Professional Documents
Culture Documents
Environmental Behaviour in Cross Cultural
Environmental Behaviour in Cross Cultural
research-article2015
EABXXX10.1177/0013916515600494Environment and BehaviorPisano and Lubell
Article
Environment and Behavior
2017, Vol. 49(1) 3158
Environmental Behavior 2015 SAGE Publications
Reprints and permissions:
in Cross-National sagepub.com/journalsPermissions.nav
DOI: 10.1177/0013916515600494
Perspective: A Multilevel eab.sagepub.com
Analysis of 30 Countries
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 multi-
level 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., composi-
tional effects). For example, our analysis finds that individual-level attitudi-
nal 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 retro-
spective 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
Pisano and Lubell 33
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.
Country level
Individual level
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 mem-
bers in the Organisation for Economic Co-operation and Development
[OECD] and one third are developing nations). The standard sampling proce-
dure is a stratified, multistage random sample considering region, household,
and person within the household. The target population is the adult popula-
tion 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 expectationmaximi-
zation 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.
Pisano and Lubell 39
Model 2 (df) 2/df CFI TLI RMSEA p (2) CFI TLI RMSEA 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 = TuckerLewis index; RMSEA = root mean square error approximation.
41
42 Environment and Behavior 49(1)
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
Pisano and Lubell 43
Yij = 0 j + rij ,
Contextual level:
0 j = 00 + u0 j ,
where
0j = mean environmental behavior for countryj,
00 = grand mean environmental behavior,
Variance (rij) = 2 = within-country variance in environmental behavior, and
Variance (u0j) = 00 = 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 coun-
tries and which is defined as ICC = 00 / (00 + 2). 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 sev-
eral 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 specificationsa constrained model that only has demographic
variables and a full model that includes demographic and environmental atti-
tude variable. These models assume that the slope coefficients for the indi-
vidual-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 dif-
ferences in group composition, that is, in the characteristics of the individuals
who make up the group (Gelissen, 2007). In other words, if individual char-
acteristics that are related to environmental behavior are unequally distrib-
uted 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
44 Environment and Behavior 49(1)
Environmental behavior = 0 j +
1 j ( gender ) + 2 j ( age ) + 3 j ( education ) + 4 j ( income ) +
5 j ( size town ) + 6 j ( knowledge ) + 7 j
( environmental risk perception ) +
8 j ( efficacy ) + 9 j ( willingness ) + rij ;
Country level:
0 j = 00 + 01 ( development ) + u0 j ,
1 j = 10 ,
9 j = 90 .
0 j = 00 + 01 ( development ) + u0 j ,
1 j = 10 + 11 ( development ) + u8 j ,
where
10 = within-country environmental risk perceptionbehavior slope,
11 = developmentj environmental risk perceptionij or the moderating effect
of development on the effect of environmental risk perception on environ-
mental behavior, and
Variance (u7j) = 11 = 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 environ-
mental 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, 2(29) = 3,507.2,
p = .000 private; 2(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 individ-
ual-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 pub-
lic behavior). Thus, these findings support our individual-level hypothesis
(Hypotheses 1a and 1b).
Table 2. Individual-Level Determinants of Private and Public Environmental Behaviors.
46
Private behavior Public behavior
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, nmacro = 30, nmicro = 38,543. ICC = intraclass correlation coefficient; REML =
restricted maximum likelihood.
*p < .05. **p < .01. ***p < .001. (two-tailed test)
Pisano and Lubell 47
Economic development
GDP .004* .001**
(.002) (<.001)
Postmaterialism .263 .055**
(.138) (.017)
Educational development
ENGO density .021 .021*
(.068) (.008)
Education index .152 .079**
(.220) (.026)
Environmental degradation
Water satisfactiona .005*** .001***
(.001) (<.001)
Environmental healtha .005* .001**
(.002) (<.001)
Combined models
Development index .049* .011**
(.020) (.002)
Explained macrovarianceb 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, nmacro = 30,
nmicro = 38,543. ENGO = environmental nongovernmental organization; REML = restricted maximum likelihood.
aReversed values.
bCompositional 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)
Pisano and Lubell 49
Funding
The author(s) received no financial support for the research, authorship, and/or publi-
cation of this article.
