The structure of nuclear energy beliefs in Norway: a psychological network analysis

1 Introduction

Achieving deep decarbonization will require both rapid expansion of low-carbon technologies and difficult societal choices about which technologies to adopt, where to site them, and how to govern their development. Public acceptance has become a key challenge for several low-carbon technologies, including onshore wind energy and electricity-transmission infrastructure, where local opposition and unresolved value conflicts have delayed, modified, or halted projects in many countries. Nuclear power is a particularly contested case: it is increasingly framed as a potentially important component of low-emission energy systems, yet acceptance varies markedly across publics and over time. Understanding how citizens evaluate nuclear power—and how those evaluations are organized—is therefore central to research on the social dimensions of the energy transition.

Nuclear power is increasingly discussed as a potential component of low-emission energy systems, given its low direct greenhouse-gas emissions and capacity to provide dispatchable electricity (European Commission, 2021; International Atomic Energy Agency, 2020). It has also been proposed as a contributor to energy security and economic development in some contexts (European Commission, 2021; Nøland et al., 2023). However, public acceptance remains contested, reflecting persistent concerns about accidents, waste management, and broader environmental and security risks (Bohdanowicz, 2023; Brugidou and Bouillet, 2023; Kim et al., 2014). How citizens form and organize attitudes toward nuclear energy is therefore directly relevant to energy-transition governance, including siting decisions, public engagement, and policy design for contested technologies.

Public opinion is not merely a backdrop to energy policy; it can shape policy development and political decision-making. A substantial body of research on policy responsiveness and dynamic representation suggests that sustained shifts in public preferences are, with varying lags and fidelity, reflected in policy outcomes (Burstein, 2003; Canes-Wrone, 2015; Page and Shapiro, 1983; Rasmussen et al., 2019; Soroka and Wlezien, 2010; Stimson et al., 1995). Because governments anticipate electoral consequences, public opinion may influence whether policymakers introduce, reframe, delay, dilute, or abandon policy proposals, while durable public consent contributes to the perceived legitimacy of decisions whose consequences span decades. These dynamics are especially visible in the energy sector. Energy infrastructure projects require siting decisions, long investment horizons, and sustained political commitment, making public acceptance an important factor in shaping what can be built, where it can be built, and under what conditions (Devine-Wright, 2009; Koecklin et al., 2021; Wüstenhagen et al., 2007).

Nuclear energy offers some of the clearest demonstrations. Following the Fukushima accident, public acceptance declined significantly across the 42-country sample studied by (Kim et al. 2013). In Germany, public opposition and political pressure following Fukushima contributed to the government's accelerated nuclear phase-out (Schreurs, 2013). In Japan, growing public concern, regulatory reforms, and local political constraints contributed to the prolonged suspension and gradual restart of reactors (Vivoda and Graetz, 2015). Whether and how a country enters—or re-enters—nuclear power is thus shaped not only by techno-economic assessments but also by political and societal factors, including public opinion, which makes an accurate, structured description of that opinion a policy-relevant scientific task.

For policy, however, the aggregate level of support is only part of the picture. Whether public opinion constrains or enables policy change depends also on its character: strong, crystallized attitudes are more stable over time, more resistant to persuasion, and more likely to be acted upon than weak or ambivalent ones (Howe and Krosnick, 2017). How the beliefs underlying an attitude are organized—how densely they are interconnected, and which considerations occupy structurally central positions—is one indicator of this character (Dalege et al., 2018, 2016). Two publics with identical headline support can therefore differ considerably in how settled that support is, and in how it may respond to new information, focusing events, or campaigns.

Research on public attitudes toward nuclear energy has identified several robust predictors of acceptance. Perceived benefits and risks consistently predict support, with perceived benefits often carrying greater weight than perceived risks in shaping overall evaluations (Bohdanowicz et al., 2023; de Groot et al., 2013; Wang and Kim, 2018). Gender differences are well established, with men typically expressing more favorable attitudes than women (Choi et al., 2000; Greenberg and Truelove, 2010). Trust in institutions (e.g., government and the nuclear industry) matters because many citizens rely on trusted sources when evaluating complex technologies (Visschers and Siegrist, 2013). Political ideology and environmental values also shape nuclear attitudes, with conservatives often more supportive and individuals with strong environmentalist orientations often more skeptical (Bohdanowicz et al., 2023; McBeth et al., 2023; Smolinski et al., 2024). While these predictors have been extensively studied, less is known about how specific beliefs about nuclear energy are structured and interconnected within individuals' broader belief systems. This question matters because, according to the Causal Attitude Network framework (Dalege et al., 2016), belief systems with different structures may respond differently to new information, framing, and deliberation: tightly interconnected belief systems may be more mutually reinforcing and potentially more resistant to change. Aggregate measures of support or opposition can mask this organization, leaving open which evaluative frameworks underpin a given level of public acceptance.

Norway provides an informative context for examining these questions. Norway's ambition to become a low-emission society by 2050 implies substantial growth in electricity demand across transportation, industry, and heating (Ministry of Climate and Environment, 2023; Statnett, 2019). Hydropower currently dominates the electricity supply, accounting for 87% of national electricity production (The Norwegian Water Resources and Energy Directorate, 2024). While plans exist to expand wind and solar energy, these initiatives have faced considerable public resistance (Energifakta Norge, 2024; Kraft and Kraft, 2025). Norway does not operate any nuclear power plants, and government policy has historically opposed incorporating nuclear energy into national energy strategy (Blaker, 2023; Norges offentlige utredninger, 2023). The political landscape remains mixed, with some parties advocating nuclear adoption while others are opposed or cautiously open (Holst, 2023), and prominent industry and environmental NGO voices have often been negative (Heggdal, 2023). Survey evidence suggests that around 60% of Norwegians express support for nuclear energy, yet a substantial proportion remains undecided, indicating that attitudes may be incompletely crystallized for many people (NTB, 2023; Tvinnereim and Liu, 2024). In such a context, examining not only what people believe, but how beliefs relate to one another, may clarify how opinions are organized.

Taken together, these features make Norway a distinctive case rather than simply a country without nuclear power: a hydropower-dominated, fossil-fuel-exporting, high-trust welfare state with substantial administrative capacity, in which nuclear power is debated as a prospective addition to an already largely renewable system rather than as a replacement for fossil generation. Each feature plausibly shapes nuclear-energy beliefs. Hydropower abundance lowers the perceived necessity of nuclear power even as electrification raises the salience of supply security and electricity prices; the oil-and-gas export economy frames nuclear power partly as a matter of industrial strategy and climate responsibility; high institutional trust may mute the distrust-driven risk perceptions documented in other national settings (Visschers and Siegrist, 2013); and the weak partisan structuring of the issue (Holst, 2023) limits the ideological cues that organize nuclear attitudes in more polarized contexts. Recent Norwegian conflicts surrounding onshore wind-power development illustrate how public beliefs, local acceptance, and perceived environmental impacts can become central issues in energy-infrastructure governance (Aas et al., 2014; Kraft and Kraft, 2025).

