The limits of current climate policy approaches

Climate change mitigation policies are increasingly adapted across national, subnational or supranational borders through “policy transfer”, defined as the process of using lessons from existing policies and institutional frameworks in one setting to guide their development in another setting [12]. However, transferred policies can have limited effectiveness when applied to new contexts for a range of contextual and socio-demographic reasons [13,14]. There are various mechanisms of policy failure. They can be ineffective in delivering the expected outcomes, they can create unintended consequences, and they can have different distributional impacts when applied in a new setting.

These failures can be due to population heterogeneity, as different groups perceive and respond differently to similar interventions. Carbon tax transfer exemplifies one of these mechanisms, exerting different impacts across population groups, increasing the burden of lower-income households more than higher-income ones and thus differentially impacting societies with higher income inequality [15]. Similarly, the impact of decarbonisation pathways differs regionally, with rural areas experiencing more developmental constraints in some scenarios [16]. These variations in impact can result in public unrest and discontent, as demonstrated by the French Yellow Vest movement against the diesel tax [17]. These failures represent what Dolowitz and Marsh [18] termed “inappropriate transfer”, which happens when policies are adopted without sufficient consideration of economic, social, and political differences between contexts.

This frequent failure of context-blind policies underscores a gap in our understanding of how population groups differ across country contexts and how they respond to interventions [2,19,20]. Evidence-based policy approaches often extrapolate from a single context without accounting for the conditions that enabled effectiveness in the original setting [21,22]. In climate policy, social comparison-based Home Energy Reports –effective in the United States [23] – proved cost-ineffective in Germany and most industrialised countries due to lower treatment effects and baseline consumption [24]. Similarly, pricing interventions that achieved major emission reductions in developed economies were largely ineffective in developing economies [25]. Knowing what meaningful behavioural patterns exist across energy, transport, and food sectors, as well as what drives these patterns, is important for informing policy transfer.

From segmentation to multilevel analysis

Segmentation approaches, particularly persona development, offer a systematic way to understand population differences in behaviours. Persona development creates detailed behavioural profiles for different stereotypical user segments in a target population [26]. Unlike variable-centred approaches such as regression analyses that treat individuals as homogeneous and evaluate factor-outcome relationships across entire populations, persona development acknowledges that determinants tend to naturally cluster in real-world settings [27]. This methodology has proven valuable across sectors, especially in understanding and targeting consumers. For example, in a systematic review of sustainable food consumer segmentation studies, Verain and colleagues [28] identified three key segments, “greens”, “potential greens”, and “non-greens”, each of whom requires different intervention strategies for effective behaviour change. In environmental policymaking, personas are useful for identifying which population groups already engage in sustainable behaviours and which face barriers to change, enabling more efficient resource allocation and targeted policy design [29–31]. For example, Rasca and colleagues [32] used segmentation to develop personas of commuting behaviours based on socio-demographic characteristics and other travel-related determinants, such as car ownership and public transport costs. The resulting personas from this and similar studies can assist policymakers in envisioning how end-users might respond to sustainable transport policies [32].

Latent Class Analysis (LCA), a common approach for persona development, categorises individuals into mutually exclusive classes based on shared characteristics [33,34]. Unlike hierarchical clustering (such as in Pettifor et al. [29]), LCA accounts for measurement error by probabilistically assigning participants to each class, reducing misspecification risk in the assignment process and making it well suited to creating behavioural-data-driven personas [35,36]. However, traditional LCA has two limitations in the environmental context. First, it relies on individual characteristics – such as attitudes and socio-demographics – that can be weak predictors of actual behaviour: household-level factors like appliance ownership often explain more variance in electricity consumption [37]. Second, LCA assumes independence of the units of analysis, ignoring how individuals nested within regions and countries share correlated behaviours [38,39]. This is problematic because common infrastructure, policies, and cultural contexts can shape behaviour in energy, transport, and food [40,41], with political divides and regional disparities (e.g., in food access) sometimes outweighing individual attitudes [42,43].

Multilevel Latent Class Analysis (MLCA) addresses these limitations by identifying segments based on individual-level characteristics while mapping their geographic distribution [38]. This reveals where specific behavioural patterns concentrate, allowing for identification of regions with similar proportions of the different population segments. For example, regions with high or low concentrations of sustainable behaviour segments can be studied to understand what contextual elements, including policies, infrastructure, or cultural factors, support or constrain their concentration. This data-driven approach replaces ad-hoc policy development with systematic insights about where certain behavioural patterns emerge, enabling evidence-based policy design and cross-country learning about what works in different contexts. By combining the intuitive appeal of personas with systematic identification of regional variation, MLCA provides policymakers with both the who and where of behavioural patterns necessary for informing effective policy design and transfer.

