4.2 Tourist satisfaction and preferences
Hypothetical relationships between tourist satisfaction and preferences were tested by three methods. First of all, the whole sample was subdivided according to a single scale of tourist satisfaction, and each segment was estimated by using the RPL model. Secondly, the four constructs of tourist satisfaction were incorporated into the evaluation model by calculating their score, as well as the socioeconomic characteristics. Finally, the LC model was employed to explore the preference heterogeneity among groups after the inclusion of latent variables of tourist satisfaction.
For the first method, this paper used tourist satisfaction as the standard for population segmentation to test the relationship between satisfaction and preferences in H1 and H2. Here the 15-item satisfaction scale in Table S1 was explored as a single construct. The mean satisfaction score of samples is 58.81 (standard error 0.214). In this paper, a three-segment equal size method (Choi and Fielding 2013) was used to divide the respondents into three groups of low, medium and high degree of satisfaction, and each group included about a third of respondents. ‘Low’ respondents had a satisfaction score of 54 or lower; ‘medium’ respondents scored above 55 and below 63; ‘high’ respondents scored higher than 64. The three groups have 89, 105 and 93 respondents, and 178, 210 and 186 observations respectively, with mean satisfaction score of 48.77, 58.23 and 69.09.
To test the impact of tourist satisfaction segmentation on each attribute, a RPL model was established to analyze each segment and all respondents independently. As showed by the statistical results in Table 4, the fit of the four models is significant according to the Chi-square statistics. The estimates for medium satisfaction group are broadly consistent with those for the full sample, while the estimates for the low and high satisfaction groups show significant differences. To be more specific, for the medium satisfaction group, the alternative specific constant (ASC) is significantly positive. The BIO and EEF of management attributes are significantly positive, NUM is not significant, and the PRICE attribute is significantly negative. These results indicate that the respondents in the medium group would like to improve the status quo in biodiversity and environmental education facilities, but at the same time, the high entrance fee would bring disutility. For the low satisfaction group, ASC and management attributes are not significant, and only the price attribute is significantly negative, which indicates that the respondents in the low satisfaction group are not willing to change the status quo and are only sensitive to price. For the high satisfaction group, only ASC and BIO are significantly positive, indicating that respondents in this group are only prone to improve biodiversity. The great difference of parameter estimates in different segments of Table 4 indicates that the groups with different satisfaction have significant preference heterogeneity, which supports the hypothesis H2.
Table 4
Estimation results from random parameter logit model of three satisfaction segments.
Attributes
|
All respondents
|
Low
|
Medium
|
High
|
ASC
|
4.786***
|
3.232
|
5.105**
|
3.539**
|
BIO
|
0.563***
|
0.586
|
0.527**
|
0.516*
|
NUM
|
-0.111
|
-0.742
|
-0.072
|
-0.031
|
EEF
|
0.401***
|
0.571
|
0.413*
|
0.049
|
PRICE
|
-0.236***
|
-0.681**
|
-0.229*
|
-0.021
|
Summary statistics
|
LL
|
-630.60
|
-195.55
|
-230.71
|
-204.34
|
Chi-squared
|
45.65***
|
26.22***
|
12.53*
|
17.46**
|
Observations
|
574
|
178
|
210
|
186
|
Note: ***, **, * ==> Significance at 1%, 5%, 10% level. ASC stands for alternative specific constant.
To further demonstrate the hypotheses of H1 and H2, the mean WTP of three attributes across different satisfaction segments was estimated as Fig. 3 shows. To calculate the WTP, the formula is WTPm=-βm⁄βp , where WTP stands for the willingness to pay of each attribute, βm is the parameter of management attribute m and βp is the parameter of price attribute p. The estimates of WTP for different satisfaction groups show significant differences. Except for the NUM attribute, the higher the tourist satisfaction, the higher the WTP, especially for the BIO attribute. This result partly supports the hypothesis H1 and fully supports H2, which indicates that tourists with higher satisfaction are more inclined to pay more for improving wetland ecotourism than tourists with lower satisfaction (Kang et al. 2018).
Tourist satisfaction continued to be incorporated into the RPL model as a single construct, and the results are shown in Model 1 of Table 5. Chi-square statistics show that the model fits well and are improved to some extent. However, this result shows that the interaction term of satisfaction is not significant and does not support H1.
Table 5
Estimation results from random parameter logit models.
