1995). Local indicators of spatial association LISA. Geographical Analysis, 27 (2), 93 115. Anselin, L., & Bao, S. (1997). Exploratory spatial data analysis lin king SpaceStat and ArcView. In M. Fisher & A. Getis (Eds.), Recent Developments in Spatial Analysis (pp. 35 59). Berlin Heidelberg NewYork: Springer. Anselin, L., & Bera, A. K. (1998). Spatial dependence in linear regression models with an introduction to spatial econometrics. In D. Giles & A. Ullah (Eds.), Handbook of Applied Economic Statistics (pp. 237 289). New York: Marcel Dekker. Archer, B. (1995). Importance of tourism for the economy of Bermuda. Annals of Tourism Research, 22 (4), 918 930. Archer, B ., & Fletcher, J. (1996). The economic impact of tourism in the Seychelles. Annals of Tourism Research, 23 (1), 32 47. Arentze, T. A., Oppewal, H., & Timmermans, H. J. P. (2005). A multipurpose shopping trip model to assess retail agglomeration effects. Jo urnal of Marketing Research, 42 (1), 109 115. Baaijens, S., & Nijkamp, P. (2000). Meta analytic methods for comparative and exploratory policy research: An application to the assessment of regional tourist multipliers. Journal of Policy Modeling, 22 (7), 82 1 858. Baaijens, S. R., Nijkamp, P., & Van Montfort, K. (1998). Explanatory meta analysis for the comparison and transfer of regional tourist Income multipliers. Regional Studies, 32 (9), 839 849. Balaguer, J., & Cantavella Jord, M. (2002). Tourism as a long run economic growth factor: The Spanish case. Applied Economics, 34 (7), 877 884. Baltas, G. (2006). Econometric models for discrete choice analysis of travel and tourism demand. Journal of Travel & Tourism Marketing, 21 (4), 25 40.
139 Baltas, G. (2007). Econometric models for discrete choice analysis of travel and tourism demand. Journal of Travel & Tourism Marketing, 21 (4), 25 40. Baum, T. (1998). Taking the exit route: Extending the tourism area life cycle model. Current Issues in Tourism, 1 (2), 167 1 75. Bernasco, W. (2010). Modeling micro level crime location choice: Application of the discrete choice framework to crime at places. Journal of Quantitative Criminology, 26 (1), 113 138. Beynon, M., Jones, C., & Munday, M. (2009). The embeddedness of tou rism related activity: A regional analysis of sectoral linkages. Urban Studies, 46 (10), 2123 2141. Bhatnagar, A., & Ghose, S. (2004). A latent class segmentation analysis of e shoppers. Journal of Business Research, 57 (7), 758 767. Bigano, A., Hamilton, J. M., Lau, M., Tol, R. S. J., & Zhou, Y. (2007). A global database of domestic and international tourist numbers at national and subnational level. International Journal of Tourism Research, 9 (3), 147 174. Brent, R. P. (1973). Algorithms for Minimization without Derivatives Englewood Cliffs, NJ: Prentice Hall. Briassoulis, H. (1991). Methodological issues: Tourism input output analysis. Annals of Tourism Research, 18 (3), 485 495. Bristow, R. S., Lieber, S. R., & Fesenmaier, D. R. (1995). The compatibil ity of recreation activities in Illinois. Geografiska Annaler. Series B, Human Geography, 77 (1), 3 15. Butler, R. (1980). The concept of a tourist area cycle of evolution: Implications for management of resources. Canadian Geographer, 24 (1), 5 12. Butler, R. (2004). The Tourism Area Life Cycle in the Twenty First Century. In A. A. Lew, C. M. Hall & A. M. Williams (Eds.), A Companion to Tourism (pp. 159 170). Malden, MA: Blackwell Pub.