Identifying who these segments consist of and where they live is a major step forward. However, for policy to be truly effective, we must also know what interventions these groups will support and whether their behaviours in one sector, like energy, are connected to their behaviours in another, like transport.

Cross-sectoral patterns and policy support

Environmental policy and research have traditionally treated energy, transport, and food as separate sectors, creating distinct policy measures and evaluation criteria for each area despite increasing calls for greater coordination [44,45]. This siloed approach ignores the interconnections between these sectors within household decision-making processes and the wider contexts, including infrastructure availability, social influences, and policy environments. Actions in one sector can create co-benefits or trade-offs in others, yet current approaches may lead to missed opportunities for coherent policy development or, in some cases, interventions that hinder progress across sectors [44,45]. To enhance cross-sectoral coordination, it becomes crucial to understand how behavioural patterns relate across different sectors, as this represents a central outcome targeted by many environmental policies.

The behavioural spillover literature offers insights into these cross-sectoral relationships, suggesting that engaging in one sustainable behaviour can influence the likelihood of engaging in others [46,47]. Spillovers can be positive, where one sustainable behaviour leads to others, or negative, where engagement in one area reduces motivation in another [46]. Positive spillovers can be explained by environmental identity, knowledge, and social norms, whilst negative spillovers are attributed to moral licensing or single-action bias [48]. However, ongoing debate persists about in what contexts and for what behaviours positive versus negative spillovers occur, with limited evidence across behaviours and sectors, and a lack of unifying theories [46,48,49]. While behavioural spillover effects remain largely ignored in policymaking, they have gained increasing attention recently [44] and can inform cross-sectoral policy initiatives that promote positive spillovers whilst mitigating negative ones. To address these gaps and better inform policy, research should examine complex interactions across sectors with high climate mitigation potential, particularly energy, transport, and food [48].

Understanding how behavioural patterns relate to policy support presents an equally critical challenge. Public support is among the most influential factors for policy adoption [50], yet the connection between sustainable behaviours and policy support remains debated. This is likely because public acceptability is shaped by many factors, including environment-related beliefs and existing behaviours, and how these vary across socio-demographic groups [41,51,52]. Experimental research has also identified a “crowding out” effect, where individuals reporting previous sustainable actions express reduced support for government policies [47,53], indicating that encouraging sustainable behaviours in one sector might unintentionally undermine support for the policies necessary for scaling. Understanding whether different population segments exhibit distinct behavioural patterns across sectors, and how these memberships influence policy support in various sectors, is therefore key to developing effective, transferable policies that account for both behavioural interconnections and the diverse regional contexts in which they unfold [54,55].

The current research

This exploratory study addressed the challenges to environmental policy transfer and coordination through the first application of MLCA across energy, transport, and food sectors, linking population segments of sustainable behaviours to policy support across 61 regions in nine countries, using the 2022 OECD Survey on Environmental Policies and Individual Behaviour Change (EPIC), a rich dataset of over 17,000 individuals [56]. In this study, “energy sector” refers to household and residential energy use, sometimes termed the building sector. We aimed to answer these research questions:

1. What are the population segments with distinct sustainable behavioural patterns based on individual-level self-reported indicators within the energy, transport, and food sectors?
2. What are the regional level differences in the distribution of these population segments?
3. To what extent do socio-demographic variables predict membership in these population segments?
4. To what extent do these population segments predict environmental policy support, and how does this relationship vary across regions?
5. What is the difference (if any) in population segment memberships across the energy, transport and food sectors?

Our approach went beyond traditional segmentation to systematically identify population segments with distinct behavioural patterns within the energy, transport, and food sectors. This segmentation allowed us to then link these population segments to their socio-demographic predictors, their cross-sectoral relationships, and their distinct levels of support for environmental policies. Finally, leveraging this multilevel framework as a proposed systematic, replicable diagnostic tool, we assessed the regional differences and similarities in the prevalence of these segments, using policy-transfer literature to illustrate how heterogeneity may affect transfer outcomes.

