Intentions Do Drive Green Behavior, But Context Decides How Strongly

People who say they care about the environment really do act on those intentions more often than not, according to a large new synthesis of decades of research, but the strength of that link depends heavily on where people live and work. A systematic review and meta-analysis published in Environmental and Sustainability Indicators pooled evidence from 52 quantitative studies to estimate how strongly pro-environmental intentions translate into actual behavior, and found a positive and statistically significant overall association. The work, led by Hammna Jillani and colleagues, challenges the popular narrative that green intentions routinely fail to materialize, while also showing that the intention-behavior relationship is far from uniform across settings.

The research team searched Scopus, Web of Science, and Google Scholar for studies published between January 2000 and February 2026, following PRISMA 2020 reporting guidelines. After screening 758 records and removing 484 duplicates, the full-text screening of 173 potentially eligible records yielded 145 studies for the systematic review. Of those, 52 provided sufficient statistical information—Spearman correlations, standardized regression coefficients, or other convertible estimates—to compute a common effect size on the Fisher’s Z scale. The included studies covered a wide range of behaviors, including energy conservation at home, recycling, sustainable consumption, food waste reduction, travel behavior, tourism, water conservation, climate-related action, workplace sustainability, and general pro-ecological behavior, in both institutional and non-institutional settings.

The headline result is a pooled Fisher’s Z of 0.39, with a 95 percent confidence interval of 0.31 to 0.46 and a test statistic of z = 9.98, p < 0.001. On the coding scheme the authors applied, that corresponds to a moderate association: stronger pro-environmental intentions are, on average, paired with stronger pro-environmental behavior. The authors are careful to note that this pooled correlation should not be read as a direct measure of the size of the intention-action gap or as a behavioral conversion rate. It synthesizes statistical associations between intentions and reported or observed behaviors, not discrepancies between what people intend and what they actually do.

What stands out most is the heterogeneity. The Q statistic reached 4070.68 with 51 degrees of freedom, and the I² value hit 99 percent, meaning nearly all of the observed variation in effect sizes reflects real differences among studies rather than sampling error. The 95 percent prediction interval stretched from −0.18 to 0.96 on the Fisher’s Z scale, an extraordinarily wide range that indicates a future comparable study could plausibly find anything from a slightly negative to a very strong positive association. The authors interpret this not as a failure of the meta-analytic estimate but as evidence that the intention-behavior relationship is structurally contingent, shaped by institutional, economic, and cultural circumstances rather than being a universal behavioral constant.

To probe that contingency, the team examined two contextual moderators. The first was institutional context: whether a study was conducted in a campus-based or other institutional environment. All 52 studies were coded for this moderator, with 29 institutional and 23 non-institutional studies, though the pooled subgroup analysis for institutional settings was restricted to 12 studies meeting predefined criteria. That subgroup showed a positive pooled association of Fisher’s Z = 0.39, with a 95 percent confidence interval of 0.08 to 0.70 and p = 0.02, alongside the same very high heterogeneity of 99 percent.

A mixed-effects meta-regression offered a more direct statistical test of whether institutional context explains between-study variation. Using non-institutional settings as the reference category, the institutional/campus coefficient was 0.186 (95 percent CI 0.035 to 0.337, p = 0.015), with an omnibus moderator test of QM = 6.02 on one degree of freedom, p = 0.013. The moderator accounted for roughly 16.4 percent of the between-study variance, though residual heterogeneity remained substantial at 95.8 percent. The authors stress that this is an exploratory study-level association, not evidence that institutional support causally strengthens the intention-behavior link. Institutional studies may differ systematically from non-institutional ones in behavioral domain, participant composition, country context, measurement instruments, and study design, and the meta-analysis did not directly measure infrastructure, organizational support, or policy strength.

