SOCIEDADES SIN PERSONALIDAD JURIDICA
A. S OCIEDADES GENERALES SIN PERSONALIDAD JURÍDICA
Table 4.5 presents the estimation results of the initial full model, the best subset regression model and the final stepwise regression model for respondents in Tianjin.
Table 4.5 Parametric WTP Models for Respondent in Tianjin Significance Level: 0.001 ***, 0.01 **, 0.05 *, 0.1 ○
Variable
Coefficient (Standard Error) Initial Full Model Best Subset
Model Final Stepwise Model (Intercept) 2.12 (2.13) 0.79 (1.14) 0.11 (0.80) Perceive-shortage Moderate Scarce 1.62 (0.81) * 1.11 (1.07) Not Abundant 0.81 (0.56) Opinion-service Acceptable Dissatisfactory -1.87 (1.01) ○ -4.10 (1.71) * -1.31 (0.70) ○ -2.68 (1.22) * Numeric a -1.01 (0.34) ** Know-bill -0.03 (0.79) Know-price 0.73 (0.74) 0.81 (0.52) Know-benefits 0.24 (0.30) 0.32 (0.22) Heard-mid-route 0.60 (0.71) 0.84 (0.46) ○ Heard-EC 1.95 (1.04) ○ 1.49 (0.66) * 1.37 (0.59) * Understand-EC 0.28 (1.49) Opinion-EC -0.70 (0.47) -0.57 (0.35) Price-increase -1.86 (0.75) * -1.50 (0.58) ** -1.80 (0.50) *** Household-size -0.37 (0.34) Income (yuan/month) 2500-4000 Above 4000 0.79 (0.83) 2.28 (1.23) ○ 0.49 (0.58) 2.15 (0.85) * Numeric 0.91 (0.33) ** Gender (Male) -0.44 (0.68) -0.73 (0.51) Age 31-40 41-50 50-60 & Above 60 0.85 (0.81) -0.84 (1.20) 1.96 (1.39) Above 50 1.62 (0.71) * Job Private Sectors Retired Unemployed -0.42 (0.69) -3.57 (1.71) * -0.34 (1.65) Without Job -1.18 (0.88) Without Job -2.67 (0.84) ** Education High School Middle School or Below
-0.37 (0.81) -1.09 (1.38) Residence 10-20 5-10 Below 5 -1.33 (1.04) -0.27 (1.38) -1.14 (2.13) Less than 20 -0.76 (0.67) Visit -1.39 (1.17) Relatives 1.36 (1.07) 1.03 (0.57) ○
Model Evaluation Indicator
AIC 149.77 129.99 161.90
Prediction Error Rate b 0.18 (109) 0.26 (111) 0.24 (155) Overall Significance 0.13×10-2 9.35×10-6 1.20×10-10
a Numeric refers to numeric variable with linear effect.
Respondents in Tianjin seemed more reluctant to give their opinions and information in the survey, which caused a large number of missing values in the dataset. For example, 57 (23.2%) respondents chose to give no opinion about the general idea of Ecological Compensation (Opinion-EC) whereas only 9 (4.5%) respondents in Beijing did so. As a result, the initial full model with all the 19 explanatory variables42
The final stepwise model selected 8 out of the 19 explanatory variables of the initial full model and substantially increased the observations from 109 to 155. The 92 omitted observations were caused by missing values of the response variable (48 respondents did not give a clear yes/no answer to the WTP question) and explanatory variables that showed significant effects on respondents’ answers to the WTP question but involved private information that some respondents refused to give such as Income (37 missing values) and Job (17 missing values). Despite having a larger AIC (161.90) than the other two models due to the substantial increase in observations, the final stepwise model maintained a moderate prediction error rate of 0.24. This means that the final stepwise model could correctly predicted 155 × (1-0.24) =118 respondents’ answers to the WTP question whereas the initial full model could only provide 109 × (1-0.18) = 89 correct predictions. More importantly, the overall model significance of the final stepwise model was much higher than the other two models (1.20×10-10). In spite of the considerable difference in the goodness of fit, the variables Opinion-service, Heard-EC, Price-increase, Income, and Job were significant at least at the 0.1 level in all the three models, which reflects the underlying consistency between the three models.
contained only 109 observations while the survey actually collected a total of 247 observations. Nevertheless, the initial full model still exhibited some extent of ability to explain respondents’ answers to the WTP question given its prediction error rate of 0.18 and overall model significance of 1.30×10-3. In comparison, the best subset model of the Tianjin survey with 11 explanatory variables and similar observations (111) reduced the AIC from 149.77 to 129.99 and raised the overall model significance from 1.30×10-3 to 9.35×10-6, but the prediction error rate rose from 0.18 to 0.26.