140 ard linkages. Journal of Travel Research, 45 (1), 36 52. Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics using Stata College Station, TX: Stata Press. Capone, F., & Boix, R. (2008). Sources of growth and competitiveness of local tourist produc tion systems: An application to Italy (1991 2001). Annals of Regional Science, 42 (1), 209 224. Cascetta, E., Pagliara, F., & Papola, A. (2007). Alternative approaches to trip distribution modelling: A retrospective review and suggestions for combining dif ferent approaches. Papers in Regional Science, 86 (4), 597 620. Cascetta, E., & Papola, A. (2008). A trip distribution model involving spatial and dominance attributes. Computer Aided Civil and Infrastructure Engineering, 23 (2), 116 124. Chang, W. H. (200 1). Variations in multipliers and related economic ratios for recreation and tourism impact analysis. Unpublished Ph.D Dissertation, Michigan State University. Chen, P. T., & Var, T. (2010). Distribution of tourism economic impacts: A longitudinal study. International Journal of Tourism Policy, 3 (2), 91 112. Clark, A., Etil, F., Postel Vinay, F., Senik, C., & Van der Straeten, K. (2005). Heterogeneity in reported well being: Evidence from twelve European countries. The Economic Journal, 115 (502), C118 C1 32. Cole, S. (2009). A logistic tourism model: Resort cycles, globalization, and chaos. Annals of Tourism Research, 36 (4), 689 714. Cole, S. (2012). Synergy and congestion in the tourist destination life cycle. Tourism Management, 33 (5), 1128 1140. Corr ado, L., & Fingleton, B. (forthcoming). Where is the economics in spatial econometrics? Journal of Regional Science
141 Croes, R., & Vanegas, M. (2008). Cointegration and causality between tourism and poverty reduction. Journal of Travel Research, 47 (1), 94 103. Crouch, G. I. (1994a). The study of international tourism demand: A review of findings. Journal of Travel Research, 33 (1), 12 23. Crouch, G. I. (1994b). The study of international tourism demand: A survey of practice. Journal of Travel Research, 32 ( 4), 41 55. Crouch, G. I. (1995). A meta analysis of tourism demand. Annals of Tourism Research, 22 (1), 103 118. Cuaresma, J. C., Doppelhofer, G., & Feldkircher, M. (2012). The determinants of economic growth in European regions. Regional Studies 1 24. Czamanski, S., & Ablas, L. A. d. Q. (1979). Identification of industrial clusters and complexes: A comparison of methods and findings. Urban Studies, 16 (1), 61 80. da Conceio Gonalves, V. F., & Roque guas, P. M. (1997). The concept of life cycle: An a pplication to the tourist product. Journal of Travel Research, 36 (2), 12 21. 1999: An exploratory spatial data analysis. The Annals of Regional Science, 39 (1), 121 148 Daniels, M. J., Norman, W. C., & Henry, M. S. (2004). Estimating income effects of a sport tourism event. Annals of Tourism Research, 31 (1), 180 199. Debbage, K. G., & Ioannides, D. (1998). The Economic Geography of The Tourist Industry: A Supply side Analysis London and New York: Routledge. Del Bo, C. F., & Florio, M. (2012). Infrastructure and growth in a spatial framework: Evidence from the EU regions. European Planning Studies, 20 (8), 1393 1414. Douglas, N. (1997). Applying the life cycle model to Melanesia. Annals of Tourism Research, 24 (1), 1 22.
142 Drakos, K., & Kutan, A. (2003). Regional effects of terrorism on tourism in three mediterranean countries. Journal of Conflict Resolution, 47 (5), 621 641. Dredge, D. (1999). Destination place planning and design. Annals of Tourism Research, 26 (4), 772 791. Dritsakis, N. (2004). Tourism as a long run economic growth factor: An empirical investigation for Greece using a causality analysis. Tourism Econo mics, 10 (3), 305 316. Durbarry, R. (2004). Tourism and economic growth: The case of Mauritius. Tourism Economics, 10 (4), 389 401. Elhorst, J. P. (2010). Applied spatial econometrics: Raising the bar. Spatial Economic Analysis, 5 (1), 9 28. Elhorst, J. P. & Oosterhaven, J. (2006). Forecasting the impact of transport improvements on commuting and residential choice. Journal of Geographical Systems, 8 (1), 39 59. Ellerbrock, M. J., & Hite, J. C. (1980). Factors affecting regional employment in tourism in th e United States. Journal of Travel Research, 18 (3), 26 32. Eugenio Martin, J. L. (2003). Modelling determinants of tourism demand as a five stage process:A discrete choice methodological approach. Tourism & Hospitality Research, 4 (4), 341 354. Eugenio Ma rtn, J. L., Martn Morales, N., & Scarpa, R. (2004). Tourism and economic growth in Latin American countries: A panel data approach : FEEM Working Paper No. 26.2004. Fesenmaier, D. R., & Lieber, S. R. (1985). Spatial structure and behavior response in out door recreation participaton. Geografiska Annaler. Series B, Human Geography, 67 (2), 131 138. Fik, T. J. (1988). Hierarchical interaction: The modeling of a competing central place system. The Annals of Regional Science, 22 (2), 48 69.
