Dynamic prolonged effects of crime on tourism demand for Thailand National Parks

Abstract

This study examines how past crime interacts dynamically with word-of-mouth (WOM) recommendations to affect tourism demand, using a panel dataset of 132 Thailand National Parks from 2010 to 2019. Using the system-Generalized Method of Moments approach, our findings demonstrate that crimes at tourist destinations not only have an immediate impact on tourism demand but also prolong ramification by diminishing the beneficial effect of WOM recommendations. This implies that the effect of WOM on destination choice does not remain constant over time but varies with changes in the conditions of personal safety at the destination. Further analysis disaggregated by the severity of crime demonstrates that Thai tourists are most sensitive to the occurrence of life-threatening crime, followed by property crime and violent crime. In addition, life-threatening crimes diminish the positive effect of WOM recommendations by the most.

Keywords:

• crime
• domestic tourism demand
• national park
• system GMM
• word-of-mouth

1. Introduction

Personal safety is a crucial factor for sustainable tourism development. Tourists are of utmost concern for their personal safety and well-being during travel. It is widely acknowledged that a tourist destination perceived to be unsafe and has high crime rates is more likely to be screened out from a bucket list during the decision-making stage (Boakye, 2012; Karl et al., 2020). As suggested by hotspot theory (Crotts, 1996), high levels of tourist arrival at a place of visit can cause increased crime. Thus, a tourism destination with ineffective security enforcement has a high chance of experiencing tourist victimization. The tourism-crime literature suggested that crimes committed at a tourist destination will gradually damage not only the image and reputation but also the competitive advantage of a destination (Alrawadieh et al., 2019; Ghaderi et al., 2017). In addition, the exaggeration of serious crime events and the use of sensational photos and headlines by the media could trigger fear among potential visitors (Giusti & Raya, 2019). Therefore, a serious crime compounded by media exposure can have a drastic impact on tourism demand (Boxill, 2012; Brown, 2015).

Referring to the fear appeal theory, crime incidents committed at a tourist destination could trigger fears among potential visitors, eventually inducing the act of postponement or cancelation of a planned visit or a change of destination (Brown, 2015; Giusti & Raya, 2019; Hem et al., 2003). These immediate responses to a change in safety conditions at a destination inevitably cause a significant decline in tourism demand. Empirical studies aimed at estimating the contemporaneous effect of crime on destination choice consistently affirmed the negative relationship between tourist arrivals and crime rates (Altindag, 2014; Johnny & Jordan, 2007; Massidda & Etzo, 2012; Walker & Page, 2007). A crime incident may be a rare or even a one-off event, but the damage caused to tourism demand is likely to be prolonged. Such damages have not been captured by past studies on the contemporaneous impact of crime on travelling demand. The prolonged or long-term effects of crime on tourism flows have rarely been investigated, and the findings have been inconsistent (Alleyne & Boxill, 2003; Altindag, 2014; Biagi et al., 2012; Lee et al., 2018; Lorde & Jackman, 2013; Mehmood et al., 2016). In this study, we postulate that the occurrence of crimes at a tourist destination exhibits not only a temporal effect on tourists’ destination choice but also a prolonged effect via word-of-mouth (WOM) recommendations from previous visitors. Therefore, our findings enrich our understanding of how the crime effect can be transmitted from one period to another.

WOM recommendation is a post-purchase evaluation formed from the overall perception of a visit, based on tourists’ experiences and well-being during visitation. Several studies have asserted that destinations that receive positive WOM recommendations from previous visitors are more likely to attract a greater number of tourist arrivals in the future (Abubakar & Ilkan, 2016; Moro & Rita, 2018). In tourism demand modelling, the WOM effect can be captured by incorporating lagged demand into the conventional demand model (Etzo et al., 2014). This dynamic specification exhibits its superiority over the static one in capturing the variation of travelling demand across time (Gallego et al., 2019). However, the tourism area life cycle (TALC) theory suggests that the marginal effect of WOM recommendations on current tourist arrivals should not be constant but should adapt according to supply side factors (Butler, 1980, 2009). Since we hypothesized that the negative feeling from the unsafe experience of a visit could be disseminated by WOM among acquaintances, any previously accumulated positive WOM effect on tourism demand could diminish as incidents of crime increase. This suggests that tourism growth could increase at a decreasing rate and could reach demand stagnation as destination degradation gradually worsens. The diminishing WOM effects due to increasing crime rates, which have been neglected in the literature, are considered in this study.

This study aims to modelling the domestic tourism demand for Thailand National Parks (TNPs), where crime is attributed to a negative supply side element. The analysis emphasizes domestic tourism because tourism activity at TNPs by Thai tourists predominates their international counterparts in terms of both tourism flow and receipt. In addition, despite its remarkable economic contributions, especially when international flows are comparatively low due to the COVID-19 pandemic (Wang, Lai & Wong, 2022), research on domestic tourism is still very limited. From an analytical point of view, we anticipate that the variation in tourism demand could be negatively influenced by crime due to both immediate and prolonged effects of WOM recommendations. We propose a dynamic tourism demand model that incorporates lagged demand as a proxy for the WOM effect and its interaction with past crime conditions. This new attempt allows the adjustment of the marginal effect of WOM recommendations on tourism demand according to changes in safety conditions.

Several studies have suggested that travelers react differently to different types of crimes because their severity is heterogeneous (Lorde & Jackman, 2013; Pizam & Fleischer, 2002). A disaggregated analysis is performed by decomposing total crime into three categories based on severity: life-threatening crime, violent crime, and property crime. The disaggregated analysis has the advantage of differentiating the degree of the adverse effects of different severities of crime on tourism demand via the contemporaneous and prolonged effects of WOM recommendations. To achieve the research objectives, our empirical analysis utilized a panel dataset consisting of 132 TNPs located across Thailand from 2010 to 2019. The proposed dynamic demand model was estimated using the system GMM procedure (Blundell & Bond, 1998) to address the problems of omitted variables, measurement errors, endogeneity, and destination-specific heterogeneity. The remainder of this paper is organized as follows. The next section reviews past studies on crime and tourism demand and explains the concept of WOM recommendations from the context of the literature. Background information on the TNPs is also provided. The methodology and data used were as follows. The results are presented in the following section. The final section concludes the paper by providing implications for the study.