Traditional approaches to nuclear energy attitude research often examine beliefs in isolation or model their independent associations with overall attitudes (Badora et al., 2021; Bisconti, 2018; Gupta et al., 2019; Kim et al., 2014; Visschers et al., 2011). Relatedly, research commonly relies on structural equation models or dimensionality-reduction techniques such as factor analysis (Hu et al., 2021; Pampel, 2011; van der Pligt et al., 1982). While valuable, these approaches may miss how beliefs cohere into configurations that mutually reinforce one another. A systems perspective—treating attitudes as networks of interrelated beliefs—offers a complementary way to describe how support and opposition are structured (Dalege et al., 2016). From an energy-transition perspective, identifying clusters of beliefs and structurally prominent beliefs can help characterize the evaluative “frameworks” through which a contested technology is understood, which may inform research on public contestation and engagement.

1.1 Attitudes toward nuclear energy

In this study, we treat attitude toward nuclear energy as an overall evaluation of nuclear power as an option for electricity generation (Maio et al., 2019), and beliefs as cognitive representations held to be true that contribute to that evaluation (Ajzen and Fishbein, 1980). Prior work indicates that supportive attitudes are often grounded in beliefs about climate mitigation, energy security, and economic benefits, whereas opposition is frequently linked to beliefs about accidents, waste management, and security threats such as warfare or terrorism (Bohdanowicz et al., 2023; de Groot et al., 2013; Hu et al., 2021; Kim et al., 2014; Smolinski et al., 2024; Wang and Kim, 2018). The key question, therefore, is not only which beliefs correlate with support or opposition, but how these beliefs relate to one another as part of an interconnected belief system.

1.2 Examining attitudes through network analysis

Psychological network analysis is a methodological approach in which psychological constructs—including symptoms, emotions, and attitudes—are represented as systems of interrelated components rather than as reflections of an underlying latent variable (Borsboom, 2017; Borsboom and Cramer, 2013). In this framework, individual components (e.g., specific beliefs) are modeled as nodes, and the statistical associations between them are modeled as edges. The resulting network can be characterized in terms of its overall connectivity (how densely beliefs are associated) and in terms of the structural prominence of individual nodes (centrality), and it can be partitioned into communities of more strongly interconnected elements (Epskamp et al., 2018). Applied to attitudes, this approach complements traditional methods such as structural equation modeling and factor analysis, which typically summarize observed items as indicators of one or more latent dimensions. By contrast, network analysis treats the items themselves as the constituents of the attitude and asks how they are organized relative to one another. This is informative when the research question concerns the structure of a belief system rather than the size of its underlying dimensions.

The Causal Attitude Network (CAN) model conceptualizes attitudes as networks of interconnected attitude elements such as beliefs (Dalege et al., 2016). In psychological network analysis, beliefs can be represented as nodes connected by edges that reflect statistical associations, allowing researchers to characterize (i) how cohesive a belief system is (connectivity) and (ii) which beliefs are structurally prominent (centrality) (Bringmann et al., 2019; Dalege et al., 2016). Because cross-sectional data cannot establish causal relationships among beliefs, we interpret network structure descriptively—as a map of how beliefs are interconnected—rather than as evidence of causal influence. Recent research also suggests caution in interpreting centrality as a guide to persuasion effectiveness; for example, (Brandt and Vallabha 2023) found that inter-attitude centrality did not predict persuasion outcomes, reinforcing the need to interpret centrality primarily as structural prominence.

To our knowledge, only one study has applied psychological network analysis to nuclear power attitudes. (Smolinski et al. 2024) examined nuclear acceptance in Poland and found political ideology to occupy a central position in their network. The present study extends this emerging literature by examining the organization of perceived advantages and disadvantages of nuclear energy in Norway, a national context without nuclear power in the current energy mix.

From this perspective, connectivity and centrality are not merely descriptive statistics. Within the CAN framework, global connectivity has been linked theoretically to attitude stability and resistance to change (Dalege et al., 2018, 2016), and structurally central beliefs index the considerations around which the wider belief system is organized. Mapping these properties—in the full sample and within attitude groups—therefore speaks to when, and for whom, public opinion on nuclear energy is likely to become consequential in future policy debates.

Accordingly, we address the following research questions:

• RQ1: How are attitudes and perceived advantages and disadvantages of nuclear power distributed in the Norwegian public, and how do they vary across demographic groups?
• RQ2: How are nuclear power beliefs organized as a network—specifically, which beliefs are structurally prominent and how do beliefs cluster into communities?
• RQ3: Do belief networks differ across individuals with positive, undecided, and negative attitudes, particularly in connectivity and the structural prominence of specific beliefs?

We pursue three corresponding objectives. First, we describe the distribution of attitudes and beliefs about nuclear power in Norway, including demographic differences. Second, we estimate the belief network in the full sample to identify central beliefs and belief communities. Third, we compare belief networks across attitude groups—positive, undecided, and negative—focusing on global connectivity and the structural prominence of specific beliefs. By mapping the relational structure of nuclear energy beliefs, this study contributes to understanding how public attitudes toward a contested low-carbon technology are organized in a national context where the technology is not yet established.

2 Methods

2.1 Participants and procedure

In February 2024, we collected data via the Norwegian Kantar Gallup Panel, an online access panel administered by Kantar/Verian (formerly Kantar Public; Norsk Gallup), which comprises approximately 40,000 individuals aged 15 years and older who have consented to participate in survey research. The panel is recruited primarily through nationally representative computer-assisted telephone interviews (CATI) and is designed to approximate the demographic composition of the Norwegian adult population. Panel administration is certified under ISO 26362:2009, the international standard for access panels in market, opinion, and social research. We note that, although the panel is designed to support representative survey research, it is an access panel rather than a probability sample drawn directly for the present study; estimates may therefore be affected by panel recruitment, non-response, and participation-related selection (see also Supplementary material, Section S2, for sample composition and benchmarking against population statistics).

For the present study, we invited 3,678 panel members aged 18 years and older and stopped data collection after obtaining 1,017 completed responses (response rate = 27.7%). The sample size is consistent with methodological recommendations for psychological network analysis, providing sufficient statistical power and stability for estimating network structures and centrality metrics (Epskamp et al., 2018).

