Social Science Research 30, 195218 (2001) doi:10.1006/ssre.2000.0694, available online at http://www.idealibrary.

com on

A Multilevel Analysis of the Determinants of Recycling Behavior in the European Countries

Daniel Guerin, Jean Crete, and Jean Mercier
Department of Political Science, Laval University, Quebec City, Quebec, Canada The main objective of this study is to permit a better understanding of recycling as a function of both individual and contextual variables. Our cross-national study aims at examining how differences in national settings and in social and institutional factors as well interact with a series of individual characteristics to inuence engagement in recycling behavior in the 15 countries of the European Union. One of our most important conclusions is that conservation behavior is greatly inuenced by the ecological mobilization in which it occurs. Indeed, the most activists participate nationally in environmental organizations, the more the population is likely to participate in programs such as the separation of domestic waste. Alone, this variable explains about 45% of the variance observed between the European Union countries on public participation in separation and recycling activities. The present contribution is the rst to do so through a multilevel modeling approach that can integrate both individual and contextual variables at the heart of conservation behavior. 2001 Academic Press

The general proposition stating that social or institutional factors can play an important role in translating environmental concerns and attitudes into action has received growing attention from several researchers since the 1980s (Black, Stern, and Elsworth, 1985; Derksen and Gartrell, 1993; Guagnano, Stern, and Dietz, 1995; Brand, 1997). Yet few studies have adopted a comparative approach in order to determine to what extent social context can account for betweencountry differences observed in levels of participation in a variety of proenvironmental behaviors. Based on a cross-national perspective, the present study aims at examining how certain factors in national social/institutional contexts combine with a series of individual characteristics to inuence participation in sorting and separating household waste for recycling in the 15 countries of the European Union. 1 Two major factors can be invoked to account for the paucity of comparative
Address reprint requests to Daniel Guerin, Department of Political Science, Quebec City, Quebec G1K 7P4, Canada. 1 Given the nature of the data available, social context was operationalized at the country level. 195 0049-089X/01 $35.00
Copyright 2001 by Academic Press All rights of reproduction in any form reserved.

196

GUERIN, CRETE, AND MERCIER

research on the contextual determinants of proenvironmental behaviors. First, the large cross-national data sets which are needed to conduct this kind of research are hardly available. Indeed, the rst surveys that enable researchers to introduce large variations in the contextual factors and generate insightful comparisons across a broad range of societies have only been accessible for a short time. Second, the statistical tools suited to handle the complications posed by the multilevel structure of comparative data sets have been developed relatively recently. Since the end of the 1970s, however, important strides have been made in developing the statistical tools suited to surmount the thorny difculties associated with the analysis of multilevel structural data (Dempster, Laird, and Rubin, 1977). In the 1990s, different statistical computing programs have become available for estimating these models: Mln, VARCL, and HLM are three such programs. The latter is a software package (Bryk, Raudenbush, Seltzer, and Congdon, 1988) that has been used mostly in the eld of educational research so far. Nevertheless, any kind of behavioral or social data having a hierarchical structure can be properly analyzed with it. HLM was used in this study in order to perform the statistical estimations needed by the multilevel modeling of recycling. We are mainly interested in the contextual determinants of recycling behavior. Our focus on this environmental behavior in particular was motivated by the emphasis that prior research has placed on a possible inuence of certain contextual variables on recycling behavior. Given the constraints of the Eurobarometer data, our analysis rests on a two-level data structure: the person level corresponds to the national populations of individuals surveyed by the Eurobarometer in the countries of the EU, 2 whereas the denition of the country level coincides with each of the 15 countries of the EU. THE INDIVIDUAL DETERMINANTS OF RECYCLING Much empirical research has been conducted in the past 2 decades to elucidate the determinants of conservation behaviors and more specically of activities such as the recycling of domestic wastes. In a detailed review of research on factors inuencing household recycling behavior, Gamba and Oskamp (1994) noted that person-level variables investigated by prior research include knowledge, environmental attitudes and concerns, motivations, and demographic characteristics. Among prior research on the role of these determinants, several studies aimed at elucidating the relationships between different conservation behaviors and environmental values and attitudes. Yet mixed results have emerged from these studies. For instance, research that used the New Environmental Paradigm framework have found a pattern of positive though modest correlations between
Eurobarometer 43.1 BIS is entitled Regional Development and Consumer and Environmental Issues. The survey was carried out between May 19 and June 26, 1995. Universe was composed by persons ages 15 and over residing in the 15 member nations of the European Union (N 16,300).
2

RECYCLING BEHAVIOR

197

environmental attitudes and behavior (Dunlap and Van Liere, 1978). On the other hand, more recent research has shown that there is a weak congruence between attitudes, values, and behavior (Buttel, 1987; Oskamp et al., 1991; Derksen and Gartrell, 1993). 3 The role of sociodemographic characteristics has also been examined. Once again, those ndings are rather inconclusive. 4 While some of the usual sociodemographic variables have been shown to be important correlates of conservation behaviors by certain researchersfor instance, some have found a positive association between higher education and recycling (Samdahl and Robertson, 1989) or higher socioeconomic status and recycling (Vining and Ebreo, 1990) other researchers have found contradictory results (McGuire, 1984; Oskamp et al., 1991). Regardless of these inconsistent ndings, it is worth pointing out that sociodemographic variables account for only a small portion of the variation in conservation behaviors. Despite the overwhelming bulk of research having used what we can call the individualistic approach in the study of conservation behaviors, one may argue that the explanation potential of this theoretical approach has now reached its limits given the contradictory ndings it has produced. THE ROLE OF CONTEXTUAL FACTORS Another strand of research has tried to link different situational factors to participation in a variety of conservation behaviors. According to CorralVerdugo (1996, p. 668), situational predictors of recycling behavior have been investigated, including convenience, information, availability of conditions for recycling, and presence of other individuals. This type of research was rst conducted to determine organizational factors that facilitate recycling behavior. 5 Geller, Brasted, and Mann (1980) demonstrated that the residents decision to recycle may be inuenced by the availability of storage containers. Other studies have reported that having convenient access to a recycling program mediates the relationship between socioeconomic factors and recycling
Many explanations have been proposed to account for the weakness of this relationship. Dunlap (1989) has pointed to the fact that public concern for the environment and propensity to act in ways that protect the environment may be weakened by the perception that the problem is being solved under the leadership of institutions. According to Gamba and Oskamp (1994), only attitudes or behavioral intentions, rather than actual behavior, have been studied (Gamba and Oskamp, 1994, pp. 592593). Usitalo (1990, p. 223) has asserted that a plausible explanation for attitude behavior inconsistency is the opposition between collective and personal interests: Expressed favorable attitudes toward environmental protection are based on existing information on collective interests and social welfare, while in actual choice situations, the decisions are still based on individual utility considerations. Balderjahn (1988) has found that the various consumption patterns are explained by different set of variables. See Tarrant and Cordell (1997) for a recent account of that question. 4 See Ebreo, Hershey, and Vining (1999) for a recent account of these results. 5 The use of extrinsic incentives to encourage participation has also been studied extensively (see Geller, Winett, and Everett, 1982). Other scholars have studied the impact of persuasive appeals on participation in recycling programs (Hopper and Nielson, 1991).
3

