Mapping quality of life in Norway: psychometric evaluation and network analysis of 15,148 responses from a public health study

Study design and data collection

This article utilises anonymised data from the third wave of the NQoLS. The cross-sectional survey was carried out during 4 weeks, from March 7th to April 6th, 2022. The online survey started with an introduction that explained the purpose of the study and included a consent form. Participants needed to click and agree to fill out the form before proceeding to the questionnaire, so their participation implied their informed consent. Statistics Norway’s survey efforts adhered to the requirements of the Statistics Act and complied with the principles outlined in the EU’s General Data Protection Regulation. The collected data were anonymised and deposited in the Norwegian Agency for Shared Services in Education and Research (SIKT) data repository, freely and openly available for scientific purposes (SIKT, 2023). Utilising anonymised data from the NQoLS does not require ethical approval, as confirmed by the Regional Committee for Medical Research Ethics, South-East Norway (REK no. 712698) (Vederhus et al., 2024).

Sample

A nationally representative sample of 40,000 individuals aged 18 years and older residing in Norway was randomly selected from the central population database, BeReg, managed by Statistics Norway. The sample was demographically reflective of the Norwegian population. Age was calculated as of March 7th, 2022, the start date for data collection. Participants were contacted through the Contact and Reservation Register, which stores residents’ contact information. The survey was conducted primarily via email and SMS, with additional contact made via phone where necessary. A small percentage of the sample (0.7%) lacked email addresses, and 7% had no registered mobile numbers. Respondents without electronic contact information were mailed a paper invitation to participate.

Survey implementation

Data collection was conducted online through a web-based survey. To manage the large sample size, Statistics Norway divided the respondents into 18 dispatch groups, each receiving email and SMS invitations in staggered batches. The survey link was included in these communications, and by completing the questionnaire, participants provided informed consent. Participation was voluntary, and respondents could withdraw at any time. They were assured that their data would be anonymised within 1 year of completion. The questionnaire was available in the two major Norwegian language versions (Bokmål and Nynorsk) and English; the median completion time was 20 min (mean: 24 min). Note that both Nynorsk and Bokmål are mandatory to learn in Norwegian schools, and share virtually identical grammar and core vocabulary, differing mainly in word endings and certain stylistic conventions. Additional demographic and socioeconomic data were sourced from various national registers, including those for education, employment, income, and living conditions, to minimise the respondent burden (Pettersen & Engvik, 2022).

Measures of health and quality of life

The survey employed well-established, freely available measures recommended by the Norwegian Directorate of Health (Nes, Hansen & Barstad, 2018; Pettersen & Engvik, 2022) to capture multiple facets of SRH and overall QoL outcomes. Details about the study, including the items and scoring of the questionnaire, are available online in an English version (Støren, 2024). An expert group with relevant linguistic and cultural expertise designed and prepared the survey in three languages. Before data collection, the group conducted structured evaluations and user testing to ensure the items were clear and culturally appropriate (Nes, Hansen & Barstad, 2018). Life satisfaction, life meaning, and optimism were assessed using single-item measures, with responses rated on a 0–10 scale, with higher scores indicating better outcomes (Cantril, 1965; Pettersen & Engvik, 2022). General health was rated using a five-point scale ranging from very bad to very good, and Satisfaction with physical or psychological health was rated on a 0–10 scale, with higher scores indicating better outcomes (Pettersen & Engvik, 2022). Pain and discomfort was assessed combined using one item from the EQ-5D-5L, rated on a five-point scale from no pain or discomfort to extreme pain and discomfort (Rabin & De Charro, 2001). In addition to single-item measures, the survey included multi-item scales to assess various dimensions of health and wellbeing. These included the Satisfaction with life scale, where the score ranges from 1 to 7, with higher scores indicating better outcomes (Diener et al., 1985); Hopkins Symptom Checklist-5 (HSCL-5), where the score ranges from 1 to 4, lower scores indicating better outcomes (Strand et al., 2003); Positive and Negative Affect Schedule, where scores ranged from 0 to 10, higher scores indicating more intense effect, either positive or negative (Hansen & Slagsvold, 2012; Watson, Clark & Tellegen, 1988); Pearlin mastery scale, where scores ranged from 1 to 5, higher scores indicating better outcomes (Hysing et al., 2020); European social survey engagement scale, where scores ranged from 0 to 10, higher scores indicating better outcomes (Charalampi, Michalopoulou & Richardson, 2020), and the UCLA loneliness scale, where scores range from 1 to 4, lower scores indicating better outcomes (Steptoe et al., 2013).

Analysis

We applied a modified version of the Wilson and Cleary framework to categorise variables into (1) biological and physiological variables, (2) symptoms and functioning (which we in this article divide into physical health, psychological health, and social health), (3) general health, and (4) overall QoL (Wilson & Cleary, 1995). In addition, we categorised variables as (5) characteristics of the individual and (6) environmental factors.