Notes
1. In most of the previous research, multilevel data have been analyzed with what
have been called aggregation and disaggregation techniques. In aggregating
individual-level data to the level of a country, variation at the individual level is
lost. In disaggregating data or assigning the same value for a variable measured
at the national level to all individual-level data, the individual cases are cor-
related, which violates the assumption of independence of measures in linear
regression. Either of these techniques can lead to incorrect estimates and results
as a consequence (Raudenbush & Bryk, 2002). Also, researchers who rely on
these strategies are prone to commit what are commonly referred to as the eco-
logical and individualistic fallacies (i.e., the invalid transfer of aggregate results
to individuals and vice versa; Alker, 1969).
54 Environment and Behavior 49(1)
2. An intensive search was carried out for relevant articles in large databases
(Google Scholar, ISI Web of Science, Scopus, etc.) covering the period from
1995 to June 2013. Keywords used were environmental behavior, multilevel
models, cross-cultural, or a series of similar terms (conservation or ecological
behavior, recycling, hierarchical, mixed-effect or random coefficient models,
cross-national comparison, etc.). A secondary search was performed using the
same databases to locate any article that had cited the three papers identified
from the previous search (Guerin, Crete, & Mercier, 2001; Hadler & Haller,
2011; Pirani & Secondi, 2011). In addition, most of the reference lists of the
studies located during the search were reviewed as well as the 2013 bibliography
list produced by the International Social Survey Program (ISSP).
3. In fact, Marquart-Pyatt (2012a) found that pathways to activism (private behav-
ior) differ across countries. In particular, she found that environmental attitudes
have no significant influences on behavior in 4 of the 16 countries analyzed, as
well as other attitudinal-behavioral gaps (5/16 for awareness, 6/16 for efficacy,
and 2/16 for willingness to make personal contribution for the environment).
4. For the sample of all countries pooled, standardized factor loadings range
from 0.42 to 0.76 for private behavior, and 0.48 to 0.70 for public behavior
(all significant, p < .001). Overall model fit statistics are very good; private:
2(2) = 305,596, p = .000, goodness-of-fit index (GFI) = .996, normed fit index
(NFI) = .990, root mean square error approximation (RMSEA) = .063; public:
2(13) = 1,484, p = .000, GFI = .989, NFI = .967, RMSEA = .054. Because
a three-item model is exactly identified within a confirmatory factor analysis
(CFA), results for public environmental behavior come from a single CFA that
include a correlation between both types of behavior. Also we conduced country-
by-country CFAs finding support for the structure proposed (more information
is available on request).
5. For the sample of all countries pooled, standardized factor loadings range from
0.57 to 0.71 for risk perception, 0.74 to 0.92 for self-perceived environmental
knowledge, 0.50 to 0.65 for efficacy, and 0.67 to 0.85 for willingness to make
personal sacrifice (all significant, p < .001). Overall model fit statistics are good
for these measures; risk perception: 2(8) = 47,140, p = .000, GFI = .986, NFI
= .974, RMSEA = .075; knowledge: 2(18) = 1,857, p = .000, GFI = .989, NFI
= .979, RMSEA = .051; efficacy and willingness: 2(8) = 677, p = .000, GFI =
.994, NFI = .989, RMSEA = .047. Because a two-item model is unidentified
within a CFA, results for self-perceived environmental knowledge come from a
single CFA that includes a correlation with risk perception and, because a three-
item model is exactly identified within a CFA, results for efficacy and willing-
ness come from a single CFA that includes a correlation between them. Also, we
conduced country-by-country CFAs finding support for the structures proposed
(more information is available on request).
6. The short Ingleharts materialism/postmaterialism values scale uses the items
maintain order in the nation, give people more say in government decisions,
fight rising prices, and protect freedom of speech. Respondents were asked
Pisano and Lubell 55
what should be the highest priority in their country and afterwards what should
be the second priority. Statements 2 and 4 are considered postmaterialistic items
and respondents who picked no postmaterialistic items are coded as 0 (= materi-
alistic), those who chose one postmaterialistic item are coded as 1 (= mixed), and
those who picked two postmaterialistic goals as 2 (= postmaterialistic).
7. In this case, the two environmental macrovariables were not recoded, thus higher
values represent higher levels of environmental quality.
