“ Green behavioral ( in ) consistencies : are pro-environmental behaviors in different domains substitutes or complements ? ”

Households’ consumption patterns and behaviors have profound influence on natural resources and environmental quality. This paper explores whether environmental behaviors and willingness to pay (WTP) in the household domains transport, energy consumption and water consumption are substitutes or complements. Using a cross-country data set from the Organization for Economic Cooperation and Development Survey on Environmental Attitudes and Behavior from 2008, a random-effects (ordered) probit model is used to answer this question for the following countries: Australia, Canada, France, Mexico, Italy, and South Korea. It is found that in most countries, actual environmental behaviors are substitutes, while WTP for environmental public goods in different domains is mostly complementary. Grounding in these results, policies aiming to encourage overall environmentally friendly lifestyles should therefore be all-encompassing of several public domains, instead of individual ones, to avoid the risk of negative spillovers.


INTRODUCTION
For ease of use we will denote the reduction of a public bad as contributing to the provision of a public good.
Households' consumption patterns and behaviors have profound influence on our natural resources and environmental quality. To reach certain environment-related Sustainable Development Goals (e.g. good health and well-being, clean water and sanitation, affordable and clean energy, responsible consumption and production (United Nations, 2015b), the general public needs to contribute to them. Moreover, private households' contribution to the provision of public goods in the form of environmental amenities is desirable for governments' public policy goals (Lanzini & Thögersen, 2014). Consumption of private environmental goods contributes to private provision of an public environmental goods. For example, choosing electricity based on renewable energy sources (RES) over conventionally generated electricity reduces greenhouse gas emissions (a public bad 1 ).
To encourage the activities of private households to increase their contributing efforts, governments in numerous countries have taken various policy measures (such as taxes, subsidies or public campaigns) to incentivize the public to take environmental aspects into consideration in their consumption decisions. Yet, before incentivizing people to follow a certain behavior, getting to know their behavior in the status quo is the foundation for inducing any behavior changes through policy. Depending on environmental attitudes, households respond differently to policies aimed at promoting energy conservation, water conservation and waste prevention (Brown, 2014;Millock & Nauges, 2010). Therefore, examining the relation of pro-environmental behavior and willingness to pay (WTP) in different areas is an important step in devising policies effective across domains.
The relation of environmental behaviors, i.e. whether they are treated as substitutes or complements, is often analyzed in the context of spillover effects. Spillover effects are changes in a non-targeted behavior as a response to a behavior targeted by a program. Positive spillover effects occur if a pro-environmental behavior in one context leads to more pro-environmental behavior in another domain. Such interventions that lead subsequently to positive spillover effects will occur under complementary pro-environmental behavior. If interventions on environmental behavior lead to less pro-environmental behavior in another domain, then this constitutes a negative spillover effect -behaviors are substitutes. According to the classification by Nilsson et al. (2017), spillover effects can be behavioral (engaging in a specific behavior affects the probability of another specific behavior), temporal (a behavior affects the probability of engaging in the same behavior at another point in time) or contextual (a behavior in one context affects the likelihood of conducing the same behavior in another context). While we also draw on the literature related to the latter two, the focus of this paper is on behavioral spillovers. Recent literature points to altruism and warm glow effects in environmental behavior and the wish for consistency in behavior and attitudes shapes consumers' behavior (see also Festinger, 1957;Thögersen, 2004).
This paper seeks to answer questions on the interrelation between pro-environmental behaviors and WTP for environmental goods in different household domains. In particular, the question of whether different environmental behaviors or consumption of environmental goods can be regarded as substitutes or as complements will be investigated.
While the concept of spillover usually has been applied to studies of interrelated behavior changes induced by incentives, this paper focuses on a series of self-reported behaviors and responses to price changes of environmental goods. The overarching perspective over several household domains that we apply allows us to draw conclusions about the relationship of several seemingly unrelated goods while taking into account their environmental characteristics. This is also one of the major contributions of this paper. The large body of research in the literature focuses on the interrelation of environmental goods within the same domain. But there is little documentation on the occurrence of behavioral spillovers across domains and the type of interrelation (Dolan & Galizzi, 2015). Complementary to that, this paper turns to the subjects of the behavior, the consumers, and asks them how they would react, without treating them with experiments. Additionally, the scope of the survey, covering Organization for Economic Cooperation and Development (OECD) countries on four continents, greatly exceeds the traditionally smaller scope of studies in this research area.
