Abstract
Given the limitations in existing resident attitude research, a new concept of resident sentiment is proposed to describe local residents’ overall perceptions of and emotional dispositions toward a dominant tourist market, in which attitude is a constituent part and behavioral response is implied. To operationalize this higher-order latent concept, this study developed measurements for its five components (cognitive and affective attitudes, identification, and two collective mentalities) identified from an earlier exploratory study. An online survey received 1,000 usable responses from Hong Kong residents to validate this construct in a nomological network. The results not only indicate the reliability and validity of the refined scales, but also provide support for resident sentiment as a better indicator of residents’ behavioral responses than attitude. Resident sentiment has the potential for significant use in extending resident attitude studies by academics, as well as being a performance measure for practitioners seeking destination sustainability.
Keywords
Residents’ attitudes toward tourism impacts and tourists in general have been extensively studied since the 1970s given the critical role of resident attitude in positive tourist experiences and sustainable tourism development. However, this line of research has encountered growing conceptual, theoretical, and methodological challenges (Li, Hsu, and Lawton 2015; Sharpley 2014). For example, attitude has been used primarily as a measure of within-group differences rather than within-group similarities in views of the world (Fraser 1994). Thus, the capability of attitudes in addressing the social essence of psychological phenomena is questionable (Howarth 2006). Numerous studies from a sociological perspective have revealed collective attitudes and emotions commonly shared by a given population, such as anti-immigrant/foreigner (Zamora-Kapoor, Kovincic, and Causey 2013) and anti-tourism/tourist sentiment (Seraphin, Sheeran, and Pilato 2018). Increasingly more tourism researchers have recognized that the collective attitudes and/or emotions permeated in a community are more than the sum of the individual members’ attitudes/emotions—they reside in individual minds but are not unique to individuals, and thus may be able to predict the intention and direction of collective actions more accurately than individualized attitudes (Hwang and Stewart 2016; Monterrubio and Andriotis 2014). Yet empirical research on community attitudes beyond an individual perspective is still in its infancy. In addition, the current resident attitude literature tends to frame residents’ support for tourism development as a rational procedure, mostly neglecting the affective dimension of resident evaluation (Hadinejad et al. 2019). As a notable exception, Woosnam and colleagues (e.g., Woosnam and Norman 2010) emphasized the role of residents’ emotional bonds with tourists in their positive perceptions of tourism impact and support for tourism development.
A vital concept in psychology and sociology, sentiment, has yet to enter mainstream tourism research. On the rare occasions when sentiment was mentioned, most likely it was used as a synonym for attitude (Williams and Lawson 2001), emotion (Brown 2015), or shared/public opinions (Alaei, Becken, and Stantic 2017). Drawing upon definitions in social psychology to enhance and expand resident attitude research, Hsu, Li, and Chen (2016) first conceptualized resident sentiment (RS) as an encompassing construct to describe local residents’ overall perceptions/views of and emotional dispositions toward tourism development, in which attitude is only one constituent part and behavior is implied. This conceptualization is well supported by psychological and sociological literature. Cattell (1950) categorized attitude and sentiment under one umbrella named dynamic traits, which encompass motives and mediate between environmental stimuli and behavioral responses. Cattell asserted that attitude, as the smallest unit among dynamic traits, is dependent on the larger, more complex concept of sentiment. Cattell (1940) distinguished sentiment from attitude, defining attitude as “an acquired neuropsychic disposition to react with belief, thought, feeling and overt behavior in a certain way toward a certain object, as part of the purposive plan of some larger sentiment or complex”; by comparison, sentiment represents “an acquired and relatively permanent major neuropsychic disposition to react emotionally, cognitively, and conatively toward a certain object (or situation) in a certain stable fashion” (p. 16). In short, attitude is a narrower, more discrete, and transient construct (Howarth 2006) whereas sentiment is more encompassing, hierarchical, and lasting (Allport 1935).
A core feature of sentiment is that it is environmental-mold—learned over time through experience with people (Ryckman 2008). This notion echoes Homans’s (1947) argument that the elements of social behaviors—sentiment, action, and interaction—are interdependent: sentiment leads to activities and interactions, which in turn modify sentiment. Contrarily, attitude research usually isolates individuals from social environment or takes the environment as a given; ignoring individual–environment interaction hinders exploration of how attitudes are shared or related, how attitudes and identities intersect, and how particular attitudes may affect social relations (Howarth 2006). Sentiment is superior to attitude in that it is inherently social (i.e., shared by many individuals) and thus constitutes a social reality that can affect individual behaviors.
Thus, RS was proposed as an indicator of social representations shared by a community and a more reliable antecedent of residents’ behavioral responses (Hsu, Li, and Chen 2016). However, to the authors’ knowledge, no consensus exists on how to measure sentiment in social science. Psychologists have mainly investigated sentiment via “systematic observation and clinical illustrations rather than by statistical or experimental means” (Cattell 1940, p. 7). Sociological studies generally examine sentiment through naturalistic, descriptive methods and present the findings in a qualitative form (Zamora-Kapoor, Kovincic, and Causey 2013). Recent media research mainly adopts bipolar classification (i.e., positive and negative) to measure media sentiment (Hao et al. 2019). Hence, what other components or indicators are contained in RS besides “attitude” is unknown. The primary purpose of this study is to take an initial step toward operationalizing RS.
Moreover, as Kock, Josiassen, and Assaf (2019) criticized, many scale development studies in tourism lack tests of nomological validity using meaningful causal models. Only theoretical development, coupled with empirical testing of nomological networks, can sufficiently demonstrate the relevance of a developed scale as well as its theoretical soundness and usefulness to explain phenomena and enhance knowledge (e.g., Tsaur, Yen, and Teng 2018). Therefore, the secondary objective of this study is to examine the nomological validity of RS. Although Homans (1947) suggested that sentiment, action, and interaction are inherently interdependent, this interrelationship has not been assessed empirically given the lack of a sound sentiment measure. The present study thus investigates the relationships among RS and associated behavioral responses.
