4.1. Descriptive Statistical Analysis
Descriptive statistics from data show that water rights reform might have played a role in reducing the irrigation application for wheat. For instance, with water rights certificates, in 2008, the irrigation application per ha for wheat was 4795 m
3, saving 2.06% of irrigation water (
Table 6). However, this was not consistent in 2014, when farmers increased their water use if they had certificates.
Increasing water prices reduced the irrigation application in 2008 and 2014. As a whole, from the lowest quarter interval (for water price) to the highest one, the irrigation application per ha for wheat declined from 6628 m
3 to 3590 m
3, and the mean irrigation times decreased from 3.73 to 2.2, dropping by 45.84% and 41.02%, respectively (
Table 7). Specifically, for irrigation application per ha, the rate of reduction was 56.61% in 2008 and 30.53% in 2014, respectively. Correspondingly, the rate of decline for irrigation times was 49.07% and 33.55%, respectively. In 2008, farmers in the second quartile of water prices reduced irrigation application rates by 28% compared to those in the first quartile. Farmers in the third quartile reduced irrigation application rates by 30% compared to those in the second quartile. However, farmers in the fourth quartile only reduced irrigation application rate by 13% although the mean price spiked from 0.19 yuan/m
3 to 0.63 yuan/m
3. This unusual result is most likely due to the villages’ water scarcity for the famers in the 4th quartile being more severe, thus farmers used more groundwater. The story in 2014 was similar.
The results above imply that water rights reform and irrigation pricing policy can possibly reduce irrigation application. However, due to many other factors affecting irrigation application, such as household characteristics or regional circumstances in nature, we cannot determine the real relationship (of irrigation application to water rights reform and irrigation pricing policy) merely by using simple descriptive statistical analysis. Therefore, multivariate econometric analysis is required to analyze the real relationship between irrigation application and water rights reform or irrigation pricing policy.
4.2. Econometric Model
Based on the above discussions, the link between irrigation application per ha for wheat and its determinants (such as water rights reform, irrigation pricing policy, and other factors) can be represented by the following equation, which applies plot level data in Zhangye:
where
Wijkc represents the average irrigation application per ha for wheat from the
ith plot of
jth household in
kth village in
cth county. The rest of the variables explain the irrigation application.
Ikc is our variable of interest, indicating the water rights reform, measured by whether the village has been issued water rights certificate (1 = yes, 0 = no).
Pijkc is the irrigation water price in the
ith plot of
jth household in
kth village in
cth county, measured by yuan/m
3.
Other control variables in the equation, represented by a matrix Zijkc, are included to represent other villages, households, and plot factors that affect irrigation application. In particular, we included two variables to hold the status of a village’s water resources constant: the percentage of areas irrigated only by surface water, and the dummy indictor, which signals whether water is scarce in the village. Household characteristics include age and the education level of the household head and share of labor. We also added three plot attributes: plot area, soil type, and rate of lined canals from water outlets to plots. Finally, our models included year dummy variable (Ykc), 1 means 2014, 0 means 2008. Rkc is a regional dummy variable representing the county in which a household is located. In addition, county dummies can capture the time invariant factors at the county level such as crop prices, general climate conditions and hydrology of the sample areas. Year dummies help capture the general trend of policy and changes in agricultural technologies. The symbols α, β, γ, ρ, θ, and λ are parameters to be estimated, and εijkc is the error term, which is assumed to be uncorrelated with the other explanatory variables in our initial equation.
4.3. Estimation Results: The Impacts of Water Rights Reform and Irrigation Pricing Policy on the Irrigation Application for Wheat
The empirical estimations perform well for wheat’s irrigation application model (
Table 8). The goodness of fit measures (adjusted
R2) are around 0.3, which sit at the upper end of the range of
R2s observed in empirical analysis that use household level repeated cross sectional or longitudinal survey data. There are examples of the similar empirical studies that have much larger sample sizes, but lower
R2. For example, Zhang and Xu (2016) [
48] reported
R2s around 0.25 with a sample of 4729 observations and Giles (2006) [
49] reported lower
R2s with a large sample of about 17,000 observations. To test the hypothesis of homoscedasticity that is the most prominent problem in the ordinary least square (OLS) regression based cross-section data, we have done the White’s test. From the results, we fail to reject the null hypothesis of homoscedasticity for each model because the
p-values of all tests (Prob > Chi2) are greater than 0.1. Even though it is unnecessary to use a robust standard, we do the robust regression for the robustness of results. Since our dependent variable is one continuous variable, independent variables in the regressions are exogenous and have no serious multicollinearity, and the error terms are also uncorrelated, it is rational for us to use robust OLS regression. Most of the coefficients for the control variables have the expected signs and are statistically significant. For instance, the results indicate that after holding other factors constant, in villages with a larger share of irrigated areas serviced only by surface water, the farmers used more water per ha for wheat. In addition, the relationship between irrigation application and the age of the household head appears in an inverted U shape, with a turning point at the age of about 53. This means that after experiencing some years of farming, farmers tended to reduce irrigation application per ha. In addition, the demographic factors in our models are continuous variables, like education level and share of farming labors in a household, which are more accurate than dummies used in Wheeler et al. (2009, 2010) [
50,
51]. Despite the differences above, the conclusions are similar that higher education and older age of farmers may reduce the irrigation application or increase the probability of water rights trade. There is also some evidence that larger plots use less water per unit of land. One reason may be that it is easier to level a larger plot so that less irrigation water is needed to reach the whole plot. Even though the coefficient of year dummy is not significant, it does control the hydrologic uncertainty in different years. If we do the pooled regression that uses all observations, changing every variable to two variables (one interacting with 2008 dummy, one interacting with 2014 dummy), the results are robust. Other control variables had no consistent effects on irrigation application, and were only significant in 2008 or 2014.
