Environmental Behavior in Cross-National Perspective

Abstract

This article seeks to explain cross-national differences on environmental behavior. After controlling for a series of sociodemographic and psychosocial factors, it was predicted that national levels of wealth, postmaterialism, education development, and environmental problems are positively related to environmental behavior. The national-level variance is to a substantial degree explained by individual-level variables, capturing compositional effects. The remaining variance is explained by the contextual-level variables. All of the country-level variables are predictors in the expected direction, with the exception of environmental degradation, which is negatively related to behavior, and education development, which has no impact on private environmental behavior. More importantly, cross-level interactions show that in more developed countries, there are stronger relationships between proecological attitudes and reported proenvironmental behavior. These findings contribute to the growing cross-cultural research on environmental behavior pointing out the necessity of simultaneously assessing the effects of both individual and contextual-level forces affecting behavior across nations.

Keywords

environmental behavior cross-cultural research multilevel analysis compositional effect cross-level interaction

Introduction

Research on environmental behavior focuses heavily on two core questions: “What are the relative roles of sociodemographic and psychological variables in explaining environmental behavior?” and “Why does environmental behavior vary cross-nationally?” Unfortunately, both of these questions have remained largely unresolved due to the mixed results across many studies. For example, whereas some studies emphasize the importance of the social and psychological factors (Marquart-Pyatt, 2012b; Oreg & Katz-Gerro, 2006), others lament the poor predictive value of sociodemographic variables (Diamantopoulos, Schlegelmilch, Sinkovics, & Bohlen, 2003) and the gap between attitudes and environmental behavior (Gifford, 2011; Kollmuss & Agyeman, 2002). Regarding the second question, there are also mixed results about the relationship between environmental behavior and country-level variables like postmaterialist values, educational levels, and environmental conditions (Dalton, 2005; Duroy, 2008).

This study argues that resolving some of these questions requires a multilevel approach that integrates both individual and country-level factors as predictors of environmental behavior. The core hypotheses of a multilevel theory is that how individual factors influence environmental behavior is conditional on country-level context (i.e., cross-level effects) and, at the same time, how country-level factors influence national levels of environmental behavior is also a function of individual-level characteristics (i.e., compositional effects). For example, our analysis finds that individual-level attitudinal factors have a stronger positive influence on environmental behavior in countries with higher levels of socioeconomic development. Furthermore, our results demonstrate how compositional effects are reducing the relative influence of national-level factors on environmental behavior.

To test these types of hypotheses, this article employs multilevel statistical models where individuals (Level 1) are nested within countries (Level 2). The major advantage of multilevel models is that both the intercepts and slope coefficients are allowed to vary across nations and that variation could be explained simultaneously by individual and national-level factors. Our analysis uses the 2010 Environmental Module from the International Social Survey Program (ISSP) that comprises 30 countries, which includes retrospective self-reports of both private and public environmental behaviors (we will use the term “environmental behavior” for narrative ease).

While other researchers have recognized the value of multilevel analysis (for a review, see Milfont, 2012), with few exceptions (Guerin, Crete, & Mercier, 2001; Hadler & Haller, 2011; Pirani & Secondi, 2011), multilevel studies are focused on other dimensions of environmental concern such as sacrifice or willingness to pay (Gelissen, 2007; Liu & Sibley, 2012; Mostafa, 2013; Nawrotzki, 2012), seriousness of local-global environmental problems (Fairbrother, 2013; Marquart-Pyatt, 2012a), environmental efficacy (Marquart-Pyatt, 2012a), or a combination of two or more of these dimensions (Franzen & Meyer, 2010; Haller & Hadler, 2008; Marquart-Pyatt, 2012a). Other researchers have also examined cross-level interactions (Franzen & Meyer, 2010; Liu & Sibley, 2012; Marquart-Pyatt, 2012a; Nawrotzki, 2012) but again did not focus on environmental behavior.

A Multilevel Conceptual Framework

Figure 1 displays a conceptual framework that represents the central hypotheses about the effects of individual and country-level variables, as well as potential cross-level and compositional effects. This section reviews the existing literature to develop the basis for each hypothesis. Following previous work (Hadler & Haller, 2011; Hunter, Hatch, & Johnson, 2004), we distinguish between private environmental behaviors such as buying certain products or saving water or energy at home, and public environmental behaviors such as signing petitions, giving money, or being a member of an environmental association. Figure 1 encompasses the joint effect of individual-level variables (Hypotheses 1a and 1b) and country-level variables (Hypotheses 2a-d) on two dimensions of environmental behavior. We also test a set of possible cross-level interactions between the level of development of a country and the strength of the attitude–behavior association (Hypothesis 3). Pisano and Hidalgo (2014) provided additional analysis about cross-level interactions between the level of development of a country and the link between sociodemographic variables and environmental behavior.

Figure 1.

Multilevel theoretical framework.

Individual-Level Predictors of Environmental Behavior

The literature on environmental behavior is traditionally divided into two major streams: “studies focused on sociodemographic factors associated with environmentalism and studies on values, beliefs and other social-psychological constructs related to environmentalism” (Dietz, Stern, & Guagnano, 1998, p. 451). The following review is focused on cross-national studies that address theoretical and methodological issues related to our work.

Regarding the social bases of environmentalism, research has found inconsistent cross-cultural effects (Diamantopoulos et al., 2003). For example, Marquart-Pyatt (2012a) tested a model of pathways to environmental activism and found that education was the only variable with a consistent positive effect across 16 countries, while gender, education, income, and place of residence were only sporadically important. Hunter et al. (2004) found similar inconsistent results for gender and public environmental behavior; women engaged in more public environmental behavior relative to men just in 3 of 22 sampled countries, while men were more engaged in 2 countries. Results were more consistent for private environmental behavior; women were more engaged than men in two thirds of the countries analyzed.

The empirical results have been more consistent for attitudes and other psychological variables related to environmental behavior. For example, Oreg and Katz-Gerro (2006) demonstrated that a series of values, attitudes, and intentions were significantly and positively associated with both private and public environmental behavior across 27 countries. They also conducted a country-by-country analysis finding support for the model cross-nationally. Pirani and Secondi (2011) using information concerning the 27 European Union member countries found that most of the attitudes, knowledge, and opinions toward the environment measured were positively affecting a series of eight different private environmental behaviors. Hadler and Haller (2011) found similar paths for both types of environmental behaviors.