In exploring the components of RS, this study focuses on Hong Kong (HK) residents’ sentiment toward the city’s dominant source market, Mainland Chinese tourists (MCTs), for two reasons. First, RS must be investigated with a specific object. Because MCTs are becoming the dominant source market for increasingly more destinations, such as Thailand and the UK, the measurement of RS toward MCTs will be increasingly important. Second, HK is the first outbound destination to receive MCTs on a large scale, leading HK residents to interact frequently with this market. Thus, HK provides an ideal setting in which to observe resident–tourist interactions and explore resultant community sentiment.
Literature Review
Resident Attitude Studies
Residents’ attitudes toward tourism development have been one of the longest-running and fruitful research areas in the tourism discipline (Sharpley 2014), given the assumption that resident attitudes can lead them to stand for or against such development. However, the predictive power of residents’ attitudes to their behavioral responses toward tourism development or tourist groups remains controversial (Carmichael 2000; Nunkoo and Gursoy 2012). Do resident attitudes actually determine how residents behave? This question remains largely unanswered; few resident attitude studies have considered actual behavioral responses (Sharpley 2014) apart from the behavioral intention variable “support for/opposition to tourism development” (Nunkoo and Gursoy 2012).
Of the few studies that considered residents’ specific behaviors, Ap and Crompton (1993) reported four types of reactions to tourism but did not test the attitude–behavior relationship empirically. Carmichael (2000) did not identify significant linkages between residents’ attitudes toward casino gambling and subsequent actions, but recommended monitoring the pulse of local feelings. Although the attitude–behavior relationship has been recognized as non-deterministic, and the strength of this relationship is influenced by many factors (Joo et al. 2018), inadequate exploration of the relationship remains a glaring weakness in relevant research (Sharpley 2014).
Moreover, tourism scholars have criticized attitude theorists for not conceptualizing the social nature of attitudes (Andriotis and Vaughan 2003; Fredline and Faulkner 2000). The most popular theoretical framework in resident attitude studies, social exchange theory, generally neglects sociocultural inputs that foster these attitudes at a community level (Hadinejad et al. 2019), although it is good at explaining individual attitude formation through a linear calculation of perceived costs and benefits. Other common frameworks, such as social representations theory and identity theory (Nunkoo and Gursoy 2012), have also received criticism. Although social representations theory highlights the effects of social interactions on individuals’ attitude formation (Fredline and Faulkner 2000), it neglects individual capability and cannot explain why a particular perception is commonly held (Sharpley 2014). Furthermore, no studies of resident attitude have incorporated the interactive and mutually constitutive relationship between the “individual” and the “social” that is central to the theory of social representations. Little is therefore known about how resident attitudes are shared within a community, how certain attitudes relate to one another, and what the relation is between attitudes and identities.
Chen, Hsu, and Li (2018) introduced social identity theory into host–tourist relation research and demonstrated its applicability through qualitative findings. This theory was also adopted as the guiding framework in the present study, to bridge the gap between social exchange and social representations theories by highlighting individual residents’ community membership; and to facilitate the understanding of community mentalities that belong to each member but transcending the sum of individual attitudes. In addition, employing social identity theory to examine resident sentiment toward a dominant tourist group is responding to the call for more affective assessment of resident attitude, because identity has been viewed as a psychological attachment to a specific group that connotes cognitive, evaluative, and emotional significance (Nunkoo and Gursoy 2012).
Sentiment as a Concept
Sentiment is a widely examined but ambiguously defined concept in disciplines including psychology, sociology, and information sciences. Most empirical studies simply adopted “sentiment” as a default term synonymous with “emotion” or “attitude” (Cattell 1940), without defining it strictly. Psychologists initially defined sentiment as an enduring, complex, organized system of dispositions to generate certain emotions under certain circumstances; this mental system can adapt itself emotionally to the changing situations of its object and persist indefinitely in correspondence with its object’s duration without becoming morbid (Shand 1922). McDougall (1923) discussed three full-grown sentiments, love, hate, and respect, viewing them as systems composed of enduring dispositions to experience emotions such as joy, gratitude, fear, anger, or shame whenever the loved, hated, or respected object comes to mind. Sentiment was later expanded beyond affective facets and defined as a more general, overarching, and complex system than attitude in social psychology, in which “a cognitive disposition is linked with one or more emotional or affective and conative dispositions to form a structural unit that functions . . . as one configuration or Gestalt” (McDougall 1923, p. 437). This intricate mental structure is thought to underlie all mental activities toward a certain object. In sociology, sentiment is conceptualized as macro-social mentalities combining emotions, social cognition, value orientations, and behavioral intention that disperse throughout society over a period of time (Wang 2013), leading to universal and consistent psychological characteristics and behavioral patterns shared by group members, which then influence individual behaviors (e.g., patriotic sentiment).
Given that sentiment is such an abiding and complex system of cognitive, emotional, and conative dispositions centered on certain objects, analyzing sentiment can be a challenge. Observing the conative tendencies of complex emotions is a potential approach (Shand 1922), as every emotion has its “characteristic conjunction of motor tendencies, which together give rise to the characteristic attitudes and expressions of the emotion” (McDougall 1923, p. 132). A systematic analysis of the myriad emotional dispositions within a sentiment system can reveal the underlying, stable sentiment toward an object. This backtracking method illustrates the determinant role of sentiment in attitudinal, emotional, and behavioral tendencies. For instance, HK residents’ self-reports normally include both positive and negative attitudes toward Mainlanders (Chen, Hsu, and Li 2018), precluding an explanation of why they oppose policies appealing to MCTs. Thus, examining their general sentiment toward MCTs may be able to clarify such resistance. As Cattell (1940, p. 11) mused, “Attitudes are twigs on the conative tree, of which the larger branches are sentiments. The attitude vectors differ from the sentiment vectors only in lesser length and, probably, in greater angular dispersion. The attitudes are architectonically dependent on the sentiment.” Because this is the first study introducing the concept of RS into tourism research, this article focuses on examining the structure of RS and testing its ability to predict residents’ behavior.