Descriptive statistics, including means, medians, standard deviations, skewness, and kurtosis, were calculated to assess data distributions. Skewness ≥2 and kurtosis ≥7 indicated significant deviations from normality, while floor and ceiling effects may be problematic if more than 15% of respondents selected the lowest or highest possible scores. Internal consistency of the scales was evaluated using Cronbach’s alpha, with values ≥0.7 deemed satisfactory (Horney & Tarlov, 1995).

The structural validity of scales with four or more items was assessed using factor analysis (CFA) for ordinal data, employing weighted least squares mean and variance (WLSMV) estimation. This tests whether all items in a scale measure the same underlying concept. Model fit was evaluated using RMSEA, SRMR, CFI, and TLI, with values close to or below .06 for RMSEA, close to or below .08 for SRMR, and close to or above .95 for CFI and TLI considered indicative of good fit (Hu & Bentler, 1999). As these indices capture different aspects of model fit, they should be interpreted together to provide a complementary evaluation of how well the items reflect the intended construct. Importantly, these cutoffs are guidelines rather than strict rules, and model evaluation should also take into account the broader measurement context. Because RMSEA can overestimate misfit when degrees of freedom are small or when categorical indicators are used, its values were interpreted in conjunction with CFI, TLI, and SRMR, which are regarded as more robust under such conditions (Shi, Maydeu-Olivares & Rosseel, 2020).

Correlations between variables were displayed using Pearson correlation coefficients. A regularised partial correlation network was estimated in JASP using the EBICglasso estimator (Bergh van den, 2018; Epskamp et al., 2012; Schönenberg et al., 2023). In the network, each health or QoL measure is shown as a node (a dot), and the lines between them (edges) show how strongly they are related. The correlation structure was determined using JASP’s automatic selection method. Model selection was based on the extended Bayesian Information Criterion (EBIC), with the tuning parameter (γ) set to 0.5. Missing data were handled using listwise exclusion. Network visualisation was generated using edge colour and thickness, reflecting the direction and magnitude of partial correlations. Three standard centrality indices—strength, closeness, and betweenness—were computed and normalised for node comparability. Additionally, the expected influence was estimated to capture net connectivity while preserving the sign of associations. Four alternative metrics—Barrat, Onnela, WS, and Zhang centrality—were also calculated to assess robustness across different centrality frameworks. Two bootstrap analyses tested the network’s stability and accuracy. First, a nonparametric bootstrap assessed edge precision by generating 1,000 resamples and re-estimating the network each time. We compared each edge’s bootstrapped mean and confidence intervals with the original estimates. Narrower intervals indicated greater reliability, whereas wider intervals suggested higher uncertainty. Secondly, a case-dropping bootstrap with 1,000 resamples evaluated the centrality indices (strength, closeness, and betweenness). We progressively removed random portions of the dataset, recalculated each metric, and correlated these with the full-sample results, thus revealing how sensitive each node’s importance was to variations in the sample.

We conducted sensitivity analyses stratified by language version—Norwegian Bokmål, Norwegian Nynorsk, and English—for all tests except the network analysis. In this case, the Nynorsk and English subgroups were too small to support reliable analyses, whereas the Bokmål subgroup was sufficiently large to approximate the pooled analysis closely. Statistical analyses were performed using JASP version 0.95.3 (JASP, 2024).

Psychometric analysis

The measures indicated relatively good health and overall QoL for many participants. Still, the variance in scores also revealed that a significant minority experiences challenges related to SRH and overall QoL (Tables 3–4). Psychometric evaluation of the seven single-item measures generally revealed acceptable measurement properties (Table 3). However, a large ceiling effect was observed for General health and Pain and discomfort. The psychometric analysis of the seven multi-item scales demonstrated acceptable properties, consistent mean and median values, appropriate skewness and kurtosis, minimal floor and ceiling effects, and satisfactory Cronbach’s alpha values (Table 4). CFA conducted on five scales showed acceptable SRMR, CFI, and TLI values. However, four scales exhibited elevated RMSEA values. Modifying covariances between error terms for specific items resulted in improved RMSEA values, rendering them adequate for all scales.

Correlational analyses revealed significant associations among the variables (Table 5). As expected, this is consistent with the probable presence of partial conceptual convergence and bidirectional causality.

