Spatial effects of tourism development on economic resilience: an empirical study of Wenchuan earthquake based on dynamic spatial Durbin model

This study highlights the importance of the effects of spatial tourism. Using panel data of 135 Wenchuan earthquake-affected counties from 2008 to 2018, this study employs the dynamic spatial Durbin model to examine the spatial effects of tourism development on post-disaster economic resilience and compares the differences in the tourism-growth nexus between severe disaster-affected counties and general disaster-affected counties. The empirical results show that tourism development contributes to economic resilience for general disaster-affected counties, which supports the tourism-led growth hypothesis, whereas there is an inverted U-shaped relationship for severely disaster-affected counties. Furthermore, the spatial spillover effects of tourism are insignificant. Plausible explanations for these results are discussed, and policy suggestions are provided.


Introduction
On May 12, 2008, the Wenchuan earthquake of 8.0 on the Richter scale led to the widespread devastation of Western China, covering an area of 0.5 million square Kilometers and over 417 counties in ten provinces. Officials confirmed 69,227 deaths, 17,923 missing people, and 20 million homeless people in this earthquake. Numerous buildings, highways, water supplies, sewage, gas, and power systems were destroyed (Yang et al. 2011). The direct economic losses were estimated to be approximately 84.5 billion Yuan. Meanwhile, this earthquake resulted in long-term adverse consequences for economic activities, such as industrial chain rupture, economic recession, and unemployment increase in disasteraffected areas.
Sichuan Province was the worst hit province, located at the epicenter of the Wenchuan earthquake. Of all the counties in Sichuan Province, over two-thirds are affected, including 39 severe disaster-affected counties (ten extremely severe counties and 29 relatively severe counties) and 100 general disaster-affected counties (see Fig. 1). The economies of all earthquake-affected counties suffered varying degrees of loss. This earthquake tremendously affected all industries in quake-hit counties in a short period of time, with no exception of local tourism (Yang et al. 2008). In particular for tourism-based areas, GDP and tourism revenue have decreased by over 10% and 50%, respectively (Cheng and Zhang 2020a, b).
After the Wenchuan earthquake, the central government, provincial, and local governments focused on recovering from these losses. A series of policies, measurements, and programs have been conducted to invigorate the physical, social, economic, political, and ecological recovery and livelihood resilience of disaster-hit areas (Cheng and Zhang 2020a, b;Liu et al. 2018), such as the Regulations on post-Wenchuan earthquake Recovery and Reconstruction (2008) and the Reconstruction Plan for the Post-Wenchuan Earthquake (2008). Regarding economic resilience, an increasing number of county governments have made the tourism industry a source of stimulating economic resilience. In particular for severely disaster-affected counties, tourism was widely considered the leading industry to boost the recovery of these relevant disaster-affected industries and has gained great success (Cheng and Zhang 2020a, b). Through a three-year effort, the tourism industry and the Fig. 1 Location of 135 disaster-affected counties in Sichuan Province a . According to the announcement of the central government of China, there are 139 Wenchuan earthquake-affected counties in Sichuan Province. However, four counties of Pengzhou (彭州), Hanyuan (汉源), Yuexi (越西) and Zhongjiang (中江) lack of the data of tourism revenue. In constructing the weight matrix of severe disaster-affected counties, Shimian is not adjacent to other counties in the geography, which were removed automatically when conducting the software of Geoda and Stata 15.1. Therefore, we select the rest of 135 counties as the study object. Source http:// www. gov. cn/ zwgk/ 2008-07/ 22/ conte nt_ 10528 35. htm overall economy in all the disaster-affected counties developed rapidly, and their economic outputs far exceeded the pre-disaster levels. However, some academic skeptics have also appeared, that is, the tourism-led growth mode in these areas is not sustainable and can be detrimental to economic growth in the long term (Zhang and Cheng 2019;Wang and Xie 2020). Therefore, it is important to investigate the effect of tourism on economic resilience and probe whether tourism can promote the economic resilience of disaster-affected areas after the Wenchuan earthquake.
