Marked socio-economic inequalities in health persist across Europe and reducing them is a key priority for most European countries (European Commission, 2021; Joint Action Prevent Non-Communicable Diseases Project, 2024), including Norway (Meld. St. 15 (2022–2023)). Addressing these disparities is also central to the United Nations’ Sustainable Development Goals (SDGs), particularly Goal 3 (Good Health and Well-being) and Goal 10 (Reduced Inequalities) (Morton et al., 2017; United Nations, 2015).

Using Norway as a case, we examine how education and household income are associated with behavioural risk factors and health outcomes (mortality) among working-age adults, leveraging both survey data (2015 and 2019) and full population register data (2015–2019).

Norway ranks among the top countries in the Human Development Index (United Nations Development Programme, 2024), and has low income inequality, a generous social security system (Barth et al., 2021; Eurostat, 2023; Molven and Ferkis, 2011) and one of the highest per capita health expenditures in Europe (WHO Regional Office for Europe, 2025). Norway has made reducing socio-economic health inequalities a political priority. A government white paper from 2007 called for targeted efforts to reduce inequalities over the next decade, particularly in smoking, nutrition and physical activity (St.meld. nr. 20 (2006–2007)). Despite this, significant socio-economic inequalities in mortality and health-related behaviours persist (Mackenbach, 2017, 2019; Mackenbach et al., 2008).

Given Norway’s focus on reducing social inequalities in health and its fast rate of adoption of international health regulations (World Health Organisation, 2025), examining the state of social inequalities in health in Norway can illustrate the regulatory challenges other European countries face in reducing inequalities.

Poor health disproportionately affects lower socio-economic groups, increasing healthcare demand, sickness absences and reliance on social security. Given the strong relationship between education, income and health outcomes, reducing inequality in these social determinants may help to mitigate health disparities. Improving healthcare services and optimising treatment outcomes, which are often worsened by behavioural risk factors, are also critical. These measures, alongside broader policies grounded in proportionate universalism, are necessary to effectively reduce health inequalities.

This study contributes to existing knowledge in three key ways. First, it employs established methods of measuring inequality (Relative Index of Inequality, RII, and Slope Index of Inequality, SII) (Mackenbach et al., 2008; Mackenbach et al., 2018) across multiple behavioural risk factors and mortality indicators, facilitating cross-country comparability. Second, it combines survey and registry data to assess whether inequalities are consistent across different data sources, and to what degree inequalities are associated with specific measures. Third, it provides up-to-date estimates of absolute and relative health inequalities in a country with a strong welfare system and a longstanding political commitment to reducing disparities. Given the rising income inequality and shifting educational demographics in Norway (Statistics Norway, 2021), understanding how inequalities emerge across different health measures remains crucial. We address the following research questions: •

To what extent do socio-economic inequalities in health exist among adults in Norway?

In 2019, Norway had a higher life expectancy (83 vs 81 years) and lower rates of all-cause mortality ¹ (701 vs 770), daily smoking (9% vs 17%), BMI of 25 or more (48% vs 56%), self-reported less-than-good health ² (37% vs 55%) and activity limitations ² (23% vs 50%) compared to other OCED countries. Levels of daily vegetable consumption (58% vs 59%) and insufficient daily physical activity (34% vs 35%) were similar to the OECD average (OECD, 2021).

Recent studies show that marked and relatively persistent differences in lifestyle, health, use of health services and mortality exist in Norway (Goldblatt et al., 2023; Gutacker et al., 2023; Kinge et al., 2015, 2023) as well as in other European countries (Enroth and Fors, 2021; Gallo et al., 2012; Mackenbach et al., 2015a; Moor et al., 2017), and that these differences are related to both education and income.

However, the magnitude of the estimates depends on which health outcomes are investigated and whether relative or absolute differences are examined, and also partly on the context (who, where and when) (Granström et al., 2015; Hu et al., 2016; Lahelma et al., 2019; Leão et al., 2018; van der Wel et al., 2018).

Since the publication of the British Black report and the Whitehall I study (Marmot et al., 1978; Townsend and Davidson, 1982), an extensive literature has documented health inequalities, particularly in self-reported health and mortality, from the early 1990s onwards (see, for instance, Hu et al., 2016; Kunst et al., 2005; Lahelma and Arber, 1994; Lahelma et al., 1997, 2002; Mackenbach et al., 1997; Strand et al., 2010; von dem Knesebeck et al., 2006; Yngwe et al., 2001).

