Abstract
In this study, participants from Beijing, China, were offered the opportunity to contribute to voluntary climate change mitigation by purchasing permits from two Chinese CO2 emissions trading schemes (ETS). Since CO2 emissions abatement is linked to other local benefits like the reduction in emissions of air pollutants, our aim is to separate the demand for local and global environmental protection. To this end, Beijing and Shenzhen ETS permits were offered. Our results indicate that a substantial part of the revealed demand for voluntary climate change mitigation in Beijing is driven by concerns for local co-benefits of CO2 emissions reduction.
1 Introduction
Local air pollution is one of the most urgent environmental problems in emerging countries. China is a prominent example. Here, coal combustion, originating from industry, power generation, and residential sources, is the single largest source of air pollution-related health impacts, and is estimated to have contributed to 366,000 premature deaths in China in 2013 (Health Effects Institute 2016). The most harmful local pollutants emitted from Chinese coal-fired power plants are SO2, NOX, and particulate matter (PM2.5/10) (Zhao et al. 2008). The different meteorological, geographic, and climatic conditions, as well as the differences in the intensity of emissions, result in concentrations of local pollutants differing considerably across the country. Fine particulate matter PM2.5, for example, is a major cause of air pollution, and the local concentrations in Chinese cities are substantially different across the country; these are typically much higher than in cities of developed countries.1
China struggles with more than just severe local environmental problems. The country is also the world’s largest emitter of CO2. Since climate change mitigation is a global public good, a strong free-rider incentive exists that makes an international cooperative solution highly unlikely. There is an intense debate about private “co-benefits” from climate change mitigation. According to the Intergovernmental Panel on Climate Change (IPCC) (2014a), co-benefits are defined as the positive effects that a policy or measure aimed at one objective might have on other objectives. Co-benefits are also referred to as ancillary benefits. Deng et al. (2017) provide a systematic review of the fast-growing research on co-benefits of reducing greenhouse gas emissions and classify them by type, mitigation sector, and geographic scope. Co-benefits from climate change mitigation policies include effects on ecosystems, economic activity, air pollution, health, resource efficiency, energy security, and technological spillover and innovation. Some of these co-benefits are clearly local. According to the IPCC (2015, 63), for example, “[climate] mitigation scenarios … are associated with significant co-benefits for air quality and related human health.”
These co-benefits from air pollution and health are particularly relevant for emerging countries with weak regulation of local pollutants. Therefore, it is expected that countries such as China, besides their incentives to contribute to the global public good climate change mitigation, have an additional incentive to mitigate CO2emissions because those emissions reductions are inevitably linked to reductions of local pollutants (Haines 2017; Zhang, Wang, and Du 2017). Reducing coal use through a carbon tax or an emissions trading system, for example, would lead to co-benefits from lowered PM2.5/10or SO2emissions. In a broader perspective, private co-benefits of climate change mitigation at home may also result from moral motivations associated with the emission reduction itself, like a “warm glow” of giving, moral satisfaction, or a positive self-image. To sum up, private co-benefits for the population at the local level from contributing to the global public good and climate change mitigation might be partly motivated by these local co-benefits.
Against this background, the central research question of our article is whether it is possible to isolate and quantify local co-benefits from climate change mitigation in individual behavior. For this purpose, we apply a revealed preference framework and use the initiation of seven pilot emissions trading schemes (ETSs) in China (Jotzo and Löschel 2014). Participants from Beijing were offered the opportunity to contribute to voluntary climate change mitigation by purchasing permits from two subnational Chinese CO2 ETSs in Beijing and Shenzhen. Purchased permits were withdrawn from the respective ETSs. Hence, CO2emissions are reduced locally in the respective region, potentially leading to local co-benefits. However, the effects of mitigating CO2—a uniformly mixed pollutant whose damage depends only on the total amount of CO2 in the atmosphere, not on the location of the emission—are identical in both cases. Because the CO2emissions reduction only leads to local co-benefits for our participants in Beijing, we are able to specify experimentally a locational preference in climate change mitigation. Under the reasonable assumption that reduced CO2emissions in Shenzhen are highly unlikely to cause positive co-benefits in Beijing, we are able to separate the actual (i.e., revealed) demand for global and for local environmental protection and provide the first empirical assessment not only of global climate change mitigation but also of local co-benefits of CO2emissions reduction and the associated willingness to contribute based on an experimental approach.
Our main results can be summarized as follows: (1) Contrary to standard economic theory, Chinese individuals contribute to CO2reduction, even though marginal benefit of contributing is practically zero while costs are positive. (2) An additional demand for CO2reduction stems from local co-benefits at low prices (i.e., more individuals contribute to climate change mitigation, and the median willingness to pay [WTP] is higher when local co-benefits are taken into account). Our results support the hypothesis that China has an additional motivation in contributing to mitigate climate change. For relatively small prices, the WTP for CO2reduction is even mainly driven by these local co-benefits. (3) The proportion of subjects who contribute to climate change mitigation quickly decreases as prices increase, that is, the revealed demand is rather elastic.
2 Related Literature and Theoretical Considerations
In economics literature, public goods are regularly treated as pure public goods, characterized by perfect nonexcludability and non-rivalry, although most public goods are not “purely public.” The main reason for doing so is the simplicity of pure public goods analysis. In reality, almost every global public good provision represents a joint production of several characteristics of different degrees of publicness, that is, global public goods production is usually an impure public goods production. Mitigating climate change as a global public good may serve as an example as it is not entirely nonrivalrous or nonexcludable. Besides the primary benefits from reducing CO2emissions, private co-benefits such as reduced local air pollution are generated. Figure 1 describes our underlying approach based on theoretical impure public good models (Cornes and Sandler 1994; Rübbelke 2003) under our assumption that CO2emissions reduction only leads to local co-benefits in Beijing.