In particular, we focus on consumer preferences, choices and WTP for renewable electricity, clean transport and water quality, and address these issues empirically using a unique data set generated by the OECD Households Survey on Environmental Attitudes and Behavior in 2008. This survey asked about 1,000 households each in eleven countries a total of 90 questions on attitudes and behavior in five environmental key areas. This gives us unique possibilities to compare environmental behavior across several OECD member states 2 . Specifically, we concentrate in this paper on the provision of the public goods "lower carbon emissions", as provided by households' behavior and consumption choices of renewable electricity and of transport, and on the provision of "clean water for all", as provided by

PREVIOUS RESEARCH
The goods that we are considering (fuel consumption for transport, electricity from RES, tap water quality) can be characterized as impure public goods, as they have characteristics of both private and public goods. The markets for environmental goods tend to be markets for impure public goods, since consumers buy a good that is both private and public in nature. This is because consumers derive utility from the private good that they consume, but they also buy the good's public good characteristics (e.g. carbon emission reduction, clean water), which are basically non-rivalrous and non-excludable.
With impure public goods consumers may additionally derive utility from their own contribution to the public good (e.g. Kotchen, 2006;Andreoni, 1990;Wichman, 2016). Yet, the evident behavior for a profit maximizing individual would be to freeride on other people's contributions. Supporting this, among others, Carbone and Gazzale (2017) found in experiments that effort contributions to a public good decreased when the share of peers who contributed with money to the public good increased. The following subsection will set the possible reasons for consumers to behave pro-environmentally and to consume environmental goods, followed by a presentation of literature on interrelations of environmental behaviors.

Private investments in environmental goods
Other than being subject to government regulations, nudges and influential framing (Croson & Treich, 2014), motives for consumers to invest in public goods are influenced by environmental knowledge, attitudes or values, and range from social expectations to altruism and status (see e.g. Truelove et al., 2014;Kollmuss & Agyeman, 2002). Impure altruists derive a warm glow from contributing to a public good (Andreoni, 1990). Welsch and Kühling (2009) investigate the modes of choice for pro-environmental consumption, in particular, the role of reference groups and routine behavior. They find that highly visible proenvironmental elements serve as status symbols among peer groups. Repeated pro-environmental behavior, irrespective of the reason, seems to enhance this kind of behavior. Also, Whitmarsh and O'Neill (2010) show that self-identity is an important predictor (though not the only one) for proenvironmental behaviors, whereas Kashima et al. (2014) argue that "environmental striving" might be a better predictor of environmental behavior, particularly more costly behavior, than environmental identity. Gneezy et al. (2011) find, in a field experiment, that costless pro-social behavior 3 does not indicate that subsequent actions will also be pro-social. However, costly pro-social behavior consistently leads to positive spillovers. They argue that this kind of behavior can be seen as part of a self-identity. The importance of low-cost vs. high-cost (or effort) behavior for subsequent behavior has also been recognized by, for example, Lanzini and Thögersen (2014), Gneezy et al. (2011), Diekmann and Preisendörfer (2003). Diekmann and Preisendörfer (2003) show that environmental behavior is determined by the law of demand and that the effect of environmental attitudes on pro-environmental behavior decreases with costs and inconvenience involved in the actual environmental behavior.
Economic attempts at explaining pro-environmental behavior trace their roots back to the characteristics demand theory suggested by Lancaster (1966) in the "New Approach to Consumer Theory", which states that when consumers buy goods, they do not actually derive utility from the goods themselves, but from the goods' characteristics. This concept can be applied to consumers buying a bundle of goods, some of which are certain environmental goods.

Literature on interrelations of environmental behaviors
Most research on interrelations of environmental goods and behaviors is restricted to one specific domain, for example, the electricity consumption of private households. Yet, the missing knowledge about both, interrelations as well as spillovers of pro-environmental behaviors between different domains has been acknowledged by many authors (e.g. Thögersen & Ölander, 2003;Stern, 2011;Dolan & Galizzi, 2015).
From the existing literature, interrelation of pro-environmental behaviors are mainly inconclusive. Thögersen and Ölander (2003), having asked whether pro-environmental behavior spreads through other areas in a "virtuous cycle", find evidence for both negative and positive spillover effects of pro-environmental behavior in one domain to other domains, all of them very small though. The effect is stronger though 4 Empirical attention has been paid to willingness to pay for certain environmental attributes or behaviors and incentives for participation in green markets (i.e. markets for environmentally friendly goods and services, see e.g. Kotchen and Moore (2007), Kotchen (2006), Cummings et al. (1994)).