198

GUERIN, CRETE, AND MERCIER

practice (Berger, 1997). In the same vein, Margai (1997) has demonstrated that the low recovery rates observed in public housing units can be explained by structural constraints in the buildings, lack of resources, and relatively poor access to the drop-off sites. Adopting a different approach, based on the role friends and neighbors play as facitators, Burn (1991) has shown that the block leader approach is an effective way to stimulate recycling behavior. More recently, Guagnano, Stern, and Dietz (1995) tested a model in which attitudinal factors and external conditions act in combination to inuence behavior. In their conceptualization, external conditions included a broad range of factors, physical, nancial, legal, and social, all potentially facilitating or curtailing behavior. Among the strongest evidence of the effects of external factors, they found that possession of a bin had a signicant effect on recycling behavior. It is not clear from these different studies how contextual variables inuence behavior. It may be possible that any of these contextual factors can inuence proenvironmental behavior either directly or indirectly. In the rst case, contextual factors might have a direct effect on participation in recycling programs through, for instance, a mechanism of psychological empowerment (Zimmerman and Rappaport, 1988). The second possibility is that environmental attitudes could be shaped by social or national contexts. From that standpoint, contextual factors could either reinforce or reduce the relationship between attitudes and behavior. The present study does not aim to clarify the sequence of the relationships between the two levels of predictors of conservation behavior. Our main objective is to determine whether a set of contextual variables measured at the national level are related to recycling behavior when modeled within a multilevel analytical framework. Four important conclusions can be drawn from this brief review of the literature on the determinants of proenvironmental behaviors. (1) A majority of studies have examined the inuence of the individual variables on different environment-related behaviors, whereas the impact of the contextual factors have generally been neglected. (2) The research on situational and organizational factors reducing barriers to conservation programs have yielded scattered evidence that contextual variables can enhance participation in conservation programs or at least reduce certain barriers limiting the access to participation. (3) Few studies that have explored the role of contextual factors have proposed an analytical approach that integrates both the contextual and the individual bases of environmental action. 6 (4) Most of the research limited its scope to a single national setting; a comparative approach has been a rarity among studies on the determinants of conservation behaviors.
The consequences of this situation for research development have been stressed recently by Dietz, Stern, and Guagnano (1998, p. 451): Theory development has been hampered because studies on national or other broad samples that have substantial demographic variation lack adequate measures of a social psychological model, whereas data sets developed to test social psychological theories usually have limited demographic variation.
6

RECYCLING BEHAVIOR

199

The multilevel modeling approach we apply in this study to identify the micro and macro determinants of recycling behavior in the 15 countries of the European Union addresses these limitations. THEORETICAL MODEL The theoretical model we have adopted in this study is Brands (1997) Context Model for Analysing Environmental Consciousness and Behavior (hereafter the Context Model). The main assumption of Brands model is that the context specically relevant for environmental consciousness and behavior in everyday life can be developed only by using different levels of relevance. The author distinguishes ve levels of relevance integrating the contextual and individual factors at the origins of environmental consciousness and behavior: (1) structural and cultural setting, (2) public environmental discourse, (3) milieu-specic life-words; (4) environmental mentalities, and (5) situational contexts of everyday life. As indicated by Brand (1997, p. 213), the Context Model provides a general framework for analysis. Yet, no specic theoretical interpretation of the interrelations between the ve analytical levels 7 are assumed by the model. This peculiarity renders the Context Model useful as a heuristic device even though data at all the levels are not available in this empirical case. At the person level, we considered the attitudes toward the environment that take the form of global environmental concerns. Many studies have shown that environmental concern is positively correlated with proenvironmental behaviors (Hines, Hungerford, and Tomera, 1986). Another variable we take into account at the individual level is to what extent people perceive the different levels of government as effective in the battle against pollution and in the management of environment problems. Indeed, Blamey has asserted that individual recycling norms are more likely to be activated when it is perceived that the government is doing its part (Blamey, 1998, p. 677). Even though the role of sociodemographics in predicting conservation behaviors has been found inconsistent in prior studies, we include a set of these variables at the person level of our model. These characteristics reect certain cleavages within each given society under investigation and from that standpoint are worth being included. At the country level, different sets of social and institutional factors have been taken into account. Among the institutional determinants of recycling behavior, we included each countrys environmental performance in the realm of waste management. Legislative instruments are being used in many European countries to facilitate reuse/recycling/recovery schemes for waste, such as used tires, bottles, and motor oil. Read (1999, p. 262) has called these instruments a new
7 These levels correspond to the levels of relevance of Brands Context Model. For the purpose of the multilevel analysis, we consider only two levels in our model: person level and country level.

200

GUERIN, CRETE, AND MERCIER

generation of waste management strategies which emphasize waste minimisation, waste re-use and waste recycling as the primary objectives. In addition, economic instruments such as eco-taxes or levies aim at the same ecological goal. Thus, there is a wide variety of ways to reduce and recycle domestic waste; in other words, to implement waste management. It is relevant to measure to what extent these differences in policies impact on public participation in these programs. This national policy indicator will be introduced into our modeling of the macro determinants of participation in recycling. Our model may help to evaluate whether these policy instruments, measured in each EU country, are related to levels of participation in recycling programs. In the perspective developed by Brand (1997, p. 213), environmental consciousness is not something that can be measured at a certain point in time. It comes out of permanent symbolic struggles between collective actors attempting to promote their own denition of problems. This is why our model will take into account the actions of certain groups which participate in this collective debate. We can argue that individual decisions to invest in a conservation activity go well beyond the mere consideration of monetary advantages associated with such behavior. People invest because they have heard from people they trust that it is worthy to invest in such behavior. We know, for example, that citizens tend to trust environmental protection groups, as the Eurobarometer data on information sources, considered the most credible, show quite convincingly. 8 Thus we included among our country-level predictors of recycling behavior-specic indicators of participation in local recycling programs and membership in environmental organizations. Our design also included, at the country level, a variable that can be referred to as awareness of concrete environmental problems, especially deforestation. We can see from Table 1 that there exist considerable differences among countries success in paper and cardboard recycling programs. If we consider the rst column of Table 1 (recycling rates for paper in 1995), we can see that there are important disparities between countries in the recycling rates. How can we explain these differences? We assert that we should go beyond person-level variables to predict the frequency of participation (self-report) in recycling activities. A multilevel modeling approach allows us to evaluate to what extent a series of contextual factors can contribute to a better account of the key antecedents inuencing recycling behavior.