Taken as a whole, results from previous research suggest that environmental behaviors are positively related across nations to education level, environmental knowledge, and a series of proenvironmental attitudes (risk perception, efficacy, intention or willingness to make sacrifice, etc.). Furthermore, sociodemographic factors have a smaller effect on behavior compared with psychological factors (Diamantopoulos et al., 2003). One potential reason for this conclusion is that survey questions measuring psychological factors refer to the same environmental content domain as the behaviors, and thus produce a stronger attitude–behavior relationship. In addition, sociodemographic dispositions may be acting indirectly through attitudes (Corral-Verdugo & Zaragoza, 2000; Milfont, Richter, Sibley, Wilson, & Fischer, 2013), which may partially explain the mixed result found between some demographic factors and environmental behavior.

The discussion presented above suggests the following individual-level hypotheses: All psychosocial factors listed in Figure 1 will be positively related to both private and public environmental behavior (Hypothesis 1a), but sociodemographic factors will have a relatively smaller influence than environmental attitudes (Hypothesis 1b).

Country-Level Predictors of Environmental Behavior

Besides the two major categories of individual-level variables, Dietz et al. (1998) suggested that “a fuller theory of environmentalism must attend to contextual effects that influence beliefs and values as well as individual level variables” (p. 466). In recent years, scholars have highlighted country-level variables like economic and educational development, postmaterialist values, and actual environmental conditions as important contextual factors.

Socioeconomic development theories suggest that modernization and national wealth expands the resource base that facilitates environmental behavior (Spaargaren & Mol, 1992). Postindustrialized societies tend to have more dense communication structures, mass education, and urbanization, which may facilitate opportunities to translate public concerns into activism (Dalton, 2005). For example, environmental nongovernmental organizations (NGOs) require financial resources to start and sustain participation in social movements, and it is easier to find a concentration of like-minded individuals in populated urban areas (Gillham, 2008). Also, in better educated societies, it is more likely that citizens will be aware of the environmental problems and possess the civic skills and political resources necessary for engagement and activism (Brady, Verba, & Schlozman, 1995; Duroy, 2008).

Ronald Inglehart (1995) argued that citizens of wealthier nations display more proenvironmental attitudes and behavior because of a general shift from materialistic (personal and national security, economic well-being) to postmaterialist values (quality of life, self-expression). As societies become more affluent, their members are less preoccupied with the economic struggle for survival and are free to pursue postmaterialistic goals, such as political freedom, individual self-fulfillment, and environmental protection. A range of other research has documented this process of value change and the relationship to environmental behavior at the national level (Dalton, 2005; Oreg & Katz-Gerro, 2006).

The direct experience or perception of environmental degradation may be an additional source of environmental concern (Dunlap & Mertig, 1995), and areas with severe environmental problems frequently have more support for taking action to solve the problems (Johnson, Brace, & Arceneaux, 2005). Environmental problems are on average more severe in the Global South and low-income and minority communities from the Global North, which may motivate the emergence of environmental activism (Escobar, 2006; Martinez-Alier, 2003; Pellow & Brulle, 2005). Considering this empirical evidence, Inglehart (1995) proposed “objective problems, subjective values hypotheses” where environmental concern could stem from a positive effect of postmaterialist values or from a negative effect of experienced environmental degradation.

To analyze which of these different national conditions best explains cross-cultural differences on environmental behavior, the following contextual-level hypotheses were proposed: national wealth (Hypothesis 2a), societal-level postmaterialism (Hypothesis 2b), national information and educational development (Hypothesis 2c), and national objective and subjective environmental problems (Hypothesis 2d) are positively related to private and public environmental behaviors.

Multilevel Studies on Environmental Behavior

Many previous studies of cross-cultural differences on environmental behavior have used only individual-level survey data (Marquart-Pyatt, 2012b; Olofsson & Öhman, 2006), national-level aggregations of such data (Dalton, 2005; Duroy, 2008), or a combination of both levels without taking into account the nested or multilevel nature of the data (Freymeyer & Johnson, 2010; Gillham, 2008; Oreg & Katz-Gerro, 2006). In considering the national and individual level simultaneously, multilevel models avoid debate about which is the appropriate level (Dunlap & Mertig, 1997; Fairbrother, 2013) and do not run the risk associated with ignoring either one (Subramanian, Jones, Kaddour, & Krieger, 2009).¹

Using data from 15 countries of the European Union, Guerin et al. (2001) provided the first work that applies multilevel modeling to explain cross-cultural differences in environmental behavior. They examined how differences in national settings and in social and institutional factors interact with a series of individual characteristics to influence engagement in recycling. At the individual level, recycling was more frequent among older individuals with higher education and income, who participate in local environmental programs, who were worried about environmental problems and believed their government was making a reasonable effort to protect the environment. At the country level, greater rates of recycling were positively related to the percentage of citizens with membership in environmental organizations.

There are at least two other studies that used multilevel regressions to predict environmental behavior cross-culturally² (Hadler & Haller, 2011; Pirani & Secondi, 2011). The individual-level effects of these studies were described above (see “Individual-Level Predictors of Environmental Behavior” section). Hadler and Haller (2011), using data from the 2000 ISSP including 23 countries, showed at the country level that linkages to world society, national political opportunity structures, and resources all had a positive impact on public and private behaviors. Pirani and Secondi (2011), however, found a positive relationship between a country’s wealth and 7 of 8 private environmental behaviors, and mixed results with national total investment for the environment and percentage of waste treated on the total waste.

All studies discussed in the last two sections assume that national factors directly influence environmental behavior. However, it is possible that those contextual factors also influence behavior indirectly via cross-level interactions (see Figure 1). From this standpoint, the broader context could be operating as a mechanism that moderates (e.g., enhances or diminishes) the gap between attitudes and behavior.³ To explain this gap, many researchers have proposed a series of internal and external barriers that make it difficult to translate real concerns into action (Gifford, 2011; Kollmuss & Agyeman, 2002). The broader social, economic, and political national contexts have been recognized as one of these structural barriers (Leiserowitz, Kates, & Parris, 2005), which opens the door to multilevel models that consider cross-level interactions.