143 Fik, T. J., Amey, R. G., & Mulligan, G. F. (1992). Labor migration amongst hierarchically competing and intervening origins and destinations. Environment and Planning A, 24 (9), 1271 1290. Fletcher, J. E. (1989). Input output analysis and tourism impact st udies. Annals of Tourism Research, 16 (4), 514 529. Forsythe, G. E., Malcolm, M. A., & Moler, C. B. (1977). Computer Methods for Mathematical Computations : Prentice Hall Professional Technical Reference. Fotheringham, A. S. (1985). Spatial competition and agglomeration in urban modelling. Environment and planning. A, 17 (2), 213 230. Fotheringham, A. S. (1988). Consumer store choice and choice set definition. Marketing Science, 7 (3), 299 310. Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geogr aphically Weighted Regression: The Analysis of Spatially Varying Relationships Chichester: John Wiley & Sons. Frechtling, D. C. (2010). The tourism satellite account: A primer. Annals of Tourism Research, 37 (1), 136 153. Frechtling, D. C., & Horvth, E. (1999). Estimating the multiplier effects of tourism expenditures on a local economy through a regional input output model. Journal of Travel Research, 37 (4), 324 332. Frenken, K., Van Oort, F., & Verburg, T. (2007). Related variety, unrelated variety an d regional economic growth. Regional Studies, 41 (5), 685 697. Garay, L., & Cnoves, G. (2011). Life cycles, stages and tourism history: The Catalonia (Spain) Experience. Annals of Tourism Research, 38 (2), 651 671. Gasparino, U., Bellini, E., Del Corpo, B., & Malizia, W. (2009). Measuring impact of tourism on urban economies: A review of literature. International Journal of Leisure and Tourism Marketing, 1 (2), 103 130.
144 Getz, D. (1992). Tourism planning and destinati on life cycle. Annals of Tourism Research, 19 (4), 752 770. Gooroochurn, N., & Hanley, A. (2005). Spillover effects in long haul visitors between two regions. Regional Studies, 39 (6), 727 738. Greene, W. H. (2007). Econometric Analysis (6th ed.). Upper Sa ddle River, NJ: Prentice Hall. Gunduz, L., & Hatemi J, A. (2005). Is the tourism led growth hypothesis valid for Turkey? Applied Economics Letters, 12 (8), 499 504. Hanink, D. M., & Stutts, M. (2002). Spatial demand for national battlefield parks. Annals of Tourism Research, 29 (3), 707 719. Annals of the Association of American Geographers, 94 (4), 715 722. Hefner, F. L. (1990). Using economic models to measure the impact of sp orts on local economies. Journal of Sport & Social Issues, 14 (1), 1 13. Heng, T. M., & Low, L. (1990). Economic impact of tourism in Singapore. Annals of Tourism Research, 17 (2), 246 269. Henry, E. W., & Deane, B. (1997). The contribution of tourism to t he economy of Ireland in 1990 and 1995. Tourism Management, 18 (8), 535 553. Hensher, D. A., Greene, W. H., & Rose, J. M. (2005). Applied choice analysis : A primer Cambridge: Cambridge Univ. Press. English Hewings, G. (1985). Regional Input Output Analys is Beverly Hills: Sage Publications. Hohl, A. E., & Tisdell, C. A. (1995). Peripheral tourism: Development and management. Annals of Tourism Research, 22 (3), 517 534. Hunt, M. A., & Crompton, J. L. (2008). Investigating attraction compatibility in an Ea st Texas city. International Journal of Tourism Research, 10 (3), 237 246.