Network analysis

The network analysis revealed a structured constellation of relationships among the 14 constructs (Figs. 1–3, Supplement 1). The resulting network was moderately dense, indicating a coherent but differentiated structure across psychological, physical, and social domains. The Satisfaction with life scale emerged as the most topologically central node, with high closeness (1.872), strength (0.445), and moderately positive expected influence (0.165). Satisfaction with psychological health showed the highest closeness (1.838) and acted as an integrative variable across affective and evaluative domains despite a slightly negative expected influence (−0.500). Life satisfaction had fewer connections (strength = −0.071) but a high expected influence (0.825), indicating meaningful contributions to well-being via fewer but consistently positive links. Positive emotions and Life meaning further supported the well-being cluster. In contrast, Engagement and Optimism, though not strongly connected, exhibited positive expected influence (0.988 and 1.267, respectively), suggesting a supportive role.

Hopkins Symptom Checklist-5 stood out for its high strength (1.715), betweenness (1.694), and closeness (0.765), marking it as a key bridge between distress-related constructs and subjective well-being. Negative emotions were moderately connected to both distress and satisfaction variables, reinforcing their role in the emotional dimension of HRQoL. UCLA loneliness scale showed low global centrality (closeness = −1.272, strength = −1.827) but high local clustering (e.g., Zhang = 1.309), suggesting that loneliness formed a dense but structurally independent subdomain.

Satisfaction with physical health, Pain and discomfort, and General Health were located at the network’s periphery. These variables exhibited low strength and closeness (e.g., node 12: strength = −0.023; expected influence = −1.453) but high local clustering (e.g., node 8: Zhang = 1.756), indicating a cohesive but isolated somatic subnetwork.

Finally, the Pearlin mastery scale displayed moderate centrality and positive influence, connecting to distress and resource indicators. Alongside optimism and engagement, it contributed to a diffuse resilience cluster linked indirectly to the central nodes.

Bootstrapped network stability analyses were conducted to evaluate model robustness (Supplement 1). The edge stability plot showed high consistency between bootstrap means and original estimates for the strongest positive and negative connections, particularly those involving Hopkins Symptom Checklist-5, Satisfaction with life scale, and Life satisfaction. However, many weaker edges showed considerable variation across bootstraps, suggesting these associations are less reliably estimated and should be interpreted cautiously. The case-dropping centrality stability analysis indicated that strength centrality exhibited the highest and most consistent stability (average correlation ∼0.35–0.45). In contrast, betweenness and closeness showed lower but relatively flat correlations across subsamples. This pattern likely reflects the stabilising effect of the large sample size and the dense interconnectivity among outcomes.

Sensitivity analysis

Participants were divided into three language-based subsamples: Bokmål (n = 13,872), Nynorsk (n = 555), and English (n = 721). The mean age was 49.9 years (SD = 17.3) in Bokmål, 51.9 years (SD = 16.9) in Nynorsk, and 38.6 years (SD = 13.3) in English. The proportion of women was 52.3%, 48.1%, and 42.9%, respectively, and the proportion of first-generation immigrants was 9.1%, 2.5%, and 73.0%. Although there were some minor differences in health and QoL across the subsamples, with the English group tending to score somewhat poorer overall, the general pattern of psychometric results indicated good agreement with a few subtle variations. This suggests that the measures’ performances were stable across languages (e-Tables S4–S9 and e-Figs. S3–S5, Supplement 1).

Psychometric analysis

The single- and multi-item scales used in the NQoLS demonstrated generally robust properties, with acceptable score distributions and satisfactory psychometric performance. However, pronounced ceiling effects were observed for the single-item measures General health and Pain and discomfort. These effects may partly reflect genuinely high levels of well-being and low pain prevalence within the general population. Alternatively, they may also arise from measurement constraints, such as a limited number of response categories and a restricted conceptual scope, which could reduce sensitivity at the upper end of the scale. Future public health studies addressing General health, pain and discomfort may therefore benefit from employing multi-item or condition-specific instruments when very high measurement precision is required (Norekvål et al., 2025). Confirmatory factor analyses supported the structural validity of the multi-item scales, although a few measures (e.g., Pearlin mastery scale, and Negative emotions) showed elevated RMSEA values that required minor model modifications. However, RMSEA values remained slightly elevated for some scales, but this was not of major concern. As RMSEA may be biased under these conditions, greater weight was placed on the other indices, which indicated excellent fit and supported the conclusion that the overall model fit was well within acceptable limits (Shi, Maydeu-Olivares & Rosseel, 2020). Taken together, these findings suggest that the NQoLS multi-item scales provide a reliable assessment of SRH and QoL. Nonetheless, targeted refinements may further improve model fit for selected measures.