Some studies on the relationship between tourism and economic resilience after natural hazards have proven that tourism significantly contributes to economic resilience (Mazzocchi and Montini 2001;Zhang and Cheng 2019;Zhang 2020a, 2020b). However, there are still some drawbacks to the existing research. First, the capacity of an area to recover from a natural disaster such as an earthquake could also depend on the real destruction caused by the tragic event. Probably adverse effects in terms of destruction differed among the areas considered. This could trigger differences in the recovery capacities of the selected areas, which may further lead to differences in the impact of tourism on economic resilience. For instance, Cheng and Zhang (2020a, b) take 35 severely disaster-affected counties as case studies and find that the impact of tourism on post-earthquake economic resilience can be different for counties with different resilience levels. Since most researchers have concentrated primarily on a single area or region at the same level of destruction (Huang and Min 2002), comparing the results across the same study areas is challenging. In light of this, it is important to compare the differences in the tourism-growth nexus across counties at different destruction levels, which has often been overlooked in tourism economic studies. Second, geographic features, resource endowments, infrastructure, and economic conditions differ across areas (Liu et al. 2021;Lee et al. 2021), which may lead to the unbalanced spatial development of tourism development and economic resilience. According to the first law of geography, near things are more related than distant things (McKercher et al. 2008). The spatial effect is an important factor that influences the transfer of tourism development into economic growth. Traditional panel methods may lead to biased results because the spatial correlation is ignored. Thus, it is necessary to capture the spatial effects of tourism on economic resilience using a spatial econometric approach. However, studies to date on how tourism development affects economic resilience from a spatial effect perspective are limited and need to be further investigated.
To bridge the aforementioned gaps: this study aims to investigate the effect of tourism on the economic resilience of counties affected by the Wenchuan earthquake from a dynamic spatial perspective. This study contributes to the literature in the following ways: First, most empirical studies on the tourism-growth nexus ignore spatial effects. Various frontier spatial econometric techniques, including the Global Moran's I index and Dynamic spatial Durbin model (DSDM), are employed to analyze the spatial effect of tourism. Second, because few comparative studies have focused on the study areas with different destruction levels, this study considers 135 Wenchuan earthquake-affected counties as the study object. To the best of our knowledge, it is the first to compare the differences in the tourism-growth nexus between two groups of study areas: severe disaster-affected counties and general disaster-affected counties. These interesting findings that tourism can boost economic resilience for general disaster-affected counties, whereas there is an inverted U-shaped relationship between tourism and economic resilience for severely disasteraffected counties, is an important supplement to the existing study on the tourism-growth nexus in the tourism disaster field.
The following three questions are addressed: (1) Can tourism development facilitate economic resilience? (2) Does the tourism-growth nexus vary across counties at different destruction levels? If so, what drives them? (3) Does tourism have a spatial spillover effect on surrounding counties' economic resilience? Answering these questions will help us further understand the prominent role of tourism in economic resilience and development processes and provide theoretical guidance for policymakers in modeling effective policies for restoring the economy in other disaster-hit destinations, thereby making tourism an important catalyst for promoting sustainable economic development.
The remainder of this paper is organized as follows. Section 2 presents the literature review. The methods and data are described in Sect. 3. Section 4 presents the empirical results of the DSDM. This is followed by a discussion and conclusions in Sect. 5.

Tourism and economic resilience
Economic resilience is a critical part of the post-disaster resilience. It refers to the ability and speed of economic sectors to recover from a disaster shock to the pre-disaster condition (Osth et al. 2015;Rose and Krausmann 2013). The economic resilience is a dynamic, complex, and multi-dimensional project, which emphasizes the synergistical recovery and development process of multiple economic sectors, rather than focus on the one industry or economic approach. Cheng and Zhang (2020a, b) suggest that the economic resilience is not confined to industrial-economic recovery alone, but encompasses also socio-economic resilience: residents' living condition, societal wellbeing and employment. As the Martin and Sunley (2015) state, economic resilience is the capacity of economy system to withstand or recover from disaster shocks to its previous developmental path, or transition to a new equilibrium state. From the time dimension, economic resilience takes a long time to recover economic sectors to the pre-disaster level due to its complex attribute. According to a recent estimate by WTTC (2019), the recovery time from different natural disasters varies greatly from 1 to 93 months.
In the sphere of tourism, tourist destination is vulnerable to the mega natural disaster. Unlike other general region, tourist destination is not able to bounce back and enter the normal track in a short time because its image may be tarnished by the natural hazards, which gives unsafe sense to tourists and negatively influences their travel decisions to destination (Cassedy 1992). Tourist destination recovery is complex and includes a series of recovery process such as a return to a feeling of tourist safety and normality, clearance and repair of physical infrastructure and the recovery of tourist numbers and hotel bookings (Khazai et al. 2018). Ritchie (2009) highlights that post-disaster tourism recovery should not focus on an industry or economic approach. There are also social and psychological dimensions. Sharpley (2009) points out that developing post-disaster tourism should consider the social, cultural, historical, and political factors. Mair et al (2016) suggest that post-disaster tourism recovery is the development and implementation of disaster-recovery actions to bring the destination back to a pre-disaster level or even an improved state. The improved state includes renewing the tourist amenities, infrastructure and building new houses catering for tourism development in disaster-affected regions. Cellini and Cuccia (2015) define the tourism economic resilience as the different reactions of tourist carrying capacity to tourism flows' shock. Based on the preceding discussion, we suggest that post-disaster tourism recovery should focus on the recovery and development of tourism industry to bring the economy, physical facilities, ecology environment and culture of disaster-affected destination back to a pre-disaster condition or an improved state.