Resident Sentiment Construct
Following Jarvis et al.’s (2003) decision rules, this study conceptualized RS as a higher-order formative construct composed of multiple lower-order latent variables, including attitude and other undetermined components. In this case, causality flows from these components to the unmeasurable latent construct of RS. The components of RS are not interchangeable—they do not necessarily co-vary, nor do they necessarily have the same antecedents and consequences. To develop measurement for a newly conceptualized construct and test it within a nomological network, we followed steps suggested by Churchill (1979) and abided by Howell, Breivik, and Wilcox’s (2013) recommendations about modeling formative indicators as separate constructs. An exploratory qualitative study was first conducted to capture RS’s content domain. In-depth interviews with the HK community identified five potential components of RS toward MCTs (Chen, Hsu, and Li 2018), falling under three main concepts: attitudes, identification, and mentalities, as described in the following subsections.
Resident attitude toward a dominant tourist market
Few studies have recognized the function of resident attitude toward a dominant source market in predicting residents’ overall attitudes toward tourism development, although contact theory is widely advocated and positive interactions with tourists are considered a reliable indicator of residents’ support for tourism (Joo et al. 2018). Residents’ perceptions of tourists from a specific country have only received limited attention from stereotype researchers who are more concerned about the stigmatized tourist experience than hosts’ experiences and reactions (Moufakkir 2015). Research into how residents’ interactions with and impressions of a specific tourist group affect their overall sentiment toward tourism development is curiously rare (Joo et al. 2018; Sharpley 2014).
The classic tripartite model of attitude structure posits that attitude consists of cognitive, affective, and conative components (Breckler 1984), among which the cognitive and affective components have received the most attention from tourism scholars (Monterrubio and Andriotis 2014). Residents’ cognitive attitude (CA) can be defined as their knowledge/perceptions/impressions of tourists, formed through direct personal experiences or secondary information sources. Affective attitude (AA) represents individual residents’ feelings about tourists. CA is often considered an antecedent to AA.
For the current research, HK residents’ CA toward MCTs was mainly derived from direct or indirect (e.g., through media or acquaintances) contact with MCTs. Many studies have revealed HK residents’ unfavorable impressions of MCTs (Siu, Lee, and Leung 2013), mainly due to MCTs’ inappropriate manners, habits, and customs (Zhang, Pearce, and Chen 2019). Residents therefore often reported negative feelings toward MCTs, such as dislike, anger, and disregard (Siu, Lee, and Leung 2013), with limited positive emotions associated with their appreciation of MCTs’ strong purchasing power or sporadic positive interactions (Fan et al. 2017). Positive relationships among HK residents’ CA and AA, as well as their behavioral responses toward MCTs, can be expected per triadic consistency theory (Norman 1975).
Resident identification with tourists
Social identity theory suggests that group identification is an indispensable variable in intra- and intergroup studies (Jackson 2002; Leach et al. 2008) because of its capability to predict in-group favoritism or collective attitudes and behaviors such as group commitment and emotional solidarity (e.g., Woosnam 2012). However, in-group members are also likely to demonstrate positive attitudes toward out-group members if they view themselves as members of a more inclusive, superordinate common group instead of as two separate groups (Gaertner, Dovidio, and Bachman 1996) or intergroup contact is cooperative and pleasant (Brewer 1996). Briefly, in-group favoritism or out-group bias can be reduced by intergroup cooperation or cognitive one-group representation.
Tourism has been shown to promote intergroup communication and understanding (Fan et al. 2017), with frequent host–tourist interactions potentially creating a new “we.” As Weinreich (1983) indicated, when one identifies with newly encountered individuals, one broadens his or her value system and establishes a new context for his or her self-definition, thereby initiating reappraisal of the self and others. However, the few tourism studies examining resident identity have mainly focused on their regional or cultural identities without considering the identification with tourists (e.g., Palmer, Koenig-Lewis, and Jones 2013). Notable exceptions are emotional solidarity studies by Woosnam and colleagues (e.g., Joo et al. 2018; Moghavvemi et al. 2017); they introduced emotional solidarity theory into the tourism literature and developed and tested a related measurement scale in multiple settings. Despite conflicting results (e.g., Moghavvemi et al. 2017), scholars generally agree that residents can develop a degree of emotional solidarity with tourists based on shared beliefs/behaviors and close interactions. Emotional solidarity can be thought of as a sense of identification with others, whereby a sentiment of “we together” is championed over the “self–other” dichotomy (Woosnam 2012). While residents’ emotional solidarity with tourists is often tied to positive perceptions of and support for tourism development, not all residents develop emotional closeness with tourists. Ye et al. (2014) noted that commonalities in residents’ and tourists’ social identities could shape residents’ attitudes toward tourists and tourism development; however, this result has not been verified quantitatively. This study aims to fill this gap by examining how residents’ identification with tourists determines their behavioral responses.
Within this study context, an unambivalent group identity, “Hongkongers,” has become increasingly apparent among the host community because of frequent intergroup comparisons made with MCTs (Chen, Hsu, and Li 2018). HK residents’ efforts in maintaining this in-group identity are reflected in claims of intergroup differences rather than an emphasis on in-group similarities. This intergroup distinction is rooted in HK’s colonial history. Before the return of its sovereignty to China, HK residents maintained a Chinese identity as a distinction from British colonizers; after its return, the regional HK identity has become increasingly salient in differentiating themselves from Mainlanders because of HK’s stronger economic power (Hong et al. 1999). However, the superior HK identity has faced unprecedented challenges in recent years as a result of the Mainland’s rising economic status and perceived threats. Such challenges have prompted more Hongkongers to reconsider their group identity and their identification with Mainlanders, which will likely alter their attitudinal and behavioral responses to MCTs and perhaps even their support for tourism development associated with this dominant tourist market. A positive relationship is thus proposed between residents’ identification with tourists and their behavioral responses.