among individuals with strong values on environmentally friendly behavior. Specifically, recycling was found to spill over into buying organic products and using a bike/public transport comparatively more, and buying organic food was negatively correlated with recycling during subsequent years. Thögersen (1999) focuses on two specific pro-environmental behaviors (recycling and packaging waste prevention) and finds that they are not independent from each other. Environmentally friendly behavior from recycling results in a negative spillover effect to attitudes towards related activities, according to Thögersen (1999), most likely because recycling is used as an excuse to not make any bigger, more radical or costly changes to lifestyle. Thögersen (1999) argues that the behavior spills over into another domain without apparent cognitive process. Mazar and Zhong (2010) conducted experiments that showed that exposing people to green products (in an online store) makes them more likely to behave pro-social (in a moral or ethical sense), while purchasing green products has negative effects on their altruism, thus suggesting "that consumption is connected to social and ethical behaviors more broadly across domains than previously thought". In fact, they show that buying green products seems to buy them the "right to violate social norms". Similar results have also been found, for example, by Besides these inconclusive study results, a number of other results indicate positive or complementary interrelations between several environmental goods. According to Margetts and Kashima (2017), positive interrelations between pro-environmental behaviors occur when they require similar resources to perform them. Lanzini and Thögersen (2014) analyzed spillovers from green purchasing behavior encouraged by praise or monetary incentives to other, self-reported pro-environmental behavior. A positive spillover was revealed, affecting mostly low-cost behavior. Martínez-Espiñeira et al. (2014) use a multivariate probit model on crosssectional data on Spanish households to find the determinants for the likelihood of engaging in any of three binary behavior variables (conserva-tion habits) and three binary investment variables (investment into efficient appliances), taking into consideration possible correlations between these variables. They find positive and significant correlations between almost all the decision variables, suggesting the existence of unobservable characteristics that drives all pro-environmental behaviors. The hypothesis is supported by Lange et al.
(2017) who find rather complementary interrelation between emission off-setting behavior and a number of pro-environmental behaviors.
Thus, socio-demographic, economic characteristics, internal motivation and knowledge, as well as characteristics of the behavior, influence the relation between pro-environmental behaviors.
Evidence for the interrelation of such behaviors is mixed, with large support for complementary in behavior, yet numerous effects working towards the contrary direction.

Theoretical background
To illustrate the issue of positive vs. negative spillover effects of environmental behavior and willingness to pay for environmental goods, we next describe two extreme cases. One, where two public goods are perfect substitutes and one where two public goods are perfect complements. Since the objective here is to isolate the effects of substitutability and complementarity, we assume that consumers do not regard two public goods as completely independent from each other. The models are here intentionally kept simple to illustrate a few points. We argue that more complex utility functions only add complexity without adding intuition about how environmental behavior might be discerned from the data.
First, let us assume that an individual has a utility function for two perfectly substitutable public goods (Z 1 and Z 2 ) that are implicitly bought when buying two private goods (x 1 and x 2 ). For demonstrative purposes, let us assume that consumption of all other goods is kept constant and that the relationship between good xi and the public good zi is multiplicative so that z i = 0 when x i = 0. This is intended to demonstrate a situation where the public good does not enter the utility function unless the associated private good is bought. It also turns the public good into an impure one, since consumers care about their own provision of the public good. The utility maximization problem for a representative consumer can then be formulated as follows: where m is income net of consumption of all other goods. The indirect utility function can then be written as ( ) 12 12 12 , , , max , .
x xx Zm Z m VZZ pm pp This implies that since goods Z1 and Z2 are perfect substitutes, consumers will choose whichever of the two public goods together with its normal good gives them most utility. When the two public goods are perceived as equal, this depends on the price of the private good. In the case of equal prices, i.e. when p x1 = p x2 = p, any combination of the two public goods is optimal, ( ) ( ) 12 1 /, since any combination of the two gives the same utility.
Moving from the representative consumer to a sample of consumers, the relationship between the two public goods that are substitutes is therefore negative. If more is spent on the good associated with the public good Z 1 , less needs to be bought of the private good associated with the public good Z 2 to achieve the same level of utility. The implication of this is that the valuation for one public good should decrease with the valuation of another substitute public good. This is graphically demonstrated in Figure 1a with a hypothetical data set. Consumers will then choose their consumption bundle such that 11 2 2 , Zx Z x = so that their consumption of one good is matched equally by consumption of the other good. A complementary public good is then valued higher with improved provision of another public good. Moving from the representative consumer to a sample of consumers, the relationship between the valuation of two complementary public goods is therefore positive, as illustrated schematically in Figure 1b, where the dots represent individual observations in a sample of hypothetical consumers.
We relate a positive relationship between the demand of the two public goods as in complements to a positive relation between WTP for them or to a positive relation between the quantity of pro-environmental behaviors connected to the two public goods. In other words, in this case, the two public goods are likely consumed together their probability of being consumed is equal. A negative relationship between the demand for the public goods represents a negative relation between WTP for the goods or a negative relation between pro-environmental behaviors. Two goods that are substitutes will then differ in their likelihood of being consumed at the same time.