With the increasing contribution of tourism to national economy, tourism was increasingly seen as a strategic pathway for invigorating the economy of many disaster-affected areas and gained great economic return (Cheng and Zhang 2020a, b). According to Bellini et al. (2017), the inextricably links between tourism and economic resilience lie in three aspects: tourism growth dynamics, tourism resilience and linkages of tourism with other economic sectors. Firstly, the rapid growth of tourism has become the most significant feature of contemporary economy in the world. Tourism industry promises to sustain the dynamic of economic growth and compensates the declining contribution of other economic sectors. Secondly, tourism is itself an industry susceptible to natural disasters and faces the resilience challenge. Lastly, the tourism industry is a large, complex sector closely linking with such economic sectors as agriculture, manufacturing, construction, retail, transportation and entertainment. Therefore, tourism recovery needs support from these industries. Meanwhile, tourism has contributed to economic resilience through various channels including foreign exchange transactions, job opportunities, and government tax revenues (Dogru and Bulut 2018). Particularly, in a tourism-based region, post-disaster resilience is intimately tied to tourism industry. This is because tourism represents a pillar one and a major economic contributor in the destination (Filimonau and Delysia 2020). Tourism resilience can also be hardly separated from the whole economic recovery process (Cheng and Zhang 2020a).

Tourism-led growth hypothesis
Over the past decades, the relationship between tourism development and economic growth has been an everlasting topic in tourism economic research. The tourism-led growth hypothesis (TLGH) was first formally proposed by Balaguer and Cantavella-Jordà (2002), who examined the impact of tourism in the Spanish long-run economic development. The TLGH offers a theoretical and empirical base for the tourism-growth nexus. Thereafter, an increasing number of empirical research attempt to verify the hypothesis, and reach four different tourism-growth hypotheses: (1) Tourism-Led Growth hypothesis suggesting that tourism development stimulates economic growth in one-way (Balaguer and Cantavella-Jorda 2002; Faber and Gaubert 2019). (2) Economy-Driven Tourism Growth hypothesis posits a positive causality from economic growth to tourism development (Ahiawodzi 2013;Oh 2005). Various economic theories provide some theoretical explanations for the tourismled growth hypothesis. TLGH was firstly originated from the export-led growth hypothesis (ELGH) that theoretically suggests that economic growth can be enhanced not only by increasing the amount of labor and capital within an economy, but also by expanding exports (Adamou and Clerides 2009;Brida et al. 2016). Lanza and Pigliaru (1999) constructed the Lucastype two-sector theoretical model which was based on the Lucas's two-sector endogenous growth model, providing a good theoretical basis for the TLGH. As clarified by the Lucastype two-sector model, small countries endowed with rich natural resources experience higher tourism specialization level and achieve higher growth 1 3 rates. Tang and Tan (2015) apply the Solow's (1956) growth framework derived from the Cobb-Douglas production function to verify the TLGH.
An increasing number of studies focus on the test for TLGH for general tourist destinations, while the impact of tourism on post-economic resilience for the disaster-affected destinations was still empirically under-researched. Qazi and Rana (2013) confirmed that tourism insignificantly spurred economic recovery in Pakistan during the years when terrorist attacks were frequent. Tang and Abosedra (2014) provided strong evidence that tourism exerted a positive impact on Lebanon's economic development for during the period of war. Dogru and Bulut (2018) took European countries as case study and revealed evidence in favor of the feedback effect between tourism and economic recovery. Zhang and Cheng (2019) revealed that tourism had various impacts on post-disaster economic growth under the different levels of socioeconomic variables for 36 Wenchuan earthquake-hit counties in China. Later, Cheng and Zhang (2020a, b) investigated the nonlinear impact of tourism on economic resilience and showed that the coefficients on tourism specialization varied under different levels of economic recovery. Lee et al. (2021) used a geographically weighted regression to analyze the effect of tourism on economic resilience in Florida, and found that positive mobility of tourism specialization led to greater resilience.