Mentalities generated from social comparisons with tourists
Social identity theory posits that people’s instinctive need for a positive self-identity motivates social comparisons that favorably differentiate in-group from out-group members (Tajfel and Turner 1979). When facing an outsider/tourist group, residents may engage in upward and downward comparisons for distinct reasons: upward comparison (with superior others) is inspired by self-evaluation and self-improvement, whereas downward comparison (with inferior others) is driven by self-enhancement. Upward comparison often results in negative affective reactions while downward comparison leads to positive ones (R. H. Smith 2000). However, Buunk et al. (1990) revealed that either comparison direction can generate various emotions, indicating that the direction does not necessarily predict people’s emotional and behavioral responses toward outgroups. Understanding the more enduring and stable sentiment triggered by intergroup comparisons may more accurately predict individuals’ specific emotional and behavioral responses.
A feeling of relative deprivation can be generated from residents’ self-assessments of their own community’s condition by comparing with the tourists’ (Seaton 1997). Specifically, material and income deprivation felt by residents can be induced by contact with tourists, tourist provisions, and tourism employment. Sense of relative deprivation can also be produced by comparisons against past living conditions and social justice philosophy (Peng, Chen, and Wang 2016). H. J. Smith et al. (2012) stated that feeling of relative deprivation, which often follows from upward comparison, can also serve as motivation affecting further judgments that one’s in-group is disadvantaged compared to a reference group. Various emotions may thus be evoked regarding the relatively advantaged group (e.g., admiration, envy, entitlement, anxiousness, depression, anger, and resentment), which can engender different behavioral responses such as assimilation, hostility, or resistance (H. J. Smith et al. 2012). In tourism research, group relative deprivation has been found to promote in-group serving bias and foster out-group prejudice, thus facilitating the prediction of collective actions such as social protests (Seaton 1997).
Residents’ downward comparison with inferior or disliked tourists may help them compensate for a feeling of deprivation and sustain/enhance self-esteem and superiority (Wills 1981). Superiority is an indicator of “narcissism” in psychology or a dimension of “nationalism/collective narcissism” in sociopsychology. Collective superiority has been deemed effective in protecting group esteem and a weak or threatened social identity (Golec de Zavala et al. 2009). Westburgh (1936) defined sense of superiority as a simple sentiment consisting of systematized beliefs and emotions to certain objects (e.g., persons, activities, and relationships), noting that the superiority–inferiority sentiment can be transferred to others or all problems presented by the environment. Once a superiority sentiment is formed, the emotions and characteristic reaction pattern it invokes (e.g., arrogance, contempt, or compassion) may be assigned to any objects through accidental association (Westburgh 1936).
In this study, two intertwined mentalities were identified permeating the HK community toward MCTs: a sense of group superiority and a feeling of relative deprivation (Chen, Hsu, and Li 2018). These contradictory mentalities were derived from residents’ conscious or subconscious comparisons with MCTs, as representatives of the HK community rather than as individuals. According to Westburgh (1936), it can be speculated that once residents form a complex sentiment featuring superiority and relative deprivation feelings, their emotional and behavioral responses to all issues related to MCTs may align with that sentiment.
Based on the above literature, a conceptual model is proposed (see Figure 1) to depict causal relationships between the five-component RS and its behavioral responses. The predictability of cognitive and affective attitudes to behavioral responses is supported by attitude–behavior theory; the predictability of the other three RS components to behavioral responses is grounded in the function of sentiment recognized by psychologists. Pear (1922) stated that the accurate predictive power to behavior is a prominent feature of sentiment. Shand (1922) explained that because sentiment embodies a system of several emotional dispositions connected with a common object and subordinated to a common end, each emotional disposition may have unique conative tendencies, leading to behavior-related predictability varies in direction and intensity.
Conceptual model.
Methodology
Measurement Development
A questionnaire was developed based on the literature and in-depth interviews (N = 39). Thirty-five cognitive attributes related to MCTs were extracted from interview results to measure HK residents’ CA toward MCTs. Measurements of AA were drawn from the attitude literature (Batra and Ahtola 1991). Respondents were asked to evaluate their affective impressions of MCTs on a 7-point bipolar scale along with three pairs of semantic differential items.
Established measures of group identification in social psychology focus on individuals’ identification with an in-group rather than an out-group (e.g., Ellemers, Kortekaas, and Ouwerkerk 1999); adaptation was thus made on the five selected items (Ellemers, Kortekaas, and Ouwerkerk 1999; Leach et al. 2008) to suit the context of this study. Respondents were asked to rate their level of identification with MCTs, an out-group visiting their community. No established measurement was identified for sense of superiority, whether at the individual or collective level; thus, six items were developed from interview results while referring to two dimensions of Golec de Zavala et al.’s (2009) collective narcissism scale (i.e., superiority and entitlement). Respondents were asked to think about their own community and indicate the extent to which they believed certain aspects were superior to MCTs’.
Given the scarcity of tourism research on resident sense of deprivation, four items were developed by referring to established measurements of relative deprivation in social psychology (Osborne, Sibley, and Sengupta 2015). Respondents were asked to compare the current pecuniary situation of HK with that of MCTs and indicate the extent to which they felt being deprived as a community. Seven-point Likert scales, with 1 = strongly disagree to 7 = strongly agree, were used to assess all the above items.