Primary and Co-Benefits from Climate Change Mitigation
Because the experiment was conducted in Beijing, we use this as our baseline treatment (called T-BJ) and the economic activity can be represented by the reduction of the cap of the Beijing ETS by 1 t CO2. The corresponding CO2abatement generates two benefits for our participants: A primary public benefit from the reduction of CO2emissions and a private co-benefit. In this case, the public benefit is almost zero, and the private co-benefit may be small but positive. In our second treatment (T-SZ), subjects were offered the opportunity to reduce the cap by 1 t CO2for the Shenzhen ETS; in this case, it may be assumed that the sole benefit of such a transaction is that resulting from the public good provision, that is, greenhouse gas abatement.
Let us briefly consider the decision situation from this stylized theoretical perspective. Assume that there are two distant locations, i and j. In our case, i denotes Beijing and j Shenzhen. Consider location i. The utility of a representative agent in location i is
[1]
where mi is the agent i’s income, gi is the environmental good produced in i, and gj the environmental good produced in j. The first utility component u(mi) is obtained as the indirect utility function, when we assume that the representative agent maximizes her standard utility function being defined over the vector of consumption goods given her income mi and the market prices of these consumption goods (see Ebert 1993, 2003). The environmental good is CO2emissions reduction. The agent’s utility depends on the sum of CO2emissions reductions in i and j, which gives the second utility component v(gi + gj). Reducing CO2 emissions at i or j also (linearly) lowers local emissions at i or j, respectively, which is represented by the third utility component w(gi + dijgj), that is, the private co-benefits of public good provision. The parameter dij, 0 <dij< 1, is the coefficient that describes the effect of CO2emissions reductions in location j on location i. It is reasonable to assume that in our case dij is close to 0. The physical reason is that Shenzhen is situated at the coast of the East and South China Sea more than 2,000 km south of Beijing. In addition, dij might display other relationships, such as charity for the local pollution in Shenzhen. However, it should be noted that the concentrations of local air pollutants in Shenzhen are rather low, and Shenzhen is an iconic city for China’s opening-up policy with, for example, a GDP per capita that is considerably higher than in Beijing.2
Given (mi, gi, gj), we now ask how much of her income agent i would be willing to spend at most to obtain an additional marginal unit of the public good produced at the same location i. This maximum WTP 
is determined by the condition that utility U(mi, gi, gj) of agent i is kept constant, that is, that the condition
[2]
is satisfied. Hence, the marginal WTP is given by
[3]
The WTPgi can be interpreted as the virtual price of the environmental good, that is, if gi were a market good, the consumers would be willing to pay this price for another unit (Baumgärtner, Chen, and Hussain 2017).3 The marginal WTP for the environmental good relates the marginal benefit of an additional unit of gi with the marginal cost of forgone consumption. This is the point of departure for our experimental approach.
For an emissions reduction in j, the marginal WTP for an increase of gj for an agent in location i is given accordingly by
[4]
If dij is positive, WTPgi>WTPgj. With dij→0, as in our case, the marginal WTP for the agent in location i for CO2 emissions reduction in i, WTPgi, is derived from the public good utility and from the private co-benefits of public good provision only in i. The marginal WTP for CO2emissions reduction in j for a consumer in i, WTPgj, is then derived from the public good utility alone.4
The empirical literature on impure public goods is surprisingly limited. Heisey et al. (1997) and Midler et al. (2015) investigate impure public good problems such as biodiversity in an agricultural context. With a laboratory experiment, Munro and Valente (2016) show that green goods with impure public good characteristics do not necessarily enhance environmentally friendly behavior. Finally, Kotchen and Moore (2007) investigate why subjects participate in green electricity programs and how a program’s incentives affect participation.
There are, however, at least two other branches of empirical literature directly related to our study. First are the several revealed preferences studies that have recently explored the question of individual demand for voluntary climate change mitigation and derived WTP for climate change mitigation in monetary units per t CO2. Löschel, Sturm, and Vogt (2013) sold EU ETS permits at different prices to a sample of 202 subjects selected from the population of Mannheim, Germany. A median WTP of 0 and a mean WTP of 12 €/t CO2were found. Similar results were observed by Diederich and Goeschl (2014), who determined the willingness to abate 1 t CO2 among the German Internet-using population. They estimate a 0 median WTP and a mean WTP of about €6/t CO2. A similar revealed preference approach was used by Uehleke and Sturm (2017) and Löschel, Sturm, and Uehleke (2017), who investigated whether the implementation of collective action in the group of participants affects the individual contribution to the global public good climate change mitigation. Diederich and Goeschl (2017) estimated the elasticity of the probability of contributing to CO2abatement for a German sample and found on average an inelastic price reaction. They conclude that for Germany, using public funds to subsidize voluntary contributions to CO2abatement is not economically meaningful. Even closer to our study is the work by Diederich and Goeschl (2018), who used a German sample to test whether the location of the CO2abatement, either in the EU or in developing countries, affects the probability that the abatement option is chosen. They found no locational preference regarding mitigation, that is, subjects were indifferent about mitigation sites in the EU and in developing countries, thus showing that CO2abatement is perceived as a global public good.
