Instruction

How much do you agree or disagree with each of these statements?

Items

Table 1

English Items of the Scale Environmental Efficacy Scale

The Environmental Efficacy scale (ISSP) was initially developed in English. Translations are available for the following languages: Afrikaans, Assamese, Bengali, Bikol, Cebuano, Chinese, Danish, German, Filipino, Finnish, French, Gujarati, Hiligaynon, Hindi, Hungarian, Iloko, Icelandic, Italian, Japanese, Kannada, Korean, Lithuanian, Malayalam, Marathi, Norwegian, Oriya, Polish, Punjabi, Russian, Slovak, Slovenian, Spanish, Swedish, Tamil, Telugu, Thai, Tswana, Xhosa, Zulu, Waray, Maguindanon. The respective questionnaires can be found on the ISSP website.

Response specifications

Participants responded on a five-point fully labelled Likert scale. The labels were 1 = Agree strongly, 2 = Agree, 3 = Neither agree nor disagree, 4 = Disagree, 5 = Disagree strongly. Moreover, participants could answer with -8 = Can’t choose, which we coded as missing value for our analyses.

Scoring

The scores of the items 2 and 7 need to be inverted, so that higher scores on the items correspond to higher environmental efficacy. After inverting, the environmental efficacy scale score can be calculated as the unweighted mean of the seven items. However, we recommend estimating the scale score only if valid responses are available for at least five of the seven items to have complete values for approximately 75% of the items (Mazza, Enders, & Ruehlman, 2015).

Application field

The environmental efficacy scale is part of the ISSP Environmental module (ISSP Research Group, 2023), which deals mainly with attitudes towards environmental issues. The module has been caried out four times, in 1993, 2000, 2010, and 2020. The environmental efficacy scale was part of each of these modules. However, the number of items increased from two to seven over the module rounds.

The scale is available in 43 languages and has been used in 28 countries and in different survey modes (PAPI, CAPI, CASI, CAWI, CATI, self-completion). However, the psychometric properties were acceptable in only 12 countries (Austria, Croatia, Denmark, Finland, France, Germany, Iceland, Japan, Lithuania, Sweden, Switzerland, and the USA). In these countries, the scale is suitable for research purposes, but not for the reliable assessment of the individual level of environmental efficacy as internal consistencies were too low for this purpose. Nevertheless, cross-cultural studies can use the economic (< 2 minutes according to authors’ estimation) environmental efficacy scale to assess environmental efficacy and compare its latent variances and covariances since partial metric invariance held across the 12 countries.

Originally introduced by Bandura (1977, 1982), perceived efficacy can be defined as a person’s belief about how well they can perform a given action and whether that action will have the intended effect. Applied to the environmental context, efficacy refers to whether people believe they can engage in pro-environmental behavior and whether this pro-environmental behavior has a positive impact on the environment.

The importance of environmental efficacy and other environmental constructs has rapidly increased with the growing impact of the climate change and the need for pro-environmental action (Steffen et al., 2015). However, although many people have positive environmental attitudes or are concerned about the impacts of the climate change, these attitudes and concerns do not necessarily translate into pro-environmental action (e.g., ElHaffar et al., 2020; Sheeran & Webb, 2016).

A theoretical framework that can explain this attitude-behavior gap is the theory of planned behavior (Ajzen, 1991). This theory posits that attitudes, norms, and perceived control (i.e., efficacy) influence behavior intention, which in turn influence the behavior. The theory of planned behavior also assumes a direct path from perceived control to behavior. This means that efficacy is an important construct to explain pro-environmental behavior as well as the attitude-behavior gap.

Several studies have empirically shown the relevance of environmental efficacy for pro-environmental behavior (Hamann & Reese, 2020; Miller, Rice, Gustafson, & Goldberg, 2022; Yuriev, Dahmen, Paillé, Boiral, & Guillaumie, 2020). Pro-environmental behavior has been operationalized in terms of political action, energy conservation, and recycling, among others. These studies also report a positive relationship between environmental efficacy and environmental attitudes. In addition, environmental efficacy seems to be related to trust in science, knowledge about climate change, connection to nature, and climate change risk perception but rather unrelated to socio-demographic variables (e.g., Bradley et al., 2020; Milfont, 2012).