Analysis of cross-level interactions and contextual effects is an emerging area of research on environmental behavior. Nawrotzki (2012) found that the relationship between ideology and willingness-to-pay for environmental protection varies as a function of country-level variables. Conservatives had the strongest opposition to environmental protection in developed countries with better environmental conditions, but appeared to be more environmentally concerned in less developed countries with poor environmental quality. Pampel (2014; see also Marquart-Pyatt, 2012a) demonstrated that the positive relationship between several different measures of environmental support and socioeconomic status is stronger in more developed nations. In a study of 34 different countries, Liu and Sibley (2012) found that in countries with a Human Development Index, there was a stronger positive correlation between the perceived importance of global warming and self-reported intentions to make personal sacrifices to protect the environment. However, none of the four studies cited above or others (Franzen & Meyer, 2010) examined cross-level effects (i.e., the effect of a Level-2 predictor on a Level-1 slope coefficient) using environmental behaviors as the dependent variable.

These considerations lead to the following cross-level hypothesis: In addition to the direct effect (see Hypothesis 2a-2d), the national context will indirectly influence the strength of association between attitudes and behavior. We predict that the relationship between environmental attitudes and behavior will be stronger in countries with higher levels of economic and social development and environmental degradation (Hypothesis 3). Developed countries have fewer barriers and more resources that facilitate environmental behavior.

Method

To test the hypotheses formulated above, the analysis uses individual-level data from the 2010 Environmental Module of the ISSP (see Haller, Jowell, & Smith, 2009). The data include more than 38,000 individuals in 30 countries: Argentina, Austria, Belgium, Bulgaria, Canada, Chile, Croatia, Czech Republic, Denmark, Finland, Germany, Great Britain, Israel, Japan, Republic of Korea, Latvia, Mexico, New Zealand, Norway, Philippines, Russian Federation, Slovakia, Slovenia, South Africa, Spain, Sweden, Switzerland, Taiwan, Turkey, and the United States. The sample of countries is composed largely of industrialized, higher income nations (two thirds of them are members in the Organisation for Economic Co-operation and Development [OECD] and one third are developing nations). The standard sampling procedure is a stratified, multistage random sample considering region, household, and person within the household. The target population is the adult population permanently living in civilian households. Sample size is about 1,000 in most countries. To preserve sample sizes, missing data were accounted for using multiple imputation procedures, specifically the expectation–maximization algorithm in SPSS 20. This procedure was implemented for each of the 30 countries separately; inferences were thus made using only data from that particular country (i.e., data from Spain were used only in the creation of the imputed data set for Spain). Country-level data come from different sources described below.

Dependent Variables: Private and Public Environmental Behaviors

Previous works using exploratory factor analyses (EFA) of ISSP data had found a two-factor structure of environmental behavior that was similar across countries (Hadler & Haller, 2011; Hunter et al., 2004). Following these results, two measures of environmental behavior are created using confirmatory factor analysis (CFA). The first latent construct, private environmental behavior, contains four items that ask survey participants how often (4 = never to 1 = always) they recycle, avoid buying certain products for environmental reasons, reduce energy at home, and save or reuse water for environmental reasons. The second factor, public environmental behavior, contains three dichotomous items (1 = yes and 2 = no) that ask survey participants if they are a member of an environmental organization, if they have signed a petition, and if they have donated money during the last 5 years to an environmental group. Factor scores of the two scales are used for the analysis, and items are scaled so that higher scores indicate proenvironmental responses. CFAs results indicate good fit for both scales.⁴

Individual-Level Independent Variables

We include the following demographic variables from the ISSP: gender (0 = male and 1 = female), age (in years), education (0 = no formal qualification to 5 = university degree), household income (z-standardized because income is reported in country-specific currencies), and adjusted for household size following Franzen and Meyer (2010) and the size of hometown ranking from rural (= 1) to urban (= 5).

The ISSP environmental module includes several questions referring to different attitudinal and knowledge dimensions that were successfully used in past environmental cross-cultural research (Hadler & Haller, 2011; Haller & Hadler, 2008; Marquart-Pyatt, 2012b; Oreg & Katz-Gerro, 2006). We applied CFAs to these questions and derived four dimensions that we call environmental risk perception, self-perceived knowledge, efficacy, and willingness to make personal sacrifice. The factor structure is similar across countries and the following describes the scales derived from CFAs of the pooled sample.

Environmental risk perception contains six items showing an awareness of the environmental consequences of societies’ modern industrial activities. Respondents were asked, “In general, do you think [air pollution caused by cars; air pollution caused by industry; pollution of country’s rivers, lakes, and streams; pesticides and chemicals used in farming; the rise in the world’s temperature caused by the ‘greenhouse effect’; and modifying the genes of certain crops] is extremely dangerous for the environment, very dangerous, somewhat dangerous, not very dangerous, or not dangerous at all for the environment?”

Self-perceived environmental knowledge is a two-item latent construct that asks respondents to self-evaluate their level of understanding of the causes and solutions for environmental problems on a 5-point scale. Environmental self-efficacy contains three items on a 5-point scale ranging from 1 = strongly agree to 5 = strongly disagree. The first indicator is “It is just too difficult for someone like me to do much about the environment.” The second is “There is no point in doing what I can for the environment unless others do the same.” The third statement is “I find it hard to know whether the way I live is helpful or harmful to the environment.”

Finally, willingness to make personal sacrifice includes three items on a 5-point scale ranging from 1 = very willing to 5 = very unwilling. The content related to the willingness to pay higher prices and taxes and to reduce one’s own living standards to protect the environment. Factor scores of the four scales are used for the analysis, and some items are scaled so that higher scores indicate proenvironmental responses. CFAs results indicate good fit for all the latent factors.⁵

Multigroup CFAs were performed to test for the metric invariance of all the individual-level measures across countries. Metric invariance is tested by constraining both the factor structure and factor loadings of each measure to be equal across groups. It is satisfied if the basic model structure and loading weights are invariant across groups, indicating that participants from different countries conceptualize the constructs and respond to the items in a similar way (Milfont & Fischer, 2010; Vandenberg & Lance, 2000). We also tested for scalar invariance by constraining the item intercepts to be equal across groups. Establishing scalar invariance indicates that observed scores are related to the latent scores; that is, individuals who have the same score on the latent construct would obtain the some score on the observed variable regardless of their group membership. The analyses presented in Table 1 indicate configural and metric invariance of the measures and scalar variance; thus, obtained ratings can be compared across groups and observed item differences will indicate group differences in the underlying latent construct.

Table 1.

Fit Indices for Invariance Test.