In terms of the methodology, one strand of literature analyzing the TLGH used various traditional economic techniques including but not limited to the Granger causality test (Solarin 2018;Tang et al. 2017), System-GMM method (Zuo and Huang 2019), VEC model (Tang and Tan 2015), ADL model (Lin and Li 2018), PTR method (Po and Huang 2008), quantile regression analysis (Cheng and Zhang 2020a, b), and panel smooth transition regression model (Li et al. 2010). However, these above methods ignore the spatial correlation and dependence of observed variables. Because of the differences in geographic factors, there are imbalance of tourism development and economic growth in geographic space. It is very important to spatial dependence and heterogeneity of tourism and economy. Moreover, due to geospatial mobility nature of tourist flow, human flow and capital flow, it may provide a possibility for tourism development to occur through spatial spillover into neighboring areas (Liu et al. 2021). Therefore, the research on the relationship between tourism development and economic growth should take spatial factor into account. The spatial econometric method is more suitable for analysis.
In the field of tourism economics, the spatial effect of tourism on economic growth has been somewhat documented (Li et al. 2020;Li and Weng 2016;Song et al. 2020;Yang and Fik 2014). Li et al. (2020), Balli and Tsui (2016), and Zhang et al. (2011) used various econometric techniques to measure the spatial spillover effect of tourist flow. Shi et al (2019) applied the spatial Durbin method to examine the inbound tourism spillover effect on urban-rural income disparity. Liu et al. (2021) used static and dynamic spatial Durbin models to analyze the effects of tourism on economic growth in China, and found that domestic tourism and inbound tourism contributed to economic growth, but the spatial spillover effect of inbound tourism was not significant.

Global Moran's I index
The global Moran's I index (GMI) was employed to examine the spatial autocorrelation between LnPGDP and LnTR which provided basic conditions for the subsequent use of the SDM (Anselin 1988). If there were no significant spatial correlation, the SDM could not be further used. The specifications of the GMI are listed: where n refers to the number of counties, x i and x j represent the values of the observed variables in county i and county j, respectively; x represents the average value of the observed variable, and W ij refers to the value of the spatial weighting matrix. Its index ranges from [− 1, 1]. The Z statistic is utilized to examine the significance of the GMI. Specifically, If -1 ≤ I < 0 and Z > 1.96, it indicates that spatial correlation is significantly negative, revealing a spatial difference distribution; if it is I = 0, it shows no spatial correlation, indicating a pattern of spatial random distribution; and if it is 0 < I ≤ 1 and Z > 1.96, the spatial correlation is significantly positive, indicating a spatial clustering pattern. The standard equation of the Z statistic lists as follows: Under a norm assumption, E(I) and Var(1) are calculated as follows: Under a random assumption, E(I) and Var(1) are calculated as follows:

Dynamic SDM
The DSDM, extending the functions of spatial Durbin model (SDM), was used to explore the dynamic spatial effect between the explanatory and dependent variables between specific and adjacent counties. It addresses the endogeneity problem, time dependence, spatial dependence, and spatial-time dependence (Elhorst 2012;Elhorst 2014). The specifications of the DSDM are written as follows (Chen et al. 2017;Feng and Wang 2020): where Y i,t refers to the variable of LnGDP, X i,t includes the key variable of LnTA and control variables, W represents the spatial weighting matrix, β is the coefficient of X i,t , denotes the coefficient of WX i,t , is the coefficient of WY i,t , i refers to the spatial-fixed (SF) effect, t refers to the temporal-fixed (TF) effect, denotes the random error vector. denotes the parameter of Y i,t−1 and denotes the parameter of the space-time lag WY i,t−1 . The formula of the DSDM is extended as follows: However, the parameter estimation in the spatial panel model only exhibits the directions and magnitudes of the explanatory variables and does not capture the marginal effect (Zhang and Xiang 2021). Therefore, this study adopted the spatial decomposition effect method to calculate the short-and long-run direct, spillover, and total effects. The direct effect estimated the impact of TR, which differed slightly from the corresponding parameter estimates. This difference arose owing to the feedback loop effects that resulted from the influences passing from adjacent counties to a specific county (Huang and Chand 2015). If the estimator of the direct effect is larger than the parameter coefficient, it indicates a positive feedback effect and vice versa. The spillover effect results from spatial dependency, so change in TR in a specific county influence the LGDP in neighboring counties. The total effect equals the total direct effect added to the spillover effect. In the time dimension, the direct and spillover effects can be classified as shortand long-run effects. The formulas for direct, spillover and total effects are as follows: Short-run direct effects: Short-run indirect effects:

Short-run total effects
Long-run direct effects: Long-run direct effects: Long-run total effects:

Spatial weight matrix
The selection of the spatial weighting matrix (W ij ) is the key to obtaining accurate results with the spatial econometric technique (Feng et al. 2019). In our study, one 0-1 adjacent weight matrix was constructed as follows.