Finally, residents’ behavioral responses to MCTs (4 items) and support for tourism development (3 items) were used to operationalize the behavioral consequences of RS. Tourism Impact Attitude scale (Lankford and Howard 1994) and Williams and Lawson’s (2001) tourist preference statements were referred when phrasing items. A 7-point Likert scale, where 1 = extremely unlikely to 7 = extremely likely, was used.
Data Collection and Analysis
The questionnaire was prepared in HK’s two official languages, traditional Chinese and English. Back-translation was used to ensure accuracy and equivalence of the Chinese questionnaire. Four experts assessed items’ face and content validity. Based on their feedback, the questionnaire was modified and pilot tested with a convenience sample of 199 HK residents to examine the scales’ content validity and reliability. All items demonstrated satisfactory reliability (all reliability coefficients exceeded 0.7) and were thus retained for the formal survey.
Formal data collection was conducted online in Nov 2016, by a research company who sent e-mail invitations to 40,000 HK permanent residents. A quota/criteria control was employed to obtain a representative sample in terms of age, gender, and household income. Ultimately, 1,000 usable responses (890 Chinese and 110 English) were obtained. Data were imported into SPSS.25 and AMOS.25 for analyses.
Data analyses started with an exploratory factor analysis (EFA) on CA to reduce the number of measurement items and identify underlying dimensions, followed by a confirmatory factor analysis (CFA) to test whether the underlying structure could form a tenable measurement model for CA (DeVellis 2003). The sample was randomly split into two halves, one serving as a calibration sample (n = 487) for EFA and the other as a validation sample (n = 513) for CFA. The multidimensional measurement model of CA derived from EFA and CFA was then cross-validated using the entire sample (N = 1,000). The reliability and validity of other unidimensional constructs were evaluated by CFA directly using the whole sample. Covariance structure analysis was then conducted to assess RS components and behavioral responses.
Results
Respondent Profile
The profile of respondents in Table 1 demonstrates a representative sample of HK permanent residents in terms of age, gender, residential area, and household income (Census and Statistics Department 2017). Most respondents (85%) did not receive any income from tourism industries. Only 4.2% had never visited the Mainland. Nearly 90% identified themselves as primarily Hongkongers. Moreover, respondents reported infrequent communication with MCTs in general (M = 3.42) but discussed MCTs with other HK residents more often (M = 4.12). Among the 456 respondents with an above-average contact frequency, 92% communicated with MCTs in a work setting and 83% lived in tourist areas. Respondents frequently obtained information about MCTs from news and social media (M = 4.36) and acknowledged the influence of the media on their feelings about MCTs (M = 4.63).
Respondent Profile (N = 1,000).
Purification of Cognitive Attitude Measures
EFA results of cognitive attitude
Table 2 presents factor analysis results for residents’ CA toward MCTs. The KMO measure of sampling adequacy (0.95) and Bartlett’s test of sphericity (prob. < 1%) indicated that factor analysis would be appropriate. Factors were extracted using the principal components method with oblique rotation to allow for correlations among factors. Four factors were retained based on two criteria (Field 2018), namely scree plot examination and Kaiser’s criterion (i.e., retaining factors with eigenvalues greater than 1), explaining 56.2% of the total variance. Five items were excluded because of low or cross-loadings. Factor loadings greater than 0.35 were considered statistically significant based on a sample size of 487 respondents (Hair, Black, and Babin 2010). All factors demonstrated satisfactory scale reliability (Cronbach’s alpha ≥ 0.6) (Hair, Black, and Babin 2010).
EFA Results for Cognitive Attitude (n = 487).
Note: EFA = exploratory factor analysis. Principal components analysis. Rotation method: Promax with Kaiser normalization.
KMO measure of sampling adequacy = .949.
Validating the Higher-order Structure of Cognitive Attitude
The measurement structure of CA derived from the EFA was verified and refined by a second-order CFA with maximum likelihood estimation using the validation sample. The initial hypothesized factor model of CA (model 1; see Table 3) comprised four latent factors with 30 manifest variables, and each pair of latent factors could co-vary.
Fit Indices for Initial, Alternative, and Refined CFA Models of Cognitive Attitudes.
Note: CA = cognitive attitude; CFA = confirmatory factor analysis; GFI = goodness of fit index; AGFI = adjusted goodness of fit index; NFI = normed fit index; TLI = Tucker–Lewis index; CFI = comparative fit index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual. Models 1 and 2 were tested using the validation sample (n = 513), while model 3 was tested using the full sample (N = 1,000).
Because of a lack of theoretical agreement, the CA model was re-specified to achieve better parsimony and fitness based on standardized residual covariance, squared multiple correlations (SMCs), modification indices, and expected parameter change statistics (Hatcher 1994). The respecified model (model 2) exhibited satisfactory goodness-of-fit indices; however, the average variance extracted (AVE) for two factors, Comparative Traits (31%) and Wealth (45.6%), fell below the recommended 50% threshold (Hair, Black, and Babin 2010) and thus did not demonstrate convergent validity. Also, the correlation between Negative Impressions and Comparative Traits (r = .92) was above the recommended criterion of 0.8. Before making a hasty decision to drop these components for better statistics, the higher-order structure of CA was retested using the whole sample (N = 1,000) to establish a tenable measurement model for the construct, because a large sample size would ensure relatively stable estimation for complicated models.
With the full sample, Comparative Traits and its indicators were removed from the model due to extremely low SMC values (<3.0) and low factor loadings (<5.0). To obtain a more parsimonious model, one item from Positive Impressions and five items from Negative Impressions were removed because of low contributions to their respective latent factors or due to cross-loadings. The refined three-factor CA measurement model (model 3; see Table 3) fit the data well. As shown in Table 4, composite reliabilities (all greater than 0.60) reflected the internal consistency of items in each construct (Bagozzi and Yi 2012). The critical ratio and AVE values lend additional evidence of convergent validity for CA and its three components. Discriminant validity was established because AVE values far exceeded the squared correlations between any pair of components.