8. Because of close overlap among contextual-level variables (most correlations
above .45), each national variable is included one at a time and each coefficient
comes from a separate model (Hadler & Haller, 2011).
9. We apply the formula of Snijders and Bosker (2011) where explained macrovar-
iance = 1 (macrovariance of final model / macrovariance of empty model). In
our case 1 (0.0143 / 0.0217) = .34 for private behavior, and 1 (0.000264 /
0.00071) = .63 for public behavior.
10. In fact, the correlation between country-level GDP and environmental footprint
is positive and very strong (r = .739, p < .01), and in a random intercept model
with Level-1 covariates similar to those in Table 3, environmental footprint was
found to be positive and significantly related to public environmental behavior
(b = .005, SE = .002, p = .009) and, although found to be positive, it was not
significant for private environmental behavior (b = .019, SE = .013, p = .171).
References
Alker, H. R. (1969). A typology of ecological fallacies. In M. Dogan & S. Rokkan
(Eds.), Quantitative ecological analysis in the social sciences (pp. 69-86).
Cambridge, MA: The MIT Press.
Brady, H. E., Verba, S., & 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, V., & Zaragoza, F. (2000). Bases sociodemogrficas y psicolgicas
de la conducta de reutilizacin: 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 envi-
ronmental group membership. Environmental Politics, 14, 441-459.
doi:10.1080/09644010500175783
Diamantopoulos, A., Schlegelmilch, B. B., Sinkovics, R., & 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, T., Stern, P. C., & Guagnano, G. A. (1998). Social structural and social psycho-
logical 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 afflu-
ence a prerequisite? Journal of Social Issues, 51(4), 121-137.
56 Environment and Behavior 49(1)
Hunter, L. M., Hatch, A., & Johnson, A. (2004). Cross-national gender variation in
environmental behaviors. Social Science Quarterly, 85, 677-694.
Inglehart, R. (1995). Public support for environmental protection: Objective problems
and subjective values in 43 societies. Political Science & Politics, 28, 57-72.
doi:10.2307/420583
Johnson, M., Brace, P., & Arceneaux, K. (2005). Public opinion and dynamic rep-
resentation in the American states: The case of environmental attitudes. Social
Science Quarterly, 86, 87-108. doi:10.1111/j.0038-4941.2005.00292.x
Kollmuss, A., & Agyeman, J. (2002). Mind the gap: Why do people act environmen-
tally and what are the barriers to pro-environmental behavior? Environmental
Education Research, 8, 239-260. doi:10.1080/13504620220145401
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.
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
Longhofer, W., & Schofer, E. (2010). National and global origins of envi-
ronmental association. American Sociological Review, 75, 505-533.
doi:10.1177/0003122410374084
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
Marquart-Pyatt, S. T. (2012b). Explaining environmental activism across countries.
Society & Natural Resources, 25, 683-699. doi:10.1080/08941920.2011.625073
Martinez-Alier, J. (2003). The environmentalism of the poor: A study of ecological
conflicts and valuation. Cheltenham, UK: Edward Elgar.
Milfont, T., & Fischer, R. (2010). Testing measurement invariance across groups:
Applications in cross-cultural research. International Journal of Psychological
Research, 3, 111-130.
Milfont, T. L. (2012). Cultural differences in environmental engagement. In S. D.
Clayton (Ed.), The Oxford handbook of environmental and conservation psychol-
ogy (pp. 181-202). New York, NY: Oxford University Press.
Milfont, T. L., Richter, I., Sibley, C. G., Wilson, M. S., & Fischer, R. (2013).
Environmental consequences of the desire to dominate and be superior. Personality
and Social Psychology Bulletin, 39, 1127-1138. doi:10.1177/0146167213490805
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
Nawrotzki, R. J. (2012). The politics of environmental concern a cross-national analy-
sis. Organization & Environment, 25, 286-307. doi:10.1177/1086026612456535
Olofsson, A., & hman, S. (2006). General beliefs and environmental con-
cern: Transatlantic comparisons. Environment and Behavior, 38, 768-790.
doi:10.1177/0013916506287388
58 Environment and Behavior 49(1)
Author Biographies
Ignacio Pisano, PhD, is an independent researcher recently interested in transper-
sonal 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 (http://environmentalpolicy.ucdavis.edu/). His research focuses on coopera-
tion and institutions in the context of environmental policy.