Variable selection and data source
In this study, the logarithm of real per capital GDP (LnGDP) as the dependent variable was used to measure economic resilience (Zhang and Cheng 2019). For the key explanatory variable, in line with the previous literature, the proportion of tourism revenue to GDP (TR) was adopted as the core variable to reflect the level of tourism development (Cheng and Zhang 2020a, b;Zuo and Huang 2018). The following control variables were considered into the models based on the existing literature and the principle of data availability: (1) INVEST: the ratio of fixed asset investment to GDP, according to the neoclassical economics theory, the capital investment may be an important endogenous factor to promote economic growth (Zuo and Huang 2018).
(2) CZZC: the proportion of government fiscal expenditure in GDP as the degree of government intervention. Liu et al. (2021) suggest that government intervention behavior exerts an important role in economic activities. In addition, in the economic literature, some important control variables including population density (POPU, the ratio of total inhabitants to total area square), employment level (EMP, the proportion of total employees in total inhabitants), and road density (ROAD, the ratio of length of highways to total area square) are suggested as key determinants of economic growth ( Table 1.

Empirical results
Figure 2 depicts the changes in tourism revenue and the tourism revenue ratio to GDP (TR) for 135 disaster-affected counties in Sichuan Province in three representative years (2008, 2013, and 2018). As can be seen from Fig. 2, tourism revenue in most counties continues to increase rapidly during the entire period. Until 2018, the ratio of tourism ((1 − )I − ( + W)) −1 ( k I n + k W) rsum ((1 − )I − ( + W)) −1 ( k I n + k W) d + ((1 − )I − ( + W)) −1 ( k I n + k W) rsum W ij = 1 i and j adjacent 0 otherwise 1 3 revenue to GDP of over 100 counties accounted for more than 20% of GDP, indicating that the tourism industry plays an increasingly important role in economic growth.
In addition, there are significant regional differences in tourism revenue among counties. The high-value zones of tourism revenue spread from the southeast to northeast of the disaster-affected regions, located in the core zones of their cities or closest to the metropolis of Chengdu. However, the counties with high TR values (over 49%) are mainly distributed in the west of the disaster-affected regions, far from their cities. This indicates that tourism has become a leading industry in the local economy. Before conducting the empirical analysis, it is necessary to determine whether multicollinearity exists among the explanatory variables. The Pearson correlation test and  the VIF test were used in this study. Table 2 shows that the correlation values of all variables are low, and the Variance Inflation Factors (VIF) values are all less than two, indicating no multicollinearity in the following regression models.
To avoid "spurious regression," this study used the four-panel unit-root tests, including the LLC, IPS, Fisher-ADF, and Fisher-PP tests, to determine the sequence stationarity in their levels. According to Table 3, all the variables (LnGDP, LnTR, LnPOPU, LnEMP, LnINVEST, LnCZZC, and LnROAD) reject the null hypothesis of the stationary test in their levels, indicating that these variables are all stationary.
Further, Table 4 shows the spatial correlation results of LnGDP and LnTR calculated by global Moran's I (GMI) in the full sample, general disaster-affected counties, and severe disaster-affected counties. In the full sample, the GMIs of LnGDP and LnTR fluctuated from 2008 to 2018. They were all above 0.09 and significant at least at the 1% statistical level. In the data sample from general disaster-affected counties and severe disasteraffected counties, the fluctuations in the GMIs of LnGDP and LnTR were similar to those of the full sample. They are significantly positive at the 1% level for most years. These results show that economic resilience and tourism revenue have significantly positive spatial autocorrelations and exhibit spatial clustering patterns due to agglomeration effects among disaster-affected counties.