8 For instance, 36.0% of the respondents surveyed in the Eurobarometer 43.1 BIS have picked out environmental organizations as their most condent source of information, followed by scientists (19.6%) and consumer associations (14.8%). These data tend to conrm what had been observed in the 1992 Eurobarometer Study (Europeans and the Environment in 1992). Worcester (1993) has noted that environmental protection organizations received 11 times the level of trust that public authorities did and 5 times that of the media.

RECYCLING BEHAVIOR TABLE 1 Recycling Rate for Paper and Self-Reported Participation in Recycling Recycling rate for paper a 1995 (%) 38 12 77 67 29 53 44 12 35 19 52 37 57 54 65 43.4 Dependent variable: Participated in recycling? (% yes) 58.9 69.9 85.7 88.5 55.1 77.4 79.6 39.1 55.5 22.5 57.7 33.9 83.5 83.2 67.6 63.8

201

Country France Belgium Netherlands Germany Italy Luxembourg Denmark Ireland UK Greece Spain Portugal Finland Sweden Austria Mean

Source: Eurobarometer 43.1 BIS. a The recycling rates for paper have been obtained from the OECD Statistics Bureau. It is worth noting that, at the country level, the dependent variable of the present study is strongly correlated with the OECD countries recycling rates for paper (r .80).

RESEARCH QUESTIONS Our rst task is to evaluate to what extent a series of individual characteristics inuence participation in recycling activities in European Union countries. This task will be performed through the specication and the estimation of the person-level model. Next, we evaluate the impact of a series of country-level variables on individual participation in recycling. Formally, the country-level specication enables us to assess whether a set of contextual factors can account for the variation, between the 15 countries of the European Union, in levels of participation in recycling activities. From the country-level standpoint, the rst question we want to elucidate is to what extent does environmental activism, whether it takes the form of membership in proenvironmental organizations or engagement in recycling programs, inuence the overall level of participation in recycling programs after controlling for the composition of each countrys population in terms of the personal characteristics of the respondents. In other words, our study aims at determining whether true contextual effects intervene to facilitate recycling. This postulated contextual effect is based on the idea that participation in these types of activities,

202

GUERIN, CRETE, AND MERCIER

even if it directly concerns only a small segment of the total population of a given country, contributes to increasing the overall level of public concern about the environment. In turn, this inuence encourages people to participate in activities that require a high degree of cooperation and trust, as is the case for the separation and recycling of waste. 9 Indeed, it is a well-known fact that a variable can take on different meanings and therefore may have distinguishable effects at the micro and macro levels. Consider, for instance, the case of participation in environmental movements or organizations. We know from prior research that individual engagement in these forms of environmental activism generally have a powerful impact on the propensity of these individuals to engage in conservationist behaviors. But when it comes time to consider the impact of mobilization as a macro phenomenon (as measured, for instance, by the proportion of the population who are members of environmental organizations or the proportion of people engaged in local conservation campaigns), a traditional linear model of analysis is not suited to capture the societal effect of collective action. It is true that we can use dummy variables for country-level variables in a more standard multivariate analysis. However, this statistical approach becomes rapidly unusable, as it increases the number of macro entities and interaction terms needed in such a specication. Another important question of substance we want to investigate is whether a governments good performance in the realm of environmental policy in general is related to public participation in recycling programs. Therefore, we evaluate the impact of the governments role in waste management. The environmental performance was measured in two distinct ways: (1) at the person level, measuring respondents perceptions of the environmental performance of the public bodies; and (2) at the country level, using an index indicating the national state of the policies implemented in the sector of waste management. In summary, country-level variables include: %OF MEMBERSHIP, a variable representing the percentage of people mobilized in environmental groups in each country; %IN COLLECTIVE ACTION, which is operationalized as the percentage of people having participated in a local collective activity such as cleaning a park; WASTE POLICIES as just described; DEFORESTATION, a variable measuring the progression of the deforestation problem between 1990 and 1995 in each country of the EU. We examine, by means of a multilevel analysis, whether those country-level variables help to explain a signicant portion of the between-country variance in the participation in programs of recycling. But, before we proceed with this task, we describe the logic and the notation of multilevel modeling.

9 Yet, the estimation of these two distinct effects poses thorny statistical problems. It is worth noting that multilevel modeling enables us to resolve what we could call an aggregation bias problem.

RECYCLING BEHAVIOR

203

METHODS: HIERARCHICAL LINEAR MODELS It is conceptually useful to view the multilevel regression model as a hierarchical system of regression equations (Hox, 1995, p. 10). For example, assume that we have collected data on N persons (i 1, . . . , N j ) in J countries ( j 1, . . . , J). Each person can be given subscripts ij. At the person level, we have the dependent variable measuring individual participation in recycling programs (Y ij ) and a number of explanatory variables (X ij ). At the country level we have a series of variables (Z j ) such as the countrys performance with regard to environmental policies or the percentage of membership in environmental organizations. For the moment, we disregard the Z variables, which are measured at the country level. A separate regression equation can be set up in each country topredict the dependent variable, Y ij , by the explanatory variable, X 1ij (we assume that there is only one micro-level explanatory variable; for instance, education): Y ij
0j 1j X 1ij

r ij ,

(1)

where 0j is the usual intercept, 1j the regression coefcient associated with the predictor X ij , and r ij is the usual residual error term. Although this approach is similar to a multiple-regression model, there is an important difference. The subscript j [which is for the countries ( j 1, . . . , J)] introduces a distinctive characteristic: the intercept 0j and eventually the slope coefcients j s are permitted to vary across the macro units (here the EU countries). The aim of multilevel modeling is to predict the variation of the coefcients by introducing explanatory variables at the country level (Z j ). We now convert this peculiarity of the multilevel model into an equation. As previously noted, Eq. (1) is referred to as the person-level model. The intercepts 0j s are conceived as parameters that vary across countries as a function of a grand mean ( 00) 10 and a random term u 0j . In this instance, 1j s are assumed not to vary across countries and are represented as a function of xed parameters ( 10). Formally:
0j 00 01 Z j

u 0j

(1.1)

and
1j 10 .