Second, there is an increasing number of papers devoted to the effect of local air pollution in emerging countries such as China on health and well-being from an economic perspective. Barwick et al. (2017) quantified the health effects of PM2.5 in China and estimate consumer WTP for improved air quality. He, Fan, and Zhou (2016) estimated the impact of PM10on mortality during the 2008 Olympic Games in Beijing. Du, Shin, and Managi (2018) evaluated the effect of air pollution on life satisfaction. Using disaggregated air pollution data for SO2, NO2, PM2.5/10, and geocoded individual respondents from original survey data, they showed that all four pollutants have significantly negative effects on life satisfaction (see Zhang, Zhang, and Chen 2017 for a similar study). Chen, Guo, and Huang (2018) quantified the causal effect of air pollution on the health status and the school attendance of Chinese students. They found a sizable negative effect of air pollution on school attendance through the health channel.
Our study complements the literature by an innovative revealed preference approach to assess local co-benefits of climate change mitigation. Exploiting the existence of subnational Chinese ETS, it is based on individual purchase decisions for Chinese ETS permits in two distant locations. Our approach opens a new way to the empirical evidence on local co-benefits from climate change mitigation.
3 Experimental Design
The aim of our study was to investigate the extent to which a sample of the Beijing population would be willing to contribute to additional global and local environmental protection from their own disposable income. To elicit the demand for environmental protection, an experimental approach of asking people to give up real money instead of a survey approach was implemented. To address the impure public good problem, Beijing and Shenzhen ETS were used as vehicles and emissions reductions were directly sold to the subjects.5 The main characteristics of both ETSs are described in Appendix Table A6_1. For our purpose, it is particularly relevant that the ETS cap is binding, that is, the price is positive, and the schemes are not linked. This means that by reducing the cap by 1 t CO2 in Beijing or Shenzhen, we can be sure that CO2 emissions are reduced by that amount in the respective ETS region.
This section presents the experimental procedures, whereby the baseline treatment, T-BJ, is used as a reference. Modifications in the second treatment, T-SZ, are also explained. Participants were recruited by the University of International Business and Economics (UIBE Beijing, China) following the random distribution of approximately 8,000 letters of invitation within the fifth ring of Beijing city, supplemented by a random online call for participation using the WeChat service (see Appendix 1for details). The information that people received at this stage was that a survey would be carried out in which they would have the opportunity to buy products and they would receive remuneration of 300 RMB (about €40) for their time. Registration was done via telephone. To avoid subjects overstating their demand, the invitation letter emphasized that the amount of 300 RMB was explicitly remuneration for participation in the survey and their travel expenses.
To elicit the individual demand, a simple and incentive compatible market mechanism was chosen (see Appendix 2 for instructions). Each participant was confronted with six different prices for permits in 1 t CO2units ordered from high to low. Subjects had to decide whether they would be willing to buy at each price. Finally, one of the six prices was randomly and openly selected by rolling a die, and the transaction was carried out at the corresponding price in privacy, that is, only the subject and the experimenter received information about the transaction.6 Participants who did not wish to buy at a specific price indicated this with “No.”
The experiment took place in March 2017 in the labs of the UIBE in Beijing, China. A total of 317 participants took part in the experiment and were randomly allocated to 11 sessions (each with 17-32 participants). At the beginning of each session, participants received 300 RMB in cash and signed and confirmed that they would obey the rules given by the research staff during the study (see letter of understanding in Appendix 1).
Participants were asked to choose a desk from which to answer the survey, and the instructions were then distributed. Participants were not permitted to communicate with one another. A research administrator and two research assistants were on hand to clarify any questions that arose. Each session lasted for approximately 90 minutes. At first, participants completed an initial questionnaire inquiring into their socioeconomic characteristics and attitudes toward climate change. The purchasing procedure was then explained (see Appendix 2). In addition, participants witnessed a first presentation of a tangible (but unrelated to CO2permits) example of the market mechanism and were asked to fill out a short test as verification of their understanding of the procedure. The explanation of the purchasing rule was included in the instructions. Following this stage, participants received information about (1) climate change and its effects on the environment and human society, including co-benefits from reduced emissions of local air pollutants; and (2) the Beijing or Shenzhen ETS (depending on the treatment). In the information about the ETS, emphasis was on the fact that buying and withdrawing permits reduces the ETS cap and thus CO2 emissions.
Finally, participants were informed that they had the opportunity to buy permits in 1 t CO2units with their own money and could therefore contribute to the overall reduction of CO2emissions. Participants were reassured that all transactions would be carried out and that the final purchases and withdrawing of permits would be announced on the UIBE webpage.7
To make individual CO2emissions more tangible, participants were provided with a second presentation with three specific examples of activities resulting in emissions of 1 t CO2.8 Thereafter, each participant was asked to indicate whether they would be willing to purchase the permit at the six different prices. Finally, participants completed a second questionnaire answering questions about expectations regarding others’ behavior and the recent price for CO2certificates, general opinions regarding climate policy and social norms. After the public price draw, participants left. Subjects who had announced purchases of 1 t CO2permits paid the corresponding amount of money they had stated in the survey.
In T-BJ, subjects were given the opportunity to buy 1 t CO2 in Beijing ETS at six different prices, in randomly allocated scenarios with higher prices (“high”) and lower prices (“low”) (see Table 1). In T-SZ, analogously, 1 t CO2from Shenzhen ETS was sold. Subjects only took part in one treatment (“be-tween-subjects design”).