Finally, Table 5 summarizes the estimations of the DSDM regression in the full sample. Based on the diagnostic test results, this study focused on the spatial effect based on the DSDM with a spatial-time fixed effect. As shown in columns 4 and 8, economic growth has significant time lag and spillover effects. On the one hand, the estimation of  W*LnGDP was 0.175 and significantly positive at the 1% level, proving that economic growth in one county can promote the current period of economic development of its surrounding counties. That is, for every 1%increase in LnGDP in local counties, LnGDP in neighboring counties would increase by 0.175%. The L.LnGDP coefficient is 0.201 and significant at the 1% level, showing that the counties with high-speed economic growth in the prior period positively impact their current economic growth. However, the coefficient of L.W*LnGDP is -0.008 but insignificant at the 10% level.
To avoid possibly insufficient conclusions due to the point estimations of the spatial panel model (Pace and LeSage 2009): the spatial decomposition effect method can be used to calculate the direct, indirect, and total effects. Table 6 shows the results for spatial decomposition with the short-and long-run direct, indirect, and total effects in the DSDM with the spatial-time fixed effect.
The regression results showed that the long-term direct and indirect effects of LnTR were all greater than their short-term direct and indirect effects, indicating that tourism development has far-reaching effects on economic resilience. Specifically, TR's short-and long-term direct effects were significantly positive. This indicated that a 1% increase in LnTR led to a short-term 0.076% and a long-term 0.095% increase in the LnGDP. These results, supporting the tourism-led growth hypothesis, were consistent with the results of Zhang and Cheng (2019) and Cheng and Zhang (2020a). They also found that tourism specialization significantly promoted economic resilience in disaster-stricken counties during the Wenchuan earthquake recovery and development period. The short-and long-term spatial indirect effects of LnTR were positive but insignificant, indicating no spatial spillover effect of tourism on surrounding counties' economic activities.
Regarding the control variables, an interesting finding was that the short-and long-term direct impacts of LnROAD were significantly negative. This indicated that road facilities in most counties have somewhat hindered economic development. This result differed from the findings of Zhang and Cheng (2019), who suggested that transport accessibility was an important factor driving economic resilience. This may be because these researchers did not capture the spatial effects. Geographically, most counties are located in remote  areas with complicated topography and poor transportation facilities, which restricts economic development. The short-and long-term indirect effects of the LnROAD were insignificantly positive, indicating that the influence of road facilities on economic resilience in adjacent counties was not obvious. The short-and long-term direct effects of LnCZZC were estimated to be significantly negative, showing that "government intervention" behavior in economic activities hinders economic development in the short and long run. That excessive "government intervention behavior," like "government failure," may lead to the distortion of resource allocations and low efficiency of the market economy. This result is consistent with those of previous Table 5 DSDM estimations in full sample SF, TF, STF refer to spatial fixed effect, time fixed effect and spatial-time fixed effect, respectively. Numbers in parentheses represent standard errors, *p < 0.1, **p < 0.02, ***p < 0.01. The LR tests with both the spatial fixed effect (4341.641, p < 0.01) and time fixed effect (2821.685, p < 0.01) were significant. Sowing that the spatial-time fixed effect model was more desirable. LM-lag statistic (52.972, p = 0.000) and robust LM-lag statistic (4.662, p = 0.000) were significant, and LM-error statistic (58.105, p = 0.000), the robust LM-error statistic (9.795, p = 0.000) are significant, implying that SEM was more efficient than SLM. Both the Wald spatial lag (40.698, p = 0.000) and LR spatial lag (40.926, p = 0.000) were significant. Likewise, the Wald spatial error (23.874, p = 0.000) and LR spatial error (67.385, p = 0.000) indicated that SDM could not be deducted to SEM. The above results showed that the estimations of the SDM were more efficient than those of the SLM and SEM. Therefore, the DSDM with a spatial-time fixed effect is the most suitable for our study  (2003), Zhao et al. (2014), and Liu et al. (2021). They believed that, among the various channels of government fiscal expenditure, economic construction expenditure and general expenditure exerted a significant negative impact on economic growth (He and Wu 2013; Liu et al. 2021). The indirect effects of LnCZZC in the short and long-term were insignificantly negative. The direct and indirect effects of LnINVEST in the short-and long term were significantly positive, indicating that fixed asset investment drove the local county's economic resilience and positively promoted the economic development of surrounding counties. This result agreed with Zuo and Huang (2018) and Liu et al. (2021).
Regarding employment level, the short-term direct effect of LnEMP was 1.316%, and the long-term direct impact was 1.649%, showing that LnEMP contributed to the increase in LnGDP. This finding is similar to that of Zhang and Cheng (2019). The indirect effects of LnEMPs were insignificant.