2. Literature review

2.1. Crimes and tourism demand

In the tourism-crime literature, the majority of studies indicated that crime exhibits adverse effects on the tourism industry to some extent. Most empirical results demonstrate that criminal acts committed at tourist destinations, especially against tourists, seriously devastate tourism development (Alleyne & Boxill, 2003; Brown, 2015; Hua & Yang, 2017; Lorde & Jackman, 2013; Mohammed & Sookram, 2015). According to the hotspot theory, Boakye (2010) explained that criminals consider tourists as easy targets because of their unfamiliarity with locations, local laws, and local communities. In addition, the motive of the offender is driven by tourists’ behavior, such as imprudence or carelessness, as well as carrying valuable belongings with them (Fowler et al., 2012). Consequently, a popular tourist attraction is typically vulnerable to crimes because it opens the opportunity for criminals to seize a suitable victim. Meanwhile, Biagi and Detotto (2014) argued that the existing crime rate of a destination indicates the chance of tourist victimization, which aligns with routine activity theory (Cohen & Felson, 1979). This implies that whenever a suitable victim appears at a destination with weak self-security, a motivated criminal would commit the crime against either the tourist or the local resident with equal chance. In addition, several recent studies viewed crime and tourism from the local community perspective. They suggested that excessive tourism flows in a particular location can destabilize the community to the extent of changing its original social fabric and causing social disorganization, which in turn lead to higher crime rates (Abbasian et al., 2020; López-Gay et al., 2021; Roth, 2021; van Holm & Monaghan, 2021). Ke et al. (2021) demonstrated that the prevalence of Airbnb listing in Boston had increased residential mobility and racial heterogeneity. These changes undermined the natural ability of the local community to prevent crime. Analogously, Maldonado-Guzmán (2023) asserted that population turnover and ethnic-cultural heterogeneity due to tourist arrival intensity mediated positively the linkage between tourism and crime rates in Barcelona.

The notion of destination carrying capacity suggests that crime is an undesirable social issue stemming from the rapid growth of tourism activities (Lee et al., 2018; Palanca-Tan & Garces, 2015). The majority of empirical evidence from testing the tourism-led crime hypothesis revealed that an expansion of tourism demand induces an escalation of crime incidents at hosted destinations (Biagi & Detotto, 2014; Mehmood et al., 2016; Mohammed & Sookram, 2015). For instance, the system GMM estimation by Biagi et al. (2012) indicated that tourism areas in Italy are exposed to a significantly higher occurrence of criminal incidents than non-tourism areas. Mehmood et al. (2016) employed the autoregressive distributed lag (ARDL) model to investigate the association between international tourist arrivals to the U.S. and crime rates. Their results revealed that an increased number of international visitors led to higher criminal incidents. In contrast, Lee et al. (2018) illustrated that the number of foreign visitors to Taiwan is negatively associated with crime rates. They explained their contradictory findings that the evolution of tourism destinations usually undergoes various developments, including safety and security programs, making tourism sites less conducive to crimes. This evidence highlights that spatial heterogeneity should not be ignored in examining the association between tourist flows and crime. Several studies argued that tourism demand and crime are spatial referenced data. Using global regression methods may yield a negative association between the two variables despite that in some locations, tourist arrivals actually accentuate crime rates (Kim & Nicholls, 2016; Maldonado-Guzmán, 2022; Xu et al., 2019). Xu et al. (2019) explored the spatial patterns of crime with respect to Airbnb lodging in Florida using geographically weighted regression that allows for spatial heterogeneity. They revealed that the local correlations between Airbnb density and crime vary from negative to positive.

Another plausible reason for the inconsistent results from tourism-led crime studies may be attributable to the use of aggregated crime, rather than that distinguished by severity, in the analysis. Cherry and List (2002) pointed out that the use of crimes disaggregated by severity is more suitable than total crimes in analyzing tourists’ decision models. Several empirical studies have revealed that tourism development induces only certain types of crime at the destination. For example, Biagi and Detotto (2014) analyzed criminal incidents in 2005 in 103 Italian provinces using the spatial lag model. Their results demonstrated that increased tourist arrivals caused increased street crimes, especially pick-pocketing offences. Montolio and Planells-Struse (2016) collected provincial panel data for Spain from 2000 to 2008 and utilized the GMM approach for the analysis. They reported that a 1% increase in tourist arrivals to Spain significantly increases serious crimes against people by 0.1% and property crimes by 0.35%. In the Philippines, Palanca-Tan and Garces (2015) applied panel Ordinary Least Squares (OLS) to analyze the data of 16 regions from 2009 to 2011. The results revealed that the regions visited mainly by foreign visitors were associated with large numbers of robberies and thefts. Additionally, increased property crimes induced by the growth of international tourism flows have been analogously observed in European countries (Altindag, 2014; Michalko, 2004), as well as on Caribbean Island (Johnny & Jordan, 2007; Mohammed & Sookram, 2015).