The survey instrument was developed on the basis of prior research on nuclear-energy attitudes, supplemented with themes observed in Norwegian public discourse, including social-media discussions. A preliminary version of the questionnaire was reviewed by 40 individuals in Norway—researchers, politicians, public officials, and representatives from environmental NGOs—and revised based on their feedback. The items analyzed in this article are provided in the Supplementary material (Section S1); the full survey is available on OSF (https://osf.io/s7t2n/).

2.2 Ethics

All respondents are members of the Norwegian Kantar Gallup Panel and provide informed consent as part of panel enrollment and participation in panel surveys. The present study analyzed de-identified (anonymized) survey responses, and no directly identifying information was available to the research team.

2.3 Sample characteristics and benchmarking

Because the sample was drawn from a non-probability online panel, traditional sampling error estimates do not capture all potential sources of survey error. For descriptive benchmarking purposes, under simple random sampling assumptions, the maximum 95% margin of error for a proportion with N = 1,017 is approximately ±3.0 percentage points; however, this should be interpreted cautiously given the panel design. We therefore benchmarked the sample demographics against population statistics from Statistics Norway (Supplementary Table S1). Gender distribution in the sample (50.8% female) was close to the population benchmark (49.7%). The median age in the sample (53 years) exceeded the population median (39.8 years), consistent with restricting the survey to adults. Respondents with higher education were overrepresented (29.8% vs. 12.5% in the benchmark). Additional benchmarking analyses, including category-level comparisons and an assessment of the impact of demographic skews on the substantive results, are reported in the Supplementary material, Section S2.

2.4 Measures

2.4.1 Demographics

Gender was recorded according to classifications used in the Norwegian population registry (male/female). Age was recorded in years. Education was categorized into five levels corresponding to the Norwegian educational system.

2.4.2 General attitude toward nuclear power

Overall attitude toward nuclear power was measured with the item: “How do you perceive the idea of developing nuclear power for electricity generation in Norway?” Responses were provided on a 7-point Likert scale from 1 (“extremely negative”) to 7 (“extremely positive”), with 4 labeled “neither negative nor positive.” We treat the midpoint as indicating an undecided stance. Two participants had missing data on this item, so analyses involving attitude group classification use N = 1,015.

2.4.3 Beliefs about nuclear power

We use the term belief to refer to the cognitive content held about nuclear power (e.g., that nuclear power generates difficult-to-handle waste), and belief item (or item) to refer to the survey statement used to index that belief. Beliefs were measured with 13 statements capturing perceived advantages and disadvantages of nuclear power (Table 1), rated on a 7-point Likert scale from 1 (“completely disagree”) to 7 (“completely agree”), with 4 (“neither agree nor disagree”) as the midpoint. While item selection necessarily constrains the network boundary, we chose these 13 items to capture core dimensions frequently discussed in nuclear attitude research (e.g., safety/risk, economics, environmental considerations, and energy security). We note that including additional constructs (e.g., institutional trust or political ideology) could yield different network structures; this limitation is addressed in the Discussion.

Table 1. Item descriptions and statistics per group.

M, mean; SD, standard deviation. N = 1,015 (two respondents with missing values on the overall attitude item were excluded from group classification). Items were rated on a 7-point Likert scale from 1 (“completely disagree”) to 7 (“completely agree”), with 4 as the midpoint. Group differences across attitude groups were tested using one-way ANOVAs with Tukey HSD post-hoc comparisons; all between-group differences were significant at p < 0.001 except for “expensiv” (expensive to build), p = 0.006. η² = eta-squared effect size for the omnibus comparison across the three attitude groups (one-way ANOVA, df = 2, 1012).

2.5 Attitude groups

To examine whether belief systems differed by overall stance, we defined three groups based on the general attitude item: Negative (1–3), Undecided (4), and Positive (5–7). Respondents at the scale midpoint are labeled “undecided” for brevity. A midpoint response can reflect genuine uncertainty, neutrality, ambivalence, limited knowledge, or a response style (Kaplan, 1972; Nadler et al., 2015); we return to this in the Discussion.

2.6 Statistical analyses

Descriptive analyses were conducted in SPSS 30 (IBM Corp., 2024). We computed descriptive statistics for the attitude item and all belief items. Differences in mean levels of agreement across the three attitude groups were examined using one-way ANOVAs. When omnibus tests were significant, we conducted post-hoc comparisons using Tukey's HSD (familywise error controlled across pairwise comparisons). Given that multiple belief items were examined, we interpret p-values as descriptive indicators of group differences and emphasize the overall pattern of results across items rather than the significance of any single item. Effect sizes (η²) are reported for all omnibus group comparisons.

Network analyses were conducted in R version 4.4.1 (R Core Team, 2024). Because all belief items were ordinal 7-point scales, associations were estimated with polychoric correlations, selected automatically by cor_auto. Networks were estimated using EBICglasso in the bootnet package [version 1.6; (Epskamp 2025)] with the default EBIC hyperparameter (γ = 0.5); robustness to alternative specifications (γ = 0.25; thresholded estimation; alternative community detection) is reported in the Supplementary material Section S8.3. EBICglasso applies graphical lasso regularization with model selection by the extended Bayesian information criterion; it is the standard estimator for psychological networks based on continuous or near-continuous data and yields sparse, interpretable networks suited to sample sizes in the present range (Epskamp et al., 2018). Edges are interpreted as regularized partial correlations. There were no missing values on the 13 belief items in the analysis dataset: “don't know” responses and item non-response (between 33 and 220 responses per item; Supplementary Table S4) were assigned the scale midpoint during data preparation, and the sensitivity of the network to this midpoint assignment is examined in the Supplementary material Section S8.1. We assessed edge-weight accuracy and node-strength stability using non-parametric bootstrapping in bootnet and report correlation stability (CS) coefficients for node strength, following (Epskamp et al. 2018), who recommend that CS-coefficients should preferably exceed 0.50 and not be below 0.25 to support interpretation of centrality differences.

Belief communities in the full-sample network were identified using walktrap community detection (cluster_walktrap) in igraph (Csárdi and Nepusz, 2006). Walktrap is a community-detection algorithm based on short random walks and does not require the number of communities to be specified in advance, which is appropriate when the latent grouping of beliefs is itself an object of investigation. Group differences in global strength (overall connectivity) and overall network structure were tested using the Network Comparison Test (NCT) package (van Borkulo et al., 2023) with 1,000 permutations; differences in node strength between groups were evaluated within the same framework. NCT is a permutation-based procedure designed specifically for comparing regularized networks across independent groups while accounting for the dependence among edge estimates. Networks were visualized with qgraph (Epskamp et al., 2012).

In these networks, each node represents a single belief, and each edge represents the association between two beliefs after accounting for all other beliefs in the network (a regularized partial correlation). A belief with high node strength is strongly connected to other beliefs in the observed network. Importantly, node strength indicates structural prominence rather than causal influence: high centrality does not imply that changing that belief would produce changes in other beliefs. Table 2 summarizes the key statistical terms in plain language.