In general, health inequalities have remained stable even during economic crises, including during the 2008 financial crisis (Leão et al., 2018; Mackenbach et al., 2018). Leão et al. attributed this stability to countries avoiding austerity policies (Leão et al., 2018). Mackenbach et al. (2018), however, observed a slowdown in the decline in less-than-good health among people with both short and long education, even in the hardest hit countries.

Most studies on health inequalities focus on education, though some also include individual or household income. Although the mechanisms linking income, education and health are complex, income can be regarded as a measure of material resources, while education represents ability, information and social and cultural capital (Dahl, 1994; Goldman, 2001). Research consistently shows that people with higher education or higher income report better health and live longer lives than their less advantaged counterparts.

However, few existing studies on health inequalities, with notable exceptions (Huijts et al., 2010; Lahelma et al., 2004; Muller, 2002), have examined the impact of both education and income. Even fewer studies have examined results from both relative and absolute perspectives across a wider range of health outcomes using both population survey and full-population samples. By including a wide range of risk factors and measures of mortality and assessing inequality in terms of both education and income, we aim to provide a comprehensive picture of socio-economic inequality in Norway using the latest available data from before the COVID-19 epidemic.

We used two data sources: representative surveys and registries. The sample included adults aged 30–69 years, as most of this group had completed their education and established an income, while excluding most retirees.

To investigate the associations between health behaviours, outcomes and socio-economic position (educational and income), we employed two gender-specific regression-based inequality indices (cf. Mackenbach et al., 2008) to pooled 2015 and 2019 data: (1) the Relative Index of Inequality (RII) for relative differences and (2) the Slope Index of Inequality (SII) for absolute differences.

Both the RII and the SII are regression-based measures that account for the distribution of socio-economic characteristics in the samples (Mackenbach and Kunst, 1997; Mackenbach et al., 2008). To estimate the RII and the SII, we calculated “ridits” for education and income using the “wridit” command (Newson, 2012) in Stata 18 (StataCorp LLC, 2023), stratified by gender and five-year birth cohort. Ridits represent cumulative probabilities on a continuous scale, ranging from zero (lowest status) to one (highest status) (Donaldson, 1998). Specifically, the ridit for value x of variable X is the proportion of values lower than x, plus half of the proportion of values equal to x. For example, the middle income tertile has a mean ridit value of 0.5 (0.33 + [0.33/2]).

The RII was calculated by regressing the ridit-transformed education and income variables on the outcomes (mortality, MEHM and self-reported dental health, obesity, smoking and fruit/vegetable intake) using generalised linear regression models (GLM) with Poisson-family and log-link. RII was the exponentiated beta coefficient.

The SII, representing the absolute difference between the lowest and the highest education and income groups, was similarly calculated using GLM with an identity-link. SII was the beta coefficient of the transformed education and income variables. Hjorthen et al. (2022) demonstrated that these calculation methods yield results equivalent to Mackenbach et al. (2008).

Regression models for self-reported health and health behaviours were adjusted or 10-year age groups and a dummy variable for calendar year. Survey weights, accounting for age, sex, education and geographical region of residence, were applied to ensure representativeness when examining self-reported health and health behaviours. Mortality models, utilising registry data, were adjusted for 10-year age groups and five-year birth cohorts (cf. Pamuk, 1985). ⁹ Statistical significance was set at 5%.

Sensitivity analyses, employing alternative model specifications for income (Supplementary material, Figures S.1 and S.2, available online at https://doi.org/10.1553/p-d6az-d2j6), demonstrated minimal impact on estimates.

RII and SII differences were influenced by the baseline level of the lowest category. SII reflected the absolute difference in the outcome between the highest and the lowest possible value of the explanatory variable. For example, a SII of 0.1 for self-reported health indicated a 10 percentage point higher likelihood of having poorer self-reported health in the lowest educated group compared to in the highest educated group.

Table 1 summarises the samples used in our analyses and highlights differences in education and income distributions between the registry sample and the survey sample. Secondary and tertiary education varied by gender and data source. In the registry sample, 20% of the population had basic education across genders, while a larger share of women (42%) than of men (33%) had tertiary education. In the survey sample, smaller shares of the population had basic education (14% of women, 16% of men), with tertiary education being overrepresented (50% of women, 37% of men).