	Model fit					Comparative fit indices
Model	χ² (df)	χ²/df	CFI	TLI	RMSEA	p (Δχ²)	ΔCFI	ΔTLI	ΔRMSEA	Comparison	Decision
1. Full configural invariance	18,573 (5183)	3.58	.942	.930	.008	—	—	—	—	—	Accept
2. Full metric invariance	22,533 (5618)	4.01	.927	.918	.009	<.0001	.015	.012	.001	Model 1 vs. Model 2	Accept
3. Full scalar invariance	82,607 (6227)	13.26	.671	.667	.017	<.0001	.256	.251	.008	Model 2 vs. Model 3	Reject

Note. CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error approximation.

Country-Level Independent Variables

There are six country-level independent variables: two measures related to the wealth of a nation, two related to educational development, and two related to environmental conditions.

To measure national wealth or affluence, we use 2010 per capita GDP in purchasing power parity (PPP US$ 2005; World Bank, 2013). Country-level postmaterialism scores are derived by averaging individuals’ postmaterialism scores within each country of the ISSP.⁶ We measure education development using the 2010 Education Index taken from the International Human Development Indicator data set (Human Development Report, 2013). The index comprises the adult literacy rate (with two-thirds weighting) and the combined primary, secondary, and tertiary gross enrollment ratio (with one-third weighting) of each country. We also consider the density of domestic environmental NGOs as a proxy for the level of diffusion of environmental information in a country (for the methodology, see Longhofer & Schofer, 2010).

For perceived environmental degradation, we gathered a single question from the Gallup (2011) World Poll Survey that measured personal satisfaction of water quality. Respondents from national representative samples were asked, “In the city or area where you live, are you satisfied or dissatisfied with the quality of water?” and we used the country mean of people that expressed satisfaction with the quality of this resource (Gallup, 2011). According to Guagnano, Stern, and Dietz (1995), “attitude theory needs to be modified to include not only the perception of external conditions but the external conditions themselves” (p. 715). Thus, we included the 2010 Environmental Health component of the environmental performance index (EPI) that covers three objective categories (environmental burden of disease, water resources for human health, and air quality for human health) through five performance indicators (Emerson et al., 2010). These two last macrovariables were multiplied by a negative one (−1) so that higher values represent lower environmental quality (or greater degradation).

Finally, using principal components analysis, we determined that the six variables consistently loaded on one factor. Therefore, we created a synthetic measure containing economic, educational, and environmental information of each country that we called “Development Index”⁷ (Cronbach’s α = .70, and standardized factor loadings greater than 0.57).

Analysis

A multilevel design was applied using the MIXED Menu Command of SPSS 20. This technique was specifically developed to examine which part of the variance in a dependent variable is attributable to individual- as opposed to country-level characteristics (Raudenbush & Bryk, 2002).

The analysis proceeded in several steps. First, a null model was estimated, that is, a model without explanatory variables. This is also called a one-way ANOVA with random effects model. Here, we modeled an individual’s level of environmental behavior (Y_ij, private or public) as follows:

Individual level:

Y_{i j} = β_{0 j} + r_{i j},

Contextual level:

β_{0 j} = γ_{00} + u_{0 j},

where

β_0j = mean environmental behavior for country_j,

γ₀₀ = grand mean environmental behavior,

Variance (r_ij) = σ² = within-country variance in environmental behavior, and

Variance (u_0j) = τ₀₀ = between-country variance in environmental behavior.

The first model allows us to estimate the intraclass correlation coefficient (ICC) that gives the proportion of the total variance that exists among countries and which is defined as ICC = τ₀₀ / (τ₀₀ + σ²). A low value indicates that there is little variance among countries. Most of the variance can then be observed among individuals.

After this basic model, the hypotheses were systematically tested in several models. We estimate one-way ANCOVA models with random effects, which include all the individual-level variables with their regression slopes fixed but exclude the contextual variables. The ANCOVA models have two different specifications—a constrained model that only has demographic variables and a full model that includes demographic and environmental attitude variable. These models assume that the slope coefficients for the individual-level variables are “country independent” or “fixed.”

The one-way ANCOVA models estimate how much of the variance within and among countries can be explained by compositional effects. Compositional effects occur when intergroup differences in an outcome are the result of differences in group composition, that is, in the characteristics of the individuals who make up the group (Gelissen, 2007). In other words, if individual characteristics that are related to environmental behavior are unequally distributed across nations, then the difference in group composition partially accounts for observed differences in environmental behavior across nations. Without the inclusion of such individual-level information, it is not possible to partition the extent to which individual versus contextual variables are related to the variance in the dependent variable (Gelissen, 2007).

For the next submodels, all individual-level variables are in a stepwise manner supplemented with country-level variables.⁸ In this way, we investigated whether differences among nations with respect to environmental behavior could be explained by either contextual effects or compositional effects (or a combination of both). Specifically, here we opted for the random intercept model with Level-1 covariates. In the random intercept model, countries differ with regard to the mean value of the dependent variable: the random intercept is the only random “country effect.” Individual-level coefficients are not allowed to vary randomly in these models. For example, a submodel including all individual-level variables and the development index can formally be presented as follows:

Individual level:

Environmental behavior = β_{0 j} + β_{1 j} (gender) + β_{2 j} (age) + β_{3 j} (education) + β_{4 j} (income) + β_{5 j} (size town) + β_{6 j} (knowledge) + β_{7 j} (environmental risk perception) + β_{8 j} (efficacy) + β_{9 j} (willingness) + r_{i j};

Country level:

β_{0 j} = γ_{00} + γ_{01} (development) + u_{0 j},

β_{1 j} = γ_{10},

β_{9 j} = γ_{90} .

Finally, we tested a series of slopes-as-outcomes models, which include country-level predictors to account for variation in the individual-level coefficients. These models include specific cross-level interactions to test whether the attitude–behavior relationship at the individual level is conditional on country-level context. The following equation shows one of the six slopes-as-outcomes submodels, using environmental risk perception as an example because it showed the strongest cross-level effect:

Individual level:

Environment behavior = β_{0 j} + β_{1 j} (environmental risk perception) + r_{i j};

Country level:

β_{0 j} = γ_{00} + γ_{01} (development) + u_{0 j},

β_{1 j} = γ_{10} + γ_{11} (development) + u_{8 j},

where

γ₁₀ = within-country environmental risk perception–behavior slope,

γ₁₁ = development_j × environmental risk perception_ij or the moderating effect of development on the effect of environmental risk perception on environmental behavior, and

Variance (u_7j) = τ₁₁ = between-country variance of the slope.