Table 2. Plain-language glossary of the main network-analytic terms used in this study.

An analysis plan for the network analyses was registered on OSF after data collection (August 2024; https://osf.io/s7t2n/). Its relation to the analyses reported here, including the exploratory analyses added during peer review, is described in the Supplementary material (“Relation to the registered analysis plan”).

2.7 Data and code availability

The dataset, codebook, questionnaire, and registered analysis plan are publicly available at https://osf.io/s7t2n/. All R scripts, including those for the analyses added during revision, are available at https://github.com/palkraft/nuclear_power_attitudes.

3 Results

3.1 General attitudes

Table 3 presents the sample's demographic characteristics and the distribution of general attitudes toward nuclear power. A substantial proportion of respondents (54.0%) reported favorable attitudes toward developing nuclear power, whereas 21.5% reported negative attitudes and 24.5% were undecided. Design-weighted estimates (calibrated on gender, age, and region) were similar: 53.8% positive, 25.9% undecided, and 20.4% negative, a maximum shift of 1.3 percentage points (Supplementary material, Section S4). The mean attitude score was 4.57 (SD = 1.54), indicating a slight lean toward positive attitudes. Overall, respondents perceived the advantages of nuclear power (M = 5.15, SD = 0.97) to slightly outweigh the disadvantages (M = 4.87, SD = 0.94), t_{(1, 016)} = 5.73, p < 0.001, mean difference = 0.28, 95% CI [0.19, 0.38], d = 0.18. Men reported more favorable attitudes (M = 5.08) than women (M = 4.06), t_{(1, 013)} = 11.26, p < 0.001, mean difference = 1.02, 95% CI [0.85, 1.20]. Age was not significantly correlated with attitudes, r = −0.05, p = 0.09. Sum score scales of perceived disadvantages and perceived advantages were moderately negatively correlated, r = −0.37, p < 0.001.

Table 3. Demographic statistics and attitudes toward nuclear power.

N = 1,017. Education categories correspond to the Norwegian educational system. The general attitude item was scored on a 7-point Likert scale from 1 (“extremely negative”) to 7 (“extremely positive”); see §2.4 for full item wording.

3.2 Attitude-group differences

The three attitude groups differed in both mean levels of agreement with belief items and demographic composition (Tables 1, 4). Concerns about terrorism/warfare risks and nuclear waste management were endorsed across groups, with the negative attitude group reporting the highest levels of concern (Table 1). The groups also differed demographically. Age differed across the three attitude groups, F_{(2, 1, 012)} = 7.28, p < 0.001, η² =.014, with Tukey post-hoc tests indicating that the negative attitude group was older than the undecided group [mean difference = 5.55 years, 95% CI (2.13, 8.97), p < 0.001]. Gender composition differed substantially across groups, χ²(2, N = 1,015) = 113.61, p < 0.001, Cramer's V = 0.335, with women overrepresented in the negative and undecided groups and men overrepresented in the positive group. Education also differed across groups, Kruskal–Wallis H(2) = 9.59, p = 0.008 (Table 4).

Table 4. Demographic and attitudes statistics per group.

N = 1,015 (two respondents excluded due to missing values on the overall attitude item). Education level is scored on a 1–5 scale (1 = primary/lower; 5 = graduate). Group differences: age, F_{(2, 1, 012)} = 7.28, p < 0.001, η² = 0.014; gender, χ²(2, N = 1,015) = 113.61, p < 0.001, Cramer's V = 0.335; education, Kruskal–Wallis H(2) = 9.59, p = 0.008.

3.3 Mean levels of agreement and correlations with overall attitude

The attitude groups differed in their mean levels of agreement with nuclear-power belief items (Table 1). Overall, the positive attitude group reported higher agreement with perceived advantages of nuclear power, whereas the negative attitude group reported higher agreement with perceived disadvantages (Table 1). Most belief differences were statistically significant at p < 0.001; the cost item (“expensive to build”) showed a smaller difference (p < 0.05). Among respondents with positive attitudes, the most strongly endorsed advantages concerned stable energy supply, reduced import dependency, and clean electricity production. Perceived advantages were positively associated with general attitudes toward nuclear power (r = 0.66, p < 0.01), whereas perceived disadvantages were negatively associated with general attitudes (r = −0.60, p < 0.01). The full bivariate correlation matrix between the overall attitude item and individual belief items is reported in Supplementary Table S2.

3.4 Network analyses (full sample)

The belief network estimated in the full sample (N = 1,017) revealed three clusters of interrelated beliefs (Figure 1): a Materialist cluster emphasizing economic and self-sufficiency considerations such as job creation and reduced dependency (orange nodes), an Environmental cluster encompassing both positive and negative environmental considerations (blue nodes), and a Burden cluster focusing on perceived risks such as terrorism/warfare, accidents, and waste management (green nodes). Node strength indicated that the belief that nuclear power provides a stable source of electricity independent of sun and wind (“stable”) was the most structurally prominent node in the network (strength = 1.7). In contrast, beliefs about job creation (“employ”; strength = 0.6) and high construction costs (“expensiv”; strength = 0.6) were among the least central. The stability of the node strength metric was acceptable (CS-coefficient = 0.75).

The “stable” node was connected to several beliefs within the Environmental cluster (e.g., greenhouse-gas emissions and accident risk) and also showed notable connections to beliefs in the Materialist cluster (e.g., reduced dependency on other countries) and the Burden cluster (e.g., terrorism/warfare risk). This pattern suggests that perceived supply stability is a cross-cutting belief linking multiple evaluative dimensions of nuclear power. Additional network results are reported in the Supplementary material (Sections S4, S8, and S10).

3.5 Comparing networks between groups

We estimated belief networks separately for respondents with positive (n = 548), undecided (n = 249), and negative (n = 218) attitudes (Figure 2). Strength centrality was stable for the positive network (CS = 0.75), moderately stable for the undecided network (CS = 0.44), and insufficiently stable for the negative network (CS = 0.28); results involving the negative network are therefore interpreted with caution. Global connectivity was essentially identical across the three groups (global strength = 5.24, 5.07, and 5.09, respectively). Network comparison tests confirmed this: neither global strength nor network structure differed between the positive and negative groups (S = 0.15, p = 0.78; M = 0.23, p = 0.18) or between the undecided and negative groups (S = 0.02, p = 0.97; M = 0.19, p = 0.81).