(1.2)

This model corresponds to what Bryk and Raudenbush call a random-intercept model: The key feature of such models is that only the intercept parameter in the Level-1 model, 0j , is assumed to vary at Level-2 (1992, p. 86). The 01 coefcient captures the inuence of country-level variables (Z j ) on the 0j s, whereas 10 predicts the nonvarying parameter, 1j .
10 Note that the gammas have no subscript j because they are not assumed to vary across countries. Consequently, they are referred to as xed coefcients.

204

GUERIN, CRETE, AND MERCIER

We followed the recommendations formulated by Bryk and Raudenbush (1992) and Hox (1995) to set up our multilevel model. The rst step was to estimate the person-level model. Then, we set up our model to predict the variation of the coefcients 0j by introducing explanatory variables at the country level. Analyses The rst step of our analysis consists in introducing all the person-level explanatory variables that could potentially account for recycling behavior. These variables were selected in light of prior research on the determinants of engagement in recycling. At the person level, RECYCLE
0j 3j 5j 6j 8j 1j

GLOBAL

2j

IDEOLOGY
4j

PUBLIC BODIES EDUCATION AGE rij .

LOCAL ACTIVISM (2)

MEMBERSHIP IN ASSOCIATION
7j

INCOME

It is worth recalling that the dependent variable, RECYCLE, takes the form of a dichotomy: It is scored 1 if the person has already recycled and 0 if not. Of interest is the expected proportion of people who have participated in sorting and separating household waste for recycling. Given the binary outcome, it is suitable to use a binomial model with the trial number equal to 1, which is also known as the Bernoulli distribution. Using the person-level Bernoulli model, the equation is as follows:
ij

log P/ 1
3j 5j 6j

0j

1j

GLOBAL
4j

2j

IDEOLOGY (3)

PUBLIC BODIES EDUCATION

LOCAL ACTIVISM AGE ,

MEMBERSHIP IN ASSOCIATION
7j

INCOME

8j

where ij is the log of the odds of success (i.e., having already recycled); GLOBAL score on the scale of global environmental concern; IDEOLOGY self-placement on LeftRight scale; PUBLIC BODIES perceptions of the effectiveness of the public bodies regarding their action to protect the environment; LOCAL ACTIVISM Taken part in a local collective initiative, for example, cleaning a beach or a park? (YES 1; NO 0); MEMBERSHIP member of an environmental association? (YES 1; NO 0); EDUCATION age of the respondent when nished education; INCOME household income quartile; and AGE age of the respondent. Note that the person-level error term

RECYCLING BEHAVIOR

205

(r ij ) is not included in Eq. (3) since our Bernoulli model is considered as a v-known case. 11 At the country level, the regression coefcients ( j s) are assumed not to vary across countries, 12 with the exception of 0j , as follows:
0j
1j 2j 3j 4j 5j 6j 7j 8j

00
10 20 30 40 50 60 70 80

01

Z 1j

02

Z 2j

03

Z 3j

04

Z 4j

u 0j

(4)

Among the country-level equations, only the intercepts (which correspond in this case to each countrys mean) have a random component denoted by the error terms that capture a unique country effect. In all other equations, each coefcient of j is obtained by calculating the corresponding . 13 Looking at 0j equation, we can see that four country-level variables (as shown in Table 4) has been introduced in order to predict the variation in the 0j s. Table 2 presents the results of the person-level model and shows that many effects of person-level variables are signicant. The most important effect on the participation in recycling programs comes from the LOCAL ACTIVISM variable. Recall that we must transform the values of the coefcients (presently expressed in log-odds) back to their original scale in order to interpret them in
11 A case in which the lowest level variance is known, its value being determined completely by the population proportion (Bryk and Raudenbush, 1992, p. 157). 12 Before proceeding to the specication of the country-level model, we had to select either a random intercept model or a random slope model. With respect to the former kind of model, only the intercept parameter varies across countries. In that case, the effects of micro-level predictors are treated as xed coefcients. In the random slope model, all or some of the micro-level coefcients are treated as random. According to Bryk and Raudenbush (1992, p. 116), theoretical considerations are primary in whether a level-1 coefcient should be conceived as random. In the case of participation in recycling programs, no theoretical foundations would permit us to assume that the effect of the person-level variables vary across countries, except for GLOBAL CONCERNS, EDUCATION, and INCOME. Indeed, one may argue that the inuence of environmental concerns and sociodemographic variables on recycling can vary according to certain contextual factors. In order to test such a possibility, we computed a likelihood ratio test comparing the deviance statistics of a model where each of the three effects was considered xed with alternative models in which the coefcients were treated as random. The results of these tests conrmed that a model that constrains the residual variance for GLOBAL CONCERNS, EDUCATION, and INCOME to zero is appropriate. 13 Substitution of the level-2 equations for qj into their corresponding level-1 terms yields a single-equation linear model with a complex error structure. The estimation of such a model is accomplished through generalized least squares.

206

GUERIN, CRETE, AND MERCIER TABLE 2 Multivariate Analysis of the Person-Level Determinants of Recycling Behavior (Population-Average Model Estimates) Micro-level model Fixed effects Intercept ( 00) Individual global concerns ( 10) Ideology ( 20) Public Bodies ( 30) Local Activism ( 40) Member of an association ( 50) Education ( 60) Income ( 70) Age ( 80) N 16,3000 Variance component Random effects 00 (u 0j ) Coefcients ( ) SE

0.53* 0.06*** 0.00 0.04*** 0.66*** 0.34*** 0.07*** 0.03*** 0.09***

(0.24) (0.01) (0.01) (0.01) (0.05) (0.06) (0.02) (0.01) (0.01)

df

p value

0.85

14

2322.53

.000

Note. All generalized least-squares estimates in this table are computed with a population-average model. Standard errors are within parentheses. Random effects ( 00 ) are estimated through a restricted maximum likelihood procedure. *** p .01. ** p .05. *p .10.

terms of proportions. Predicted log-odds can be converted to a predicted proportion by using the inverse transformation for the logistic link function given by the following equation: exp x / 1 exp x .