Results

From the two-level random intercept null model, the point estimate for the grand mean of private environmental behavior (private) is 1.59 and for public environmental behavior (public) is 0.45. The variance among countries (.022 private, .0007 public) turns out to be much smaller than the variance among individuals within countries (.16 private, .006 public). This is also reflected in the value of the ICC, which is .12 for private and .10 for public (12% and 10% of the total variance is among countries). However, a chi-square test of the estimated between-country variance component proved to be significant, χ²(29) = 3,507.2, p = .000 private; χ²(29) = 8,619.9, p = .000 public. Thus, the evidence indicates a small but significant variation among nations in their level of environmental behavior can potentially be explained by contextual characteristics.

The one-way ANCOVA with random effects models (Table 2) examine the extent to which individual-level explanatory variables are related to individual-level environmental behavior. Most of the direct associations found in this model are in accordance with our expectations. In Model 1 (attitudinal variables excluded), all sociodemographic indicators show positive relations with both types of environmental behaviors (with the exception of income that is negative but not significant for private behavior). There are higher levels of behaviors among women, adults, people with a high socioeconomic status, and those living in a large town. Nevertheless, these variables explain only a small amount of microvariance: 3% for private and 6% for public behavior. Moreover, if we add the attitudinal variables (Model 2), 4 of the 10 previous relationships are either no longer significant, have changed the sign, or became significant, while all attitudinal variables strongly and positively influence behavior (total explained microvariance: 30% for private and public behavior). Thus, these findings support our individual-level hypothesis (Hypotheses 1a and 1b).

Table 2.

Individual-Level Determinants of Private and Public Environmental Behaviors.

	Private behavior		Public behavior
	Model 1	Model 2	Model 1	Model 2
One-way ANCOVA with random effects	b (SE)	b (SE)	b (SE)	b (SE)
Intercept	1.593*** (.027)	1.593*** (.021)	0.452*** (.005)	0.451*** (.003)
Social bases
Gender (1 = female)	0.062*** (.004)	0.048*** (.003)	0.002* (.001)	0.002* (.001)
Age	0.003*** (<.001)	0.003*** (<.001)	0.001*** (.000)	0.001*** (<.001)
Education	0.036*** (.002)	−0.005*** (.001)	0.011*** (<.001)	0.002*** (<.001)
Household income	−0.001 (.002)	−0.014*** (.002)	0.006*** (<.001)	0.002*** (<.001)
Size of hometown	0.005** (.002)	−0.001 (.001)	0.001* (<.001)	0.000 (<.001)
Environmental attitudes
Self-perceived environmental knowledge	—	0.052*** (.002)	—	0.011*** (<.001)
Environmental risk perception	—	0.142*** (.003)	—	0.005*** (.001)
Efficacy	—	0.165*** (.003)	—	0.037*** (.001)
Willingness to make sacrifice	—	0.087*** (.002)	—	0.020*** (<.001)
ICC (null model)	.120	.120	.103	.103
Conditional ICC (with micro model)	.119	.112	.096	.058
Explained microvariance	3%	30%	6%	30%
Explained macrovariance	0%	34%	13%	63%

Note. Unstandardized coefficients, with standard errors in parentheses. Standard errors less than .001 indicated with <.001. Individual-level independent variables are grand mean centered. REML estimation, n_macro = 30, n_micro = 38,543. ICC = intraclass correlation coefficient; REML = restricted maximum likelihood.

p < .05. **p < .01. ***p < .001. (two-tailed test)

When we include all of these individual-level explanatory variables, the contextual-level variance goes down to .014 for private and .0003 for public environmental behavior (null model: .022 private, .0007 public). This indicates that compositional effects explain .34 and .63 of the variance at the contextual level, respectively.⁹ Here, the social composition of the populations under study differs substantially with respect to the individual-level explanatory variables, which causes the averages of these nations on these characteristics to differ.

Even after including all of the individual-level explanatory variables, there is still significant variability between countries (Wald Z = 3.783, p = .000, and Conditional ICC = .11 for private; Wald Z = 3.750, p = .000, and Conditional ICC = .06 for public). This supports our choice to estimate a series of random intercept models with Level-1 covariates that investigate contextual-level effects controlling for compositional effects.

Table 3 shows that national wealth and average postmaterialism are both positively related to higher levels of private and public environmental behavior. The inclusion of these variables increases the explanatory power of the macro model by about 10%. Hypotheses 2a and 2b are thus supported. Regarding the national information and educational development influences, Hypothesis 2c is partially confirmed. The level of education of a population and the presence of environmental NGOs are positively related to public environmental behavior but with only a small positive effective on private behavior. For national objective and subjective environmental problems, we found significant effects, but the direction of the relationship is opposite to what was expected (the higher the level of environmental degradation, the lower the level of private and public environmental behavior), thus Hypothesis 2d is not supported. Finally, the combination of all the contextual-level variables through the development index shows a significant positive effect on both types of behaviors with a final explained macrovariance of 44% for private and 79% for public environmental behavior.

Table 3.

The Effects of the Context on Private and Public Environmental Behavior.

Random intercept model with Level-1 covariates	Private behavior							Public behavior
Random intercept model with Level-1 covariates	b (SE)							b (SE)
Economic development
GDP	.004* (.002)	—	—	—	—	—	—	.001** (<.001)	—	—	—	—	—	—
Postmaterialism	—	.263^† (.138)	—	—	—	—	—	—	.055** (.017)	—	—	—	—	—
Educational development
ENGO density	—	—	.021 (.068)	—	—	—	—	—	—	.021* (.008)	—	—	—	—
Education index	—	—	—	.152 (.220)	—	—	—	—	—	—	.079** (.026)	—	—	—
Environmental degradation
Water satisfaction^a	—	—	—	—	−.005*** (.001)	—	—	—	—	—	—	−.001*** (<.001)	—	—
Environmental health^a	—	—	—	—	—	−.005* (.002)	—	—	—	—	—	—	−.001** (<.001)	—
Combined models
“Development index”	—	—	—	—	—	—	.049* (.020)	—	—	—	—	—	—	.011** (.002)
Explained macrovariance^b	43%	40%	34%	34%	60%	42%	44%	74%	72%	69%	72%	79%	72%	79%

Note. Unstandardized coefficients, with standard errors in parentheses. Standard errors less than .001 indicated with <.001. Individual-level independent variables are grand mean centered (not shown). Each national variable is included one at a time and each coefficient comes from a separate model. REML estimation, n_macro = 30, n_micro = 38,543. ENGO = environmental nongovernmental organization; REML = restricted maximum likelihood.