In summary environmental efficacy refers to people’s beliefs about their ability to engage in pro-environmental behavior. The relevance of environmental efficacy is established not only on theoretical grounds, e.g., in the theory of planned behavior, but also empirically through its links to various environmental constructs, such as pro-environmental attitudes and behaviors.

Item generation and selection

The environmental efficacy scale was first used in the 1993 ISSP environmental module (ISSP Research Group, 1995). At that time, the scale included the first two items of the current scale (Item 1 and Item 2). The number of items was increased to the first five items in ISSP 2000 (ISSP Research Group, 2003). The sixth item was added in ISSP 2010 (ISSP Research Group, 2019) and the seventh item was added in ISSP 2020 (ISSP Research Group, 2023). All items were originally developed in English. Translations are available for 43 languages.

Samples

The data set of the ISSP 2020 environmental module includes data from 28 countries with a total of 44,100 participants (ISSP Research Group, 2023). The data were collected between February 2020 and May 2023 using simple random sampling or multistage sampling. The survey modes were PAPI, CAPI, CASI, CAWI, CATI, and self-completion. Of the 44,100 participants, we excluded 1,268 cases due to missing values for at least three of the seven items of the environmental efficacy scale. Of the remaining N = 42,832 participants 20,107 (46.9%) reported male gender and 22,662 (52.9%) reported female gender (0.15% missing). Regarding education, 1,205 (2.8%) participants reported no formal education, 2,852 (6.7%) reported primary education, 5,931 (13.9%) reported lower secondary education, 14,266 (33.3%) reported upper secondary, 1,317 (3.1%) reported post-secondary education, 4,316 (10.1%) reported short cycle tertiary education, 6,765 (15.8%) reported lower tertiary education, 4,779 (11.16%) reported upper tertiary education, and 653 (1.5%) reported a PhD as highest level of education (1.8% missing). Except for Finland (15+) and South Africa (16+) participants were 18 years or older, with a mean age of M =50 years (SD = 17.4; 0.9% missing). We report country-specific background variables in the Appendix Table A1.

Item analyses

All analyses were performed in R (4.4.1; R Core Team, 2024). We used the following packages for our analyses: psych (2.4.6.26; Revelle, 2022), lavaan (0.6-18; Rosseel, 2012), semTools (0.5-6; Jorgensen et al., 2022), purrr (1.0.2; Wickham & Henry, 2023), and metafor (4.6-0; Viechtbauer, 2010). We provide our code as Appendix.

As a first step, we tested the factorial structure of the environmental efficacy scale. We did this by running single group CFAs in each country, assuming a one-factorial structure for the seven items[1]. We used a robust maximum likelihood estimator (MLR), identified the model via standardized latent variables (latent variance equals unity, latent mean fixed to zero), and used full information maximum likelihood (FIML) to handle missing values. We report the model fit for each country in Table 2.

Table 2

Fit Statistics for Single-Group CFAs Assuming a one-Factorial Model

Examination of the model fits and the correlation residuals revealed that the one-factor model was unable to account for the covariance between Item 2 and Item 7 of the environmental efficacy scale. Since these items are the two inverted items, we allowed the residuals of these items to be correlated to improve the local model fit. Table 3 shows the model fit for each country. The one-factor with one residual covariance fitted the data better in most countries. However, the model fit was not acceptable in all countries. Referring to Hu and Bentler (1999) and the large sample sizes per country (usually exceeding 1,000), we excluded countries with CFI < .90, RSMEA > .10, or SRMR > .06. These rather liberal criteria ensured a low type I error, while we aimed to reduce type II error by further investigating the factorial structure. Nevertheless, following these model fit criteria resulted in the exclusion of 10 of the 28 countries. Australia, Taiwan, India, South Korea, New Zealand, Norway, the Philippines, Slovakia, Spain, and Thailand were excluded from the following analyses.