The direct and indirect effects of LnPOPU were all significantly negative in the shortand long-term, indicating that population growth curbed the economic resilience of local counties and adversely impacted the economic growth of neighboring counties. The underlying reason may be that large-scale population growth in a county will lead to a reduction in labor productivity, deterioration of the ecological environment, increase in the unemployment rate and crime rate, and a decrease in the investment rate, unfavorable to the economic resilience of the local county. Meanwhile, a large proportion of population influxes from surrounding counties lead to a labor shortage in the outflow counties.
For the vast territory of 135 counties, the Wenchuan earthquake may have caused other destruction in these areas. To compare the heterogeneous influence of tourism development in different regions on economic growth: the full sample is further divided into two subsamples: general disaster-affected counties and severe disaster-affected counties, according to the assessment of the Wenchuan earthquake disaster scope reported by the central government of China in 2008. The spatial effects of tourism development on economic growth were then analyzed regionally. Table 7 summarizes the results of the DSDM estimations for severely disaster-affected counties and general disaster-affected counties. For general  disaster-affected counties, tourism exerted a significantly positive effect on economic resilience linearly. However, there is an inverted U-shaped relationship between tourism and economic resilience in severe disaster counties. When LnTR is lower than 0.908, the tourism economy promotes economic resilience. Once LnTR exceeded 0.908, the tourism economy played a hindrance role to economic resilience. This conclusion is consistent with that of Zhang and Cheng (2019), who suggested that for severely disaster-affected counties when the industrial structure level (IS) was less than 0.34. Tourism specialization stimulated economic growth, whereas IS was higher than 0.34, indicating that tourism specialization stunted economic growth. Moreover, the short-and long-term indirect effects of LnTR and (LnTR)^2 are insignificantly negative at the 10% level, indicating that no spatial spillover effects of the tourism economy exist. Table 8 shows the direct, indirect, and total effects in the DSDM with the spatial-time fixed effect. For the general disaster-affected counties, the magnitudes and dimensions of the estimated results for the short-and long-term direct, indirect, and total effects of LnTR and control variables were similar to those of the full sample. The short-and long-term direct effects of LnTR were significantly positive at the 5% level, proving that tourism development contributed to economic resilience. The indirect effects of lnTR in the shortand long-term were negative but insignificant. Tourism had negative spillover effects on its surrounding counties, but these effects were not obvious. However, for severely disasteraffected counties, the short-and long-term direct effects of LnTR were significantly positive, whereas the short-and long-term direct effects of (LnTR)^2 were significantly negative, which indicated that there was an inverted U-shaped relationship between tourism development and economic resilience. In the short and long term, the direct and indirect effects of LnTR and (LnTR)^2 were insignificant.

Discussion and conclusion
Based on panel data of 135 counties affected by the Wenchuan earthquake from 2008 to 2018, this study applied a dynamic spatial Durbin model to understand the spatial effect of tourism on economic growth during the recovery and development stages. The decomposition results of the DSDM with the spatial-time fixed model provided strong evidence that tourism development enhances economic growth, which supports the application of the TLGH to disaster-affected destinations. These results are consistent with Zhang and Numbers in parentheses represent standard errors, *p < 0.1, **p < 0.02, ***p < 0.01 Cheng (2019) and Cheng and Zhang (2020a, b). They found evidence that tourism can be an effective catalyst for stimulating post-disaster economic growth in the Wenchuan Earthquake Harder-hit areas.
