Attitudes to climate change risk: classification of and transitions in the UK population between 2012 and 2020

The UK aims to reach net-zero emissions by 2050, a goal that requires not only policy changes but also shifts in individual behaviours, which are influenced by public beliefs about climate risk and urgency. Prior UK surveys show many people acknowledge climate change but skepticism and uncertainty persist. Single-item measures may oversimplify risk attitudes, and cross-sectional segmentation leaves questions about change over time. This study investigates how attitudes to climate change risk (ACCRs) are organised within the UK population, how socio-demographic factors relate to these segments, and how both segment structures and memberships changed between 2012–2014 (wave 4) and 2018–2020 (wave 10). Research questions: (1) What ACCR groups exist in the UK? (2) How do socio-demographics relate to ACCR group membership? (3) How do group structure and membership change over time?

The paper conceptualises ACCRs as comprising cognitive elements (beliefs about likelihood, timing, scope, and control of climate change) and affective responses (worry and concern), drawing on van der Linden (2017). Temporal and spatial perceptions shape whether risks are seen as personally relevant, with potential optimism biases. Prior segmentation studies using LCA or clustering have identified multiple audience segments (e.g., Six Americas; national typologies in the UK, New Zealand, Australia, and Europe), but often relied on cross-sectional data or assumed stable class structures over time. Evidence on socio-demographic correlates is mixed: gender, age, income, education, and political affiliation variably predict climate beliefs and concern, with political orientation often the strongest predictor and interactions with education sometimes reported. The review highlights the need for data-driven segmentation of climate risk attitudes specifically, using large-scale probability samples, and to track transitions without assuming fixed class solutions.

Data: UK Household Longitudinal Study (UKHLS), a national annual panel begun in 2009. Waves analyzed: 4 (2012–2014) and 10 (2018–2020). Adults aged 16+ completed interviewer-administered questionnaires (face-to-face/telephone). After listwise deletion on ACCR items, valid Ns were 38,037 (wave 4; valid rate 80.81%) and 31,498 (wave 10; valid rate 91.78%). Missingness diagnostics suggested MAR. Measures (ACCRs): Five items capturing climate risk attitudes, answered on 5-point Likert scales (1=Strongly disagree to 5=Strongly agree) or dichotomous (0/1) as provided by UKHLS: - Beyond control: “Climate change is beyond control, it's too late to do anything about it.” - Too far in future: “The effects of climate change are too far in the future to really worry me.” - Affected within 30 years (dichotomous): “People in the UK will be affected by climate change in the next 30 years.” - Major disaster: “If things continue on their current course, we will soon experience a major environmental disaster.” - Crisis exaggerated: “The so-called 'environmental crisis' facing humanity has been greatly exaggerated.” Socio-demographics: Sex (male=1, female=0), age (continuous), gross monthly personal income in £1000s (can be negative due to reported losses), education (ordinal: 1=No qualification to 5=Degree), political affiliation (party closest to respondent). Political affiliation recoded into Right-wing (reference), Left-wing, Other, based on Chapel Hill Expert Survey party placements; party lists differ across waves reflecting UK party system changes. Age and income were mean-centred for regression. Analysis: K-means clustering on the five ACCR items per wave. Optimal k determined using three criteria (Makles, 2012): elbow in within-cluster sum of squares (WSS) and log(WSS), η² (WSS/TSS), and proportional reduction of error (PRE) comparing k and k−1 solutions. Stability checked by repeating k-means 50 times with random starts. ANOVAs tested mean differences across clusters for each item. Participants were assigned to nearest centroid cluster. Multinomial logistic regression predicted cluster membership from socio-demographics (education modelled as categorical improved AIC over continuous). Independence of Irrelevant Alternatives checked via Hausman tests; sensitivity analyses with survey weights showed similar results.