Table 3

Fit Statistics for Single-Group CFAs Assuming a one-Factorial Model With one Residual Covariance

Note. The residuals of item 2 and item 7 were allowed to covary.

Item parameters

We computed mean, standard deviation, skewness, excess, percentage of maximum possible (POMP) and percentage missing for each item in each country and across countries. Moreover, we report standardized factor loadings of the one-factor model with one residual covariance as item selectivity. While we report country specific statistics in the Appendix Table A2, we present overall statistics here. Overall, the item means varied between 3.18 and 3.47. Neither skewness (max. |skewness| = 1.44) nor excess (max. |skewness| = 1.57) showed worrisome levels for any item in any country (West, Finch, & Curran, 1995). Missing value rates were low. A higher missing percentage occurred only for the following countries and items: Denmark: Item 6 (6.55%), Item 7 (13.54%); India: Item 6 (8.12%); Japan: Item 7 (5.41%); Lithuania: Item 5 (7.15%); Russia: Item 6 (5.85%), Sweden: Item 7 (5.54%).

Table 4

Overall Descriptive Item Statistics

Note. Statistics were estimated across all 28 countries.

Selectivities (i.e., the correlation between the item and the latent environmental efficacy variable) were measured via standardized factor loadings of the CFAs and were low (< .20) for some items in some countries. Therefore, we also excluded countries with low selectivities. This resulted in the exclusion of six countries, namely China, Hungary, Italy, Russia, Slovenia, and South Africa. We report the overall selectivties based on a CFA assuming the one-factor model with one residual covariances using country as cluster in Figure 1. Note that we only included the 12 countries with sufficient model fit and selectivities in this analysis (N = 19,164). Selectivities for all countries can be found in the Appendix Table A3. For the remaining 12 countries, selectivities ranged from λ = .47–.72 (Lithuania and Japan) for Item 1, λ = .27–.52 (Denmark and Finland) for Item 2, λ = .47–.69 (Croatia and Sweden) for Item 3, λ = .49–.72 (Japan and Finland) for Item 4, λ = .47–.71 (Iceland and Austria) for Item 5, λ = .30–.65 (Finland / Sweden and Austria) for Item 6, and λ = .20–.43 (Denmark and the USA) for Item 7.

Figure 1

Measurement Model of Environmental Efficacy

Note. Standardized factor loadings of a CFA using factor as cluster are displayed ((13) = 201.877, CFI = .934, RMSEA = .074, SRMR = .035)

Objectivity

Data were collected either by trained interviewers or self-completion. For both, standardized formats and written instructions ensured the objectivity of the application. The written instructions, the ordered and labeled categories, the coding scheme for missing/ambiguous responses, the scoring procedure, and the country-specific statistics provided in the Appendix ensure the objectivity of the scoring and interpretation.

Reliability

We estimated Cronbach’s alpha as well as McDonald’s omega as indicators of internal consistency (Cronbach, 1951; McDonald, 1999). We did this only for the 12 countries, where the one-factor model with one residual covariance provided an acceptable fit. Values for omega varied between ω = .67–.78 and alpha varied between α = .67–.77 (Denmark and Austria). Table 4 shows the exact country-specific values.

Table 4

Country-Specific values for Cronbach’s Alpha and McDonald’s Omega

Validity

Content Validity

The definition of environmental efficacy has two parts (Bandura, 1977, 1982). The first part refers to whether people believe that they can engage in pro-environmental behavior. These beliefs are covered by items 1, 2, and 6. The second part refers to whether people believe that their behavior and the environment are mutually influencing each other. These beliefs are captured by items 3, 4, 5, and 7. Thus, the items of the environmental efficacy scale cover the content of the definition of environmental efficacy.

Factorial Validity

We tested factorial validity above (see Tables 2 and 3). For the remaining 12 countries, the model fit was acceptable and the standardized loadings sufficiently high. Therefore, we can assume factorial validity for these 12 countries.

Nomological network

We examined convergent, divergent, and criterion-related validity using manifest items and scale scores. We estimated product-moment correlations between the environmental efficacy scale and several other constructs for each of the remaining 12 countries (N = 19,164). In addition, we used meta-analytic techniques (random effects model with DerSimonian-Laird estimator; DerSimonian & Laird, 1986) to estimate the mean correlation across countries.