The tourism-crime literature also asserts that the degree of the negative effect of crime on destination choice depends largely on either the frequency or severity of criminal incidents (Alleyne & Boxill, 2003; Pizam & Fleischer, 2002). The regularity of crimes reflects not only weak security systems, but also inefficient criminal preventive programs implemented by destination authorities of the destinations (Altindag, 2014; Michalko, 2004). Therefore, a tourist destination with a high crime rate is gradually perceived as unsafe by travelers. Considering crime severity, on the other hand, different types of crimes may affect the tourists’ decision differently. Lorde and Jackman (2013) illustrated that the act of murder has the largest negative effect on tourist arrivals to Barbados, followed by assault with intent to rob, rape, and burglary. In addition, Hua and Yang (2017) indicate that the negative influence of violent crime on lodging business performance in Houston is more substantial than that of property crimes. Similarly, Mohammed and Sookram (2015) reported that violent crime has a larger adverse effect on tourism demand for Jamaica than property crime, whereas international visitors to Trinidad and Tobacco are more sensitive to property crime.

Several tourism-crime studies revealed that serious crime committed at tourism sites or against tourists compounded with public exposure by the media would gradually damage their image and reputation, resulting in these destinations losing their attractiveness among travelers (Boxill, 2012; Brown, 2015; Giusti & Raya, 2019). From a journalism perspective, Vettehen et al. (2008) described that the magnitude of life-threatening crimes is more likely to be exaggerated by the media because it serves viewers’ thirst for sensational news. Therefore, a combination of a serious crime and uncontrolled exaggerations by the media would have a negative influence on holidaymakers’ decisions, causing a sharp decline in tourist demand after its occurrence. In Lorde and Jackman (2013), the impulse response function analysis demonstrated that tourist arrivals to Barbados could take up to 61 months to recover to their original level if the murder rate increases by 1 percent compared to 48 and 51 months for rape and assault with rob, respectively. Ample evidence suggests that different types of crime negatively affect tourism demand differently. However, comparative analyses of crime effects by crime severity are very restricted in tourism-crime literature. Most studies have highlighted the role of crimes on travel intention, while their role in tourists’ behavior after visitation has received inadequate attention from researchers. In particular, the link between crime disaggregated by severity and WOM recommendations, and its impact on tourism demand remains unexamined.

2.2. Word of mouth recommendation in the tourism context

Generally, WOM recommendations are used to assess and exchange information regarding products and services among friends, family, colleagues, and acquaintances. WOM recommendation, by definition, is a form of interpersonal communication that directly informs other potential consumers of either positive or negative perceptions and experiences after using a particular good or service (Chen & Law, 2016). Compared to conventional advertising media, informal and non-commercial information shared among close relations or acquaintances allows WOM recommendations to gain more credit and trust (Gupta & Harris, 2010). Marketing researchers have indicated that WOM recommendation is a more effective strategy than direct sales or broadcasting advertising (Cheung & Thadani, 2012). In the context of tourism, analogously, having past visitors recommend a tourist destination via their positive WOM could be a great tool for acquiring new demand for tourism (Abubakar & Ilkan, 2016). Recent studies have demonstrated that optimistic WOM promotes awareness of a tourist attraction among unfamiliar travelers (Moro & Rita, 2018).

As WOM recommendation is a post-purchase evaluation, the overall experiences obtained during a visit play an essential role in promoting positive information. Tourism literature suggests that various psychological factors such as satisfaction, service quality, loyalty, and destination attributes, namely, tourism product, destination image, accommodation, and local community, are strongly correlated with positive WOM recommendations (George & Swart, 2015; Lai & Wong, 2022; Phillips et al., 2011; Zhou et al., 2022). Given the importance of personal safety and well-being to most tourists during their travel, it can be postulated that crime occurrence at a tourist destination could have a negative influence on travelling experience, which eventually produces an adverse WOM recommendation. Sönmez and Graefe (1998) revealed that tourists’ perception of crime, affected by their past trip experience, has a negative impact on future recommendations. George and Swart (2015) conducted interviews with 354 tourists attending the London 2012 Olympic Games to examine the linkage between perceptions of personal safety and WOM recommendations. The results indicated that 22.5% of the interviewees would not recommend London to others because they felt unsafe during the visit.

Along the same line, Lai et al. (2018) investigated the role of safety experience on WOM recommendation among Mainland Chinese tourists visiting an urban destination at Macau. Their empirical analysis based on structural equation modelling (SEM) showed that safety experience has both direct and indirect effects on trip satisfaction via WOM recommendation. However, several empirical studies have found contradictory results. George (2003), for instance, found that most visitors (69%) were very likely to recommend Cape Town to others, even though they felt unsafe during the trip. Subsequently, George (2010) conducted a survey of tourists visiting Table Mountain National Park (TMNP), Cape Town. The results indicated that the respondents were willing to recommend TNP as a destination, although their concern about safety was high during visitation. These contradictory results reflect the complexity of how WOM recommendations are generated; however, this field of research still receives little attention from academics and practitioners.

Regarding tourism demand modelling, a few studies have applied the dynamic demand model to examine the association between WOM recommendations and current demand. Using panel data, Massidda and Etzo (2012) explain that Italian domestic demand is positively influenced by recommendations from previous visitors. Bento (2014) examined the factors influencing international academic arrivals in Portugal under the Erasmus student mobility program. The empirical results show that shared beliefs from previous visitors positively affect tourism flows. Similarly, the estimated results obtained from the system GMM estimator affirmed the positive linkage between WOM recommendations and Italian inbound tourism demand (Etzo et al., 2014; Massidda et al., 2015) and Malaysia (Tang, 2018). Nevertheless, all these studies explicitly assume that the positive WOM effect on tourism demand does not vary over time, although this is questionable. The influential power of WOM recommendations can be adapted contingent on the stage of destination evolution rather than fixed throughout time (Butler, 2009). In particular, the impact of WOM recommendations on destination choice should be modified according to changes in safety conditions at the destination. However, the dynamic influence of demand on the interaction between WOM recommendations and crime remains neglected in the literature.

