What attributes of mandatory waste management policy can enhance the separation intention of residents in China? A behaviour choice experiment

Abstract

China is facing a severe waste siege. Notwithstanding the several years of mandatory policy implementation, the residents’ waste separation intention remains insignificantly enhanced. Consequently, it is urgent to theoretically answer the question of what type of mandatory policy is most effective. Thus, in the current research, we propose the mandatory policy as a combination of five attributes, that is, economic penalty, social penalty, supervision, charging, and community governance, and conduct a policy choice experiment with 354 participants. The results of the randomized conjoint analysis show that economic penalty, supervision, and community governance are influential determinants of separation intention. The mandatory policy has an interaction effect with auxiliary policies; specifically, community governance and supervision are more effective, while residents are also more willing to pay fines when a points redemption policy or a well-developed infrastructure system is present. The effect of mandatory policy also varies across residents. Females, highly educated, or high-income individuals are relatively more inclined to separate waste, while elders, renters, and residents who have lived in an environment for more than 20 years are less likely to separate. The results of the welfare gain analysis show that strengthening the penalty, supervision, and community governance are also helpful toward improving social welfare. In order for mandatory policies to work better, it is important to continuously improve community governance, raise the cost of violations, vigorously promote waste supervision, perfect the reward mechanism, and improve waste separation infrastructure.

Appendix

Estimation of welfare gains

Suppose a mandatory policy consists of L attributes. The utility of the current policy is referred to as the status quo utility, denoted by U₀. The utility of a hypothetical policy is represented by U(c, a), where c refers to the individual burden of implementing the policy and a is a vector of discrete attributes. The probability of choosing a hypothetical policy combination rather than the status quo can be defined as

$$q(c,a) = \Pr [U(c,a) \ge U_{0} ]$$

q(c, a) is Bhattacharya’s structural choice probability [87]. Then, the marginal structural choice probability can also be defined as

$$Q(c) = \sum\limits_{a} {\Pr [U(c,a) \ge U_{0} ] \times p(a)}$$

(2)

where p(a) denotes the conjoint uniform distribution of profile attributes. Therefore, Eq. (2) provides explanations for estimating the probability of structural choice, which is the proportion of residents who prefer to improve the policy rather than maintain the status quo.

The subsequent discussion is how to define the WTP, c^WP(a). Since it is difficult to estimate the acceptable burden for each resident, we also consider that U₀ = U(c^WP (a), a) ≥ U(c, a) if and only if c ≥ c^WP (a). Therefore, the marginal distribution function of WTP can be defined as

$$F^{{{\text{WP}}}} (C) = \sum\limits_{a} {\Pr [U_{0} \ge U(c,a)\left| a \right.] \times p(a)}$$

(3)

Combining Eqs. (2) and (3) yield

$$F^{{{\text{WP}}}} (c) = 1 - Q(c)$$

(4)

Equation (4) indicates that the structural choice probabilities are sufficient statistics to recover the WTP distribution. Thus, the marginal average WTP can be obtained from the summary statistics of the WTP distribution as follows:

$$E[c] = \int\limits_{0}^{\infty } {cdF^{{{\text{WP}}}} (c)} = \int\limits_{0}^{\infty } {cd[1 - Q(c)]}$$

(5)

The average WTP can be interpreted as a monetary measure of the welfare gain from policy implementation, which implies identifying the welfare gains of the policy from the choice experiment data. As the experimental data for this study only provide estimates of the probability of structural choice for CN¥0, 2, 5, and 10, it is difficult to obtain nonparametric point estimates of the monetary welfare gain. Therefore, the marginal average WTP can be rewritten as

$$E[c] = \sum\limits_{i = 0}^{k} {\int\limits_{{c_{i} }}^{{c_{i + 1} }} {cd[1 - Q(c)]} }$$

(6)

where c_i is the ith attribute level of the payment attribute c, k is the number of attribute levels, and c₀ = 0 and c_k+1 = ∞. In this experiment, k = 4, while c₁ = 0, c₂ = 2, c₃ = 5, and c₄ = 10.

Based on the mean value theorem combined with Eq. (6), the lower bound of the marginal average welfare gain can be derived as

$$\underline{C} = \sum\limits_{i = 0}^{k} {c_{i} [Q(c_{i} ) = Q(c_{i + 1} )]}$$

(7)

Likewise, from Eq. (6), the lower bound of the conditional average welfare gain is derived as

$$\underline{C} \left| {a{}_{l}} \right. = \sum\limits_{i = 0}^{k} {c_{i} [Q(c_{i} \left| {a_{l} } \right.) - Q(c_{i + 1} \left| {a_{l} } \right.)]}$$

(8)

Equations (7) and (8) indicate that the lower bounds of both marginal and conditional welfare gains can be identified by the estimates of the choice probabilities, as no other unknown parameters are included in these equations.

According to Kaneko’s research on WTP [81], estimating the lower bound of the marginal average WTP needs to first determine the structural choice probabilities of the policy program. The probability value can be regressed by the following equation:

$$Y_{ijk} = \beta_{0} + \beta_{1} \times X_{1} + \beta_{2} \times X_{2} + \beta_{3} \times X_{3} + \delta_{ijk}$$

where X_1–3 are dummy variables for the policy payment attribute as CN¥5, 2, and 0, respectively; β_1–3 are their corresponding coefficients. Thus, the estimators of the marginal structural choice probabilities are $\widehat{Q}(0) = \widehat{\beta }_{0} + \widehat{\beta }_{3}$, $\widehat{Q}(2) = \widehat{\beta }_{0} + \widehat{\beta }_{2}$, $\widehat{Q}(5) = \widehat{\beta }_{0} + \widehat{\beta }_{1}$, and $\widehat{Q}(10) = \widehat{\beta }_{0}$, where a hat ($\wedge$) denotes an estimated coefficient.

Combining Eq. (7) yields the estimator of the lower bound, as follows:

$$\begin{aligned} \underline{{\hat{C}}} & = 0 \times \left[ {\widehat{Q}(0) - \widehat{Q}(2)} \right] + 2 \times \left[ {\widehat{Q}(2) - \widehat{Q}(5)} \right] \\ & \quad + 5 \times \left[ {\widehat{Q}(5) - \widehat{Q}(10)} \right] + 10 \times \widehat{Q}(10) \\ & = 2 \times \widehat{\beta }_{2} + 3 \times \widehat{\beta }_{1} + 5 \times \widehat{\beta }_{0} \\ \end{aligned}$$

(9)

Similarly, the conditional choice probabilities can be estimated by the following regression equation:

$$Y_{ijk} = \beta_{0}^{{d_{l} }} + \beta_{1}^{{d_{l} }} \times X_{1}^{{d_{l} }} + \beta_{2}^{{d_{l} }} \times X_{2}^{{d_{l} }} + \beta_{3}^{{d_{l} }} \times X_{3}^{{d_{l} }} + \delta_{ijk}^{{d_{l} }}$$

Using Eq. (8) yields the lower bound estimator for the conditional average welfare gain as follows:

$$\underline{{\hat{C}}} \left| {a_{l} } \right. = 2 \times \widehat{\beta }_{2}^{{d_{l} }} + 3 \times \widehat{\beta }_{1}^{{d_{l} }} + 5 \times \widehat{\beta }_{0}^{{d_{l} }}$$

(10)

The abovementioned equations indicate that the estimated average welfare gain is essentially an increasing function of the constant and the coefficient.