145 Huse, M., Gustavsen, T., & Almedal, S. (1998). Tourism impact comparisons among Norwegian towns. Annals of Tourism Research, 25 (3), 721 738. Huybers, T. (2003a). Domestic tourism d estination choices A choice modelling analysis. International Journal of Tourism Research, 5 (6), 445 459. Huybers, T. (2003b). Modelling short break holiday destination choices. Tourism Economics, 9 (4), 389 405. Huybers, T. (2005). Destination choice m odelling: what's in a name? Tourism Economics, 11 (3), 329 350. Hwang, Y. H., & Fesenmaier, D. R. (2003). Multidestination pleasure travel patterns: Empirical evidence from the American Travel Survey. Journal of Travel Research, 42 (2), 166 171. Hwang, Y. H., Gretzel, U., & Fesenmaier, D. R. (2006). Multicity trip patterns: Tourists to the United States. Annals of Tourism Research, 33 (4), 1057 1078. Isard, W., Azis, I. J., Drennan, M. P., Miller, R. E., Saltzman, S., & Thorbecke, E. (1998). Metho ds of Interregional and Regional Analysis Aldershot, Hants, England: Ashgate. Jeng, J. M., & Fesenmaier, D. R. (1998). Destination compatibility in multidestination pleasure travel. Tourism Analysis, 3 (2), 77 87. Karplus, Y., & Krakover, S. (2005). Stochastic multivariable approach to modelling tourism area life cycles. Tourism and Hospitality Research, 5 (3), 235 254. Khadaroo, J., & Seetanah, B. (2007). Transport infrastructure and tourism development. Annals of Tourism Resear ch, 34 (4), 1021 1032. Khadaroo, J., & Seetanah, B. (2008). The role of transport infrastructure in international tourism development: A gravity model approach. Tourism Management, 29 (5), 831 840 Kim, S. i., & Fesenmaier, D. R. (1990). Evaluating spatial structure effects in recreational travel. Leisure Sciences, 12 (4), 67 81.
146 Klijs, J., Heijman, W., Maris, D. K., & Bryon, J. (2012). Criteria for comparing economic impact models of tourism. Tourism Economics, 18 (6), 1175 1202. Koo, T. T. R., Wu, C. L., & Dwyer, L. (2012). Dispersal of visitors within destinations: Descriptive measures and underlying drivers. Tourism Management, 33 (5), 1209 1219. Krakover, S. (2004). Tourism development centres versus peripheries: The Israeli experience during the 1990s International Journal of Tourism Research, 6 (2), 97 111. Kweka, J., Morrissey, O., & Blake, A. (2003). The economic potential of tourism in Tanzania. Journal of International Development, 15 (3), 335 351. Lazzeretti, L., & Capone, F. (2009). Spatial spi llovers and employment dynamics in local tourist systems in Italy (1991 2001). European Planning Studies, 17 (11), 1665 1683. Le Gallo, J., & Ertur, C. (2003). Exploratory spatial data analysis of the distribution of regional per capita GDP in Europe, 1980 1995. Papers in Regional Science, 82 (2), 175 201. Lee, C. C., & Chang, C. P. (2008). Tourism development and economic growth: A closer look at panels. Tourism Management, 29 (1), 180 192. Lee, C. K. (1996). Input output analysis and income distribution p atterns of the tourism industry in South Korea. Asia Pacific Journal of Tourism Research, 1 (1), 35 49. Leeflang, P. S. H., Wittink, D. R., Wedel, M., & Naert, P. A. (2000). Building Models for Marketing Decisions Boston: Kluwer Academic. LeSage, J. P., & Fischer, M. M. (2008). Spatial growth regressions: Model specification, estimation and interpretation. Spatial Economic Analysis, 3 (3), 275 304. LeSage, J. P., & Pace, R. K. (2009). Introduction to Spatial Econometrics Boca Raton, FL: CRC Press.