Data

The EPIC survey assessed environmental attitudes, behaviours, and policy support through self-reported measures across nine countries (United States, Canada, United Kingdom, Sweden, Switzerland, the Netherlands, France, Belgium, Israel) between 21 June and 20 July 2022. The dataset for analyses included 17,215 respondents collected through quota sampling. Respondents were randomly assigned to answer questions on two of four sectors (energy, transport, waste, and food). Further details on data collection are available here [56]. The OECD provided us access to the anonymised survey data on 21 June 2024; we had no access to information that could identify individual participants. Ethical approval was not required for this secondary data analysis.

Variable selection

Statistical analyses

Population segments of sustainable behaviours and their related policy support across sectors

Individual-level population segments for each sector were generated based on the sector-specific behavioural indicators shown in Fig 2, where the names of the resulting segments are explained. Fig 3 illustrates the proportion of individuals in each segment across sectors, as well as the likelihood of individuals possessing key socio-demographic characteristics (e.g., sex, income) within each segment relative to the least sustainable segment. Our approach to describing a population segment as the most or least sustainable is based on the likelihood of individuals engaging in sustainable behaviours in each sector, although we weighted certain behavioural indicators differently when classifying segments as more or less sustainable. For energy, less emphasis was given to the electricity bill indicator as it implies more contextual constraints than individual choices; for transport, more emphasis was given to the degree of car use, with regular household car use or primarily using automobiles considered less sustainable; for food, more emphasis was given to meat consumption over sustainable purchases, with eating less red and white meat considered more sustainable than buying seasonal or minimally packaged products. To consistently describe the likelihood of population segments performing each behaviour, we use the probability yardstick (0–100%) from the UK’s Defence Intelligence [71], divided into seven distinct numerical ranges. Specifically, 0–5% means “Remote chance”, 10–20% “Highly unlikely”, 25–35% “Unlikely”, 40–50% “Realistic possibility”, 55–75% “Likely or probable”, 80–90% “Highly likely”, and 95–100% “Almost certain”.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Description of indicators and definitions of the resulting individual-level population segments.

The “Indicator” column (left) defines the indicators used in the MLCAs in the energy, transport and food sectors (from top to bottom). The “Segments” column (right) describes the distinct features of the resulting segments in each sector. The electricity bill was originally measured in equivalent value ranges in local currencies in the OECD EPIC survey. “Primary mode” = most-frequent commute/leisure mode; “regular household car use” = household has a car available, which can coexist with public transport. The OECD converted these values to EUR in the dataset before sharing it with the authors. Icons used are from Font Awesome Free (https://fontawesome.com), licensed under CC BY 4.0.

https://doi.org/10.1371/journal.pclm.0001039.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Individual-level population segments and their socio-demographic covariates.

The line graphs (left column) show the segment-specific response probabilities for behavioural indicators based on a four-segment solution for energy, a five-segment solution for transport, and a four-segment solution for food. These solutions are based on the individual-level LCA output. Segment lines are shaded from darkest (least sustainable) to lightest (most sustainable), based on the likelihood of individuals engaging in sustainable behaviours in each sector. For energy (top row), the bar plot in the middle displays the response probabilities for the three-category nominal variable on electricity bills across the four energy segments (EUR 75 or less was treated as low bill, EUR 76 to over 200 was treated as a high bill). The forest plots (right) illustrate the impact of various socio-demographic characteristics on the relative likelihood of belonging to different segments compared to the least sustainable segment. Dots represent estimated odds ratios from multinomial logistic regressions, with error bars indicating 95% confidence intervals. Positive values (pink) indicate a higher likelihood of belonging to a given population segment compared to the reference (i.e., least sustainable) segment, while negative values (purple) indicate a lower likelihood. Non-significant odds ratios are in grey. The four age groups were dummy variables compared against the youngest participant group aged between 18 and 24. Income here referred to the after-tax monthly household income. The four income groups were also dummy variables compared against the lowest income group (below EUR 1450). Income in the OECD EPIC survey was originally measured in equivalent value ranges in local currencies. The OECD converted these values to EUR in the dataset before sharing it with the authors.