In economic terms, sensitivity of consumers' WTP to the order of goods purchased is also related to the order of income changes (cf. Johansson, 1987, p. 22). For example, if a respondent pays first for good i, then his income is reduced by WTPi, he is less wealthy. If he is then asked to value a second good j, his base line income is m − WTPi and he will value. This second good assumes lower income, which will result in a reduced WTP for good j. The situation is turned around if the first good to be valued is good j and the second one is good i. But under some conditions WTP will not be affected by the sequence of valuations. These conditions are known as path independence conditions. Total WTP does not depend on the order of questions if cross-price effects on demand are equal, such that ij ≠ This is the case when marginal utilities are constant. Thus, a sufficient condition of path independence is linearity in the utility function with respect to the parameters of interest, since then marginal utilities will be constant. However, a sequencing effect is not very likely to arise in our data set, since valuation questions were asked interspersed with non-valu- Figure 1а. Schematic illustrations of demand for two public goods, if they are substitutes ation questions. Additionally, the money spent on the three domains represents a small share of consumption. We can therefore assume that respondents considered their original income as base line for answering the valuation questions.

Econometric method
Given the nature of the available data set, we approach the question of consistency of pro-environmental behavior by means of probabilities of engaging in them in different household domains. For this we adapt the approach used by Lange et al. (2017, p. 80).
In a first step, we estimate a binary probit model due to the binary setup of our pro-environmental behavior variables in the three domains. Thus, we can estimate whether probabilities of engaging in above-median pro-environmental behavior in the domains transport and water differ from those in the energy domain, which we use as base line: where the dummy variable i Y takes value 1 if for a particular domain the respondent engages in a number of pro-environmental behaviors above the median, and zero otherwise. We take this to be the observable representation of unobservable overall pro-environmental behavior of * .
i Y Each domain is then identified by a dummy variable it-self, T for transport and W for water, where E (energy) serves as the base and is not included as a regressor. We include ij x as the vector of explanatory variables described in Table 1, including the dummies for the respondents' country with Australia as base line, and ij β is their parameter vector. Since we assume that pro-environmental behaviors might be correlated across domains, we apply random effects model where i u represents unobserved heterogeneity. This is appropriate since the null hypothesis of no unobserved heterogeneity was rejected by likelihood ratio test. The error term ij v for variation within households and domains is normally distributed. Results are reported as average discrete effects in Table B1, column 1.
Second, we estimate the same model for each country, since the results from the first step suggest variations among countries in the relation between environmental behaviors in the domains (reported as average discrete effects in Table B1, Building on the previous, if the public goods involved in transport and energy and water consumption have a positive interrelation (and thus are considered as behavioral complements), we would expect the discrete probability effects of the domain dummies i W and i T to be zero. Then, any given outcome is equally likely to occur for each of the domains. If they are significantly different Demand for good Z 1 We expect respondents who have shown an above-median level of pro-environmental behavior to be more likely to respond with higher WTP in all domains, as an expression of more pro-environmental attitude. The interaction coefficients are therefore expected to counteract the domain coefficients. Limitations in our model choice can arise due to unobserved underlying characteristics and due to the order in which the questions were asked 5 . As mentioned above, distorting effects, in the form of income effects or substitution effects, can appear when several WTP are elicited in a sequence. 5 When there is a possibility of correlations of errors in between the three models, modelling the data in an equation system seems to be the straightforward method of choice. Greene (2008, chapt. 10) presents the formal model of multivariate choice estimations for the bivariate case, but point out that the bivariate model involves a substantial amount of computation, not to speak of equation systems of order three. 6 For a full description of the survey and the data see OECD (2008a). 7 Note that whereas the answer options in the energy and water domain contain only behaviours, the transport domain includes the investment option that respondents "changed a car for another one, which uses less polluting fuel". This option is obviously related to income.

DATA
For the empirical analysis, we use existing data from the Web Survey on Environmental Attitudes and Behavior conducted by OECD in 2008. This unique data set contains approximately 1,000 responses from each of ten OECD countries. The survey contained in total about 90 questions on household characteristics, attitudinal characteristics and the five key areas such as Waste, Transport, Energy, Organic Food, and Water 6 . We focus on such domains as transport, energy and water, since they have a clear connection to environmental behavior and questions were posed in a comparable way. Item non-response was in many cases large. Responses that were not answered or answered with "I don't know" are treated as missing. From this data set, we selected six countries, which had more than 100 complete and usable responses: Canada, France, Mexico, Italy, Australia and South Korea.
As related above, we use two sets of dependent variables: one variable indicating pro-environmental behavior and second one indicating WTP for pro-environmental goods. For each domain, we consider these two variables of interest. For the first, respondents were asked if or how often they perform certain pro-environmental activities; these are the behavioral variables (see Q1.2, Q2.2 and Q3.2 in Appendix A). We recoded them into binary variables where 1 signifies that respondents are engaged in an above-median number of environmental behaviors, and 0 otherwise 7.