Revised CFA Model of Cognitive Attitudes.
Note: CFA = confirmatory factor analysis; AVE = average variance extracted. CFA modification on CI measurements, with the full sample (N = 1,000).
All loadings are significantly above 0.5 (p < .001).
AVE values were shown as decimals for ease in comparison with the squared correlations.
Values shown below the diagonal are correlation estimates (at two-tailed p level of .001).
Values above the diagonal are the squared correlations.
Next, to reduce the second-order CA to a first-order variable, the mean of two indicators of Wealth was computed to represent this component, while two one-factor congeneric models were tested for Positive and Negative Impressions of MCTs, each having multiple indicators. By fitting one-factor congeneric models, factor score weights were employed to compute composite scores for these two CA components. The benefits of weighted composites include minimizing measurement error in items and increasing the reliability and validity of computed scores (Holmes-Smith and Rowe 1994; Table 5). Discriminant validity was achieved because AVE values were greater than the square of the correlations between the three components. Thus, CA can be considered a first-order latent variable accounting for the covariance of three reliable and valid composite variables, which were used as indicators in subsequent analyses.
Correlations, Means, Standard Deviations of Three CA Facets.
Note: CA = cognitive attitude. N = 1,000.
Coefficient H is the maximized reliability coefficient developed by Hancock and Mueller (2001).
Values above the diagonal are the squared correlations.
The scale: 1= Strongly disagree, 7 = Strongly agree.
Values shown below the diagonal are correlations.
Correlation is significant at the p level of .01 (2-tailed).
Overall Measurement and Structural Models
CFA was first employed to assess the reliability and validity of construct measures. The CFA resulted in the elimination of one item each from CA, “support tourism development in HK” (Support-TD), and “identification” (ID) with MCTs due to low factor loadings or SMC values. Multiple fit indices indicated that the revised measurement model fit the data well (χ²/df = 3.04, p = .000; goodness of fit index [GFI] = 0.942; adjusted goodness of fit index [AGFI] = 0.924; normed fit index [NFI] = 0.95; comparative fit index [CFI] = 0.966; root mean square error of approximation [RMSEA] = 0.045; standardized root mean square residual [SRMR] = 0.048).
As listed in Table 6, the composite reliabilities (all greater than 0.7) reflect the internal consistency of items in each construct. Convergent validity can be claimed because the minimum standardized factor loading (= 0.57) exceeded the 0.5 threshold, and the minimum critical ratio (CR = 17.13) was much higher than the threshold of 2 (Hair, Black, and Babin 2010). Moreover, AVE scores were mostly above 50% except for “sense of superiority” (SUP, 0.44) and “feeling of relative deprivation” (RD, 0.47). Netemeyer, Bearden, and Sharma (2003) indicated that for new scales, a low AVE value near the threshold of 0.45 could be reasonable and acceptable. These two variables were thus retained due to the exploratory nature of this study.
CFA Results for All Constructs.
Note: CFA = confirmatory factor analysis; AVE = average variance extracted; MCTs = Mainland Chinese tourists. CFA on all constructs’ measurements, with the full sample (N = 1,000).
All loadings are significantly above 0.5 (p < .001).
AVE values were shown as decimals for ease in comparison with the squared correlations.
Values shown below the diagonal are correlation estimates (***at 2-tailed p level of .001; ** at p level of .05).
Values above the diagonal are the squared correlations.
AVE values for any pair of factors were compared to the square of the correlation between these factors, substantiating the discriminant validity of AA, SUP, and RD. The discriminant validity of CA and ID was questionable; their AVE values were much lower than their squared correlations, and marginally high correlations were observed between CA, ID, and AA (>0.83). However, the more rigorous chi-square difference test showed that CA was still significantly different from ID. A multicollinearity test was also performed, and low variance inflation factors (ranged from 1.45 to 2.64) indicated no multicollinearity threat (Hair, Black, and Babin 2010).
The next step was to examine the hypothesized structural model (i.e., Figure 1), including 10 gamma paths between the five predictor variables and the two dependent variables. Covariance between CA, AA, and ID was allowed due to high correlations between these constructs. This baseline model did not display an optimal fit (χ² = 854.62, df = 253, p = .000; χ²/df = 3.38; GFI = .937, AGFI = .919, NFI = .943, TLI = .952, CFI = .959, RMSEA = .049, SRMR = .095) according to the stringent cutoff values adopted here (Hair, Black, and Babin 2010; Hu and Bentler 1999). The path diagram is illustrated in Figure 2.
Baseline model with standardized paths.
The SMC (R²) indicated that RS explained only 14.3% of the variance in Support-TD but 79% of the variance in Behavioral Response to MCTs (Beh-Resp). Path coefficients revealed that neither CA nor AA exerted a significant effect on both dependent variables (p > 0.05), which was surprising because the attitude–behavioral intention relationship has been widely confirmed (Palmer, Koenig-Lewis, and Jones 2013). Contrarily, ID had a dominant positive influence (β = .94) on Beh-Resp. A dominant exogenous variable leading to nonsignificant coefficient estimates of the other predictors is a telltale sign of multicollinearity, which may also lead to “wrong” coefficient signs and unstable parameter estimates (Grewal, Cote, and Baumgartner 2004). To mitigate the effects of multicollinearity, alternative models were tested by considering other potential factors that may influence the accuracy of estimation results (Mason and Perreault 1991). This approach can make a case for the most defensible solution and ensure a reasonable interpretation (Marsh et al. 2004).