3. Methodology

3.1. Data and model specification

We formulate a demand function based on consumer behavior theory to estimate the impact of crimes on domestic tourism demand for TNPs. The theory states that the variation in demand can be explained by changes in income and price-type factors. The literature on domestic tourism also suggests that tourists’ behavior may be influenced by other non-economic and exogenous factors, including crime and WOM recommendations (Garín-Muñoz, 2009; Massidda & Etzo, 2012). Thus, our theoretical framework can be expressed mathematically in generic form as follows:

(1) $T{A}_{i,t}={\alpha }_0\mathit{\text{GD}}{P_t}^{{\beta }_1}{P_{i,t}}^{{\beta }_2}{Z_{i,t}}^{\gamma }{e}^{{\epsilon }_{i,t}}$(1)

where $T{A}_{i,t}$ represents the domestic tourism demand for TNP, determined by the number of visitors to national park (i) in year (t). $\mathit{\text{GD}}{P}_t$ is the real gross domestic product (GDP) per capita representing the income of a local Thai in year (t). $P_{i,t}$ is constructed by dividing the consumer price index (CPI) of the province where national park (i) is located by the country’s corresponding CPI. $Z_{i,t}$ is a vector of exogenous factors.

Crime and WOM recommendations are considered in the vector $Z_{i,t}$. Three aspects are emphasized in this consideration. First, to examine the role of personal safety in domestic tourism demand, the crime rate was employed to approximate the safety conditions of the destinations. Second, tourism demand is fundamentally dynamic, as previous visitors can affect prospects through WOM recommendations (Morley, 1998, 2009). Specifically, the lagged dependent variable capturing the WOM effect was incorporated into the demand model. Lastly, we argue that the crime rate at a destination not only has a contemporaneous impact on tourism demand, but its past values also affect current tourism flows via WOM recommendations from previous travelers who had bad experiences or felt unsafe conditions due to crime. With respect to these three aspects, the demand function expressed in Equation (1)(1) $T{A}_{i,t}={\alpha }_0\mathit{\text{GD}}{P_t}^{{\beta }_1}{P_{i,t}}^{{\beta }_2}{Z_{i,t}}^{\gamma }{e}^{{\epsilon }_{i,t}}$(1) can be transformed into an estimation model as follows:

(2) $T{A}_{i,t}={\alpha }_0+{\beta }_1\mathit{\text{GD}}{P}_t+{\beta }_2{P}_{i,t}+{\gamma }_{1}T{A}_{i,t-1}+{\gamma }_{2}C{R}_{i,t}+{\gamma }_3{\left(\mathit{\text{TA}}\times \mathit{\text{CR}}\right)}_{i.t-1}+{\epsilon }_{i,t}$(2)

where $T{A}_{i,t-1}$ is the lagged dependent variable that captures the WOM recommendation effect. $C{R}_{i,t}$ is the ratio of criminal cases to the number of visitors to national park (i) in year (t). According to safety perceptions, a high ratio implies unsafe conditions at a destination. The non-linear term ${\left(\mathit{\text{TA}}\times \mathit{\text{CR}}\right)}_{i.t-1}$, measures the interaction effect of safety and WOM recommendations on current tourism demand. This variable captures the adaptive WOM effect, which varies over time according to the evolution of a destination’s safety condition. Given the differences in the negative impacts of crimes distinguished by their severity, four models were estimated in this study. One is for analyzing the total crime, while the remaining are for assessing life-threatening crime, violent crime, and property crime. ${\epsilon }_{i,t}$ denotes the disturbance term. All variables presented in Equation (2)(2) $T{A}_{i,t}={\alpha }_0+{\beta }_1\mathit{\text{GD}}{P}_t+{\beta }_2{P}_{i,t}+{\gamma }_{1}T{A}_{i,t-1}+{\gamma }_{2}C{R}_{i,t}+{\gamma }_3{\left(\mathit{\text{TA}}\times \mathit{\text{CR}}\right)}_{i.t-1}+{\epsilon }_{i,t}$(2) , except for crime rate, are in natural logarithms; thus, these coefficients can be interpreted as elasticity.

According to economic theory and the empirical evidence provided in Sections 2 and 3, it is anticipated that ${\beta }_1$ and ${\gamma }_1$ would have a positive sign, while ${\beta }_2$ and ${\gamma }_2$ are expected to show a negative sign. The interaction between WOM recommendation and crime rate suggests that the marginal effect of WOM recommendation on the current tourism demand depends not only on ${\gamma }_1$, but also on the product of ${\gamma }_3$ and the previous level of crime rate, that is, $\mathrm{\partial }T{A}_{i,t}/\mathrm{\partial }T{A}_{i,t-1}={\gamma }_1+{\gamma }_{3}C{R}_{i,t-1}$ as $\mathrm{\Delta }T{A}_{i,t}\to 0$. If ${\gamma }_1$ is positive and ${\gamma }_3$ is negative, any positive effect of WOM recommendation on current tourism demand would be diminished by the past crime rate. However, if ${\gamma }_3$ is not significantly different from zero, the assumption of constant elasticity of tourism demand with respect to WOM effect is validated.

To achieve the objectives of this study, a balanced panel dataset consisting of annual data on 132 national parks located across the nation during the 10-year period between 2010 and 2019 was used for our analysis. The use of a panel structure enlarges the degrees of freedom, reduces multicollinearity, promotes the accuracy of the estimates, and mitigates omitted variable bias (Garín-Muñoz, 2009; Massidda & Etzo, 2012). Data on the number of domestic visitors to national parks were collected from the DNP of Thailand. Data on economic variables and criminal cases were extracted from the National Statistical Office (NSO) of Thailand.