Attributes
|
Model 1
|
Model 2
|
Model 3
|
Model 4
|
ASC
|
0.717*
|
0.812*
|
1.024*
|
1.233**
|
BIO
|
0.565***
|
0.571***
|
0.571***
|
0.561***
|
NUM
|
-0.103
|
-0.090
|
-0.095
|
-0.086
|
EEF
|
0.401***
|
0.392***
|
0.396***
|
0.373***
|
PRICE
|
-0.230***
|
-0.218***
|
-0.227***
|
-0.206***
|
Satisfaction1
|
-0.142
|
|
|
|
Gender1
|
|
0.778***
|
|
0.804***
|
Age1
|
|
-0.099
|
|
-0.151*
|
Education1
|
|
-0.144
|
|
-0.147
|
Income1
|
|
0.242
|
|
0.254*
|
Job1
|
|
-0.504***
|
|
-0.417
|
NRS1
|
|
|
-0.522**
|
-0.520**
|
EGS1
|
|
|
-0.464**
|
-0.465**
|
FS1
|
|
|
-0.238
|
-0.257
|
TES1
|
|
|
0.609***
|
0.602***
|
Summary statistics
|
LL
|
-607.71
|
-601.24
|
-598.74
|
-589.18
|
Chi-squared
|
45.79***
|
58.72***
|
63.72***
|
82.83***
|
1 They are all interaction terms between alternative specific constant and latent variables.
To further test the hypotheses, tourist satisfaction continued to be included into the RPL model as a multidimensional construct in the second approach. The weight of tourist satisfaction evaluation indicators was constructed and the score of each respondent on the four constructs reflecting satisfaction was calculated based on the estimated results of Table 3. The formula for calculating the weight is , where Wic and Fic are the weight and factor loading of the indicator i with construct c, respectively. The calculation results of the indicator weight of four constructs of tourist satisfaction are presented in Table S3. According to the indicator weight, the score of each construct was calculated.
Socioeconomic attributes and the score of the four constructs were respectively included in the RPL model, and Model 2, Model 3 and Model 4 in Table 5 were obtained. The results show that the model fits are significant according to the Chi-squared statistics and are improving gradually. Model 4 has the best fitting result. Same with previous literature (Andreopoulos et al. 2015; Birol et al. 2006; Xu et al. 2020), the interaction terms between ASC and socioeconomic profiles show that gender and income are significantly positive, and age is significantly negative, indicating women, young people, and high-income respondents are more willing to change the status quo. Among the interaction terms between ASC and the four constructs of tourist satisfaction, NRS and EGS are significantly negative, FS is not significant, and TES is significantly positive. This indicates that the respondents with low satisfaction on natural resource and environmental governance and high satisfaction on travel experience prefer to improve the status of wetland ecotourism. Therefore, the results support the hypothesis of H3, H4 and H6, but not H5.
To explore the preference heterogeneity among groups after the inclusion of latent variables of tourist satisfaction, in the third method, the LC model was employed. Using the same three-segment equal size method as Method 1, the score of four constructs calculated in Method 2 was assigned as 1, 2, and 3, respectively, on behalf of low, medium and high level. The assigned values of four constructs were incorporated into the LC model.
In order to determine the number of classifications, we compared the information criteria statistics of different segmentations in Table S4, including Bayesian information criteria (BIC), Akaike’s information criteria (AIC), AIC with a penalty factor of 3 (AIC3) and consistent AIC (CAIC) (Boxall and Adamowicz 2002; Shoji and Tsuge 2015). Although BIC, AIC3, and CAIC decline until the inclusion of two classes, then increase as more classes are added, the AIC criteria identify the three-class model as the best (lowest AIC). Notice also that the R2 increases from 0.02 for the standard aggregate (one-class) model to 0.47 for the three-class model, which assesses the percentage variance explained in the dependent variable. Therefore, the three-class model is recommended for analysis.
Figure 4 shows the relative importance of attributes in the three-class model. Of all the respondents, 43.31% are classified as Class 1, 33.68% as Class 2, and 23.01% as Class 3. Class 1 includes the majority of respondents who consider environmental entertainment facility as the preferred attribute, while other attributes are of less relative importance, so named as “entertainment-preferred”. Members of Class 2 have a particularly high preference for price attribute, hence called “price-sensitive”. In addition, Class 3 members show a higher preference for biodiversity attribute and lower preference for other improvement attributes, thus named as “eco-friendly”.