By contrast, comparing the positive group with a pooled “not-positive” group (negative and undecided combined, n = 467) indicated substantially higher connectivity in the pooled group (global strength = 6.66 vs. 5.24). A matched-n subsampling analysis showed that this difference reflects sample composition rather than the belief systems of non-positive individuals: at identical sample sizes, mixed-attitude samples yielded systematically denser networks than homogeneous samples (mean global strength 5.97 vs. 5.23 at n = 218), whereas increasing sample size within a homogeneous group had virtually no effect (5.23 vs. 5.27 for n = 218 vs. 467). Pooling subgroups with different mean belief profiles introduces between-group covariance that inflates estimated connectivity (Supplementary material, Section S7).

The groups did differ in which beliefs were most central. The belief that nuclear power provides a stable energy supply was the strongest node among positive respondents (strength = 1.77) but not among negative respondents (0.98); this was the only node-level difference that remained significant after correction for multiple comparisons [difference = 0.79, p = 0.002, remaining significant after Holm correction (adjusted p = 0.026)]. Within the negative network, terrorism/warfare risk was descriptively the most central belief (1.00), although this network's low stability precludes strong conclusions. Treating attitude as a continuous moderator rather than a grouping variable yielded the same conclusion: the conditional-association structure was largely invariant across the attitude continuum (Supplementary material, Section S10).

4 Discussion

This study investigated attitudes toward nuclear energy in Norway using psychological network analysis. We (1) described the distribution of nuclear attitudes and related beliefs, (2) examined how these beliefs were organized as an interconnected network, and (3) compared belief networks across the three attitude groups (positive, undecided, and negative). The findings provide insights into how nuclear energy beliefs are structured in a national context without nuclear power in the current energy mix, while highlighting limitations inherent to cross-sectional network analysis.

4.1 General attitudes toward nuclear power

Overall, attitudes in our sample were generally favorable, with 54% of respondents reporting positive views. Support was primarily associated with perceived advantages such as stable energy supply and reduced import dependency. This level of support is consistent with recent Norwegian survey findings (NTB, 2023; Tvinnereim and Liu, 2024) and aligns with broader European evidence that perceived benefits are strongly related to nuclear acceptance (Kim et al., 2014; Wang and Kim, 2018).

At the same time, a substantial proportion of respondents (24.5%) were undecided, indicating that many respondents had not formed a clear evaluative stance on nuclear power. One plausible explanation is limited direct experience with nuclear power in Norway, as unfamiliar technologies often produce greater uncertainty in risk perception and evaluation (Slovic, 1987). In addition, Norway's electricity system is dominated by hydropower, which may reduce the perceived immediacy or personal relevance of nuclear power for some respondents. Finally, Norway's political and public discourse on nuclear energy remains fragmented (Holst, 2023), potentially limiting clear partisan cues that often guide public attitudes. These interpretations remain speculative, and future research should examine how familiarity, perceived relevance, and political cues contribute to attitude formation in this context.

Examining the undecided group descriptively—using the demographic and agreement data summarized in Tables 1, 4—reveals a profile that is distinguishable from both the positive and the negative groups. Undecided respondents were predominantly women (69.5%) and were somewhat younger than the negative group (mean age 48.6 vs. 54.1 years). Their agreement pattern was characteristically intermediate: across both perceived advantages and perceived disadvantages, undecided respondents tended to report agreement levels between those of the positive and negative groups, with no single belief domain showing the kind of pronounced agreement seen in either of the more decided groups. This pattern is consistent with an interpretation in which uncertainty about nuclear power, in the Norwegian context, is associated with the absence of a strongly endorsed evaluative anchor—whether positive (e.g., supply stability) or negative (e.g., terrorism/warfare risk)—rather than with a distinctive belief profile of its own. Whether this descriptive pattern reflects genuinely incomplete attitude crystallization, ambivalence, or low engagement with the issue cannot be determined from cross-sectional data alone, and disentangling these possibilities is an important task for longitudinal and qualitative research.

Two further observations bear on what the midpoint represents in this population. First, “don't know” responding on the belief items was heavily concentrated in this group: 23.3% of undecided respondents answered “don't know” to at least 7 of the 13 items, against 6.9% of negative and 1.5% of positive respondents. Second, the undecided-group network estimated separately (Figure 2; Supplementary material, Section S6) showed the same three-community organization and closely similar connectivity to the decided groups, with supply stability again the most central belief, though less pronounced than among positive respondents. Together, these patterns are most consistent with interpreting midpoint responses as reflecting relatively weakly developed or weakly crystallized attitudes, potentially arising from limited knowledge or engagement with the issue. In this sense, they may be closer to the “nonattitudes” discussed by (Schuman and Presser 1980) than to firmly held neutral positions. At the same time, midpoint responses may also reflect genuine ambivalence, uncertainty, indifference, or other response tendencies (Kaplan, 1972; Nadler et al., 2015), and cross-sectional survey data cannot distinguish among these interpretations at the individual level. The still-hypothetical status of nuclear power in Norway plausibly contributes to several of these response processes.

We observed a substantial gender difference, with men reporting more favorable attitudes than women, consistent with prior work on gendered risk perception (Choi et al., 2000; Greenberg and Truelove, 2010). Age showed no simple bivariate association with attitudes, although age differed modestly across the three attitude groups. Education also differed across groups, indicating that demographic associations may be more nuanced than single-variable correlations suggest.

Gender was the strongest demographic correlate of attitude levels in our data, with men markedly more often positive and women more often undecided. Whether this difference in levels extends to the organization of beliefs is a separate question, and the answer appears to be no: separate networks for men and women showed adequate stability and were statistically indistinguishable in global strength, overall structure, and the strength of every individual node, with supply stability the most central belief in both (Supplementary material, Section S9). Notably, this equivalence holds even though the two gender groups pool very different attitude compositions. The gender gap in nuclear attitudes thus appears to reflect differences in belief levels and in the distribution across attitude groups rather than differences in belief-system architecture—paralleling our finding for the attitude groups themselves, where connectivity was likewise equivalent once groups were examined separately.

Patterns in mean agreement reflected established findings in nuclear attitude research. Concerns about terrorism/warfare and waste management were strongly associated with negative attitudes (Kim et al., 2014; Uji et al., 2021), whereas perceived benefits related to stable energy supply, energy security, and clean electricity production were associated with positive attitudes (De Groot and Steg, 2010). The correlations between perceived advantages and positive attitudes (r = 0.66) and between perceived disadvantages and negative attitudes (r = −0.60) underscore the evaluative importance of these belief dimensions, although the cross-sectional design precludes conclusions about the direction of influence.