3.2. Estimation methodology

In modelling the dynamic demand function with panel data, OLS can yield biased and inconsistent estimators. In addition, incorporating a lagged dependent variable into the dynamic demand model potentially encourages dynamic panel bias because of its correlation with the unobservable individual-specific effect (Nickell, 1981). Although individual-specific heterogeneity can be mitigated using either the transformation of the fixed-effects or random-effects model, this specification inevitably introduces dynamic endogeneity caused by the correlation between the past realization of the dependent variable and error term. Conversely, consider a generic dynamic model expressed in Equation (3)(3) $Y_{i,t}={\eta }_i+\alpha Y_{i,t-1}+X_{i,t}^{\prime }\beta +{\epsilon }_{i,t}$(3) :

(3) $Y_{i,t}={\eta }_i+\alpha Y_{i,t-1}+X_{i,t}^{\prime }\beta +{\epsilon }_{i,t}$(3)

where $Y_{i,t}$ and $Y_{i,t-1}$ represent the dependent and lagged dependent variables, respectively. ${\eta }_i$ is the unobserved individual-specific effect that varies cross-sectionally but is fixed within an individual over time. $X_{i,t}^{\prime }$ is a vector of exogenous variables and ${\epsilon }_{i,t}$ is the error term. Arellano and Bond (1991) proposed the difference GMM method by taking the first difference of Equation (3)(3) $Y_{i,t}={\eta }_i+\alpha Y_{i,t-1}+X_{i,t}^{\prime }\beta +{\epsilon }_{i,t}$(3) to remove the unobserved individual-specific variable (${\eta }_i$). Because $\mathrm{\Delta }{Y}_{i,t-1}$ may be correlated with the error term, the lagged level of endogenous variables is used as the instrument variable (IV) to mitigate endogeneity. The first differencing model is expressed as follows:

(4) $\mathrm{\Delta }{Y}_{i,t}=\alpha \mathrm{\Delta }{Y}_{i,t-1}+\beta \mathrm{\Delta }{X}_{i,t}^{\prime }+\mathrm{\Delta }{\epsilon }_{i,t}$(4)

Blundell and Bond (1998) point out that employing lagged levels of endogenous variables as IVs for equations in the first difference form may promote the selection of weak instruments. They improved the estimation procedure by introducing a system-GMM approach. In this approach, a system of equations is formed by combining the equation in the first differences and the equation in the levels, whereby the former and latter are instrumented by lagged levels and lagged first differences, respectively. In this study, the system GMM method (Blundell & Bond, 1998) is employed to modelling the dynamic demand model expressed in Equation (2)(2) $T{A}_{i,t}={\alpha }_0+{\beta }_1\mathit{\text{GD}}{P}_t+{\beta }_2{P}_{i,t}+{\gamma }_{1}T{A}_{i,t-1}+{\gamma }_{2}C{R}_{i,t}+{\gamma }_3{\left(\mathit{\text{TA}}\times \mathit{\text{CR}}\right)}_{i.t-1}+{\epsilon }_{i,t}$(2) . This approach addresses the problems of omitted variables, measurement errors, endogeneity, and destination-specific heterogeneity simultaneously. The consistency of the system GMM estimator and IVs is assessed by Hansen’s (1982) J-test of over-identifying restrictions and Arellano and Bond’s (1991) test of autocorrelation between the error terms.

4. Empirical results

This section reports and discusses the results. The model in Equation (2)(2) $T{A}_{i,t}={\alpha }_0+{\beta }_1\mathit{\text{GD}}{P}_t+{\beta }_2{P}_{i,t}+{\gamma }_{1}T{A}_{i,t-1}+{\gamma }_{2}C{R}_{i,t}+{\gamma }_3{\left(\mathit{\text{TA}}\times \mathit{\text{CR}}\right)}_{i.t-1}+{\epsilon }_{i,t}$(2) was estimated four times to examine the negative impact of crimes distinguished by their severity. The estimated results obtained from the system-GMM approach are presented in Table 1. The analysis of total crime is in column 2, while the disaggregated analysis of life-threatening, violent, and property crimes are shown in columns 3, 4, and 5, respectively. At the 5% significance level, diagnostic tests validated the desirable properties of the proposed models. The Arellano-Bond test shows that the residuals exhibit first-order autocorrelation, while the second order is auto-uncorrelated. In addition, the Hansen J-test results confirm the joint validity of the instruments because they fail to reject the null hypothesis of no over-identifying restrictions. Given these justifications, further inferences and interpretation of the results can be made.

Table 1. Results of dynamic panel GMM estimation for Thailand National Parks

Explanatory Variable	Crime Severity
	Total	Life-threatening	Violent	Property
$\mathit{\text{GD}}{P}_t$	0.074	0.068	0.050	0.064
$P_{i,t}$	−9.624***	−9.210***	−9.301***	−10.086***
$T{A}_{i,t-1}$	0.288***	0.345***	0.274***	0.301***
$C{R}_{i,t}$	−0.019***	−0.045***	−0.018***	−0.025***
$T{A}_{i,t-1}.C{R}_{i,t-1}$	−0.001***	−0.007***	−0.003***	−0.003***
Constant	7.368***	5.958***	6.961***	8.042***
AR(1) (p-value)	0.002	0.001	0.002	0.003
AR(2) (p-value)	0.623	0.445	0.681	0.593
Hansen test (p-value)	0.174	0.17	0.136	0.233
No. of instruments	12
No. of groups	132
No. of observations	1188

Note: The dependent variable is tourist arrival ($T{A}_{i,t}$) at TNP. All variables, except for crime rate, were log-transformed. ***, ** and * denote statistical significance at the 1%, 5% and 10% levels, respectively.