The intercept (0.53) is the expected log-odds of having recycled for a person scoring at the mean for all group mean centered variables 14 and scoring zero on MEMBERSHIP and LOCAL ACTIVISM measured at the person-level. In this case, this expected log-odds corresponds to an expected proportion of 0.63. This means that the expected proportion for persons with the mean value on all group mean centered predictors (and with a value of 0 for both MEMBERSHIP AND LOCAL ACTIVISM) to have recycled is about 0.63. Recall that the mean proportion of the dependent variable is 0.65 (overall
14 Because we centered all the continuous variables around their group mean, the interpretation of the intercept is the value of the response variable when these explanatory variables are at the mean rather than at zero. MEMBERSHIP and LOCAL ACTIVISM, which were not centered around their mean, are dichotomous variables that take the values 0 and 1.

RECYCLING BEHAVIOR

207

proportion of people having done recycling in our sample), which is quite close to the intercept. Such a slight difference is expected since we used a nonlinear link function. The large positive coefcient for the variable measuring the effect of individual local activism (0.66) means that the expected proportion of having recycled is much greater for persons who have participated in local ecological initiatives to protect the environment. We can calculate how much higher by computing as follows: 0.53 (INTERCEPT) 0.66 (LOCAL ACTIVISM) 1.19. Converting this number to a proportion gives 0.77, which means that, for persons who have been engaged in this form of activism (scoring at the mean value on all the group centered predictors and 0 on MEMBERSHIP), the expected proportion of participation in recycling activities is 0.77. Thus, the expected proportion of participation in recycling is multiplied by 1.22 (0.77/0.63) for a group of persons who have taken part in a collective local activity such as cleaning a park. The global environmental concerns (GLOBAL) also have a signicant inuence on recycling behavior, though of a weaker magnitude (0.06). Results from the person-level model also show that there is a signicant effect on participation in recycling coming from the PUBLIC BODIES variable (0.04), which measures individuals trust toward the effectiveness of the environmental policies of different levels of government. As for the socioeconomic variables, two of them have a small effect on recycling behavior in the expected direction: education and income. As for age, the positive coefcient means that those who are younger are less likely to have recycled. 15 However, their effect appears to be quite small, a result which is consistent with the demographic shift hypothesis (Howell and Laska, 1992). Traditional ideological orientation did not seem to have an impact on participation in household waste recycling. Let us now look at the lower panel, which represents the variance components of the model. The chi-square statistics accompanying the variance at the countrylevel (0.85) shows that there is signicant variation in country means. We now enter the country-level phase of the modeling process, where the aim is to account for this variation by introducing a series of contextual variables of the determinants of recycling into our model. THE CONTEXTUAL DETERMINANTS OF RECYCLING Is participation in recycling activities inuenced by social and environmental contextual variables? In other words, to what extent can a series of contextual
Past research has indicated that age is often negatively associated with environmental concern (see Mohai and Twight, 1987). Our nding concerning the association between age and recycling seems to differ in direction from this pattern. Yet, this difference can be explained by the fact that 17% of the sample (2776) of Eurobarometer 43.1 is in the 15-to-24 age group. Therefore, people in this age group are less likely to be homeowners and to recycle household waste. This hypothesis was tested by excluding all these respondents from our analyses. Within the new sample, the association between age and recycling behavior was found to be negative.
15

208

GUERIN, CRETE, AND MERCIER

variables account for the variation we just noticed between countries in the levels of participation? We try to answer this question by developing a country-level component to our model. Following Byrk and Raudenbushs suggestion (1992, p. 211), we divided the country-level predictors into conceptually distinct subsets. After having tted a submodel for each, the strongest predictors were combined in an overall model. Recall that our model posits that country effects modify only the mean level of the outcome for the country. In this kind of model, only the intercept ( 0j ) varies across countries, which is the reason we introduce the country-level variables into the 0j equation only. Following Bryk and Raudenbushs (1992) suggestion, we regrouped our country variables into three conceptually distinct subsets: collective action, environmental policies, and real-world environmental outcomes. Ecological collective action is operationalized here in the form of two variables: the percentage of people in each country who are members of an association for the protection of the environment and the percentage of the population participating in local cleaning activities in each country. In the environmental policies category, we constructed a variable that draws from the state of waste legislation and policy concerning domestic waste (European Environment Agency, 1998). Another contextual variable deals with concrete ecological problems (loss of forest area). According to Guagnano et al. (1995), Attitude theory needs to be modied to include not only the perception of external conditions but the external conditions themselves. One can argue that environmental problems are an important factor in the formation of environmental consciousness and behavior. Table 3 describes the country-level variables used in the analysis. We start by introducing the three collective action variables in our countrylevel model. Then we include the variables from other categories in the later phases of the modeling process. Table 4 shows the results of these different models, with the lower panel presenting the ndings of the nal model. The person-level model of Eq. (3) remains unchanged. At the country level, the rst model we tested includes the two ecological action variables in an equation that aims to predict the variation in the intercepts, 0j s (corresponding to the mean of the participation variable), as follows:
0j 00 01

%OF MEMBERSHIP
02

%IN COLLECTIVE ACTION

u0j .

In the different panels of Table 4, the results of these models are presented, starting with the Collective Action Model. Only the country-level variables having a signicant effect were incorporated into the subsequent models. Each model introduces a new group of macro-level variables. At the bottom of each panel, a 00 indicates the proportion of variance remaining in the intercepts (average participation in each country) after taking into account these variables,

RECYCLING BEHAVIOR TABLE 3 Country-Level Variables a % of membership in ecological groups 4.2 6.6 21.7 6.1 6.1 24.8 17.5 6.7 7.3 2.4 3.9 3.4 9.4 15.2 11.4 % of participation in a collective initiative 8.7 7.5 7.9 16.4 8.4 22.0 9.7 15.5 12.1 9.4 7.0 6.0 42.9 32.9 15.4

209

Countries France Belgium Netherlands Germany Italy Luxembourg Denmark Ireland UK Greece Spain Portugal Finland Sweden Austria

Waste policy (11-point scale) 10.0 10.0 10.0 9.5 9.0 7.0 10.0 8.0 9.0 9.0 6.0 6.0 10.0 9.0 10.0

Average annual rate of deforestation b (19901995) 1.1 0.5 c 0.0 0.0 0.1 0.2 c 0.0 2.7 0.5 2.3 0.0 0.9 0.1 0.4 c 0.0