Number of Respondents in Each Treatment
The total quantity of permits purchased by the participants equated to 60 t CO2 (including Beijing ETS 55 t CO2 and Shenzhen ETS 5 t CO2). This amount of permits was bought and then deleted.9 The revenue collected by those subjects who completed transactions totaled 1,184 RMB. The process was published at the UIBE webpage.
4 Hypotheses
We start the discussion of the hypothesized behavior with T-SZ. Since subjects living in Beijing are barely affected by co-benefits in Shenzhen, contributions in this treatment are in practice solely motivated by climate change concerns (see our stylized theoretical framework in Section 2). Standard economic theory based on selfishness predicts zero contributions to the global public good from climate change mitigation, as marginal benefit of contributing is virtually zero and costs are positive. Accordingly, the proportion of subjects who buy certificates, pcert, is 0. However, there is considerable empirical evidence from previous revealed preferences studies (Löschel, Sturm, and Vogt 2013; Diederich and Goeschl 2014, 2017; Uehleke and Sturm 2017) and the literature on donations (e.g., Andreoni 1990) showing that contributions in such decision situations are positive. On the one hand, positive contributions can be explained by moral motivations, which are associated with contributing to the public good rather than with the effect of the contribution (Cooper, Poe, and Bateman 2004). For example, subjects could receive a warm glow of giving (Crumpler and Grossman 2008), could buy moral satisfaction instead of ascribing an economic value to the public good (Kahneman and Knetsch 1992), gain from a positive self-image (Johansson-Stenman and Svedsäter 2012), or follow deontological decision rules that cause them to disregard consequences and instead decide on the basis of morally mandated duties to “do the right thing” (Spash 2006). This behavior can be described as unconditionally cooperative. On the other hand, it is possible that some subjects are willing to contribute only under the condition that others also do so (e.g., Sugden 1984; Fischbacher, Gächter, and Fehr 2001). In our design, subjects had to build their own expectations regarding the behavior of other subjects; consequently, only those conditionally cooperative subjects who expected that others would also “bear their share” would contribute. However, because free-riding within the group is possible, strong incentives exist to understate the individual demand for the public good.
Based on these considerations, we can state our first hypothesis H1 regarding, the proportion of subjects who buy:
Hypothesis H1: 
versus 
.
Because of the local co-benefits from climate change mitigation in T-BJ, subjects in this treatment should have an additional incentive to contribute compared with T-SZ. This effect is also illustrated by our theoretical considerations in Section 2. However, the provision of cleaner air as an example for local co-benefits represents a (local) public good and the marginal benefit of contributing is virtually nothing in this decision situation. Thus, purchase decisions in both treatments are subject to strong free-riding incentives, and it is an empirical question whether and to what extent subjects react to the treatment effect. Therefore, we derive our second hypothesis H2:
Hypothesis H2: 
versus 
.
In the case that subjects contribute, we nevertheless would expect that the law of demand holds, that is, the price should have a negative effect of the proportion of subjects who buy. This is our third hypothesis H3, which holds for both treatments:
Hypothesis H3: H0 : no price effect on pcert versus HA : pcert decreases with price.
5 Results
Pool of Participants and Their Environmental Attitudes
Appendix Table A3_3 panel a-k present the participants’ socioeconomic characteristics. Our subject pool covers all age groups from 18-75 years for men and women. The sample is characterized by an underrepresentation of male subjects in general and subjects in the age group 40-49 years. Furthermore, subjects with higher education (undergraduate or higher) are overrepresented in our sample.10
Appendix Table A3_1 panel b presents participants’ attitudes toward climate change (see also Appendix Table A6_4). Thirty-nine percent of our subjects are concerned about human-induced climate change. Meanwhile, 75% of the sample are concerned about local air pollution caused by pollutants in the north (including Beijing), but only 18% are concerned about local air pollution caused by pollutants in the south (including Shenzhen). This difference is statistically significant (Wilcoxon matched pairs test, p-value<0.001).
Univariate Analysis of the Treatment Effect
In a first step, we compare individual behavior in T-BJ and T-SZ by defining two types: (1) subjects do not buy for any price (“no contribution”); and (2) subjects buy for at least one price (“contribution”). Based on the distribution of types (see Figure 2), we can state that in both treatments the proportion of subjects who contribute is clearly above 0. Furthermore, the proportion of subjects who contribute significantly increases from 44% in T-SZ to 64% in T-BJ (exact Fisher test, p-value = 0.001). Thus, we can reject our null hypotheses in H1 and H2.
Types of Individual Behavior
In the next step, we analyze subjects’ implicit WTP. For this purpose, we denote the highest price a subject i is willing to pay as minimum WTP (WTPmin). For subjects who do not buy at any price, we set WTPmin = 0.11
The descriptive statistics for WTPmin are shown in Table 2 (panel a). Both distributions are highly skewed to the right, and therefore the median is the appropriate measure for the center. While the median WTPmin in T-BJ is 5 RMB, it is 0 RMB in T-SZ. This difference is also significant (two-sided Mann-Whitney U-test, p-value = 0.0199). There is no difference regarding the mean WTPmin (two-sided t-test, p-value = 0.874). This result shows that for small prices, that is, in the range [2, 5], the change of the treatment from T-SZ to T-BJ has a positive effect on the WTP. To put it another way, the median WTP for permits is completely determined by the preference for local co-benefits of CO2 emissions reduction.