A noteworthy feature of the Norwegian context is the apparent divergence between public attitudes and elite or policy positions on nuclear power. While our findings, consistent with other recent surveys (NTB, 2023; Tvinnereim and Liu, 2024), indicate majority public support, the most recent authoritative policy assessment recommends against initiating a comprehensive process to introduce nuclear power in Norway in the near term. The Norwegian Nuclear Power Committee concluded that, although nuclear power can in principle be established and operated with low risk in Norway, it is unlikely to be socioeconomically viable under current cost and market expectations, would require an extensive regulatory and competence apparatus, and would contribute only modestly to climate targets toward 2050 given the alternatives available (NOU 2026: 4, 2026). Such elite–public divergence is not unusual in contested energy domains and has been documented for other technologies in Norway and elsewhere (Kraft and Kraft, 2025). One implication is that the relevance of public attitudes for energy-transition governance lies not only in their aggregate level, but in how the underlying belief systems are organized and how they may respond to evolving expert assessments, policy signals, and information environments—questions that the network approach is well suited to address.

4.2 Network structure and belief organization

Network analysis identified three clusters of interconnected beliefs: a Materialist cluster emphasizing economic and self-sufficiency considerations, an Environmental cluster encompassing both climate-related benefits and ecological concerns, and a Burden cluster focusing on risks and costs. These clusters are consistent with prior research suggesting that nuclear attitudes are multidimensional and structured around distinguishable themes (De Groot and Steg, 2010; Drottz-Sjöberg and Sjöberg, 1990; Hu et al., 2021; Neal and Neal, 2021). More broadly, the clusters may reflect common evaluative frameworks used when citizens assess contested energy technologies: utilitarian/economic reasoning, environmental trade-offs, and perceived burdens and risks.

These three clusters partly converge with, and partly extend, dimensional structures reported in earlier work. The distinction between perceived benefits and perceived risks is well established (de Groot et al., 2013; De Groot and Steg, 2010; Visschers et al., 2011), and our Burden cluster maps closely onto the risk dimension typically identified in this literature, with terrorism/warfare, accidents, and waste management forming a tightly interconnected group of concerns (Drottz-Sjöberg and Sjöberg, 1990; Kim et al., 2014). Less typical is the separation between a Materialist cluster, organized around economic and self-sufficiency considerations (jobs, electricity price, reduced import dependency), and an Environmental cluster combining climate-related benefits with ecological and spatial concerns (greenhouse-gas emissions, accident risk, water for heating, local burden, and biodiversity considerations). In multidimensional studies that rely on factor analysis, “benefits” are commonly treated as a single dimension or split along benefit-domain lines (e.g., Hu et al., 2021; van der Pligt et al., 1982); the present analysis suggests that, in the Norwegian sample, these perceived benefits are organized into two empirically distinguishable subsystems with different evaluative content. The only previous psychological-network analysis of nuclear attitudes (Smolinski et al., 2024) was conducted in Poland and included political ideology as a node, which occupied a central position; the present study extends this emerging literature by examining how perceived advantages and disadvantages are organized in a country without nuclear power in the current energy mix and where political-ideology cues are comparatively weak (Holst, 2023).

The belief that nuclear power provides stable energy supply emerged as the most structurally prominent node in the full-sample network, indicating strong aggregate connections to beliefs across multiple clusters. This suggests that perceptions of reliability and supply stability are closely intertwined with how respondents evaluate nuclear power's benefits and risks. However, this structural prominence should be read as a clue about how the belief system is organized, not as a demonstrated lever for attitude change. Centrality in a cross-sectional network does not establish causal influence, and recent evidence indicates that central attitudes are not necessarily more effective persuasion targets (Brandt and Vallabha, 2023). The direction of influence is also ambiguous: supporters may emphasize supply stability because they are supporters, and unmodeled variables such as political ideology, institutional trust, or the salience of energy security could generate the observed connections. The group comparison strengthens the descriptive case—supply stability was significantly more central among positive than negative respondents, a contrast robust to multiple-comparison correction and to alternative treatment of “don't know” responses (Section 3.5; Supplementary material, Sections S6 and S8.1)—but any communication or policy implication drawn from it remains hypothesis-generating rather than evidence of intervention effects. Longitudinal or experimental designs are needed to test whether changes in specific beliefs propagate to other beliefs and overall attitudes.

The central position of stability beliefs may reflect contextual salience: energy security and electricity-price debates have been prominent in recent European discourse, and reliability may function as a key criterion through which nuclear power is evaluated. Alternatively, stability may be a focal belief because nuclear power is commonly framed as a baseload energy source. These explanations remain tentative and should be evaluated in future comparative research across national contexts and time periods.

Taken together, these network-based findings characterize nuclear-energy beliefs in Norway as an organized system rather than a set of independent considerations. Three features stand out. First, beliefs cohere into three empirically distinguishable evaluative frameworks—Materialist, Environmental, and Burden—that correspond to recognizably different ways of reasoning about nuclear power. Second, one belief—perceived supply stability—bridges these frameworks, suggesting that perceptions of reliability are not confined to a single evaluative dimension but are interwoven with economic, environmental, and risk-related considerations. Third, while overall network structure was broadly similar across attitude groups, the three attitude groups organized the same set of beliefs into systems of similar overall density but with different focal beliefs, with supply stability more central in the positive network and terrorism/warfare concerns descriptively most central among negative respondents. None of these features is visible in aggregate measures of support, nor is it readily recoverable from regression-style or factor-analytic models that treat beliefs as parallel indicators of an underlying construct. They become visible only when beliefs are modeled as a system. We interpret these patterns descriptively, given the cross-sectional design, but they illustrate what a network perspective adds to the study of attitudes toward contested energy technologies: a structured map of how the constituent beliefs relate to one another, rather than only an account of which beliefs are most strongly associated with overall support or opposition.

4.3 Comparing networks across attitude groups

Our group comparisons yielded an instructive result. When belief networks were estimated separately for supporters, undecided respondents, and opponents, global connectivity was essentially identical across the three groups, and pairwise network comparison tests showed no differences in either connectivity or structure. The substantially higher connectivity of a pooled “not-positive” network proved to be compositional: combining subgroups whose mean belief profiles differ introduces between-group covariance that inflates estimated connectivity (Section 3.5; Supplementary material, Sections S6, S7). Within the CAN framework, network connectivity has been theorized to relate to attitude stability, as activation may spread more readily through tightly connected belief systems (Dalege et al., 2016). Applied here, this theoretical link gives no grounds for expecting supporters and opponents to differ in attitude stability: neither camp holds a distinctly more consolidated web of beliefs, and our results caution against inferring group differences in connectivity from pooled samples.