Source: Eurobarometer 43.1 BIS. a See Appendix 1 for descriptions of the variables. b A positive number indicates a loss of forest area, a negative number a gain. c The average annual rate of deforestation for this country is an estimate based on parameters obtained by regressing deforestation on the percentage of species (plants) threatened in each country. (The r between deforestation and the percentage of species threatened is .60 in the 12 countries where both variables were available).

as measured at the country level. In principle, if these variables explain a signicant portion of this variance, the 00 should be inferior to the one estimated in the basic model (person-level predictors with only the intercept varying at the country level), which was 0.85. The next line below it indicates the proportion of overall variance explained by the country-level model, whereas the last line shows the incremental variance, i.e., the increase in variance accounted for in comparison to the previous model. The results from the Collective Action Model indicate that the %OF MEMBERSHIP variable has a signicant effect on the average participation in the countries recycling activities. The coefcient associated with this variable (0.07) have a positive inuence on the intercept 0j s. This implies that the average recycling participation in countries having more membership in environmental associations is higher after controlling for membership measured at the person level and for the composition of each country sample in terms of the other individual characteristics of the respondents. As for the % IN COLLECTIVE ACTION variable, its coefcient (0.02) is nonsignicant. Let us consider the interpretation of the %OF MEMBERSHIP variable. Its

210

GUERIN, CRETE, AND MERCIER TABLE 4 Effects of Country-Level Variables on Recycling Behavior Coefcients ( ) SE

Collective Action model Fixed effects %OF MEMBERSHIP ( 01) % IN COLLECTIVE ACTION ( 02) Random effects 00 (u 0j ): 0.46 Proportion of variance accounted for: .45 Legislative Facilitators plus Collective Action Fixed effects WASTE POLICY ( 03) %OF MEMBERSHIP ( 01) Random effects 00 (u 0j ): 0.41 Proportion of variance accounted for: .51 Incremental variance explained by this model: .06 Collective Action plus Concrete Environmental Problems Fixed effects DEFORESTATION ( 04) WASTE POLICY ( 03) %OF MEMBERSHIP ( 01) Random effects 00 (u 0j ): 0.20 Proportion of variance accounted for (at the country level): .76 Incremental variance explained by this model: .25 *** p ** p *p .01. .05. .10.

0.07** 0.02

(0.03) (0.02)

0.20* 0.07***

(0.12) (0.03)

0.53*** 0.16* 0.06**

(0.15) (0.08) (0.02)

coefcient is 0.07. This effect can be interpreted as follows. On average, a 1% increase in membership at the country level yields approximately a 0.5-point increase in countries proportion of people participating in recycling activities. This result can be interpreted as a powerful contextual effect of membership on participation in recycling programs. This rst country-level model helps us to considerably reduce the proportion of the variance between the countries intercepts, since the now stands at 0.46. If we compare this variance with the one registered in the baseline model where the was 0.85, the variance explained by the Collective Action Model is 0.45. The second panel presents the results of a model that adds to the signicant variable of the Collective Action Model, a variable measuring the environmental performance of each country in terms of legislative initiatives in the realm of waste management and recycling. One may argue that favorable norms and

RECYCLING BEHAVIOR

211

attitudes for the protection of the environment are more easily activated when the citizens have the feeling that governments and public authorities have done their share in managing the environment. We have already found that individuals perceptions on this aspect of the governmental performance have an impact on the citizens willingness to participate in recycling activities. We now examine whether a similar effect can be noticed when we examine policies implemented by national governments. WASTE POLICY is an index which varies between 0 and 11 and its effect was measured in the Legislative Facilitators plus Collective Action model. We can see from panel 2 that the coefcient measuring the inuence of the state of waste legislation and policy (0.20) is in the expected direction and signicant at the statistical threshold of 0.10. This result seems to indicate that countries that have promulgated environmental laws and policies to discourage the use of landlling and facilitate reuse/recycling/recovery schemes for waste tend to exhibit slightly higher levels of public participation in recycling activities. Unfortunately, the addition of this variable accounts for a rather small part of the between-country variance, that is, 0.06. We can suggest an hypothesis to explain this weak effect. Environmental laws and policies enacted by these countries do not necessarily translate in every case into concrete measures to protect the environment. According to Read (1999, p. 264), a common problem has emerged in countries that have embarked on policies promoting greater sustainability in waste management through recycling and reduction. The pace of policy making has not been matched by an equal effort to provide mechanisms for effective policy implementation. It is possible that this gap could explain the weakness of this relationship between waste management policies and public participation in recycling programs. This is why we contend that it is important to examine an indicator of the real condition of the environment rather than limit the investigation to the content of the policies. Several indicators can be used to assess programs in the eld of environmental protection. These indicators should be familiar to the public and related to dimensions that have received sizeable attention from the media. Moreover, a suitable indicator should be substantively linked to the issue of recycling. Such is the case, we think, for the deforestation problem. Therefore, we built a new model (Collective Action plus Concrete Environmental Problems) in which we included a deforestation variable along with the two other signicant predictors in the macro-level model. Results from this interactive model are presented in the lower panel. These results show that the hypothesis regarding the inuence of concrete environmental conditions is conrmed by our model. Indeed, the resulting coefcient associated with this new variable is signicant and of a large magnitude (0.53). The positive sign means that a lower rate of increase in forest area (re-forestation) between 1990 and 1995 is associated with higher levels of recycling behavior as measured in 1995. According to Brands framework (1997, p. 213), a suggestive interpretation of this nding can be formulated as follows:

212

GUERIN, CRETE, AND MERCIER

The disappearance of forests has inuenced participation through increasing awareness of global environmental concerns. 16 In turn, more acute concerns among respondents about the destruction of their own countrys forests may be conducive to adequate environmental behavior that, in the present case, takes the form of recycling. However, the conclusion in this regard cannot be considered as denitive but only suggestive given the limitations of the evidence presented. In the lower part of this model, we have a associated with this model and the proportion of variance that is explained by the interactive model. If we compare this latter (0.20) with the one obtained in the baseline model (0.85), we now have an explained variance of 0.76 at the country level. This means that 76% of the variance observed at the outset between the average participation in recycling activities of the EU countries was explained by three country-level variables included in our nal model, i.e., %OF MEMBERSHIP, WASTE POLICY, and DEFORESTATION. We now initiate a discussion of the main ndings obtained at both levels of analysis. RESULTS Findings at the Person Level From the individual standpoint, our results indicate that people who participate in local programs to protect the environment (LOCAL ACTIVISM) tend also to engage in recycling behaviors. We also found that global environmental concern has a positive impact on the individual propensity to recycle. Results from the person-level model also show that people who believe that their government is making a reasonable effort to protect the environment are more inclined to adopt environmentally friendly behaviors such as recycling. As for the different markers of socioeconomic status, two of them, education and income, make modest impact on recycling behavior, as expected. As early as 1984, Stern and Aronson were calling for a wider perspective, one that included social and cultural contexts, in generating individual conservation decisions within the eld of environmental policy analysis. The aim of this study was to empirically test a theoretical model that includes such contexts. A summary of our ndings concerning the interrelationships between context and recycling behavior is presented below. Findings at the Country Level With regard to the country-level part of the analysis, one of our main conclusions is that conservation behavior is greatly inuenced by the context
It is interesting to note that among the items used to create the global concerns variable (whose inuence was found signicant), three of them referred to the disappearance of forests and other natural habitats.
16