For the second, respondents are asked to state their fuel consumption change in response to change in price (transport), and to value the maximum price change they would accept to change the quality of their consumption (renewable-energy-only and improved water quality in the respective domains). These are the WTP variables (see Q1.1, Q2.1 and Q3.1 in Appendix A), which are interval-valued and they are attributed ordinal values. When respondents consider the questions, they are actually asked to consider the consumption good and the underlying public good, i.e. the environmental qualities of the good. While the underlying public goods might be the same or similar, the associated private goods are not. The effect of an increase in fuel price on motorized vehicle transport includes not only the choice of lower emission levels due to lower fuel consumption, but also the choice for preferred vehicle and its convenience. Similarly, the choice between conventional electricity generation and renewable-based electricity generation implies a choice of the amount of pollutant emissions. The choice to pay for cleaner water asks implicitly to value the reduction of pollutants in the water cycle. Whitmarsh et al. (2011) have shown for the public bad carbon emissions that carbon is not a normal part of everyday decision-making. Knowledge, skills, and judgement are misaligned with behaviors. It is conceivable that households engaging in pro-environmental behavior differentiate between these behaviors even if the behaviors are, in fact, related to the same public good. Therefore, although pollutant emission as the underlying environmental attribute of transport and energy is the same, we theorize that respondents still perceive the public goods as different from each other in that they are transmitted through different private goods. The survey primes respondents to consider their environmental behavior, therefore, it is valid to assume that they actually value the public good in their private consumption decision.
As explanatory variables we consider environment related attitudinal variables: an indicator for membership of or contribution to an environmental organization, an indicator for environmental concern relative to other global concerns ranked from one to six, and two indicators if the household has invested into an above-median amount of energy-or water-efficient appliances, respectively. As socio-demographic control variables we consider gender, age, household size, an indicator for a university education, self-reported household income and an indicator for urban residency, since these factors are to be expected and partly have been shown to influence environmental behavior and WTP (see e.g. Kriström   Compared to the sample, the national populations are structurally different in certain aspects. Since household income is self-reported net of taxes and the population mean is household net adjusted disposable income, values might be expected to somewhat differ. More survey respondents than average have a university degree. The OECD's data corroboration admits certain differences in the sample versus the population for some countries. This difference is particularly large and raises concerns about bias only in Mexico. The difference between sample mean age and population mean age is also the largest in Mexico. The deviation of the Mexican sample from the population is due to low internet access compared with other countries (OECD, 2008a). iors and investments. A majority of respondents would not change their fuel consumption if fuel prices increased and about 20% of respondents reported that they would reduce their fuel consumption by less than 10%, between 10% and 20%, and more than 20%, respectively. Seventeen percent of respondents did not answer this question and an equal share responded with "I don't know".
Thirty percent of respondents each would either not pay anything additional or less than 5% additional on their energy bill to only use electricity from RES, while fewer respondents reported willingness to pay higher percentages of their energy bill towards this objective. Fifteen percent of respondents answered this question with either "I don't know" or "I prefer not to answer". For improved water quality, each of the answer possibilities "would not pay anything", "would pay less than 5%" and "would pay between 5% and 15%" received 30% of responses. The remaining responses are split between respondents who are willing to pay more than 16% and more than 30% above their current water bill. More than 50% of responses are missing for this question. Respondent fatigue might account for this. All domains have  in common that responses seem to accumulate along the lower range of the WTP scale. In the analysis below, respondents who reported "don't know" or "prefer not to answer" were treated as missing information.
From the pro-environmental behaviors to choose from, between 44% and 47% of the sample reported engagement in more than the median number of pro-environmental behaviors in each particular domain. A slightly lower share has made investments into energy and water efficient devices during the past ten years. Note that surveyed behaviors in the different domains vary in their environmental impact (turning off the lights has low impact, whereas not using the car for commuting has more likely high impact on emission reductions).

RESULTS
In order to determine whether the relationships between pro-environmental activities and between WTP for goods different domains vary among the countries, we estimated the probit model for each country. Table 4 provides the excerpt of the regression results showing marginal probability differences of engaging in pro-environmental behavior in the specified domains (1), and parameter estimates of differences in WTP for pro-environmental goods (2). The full table of results can be found in Appendix B in Table B1 and Table B2 for (1) and in Table B5 and Table B6 for (2), where the respec-tive first table has energy as base domain and the second table has transport as base domain.