Table 6 reports the estimated results for the three-class model. The utility function of wetland ecotourism management attributes and the class membership function of socioeconomic profile and four latent variables of tourist satisfaction was estimated. The results show considerable heterogeneity in preferences between groups of wetland ecotourism improvement policies, which supports the hypothesis H2. From the perspective of utility function, only EEF attribute is significantly positive in Class 1, indicating that the members of Class 1 only have a positive tendency to improve environmental education facilities. Only the BIO attribute is significantly positive in Class 3, meaning that the respondents in Class 3 are only willing to improve biodiversity. In Class 2, the price attribute is significantly negative at the 1% level, the NUM attribute is significantly negative at the 5% level, and the EEF attribute is significantly positive at the 10% level. The result indicates that the members of Class 2 have negative expectations of the number of visitors besides price, but at the same time, surprisingly, they are willing to change the status quo of environmental education facilities.
Table 6
Estimation results of latent class model.
Variable
|
Class 1 Entertainment- preferred
|
Class 2 Price-sensitive
|
Class 3 Eco-friendly
|
Utility function: ecotourism management attributes
|
BIO
|
0.326
|
-0.141
|
4.453*
|
NUM
|
0.176
|
-1.057**
|
0.745
|
EEF
|
0.496**
|
0.633*
|
2.950
|
PRICE
|
-0.111
|
-1.636***
|
-0.299
|
Class membership function: socioeconomic profile and multidimensional tourist satisfaction
|
Intercept
|
-2.531
|
-0.653
|
3.183***
|
Gender
|
1.065**
|
-0.217
|
-0.848*
|
Age
|
0.224
|
0.205**
|
-0.428***
|
Education
|
-0.234
|
0.107
|
0.127
|
Income
|
0.317
|
-0.134
|
-0.183
|
Job
|
-0.024
|
-0.040
|
-0.016
|
NRS
|
-0.032
|
0.071
|
-0.039
|
EGS
|
-0.547**
|
0.273
|
0.274
|
FS
|
0.029
|
-0.211
|
0.183
|
TES
|
0.807***
|
0.088
|
-0.895***
|
Note: ***, **, * ==> Significance at 1%, 5%, 10% level.
The results of the class membership function reflect the individual sources of preference heterogeneity from the aspects of socioeconomic attributes and multidimensional tourist satisfaction. For Class 1, the statistically significant class membership coefficients indicate that being female, having a lower level of environmental governance satisfaction and a higher level of travel experience satisfaction increase the possibility that the respondents are more inclined to pay attention to the improvement of recreational facilities related to wetland ecotourism. In Class 2, only age is significantly positive, revealing that older respondents are more likely to become members of Class 2. For Class 3, the significantly negative parameters of gender, age and TES indicate that male, young people and respondents with lower level of travel experience satisfaction increase the possibility of becoming members of Class 3 and are more willing to improve the biodiversity attribute related to wetland ecotourism.
Figure 5 depicts a ternary phase diagram of probability means of the three-class model that can be used to locate each class in an informative 2-dimensional barycentric coordinate display. Each vertex of a ternary phase diagram corresponds to a class. The level assignments of each variable in Fig. 5 are described in Table S5. Figure 5 provides more specific information on preference heterogeneity and class membership characteristics between different classes. As showed in the figure, Class 1 respondents wish to add the environmental education information boards and the payment level is high. Female respondents, those over 30 years old, those with a monthly income of more than 5000 RMB, and those who have jobs are more likely to fall into Class 1, as well as those who are more satisfied with the facilities and travel experience. Class 2 members have no desire for improvement other than a decrease in the number of visitors and their WTP is relatively low. Male, respondents with high satisfaction of environmental governance and low satisfaction of facilities have more probability to become the members of Class 2. For Class 3 members, they prefer the option of increasing the number of species by 10%, increasing the number of visitors, and increasing the plants and animals’ specimen museums, with the medium payment level. Respondents younger than 30 years old, earning less than 3,000 RMB a month, not working, and having lower satisfaction with natural resources are more likely to fall into Class 3. Unlike the previous results (Birol et al. 2006), education is not clearly differentiated here, so it is not included in the figure. The above results support the hypotheses of H3, H4, H5 and H6, and the four constructs have an influence on the preference heterogeneity among different classes.
Figure 6 shows the marginal WTP of attributes related to wetland ecotourism improvement policies for different classes as calculated by the LC model. The estimates of marginal WTP provide us with the implicit benefits of various variations at the attribute level (Hoyos et al. 2015). The figure turns out that WTP varies a lot from class to class, which also confirms the hypothesis of H2. Eco-friendly members are most willing to pay for wetland ecotourism policy improvements, followed by entertainment-preferred respondents, while price-sensitive respondents are almost unwilling to pay anything for improvements.