https://doi.org/10.1371/journal.pclm.0001039.g003

In the energy sector, four segments were identified among the 8,535 respondents randomly assigned to answer questions on this sector: ‘Energy investors’ (66%), ‘Disengaged energy users’ (12%), ‘Engaged energy users’ (11%), and ‘Energy conservers’ (11%). Only ‘Engaged energy users’ were almost certain (100%) to report having had a smart meter installed, while other segments were unlikely (23%) or highly unlikely (17%) to have done so. ‘Energy investors’ and ‘Energy conservers,’ comprising most of the sample, were likely or highly likely to report frequent energy-saving behaviours (60% and 81%, respectively) and efficiency installations (89% and 67%, respectively). However, the ‘Energy investors’ segment presented a mixed profile: despite high rates of efficiency installations (89%) and relatively active engagement in energy-saving behaviours (60%), they were also highly unlikely (17%) to report smart-meter uptake. Lastly, ‘Disengaged energy users’ were unlikely (33%) to report frequent energy-saving behaviours and had a remote chance (<1%) of reporting efficiency installation behaviours. In terms of socio-demographic characteristics, sex and income were generally not indicative of segment membership. The youngest age group (18–24) and those with children were more likely to be classified as ‘Disengaged energy users’ rather than ‘Engaged energy users’ or ‘Energy investors,’ while those with children were more likely to be ‘Energy conservers’ than ‘Disengaged energy users.’

In the transport sector, five segments were identified among the 8,695 respondents randomly assigned to this sector, and were named based on the primary transport mode used for commuting and leisure: ‘Auto travellers’ (61%), ‘Active & auto travellers’ (17%), ‘Active travellers’ (8%), ‘Public transport & long-trip travellers’ (8%) and ‘Public transport travellers’ (7%). ‘Auto travellers’ and ‘Active & auto travellers,’ comprising most of the sample, were almost certain (96% and 93%, respectively) to use cars regularly in their households. ‘Public transport & long-trip travellers’ were almost certain (100%) to primarily use public transport, while ‘Public transport travellers’ were highly likely (81%) to do so. ‘Active & auto travellers’ and ‘Active travellers’ were almost certain (100%) to use active transport. 86% of the sample belonged to segments that were at least highly likely (89%) to take annual long-distance high-emission trips. Regarding socio-demographics, individuals in the most sustainable transport segments (‘Public transport travellers’ and ‘Active travellers’) were more likely to be aged 25–34, in the lowest income group, live in (sub)urban areas, have children, hold higher education qualifications, and be employed. In contrast, ‘Auto travellers’ were more likely to be over 35, have a higher income, live in rural areas, and were less likely to have children, higher education, or employment. Finally, males were slightly more likely to be in segments that feature active transport than in the ‘Auto travellers’ segment.

In the food sector, four segments were identified among the 8,625 respondents randomly assigned to this sector. In the segment names, we used “sustainable consumers” to denote segments with low probabilities of eating (consuming) meat and “sustainable purchasers” for segments with high probabilities of purchasing in-season and minimally packaged products. ‘Non-sustainable consumers & purchasers’ (50%) and ‘Sustainable purchasers’ (24%), comprising most of the sample, were highly likely (81–84%) or almost certain (93–94%) to frequently eat meat. ‘Sustainable purchasers,’ however, were also highly likely (78%) to engage in sustainable purchasing behaviours (i.e., buying in-season and minimally packaged products). ‘Sustainable consumers’ (13%) and ‘Sustainable consumers & purchasers’ (12%), comprising smaller portions of the sample, were less likely to consume meat (22% and 7% for red meat, respectively), with ‘Sustainable consumers & purchasers’ likely (60–86%) to perform sustainable purchasing behaviours. Regarding socio-demographics, females were more likely to be ‘Sustainable consumers & purchasers’ than ‘Non-sustainable consumers & purchasers.’ Individuals aged 35–54 were more likely to belong to a frequent meat-eating segment than those aged 18–24. Income showed no clear pattern across food segments, except insofar as the highest income group tended to report more sustainable purchasing behaviours compared to the lowest income group. However, individuals living in a (sub)urban area and having children were more likely to be ‘Sustainable consumers & purchasers.’ Higher education was associated with a higher likelihood to eat less meat.

Linear regressions predicting policy support by segment membership (Fig 4) indicated that segments with more sustainable behaviours (e.g., ‘Engaged energy users,’ ‘Active travellers,’ ‘Sustainable consumers & purchasers’) generally showed stronger support for environmental policies within their respective sectors than those with a low likelihood of sustainable behaviours (e.g., ‘Disengaged energy users’, ‘Auto travellers,’ ‘Non-sustainable consumers & purchasers’). However, these differences were small (Cohen’s d < 0.2, where common thresholds for small, moderate and large effect sizes are 0.2, 0.5 and 0.8 respectively). This pattern held across most policy types, with a few exceptions, such as low-carbon car subsidies.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Policy support across population segments with distinct behavioural patterns and across sectors.