WTP Measures
In the third step, we calculate the share of buyers of certificates per price in each treatment. Figure 3 shows the proportion of subjects who buy for the 12 different prices. The demand curves do not decrease monotonically, but the fitted values show a clear downward trend as prices increase. The difference in proportions between the treatments is quite large for low prices, particularly for the smallest prices of both price vectors, that is, for the price range [2, 5]. Here, the proportion of subjects who buy in T-BJ is about 20 percentage points higher than in T-SZ. For larger prices, that is, in the range [9, 300], the demand curves overlap for many prices, suggesting that the treatment effect is only valid for low prices. In general, the proportion of subjects who buy for prices above 45 RMB is quite low in both treatments.
Price and Proportion of Subjects Who Buy
We test the null hypothesis of independence between the purchase decision in T-BJ and T-SZ with the Fisher exact test for count data (see Appendix Table A6_2). For all prices in T-BJ, in 22.1% of all cases subjects purchase certificates, compared with 17.9% in T-SZ (p-value = 0.042). Testing at the individual price level leads to a differentiated picture. For the smallest price of both price vectors, we can reject the null hypothesis at least at a 10% level of significance (at P = 2 with p-value = 0.040 and at P = 5 with p-value = 0.054). Thus, there is weak statistical evidence that for low prices in the range [2, 5] subjects purchase permits more often in T-BJ than in T-SZ. For [9, 300] there are no significant effects. Because observations for P = 2 and P = 5 belong to different price vectors and thus are independent, we can jointly test whether the null hypothesis of independence between the purchase decision in T-BJ and T-SZ can be rejected. From the 173 purchase decisions in this price range, 133 (76.9%) took place in T-BJ and 40 (23.1%) in T-SZ. Of the 144 no-buy decisions, 89 (61.8%) occurred in T-BJ and 55 (38.2%) in T-SZ. For both prices, the null hypothesis of independence between the purchase decision in T-BJ and T-SZ can be rejected at a p-value = 0.004 (Fisher exact test).
We can summarize that the proportion of subjects who buy is positive for almost all prices and quickly decreases with price. Null hypotheses in H1 and H3 therefore must be rejected. Furthermore, in the price range [2, 5], the proportion of subjects who buy is significantly higher in T-BJ than in T-SZ, meaning that the null hypothesis in H2 is rejected for low prices.
In both treatments, T-BJ and T-SZ, we asked subjects at each price about their expectations regarding the share of all other participants they believed would purchase the permit at the respective price levels.12 Figure 4 shows that the mean percentage of individual expectations in T-BJ is constantly above the mean in T-SZ.13 Furthermore, mean expectations decrease with price.
Expectations Regarding the Percentage of Other Subjects Who Buy
For the price range [2, 9] mean expectations in T-BJ are significantly higher than in T-SZ at a p-value<0.05 (Appendix TableA6_3). The price range broadens to [2, 27] if we accept a 10% level of significance.
Econometric Analysis
This section presents logit models to estimate treatment and covariate effects on the probability to buy the certificate. Therein, the purchase of certificate (yes/no) is the dependent variable. Appendix Table A6_4 summarizes the socioeconomic covariates of the following models. The covariates contain standard demographic variables such as gender, age, income, and education. We also include dummy variables for religion and risk preference, membership in the Communist Party, and having children below 6 years and between 6 and 18 years. Furthermore, commuting time is included. We control for individual attitudes toward the environment and climate policies. Thirty-nine percent of respondents are concerned about climate change. Seventy-five percent (18%) are concerned about pollution in the north including Beijing (south including Shenzhen). Twenty-two percent of respondents agree with the statement “It is pointless to try to do something against climate change as an individual.” We use this statement as a proxy for dilemma awareness, which measures the degree to which the sample is aware of the social dilemma of emissions reductions as dilemma awareness has been found to affect WTP for public goods (Liebe, Preisendörfer, and Meierhoff 2011; Uehleke and Sturm 2017). We measure the degree of pro-environmental behavior with the Personal Norm Scale (Stern et al. 1999; Steg, Dreijerink and Abrahamse 2005; Steg, van den Berg, and de Groot 2013), which explains support for proenvironmental action.
The question wording and the scale properties are given in Appendix Table A3_3( panel e). Finally, 42% stated that they trust that the ETS is fit to reduce CO2emissions.
The results for the logit models are presented in Table 3 (panel a), where coefficients are presented as odds ratios. Model 1 includes only the price as an explanatory variable. Model 2 adds a dummy variable for T-BJ (the reference is T-SZ in this case). In model 3 an interaction dummy variable for T-BJ and prices in [2, 5] is added. Model 4 adds socioeconomic characteristics and various environmental attitudes.14 Overall, the logit results confirm the univariate results of the treatment influence, meaning that we have to reject the null hypothesis in H2 for low prices. For prices in [2, 5], the odds ratio of being in the T-BJ group over being in the T-SZ group is 4.407, indicating that the odds of buying a certificate are more than three times higher in the T-BJ group than in the T-SZ group for this price interval, when all other variables remain constant.
Econometric Results
Furthermore, in model 4 we find evidence for the factors underlying the decision (ceteris paribus for p-value<0.05). First, for individual decisions, the price of the certificate reduces the odds of buying by 2.8% for each additional RMB. Second, increasing individual risk attitude has a positive effect on the purchase probability. With each additional point on the risk scale, the odds of buying increases by 26.7%. Subjects who trust in the ETS have 1.785 times higher odds in making a transaction than subjects who do not trust in the ETS.15 Finally, subjects with children aged 6 to 18 years have 69.6% smaller odds of buying than subjects not in this group. To analyze whether subjects who are particularly concerned about the pollution in the north behave differently, we generated the variable excess concern for pollution in the north (“ex. conc.poll.north”), which is the difference between the concern for pollution in the north (including Beijing) and in the south (including Shenzhen). Interestingly, the interaction term between ex.conc.poll.north and the Beijing treatment dummy variable, T-BJ, does not have a statistically significant effect on the purchase decision.