As shown in Table 7, to test the statistical significance of the difference between the path coefficients of CA, ID, and AA and perhaps resolve the multicollinearity issue, six paths in the alternative model 2 (i.e., between the three highly correlated exogenous variables [CA, ID, and AA] and two dependent variables) were constrained as equal. If their path coefficients do differ significantly, then the baseline model (i.e., model 1) should fit the data significantly better than model 2. Results indicated that the difference in chi-squares for the two models (∆χ² = 417.3, ∆df = 7) was statistically significant (p < 0.001). Other fit indices showed that the fit of model 2 was worse than that of model 1, and the correlations between CA, ID, and AA were the same in both models. model 2 was thus rejected.
Structural Equation Modelling Results (N = 1,000).
Note: β: standardized regression weight; CR = critical ratio; GFI = goodness of fit index; AGFI = adjusted goodness of fit index; NFI = normed fit index; TLI = Tucker–Lewis index; CFI = comparative fit index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual.
p < .05, **p < .01, ***p < .001.
In model 3, a second-order factor was proposed as a highly abstract representation of CA, AA, and ID; thus, these three lower-order factors were construed as three correlated dimensions of the higher-order factor (Bagozzi and Yi 2012). Causal paths were added between this factor and two dependent variables. Fit indices demonstrated that this model fit the data well. High factor loadings and low error variances further indicated that CA, AA, and ID captured the proposed abstract factor well. However, this higher-order factor is not a real, justified concept with literature support; the tripartite attitude structure consists of cognition, affect, and conation (Breckler 1984) rather than identification. Therefore, model 3 was not an ideal choice although its fit indices were better than those in the baseline model.
Model 4 was proposed by exploring the potential causal relationship between the five exogenous variables. As Shieh (2010) reminded, misconceptions of multicollinearity may hinder the detection of interaction effects. Inter-correlations among predictor variables do not always indicate a multicollinearity problem that should be mitigated by dropping variables, and advanced methodology should be proposed to “fully account for the intertwined structure in order to help advance social science theory” (p. 485). This proposition is echoed by Kline’s (2015) argument that mediation relationships are often overlooked when multicollinearity exists between a causal variable and mediator. Therefore, potential causal relationships between the five RS components were explored by considering the construct’s definition, interview results, and evidence from the literature. The recognized hierarchy model of cognition–affect–conation (Lutz 1977) suggests that cognition serves as the primary activator in individual response process. In this context, impressions of MCTs’ salient features provided HK residents a cognitive foundation upon which feelings and identification could be developed. Therefore, model 4 was proposed by adding causal paths between CA and the other four RS components.
After several rounds of modification by dropping nonsignificant causal paths based on modification indices, full mediation effects were identified among the five RS components (see Figure 3): ID and AA served as full mediators between CA and the dependent variables, while SUP and RD served as full mediators between AA and the dependent variables. No partial mediation effect was identified; because once the mediators (AA, ID, SUP, and RD) were controlled, the effects of CA and AA on the dependent variables became nonsignificant (as shown in model 1). In other words, CA and AA did not exert direct influences on behavioral responses but exhibited indirect effects through the other three components. In addition, a significant positive causal path was identified between ID and RD. As shown in Table 7, causal paths from the five RS components to Beh-Resp were all significant at p = .001, and Beh-Resp exhibited a high R2 value of 0.8 (i.e., 80% of the variance in Beh-Resp could be explained by the five RS components). Conversely, only 13% of the variance in Sup-TD could be explained by CA, AA, and SUP.
Final model with standardized paths.
Finally, to examine whether RS has a stronger predictive power than the attitude construct to behavioral responses, an attitudinal–behavioral response model was tested by removing the three newly identified RS components (i.e., ID, RD, and SUP); only CA and AA were retained as predictors of dependent variables. This model demonstrated a moderate fit (χ² = 279.4, df = 54, p = .000; χ²/df = 5.18; GFI = 0.960, AGFI = 0.932, NFI = 0.968, TLI = 0.963, CFI = 0.974, RMSEA = 0.065, SRMR = 0.068). All causal paths were significant at p = .001 (Figure 4), with AA as a full mediator between CA and Sup-TD. While 67% of the variance of Beh-Resp could be explained by CA, only 3% of the variance of Sup-TD could be explained by CA and AA, far below the variance explained by the five-component RS. Therefore, sentiment was found to be a better predictor than attitude in explaining residents’ behavioral responses to tourists and tourism development.
Resident attitude–behavioral response model.
Discussion and Conclusion
Resident Sentiment
The aim of this study is to operationalize RS, a formative construct including five components. Following a scientific scale development procedure, measurements were developed for all five components and validated in a structural model with consequence variables. RS was not calculated as a composite of its components, because a composite cannot explain more variance in a dependent variable than regression based on the components of the composite (Howell, Breivik, and Wilcox 2013). Noteworthy differences were identified among the five components’ predictability of the dependent variables.
Specifically, HK residents’ CA toward MCTs was reflected in the perceived positive and negative behaviors/features of MCTs, supporting Chen, Hsu, and Li’s (2018) results. However, different from prior resident attitude research results (e.g., Ribeiro et al. 2017), no significant direct causal effects were identified from either CA or AA to behavioral responses. The causal influences of HK residents’ CA on behavioral responses were fully mediated by their identification with the tourist group and all affective components of RS (i.e., AA, RD, and SUP). This pattern substantiates Norman’s (1975) belief that emotions, especially those derived from personal experiences, are essential driving forces to transfer cognitions into actions. The final model also affirms social identity theory, which suggests social groups offer the characteristics that define members’ self-concept by providing normative values and emotional attachment associated with group membership (Tajfel and Turner 1979). In this context, once HK residents increase their identification with MCTs, their self-concept may encompass a greater proportion of Chinese identity that is shared by MCTs; subsequently, the emotional and evaluative significance attached to the Chinese will be enhanced. CA provides cognitive materials and principles for HK residents’ identification process (Turner et al. 1987).