In contrast to previous studies, this study highlights differences in the tourism-led growth nexus between counties affected by severe disasters and counties affected by general disasters. For the general disaster-affected counties, tourism yielded significantly positive effects on economic resilience, whereas there was an inverted U-shaped relationship between tourism and economic resilience for severely disaster-affected counties. One possible reason is the economic structure of tourism in the national economy. For the general disaster-affected counties, on the one hand, most counties are underdeveloped areas, where they lack well-developed secondary sectors, or most manufacturing industries were destroyed by the Wenchuan earthquake. Thus tourism  (Lin et al. 2018). Hence, it is easier to choose a pathway for economic resilience. However, on the other hand, the tourism industry was an important but not a major part of the economy in these counties. The tourism ratio to GDP in most areas was less than 20%, showing that the economic growth does not rely on the tourism industry. Low tourism specialization can promote economic resilience to a larger extent. As Yang and Fik (2014) suggest, regions with less-developed tourism industries may experience faster economic growth. Figini and Vici (2010) and Du, Lew, and Ng (2014) find that non-tourism-dominated areas experience a higher economic growth rate than tourismdominated areas. Tourism significantly promotes economic resilience for most severely disaster-affected counties, which was also supported by Zhang and Cheng (2019) and Cheng and Zhang (2020b). However, among some of them, including Jiuzhaigou and Songpan counties, the tourism sector accounts for more than 147.9% of the GDP, indicating that these counties' economies are too heavily dependent on tourism sector. Under these circumstances, overdependence on the tourism industry is detrimental to sustainable economic growth (Adamou and Clerides 2010;Parrilla et al. 2007), leading to unfavorable outcomes, such as de-industrialization and industry hollowing out. Specifically, although rapid tourism development may promote economic growth in the short term, it leads to a reallocation of resources such as material capital and labor to flow into the tourism-oriented sectors from traditional agriculture and manufacturing, hampering the competitiveness of traditional productive sectors and compromising the long-term economic growth (Capó et al. 2007a(Capó et al. , 2007b. For example, after the Wenchuan earthquake, many high-energy-consumption and heavy-pollution industries were discarded, and the tourism industry was prioritized in some severely disaster-affected areas. In addition, the tourism-reliant mode also caused a rapid increase in local land prices and rental costs, a crowding-out effect on local enterprises, and an overall loss in social welfare (Zhang and Cheng 2019). Zuo and Huang (2018) found that when the tourism revenue ratio to GDP exceeded its highest point, the destination went into a lock-in situation owing to path dependence, resulting in a negative impact on economic performance through higher salaries and rents, overrunning operational costs, outdated knowledge, and technology.
For spatial spillovers, the short-and long-term indirect effects of LnTR are insignificant, showing that no spatial spillover effects of tourism exist for disaster-affected counties. This may be because of the limitation of the small-scale tourism market at the county level, which has an insignificant impact on the economic growth of the surrounding counties. As Liu et al. (2021) suggested, a spillover effect may occur among high-level cities, but that in the surrounding area is smaller and insignificant.
The results of this study have several theoretical and practical implications. First, it extends the scope of tourism economic research by measuring and decomposing the spatial effect of tourism on economic recovery after a disaster shock from a dynamic spatial perspective. Second, this study is the first substantial study of the spatial effect of tourism on economic resilience in the field of tourism disasters to reveal new findings and enhance the credibility of empirical results. This has revealed a complex relationship between tourism development and economic resilience. The significance of the short-and long-run marginal effects of tourism development on economic resilience within and across counties makes it important for scholars, not just in the tourism research field but also beyond, to recognize how tourism development influences economic resilience across counties from a long-term perspective. Third, this study horizontally compares the difference in the tourism-growth nexus between counties affected by severe disasters and counties affected by general disasters, further theoretically verifying the validation of the TLG hypothesis.
These conclusions provide valuable policy recommendations for tourism in disasteraffected areas. First, these disaster-affected counties, especially the undeveloped ones, should emphasize regional cooperation and coordinated development and fully use the advantage of cross-county tourism spillover to support local tourism development, ultimately driving an economic boom for the entire disaster-affected area. For example, regional governments should jointly conduct marketing campaigns and plan travel packages or routes that link multiple destinations (Santos and Vieira 2020). Second, due to overdependence on the tourism economy, tourism development in some disaster-hit counties may hinder sustainable economic development. Thus, policymakers should emphasize the linkages between tourism and relevant industries, such as agriculture, manufacturing, and other service industries, to motivate tourism vitality and strengthen the competitiveness of other economic sectors (Zhang and Cheng 2019). Third, tourism vulnerability should not be neglected, and local policymakers should implement a series of measures to strengthen the resistance of the tourism industry to natural hazards. Fourth, local governments should reduce irrational government interference in the tourism market economy, invigorate the potential of markets, and eliminate local protectionism to generate a win-win situation for the local and entire economies. Finally, considering that poor road facilities in some disaster-affected counties have become a big obstacle to economic development, governments should improve transportation facilities and thereby enhance the accessibility of tourist destinations.
Our study has some limitations. Because of the unavailability of data at the county level in some instances, the total data for tourism could not be further disaggregated into domestic and international tourism. Theoretically, tourist arrivals may have various impacts and spillover effects on earthquake-affected county economies. For example, Ma et al. (2015) found that domestic tourist arrivals exerted a positive spillover on the urban economy, whereas the number of international tourist arrivals yielded a significant negative spatial spillover effect. Therefore, separate models of these two tourist arrivals should be considered. Second, this study considers only disaster-affected counties; therefore, its conclusions may differ from those of traditional destinations. Future studies should focus on the differences in the spillover of the tourism-growth nexus between disaster-affected destinations and traditional destinations.