The sign of the coefficient of the economic factors is consistent with our hypothesis. Despite its statistical insignificance, the results indicate a positive income elasticity for domestic tourism demand. Regardless of crime severity, the coefficients for income are less than 1, suggesting that holidays at TNP are not regarded as luxury goods. Since travelling to National Parks is not sensitive to income change, they are destinations for leisure for Thai people from all walks of life. The price factor is negative and statistically significant for TNP for all types of crime. The demand for TNP decreased from 9.21% to 10.086% in response to a 1% increase in the relative price. The sign of the price coefficient not only supports our hypothesis but also supports the findings of previous studies on domestic tourism (Garín-Muñoz, 2009; Massidda & Etzo, 2012; Nguyen & Paula Remoaldo, 2021).

In all models, crime rate has a significant negative effect on contemporaneous tourism demand. Our findings indicate that for total crime, a 1% increase in the current crime rate per tourist could lead to a 1.9% reduction in the domestic demand for TNP in the current year. As hypothesized earlier, tourists react differently to different types of crimes. When analyzing the effect of crime on tourism demand, we found that the impact differs across severity levels, indicating that it is not uniform across all three levels. In this study, life-threatening crimes included murder, assault to death, and attempted murder, while the acts of other assaults and rape were categorized as violent crimes. In addition, robbery and carjacking are grouped as property crimes. The results reveal that life-threatening crime exhibits the largest negative effect on domestic demand, followed by property and violent crime. Our results are consistent with those of Barbados (Lorde & Jackman, 2013), Houston (Hua & Yang, 2017), and Jamaica (Mohammed & Sookram, 2015). In addition, it should be noted that the coefficients of life-threatening and property crimes are larger in magnitude than those of total crime. These findings reinforce the need to use disaggregated crimes, rather than total crimes, for tourism demand modelling.

Focusing on the WOM recommendation effect, the coefficient of tourist arrivals in the previous period ($T{A}_{i,t-1}$) is positive and statistically significant at the 1% level for all categories. Approximately 28.8% of the change in domestic demand is attributable to the WOM effect in the total crime model. This result highlights the remarkable impact of WOM recommendations on promoting domestic tourism on TNP. Our findings align with the domestic tourism literature, which articulates that WOM is a crucial factor in determining domestic tourism demand (Garín-Muñoz, 2009; Massidda & Etzo, 2012). However, the marginal effect of WOM recommendations on domestic demand is inconsistent. Instead, it varies depending on the crime rate at the destination. This is supported for all four crime categories, where the estimated coefficient of the interaction term ${\left(\mathit{\text{TA}}\times \mathit{\text{CR}}\right)}_{i.t-1}$ is negative and statistically significant at the 1% level. The negative coefficient suggests that past crime rates diminish the marginal effect of WOM recommendations on current domestic tourism demand. Thus, a crime situation dynamically changes the WOM effect. The marginal WOM effect on domestic tourism demand for TNP can become negative if the past crime rate is high. The inflection point for the marginal WOM effect was computed to obtain the threshold crime level before the effect became negative. The threshold is 2.88 crimes per tourist for total crime. Since the coefficient of the interaction term is relatively small compared to the coefficient of tourist arrivals in the previous period, the marginal WOM effect is still positive given the current crime situation, where the crime rate has a mean of 0.066, which is below the threshold. corresponding thresholds are 0.49 for life-threatening crime, 0.91 for violent crime, and 1.00 for property crime, respectively. The low tolerance threshold for life-threatening crimes is noteworthy. This is almost six times lower than the threshold for total crimes. The results highlight the importance of analyzing the effects of crime by crime category.

We further conducted two subsample analyses by separating the TNPs into two groups: Terrestrial National Parks (NPs) and Marine National Parks. Among the 132 TNPs, 109 were terrestrial and 23 were marine. Terrestrial and marine NPs have different characteristics and attractions that appeal to tourists with different interests (Biagi & Detotto, 2014). Thus, the subsample analysis examines the different impacts of crime and WOM recommendations by distinguishing the characteristics of parks and visitors. Table 2 presents the estimated results for Terrestrial NPs, whereas the analysis of Marine NPs is presented in Table 3. In both sets of estimations, diagnostic tests confirm that the estimated models are valid at the 5% significance level. The Arellano-Bond autocorrelation test suggests only first-order autocorrelation among the residuals. Additionally, the Hansen J-test finds no evidence of misspecified instruments that validate the size of the instruments.

Table 2. Results of dynamic panel GMM estimation for terrestrial national parks

Explanatory Variable	Crime Severity
	Total	Life-threatening	Violent	Property
$\mathit{\text{GD}}{P}_t$	0.054	0.038	0.095	0.069
$P_{i,t}$	−7.601***	−7.090***	−7.526***	−7.595***
$T{A}_{i,t-1}$	0.264*	0.222*	0.238*	0.216*
$C{R}_{i,t}$	−0.008***	−0.043***	−0.017***	−0.025***
$T{A}_{i,t-1}.C{R}_{i,t-1}$	−0.002***	−0.009***	−0.004***	−0.004***
Constant	8.125***	6.357***	7.701***	8.818***
AR(1) (p-value)	0.021	0.010	0.016	0.034
AR(2) (p-value)	0.863	0.929	0.845	0.834
Hansen test (p-value)	0.115	0.102	0.109	0.125
No. of instruments	12
No. of groups	109
No. of observations	981

Note: The dependent variable is tourist arrival ($T{A}_{i,t}$) for Terrestrial NPs. All variables, except for crime rate, were log-transformed. ***, ** and * denote statistical significance at the 1%, 5% and 10% levels, respectively.