On the left, the plot lists all policies coloured by sector. Then along the top, moving from left to right, each sector has a tile plot that presents mean policy support ratings on a 5-point Likert scale from “1 = strongly against” to “5 = strongly support” for each reference segment. The reference segment is always the least sustainable segment, for example, ‘Disengaged energy users’ in energy. After the tile plots are created, forest plots display Cohen’s d effect sizes with 95% confidence intervals, showing differences in support between each population segment and the reference segment on that row across all sectors. For example, ‘Engaged energy users’ have significantly higher support for energy subsidies than the ‘Disengaged energy users’, but ‘Active travellers’ have no difference in support in comparison to ‘Auto travellers’, as denoted by the grey coloured circle. Effect sizes were based on linear regressions predicting policy support by segment membership with no control variables.

https://doi.org/10.1371/journal.pclm.0001039.g004

Regarding cross-sector policy support, individuals in more sustainable segments in one sector also tended to express greater support for environmental policies in other sectors, though the differences remained small. This was particularly evident in food segments, where individuals in more sustainable segments showed stronger support for energy and transport policies. In contrast, individuals in the relatively sustainable transport segments did not differ significantly from ‘Auto travellers’ in their support for food policies.

The cross-sector comparisons in Fig 5 indicated associations between segment membership across sectors. While belonging to the most sustainable segment in one sector did not always predict membership in the most sustainable segment of another, it was associated with membership in a segment with at least some sustainable behaviours in other sectors. For instance, ‘Sustainable consumers & purchasers’ were often ‘Energy conservers’ — not the most sustainable energy segment, but one with above-average sustainable behaviours. In contrast, belonging to the least sustainable segment in one sector strongly predicted membership in the least sustainable segment of another. For example, ‘Auto travellers’ were often ‘Non-sustainable consumers & purchasers’ and were also likely ‘Disengaged energy users.’ However, no significant association was found between membership in the least sustainable energy and transport segments.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Individual latent class membership associations across sectors.

Circle size and colour intensity represent the magnitude and direction of association: bright pink circles indicate positive associations, while dark purple circles indicate negative associations. Food segments (rows, top panel) show significant associations with energy segments (columns), as do transport segments (rows, bottom panel) with both energy segments (left columns) and food segments (right columns). The heat-map displays Pearson residuals from Chi-squared tests of independence (indicated by chi-squared values with degrees of freedom shown for each sector pairing). | ± 1.96|: significant at p = 0.05, | ± 2.58|: significant at p = 0.01.

https://doi.org/10.1371/journal.pclm.0001039.g005

Trends in behavioural patterns across country regions

Regional-level groups for each sector were derived from the country regions listed in S1 Table. MLCA grouped the country regions based on similarities in the distribution of individuals across the population segments, capturing distinct patterns within and between countries. Four regional groups were identified for energy, and three for transport and food, respectively, as shown in the left panel of Figs 6 (energy), Fig 7 (transport), and Fig 8 (food), which provides an overview of the distribution patterns of individual-level population segments across regional groups. In transport and food, most individuals in each region group belonged to the two segments with the least sustainable behavioural patterns (darker pie slices), although the exact distribution varied between region groups. In the energy sector, regional patterns were shaped primarily by the prevalence of Energy investors (slightly darker pie slices), a segment with a mixed behavioural profile, making direct comparison with transport and food less straightforward. The right panel lists the regions assigned to each region group and the country they belong to.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Region-level distribution of population segments and summary by country for the energy sector.

The left panel displays pie charts of the four region groups derived from the country regions listed in S1 Table. Regions are grouped based on similarities in the distribution of the population segments of sustainable energy behaviours in MLCA and labelled using Roman numerals (I, II, III, IV). The percentage of individuals in each population segment within each regional group is shown in the pie charts. The right panel lists the regions assigned to each region group and the country each region belongs to.

https://doi.org/10.1371/journal.pclm.0001039.g006

In the energy sector, the segment with the clearest unsustainable profile was ‘Disengaged energy users’; ‘Energy investors’, despite lower engagement in some practices, showed substantially higher rates of efficiency installations and energy-saving behaviours. Region Group I (e.g., most Belgian regions and New England) had the highest proportion of these less sustainable segments (96.7%), with 5.5% classified as ‘Disengaged energy users.’ Region Group II (most US and Canadian regions) had slightly fewer individuals in these segments (93.2%), but a higher proportion of ‘Disengaged energy users’ (17.9%). Region Group III had 73.3% in these segments, with the highest proportion of ‘Disengaged energy users’ (19.4%) across all groups. Region Group IV (e.g., all Israeli and Dutch regions, British Columbia) had the lowest proportion in less sustainable segments (55.7%), with 5.9% classified as ‘Disengaged energy users.’ This group also had the highest percentage (30.3%) of individuals in the most sustainable ‘Engaged energy users’ segment, compared to just 6.8% in Group II and none in Groups I and III.