Reversed values.

Compositional effect explains 34% of the macrovariance for private environmental behavior and 63% for public environmental behavior (see micro model on Table 1).

†

p < .10. *p < .05. **p < .01. ***p < .001. (two-tailed test)

Finally, the slopes-as-outcomes models examine why in some countries the association between environmental attitudes and behavior is stronger than in others. Table 4 shows all the significant cross-level interactions detected. In particular, we found that two of four possible interactions are significant and positive for private behavior, and four of four for public behavior. Consistent with Hypothesis 3, these results indicate that the extent of economic, educational, and environmental development in a country enhances the positive relationship between several environmental attitudes and both types of environmental behaviors. In other words, as the level of development of a country increases, there is a stronger correlation between environmental attitudes and environmental behavior.

Table 4.

Cross-Level Interactions of Different Attitudinal-Behavior Slopes and Development Index.

	Private behavior	Public behavior
Slopes as outcomes model	b (SE)	b (SE)
Knowledge–behavior slope	.147*** (.023)	.037*** (.001)
Knowledge × Development index	.006 (.007)	.008*** (.001)
Environmental risk–behavior slope	.255*** (.012)	.031*** (.002)
Environmental risk × Development index	.046** (.012)	.018*** (.002)
Efficacy–behavior slope	.273*** (.013)	.063*** (.003)
Efficacy × Development index	.012 (.013)	.016*** (.003)
Willingness to make sacrifice–behavior slope	.157*** (.005)	.037*** (.001)
Willingness × Development index	.017** (.005)	.010*** (.001)

Note. Unstandardized coefficients, with standard errors in parentheses. Standard errors less than .001 indicated with <.001. Individual-level independent variables are group mean centered and entered only one at a time as covariates. REML estimation, n_macro = 30, n_micro = 38,543. REML = restricted maximum likelihood.

p < .05. **p < .01. ***p < .001. (two-tailed test)

To visualize this important finding, Figure 2 graphs predicted environmental behaviors values by level of environmental risk perception for countries at 1 standard deviation below mean development index, at the mean, and 1 standard deviation above the mean development index (corresponding to absolute development index values of −1.74, −0.07, 1.16, values roughly of Turkey, South Korea, and Germany). The environmental risk perception gradient is successively more positive for the nations: 0.051 for the low-developed nation, 0.095 for the mean nation, and 0.099 for the high-developed nation for private environmental behavior; and 0.0055 for the low-developed nation, 0.0145 for the mean nation, and 0.0205 for the high-developed nation for public environmental behavior.

Figure 2.

Slopes relating environmental risk perception to environmental behaviors by level of development.

Discussion and Conclusion

This article provides new cross-national evidence of how and to what extent environmental behavior is related to individual and contextual-level characteristics. The multilevel analysis of the ISSP 2010 demonstrates that environmental behavior varies between countries and within countries. For the within-country differences, similar to previous research (Hadler & Haller, 2011; Marquart-Pyatt, 2012a; Oreg & Katz-Gerro, 2006), we found that sociodemographic characteristics have a significant effect, although small compared with the effect exerted by environmental attitudes, and all together these variables explain a fair amount of micro-level variance.

The between-country differences highlight the importance of considering compositional effects (Gelissen, 2007). We found that a substantial proportion of the macro-level variance in environmental behavior is explained by the individual-level variables. These results demonstrate how compositional effects are reducing or outshining the relative influence of national-level factors, and this fact could be explaining the contradictory results found in the literature about competing national-level predictors. Also, if we consider that this effect has not been estimated in the vast majority of the cross-national studies reviewed, our analyses represent an original contribution to the field and a call for future research to consider this issue.

Even after controlling for compositional effects, citizens of countries with high levels of national wealth and postmaterialist values appear to have more individual involvement in environmental protection than citizens of countries with lower levels of wealth and postmaterialism. These findings are similar to previous research (Dalton, 2005; Freymeyer & Johnson, 2010; Inglehart, 1995; Oreg & Katz-Gerro, 2006; Pirani & Secondi, 2011), thus confirming the positive effect of wealth and postmaterialism in facilitating the emergence of environmental behavior. Also, we found that postindustrialized societies that have more dense communication structures and mass education (measured through environmental NGOs density and level of education of the population) tend to display more environmental engagement and activism than developing countries. Thus, all these interrelated factors seem to be creating better conditions and opportunities to foster environmental behavior.

Contrary to our expectation, environmental degradation negatively affects environmental behavior, which contradicts Inglehart’s “objective problems” hypothesis. Several explanations of this result are possible. First, in a cross-section data set, it is hard to observe the direction of causal influence between environmental behavior and objective conditions. Countries with poor environmental conditions may not have enough environmental behavior to change those conditions (Freymeyer & Johnson, 2010). Second, the overall level of development could be jointly influencing both environmental behaviors and conditions, which would create a spurious negative correlation. Supporting this idea is the high correlation between indicators of environmental conditions and other contextual variables related to development (e.g., r = −.842, p < .01 for unhealthy environment and GDP). Third, we used environmental indicators where developed countries perform better but could instead consider environmental problems related to overindustrialization and consumption, where developed countries perform worse.¹⁰ In this case, the objective conditions hypothesis would need to differentiate between the basic environmental problems of developing countries versus the environmental problems related to patterns of production and consumption (Dalton, 2005). Future research must consider this dimensionality argument and test the effect of different environmental indicators in cross-national studies.

Finally, and most importantly, we found that the attitude–behavior correspondence is stronger in more developed countries. Developed countries have fewer “external barriers” that make it harder to match people’s intentions to behavior, thus making it easier to translate a set of environmental attitudes into real actions. Apart from the external barriers, “internal or psychological barriers” may be possible explanations for the cross-national variations in attitude–behavior correspondence. For instance, if people in a country generally do not trust others, or do not believe that things are within their control, they might think that environmental behavior will not pay off even if they are concerned about current environmental conditions. The moderation effect detected may also contribute to the mixed results found in the literature about the lack of correspondence between attitudes and behavior. Therefore, these findings represent a novel contribution to the field and offer a promising avenue for examining more complex and interactive models of environmental behavior.