Similarly, the causal effect of HK residents’ AA on their behavioral responses was entirely mediated by two contradictory mentalities widely shared by HK community members. Even so, this pair of ambivalent mentalities influenced residents’ behavioral responses differently. Sense of superiority was the only direct predictor of HK residents’ conditional support for tourism development, namely by diversifying the tourist population and controlling the number of MCTs. This positive relationship implies that the stronger HK residents’ sense of superiority over MCTs, the more likely they are to support the tourism development strategy of attracting visitors from other markets. However, a strong sense of superiority over tourists does not necessarily indicate an opposition to interacting with them; on the contrary, in this study, a weak positive relationship was uncovered between SUP and Beh-Resp to MCTs. Perhaps HK residents’ strong collective confidence underlying their sense of superiority inspired this counterintuitive result (Hong et al. 1999).
Conversely, the weak negative relationship between RD and Beh-Resp suggests that the stronger HK residents’ feeling of relative deprivation compared to tourists, the lower residents’ desire to interact with them; this trend supports our postulation. Yet HK residents’ feeling of relative deprivation did not influence their support for tourism development because perceived deprivation was caused by MCTs, not all tourists. Additionally, an unanticipated causal relationship was observed between residents’ identification with MCTs and their feeling of relative deprivation. This finding can be explained by relative deprivation theory: the stronger one’s identification with an out-group, the fewer differences will be perceived between the in- and out-group; as such, people may assume that the magnitude of rewards received by these two groups should be the same. When the difference in reward magnitude is perceived as noticeably large or illegitimate, the disadvantaged group will feel deprived and treated unfairly (Seaton 1997). Thus, HK residents may firmly believe they deserve the same, even better, economic conditions than MCTs, given HK’s more democratic and civilized social system compared to the Mainland.
Contributions to Knowledge and Implications
The major contribution of this study comes through clarifying the structure of RS and developing measurements to assess its formative components, which can be employed by policy makers, destination marketers, and researchers in future work. This contribution is meaningful for several reasons. First, constructing and evaluating a formative RS enables an abstract, unmeasured concept to be measured in its broadest sense, which facilitates comparison of RS across destinations or tracking RS of one destination through its tourism development trajectory. Hair, Black, and Babin (2010) stated that construction of a formative index is valuable when a concept is not theoretically established. The proposed RS scale can be adapted to examine various communities’ sentiment toward particular tourist groups and overall tourism development. RS may have stronger explanatory power and wider applicability than resident attitude or emotional solidarity because it can accommodate positive and negative attitudinal and emotional dispositions held by individuals and groups.
Second, this study presented a scale development and validation procedure that is somewhat unique from previous studies. The conceptualization of RS provides a feasible framework to facilitate its operationalization, thereby avoiding the controversial MIMIC model measuring formative constructs (Lee, Cadogan, and Chamberlain 2013). To validate RS scale in a nomological network, an initial exploration and identification of potential intercorrelations among predictors of a formative construct was conducted; this approach has been suggested but rarely practiced by tourism researchers.
Third, relevant dependent variables were incorporated into the scale development to test the nomological validity of RS. This study is the first to examine the causal relationship underlying RS and conative tendencies, offering empirical support for the construct’s utility in explaining residents’ behavioral responses to tourists and tourism development. As Kock, Josiassen, and Assaf (2019) advised, integrating a new scale with other proper variables demonstrates a level of sophistication and theoretical consideration, which rarely found in traditional scale development research. Linking the new RS scale to dependent variables can integrate the new concept and its scale into existing research findings comprehensively, and thereby more convincingly introduce the new concept to relevant fields (Kock, Josiassen, and Assaf 2019). The structural model testing results can also produce more intriguing implications for managers and tourism researchers.
The findings of this study also suggest practical implications for destination management organizations. First, a formative construct is a combination of multiple domains representing a latent construct that is formed throughout a process, thus has a higher degree of practical implication for managerial purpose (Coltman et al. 2008). For example, the relative magnitudes of direct and indirect effects may have distinct treatment or policy implications. In this study, ID was identified as the key mediator between CA and Beh-Resp toward MCTs, indicating that efforts to boost HK residents’ identification with MCTs may improve their positive behavioral intentions more than altering their impressions. Second, measuring and evaluating RS on a continued basis will enable destination managers to timely adjust strategies and regulate tourism development within the psychological bearing capacity of the host community.
Limitations and Future Research
The conceptualization and operationalization of RS remain preliminary. Our findings are focused on a specific community facing a dominant source market, such that the five identified components are associated with a specific tourist group. These results may thus not be applicable to destinations and communities with different tourism development scenarios. However, as the fastest-growing source market in the world, MCTs have remained the worlds’ top spender in international tourism since 2012 and become the dominant source market for increasingly more destinations (Arlt 2016). Therefore, other communities facing the same dominant source market can benefit from this study by obtaining a deeper insight into the resident-tourist relations and a tool to assess their residents’ sentiment toward MCTs. Researchers and practitioners are encouraged to replicate and expand the RS study in other destinations to further validate domains and measurement of RS, or to identify RS unique to their communities. When designing RS questionnaire applicable to other communities, relevant literature needs to be considered. For instance, Woosnam’s (2012) emotional solidarity scale could be incorporated when examining host communities that express positive sentiment toward tourists.
This study only incorporated consequences of RS to verify its nomological validity. Subsequent studies could explore the various antecedents of RS, because the five constituent components may have distinct antecedents. Other intervening factors, such as personal experience with tourism, social interactions with other residents, and perceived media sentiment, could also be examined. Finally, RS is shared among a community and is not unique to individuals; Fraser (1994) recommended that social representations be investigated via numerous externalized and institutionalized channels (e.g., books and mass media) and through social, legal, and religious practices and codifications. Therefore, future RS studies could adopt mixed research methods along with a media content analysis to produce new knowledge and advance theoretical development.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The work described in this article was fully supported by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China. (Project No.: PolyU 155024/14B)