Note in Table 4 that effects that are significantly different from zero indicate substitutes, while those that are non-significant indicate complements. We see that there are differences in the probability levels of most domains in all countries. High engagement in pro-environmental activities in the transport domain is less likely compared to the energy domain in Canada, France, and Italy, whereas they are more likely in Mexico and South Korea. Above-average engagement in water-efficient activities, on the other hand, is more likely in Australia and less likely in Mexico, Italy and South Korea. Past investments into energy-and water-efficient appliances positively significantly affect the probability to engage in above-average pro-environmental activities (Table B1). Contrary to the prevalently substitutional relations of behaviors in the studied countries, WTP seem to be more likely to be complements. Some substitutionary relations exist in Italy and South Korea, higher WTP are significantly more likely in the transport than in energy domain. In Italy and Mexico, respondents are more likely to report higher WTP for water quality than for electricity from RES, whereas in France, this relation is the opposite. Additionally, interaction terms with domain dummies T and W in Table B5 and E and W in Table B6 show relations between WTP for pro-environmental goods and the effects that above-average pro-environmental behavior has on these relations. Notably,

Behaviors
Choices almost all significant interactions have the opposite sign to the pure domain effect, thus attenuating differences in probability between WTP for different domains when above-average pro-environmental behavior is taken into account. In other words, looking for example at France, respondents with above-average pro-environmental behavior in the energy domain show rather complementary WTP for water and energy domains.
In Italy, respondents with above-average pro-environmental behavior in the transport and water domain are more likely to display complementary WTP for water and energy. All interrelations found above for every individual country are summarized in Table 5 and shown graphically in Figures 2 and 3. There are no consistent country differences in the interrelation between environmental goods and behaviors. In Note: S -substitutionary interrelation, C -complementary interrelation, (C) -complementary interrelation if above-average pro-environmental activities.
other words, the relation of pro-environmental behaviors and the relation of WTP in different domains is similar in all countries in the study. The covariates' significance and signs are similar across countries. Household size, gender, education, living in an urban environment, satisfaction with water quality and distance travelled by car are insignificant in all countries. Age is only significant and positive in Italy and South Korea; income is only significant and negative in Australia and France. Environmental concern is positively significant in Australia, Mexico and Italy and membership in an environmental organisation and having taking special measures to buy electricity from renewable sources from the electricity provider positively affect engagement in pro-environmental activities in almost all countries. See Tables B1-B4 for full results of average discrete effects.
Considering the aggregate level of all countries combined, discrete probability effects for each country are given by the country dummies at the end of Tables B1 and B2 (column 1) and  Tables B3 and B4, with Australia, as the reference country 9 .
Relative to Australia, respondents in Mexico are more likely to report above-average pro-environmental activities and in South Korea they are less likely to report high engagement in pro-environmental activities. In Tables B3 and B4, given that Australia has the lowest average 9 Average WTP outcomes were the lowest in Australia compared to the other countries in the sample (see Table 2).
WTP outcomes, respondents in all other countries have a significantly higher propensity to report higher order positive WTP. Also from the first column in Table B1, we find that relative to energy-efficient behavior, respondents are neither more nor less likely to engage in above-median pro-environmental behavior in the transport or water domain. Having invested in water-efficient or in energy-efficient devices in the last ten years has a positive effect on reporting above-average pro-environmental behavior.
The first two lines of results from the ordered probit model including all countries (with random effects fixed at its mean) in Table B3 show that the discrete probability effect of transport and water on reporting zero WTP (outcome 1) is negative, but that there are significantly higher propensities to report higher order WTP (outcomes 3 to 5). At the same time, there is no significant discrete probability effect of transport and water on outcome 2. Two things are interesting to note here: first, the significant difference in probability levels for the outcomes as compared to WTP for renewable-energy-only suggest a substitutionary relationship between the WTP for the three domains. Second, these differences are more pronounced in the extremes, i.e. at zero WTP and at WTP higher than the minimum positive choice (up to 5 or 10%). Thus, when looking at very low positive WTP, environmental goods are treated as com- plements. On the other hand, larger discrete probability difference between higher order WTP in the different domains might indicate stronger preferences of one domain over another, thus leading to treating them as substitutes. Both can be interpreted as an expression of a fixed budget for environmental consumption.
As for socio-economic characteristics, we find that basically all such characteristics have, in general, an insignificant or a very small effect on engaging in pro-environmental behaviors, on reporting positive WTP (Table B1), as well as on the individual outcomes (Table B3). Being female has a small negative effect on higher order outcomes in the ordinal case. As expected, environmental concern, being a member of an environmental organization and having taken measures to buy electricity from renewable sources has positive probability effects on the binary outcomes, as well as on higher order WTP outcomes. Outcomes are not affected by weekly distance travelled by car or by reported satisfaction with tap water. tive predisposition towards emission abatement returns positive WTP for more environmentally friendly vehicles.