RECYCLING BEHAVIOR

213

of ecological mobilization in which it occurs. Indeed, the more people participate in recycling programs by being a member of an ecological association, the more the countrys population at large is likely to engage in behaviors such as sorting and separating domestic wastes. It is worth recalling that such a conclusion can be stated after controlling for the composition of each country sample in terms of the individual characteristics of the respondents, including individual membership. In fact, this one dimension of environmental mobilization measured at the country level explains as much as 45% of the variance observed between the European Union countries in their recycling participation levels. Another interesting result, also regarding contextual determinants of participation, is the complex relationship we observed between the state of the environment, environmental consciousness, and recycling behavior. Indeed, we were able to determine that a countrys mean level of participation is related to a concrete environmental problem, deforestation, and perhaps also to the degree of consciousness inspired by the ecological damage caused by this problem. Our study, however, is not exempt of caveats. First, we have only considered one type of proenvironmental activity, i.e., recycling behavior, as operationalized by a dichtomous self-report measure. We know that the antecedents of each conservation behavior are different. Accordingly, the extent to which contextual factors inuence environmental behavior could also vary according to the nature of the behavior. This is why it would be interesting to replicate this study on a series of proenvironmental behaviors. A second limitation is the fact that we did not address how the contextual variables inuence recycling; that is, the precise mechanisms by which social factors and individual behavior are linked. We think that a next step could be the specication of a multilevel structural equation model (Heck and Thomas, 2000, p. 154) which aims to assess different hypotheses concerning the structural relationships between the multilevel determinants of recycling. Finally, this research was done in European Union countries. It would be interesting to extend this kind of comparative analysis to other regions of the world, notably to North America, where we could generalize our conclusions by developing a similar analysis that would take into account characteristics of different subnational entities such as American states or Canadian provinces. This kind of research would be a signicant step forward in the development of theory about the specic contribution of the contextual and individual bases of environmental consciousness and behavior.

214

GUERIN, CRETE, AND MERCIER

APPENDIX 1 Dependent Variable Participation in Recycling Participation in recycling was measured by a binary outcome drawn from v190 of the questionnaire. A respondent was given a score of 1 for a yes to this variable and a score of 0 for a no.
v190 Already done for environment protection: Sort out certain types of household waste (glass, paper, motor oil, batteries, . . . ) for recycling. 0. Not mentioned 1. Mentioned

Independent Variables Person-Level Variables Global concerns. Individual global concerns was measured as an index created by summing the responses to variables 135 to 140.
Are you very worried, somewhat worried, not very worried or not at all worried about the following problems? v135 The disappearance of certain types of plants, animals and habitats throughout the world. 3. 2. 1. 0. Very worried Somewhat worried Not very worried Not at all worried

v136 Worry about environment problems: Using up natural resources throughout the world. 3. 2. 1. 0. v137 3. 2. 1. 0. v138 3. 2. 1. 0. Very worried Somewhat worried Not very worried Not at all worried Worry about environment problems: The disappearance of tropical forests. Very worried Somewhat worried Not very worried Not at all worried Worry about environment problems: Global warming (the greenhouse effect). Very worried Somewhat worried Not very worried Not at all worried

RECYCLING BEHAVIOR v139 3. 2. 1. 0. Worry about environment problems: The destruction of the ozone layer. Very worried Somewhat worried Not very worried Not at all worried

215

v140 Worry about environment problems: The risk that pollution from industrialized countries spreads to less industrialized countries. 3. 2. 1. 0. Very worried Somewhat worried Not very worried Not at all worried

Ideology. Ideology was drawn from the LeftRight scale (variable 339).
V339 In political matters people talk of the Left and the Right. How would you place your views on this scale?

Local activism. Local activism was measured by variable 191: Already done something for environmental protection?
Taken part in a local environmental initiative, for example, cleaning a beach or a park. (YES 1; NO 0).

Public bodies. The Public bodies index was constructed by summing variables 229 to 233. This index ranges from 0 to 5.
Public bodies at different levels can act to protect the environment. In your opinion, do public bodies act efciently or not . . . ? v229 At local level.

1. Efciently 0. Not efciently v230 Public environment protection: At regional level.

1. Efciently 0. Not efciently v231 Public environment protection: At national level.

1. Efciently 0. Not efciently v232 Public environment protection: At European Community level.

1. Efciently 0. Not efciently v233 Public environment protection: At worldwide level.

1. Efciently 0. Not efciently

216

GUERIN, CRETE, AND MERCIER

Education. The Education index corresponds to v343.

v343 1. 2. 3. 4. How old were you when you stopped full-time education? Up to 15 years 16 19 years 20 years and older Still studying

Income. The Income variable corresponds to v356 and collapses answers into four groups representing approximate quartiles.
v356 Household incomeapproximate quartiles.