147 Leung Y., Mei, C. L., & Zhang, W. X. (2000). Statistical tests for spatial nonstationarity based on the geographically weighted regression model. Environment and Planning A, 32 (1), 9 32. Lew, A., & McKercher, B. (2006). Modeling tourist movements: A local des tination analysis. Annals of Tourism Research, 33 (2), 403 423. Lin, C. H., & Morais, D. B. (2008). The spatial clustering effect of destination distribution on cognitive distance estimates and its impact on tourists' destination choices. Journal of Travel & Tourism Marketing, 25 (3 4), 382 397. Liu, J., & Var, T. (1983). The economic impact of tourism in Metropolitan Victoria, B.C. Journal of Travel Research, 22 (2), 8 15. Liu, J. C. (1986). Relative economic contributions of visitor groups in Hawaii. Jour nal of Travel Research, 25 (1), 2 9. Lloyd, C. D. (2011). Local Models for Spatial Analysis (2nd ed.). Boca Raton, Fla: CRC Press. Louis J, D. A. (1988). Tourism a vital force for peace. Tourism Management, 9 (2), 151 154. Loviscek, A. (1982). Industrial cluster analysis backward or forward linkages? The Annals of Regional Science, 16 (3), 36 47. Lozano, J., mez, C. M., & Rey Maquieira, J. (2008). The TALC hypothesis and economic growth theory. Tourism Economics, 14 (4), 727 749. Lue, C. C., Crompton, J. L., & Fesenmaier, D. R. (1993). Conceptualization of multi destination pleasure trips. Annals of Tourism Research, 20 (2), 289 301. Lue, C. C., Crompton, J. L., & Stewart, W. P. (1996). Evidence of cumulative attraction in multidestination recreational trip decisions. Journal of Travel Research, 35 (1), 41 49. Lundtorp, S., & Wanhill, S. (2001). The resort lifecycle theory: Generating processes and estimation. Anna ls of Tourism Research, 28 (4), 947 964.
148 Luo, H., & Yang, Y. (2012). Spatial pattern of hotel distribution in China Disscussion Paper, School of Business, Sun Yat sen University, Guangzhou. Mannell, R. C., & Iso Ahola, S. E. (1987). Psychological nature o f leisure and tourism experience. Annals of Tourism Research, 14 (3), 314 331. Mbaiwa, J. E. (2005). Enclave tourism and its socio economic impacts in the Okavango Delta, Botswana. Tourism Management, 26 (2), 157 172. McGregor, P. G., Swales, J. K., & Yin, Y. P. (1996). A long run interpretation of regional input output analysis. Journal of Regional Science, 36 (3), 479 501. McKercher, B., & Wong, D. Y. Y. (2004). Understanding tourism behavior: Examining the combined effects of prior visitation history and destination status. Journal of Travel Research, 43 (2), 171 179. Medlik, S. (2003). Dictionary of Travel, Tourism and Hospitality (3rd ed.). Boston: Butterworth Heinemann. Midmore, P., Munday, M., & Roberts, A. (2006). Assessing industry linkages using r egional input output tables. Regional Studies, 40 (3), 329 343. Miller, R. E., & Blair, P. D. (1985). Input Output Analysis: Foundations and Extensions Englewood Cliffs, N.J: Prentice Hall. Mings, R. C., & Mchugh, K. E. (1992). The spatial configuration of travel to Yellowstone National Park. Journal of Travel Research, 30 (4), 38 46. Mitchell, L. S., & Murphy, P. E. (1991). Geography and tourism. Annals of Tourism Research, 18 (1), 57 70. Moore, W., & Whitehall, P. (2005). The tourism area lifecycle and regime switching models. Annals of Tourism Research, 32 (1), 112 126. Narayan, P. K. (2004). Economic impact of tourism on Fijis economy: empirical evidence from the computable general equilibrium model. Tourism Economics, 10 (4), 419 433.
149 Nepal, S. K. (20 09). Traditions and trends: A review of geographical scholarship in tourism. Tourism Geographies, 11 (1), 2 22. Neumayer, E. (2004). The impact of political violence on tourism: Dynamic cross national estimation Journal of Conflict Resolution, 48 (2), 259 2 81. Newman, J., & Bernardin, V., Jr. (2010). Hierarchical ordering of nests in a joint mode and destination choice model. Transportation, 37 (4), 677 688. Nicolau, J. L., & Ms, F. J. (2005). Stochastic modeling: A three stage tourist choice process. Anna ls of Tourism Research, 32 (1), 49 69. Nicolau, J. L., & Ms, F. J. (2006). The influence of distance and prices on the choice of tourist destinations: The moderating role of motivations. Tourism Management, 27 (5), 982 996. Nicolau, J. L., & Ms, F. J. (2008). Sequential choice behavior: Going on vacation and type of destination. Tourism Management, 29 (5), 1023 1034. hUallachin, B. (1984). The identification of industrial complexes. Annals of the Association of American Geographers, 74 (3) 420 436. hUallachin, B. (1985). Complementary linkages and the hierarchical structure of regional economies. Geographical Analysis, 17 (2), 130 142. Oh, C. O. (2005). The contribution of tourism development to economic growth in the Korean economy. T ourism Management, 26 (1), 39 44. Oksanen, E. H., & Williams, J. R. (1992). An alternative factor analytic approach to aggregation of input output tables. Economic Systems Research, 4 (3), 245 256. Oppermann, M. (1993). Tourism space in developing countrie s. Annals of Tourism Research, 20 (3), 535 556. Ord, J. K., & Getis, A. (1995). Local spatial autocorrelation statistics: Distributional issues and an application. Geographical Analysis, 27 (4), 286 306.