DISCUSSION
Our choice of goods with an environmental attribute originates from the clear environmental quality inherent in the choice of these goods such that respondents might be able to make the connection between their stated choice and a certain corresponding environmental quality. Given this premise, of the goods with an environmental public good nature asked about in the OECD survey, the relevant ones were: electricity from renewable sources (instead of conventional sources), fuel for personal car transport, and clean water. The former two relate to the public good reduction of GHG emissions, while the latter refers to the public good clean water access. Usually GHG emissions are referred to as public bad with negative externalities on people. Since our study asks about the opposite, i.e. a reduction of the public bad, we talk about a public good.
The six countries in our sample were chosen particularly, because they represent very diverse countries (within the scope of member states of OECD and those that participated in the survey) in terms of geographic location, natural resource endowment and socio-economic background. When comparing WTP in different countries in this type of analysis, it must be borne in mind that the base line levels and prices of the environmental goods in question vary considerably across the included OECD countries. This influences the empirical effects of complementarity in pro-environmental behavior, since if for instance tap water is already seen as excellent in a country this will effectively truncate the WTP to improve it further, and thus "hide" the true preferences for high quality tap water. To take this into account in this survey, only people reporting that they were not satisfied with tap water quality were asked about their WTP for better tap water quality 10 . Beaumais et al. (2014) find in a cross-country analysis that WTP for better tap water quality is higher in countries in which respondents are to a larger share unsatisfied with tap water quality. Also other water attributes, like an appalling odor or taste, which are perceived as health risk contribute to an increased WTP for improved water quality (Dupont, 2005). In our sample, household satisfaction with tap water quality has only insignificant probability effects on engaging in above-average pro-environmental behavior in all countries. In all countries, irrespective of satisfaction with tap water, the behavior domains water and transport are substitutes. This might be because behavior changes in the water domain, e.g. turning off water when brushing teeth or plugging the sink when washing the dishes, are more effortlessly implemented than changes in the transport domain, e.g. use car-sharing or change a car for another one that uses less fuel. Thus, easy pro-environmental behaviors in the water domain may substitutes for higher effort behaviors in the transport domain.
Only in Canada and France, two countries in which a relatively larger share of respondents has given a positive answer to satisfaction with tap water quality (see Table 2 for shares), behaviors in the domains water and energy are complements whereas they are substitutes in the other domains. For WTP, we cannot identify a regularity among the different countries corresponding to their tap water satisfaction. But for energy and water WTP interrelations are complements apart from Mexico, which also happens to show lowest satisfaction with tap water.
There, water quality might be a driving factor behind the substitute relation we find.
Similarly, if a country already produces a high share of its electricity from renewable sources, and this is common knowledge, the WTP to increase this share could be low. To choose this option, a country also needs to have the option including the technology available to produce larger shares of electricity from renewable sources. Large countries with a less developed public transport system might naturally be more reliant on personal car transport which hides true WTP for the public good associated with lower fuel consumption. To reduce fuel consumption or change behavior, alternatives, like transportation network or low-fuel cars, need to be available and affordable for households. In Canada, more than half of the electricity comes from renewable sources, which is high by international comparison (see Table 6 for renewable energy shares by country). The provincial policies and renewable energy goals vary considerably, so do the shares of electricity from renewable energies over all of Canada's provinces. Australia has an electricity production from renewable sources of 15%, with a high potential for a larger share resulting from more extensive use of solar and wind power plants. Support mechanisms include carbon pricing, a variation of small-scale renewable support schemes, as well as feed-in tariffs in several regions. A government program allows households to pay extra to get electricity from renewable energy sources only. Both Canada and Australia are vast countries in which personal car transport plays a major role. Even in urban areas only a small share of people use public transport for commuting (Statistics Canada, 2017). Fuel reduction by switching to public transport may therefore seem like an unviable option. Public transport networks in Italy, France and South Korea are generally expansive and electricity from RES shares are 34%, 20% and 7%, respectively. In Italy, renewable energy installations are promoted by a guaranteed minimum price and certain plants are eligible for reduced VAT. France has as support mechanisms, among others, feed-in-tariffs and premium tariffs in place. South Korea relies mainly on thermal electricity production from imported fuels, with renewable reaching at most 7%. Increasing the renewables share is thus not only a way to fight climate change, but also a way to reach energy independence. Mexico has a share of about 25% electricity from renewable sources and at the time of the survey had just passed a law for the development of renewable energy and energy transition. A complication with this survey data from Mexico is that internet penetration in Mexico is quite low, therefore the sample is skewed towards urban, rather wealthy households and not representative of the Mexican population. Wealthier households are more likely to have a car instead of relying on local public transport (Guerra, 2015).
We are controlling for electricity from renewable sources and personal car relevance on a household level by including whether a household has already taken measures to invest into electricity from renewable sources and the weekly distances covered by car. Only the former does significantly and positively affect pro-environmental behaviors and WTP in some countries. This falls in line with other activities that can be interpreted as the manifestations of pro-environmental attitudes, i.e. investments into water-or energy-efficient appliances and being member of an environmental organization, or contributing to one, remain as determinants of pro-environmental behaviors and WTP for environmental goods in most countries.