The second moderator was country-income classification. Across 50 studies that could be assigned to a single income category—two were excluded because they spanned multiple countries without disaggregating results—positive associations emerged in all three groups. High-income economies showed a pooled Fisher’s Z of 0.45 (95 percent CI 0.34 to 0.57, k = 23), emerging economies 0.32 (95 percent CI 0.19 to 0.45, k = 22), and developing economies 0.52 (95 percent CI 0.13 to 0.91, k = 5). Yet the formal between-subgroup test was not statistically significant (Qb = 2.49, p = 0.29), meaning the hypothesis that country-income status systematically moderates the relationship was not supported. The authors frame these differences as exploratory descriptive patterns, cautioning that the number of studies in some categories is small and unevenly distributed.

Robustness checks reinforced the core finding. Leave-one-out sensitivity analysis showed no single study disproportionately influenced the pooled estimate, with confidence intervals remaining stable and the effect staying positive and significant in every case. Estimator-specific sensitivity analyses compared model-derived or converted effects (Fisher’s Z = 0.38, k = 42) with directly reported correlation-based effects (Fisher’s Z = 0.40, k = 10); the difference between the two was not statistically significant in an approximate test (z = −0.18, p = 0.85). Publication bias diagnostics, including funnel plot symmetry and a fail-safe N of 8000—well above the conventional threshold of 5k + 10—did not provide strong evidence of substantial small-study or publication bias, though the authors acknowledge such diagnostics cannot rule out selective reporting entirely. A GRADE-guided assessment rated the overall body of evidence as moderate certainty, citing the extreme heterogeneity as a consistency concern while noting that narrow confidence intervals support precision.

The theoretical upshot is a reframing of the intention-action gap itself. Rather than treating it as an individual-level motivational failure, the authors position it as a context-dependent phenomenon in which structural conditions influence the strength of the association. Drawing on institutional theory, they argue that universities and similar environments can function as behavioral infrastructures that lower implementation costs and increase behavioral visibility, thereby strengthening the link between what people intend and what they do. The meta-regression result is consistent with that proposition, but the authors repeatedly emphasize that mechanisms such as infrastructure, norms, and policy were not directly measured, so the finding should be read as evidence of contextual heterogeneity rather than a causal institutional-support mechanism.

For practitioners, the message is that strengthening intentions alone is unlikely to be enough. The authors suggest that effective sustainability strategies should shift from awareness campaigns toward structural enablement—improving infrastructure, reducing behavioral costs, and embedding sustainability into institutional routines. They also flag important limitations: most included studies used cross-sectional survey designs, so the synthesized effect sizes are indices of association that cannot establish temporal order or causality, and self-reported behavior scales mean common method bias cannot be fully excluded. No formal study-level risk-of-bias tool such as Cochrane ROB2 or the Newcastle-Ottawa Scale was applied, given the methodological diversity of the included designs. The wide prediction interval further warns against treating the pooled estimate as a universal effect. Future work, the authors suggest, should employ longitudinal, panel, and experimental designs, along with multi-level meta-regression, to identify the causal mechanisms underlying behavioral implementation and to test whether the descriptive country-income patterns hold up with larger and more balanced samples.

Subject of Research: Meta-analysis of the relationship between pro-environmental intentions and actual behavior across institutional contexts and country-income groups

Article Title: The pro-environmental intention-behaviour relationship in institutional contexts: A systematic review and meta-analysis across country-income groups

Article References: Jillani, H., Zahid, H., Králová, V., & Dvořák, J. (2026). The pro-environmental intention-behaviour relationship in institutional contexts: A systematic review and meta-analysis across country-income groups. Environmental and Sustainability Indicators, 32, Article 101589. https://doi.org/10.1016/j.indic.2026.101589

Image Credits: AI Generated

DOI: 10.1016/j.indic.2026.101589

Keywords: pro-environmental behavior, intention-behavior gap, meta-analysis, institutional context, country-income groups, systematic review, Theory of Planned Behaviour, heterogeneity, campus sustainability, behavioral implementation, Fisher's Z, structural moderators