Although these findings point out the necessity of simultaneously assessing the effects of individual and contextual-level characteristics on environmental behavior, a number of limitations deserve mentioning. The sample of countries analyzed here is composed largely of industrialized, higher income nations thus posing problems for the generalizability of the presented findings. The cross-sectional research design has a limited capacity to untangle complex causal relationships, such as how changes in objective environmental conditions might influence environmental behaviors and vice versa. These two limitations can be addressed in future studies with the use of databases that cover more countries (e.g., the World Value Survey for a larger number of countries) or collect data over time (e.g., Understanding Society: The U.K. household longitudinal study for panel data). Finally, there is still a substantial amount of unexplained variance in the models. Besides the use of micro- and macro-level data, future work should consider and incorporate meso-level data (e.g., household-level factors, social networks, subnational communities) for a more integrative multilevel theory of environmental behavior.

In addition to the theoretical and methodological contribution, the results suggest some practical implications. Considering that environmental behavior depends on the interaction between individual and contextual factors, successful interventions must consider them simultaneously—that is, reducing contextual barriers and enhancing personal attitudes and disposition. Social scientists and policy makers who emphasize internal processes advocate interventions such as education and persuasion as the best way to change undesirable behavior and motivate desirable ones. Scientists and policy makers who emphasize external factors advocate interventions such as regulations or taxes to change behavior. These policy interventions will fall short if they neglect the interaction between individual and contextual factors (Guagnano et al., 1995; Thøgersen, 2005).

The results also imply that contextual barriers to individual action will reinforce inequality in environmental outcomes across countries with different levels of development. Even if individuals within less developed countries desire higher levels of environmental activism, they will have difficulty finding the socioeconomic or cultural resources to help them act on their intentions. To the extent that citizen environmental behavior is necessary for overall environmental improvement, this negative feedback pattern is one reason for the sustained gap between more and less developed countries. Hence, effective environmental and economic development policy may have the indirect effect of reducing economic and social barriers that inhibit environmental behaviors. These same types of barriers may also affect individual behavior in other policy areas like health or education, where successful policy requires both private cooperation and public political activism.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Ignacio Pisano, PhD, is an independent researcher recently interested in transpersonal psychology and in issues such as spiritual self-development, self beyond the ego, altered states of consciousness, spiritual practices, fasting, meditation, and other sublime expanded experiences of living.

Mark Lubell is a professor of environmental science and policy at the University of California Davis, where he codirects the Center for Environmental Policy and Behavior (). His research focuses on cooperation and institutions in the context of environmental policy.

References

Alker

H. R.

(1969). A typology of ecological fallacies. In Dogan

Rokkan

(Eds.), Quantitative ecological analysis in the social sciences (pp. 69-86). Cambridge, MA: The MIT Press.

Brady

H. E.

Verba

Schlozman

K. L.

(1995). Beyond SES: A resource model of political participation. American Political Science Review, 89, 271-294. doi:10.2307/2082425

Corral-Verdugo

Zaragoza

(2000). Bases sociodemográficas y psicológicas de la conducta de reutilización: un modelo estructural [Socio-demographic and psychological basis of reuse-behavior: A structural model]. Medio Ambiente Y Comportamiento Humano, 1, 9-29.

Dalton

R. J.

(2005). The greening of the globe? Cross-national levels of environmental group membership. Environmental Politics, 14, 441-459. doi:10.1080/09644010500175783

Diamantopoulos

Schlegelmilch

B. B.

Sinkovics

Bohlen

G. M.

(2003). Can socio-demographics still play a role in profiling green consumers? A review of the evidence and an empirical investigation. Journal of Business Research, 56, 465-480. doi:10.1016/S0148-2963(01)00241-7

Dietz

Stern

P. C.

Guagnano

G. A.

(1998). Social structural and social psychological bases of environmental concern. Environment and Behavior, 30, 450-471. doi:10.1177/001391659803000402

Dunlap

R. E.

Mertig

A. G.

(1995). Global concern for the environment: Is affluence a prerequisite? Journal of Social Issues, 51(4), 121-137.

Dunlap

R. E.

Mertig

A. G.

(1997). Global environmental concern: An anomaly for postmaterialism. Social Science Quarterly, 78, 24-29.

Duroy

Q. M. H.

(2008). Testing the affluence hypothesis: A cross-cultural analysis of the determinants of environmental action. The Social Science Journal, 45, 419-439. doi:10.1016/j.soscij.2008.07.001

10.

Emerson

Esty

Levy

Kim

Mara

de Sherbinin

Srebotnjak

(2010). 2010 Environmental Performance Index. New Haven, CT: Yale Center for Environmental Law & Policy.

11.

Escobar

(2006). An ecology of difference: Equality and conflict in a glocalized world. Focaal, 47, 120-137. doi:10.3167/092012906780646415

12.

Fairbrother

(2013). Rich people, poor people, and environmental concern: Evidence across nations and time. European Sociological Review, 29, 910-922. doi:10.1093/esr/jcs068

13.

Franzen

Meyer

(2010). Environmental attitudes in cross-national perspective: A multilevel analysis of the ISSP 1993 and 2000. European Sociological Review, 26, 219-234. doi:10.1093/esr/jcp018

14.

Freymeyer

R. H.

Johnson

B. E.

(2010). A cross-cultural investigation of factors influencing environmental actions. Sociological Spectrum, 30, 184-195.

15.

Gallup. (2011). World Poll Survey. Retrieved from http://www.gallup.com/poll/154646/air-quality-rated-better-water-quality-worldwide.aspx#2

16.

Gelissen

(2007). Explaining popular support for environmental protection: A multilevel analysis of 50 nations. Environment and Behavior, 39, 392-415. doi:10.1177/0013916506292014

17.

Gifford

(2011). The dragons of inaction: Psychological barriers that limit climate change mitigation and adaptation. American Psychologist, 66, 290-302. doi:10.1037/a0023566

18.

Gillham

P. F.

(2008). Participation in the environmental movement analysis of the European Union. International Sociology, 23, 67-93. doi:10.1177/0268580907084386

19.