Noticing that the Income variable was categorized into three levels (the reference level of below 2500 yuan/month is not listed in Table 4.5) in the
42
The variable of Heard-SNWTP and Understand-EC were removed for a similar reason explained in Footnote 41.
initial full model but only one of the levels (above 4000 yuan/month) was significant at the 0.1 level. In the best subset model which had less explanatory variables and better goodness of fit, the income level of “above 4000 yuan/month” became significant at the 0.05 level. Then in the final stepwise regression model with even less variables, better goodness of fit and more observations, the variable Income was treated as a numeric variable (with linear effects) and became significant at the 0.01 level. This is an example of improving model fitting and better revealing variables’ effects on respondents’ answers to the WTP question by removing unnecessary explanatory variables and adjusting the coding of variables. More examples can be easily found on other variables (such as Price-increase, of which the significance level was 0.05, 0.01 and 0.001 respectively in the three models) and in the results of other cities. Similar justification will not be repeated in later discussion.
Figure 4.4 illustrates the eight explanatory variables in the final stepwise model for the respondents in Tianjin. The variables Age (above 50), Heard- EC, Relatives, Income and Heard-mid-route were found to be positively influential on respondents’ yes answer to the WTP question with the coefficients of 1.62, 1.37, 1.03, 0.91 and 0.84 respectively. That is to say, the odds of accepting an increase in water price by respondents older than 50 were 5.05 (e1.62) times larger than the odds of younger respondents. Those who had heard about Ecological Compensation were more likely to accept an increase in water price and the odds of yes answer were 3.94 (e1.37) times larger.
Respondents who had relatives or close friends living in the water source area had the acceptance odds 2.8 (e1.03) times larger. Compared with respondents whose gross income was below 2500 yuan/month, the acceptance odds of respondents who earned 2500-4000 yuan/month and more than 4000 yuan/month were 2.48 (e0.91) and 6.17 (e0.91×2) times larger respectively. Additionally, respondents who had heard about the specific middle route of the SNWTP had the acceptance odds 2.32(e0.84) times larger. For the five positive variables, Income was at the highest significance level of 0.01, Age (above 50) and Heard-EC were both at the 0.05 significance level while Relatives and Heard-mid-route were only significant at the 0.1 level.
Figure 4.4 Explanatory Variables in the Final Model of Tianjin
Without Job is the merged category of the variable of Job, referring to respondents
who were retired or unemployed.
The variable Job (retired & unemployed, i.e. “Without Job” in Figure 4.4), Price-increase and Opinion-service were found to be negatively influential on respondents’ yes answer to the WTP question with the coefficients of -2.67, - 1.80 and -1.01 respectively. This means that the odds of accepting a higher water price by respondents in retirement or unemployment were only 6.9% (e-2.67) of the odds of respondents who had a job. For an increase of 1 yuan/m3 in the water price, the acceptance odds would decline by 83% (1 - e-1.80). Compared with respondents who felt satisfied about the current tap water service, those who thought it acceptable and dissatisfactory were less likely to accept an increase in water price and the odds of yes answer declined by 64% (1-e-1.01) and 87% (1-e-1.01*2) respectively.
The finding of smaller odds of accepting a higher water price by respondents who were less satisfied about the current tap water service seems unexpected in the first place since the water transfer project is assumed to improve the tap water quality in the northern cities, so respondents should be more willing to pay a higher water price if they were not satisfied about the current tap water service. However, further analysis of the survey results found that among the 30 respondents in Tianjin who had given the reasons for their dissatisfaction about the tap water service, 25 of them mentioned water quality issues (turbidity, odour and taste), but 28 of them did not know
the water transfer project could improve tap water quality in the northern cities. In other words, these respondents were less willing to pay a higher water price probably because they did not want to pay additional money for a service that (they thought) could not solve their water quality problem. Lastly, among the three negative variables, Job and Opinion-service were both significant at the 0.01 level, and Price-increase was highly significant at the 0.001 level, which corresponds with the findings in the section of non- parametric model that the data of Tianjin survey behaved quite well in terms of monotonicity (Figure 4.1).