In general, we can see that pro-environmental behaviors are more likely substitutes, whereas pro-environmental WTP are more likely to be complements in all countries. We can interpret WTP, which is a hypothetical willingness, as the theoretically desired lifestyle, which seems to be to pay consistently for all pro-environmental domain -either consistently little or consistently high. However, actual behavioral patterns as reported for the previous show a different lifestyle where pro-environmental behavior in one domain substitutes for non-environmentally friendly behavior in other domains. The exact behaviors referred to can be found in Appendix A.

CONCLUSION
Do consumers in different countries consistently make "green" lifestyle choices and are equally likely to pay more for pro-environmental goods in several domains of their life? Possible spillover effects of environmental behavior have received some attention from behavioral research. However, economic studies of the linkages between environmental behavior and goods across different household domains, and comparisons of the linkages across different countries are still relatively few. This study contributes to filling this gap through a multi-country approach, based on a unique data set assembled by the OECD.
The study uses the random effects probit and ordered probit models on pro-environmental behaviors and WTP for pro-environmental goods in three domains: transport, energy and water. We chose to focus on six of the countries surveyed in the 2008 OECD Survey on Environmental Attitudes and Behavior: Australia, Canada, France, Mexico, Italy and South Korea.
The first main finding is that while past pro-environmental behaviors in different domains are revealed to be substitutes, WTP for pro-environmental goods tend to have a rather complementary interrelation in the countries under analysis. This finding of complementary WTP for pro-environmental goods echoes the positive correlations between pro-environmental investments found by Martínez-Espiñeira et al. (2014) and the partly positive spillover effects found by Thögersen and Ölander (2003) and Lange et al. (2017). A desire to consistently invest in green choices can account for complementarity of WTP for environmental goods, since this is a hypothetical decision, whereas behaviors are reported actual behaviors, which unveil that reality cannot keep up with desires.
Second, and adding to the above, a key finding is that complementary WTP in different domains is more likely for respondents with above-average pro-environmental behavior. Past pro-environmental behavior can therefore be interpreted as an indicator for a more widespread environmentally friendly attitude throughout several domains of life. This is a point of discussion when comparing our mostly low-cost pro-environmental behaviors with conclusions by Gneezy et al. (2011). They find that costless pro-social behaviors do not indicate that subsequent actions will also be pro-social, whereas costly prosocial behavior consistently leads to positive spillovers. While Gneezy et al. (2011) and Thögersen and Ölander (2003) refer to temporal spillover effects, we give support to their findings within the field of behavioral spillover effects.
Since stated WTP could suffer from hypothetical bias, reported behaviors should be a stronger indication of interrelations between domains than stated WTP. Behaviors for most studied countries are revealed to be substitutes. This suggests that policies that focus narrowly on encouraging pro-environmental behavior in one domain risk negative spillovers into other domains. Therefore, policies aiming at kick-starting an overall green lifestyle should be all-encompassing of several public domains. Carbon pricing is a good example of such a policy, since it covers emissions irrespective of domain. Given that membership in an environmental organizations is a significant predictor, environmental education could potentially aid in such a kick-starting process.
Lastly, the interpretations of the results must be taken with some caution given the sample bias in the dataset. Survey respondents had a higher household income, higher education levels and larger household sizes than the population average. This difference was most substantial for Mexico. In our results, there are no major country differences in the interrelation between environmental goods and behaviors; the relation of pro-environmental behaviors and the relation of WTP in different domains is relatively similar in all countries in the study. A more representative study could improve the tentative interpretations in this paper and point towards particularities of individual countries. Our results are contingent on the three domains under study. Other domains of morally or ethically desirable behavior might affect the interrelation 11 .
With a focus on behavioral spillovers, this study does not take into account any time effects. The OECD study used in this paper was followed up with another survey in 2011. Prospective research should consider combining these data sets in a panel-like setup and look into temporal issues of the consistency of pro-environmental behaviors. With the given data, we cannot completely disentangle the public and private aspects of the goods, but since the survey is labelled as an environmental study, subjects are primed to think about the public good aspect in their choices. These issues certainly warrant a more detailed analysis with data sets specifically aimed at eliciting the relationship between pro-environmental behavior in different domains and underlying attitudes.  Note:* p < 0.05, ** p < 0.01, *** p < 0.001, 1 base country: Australia. Note: * p < 0.05, ** p < 0.01, *** p < 0.001, 1 base country: Australia. Note: Base country: Australia, * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Base country: Australia, * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Base domain: energy, * p < 0.05, ** p < 0.01, *** p < 0.001.