Age. Age was measured by v346.

v346 1. 2. 3. 4. Respondents agefour collapsed categories. 1524 years 2539 years 40 54 years 55 years and older

Country-Level Variables Percentage of Membership (in associations). Membership was measured as the countrys percentage of people who are member of an association for the protection of the environment (v188). Collective Action. This variable was measured by the countrys percentage of people who have taken part in a local environmental initiative (v191). Deforestation. Average annual rate of deforestation (1990 1995) computed by the United Nations Development Programme (1998). Waste Legislation and Policy. The waste legislation index was computed by summing each countrys scores based on their policy initiatives concerning the different aspects of waste management (see European Environment Agency, 1998). APPENDIX 2 Correlation Matrix of the Country-Level Variables
1 1. 2. 3. 4. 5. COLLECTIVE ACTION MEMBERSHIP %HAVING RECYCLED DEFORESTATION WASTE POLICY 1.000 0.504 0.479 0.025 0.337 2 3 4 5

1.000 0.719 0.313 0.324

1.000 0.648 0.601

1.000 0.531

1.000

RECYCLING BEHAVIOR

217

REFERENCES
Balderjahn, I. (1988). Personality variables and environmental attitudes as predictors of ecologically responsible consumption patterns, Journal of Business Research 17, 5156. Berger, I. E. (1997). The demographics of recycling and the structure of environmental behavior, Environment and Behavior 29(4), 515531. Black, J. S., Stern, P. C., and Elsworth, J. T. (1985). Personal and contextual inuences on household energy adaptations, Journal of Applied Psychology 70, 321. Blamey, R. (1998). The activation of environmental norms: Extending Schwartzs model, Environment and Behavior 30(5), 676 708. Brand, K.-W. (1997). Environmental consciousness and behavior: The greening of lifestyles, in The International Handbook of Environmental Sociology, (M. Redclift and G. Woodgate, Eds.), pp. 204 217, Edward Elgar, Cheltenham, UK. Bryk, A. S., Raudenbush, S. W., Seltzer, M., and Congdon, R. (1988). An Introduction to HLM: Computer Program and Users Guide, 2nd ed., Univ. of Chicago Press, Chicago. Bryk, A. S., and Raudenbush, S. W. (1992). Hierarchical Linear Models, Sage, Beverly Hills, CA. Burn, S. (1991). Social psychology and the stimulation of recycling behaviors: The block leader approach, Journal of Applied Social Psychology 21, 611 619. Buttel, F. (1987). New directions in environmental sociology, Annual Review of Sociology 13, 465 488. Corral-Verdugo, V. (1996). A structural model of re-use and recycling in Mexico, Environment and Behavior 28(5), 665 696. Dempster, A. P., Laird, N. M., and Rubin, D. B. (1977). Maximum likelihood from incomplete data via the EM algorithm, Journal of the Royal Statistical Society, Series B, 39, 1 8. Derksen, L., and Gartrell, J. (1993). The social context of recycling, American Sociological Review 58, 434 442. Dietz, T., Stern, P. C., and Guagnano, G. A. (1998). Social structural and social psychological bases on environmental concern, Environment and Behavior 30(4), 450 471. Dunlap, R., and Van Liere, K. (1978). The new environmental paradigm: A proposed measuring instrument and preliminary result, Journal of Environmental Education 9, 10 19. Dunlap, R. E. (1989). Public opinion and environmental policy in Environmental Politics and Policy (J. P. Lester, Ed.), pp. 87134, Duke Univ. Press, Durham, NC. Ebreo, A., Hershey, J., and Vinning, J. (1999). Reducing solid waste: Linking recycling to environmentally responsible consumerism, Environment and Behavior 31(1), 107124. European Environment Agency. (1998). Europes Environment: The Second Assessment, Ofce for Ofcial publications of the European Union. Elsevier Science Ltd., Luxembourg/Oxford. Gamba, R. J., and Oskamp, S. (1994). Factors inuencing community residents participation in commingled curbside recycling programs, Environment and Behavior 26(5), 587 612. Geller, E. S., Brasted, W., and Mann, M. (1980). Waste receptacle designs as interventions for litter control, Journal of Environmental Systems 9, 145160. Geller, E. S., Winett, R. A., and Everett, P. B. (1982). Preserving the Environment: New Strategies for Behavior Change, Wiley, New York. Guagnano, G. A., Stern, P. C., and Dietz, T. (1995). Inuences on attitude behavior relationships: A natural experiment with curbside recycling, Environment and Behavior 27(5), 699 718. Heck, R. H., and Thomas, S. L. (2000). An Introduction to Multilevel Modeling Techniques, Erlbaum, Mahwah, NJ. Hines, J. M., Hungerford, H. R., and Tomera, A. N. (1986). Analysis and synthesis of research on responsible environmental behavior: A meta-analysis, The Journal of Environmental Education 18, 1 8. Hopper, J. R., and Nielson, J. M. (1991). Recycling as an altruistic behavior: Normative and behavioral strategies to expand participation in a community recycling program, Environment and Behavior 23, 195220.

218

GUERIN, CRETE, AND MERCIER

Howell, S. E., and Laska, L. B. (1992). The changing face of the environmental coalition: A research note, Environment and Behavior 24, 134 144. Hox, J. J. (1995). Applied Multilevel Analysis, TT-Publikaties, Amsterdam. Margai, F. L. (1997). Analyzing changes in waste reduction behavior in a low-income urban community following a public outreach program, Environment and Behavior 29(6), 769 792. McGuire, R. H. (1984). Recycling: Great expectations and garbage outcomes, American Behavioural Scientist 28, 93114. Mohai, P., and Ttwight, B. W. (1987). Age and environmentalism: An elaboration of the Buttel model using national evidence, Social Science Quarterly 68, 798 815. Oskamp, S., Harrington, M., Edwards, T., Sherwood, D., Okuda, S., and Swanson, D. (1991). Factors inuencing household recycling behavior, Environment and Behavior 23, 494 519. Raudenbush, S. W., and Sampson, R. (1999). Assessing direct and indirect effects in multilevel designs with latent variables, Sociological Methods and Research 28(2), 123153. Read, A. D. (1999). Making waste work: Making UK National Solid Waste Strategy work at the local scale, Resources, Conservation and Recycling 26, 259 285. Samdahl, D. M., and Robertson, R. (1989). Social determinants of environmental concern: A specication and test of the model, Environment and Behavior 21, 57 81. Stern, P. C., and Aronson, E. (Eds.). (1984). Energy Use: The Human Dimension. San Francisco: Freeman. Tarrant, M. A., and Cordell, H. K. (1997). The effect of respondent characteristics on general environmental attitude behavior correspondence, Environment and Behavior 29(5), 618 637. Usitalo, L. (1990). Are environmental attitudes and behavior inconsistent? Findings from a Finnish study, Scandinavian Political Studies 13, 211226. Vining, J., and Ebreo, A. (1990). What makes a recycler? A comparison of recyclers and nonrecyclers, Environment and Behavior 23, 494 519. Worcester, R. M. (1993). Public and elite attitudes to environmental issues, International Journal of Public Opinion Research 5(4), 315334. Zimmerman, M. A., and Rappaport, J. (1988). Citizen participation, perceived control and psychological empowerment, American Journal of Community Psychology 16, 725750.