Table 3. Results of dynamic panel GMM estimation for marine national parks

Explanatory Variable	Crime Severity
	Total	Life-threatening	Violent	Property
$\mathit{\text{GD}}{P}_t$	0.483	0.532	0.755	0.293
$P_{i,t}$	−4.645**	−5.869**	−5.689**	−5.372**
$T{A}_{i,t-1}$	0.468**	0.518***	0.570*	0.675***
$C{R}_{i,t}$	−0.059***	−0.321***	−0.083***	−0.167***
$T{A}_{i,t-1}.C{R}_{i,t-1}$	−0.005*	−0.009*	−0.002	−0.005
Constant	−0.139	−1.261	−1.328	−0.021
AR(1) (p-value)	0.018	0.007	0.004	0.002
AR(2) (p-value)	0.42	0.532	0.372	0.826
Hansen test (p-value)	0.408	0.602	0.415	0.366
No. of instruments	12
No. of groups	23
No. of observations	207

Note: The dependent variable is the tourist arrival ($T{A}_{i,t}$) at Marine NPs. All variables, except for crime rate, were log-transformed. ***, ** and * denote statistical significance at the 1%, 5% and 10% levels, respectively.

The coefficients of the economic variables obtained from both sub-sample analyses are as expected and are consistent with consumer theory. The estimated coefficients demonstrate positive elasticity of demand with respect to income, although these coefficients are not statistically significant. The price factor is still significant in explaining the variation in local tourist visits for both terrestrial and marine NPs. It should be noted that visitors of Terrestrial NPs are more responsive to price changes than those travelling to Marine NPs. This finding may be attributable to the fact that many terrestrial NPs are located in areas near to other places of attractions. These alternative destinations are substitutes for leisure activities, thereby making tourists more price-sensitive since their availability allow visitors to explore other destinations or activities that offer better value for money. Although the non-economic factors are statistically significant in both groups of national parks, the coefficients of previous arrivals reveal that WOM recommendations exhibit a greater effect in stimulating domestic tourism demand for Marine NPs compared to Terrestrial NPs. The analysis of destination attributes revealed that crimes, regardless of their severity level, exhibit an adverse effect on destination choice for both Marine and Terrestrial NPs. Tourists visiting Marine NPs, however, show greater concerns about personal safety than those travelling to Terrestrial NPs. The plausible explanation is that most of the marine NPs in Thailand are world-renowned destinations that attract not only domestic visitors but also a significantly higher number of international tourists compared to the terrestrial NPs. These tourism hotspots, according to the hotspot theory, provide the opportunity for criminals to commit crime and raise safety concerns among visitors.

The effect of crimes, distinguished by their severity, on current tourism demand exhibits a pattern similar to that discussed in the full-sample analysis. The effect of life-threatening crime on tourism demand is the largest, followed by property and violent crimes. The occurrence of life-threatening crimes at Marine NPs evokes the greatest concern. The estimated coefficient indicates that a one percentage point increase in life-threatening crime at Marine NPs could shrink tourism demand by 32.1% in the same year, compared to 4.3% for Terrestrial NPs. Correspondingly, property crime also showed a larger adverse effect (16.7%) on domestic tourism demand for Marine NPs than on the demand for Terrestrial NPs (2.5%).

The estimated coefficients of the interaction term for all crime severity levels are negative and statistically significant for Terrestrial NPs, while those of the total crime and life-threatening crime for Marine NPs are also significant. The estimated coefficients of the interaction term for total crime are 0.002 and 0.005 for terrestrial and marine NPs, respectively. This implies that the WOM recommendations from domestic tourists visiting Marine NPs are more sensitive to the change in crime rate at the destination than those visiting Terrestrial NPs. The significance of the interaction term demonstrates that the marginal effect of WOM recommendations on domestic demand for both terrestrial and marine NPs is not constant but diminishes if the crime rate at the destination increases. Finally, the analysis disaggregated by crime severity also confirms that local tourists react differently in making their recommendations to dissimilar types of crime. Meanwhile, the negative WOM effect of life-threatening crime is still predominant among different crime categories.

5. Conclusion and policy recommendations

This study examines the dynamic impact of crime on domestic tourism demand for TNPs. We hypothesize that criminal incidents committed at the destination exhibit not only a temporal effect on tourists’ destination choice but also prolong the negative WOM recommendation effect from previous visitors. A conventional dynamic demand model incorporating economic, lagged dependent, and crime variables is formulated for the analysis. A variable generated by interacting the WOM effect and past crime rate was introduced to examine the dynamic impact of past crimes on domestic demand through WOM effects. This model specification allows for variations in the WOM effect over time, which are attributable to changes in a destination’s safety conditions. Annual panel data covering 132 National Parks from 2010 to 2019 were used, and the system GMM approach was employed for estimations.

The empirical results reveal that Thai tourists visiting TNPs are very sensitive to changes in relative prices. In contrast, income is not a significant deciding factor affecting tourism demand for holidays in TNPs. Apart from these economic factors, the statistically significant and positively large coefficients of the WOM effect indicate that recommendations from previous visitors have a strong influence on the potential demand for domestic travelling to TNPs. Although local tourists are generally perceived to be less vulnerable to crime victimization than international visitors (Michalko, 2004; Palanca-Tan & Garces, 2015), our results indicate that safety conditions measured by crime rate at the destination still significantly affect domestic travel to TNPs. Crimes at destinations not only have an immediate impact on tourism demand, but also have a prolonged negative effect on WOM recommendations. This finding contributes to tourism literature by increasing the understanding of the channel through which the crime effect can be transmitted from one period to another. In addition, the findings show that the effect of WOM on destination choice does not remain constant but varies with changes in the conditions of tourism supply, and in this context, personal safety at the destination.

Further analysis disaggregated by the severity of crime was conducted to explore whether Thai tourists respond differently to different types of crime to increase the understanding of the effect of crime on domestic tourism. The estimated results demonstrate that domestic demand responds unequally to changes in different types of crime. Thai tourists are most sensitive to the occurrence of life-threatening crimes, followed by property and violent crimes. Similarly, life-threatening crimes diminish the positive effects of WOM recommendations. A sub-sample analysis, categorized by the geographical characteristics of NPs (terrestrial and marine parks), was conducted to explore the differences in the behavior of tourists toward unsafety and insecurity according to these two destination characteristics. First, we found that local tourists visiting marine NPs are less sensitive to price changes and are more influenced by positive WOM recommendations than those visiting terrestrial parks. Second, the degree of response to crime to the occurrence of life-threatening crimes at marine NPs is most pronounced. Crime was also found to have a diminishing marginal effect on WOM recommendations for both categories of NPs.