150 Pez, A., Uchida, T., & Miyamoto, K. (2002). A gener al framework for estimation and inference of geographically weighted regression models: 2. Spatial association and model specification tests. Environment and Planning A, 34 (5), 883 904. Papatheodorou, A. (2004). Exploring the evolution of tourism resorts. Annals of Tourism Research, 31 (1), 219 237. Pearce, D. G. (1995). Tourism Today: A Geographical Analysis (2nd ed.). Harlow, Essex, England: Longman Scientific & Technical. Pellegrini, P. A., & Fotheringham, A. S. (1999). Intermetropolitan migration and hierarchical destination choice: A disaggregate analysis from the US Public Use Microdata Samples. Environment and Planning A, 31 (6), 1093 1118. Pellegrini, P. A., & Fotheringham, A. S. (2002). Modelling spatial choice: a review and synthesis in a migration context. Progress in Human Geography, 26 (4), 487 510. Polo, C., & Valle, E. (2009). Estimating tourism impacts using input output and SAM models in the Balearic islands. In . Matias, P. Nijkamp & M. Sarmento (Eds.), Advances in Tourism Economics (pp. 121 143): Physica Verlag HD. Pratt, S. (2011). Economic linkages and impacts across the TALC. Annals of Tourism Research, 38 (2), 630 650. Roepke, H., Adams, D., & Wiseman, R. (1974). A n ew approach to the identification of industrial complexes using input output data. Journal of Regional Science, 14 (1), 15 29. Ryan, C. (2003). Recreational Tourism: Demand and Impacts Buffalo, N.Y. : Multilingual Matters. Salman, A. K., Shukur, G., & vo n Bergmann Winberg, M. L. (2007). Comparison of econometric modelling of demand for domestic and international tourism: Swedish data. Current Issues in Tourism, 10 (4), 323 342. Santos, G. E. d. O., Ramos, V., & Rey Maquieira, J. (2011). A microeconomic mo del of multidestination tourism trips. Tourism Economics, 17 (3), 509 529.
151 Santos, G. E. d. O., Ramos, V., & Rey Maquieira, J. (2012). Determinants of multi destination tourism trips in Brazil. Tourism Economics, 18 (6), 1331 1349. Seddighi, H. R., & Theoc harous, A. L. (2002). A model of tourism destination choice: a theoretical and empirical analysis. Tourism Management, 23 (5), 475 487. Smith, S. L. J. (1983). Recreation Geography New York: Longman. Song, H., Dwyer, L., Li, G., & Cao, Z. (2012). Tourism economics research: A review and assessment. Annals of Tourism Research, 39 (3), 1653 1682. Song, H., & Li, G. (2008). Tourism demand modelling and forecasting A review of recent research. Tourism Management, 29 (2), 203 220. Song, H., & Witt, S. F. (2000). Tourism Demand Modelling and Forecasting: Modern Econometric Approaches Oxford: Pergamon. Song, H., Witt, S. F., & Li, G. (2003). Modelling and forecasting the demand for Thai tourism. Tourism Economics, 9 (4), 363 387. Stewart, S. I (1997). Multi destination trip patterns. Annals of Tourism Research, 24 (2), 458 461. Stoeckl, N. (2007). Using surveys of business expenditure to draw inferences about the size of regional multipliers: A case study of tourism in northern Australia. Regi onal Studies, 41 (7), 917 931. Streit, M. E. (1969). Spatial associations and economic linkages between industries. Journal of Regional Science, 9 (2), 177 188. Sugiyarto, G., Blake, A., & Sinclair, M. T. (2003). Tourism and globalization: Economic Impact in Indonesia. Annals of Tourism Research, 30 (3), 683 701. Sun, Y. Y. (2005). Marginal economic impacts of inbound tourism to Taiwan and policy evaluation. Asia Pacific Journal of Tourism Research, 10 (3), 309 327. Tideswell, C., & Faulkner, B. (1999). Mul tidestination travel patterns of international visitors to Queensland. Journal of Travel Research, 37 (4), 364 374.