In the transport sector, the two segments with least sustainable behaviours were ‘Active & auto travellers’ and ‘Auto travellers.’ Region Group I (e.g., all US and Canadian regions) had the highest proportion (88.8%) in these segments, with 75.4% classified as ‘Auto travellers.’ This group also had the lowest proportion (3.1%) in the most sustainable segment (‘Public transport travellers’). Region Group II (e.g., all Swiss regions) had 66% in the two least sustainable segments, with 50.9% classified as ‘Auto travellers.’ The remaining individuals were relatively evenly distributed (10%) across population segments with higher public and active transport use. Region Group III (e.g., all Dutch regions) had 73.1% in the two least sustainable segments, with 44.1% classified as ‘Auto travellers.’ Most remaining individuals were in the two most sustainable segments (7.3% ‘Public transport travellers’ & 14.2% ‘Active travellers’).

In the food sector, the two segments with least sustainable behaviours were ‘Sustainable purchasers’ and ‘Non-sustainable consumers & purchasers.’ Region Group I (e.g., all US and Canadian regions, most UK and Israeli regions) had the highest proportion (84.4%) in the least sustainable segments, with 57.6% classified as ‘Non-sustainable consumers & purchasers.’ Region Group II (e.g., all Swedish regions) had 66.3% in the least sustainable segments, with 50.1% classified as ‘Non-sustainable consumers & purchasers.’ Region Group III (e.g., all Swiss regions) had the lowest proportion (63.6%) in these segments, with just 34.8% classified as ‘Non-sustainable consumers & purchasers.’ This group had the highest share (27.9%) in the most sustainable segment (‘Sustainable consumers & purchasers’).

Overall, looking at regions within the countries (see the right panels of Figs 6–8), the country regions were typically grouped geographically, with most regions from one country ending up in the same region group in the three sectors. Particularly, within the Netherlands, the US, Canada, and Switzerland, most of their regions were grouped similarly across all three sectors in terms of behavioural-profile composition, with only a few exceptions. This suggests that country-level factors strongly influence individual consumption behaviours. However, some countries’ regions diverged from the most common regional group, highlighting within-country variation that may be relevant for policy differences. For instance, British Columbia belonged to a different energy region group than the rest of Canada, with a higher proportion of ‘Engaged energy users.’

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Region-level distribution of population segments and summary by country for the transport sector.

The left panel displays pie charts of the three region groups derived from the country regions listed in S1 Table. Regions are grouped based on similarities in the distribution of the population segments of sustainable transport behaviours in MLCA and labelled using Roman numerals (I, II, III). The percentage of individuals in each population segment within each regional group is shown in the pie charts. The right panel lists the regions assigned to each region group and the country each region belongs to.

https://doi.org/10.1371/journal.pclm.0001039.g007

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Region-level distribution of population segments and summary by country for the food sector.

The left panel displays pie charts of the three region groups derived from the country regions listed in S1 Table. Regions are grouped based on similarities in the distribution of the population segments of sustainable food behaviours in MLCA and labelled using Roman numerals (I, II, III). The percentage of individuals in each population segment within each regional group is shown in the pie charts. The right panel lists the regions assigned to each region group and the country each region belongs to.

https://doi.org/10.1371/journal.pclm.0001039.g008

Looking further across the countries, there were similarities in regional group profiles. For example, the US and Canada shared largely the same regional classifications across all sectors, with low proportions of individuals engaging in the most sustainable energy, transport, or food behaviours.

Heterogeneous effects of socio-demographics & the importance of regional context

Socio-demographics were inconsistently associated with segment membership across sectors (Fig 3), echoing growing evidence that demographic predictors of sustainable behaviour vary with individuals’ existing engagement and are unreliable when applied uniformly across diverse populations [72–74].

While previous studies identified the limited predictive power of socio-demographics in single sectors such as food [28] and across European climate-policy preferences, particularly among the policy-sensitive middle [75], our cross-sector analysis indicates the problem is systematic: sex and income showed little or unclear association with population segments across any sector, whilst age showed inconsistent associations that varied by sector and segment. Prior research showed that demographics identify who differs but not why, because behavioural differences stem from infrastructural, normative, and contextual factors [37,76,77]. Our evidence strengthens the case for climate policies that target actual behavioural drivers rather than demographic proxies [78].