The results of our regression analysis (model 4) can be supported by literature on the issue. The observed effects for price and trust in ETS are qualitatively consistent with similar studies executed with EU ETS permits (e.g., Uehleke and Sturm 2017). Meanwhile, the empirical evidence regarding the effect of risk attitude on contributions to environmental goods is limited. Contrary to our results, Bart-czak et al. (2016) find that risk seekers contribute less to the local environmental good species protection. Given the literature, the insignificance of effects for the variables dilemma awareness and personal norm and the environmental concern variables is surprising. The negative effect of having older children on the purchase decision might be explained by the fact that those subjects have a tighter budget constraint. The insignificant effect of the interaction term between the excess concern for pollution in the north and the Beijing treatment variable suggest that other motives, not related to local environmental concerns, could also have an effect on the purchase decision.
To illustrate the identified effects in Table 3 (panel b) average marginal effects on the probability to buy are presented. For the price range [2, 5] in T-BJ, for example, subjects on average have a 16.5 percentage points higher probability to purchase a certificate than in T-SZ.
Elasticity
Based on the logit estimates presented earlier and following the analysis of Diederich and Goeschl (2017), we calculate the corresponding average marginal effects and the average elasticity of the probability of purchasing with respect to the price (LeClere 1992). The elasticity of the probability of purchasing captures the change in the probability to buy certificates caused by a 1% change in price and is calculated as follows:
[5]
where Yi is an indicator variable that takes the value of 1 for a contributor and 
is the marginal effect of a price increase on probability. An elasticity below 1 (in absolute value) then describes a less-than-proportion-ate change in the probability relative to the price P, and vice versa.
In Table 3 (panel c), the results are shown for the specification with the price as the only explanatory variable.16 For all observations, the absolute effect of the marginal effect (1 unit = 1 RMB) decreases across the price range. This is consistent with the purchase behavior depicted in Figure 3. A price increase by 1 RMB has a rather strong effect on the probability to buy when the price level is low compared with a broader price range including higher prices. For T-BJ, the marginal effect of a price increase on the probability to buy is stronger than for T-SZ in all price ranges. This observation is also in line with Figure 3 as the proportion of subjects who buy is much higher in T-BJ than in T-SZ at low prices while both proportions converge quickly to 0 as prices increase.
Across the whole price range, the elasticity is estimated at ηPr=−2.24. A 1% increase in price on average leads to a decrease of the purchase probability by 2.24 percentage points. Thus, we observe an elastic price reaction. Starting from low prices in the range [2, 14], the absolute value of the elasticity increases, that is, the price reaction becomes more elastic. This effect is caused by the observed pattern of purchase decisions (see Figure 3). For low prices in [2, 14], the probability to buy is in the range of 50%. A 1% increase of the price causes an inelastic reaction in this case. For high prices the probability to buy quickly reaches values near 0. A 1% increase in the price therefore causes an elastic reaction. As expected, the absolute value of the elasticity is higher in T-BJ than in T-SZ for all price ranges. Overall, the observed elasticity for T-BJ is ηPr=−2.65 and for T-SZ ηPr=−1.47.
It is interesting to note that in a similar setting, Diederich and Goeschl (2017) report an elasticity of ηPr=−0.3 across the entire price range for a German sample. It follows from this observation that rising prices will not matter much for the demand for voluntary CO2 reductions. In contrast to the inelastic price reaction observed in Germany, our subjects in T-SZ, where only the public good characteristic should matter, exhibit price elastic behavior. Obviously, the demand reaction to price increases of a global public good in emerging economies such as China is much more price sensitive than in developed economies such as Germany.17 When interpreting these results, one should take into account that in Died-erich and Goeschl (2017), only one price was presented to the subjects, that is, each subject faced only a take-it-or-leave-it offer at a fixed price. While monetary incentives should eliminate possible “psychological biases,” we cannot rule out that the observed difference in elasticities is caused by design differences.18
Additional WTP Estimates
To confirm the robustness of the results obtained so far, we present additional WTP estimates, which should be interpreted as lower bound because in both treatments strong free-riding incentives exist. To estimate the WTP from dichotomous responses we use the lower-bound Turnbull (LBT) estimator (see Turnbull 1976; Haab and McConnell 2003), which is a nonparametric estimation method. The advantage of this approach is that it relies only on the respondents’ information, namely, that the WTP is at least the presented price if the certificate is purchased. The calculation of the LBT estimator is explained in Appendix 5.
The WTP estimates are shown in Table 2 (panel b) for all observations and for both treatments. The mean WTP for T-BJ (11.39 RMB) is only slightly larger than for T-SZ (11.10 RMB), and both confidence intervals clearly overlap. In addition, due to the low number of purchases at higher prices, the standard deviation is rather large (see Appendix 5), and consequently the confidence intervals are rather broad. The median, which is robust to extreme observations, is much lower than the mean in both treatments. The reason for this observation is that the mean for the treatments is biased to the right by a few purchases at the highest prices above 70 RMB (see Figure 3 and Appendix 5). Therefore, we focus on the median here. The median is in the price range [2, 9] for T-BJ and in [0, 2] for T-SZ. This difference is mainly driven by the large difference in acceptance rates at low prices (see Figure 3). According to the linearly interpolated median in T-BJ, 50% of the subjects have a WTP of 4.95 RMB or more, while in T-SZ 50% of the subjects have a WTP of only 1.73 RMB or more. Thus, we can state that based on the linearly interpolated median in T-BJ, 65% of the WTP for a certificate is driven by a preference for local co-benefits and 35% by a preference for reducing climate change. Obviously, the weights of the preferences for local co-benefits on the one hand and global CO2emissions reduction on the other hand depend on the chosen metric (see our earlier calculation. Based on our results, we can conclude that regarding the median WTP the preference for local co-benefits (such as the reduction of local air pollution) seems to be more important than the preference to avoid global climate change.