- Optimal clusters: k=3 in both waves. Moving from k=2 to k=3 reduced WSS substantially (η² indicated ~44% reduction; PRE ~24% in wave 4; η² ~50%, PRE ~26% in wave 10). Solutions were stable across 50 random starts (k=3 selected 49/50 in wave 4; 47/50 in wave 10). ANOVAs showed significant between-cluster differences for all items in both waves (e.g., Major disaster_w4 F(2,38034)=5056.80, p<0.001; Crisis exaggerated_w10 F(2,31497)=13973.79, p<0.001). - Cluster profiles (consistent across waves): - Sceptical: downplay risk—more agreement with negative statements (exaggerated, too distant, beyond control), lowest agreement that the UK will be affected within 30 years. - Concerned: lowest agreement with negative statements; highest agreement with positive statements (near-term impacts; major environmental disaster if current course continues); support urgency. - Paradoxical: ambivalent/mixed—agree people will be affected within 30 years yet also endorse “too far in the future” and “beyond control,” implying worry but low perceived efficacy; moderate disagreement about imminent major disasters. - Cluster sizes: - Wave 4 (2012–2014): Sceptical 21.3%; Concerned 35.4%; Paradoxical 43.2% (largest). - Wave 10 (2018–2020): Sceptical 15.6%; Concerned 43.8% (largest); Paradoxical 40.7%. - Transitions (wave 4 → wave 10): - Concerned retained 71.6% within-cluster; 23.8% moved to Paradoxical. - Paradoxical: over half shifted to Concerned (≈53.6% remained; 23.8% of Concerned moved to Paradoxical; table indicates large flows Paradoxical→Concerned). - Sceptical: ≈20% moved to Concerned; 43.2% moved to Paradoxical; 37.1% remained Sceptical. Overall trend: net movement away from Sceptical/Paradoxical toward Concerned, with the 3-cluster structure stable. - Socio-demographic predictors (multinomial logit): - Sex: Males ≈50% more likely than females to be Sceptical vs Concerned; 11–16.8% more likely to be Paradoxical vs Concerned (both waves, p<0.001). - Age: Wave 4—older more likely Paradoxical vs Concerned (~5% per decade); no significant Sceptical vs Concerned difference. Wave 10—younger more likely Sceptical vs Concerned (~5% per decade); Paradoxical vs Concerned not significant. - Income: Higher income associated with Concerned vs Sceptical (≈6.5% per £1000 in wave 4; ≈8% in wave 10) and Concerned vs Paradoxical (≈3.3% and 4.7% per £1000, respectively). - Education: Strongest predictor—higher qualifications markedly increased odds of being Concerned vs Sceptical/Paradoxical; largest jump between No qualification and GCSE; those with a degree were at least 70–79.5% more likely than those with no qualifications to be Concerned vs others. - Political affiliation: Left-wingers substantially more likely than right-wingers to be Concerned vs others (≥36.7% in wave 4; ≥58.7% in wave 10). “Other parties” aligned more with right-wingers in wave 4 and with left-wingers in wave 10 regarding cluster membership. - Education and political orientation effects were independent (no interaction detected).

The study answers the three research questions by showing: (1) UK ACCRs coalesce into three robust segments—Sceptical, Concerned, and Paradoxical—defined by coherent cognitive and affective risk patterns; (2) socio-demographic factors relate meaningfully to segment membership, with education and political affiliation the strongest correlates, and additional roles for sex, age, and income; (3) while the 3-cluster structure is temporally stable, individual memberships are fluid, with a net shift toward the Concerned segment from 2012–2014 to 2018–2020. These findings suggest substantial potential for targeted communication and policy interventions. The sizable and least-stable Paradoxical group appears particularly amenable to movement toward Concerned with messages that bolster efficacy and address ambivalence. For Sceptical individuals, clearer risk quantification and salient evidence of impacts and effective mitigation may move attitudes toward the middle. Segment-tailored messaging strategies, disseminated via trusted and demographically relevant media channels, could increase engagement and support for climate action. The fluidity of individual transitions indicates that public opinion is more dynamic than public discourse may imply, opening opportunities for effective, evidence-based communication campaigns.

Using UKHLS panel data and data-driven k-means clustering, the study identifies three stable UK attitudes to climate change risk segments—Sceptical, Concerned, and Paradoxical—and documents a net transition toward Concerned between 2012–2014 and 2018–2020. Education and political affiliation are the strongest correlates of cluster membership, with additional effects of sex, age, and income. The Paradoxical group remains large, indicating a priority target for communications and policy interventions to strengthen efficacy and reduce ambivalence. Future research should examine the characteristics of individuals who transition in different directions, incorporate additional constructs (knowledge, policy preferences, behaviours), explore media use and trust across segments, and compare typologies across countries.

- Use of secondary survey data limits construct scope and granularity; the climate module was brief within a broad survey. - Two items referenced “environmental disaster/crisis” rather than climate change explicitly; interpretations may vary (e.g., pollution vs climate), though climate is likely salient. - Focus solely on risk attitude items excludes knowledge, policy preferences, and behaviours; this avoids confounding but oversimplifies multi-faceted attitudes and may limit predictive power for behaviours. - Clustering involves subjective choices (variable selection, algorithm, starting centroids); despite robustness checks, results depend on these decisions. - Common method and interview-mode biases possible (face-to-face interviewing, interviewer effects, consistency motifs). - Structural age effects may confound education/income; age-period effects cannot be fully disentangled with two waves of the same cohort. - Generalisability limited to the UK context; structures may differ in other countries, particularly developing contexts.