The validation constructs were age, gender, left-right scale, nationalism, institutional trust (trust in media, science, parliament, industry), environmental concerns, cause of climate change (natural vs. man-made), impact of climate change (for world and country), environmental attitudes, willingness to change (act pro-environment despite higher costs), consumption (recycling and reduce consumption), appreciation of nature, as well as political behavior in the form of signing a petition, donating money, and participating in a demonstration. We report the items, their scale, and their labels in the ISSP data set in the Appendix. Table 5 shows the country-specific correlations of the environmental-efficacy scale, as well as the mean effect and its 95%-confidence interval (CI). Overall, the environmental efficacy scale showed high associations with environmental knowledge, the attitudes, and behaviors, which is consistent with previous studies (Bradley et al., 2020; ElHaffar et al., 2020; Hamann & Reese, 2020; Milfont, 2012; Miller et al., 2022; Sheeran & Webb, 2016; Steffen et al., 2015; Yuriev et al., 2020). Furthermore, the lower correlations with socio-demographics also align with previous studies (Bradley et al., 2020; Milfont, 2012). Thus, the correlations found indicate good convergent, divergent, and criterion-related validity.

Table 5

Country-Specific and Mean Correlates of the Environmental Efficacy Scale

Descriptive statistics (scaling)

Table 5 presents country-specific descriptive statistics of the environmental efficacy scale scores for the remaining 12 countries.

Table 5

Country-specific Scale Statistics

Further quality criteria

We tested measurement invariance for the 12 countries, where the psychometric properties were acceptable, using multi-group CFA. We did this by successively testing three levels of measurement invariance (Leitgöb et al., 2023). The first level is configural invariance. This level is tested by estimating an unconstrained multi-group CFA. If the absolute model fit is acceptable, configural invariance can be accepted. This means that measurement models are the same for the countries examined. The second level is metric invariance. For metric invariance, factor loadings are constrained to be equal across countries. If the delta CFI is less than .01 compared to the configural model, metric measurement invariance can be accepted (Chen, 2007). This means that latent variances and covariances can be compared. The third level is scalar invariance. For scalar invariance, the intercepts are constrained to be equal in addition to the constrained loadings, but the latent means are allowed to vary across countries. If the delta CFI is smaller than .01 compared to the metric model, scalar measurement invariance can be accepted (Chen, 2007). This means that latent means can be compared across countries.

The fit of the configural model was acceptable (CFI = .932, RMSEA = .077; SRMR = .036). However, the constraints of the metric model led to a substantial decrease in fit (ΔCFI = .019). We therefore tested for partial metric invariance by successively freeing loadings as suggested by the modification indices. According to Pokropek, Davidov, & Schmidt (2019) having at least 20% of invariant items is sufficient to receive unbiased latent (co-)variances. This criterion was met since item 4 and 7 (29% of all items) were invariant across all countries. We needed to free the loadings of the other five items in four countries (Item 1 and Item 2 in Japan, Item 3 and Item 5 in Croatia as well as Item 6 in Finland and Sweden) to achieve an acceptable decrease in fit of the metric model (ΔCFI = .010). We, thus, can compare latent (co-)variances and tested for scalar measurement invariance in a next step. A comparison of the fit of the scalar model with the partial metric model revealed a substantial decreased in fit (ΔCFI = .265). Because of the large decrease, we decided not to test for partial scalar measurement invariance. This means that neither latent means, nor manifest (co-)variances and means can be compared across countries.

In summary, both configural and metric measurement invariance could be accepted. This means that measurement models as well as latent variances and covariances of the environmental efficacy scale can be compared across Austria, Croatia, Denmark, Finland, France, Germany, Iceland, Japan, Lithuania, Sweden, Switzerland, and the USA.

Table 6

Fit Indices for Measurement Invariance Across Countries

Note. N = 19,164.

Acknowledgement

We would like to thank Piotr Koc for his feedback on this documentation.