What distinguished the groups was the content around which their belief systems were organized. Among supporters, the belief that nuclear power provides a stable energy supply was the most central node—the only node-level group difference that survived permutation testing with correction for multiple comparisons—whereas terrorism/warfare concerns were descriptively most central among opponents, subject to the caution that the negative-only network's centrality estimates were unstable. Within the CAN framework, structurally central beliefs are those most strongly entwined with the rest of the belief system. The contrast therefore suggests that, in Norway, support for nuclear power is anchored in perceived supply security, whereas opposition centers on perceived security threats. Whether these differently anchored belief systems also differ in temporal stability or resistance to persuasion cannot be determined from cross-sectional data and is a question for longitudinal work.

Beyond the specific findings reported above, the present study contributes to the literature in three ways. First, it extends the emerging network-analytic literature on nuclear-energy attitudes to a further national context—one without nuclear power in the current energy mix and with comparatively weak partisan structuring of the issue. Second, it highlights the importance of examining undecided respondents as a substantive group in their own right: contrary to our registered expectation, the belief system of undecided respondents was no less connected than those of respondents with positive or negative attitudes, suggesting that indecision does not necessarily reflect a weakly organized belief structure. Third, the analyses indicate that pooling respondents with different overall attitudes can inflate estimated network connectivity, illustrating how group composition may affect comparisons of belief-system structure. Taken together, these findings contribute both to research on public attitudes toward nuclear energy and to the methodological literature on the analysis and comparison of attitude networks.

4.4 Limitations

Several limitations warrant careful consideration when interpreting our findings.

4.4.1 Cross-sectional design

The cross-sectional nature of the data precludes causal inference regarding relations among beliefs and between beliefs and overall attitudes. Although psychological network analysis provides a useful descriptive map of conditional associations among belief items, cross-sectional networks cannot establish temporal ordering or test whether changes in one belief would lead to changes in others. The observed network represents a snapshot of belief organization at one point in time and may evolve in response to new information, salient events, or policy developments. Longitudinal designs, experience-sampling approaches, or experimental studies are therefore needed to test whether specific beliefs play a causal role in attitude change, and whether connectivity predicts temporal stability.

4.4.2 Item selection and network boundaries

Network results depend on the boundary conditions implied by the selected set of nodes (Neal and Neal, 2021). Centrality and connectivity metrics are sensitive to which belief items are included or excluded. The present survey was designed to map the structure of perceived advantages and disadvantages of nuclear power and accordingly included 13 belief items spanning safety/risk, economic, environmental, and energy-security domains, alongside an overall attitude item and demographic measures (see Supplementary material, Section S1, for the survey items). Other psychologically important constructs were therefore deliberately not assessed in this study. In particular, political ideology, environmental values, and institutional trust have been shown to predict nuclear attitudes and could occupy influential positions in more comprehensive networks (Bohdanowicz et al., 2023; McBeth et al., 2023; Smolinski et al., 2024)—(Smolinski et al. 2024), for instance, found political ideology to be structurally prominent in a Polish sample. Including such constructs in a single, integrated network is a distinct research question that we identify as a priority for future work; our aim here was to provide a focused structural account of nuclear-specific beliefs. Future research should extend node sets to include broader psychological and sociopolitical factors and examine how network structures change when additional determinants are modeled.

4.4.3 Interpretation of centrality

Centrality in cross-sectional networks should be interpreted as structural prominence rather than causal importance. Although central nodes are often discussed as potentially influential, empirical evidence suggests that centrality may not reliably predict attitude change following persuasive interventions (Brandt and Vallabha, 2023), and debates remain regarding the interpretation and robustness of centrality metrics in psychological networks (Bringmann et al., 2019; Rodebaugh et al., 2018). We therefore caution against drawing practical recommendations based solely on centrality patterns without longitudinal or experimental validation.

4.4.4 Group comparisons and network stability

Our analyses illustrate that combining attitudinally heterogeneous respondents is not innocuous: pooling negative and undecided respondents produced an apparent connectivity difference that disappeared when the groups were analyzed separately. The two subgroups should not be treated as interchangeable—opponents may hold relatively coherent opposing beliefs, whereas a substantial share of the undecided appears not to hold crystallized beliefs at all (23.3% answered “don't know” to at least seven of the 13 belief items). Results for any combined group should accordingly be read as properties of the mixture rather than of its members. Relatedly, the negative-only network's centrality estimates were insufficiently stable (CS = 0.28) to support strong node-level conclusions about opponents; larger samples of opponents are needed to characterize their belief structure with confidence.

4.4.5 Measurement and scaling assumptions

Beliefs were measured using 7-point Likert items and modeled as approximately continuous in a Gaussian graphical model framework. Following the cor_auto procedure, polychoric correlations were used; alternative specifications (γ = 0.25, thresholded estimation, alternative community detection) yielded closely similar structures (Supplementary material, Section S8.3). Likewise, edges represent regularized partial correlations and should be interpreted as conditional associations rather than direct causal relationships. Replication using alternative estimation strategies would strengthen confidence in the robustness of the observed patterns.

4.4.6 Sample characteristics, non-response, and generalizability

The study relied on the Norwegian Kantar Gallup Panel, a well-established survey infrastructure that nonetheless constitutes a non-probability sample. The overall response rate was 27.7%, and participation was strongly selective on age (from 11% in the youngest invitation band to 59% in the oldest) and moderately on education, though essentially not on gender or region (Supplementary material, Section S3). Such selectivity matters for a network analysis in a specific way: if more engaged or better-educated individuals participate at higher rates, the belief system observed in the sample may appear more coherent than in the broader population, since better-crystallized respondents contribute stronger inter-item associations; estimated connectivity should therefore be regarded as an upper bound. Weighted estimates shifted the attitude-group proportions by at most 1.3 percentage points, and a weighted re-estimation left the network essentially unchanged (edge correlation = 0.995; Supplementary material, Section S4), but the design weight does not include education, and no weighting can correct for unobserved differences between respondents and non-respondents. While the overall distribution of attitudes was similar to other Norwegian surveys (NTB, 2023; Tvinnereim and Liu, 2024), the estimated network structure might still differ in underrepresented subgroups.

4.4.7 Response styles and social desirability

Self-reports of this kind are also subject to social-desirability and response-style biases. The direction of social-desirability pressure is ambiguous for nuclear power, so we make no assumption about its net effect on levels. Response styles—acquiescence, midpoint over-use, straight-lining—are the more direct threat to network estimates, because they inflate inter-item correlations and can thereby exaggerate edges, connectivity, and community cohesion. Our sensitivity analyses (Supplementary material, Section S8.2) indicate they do not drive the present results: excluding flagged respondents left the network essentially unchanged (edge correlations ≥0.99 across exclusion sets), and partialling out a general agreement component preserved the three-community architecture while, as expected, attenuating global strength and the centrality of the stable node. The latter is informative in its own right: part of that belief's centrality is constituted by its tie to the general evaluative core of the attitude, rather than by content-specific links alone.