152 Tideswell, C., & Faulkner, B. (2003). Identifying antecedent factors to the travelers pursuit of a multidestination travel itinerary. Touris m Analysis, 7 (3 4), 177 190. Tyrrell, T. J., & Johnston, R. J. (2006). The economic impacts of tourism: A special issue. Journal of Travel Research, 45 (1), 3 7. van Leeuwen, E. S., Nijkamp, P., & Rietveld, P. (2006). Economic impacts of tourism: A meta a nalytic comparison of regional output multipliers. In G. M & N. P (Eds.), Tourism and Regional Development: New pathways (pp. 115 132). Aldershot: Ashgate. van Leeuwen, E. S., Nijkamp, P., & Rietveld, P. (2009). A meta analytic comparison of regional outpu t multipliers at different spatial levels: Economic impacts of tourism. In . Matias, P. Nijkamp & M. Sarmento (Eds.), Advances in Tourism Economics: New Developments (pp. 13 33): Physica Verlag HD. Vaughan, D. R., Farr, H., & Slee, R. W. (2000). Estimatin g and interpreting the local economic benefits of visitor spending: An explanation. Leisure Studies, 19 (2), 95 118. Weidenfeld, A., Butler, R., & Williams, A. W. (2010a). Clustering and compatibility between tourism attractions. International Journal of Tourism Research, 12 (1), 1 16. Weidenfeld, A., Butler, R., & Williams, A. W. (2010b). The role of clustering, cooperation and complementarities in the visitor attraction sector. Current Issues in Tourism, 14 (7), 595 629. West, G., & Gamage, A. (2001). Ma cro effects of tourism in Victoria, Australia: A nonlinear input output approach. Journal of Travel Research, 40 (1), 101 109. West, G. R. (1995). Comparison of input output, input output econometric and computable general equilibrium impact models at the regional level. Economic Systems Research, 7 (2), 209 227. Wu, B., Zhu, H., & Xu, X. (2000). Trends in China's domestic tourism development at the turn of the century. International Journal of Contemporary Hospitality Management, 12 (5), 296 299.
153 Wu, C. L. & Carson, D. (2008). Spatial and temporal tourist dispersal analysis in multiple destination travel. Journal of Travel Research, 46 (3), 311 317. of destination and t ravel party with an integrated latent class and nested logit model. Tourism Management, 32 (6), 1407 1413. Wu, L., Zhang, J., & Fujiwara, A. (2011b). A tourist's multi destination choice model with future dependency. Asia Pacific Journal of Tourism Researc h, 17 (2), 121 132. 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 (6), 827 837. Yang, Y., Wong, K., & Zhang, J. (2011). Determinants of length of stay for domestic tourists: Case study of Yixing. Asia Pacific Journal of Tourism Research, 16 (6), 619 633. Yang, Y., & Wong, K. K. F. (2012a). Spatial distribution of tourist flows to China's cities. Tourism G eographies 10.1080/14616688.14612012.14675511. Yang, Y., & Wong, K. K. F. (2012b). A spatial econometric approach to model spillover effects in tourism flows. Journal of Travel Research, 51 (6), 768 778. Zhou, D., Yanagida, J. F., Chakravorty, U., & Leun g, P. (1997). Estimating economic impacts from tourism. Annals of Tourism Research, 24 (1), 76 89. Zurick, D. N. (1992). Adventure travel and sustainable tourism in the peripheral economy of Nepal. Annals of the Association of American Geographers, 82 (4), 608 628.
154 BIOGRAPHICAL SKETCH Yang Yang was born in 1985 in the city of Nanchang, China, where he also grew up. He earned his Bachelor of Science degree in resource management from Nanjing University in 2006 and Master of Philosophy degree in hotel and tourism management from the Hong Kong Polytechnic University in 2009. In 2009, Yang entered the Ph.D program in Department of Geography at University of Florida. He received his Ph.D. from the University of Florida in the summer of 201 3. His research interests focused on tourism and recreational geography, as well as economic analytics in hospitality and tourism management. Upon graduation, he will join the School of Tourism and Hospitality Management at the Temple University as an assistant professor