Our regional findings highlight the importance of contextual factors, with within-country as well as between-country variation. At the within-country scale, British Columbia, for example, had higher proportions of ‘Engaged energy users’ and ‘Energy conservers,’ compared to other Canadian provinces. Importantly, regional similarities and differences can be attributed to differences in contextual factors by triangulating evidence from policy documents, such as more advanced smart meter infrastructure in British Columbia compared to other provinces [79], partially explaining the smart-meter uptake behaviour of ‘Engaged energy users.’ Notably, this segment also exhibits high rates of energy-saving behaviours and efficiency installations that cannot be explained by smart meter presence alone. A similar pattern is visible in Belgium, where Flanders’ 35-year bicycle policy has produced an 18.5% cycling modal share, substantially exceeding rates in Brussels and Wallonia [80]. These within-country regional differences reflect broader contextual factors including implementation approaches, complementary policies, and local cultures.

Between countries, these regional patterns also align with research showing the influence of material and social contexts on sustainable behaviours [37,81]: cycling infrastructure in the Netherlands enables higher cycling rates [82], and Belgium’s lagging smart-meter adoption despite EU-wide regulations illustrates how local implementation modifies policy effectiveness (e.g., lack of a target year period for wide scale roll-out) [83]. Policymakers therefore need systematic approaches to identify where specific behavioural patterns emerge and understand why certain contexts enable or constrain sustainable behaviours.

Implications for policy design and cross-sector coordination

We found that individuals in sustainable segments of one sector were more likely to behave sustainably in others, while those in unsustainable segments showed consistent patterns across sectors (see Fig 5). These patterns align with the concepts of positive and negative behavioural spillover patterns [46]. Association strength varied between sectors: ‘Auto travellers’ showed a strong likelihood of being ‘Non-sustainable food consumers & purchasers,’ while associations between energy and transport behaviours were weaker. These differences may be attributed to variations in behavioural similarity, as the behaviours analysed in the food and transport sectors were frequent, habitual behaviours (e.g., daily travel, weekly food consumption), whereas certain energy behaviours were more infrequent (e.g., installing appliances and smart meters) [84]. These patterns challenge traditional siloed policy approaches and highlight the need for coordinated strategies that account for how behaviours in one sector influence others [44].

The association between population segment membership and policy support across sectors also indicates a need for policy coordination. We found that sustainable behaviours were associated with greater policy support across all sectors, contradicting the “crowding-out” theory, which posits that engaging in sustainable behaviours reduces policy support through moral licensing [85]. However, the small effect sizes (d < 0.2) suggest that behavioural engagement and policy support likely share common underlying factors, such as environmental values [86] or structural contexts, rather than having strong causal relationships.

Intervention design should align with segment composition and regional context. Least sustainable segments (e.g., ‘Auto travellers’, ‘Disengaged energy users’) need behaviour change interventions that offer alternatives, incentives, or impose restrictions [29], and in highly auto-dependent countries like the United States and Canada, infrastructure transformation should precede individual behaviour change efforts [81]. Conversely, already sustainable segments (e.g., ‘Public transport travellers,’ ‘Engaged energy users’ and ‘Sustainable food consumers & purchasers’) need maintenance strategies that protect enabling infrastructure and prevent regression [87,88]. Mixed segments such as ‘Active & auto travellers’ need a combination of targeted maintenance of existing sustainable behaviours and adoption-barrier solutions. This dependence of intervention design on regional contexts argues for systematic regional assessment of segment distributions before selecting interventions [89].

A framework for evidence-based policy transfer

Current policy transfer relies heavily on informal networks and opportunistic timing rather than systematic frameworks [12,90,91]. MLCA provides a systematic alternative by quantifying population segment distributions across regions, enabling evidence-based identification of geographic clustering patterns. While we could not formally model policy and infrastructure variables due to regional sampling constraints, these patterns offer insights for evidence-based transfer.

Regions with high proportions of sustainable segments can serve as source regions for policy learning when their infrastructure and institutional contexts are comparable to those of target regions. For example, the lower auto-dependency in the Netherlands, Belgium, and some Swedish regions (44–51% versus 88.8% in the US and Canada) is supported by cycling infrastructure and urban design [82] – features that would require substantial adaptation if transferred to North American urban contexts.