4.4.8 National context and generalizability

The Norwegian setting bounds the generalizability of these findings in identifiable ways. Some results are plausibly context-dependent. The structural prominence of supply stability, and the salience of price- and dependency-related beliefs, likely reflect a hydropower-dominated system facing electrification-driven demand growth, in which nuclear power is debated as a prospective addition rather than an operating reality. In countries with operating reactors or direct accident experience, risk- and waste-related beliefs may occupy more central positions, as the post-Fukushima trajectories of Germany and Japan suggest (Schreurs, 2013; Vivoda and Graetz, 2015). Similarly, the weak partisan structuring of the Norwegian debate may permit a less ideologically anchored belief architecture than would emerge in polarized settings. Other features are better candidates to travel: that nuclear beliefs organize into multiple distinguishable evaluative clusters rather than a single pro–con dimension, and that belief-system connectivity was similar across attitude groups, invite cross-national replication. Aggregate acceptance is known to vary substantially across countries (Kim et al., 2013), but structural comparisons remain rare; fossil-intensive, low-trust, and nuclear-experienced settings would provide the most informative test cases.

4.4.9 Temporal context

Attitudes toward nuclear energy are likely to change in response to contextual factors such as policy debates, media coverage, international events, and energy market dynamics. Data were collected in February 2024, during a period when energy security and electricity prices were salient topics in Europe. The network structure observed here may therefore not generalize to other time periods. Replication across multiple time points would help establish the stability of belief clusters and the consistency of central nodes under changing conditions.

4.5 Implications and future directions

Given the limitations outlined above—particularly the cross-sectional design, boundary specification, and uncertainty about the practical meaning of centrality—we interpret the following implications as tentative and primarily as directions for future research rather than prescriptive recommendations.

4.5.1 Advancing research on attitude formation and change

A notable feature of the Norwegian context was the substantial proportion of respondents who reported being undecided about nuclear power. This underscores the importance of studying how attitudes toward contested energy technologies crystallize over time in contexts where technologies are not yet deployed. Longitudinal panel studies could track how belief networks evolve as individuals move from undecided to more clearly positive or negative positions, and whether changes in network connectivity predict subsequent attitude stability. Relatedly, experimental designs could test whether providing information about specific belief domains (e.g., waste management, accident risk, energy security, costs) induces broader changes in belief networks and overall attitudes, or whether effects remain localized.

4.5.2 Comparative and cross-national replication

The structural prominence of the stable energy supply belief and the division of beliefs into Materialist, Environmental, and Burden clusters should be examined in other national contexts. Cross-national studies could clarify whether stability is consistently central in nuclear belief networks or whether it reflects the specific salience of energy security debates and electricity prices in Norway. Similarly, replication across countries with existing nuclear power, planned expansion, or strong anti-nuclear histories could identify which network features are context-dependent and which may be more universal.

4.5.3 Broader network boundaries and integrated models

Future research would benefit from integrating nuclear-specific beliefs with broader psychological and sociopolitical predictors, including institutional trust, political ideology, environmental values, perceptions of fairness and procedural legitimacy, and attitudes toward energy transitions more generally. Prior research suggests these constructs may be important for nuclear acceptance, and they may occupy structurally prominent positions in more comprehensive networks. Expanding node sets would also allow researchers to evaluate whether belief clusters remain stable when additional determinants are included, and whether central nodes shift as the network boundary becomes more complete.

4.5.4 Implications for sustainability transitions and public engagement

From a sustainability transitions perspective, the identified belief clusters suggest that nuclear energy is evaluated through multiple coexisting frameworks: economic and self-sufficiency considerations (Materialist), environmental trade-offs (Environmental), and perceived burdens and risks (Burden). Mapping these configurations can help characterize the diversity of public evaluations that may surface in deliberation and controversy around contested technologies. While cross-sectional networks cannot determine which beliefs “drive” support or opposition, the observed structure highlights that different segments of the public may approach nuclear energy through different evaluative lenses. For public engagement processes, this implies that a single framing of nuclear energy is unlikely to address the range of concerns and perceived benefits present in the population. Engagement efforts that acknowledge trade-offs and address multiple belief domains may be more appropriate than narrow messaging strategies.

4.5.5 Toward more standardized measurement and cumulative evidence

Finally, network-based studies of nuclear energy attitudes would benefit from greater standardization of measurement. Collaborative efforts across research groups to develop shared item sets and reporting conventions—for example, consistent measurement of risk, benefit, trust, and justice considerations—would improve comparability across contexts and facilitate cumulative knowledge development. Replication using different estimation choices (e.g., ordinal models) and repeated data collection over time would strengthen confidence in the robustness of observed structures and clarify their relevance for understanding attitude dynamics.

5 Conclusions

This study applied psychological network analysis to examine the structure of beliefs underlying public attitudes toward nuclear energy in Norway. Across the full sample, we identified three clusters of interrelated beliefs—Materialist (economic and self-sufficiency considerations), Environmental (climate-related benefits and ecological concerns), and Burden (risks and perceived costs)—highlighting the multidimensional character of nuclear energy perceptions in a national context where nuclear power is not part of the current energy system.

A substantial proportion of respondents reported favorable attitudes toward nuclear power, while nearly one-quarter were undecided, underscoring that attitudes are not fully crystallized for a substantial share of the public. Within the belief network, the perception that nuclear power provides a stable energy supply independent of sun and wind emerged as the most structurally prominent belief, indicating that stability considerations are strongly interconnected with multiple belief domains. Network comparisons indicated that belief-system connectivity was equivalent across attitude groups once these were examined separately; the groups differed instead in which beliefs were most central, with supply stability significantly more central among positive respondents.

These findings should be interpreted in light of important limitations, particularly the cross-sectional design, which precludes causal inference, and the sensitivity of network metrics to boundary specification and item selection. The belief set did not include broader determinants such as institutional trust, political ideology, or perceived procedural fairness, which may play important roles in more comprehensive models. Future research using longitudinal and experimental designs, and incorporating a broader range of psychological and sociopolitical predictors, is needed to test whether network structure predicts attitude change and how belief systems evolve as policy debates and energy-system conditions shift.

Despite these limitations, the study contributes to research on energy attitudes by providing a structured description of how nuclear energy beliefs are organized and interconnected in the Norwegian public. Mapping such belief configurations can support sustainability transitions research by clarifying the evaluative frameworks through which contested low-carbon technologies are perceived, and by informing future work on public engagement and the dynamics of attitude formation.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsuep.2026.1860422/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Gilbert Michaud, Loyola University Chicago, United States

Reviewed by: Yiliu Liu, Norwegian University of Science and Technology, Norway

Anton Brännlund, Uppsala University, Sweden