6 Conclusion
China, one of the world’s largest CO2 emitters, struggles with severe environmental problems such as local air pollution. Because of the link between CO2 emissions and especially local air pollution, it is often suggested that the country have additional private incentives, the co-benefits, to contribute to CO2 reduction. Co-benefits materialize via different channels. On an individual basis, private co-benefits of local CO2reduction might also be linked to moral motivations. In this article, we present the first experimental evidence on these co-benefits from climate change mitigation, which are observable in real individual decisions. We use the fact that in China several separate subnational ETSs exist. In our experiment, we sell permits from Beijing and Shenzhen ETSs to a sample of subjects from Beijing. Both regions, are sufficiently unrelated to avoid significant co-benefits in Beijing caused by emission reductions in Shenzhen. Our design allows us to separate the demand for local environmental protection on one hand and mitigating anthropogenic climate change on the other. Our core result is that Chinese subjects have a positive demand for climate change mitigation and—at low prices only—an additional positive demand for local environmental protection as the latter generates local co-benefits. In our experiment, the sample shows a locational preference or, in other words, a “home” bias regarding climate change mitigation. The demand reaction to the CO2price has the expected negative sign and is relatively elastic. Interestingly, subjects expect the observed demand behavior regarding price and treatment effect. From a regulatory perspective, our results may also help explain China’s active role in recent climate policy: China not only benefits from mitigating climate change, it does so from the associated local co-benefits. This observation is in line with the growing economic literature on the strong effect of local air pollution in emerging countries on health and well-being.
Three qualifications are in order regarding the policy implications of our results. First, there are alternative explanations for the observed WTP differences. For example, besides air pollution and other environmental co-benefits, subjects living in Beijing may have a locational preference for CO2 abatement at home for other reasons (such as higher level of trust in their local governmental officials and the local ETS than in officials in an unfamiliar location). Our experimental design captures an overall effect and is not able to identify the exact channel at work. However, our experimental framing reflects the high attention the effects of local pollution receive in China and strongly supports the co-benefits interpretation. Second, the identified value added from local co-benefits seems to be relatively small and is limited to low prices only. In addition, we must also take into account that our design frames reducing CO2 emissions as a primary benefit and the reduction of local air pollution as a possible co-benefit only. Interestingly, even under this framing we were able to show a treatment effect. For the median WTP the preference for local environmental protection is even stronger than the preference for contributing to climate change mitigation. It might be the case that the additional demand for local environmental protection is higher when reducing CO2emissions is treated as the side benefit and local effects are listed as the primary benefit. Thus, the effect identified in our study can be seen as “lower bound” for the true treatment effect.
Third, because we measure individual demand decisions in a public good context with strong free-rider incentives, the data gathered do not reflect the “true” demand or WTP for the provision of the public good. We would expect that due to the massive local air pollution in Chinese cities, the marginal benefit from reducing local air pollution is much higher than that from mitigating climate change for Chinese subjects. Furthermore, we can assume that the individual demand for the global public good, and thus the WTP, depends on the decision of other subjects and on the level of collective action at the regional, national, and international levels. Given the public good characteristic of our local co-benefit assessment for individuals, we might observe only a fraction of the valuation. Our results show that even under complete absence of collective action, Chinese subjects are nevertheless willing to sacrifice some of their own disposable income to mitigate climate change and provide local co-benefits such as reduced air pollution. The recent investments in air purifiers in private homes and cars, for example, indicate that the individual valuation for clean air is indeed quite substantial (see Ito and Zhang 2020) if the benefits can be privatized. This indicates the potential role of regulating local environmental problems in China.
Acknowledgments
We thank the editor and two anonymous referees for constructive suggestions and comments. We are also indebted to workshop participants at Beijing Normal University, Tianjin University, UFZ Leipzig, ZEW Mannheim, and to Martin Achtnicht, Carlo Gallier, Andrea Gauselmann, Ian Mills, Zhongxiang Zhang, and Yexin Zhou for valuable comments. Financial support by the National Natural Science Foundation of China (no. 41675139; 72042003), the Program for Excellent Talents, UIBE (no. 17JQ11), and the 111 Project (B18014) is gratefully acknowledged.
Fred Engst and Hongyu Pan are gratefully acknowledged for their help with the developing of the sampling method. Thanks also go to Wenzhan Li, Yongjie Liu, Jiatong Jiang, Zhuqi Shen, Linshu Wang, Shuai Wang, and Jiawei Zhang for their dedicated work sending out the invitation letters, receiving phone calls and emails, and related assistance. Further, we used the WeChat as a dissemination tool to call for more participants, and one-fourth to one-third of subjects are enrolled through this channel. After registering the subjects through invitation letters, we did a rough estimate for representativeness and found that we seemed to lack male participants between the ages of 40 and 55. Then, we posted the invitation letter online (a function can be used in WeChat, namely “pengyouquan”) and made it explicit that the targeted individuals are male between the ages of 40 and 55. It was a supplement to the invitation letter, and the contents were the same as the invitation letter we sent before, with the only difference being the age range. Again, we tried to mirror the age and gender structure of Beijing’s population.