Within-country variation also offers promising transfer opportunities, where some regions outperform their national context despite shared policies. The higher proportion of ‘Engaged energy users’ in British Columbia compared to other Canadian provinces has been associated with more advanced smart-meter rollout infrastructure [79] and a higher acceptance of smart grid technology [92]. Notably, British Columbia’s rollout was successful despite media coverage in the region, which often presented smart meters in a negative light, focusing on risks such as privacy [93]. The different patterns in Ontario and Quebec, despite similar infrastructure [79], suggest implementation strategies and local adaptation matter more than policy design alone. For instance, learning from British Columbia, policy communication could be improved by highlighting the benefits and using environmental framing to boost support [93].

As such, we propose a framework for evidence-based identification of policy transfer opportunities based on our MLCA approach (Fig 9). Specifically, Step 1 highlights the importance of collecting cross-country data on sustainable behaviours and their determinants to enable research on trends across contexts (e.g., survey datasets such as EPIC, but also more objective sources such as smart-meter or mobility data). Steps 2 and 3 involve applying statistical analyses to these data to identify population segments and regional groupings, and to assess policy support and need for intervention (e.g., distinguishing populations requiring behaviour change versus behaviour maintenance). Steps 4 and 5 are proposed extensions that go beyond the current research. Step 4 involves identifying clustered regional patterns, selecting source regions (those with high sustainable-segment proportions and documented policy success) and candidate target regions, and assessing contextual similarity for transferability. For instance, desk research on high-performing regions could reveal differences in policy and infrastructure that may explain their outcomes, as we did for British Columbia and smart-meter uptake. Building on Step 4, Step 5 requires adapting and piloting interventions, monitoring effectiveness over time, and prioritising within-country transfers where contextual similarity is greatest.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 9. Proposed five-step framework for evidence-based policy transfer.

Steps 1-3 are based on our data analysis processes and Steps 4-5 are proposed elements.

https://doi.org/10.1371/journal.pclm.0001039.g009

Although our analysis is limited to the nine OECD countries, this proposed five-step framework, from understanding population segment behaviours and their regional distribution to informing policy transfer strategies, can be applied more widely as large cross-sectional surveys like the OECD EPIC survey are regularly repeated and MLCA modelling becomes increasingly accessible through open-source software [94]. The proposed, unvalidated framework supplements traditional approaches to policy transfer with evidence-based targeting, though its effectiveness depends on data quality, cultural comparability, and institutional similarity, requiring future validation.

Limitations and future directions

This study has several limitations. First, population segments were identified based on self-report measures, which previous meta-analyses show do not accurately reflect actual sustainable behaviours [95]. While objective measures are impractical for large-scale surveys of this scope, future research should validate self-reported patterns against objective indicators where possible. Second, the cross-sectional design of the survey meant that we could not capture potential changes in behavioural patterns over time or confirm causal spillovers. Third, due to model constraints, we could not include regional-level predictors such as infrastructure availability or economic development in our MLCA models, limiting our ability to explain why certain regions cluster together. Fourth, several factors complicate interpretation of energy-sector findings specifically. Smart-meter uptake, our key behavioural indicator, depends substantially on utility infrastructure rollout rather than individual choice. Electricity bill levels are confounded by household size and dwelling characteristics, which we could not disentangle. Data were also collected in mid-2022, during the early months of the European energy crisis, which may have influenced reported bills and energy-saving behaviours. The energy sector’s lower entropy (0.672) and our most-likely class assignment also imply greater classification error and possible bias in its policy-support analyses. These factors mean that a strict least-to-most-sustainable ordering of energy segments is harder to justify than in transport and food. Finally, our sample of nine OECD countries limits generalisability, particularly to Global South contexts where infrastructure, institutional capacity, and behavioural determinants may differ substantially.

Future research can address these limitations through several avenues. Longitudinal studies employing repeated MLCA could track how segment membership evolves over time and test whether interventions in one sector genuinely cause changes in others [85,96]. Including regional-level predictors in multilevel models would help identify which infrastructure and policy variables drive the regional clustering patterns we observed. The proposed framework presented in Fig 9 should be validated through actual policy transfer experiments, testing whether interventions successful in source regions achieve similar outcomes when adapted to target regions with comparable contexts. Research should also expand beyond OECD countries to test whether similar behavioural patterns and policy transfer opportunities exist in different economic and cultural contexts. Such extensions would determine whether the segmentation approach and transfer framework generalise globally, or require fundamental adaptation for different developmental contexts.