Network analysis

Network analyses provided insights into how SRH and QoL outcomes were interlinked. Three measures emerged as especially central: the Satisfaction with life scale, the Hopkins Symptom Checklist-5, and Satisfaction with psychological health. These constructs bridged affective indicators (positive and negative emotions) with overall life evaluations, underscoring the pivotal role of psychological health in shaping wellbeing. While the Satisfaction with life scale anchored the central cluster, Satisfaction with psychological health was topologically closest to all other nodes, reinforcing its integrative function. The Hopkins Symptom Checklist-5 served as a bridge between wellbeing and distress, supporting models of mental health that emphasise measuring both symptoms and strengths. Life meaning and Positive emotions added further conceptual coherence to this cluster. Additionally, Life satisfaction—while less connected, had a strong indirect influence, indicating that global well-being evaluations can influenced by a few but high-quality connections. Psychological variables like Optimism, Engagement, and Mastery did not hold structurally central positions but consistently showed positive expected influence. Their role as facilitators of wellbeing aligns with theories of resilience and stress-buffering, suggesting that they may amplify or safeguard psychological functioning without directly anchoring the system.

The UCLA loneliness scale displayed a distinct structural profile. Despite having low global connectivity, its high clustering indicates that loneliness forms a distinct emotional domain with indirect links to broader wellbeing. These results align with recent research framing loneliness as a unique psychosocial determinant with important health implications (Klein et al., 2021). In contrast, physical health variables—Satisfaction with physical health, Pain and discomfort, and General health—occupied peripheral network positions. Although general health is often conceptualised as a summary indicator, it was weakly integrated with the network. This finding is consistent with previous studies suggesting that general health may be associated more with physical rather than psychological health (Andersen et al., 2022). While these variables formed a locally cohesive subnetwork, their limited reach into the broader system indicates that the current structure of the NQoLS may underrepresent somatic domains (Pettersen & Engvik, 2022). Overall, the network supports a multidimensional model in which psychological well-being and satisfaction with life form a densely integrated core, while distress, loneliness, physical complaints, and psychological resources occupy distinct yet interconnected peripheries. These results suggest that public health assessments and interventions should consider global well-being and the specific pathways through which emotional, somatic, and social experiences influence perceived health.

According to the Wilson and Cleary model, health outcomes form a pathway from biological variables and symptom status to broader domains of functioning, perceived health, and overall QoL (Wilson & Cleary, 1995). Our network findings partly support this idea, illustrating how psychological health indicators (e.g., Hopkins Symptom Checklist-5) can bridge emotional functioning and global life satisfaction. Notably, general health—a pivotal construct in Wilson and Cleary’s hierarchy—exhibited relatively low connectivity. This has implications that merit further consideration. For example, the weak integration of general health in the network highlights the possible limitation of relying on this item in population surveys, particularly when complex health processes are involved. It also raises questions about how individuals weigh different aspects of SRH and QoL. The central positioning of psychological variables suggests that perceptions of SRH and QoL may be shaped more by emotional and cognitive states than by simple measures of physical health, as in this study. This highlights the importance of including subjective well-being and symptom burden, as well as more robust and comprehensive measures of physical health, in future research.

Strengths and limitations of the NQoLS

A strength of the NQoLS is its broad scope, encompassing a wide range of variables. In addition to measuring SRH and QoL directly, it includes many factors that may influence these outcomes. This extensive coverage provides a strong foundation for population-level monitoring, enables the identification of vulnerable subgroups, and supports the design of targeted interventions to reduce health inequalities. However, the lack of alignment between the NQoLS and other major Norwegian cohort studies such as the HUNT Study and the TromsøStudy. represents a structural limitation. Using different SRH and QoL instruments across public health studies restricts comparability across cohorts and reduces the potential for longitudinal and cumulative analyses (Kaplan & Hays, 2022). Although user involvement was included in developing the NQoLS, harmonisation with established national surveys was not prioritised. Future iterations would benefit from incorporating a core set of indicators already used in large, longstanding studies, improving comparability and enabling integration across cohorts. Finally, although an expert group with relevant linguistic and cultural expertise designed and prepared the survey in three languages, the original report provides little information on this process (Nes, Hansen & Barstad, 2018). This may be considered another limitation of the NQoLS that readers should bear in mind when interpreting the results.

Strengths and limitations of the current study

The current study enhances the NQoLS by integrating psychometric evaluation with network analysis, applied to a large and diverse sample. This approach offers a map of how SRH and QoL outcomes interrelate in a general population. The response rate was acceptable for studies of this, with minimal missing data. Nonetheless, several limitations should be acknowledged. The cross-sectional design prevents causal inference, and despite weighting, some demographic bias likely remains, particularly underrepresentation of older and less-educated individuals. It is also important to acknowledge that selection bias is inevitable in studies of this kind. Health problems and life difficulties may lead some individuals to refrain from participation, which means that the results are likely skewed toward better outcomes compared to what would have been observed if all eligible participants had taken part. Moreover, while the network structure was stable for strong associations, weaker edges showed more variability across bootstrap samples and should be interpreted cautiously. Thus, future research should aim to replicate these results and employ alternative methodological approaches to more robustly evaluate network patterns and their stability.