The outcomes of this study have significant implications for stakeholders in the tourism industry. First, our results indicate that in choosing domestic destinations, Thai tourists are highly sensitive to price changes. Therefore, it is recommended that the pricing of tourism products and services at TNPs and their neighboring areas be strategically designed. The Ministry of Tourism and Sports (MOTS) could consider pricing intervention by allocating the central budget to fully subsidize the NP entrance fee and the expenditure of local tourists, for instance, through tax deduction incentives (Wang et al., 2022). As anticipated by the MOTS, it could take at least three years before international tourist arrivals and tourism receipts would reverse the pre-Covid-19 pandemic levels. Lowering prices could be an effective way to jump start economic activities driven by the tourism sector by helping to stimulate tourism demand and recover the tourism multiplier effect to re-activate both tourism-related up- and downstream activities, which eventually promote the economic growth of the nation.

Second, we find that WOM recommendations from previous visitors have a significant impact on domestic tourism demand. Since spoken words are highly associated with the overall experience and satisfaction of visitors, MOTS and DNP may consider utilizing big data on tourists to conduct comprehensive data analytics to elevate the image and reputation of TNPs by managing tourist experience, offering personalized marketing and communications, and creating engagement at every touch point throughout the customer journey (Fu & Pulido-Fernández, 2020). In addition to the psychological perspective, significant investment in renovating and restructuring obsolete supply side elements, such as facilities, transportation, lodging, and safety, is needed. Local authorities and relevant businesses must act coherently to support infrastructural upgrading initiatives.

Third, personal safety has a significant impact on destination choice among Thai people. In this context, the safety concerns of domestic tourists are riled up the most by life-threatening crime, followed by property crime. The Royal Thai Police should manage the police patrols comprehensively; expand the coverage area of closed-circuit televisions; and exploit the benefits of the IoT ecosystem by developing mobile applications, smart wearable devices, and smart closed-circuit televisions integrated with facial recognition systems to support crime prevention, reporting, and warning programs. These measures should be carefully implemented to avoid excessive police presence on the street that can cause fear of insecurity. Tourism demand increases when better security measures are put in place. As discussed above, the empirical literature documented evidence of worsening crime with higher volume of tourist inflows. The security measures alone are necessary but not sufficient to deal with the problem. It calls for the involvement of the local community whereby the possibility of any social discord due to high tourist arrivals must be eliminated. The security measures should be supplemented by initiatives to protect and strengthen local community cohesion. The local governments may consider several measures aim at avoiding social disorganization in the local community due to savage tourism such as offering tax incentive for tourism related local enterprises, enhancing employment in the tourism industry for the residents as well as promoting local cultural heritage. These measures can foster ties among the residents and tourists and at the same time realize the economic benefits from tourism that trickle down directly to the local community. The weaving of tourism into the social and economic fabric of the local community will reinforce an informal social control mechanism among the local community to collectively help to prevent crime in the neighborhood. As is widely recognized, a serious crime compounded by media exposure could have a drastic impact on tourism demand. Soft interventions by local authorities may be adopted to maintain a good balance between freedom of news reporting and the media code of conduct. This intervention could help preserve the destination image by reducing exaggeration of crime events and the use of eye-catching photos and headlines.

Finally, we highlight that unsafe conditions perceived by tourists during visitation will be communicated to other potential tourists in subsequent periods through WOM. Therefore, crime prevention policies aimed only at reducing crime rates may not be adequate for avoiding demand stagnation. Collaboration between MOTS and the Royal Thai Police is also crucial for the dissemination of news to the public to elucidate case handling, punishment sentenced, and additional initiatives implemented to deter repetitions of crime occurrence. These communications would re-instill confidence among potential visitors and remove the negative perception of unsafe conditions while allowing them to reevaluate their destination choice, which could eventually alleviate the negative WOM outcome from crime occurrence. Efforts and actions actively pursued by the local authorities to address destination degradations, whether in terms of safety conditions, infrastructure, or supporting industries, would help to sustain attractiveness as well as extend the product life cycles of tourism destinations.

It should be noted that any generalization of the results presented above should be done with caution since the patterns of domestic tourism demand are heterogeneous. Exploring dynamic patterns of domestic demand at regional level or in other countries could be of interest for future research. Our conceptual framework is constructed based upon the crime opportunity and hotspot approaches. Future studies should examine other variables derived from the social disorganization theory that may affect the association between tourism and crime. Likewise, it is also plausible to investigate other factors that dynamically influence tourism demand via WOM recommendations. Finally, several recent studies suggested that local regression methods allow researchers to discover the spatial relationship between tourism and crime. Since tourism demand and crime are spatial data embedded in specific locations, future studies may consider employing local regression methods, such as geographically weighted regression for exploring spatial heterogeneity.

Acknowledgments

We are grateful to the comments and suggestions by the reviewers that improved the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Faculty of Business Administration, Burapha University, under Grant number 015/2564.

Notes on contributors

Karoon Suksonghong

Karoon Suksonghong is Assistant Professor at the Faculty of Business Administration, Burapha University, Thailand. His research interests include tourism demand modelling, carrying capacity management, portfolio selection and application of optimization algorithms in business management and finance.

Kim-Leng Goh

Kim-Leng Goh is Honorary Professor at the Faculty of Business and Economics, University of Malaya, Malaysia. His area of expertise is econometrics. His research interests include economics modelling and projection, and application of statistics in business and economics.