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 As an example: according to World Health Organization data (for 2016) the annual means for PM2.5 and PM10 in Beijing (Shenzhen) were 73 and 92 μg/m3 (27 and 42 μg/m3). In Berlin, the annual means were 16 and 23 μg/m3 (World Health Organization 2019).
↵2 See Sun et al. (2015) for potential source contribution functions for fine particles in China. According to the Ministry of Ecology and Environment (2017), Shenzhen was among the top 10 cities with relatively good air quality, ranking no. 7 in the list of the air quality comprehensive index and primary pollutants of 74 cities in China, with its comprehensive index of 3.49, compared with Beijing (comprehensive index 5.87, rank 56). The GDP per capita of Shenzhen in 2017 was among the top of all cities in China, reaching 183,544 RMB (China Statistics Press 2018b). The GDP per capita of Beijing in 2017 was 128,994 RMB (China Statistics Press 2018a), considerably lower than that of Shenzhen.
↵3 This is the marginal version of the “compensating variation” (e.g., Fleurbaey 2011), which is a key concept for the theoretical foundations of monetary valuation of environmental quality changes (e.g., Bockstael and Freeman 2005; Buchholz and Rübbelke 2019, ch. 3).
↵4 If we assume that subjects in Beijing to some extent have also preferences for local air pollution in Shenzhen, the value of local co-benefits in Beijing is underestimated. The observed difference WTPgi − WTPgj is a lower bound for the individual value of local co-benefits. Furthermore, in this case, the valuation of the global public good component is overestimated.
↵5 We decided to use a second Chinese ETS, the Shenzhen ETS, instead of the EU ETS to avoid any potential biases caused by international aspects. See Löschel, Sturm, and Vogt (2013), Löschel, Diederich and Goeschl (2014, 2017, 2018), Sturm and Uehleke (2017), and Uehleke and Sturm (2017) for revealed preference studies using the EU ETS.
↵6 This mechanism is a modification of the Becker-DeG-root-Marschak mechanism (see Becker, DeGroot, and Mar-schak 1964).
↵8 The following examples for activities generating 1 t CO2 were chosen: (1) a 7,200 km drive with a VW Lavida 1.4 TSI, (2) the electricity consumption of one person in 870 days, and (3) 13.2% of the annual average per capita CO2 emissions in China.
↵9 The real costs for purchasing the 55 t CO2 in April 2017 in Beijing ETS were 55 t CO2 × 39.8 RMB/t CO2 = 2,189 RMB; and the 5 t CO2 in August 2017 in Shenzhen ETS were 5 t CO2 × 24.8 RMB/t CO2 = 124 RMB; totaling 2,313 RMB.
↵10 All comparisons concerning representativeness are based on χ2tests with p<0.05 level of significance. The population of the city of Beijing (census data from 2010) is the population of interest. The fact that our sample is not purely random is uncritical for the comparison of our treatments and corresponding conclusions. It could limit the external validity of our results.
↵11 Subjects who do not behave in an economically consistent manner, that is, who have a partially increasing demand function (4%), are excluded from this analysis.
↵12 In T-BJ, subjects who faced a second decision situation (see the notes to Table 1) were not asked about their expectations.
↵13 We also asked subjects about their ETS price expectations (see Appendix Table A3_2 panel a). There is no significant difference in the ETS price expectations between both regions (Mann-Whitney U-test, p-value = 0.733). Furthermore, 80% (79%) of subjects in T-BJ (T-SZ) indicated that they do not know the ETS price (see Appendix Table A3_2, panel a). Because of this observation and the fact that private access to the Chinese ETS markets is practically impossible, we can assume that field prices did not affect individual purchase decisions in our experiment.
↵14 Models 5-7 (see Appendix 4) include interaction terms between the T-BJ condition and a dummy variable for prices in [2, 14], [2, 27], and [2, 45]. Owing to the higher goodness-of-fit measure, we focused on model 4.
↵15 Overall, 42% of our subjects trust the effectiveness of the ETS (Question B03 in Appendix 2). Although this proportion seems to be low, it is actually higher than the 33% observed in Uehleke and Sturm (2017), who used a similar design for a German sample. If we restrict our sample to the subjects who trust in the ETS, the main result of the logit regression in Table 3 (panel a) (i.e., the treatment effect for low prices in [2, 5] RMB) remains unchanged. The results of these regressions are available from the authors on request.
↵16 See Appendix Table A4_2 for estimates with price and other covariates in model 4 (Table 3, panel a), which are similar to the values presented here for the entire price range.
↵17 For given prices of the environmental good, higher budget shares that are spent on the environmental good are related to a higher price elasticity. However, it is not clear how the prices of the environmental good differ between countries and what climate budget share would follow. It seems reasonable that high-income countries have higher budget shares for climate protection and hence a lower elasticity of CO2 reduction. The health economics literature also reports highly elastic demand for goods such as preventive health care in developing countries (see Dupas 2011). Although the similarity in demand behavior is surprising, we have to take into account that health goods are usually private goods, whereas in our setting, subjects could contribute to a public good.
↵18 In fact, when we calculated the elasticities for both price vectors separately, in Treatment T-SZ, demand for the low price as well as the high price vector is rather elastic compared with Diederich and Goeschl (2017). The elasticity values based on model 1 are ηpr=−2.373 for the low-price vector and ηpr=−1.144 for the high price vector (similar values are obtained based on model 4; results are available from the authors on request).