Clearly, water price motivates farmers to reduce irrigation application for wheat all the time. The regression results show that the coefficients of water price are all negative and significant at the 1% level. Especially in 2014, when water price increased a great deal (
Table 4 and
Table 7), the negative effects on irrigation application were relatively huge. Moreover, the overall price elasticity of agricultural water demand, calculated based on the estimated results, was approximately −0.17; this means that irrigation application per ha for wheat would lessen by 0.17% if water price rose by 1%. In addition, when we analyze the marginal effects of prices for both surface water and groundwater in the respective regression, the marginal effect of groundwater price is bigger than that of surface water. Moreover, farmers use less surface water when irrigation is charged in the time-based scheme.
However, regarding water rights reform, water rights certificates do not have a sustainable function on reducing irrigation application. The results signal that changes in water rights were possible at the beginning of reform, but failed in the end. For example, the coefficient of the water rights certificate was negative and significant at the 10% level in 2008. In other words, if the institution was implemented, the irrigation application per ha of wheat could be reduced by 23%. On the contrary, the results from 2014 are not significant.
A lack of effective implementation is a possible reason for the ineffectiveness of the certificates. We can demonstrate this based on three aspects. First of all, there are more and more irrigated areas that cannot obtain water rights, which leads to more water demand for excrescent land. For example, the share of villages with actual irrigated areas was greater than the number of areas listed on the water rights certificates; this proportion of villages increased from 43% in 2008 to 50% in 2014, while the corresponding equality relation declined from 50% to 36% (
Table 9).
Secondly, most villages with certificates believe that the water rights amount listed on the certificate cannot meet their demand for irrigation water. Our survey data show that in 2008, half of these villages thought they should receive more water than the amount listed on their certificates; this figure rose to 87% in 2014. Finally, as expected, the actual irrigation application exceeded the water rights amount listed on the certificates. For instance, 67% of villages were using more water than the regulated amount in 2008; this share grew to 75% in 2014.
The specific reasons why the actual demand for irrigation water was greater than the water rights amount can provide a deeper insight into water rights reform. In 2008, most of the villages, about 57%, which were limited by their certificates thought that water scarcity was the main cause of why they were not assigned more water; 29% of villages believed this was due to an administrative order being executed that stipulated agriculture should save water in terms of irrigation application (
Table 10). However, water resources were relatively abundant, and there was no intense mandatory requirement of water conservation in 2014. Hence, the primary explanations were unscientific water management and extensive irrigation patterns, occupying 75% of villages that were limited by their certificates.
In addition, farmers receive almost no punishment if they exceed their amount of water rights, which encourages them to use yet more water. Our survey data show that in 2008 and 2014, approximately 80% of villages had no punitive measures to penalize those who violated their amount of water rights (
Table 11). Even if there was a regulated punishment, it was never implemented. Only about 20% of the villages had the two really effective measures of penalty or cutting water source, yet the penalty disappeared in 2014.
The facts above indicate that the institution of water rights certificates was not implemented strictly and accurately, which emboldened farmers to use more irrigation water. This implies that besides the impetus of natural conditions, water rights reform in Zhangye was conducted rapidly in the beginning stages, with distinctive administrative features. Once the passion of reform winded down, the implementation of the reform was ineffective. This indicates that superficial water rights reform may be effective in the short term. Based on the above investigation, we can see that water price persistently played a role on reducing irrigation application. In addition, water rights reform was only helpful in the early phases; it failed in the later stages, meaning that it is not sustainable in the long run.
The analysis in the third part of the paper demonstrates how water rights reform in Zhangye, which has unique administrative characteristics, lacks effective implementation in the long term. In the beginning of reform, having a water rights certificate could mean that reform was relatively normative, then reached the expected goal of reducing irrigation application. However, without successful execution in the later phases, water rights certificates could simply become a form of water rights reform because the true basis of water allocation was water rights areas that every village had. In this case, water rights certificates had no distinct role in reducing irrigation application, which means that nominal water rights reform was ineffective and unsustainable. The econometric results provide evidence for our reasoning.