Guagnano

G. A.

Stern

P. C.

Dietz

(1995). Influences on attitude-behavior relationships: A natural experiment with curbside recycling. Environment and Behavior, 27, 699-718. doi:10.1177/0013916595275005

20.

Guerin

Crete

Mercier

(2001). A multilevel analysis of the determinants of recycling behavior in the European countries. Social Science Research, 30, 195-218. doi:10.1006/ssre.2000.0694

21.

Hadler

Haller

(2011). Global activism and nationally driven recycling: The influence of world society and national contexts on public and private environmental behavior. International Sociology, 26, 315-345. doi:10.1177/0268580910392258

22.

Haller

Hadler

(2008). Dispositions to act in favor of the environment: Fatalism and readiness to make sacrifices in a cross-national perspective. Sociological Forum, 23, 281-311. doi:10.1111/j.1573-7861.2008.00059.x

23.

Haller

Jowell

Smith

T. W.

(Eds.). (2009). The International Social Survey Programme, 1984-2009: Charting the globe (1st ed.). Abingdon, UK: Routledge.

24.

Human Development Report. (2013). Public data explorer. Retrieved from http://hdr.undp.org/en/data/explorer/

25.

Hunter

L. M.

Hatch

Johnson

(2004). Cross-national gender variation in environmental behaviors. Social Science Quarterly, 85, 677-694.

26.

Inglehart

(1995). Public support for environmental protection: Objective problems and subjective values in 43 societies. Political Science & Politics, 28, 57-72. doi:10.2307/420583

27.

Johnson

Brace

Arceneaux

(2005). Public opinion and dynamic representation in the American states: The case of environmental attitudes. Social Science Quarterly, 86, 87-108. doi:10.1111/j.0038-4941.2005.00292.x

28.

Kollmuss

Agyeman

(2002). Mind the gap: Why do people act environmentally and what are the barriers to pro-environmental behavior? Environmental Education Research, 8, 239-260. doi:10.1080/13504620220145401

29.

Leiserowitz

A. A.

Kates

R. W.

Parris

T. M.

(2005). Do global attitudes and behaviors support sustainable development? Environment: Science and Policy for Sustainable Development, 47(9), 22-38.

30.

Liu

J. H.

Sibley

C. G.

(2012). Hope for the future? Understanding self-sacrifice among young citizens of the world in the face of global warming. Analyses of Social Issues and Public Policy, 12, 190-203. doi:10.1111/j.1530-2415.2011.01275.x

31.

Longhofer

Schofer

(2010). National and global origins of environmental association. American Sociological Review, 75, 505-533. doi:10.1177/0003122410374084

32.

Marquart-Pyatt

S. T.

(2012a). Contextual influences on environmental concerns cross-nationally: A multilevel investigation. Social Science Research, 41, 1085-1099. doi:10.1016/j.ssresearch.2012.04.003

33.

Marquart-Pyatt

S. T.

(2012b). Explaining environmental activism across countries. Society & Natural Resources, 25, 683-699. doi:10.1080/08941920.2011.625073

34.

Martinez-Alier

(2003). The environmentalism of the poor: A study of ecological conflicts and valuation. Cheltenham, UK: Edward Elgar.

35.

Milfont

Fischer

(2010). Testing measurement invariance across groups: Applications in cross-cultural research. International Journal of Psychological Research, 3, 111-130.

36.

Milfont

T. L.

(2012). Cultural differences in environmental engagement. In Clayton

S. D.

(Ed.), The Oxford handbook of environmental and conservation psychology (pp. 181-202). New York, NY: Oxford University Press.

37.

Milfont

T. L.

Richter

Sibley

C. G.

Wilson

M. S.

Fischer

(2013). Environmental consequences of the desire to dominate and be superior. Personality and Social Psychology Bulletin, 39, 1127-1138. doi:10.1177/0146167213490805

38.

Mostafa

M. M.

(2013). Wealth, post-materialism and consumers’ pro-environmental intentions: A multilevel analysis across 25 nations. Sustainable Development, 21, 385-399. doi:10.1002/sd.517

39.

Nawrotzki

R. J.

(2012). The politics of environmental concern a cross-national analysis. Organization & Environment, 25, 286-307. doi:10.1177/1086026612456535

40.

Olofsson

Öhman

(2006). General beliefs and environmental concern: Transatlantic comparisons. Environment and Behavior, 38, 768-790. doi:10.1177/0013916506287388

41.

Oreg

Katz-Gerro

(2006). Predicting proenvironmental behavior cross-nationally: Values, the theory of planned behavior, and value-belief-norm theory. Environment and Behavior, 38, 462-483. doi:10.1177/0013916505286012

42.

Pampel

F. C.

(2014). The varied influence of SES on environmental concern. Social Science Quarterly, 95, 57-75. doi:10.1111/ssqu.12045

43.

Pellow

D. N.

Brulle

R. J.

(Eds.). (2005). Power, justice, and the environment: A critical appraisal of the environmental justice movement. Cambridge, MA: The MIT Press.

44.

Pirani

Secondi

(2011). Eco-friendly attitudes: What European citizens say and what they do. International Journal of Environmental Research, 5, 67-84.

45.

Pisano

Hidalgo

M. C.

(2014). Testing the cross-national social bases of environmentalism: A current and comparative analysis of conservation behaviors. Psico, 45, 395-405.

46.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models: Applications and data analysis methods (2nd ed.). Thousand Oaks, CA: Sage.

47.

Snijders

T. A. B.

Bosker

R. J.

(2011). Multilevel analysis: An introduction to basic and advanced multilevel modeling. London, England: Sage.

48.

Spaargaren

Mol

A. P. J.

(1992). Sociology, environment, and modernity: Ecological modernization as a theory of social change. Society & Natural Resources, 5, 323-344. doi:10.1080/08941929209380797

49.

Subramanian

S. V.

Jones

Kaddour

Krieger

(2009). Revisiting Robinson: The perils of individualistic and ecologic fallacy. International Journal of Epidemiology, 38, 342-360. doi:10.1093/ije/dyn359

50.

Thøgersen

(2005). How may consumer policy empower consumers for sustainable lifestyles? Journal of Consumer Policy, 28, 143-177. doi:10.1007/s10603-005-2982-8

51.

Vandenberg

R. J.

Lance

C. E.

(2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3, 4-70. doi:10.1177/109442810031002

52.

World Bank. (2013). World DataBank. Available from http://databank.worldbank.org