Table 1 indicates a pronounced social gradient in most health outcomes for both men and women. The prevalence difference between women at the top and the bottom of the education distribution was especially large for less-than-good self-reported health (27 percentage points [pp]), activity limitations (22 pp), longstanding health problems (18 pp) and current smoking (21 pp). Among men, absolute differences were largest for current smoking (23 pp) and low physical activity (22 pp). Differences according to income were also substantial, but smaller. Absolute differences in BMI were relatively small, regardless of the measure used. A gradient for both income and education was also observed for all three mortality outcomes (cf. Supplementary material Table S.1 for further details on mortality).

Figure 1 depicts RIIs for self-reported health behaviours and outcomes among men (solid markers) and women (hollow markers) for education (left panel) and income (right panel). Detailed estimates are reported in Supplementary material Table S.2. Across all health behaviours (triangles), BMI (squares) and self-reported health (circles), the RII coefficients were consistently positive with a 95% confidence interval (CI) above one, indicating statistically significant differences, except for income inequality in fruit intake among men (RII=1.09, CI 0.98–1.21). ¹⁰ Associations were especially strong for less-than-good self-reported health, activity limitations and current smoking.

Figure 2 similarly shows pronounced relative inequalities for all-cause premature mortality, medical intervention-amenable mortality and smoking-related mortality (see Supplementary material Table S.3). Educational attainment (left panel) showed substantial relative inequalities for both genders for all-cause premature mortality ( RII women = 3.5 , 95% CI=3.3–3.7 and RII men = 3.6 , 95% CI=3.5–3.8), medical intervention-amenable mortality ( RII women = 2.4 , 95% CI=2.1–2.7 and RII men = 3.8 , 95% CI=3.2–4.4) and especially smoking-related mortality ( RII women = 6.2 , 95% CI=5.4–7.1 and RII men = 6.2 , 95% CI=5.5–7.0). No substantial gender differences were observed for all-cause premature mortality or smoking-related mortality for education. However, women exhibited lower income-related mortality estimates than men. Among men, income had a strong association with all-cause mortality (RII=6.2, 95% CI=5.9–6.5), with the greatest inequality observed in deaths amenable to medical intervention (RII=9.5, 95% CI=7.8–11.6).

Figure 3 illustrates absolute inequality (SII) for self-reported health outcomes, with estimates detailed in Supplementary material Table S.2. Note that the axes are different from the axes in Figure 1, reflecting smaller numerical inequalities on an absolute scale, where values greater than zero indicate statistically significant differences.

The SII patterns differed somewhat from those for RII, with the largest absolute differences observed for smoking, exercise and less-than-good dental health (at least for income) among men, and for smoking, exercise, self-reported health and activity limitations among women. BMI had the smallest SII estimates among both genders, except for income inequalities in fruit intake among men, which was not statistically different from zero (p=0.09).

Figure 4. (and Supplementary material Table S.3) shows that compared to relative inequality (Figure 2), estimates for absolute inequality in all-cause mortality (per 1,000) were higher compared to medical intervention-amenable and smoking-related mortality, which was partly attributable to low baseline mortality in adults under 70. The large absolute differences for all-cause mortality reflect the higher number of overall deaths compared to deaths from select causes, with a larger impact among men. Among men, absolute estimates for income and education were similar (13/1,000), unlike the relative scale, which showed greater disparities. Among women, SII was higher for education than for income for all-cause and smoking-related mortality.

Our findings, which describe the situation in recent years, are similar in magnitude to findings from several European countries (Mackenbach et al., 2008; 2015a,b, 2018; 2019a). This likely reflects shared health and welfare systems characteristics and similar societal views on social inequalities, as observed in countries like the UK, the Netherlands, Sweden and Finland (Lahelma et al., 2019). Supporting an earlier study of 22 countries (Mackenbach et al., 2008), we found significantly higher mortality and poorer self-reported health in lower socio-economic groups. While Mackenbach et al. (2008) attributed these disparities to smoking, alcohol and intervention-amenable mortality, our RIIs for poorer self-reported health and current smoking were generally larger. Our finding that income was associated with poor health, particularly in relative terms, is consistent with previous studies (Siegel et al., 2014; Wachtler et al., 2019). Notably, a systematic review found that material factors had a direct impact on health inequalities (Moor et al., 2017).

This study presents a snapshot of the current situation. Longitudinal studies indicate that inequalities have increased over time, although their findings vary depending on whether education or income is considered. While educational inequalities have remained fairly stable (as noted above), many studies have reported a growing trend in income-related disparities (Degerlund Maldi et al., 2019; Pförtner and Elgar, 2016), albeit with some exceptions.

The observed socio-economic disparities in self-reported health and mortality align with the existing literature (Groeniger et al., 2020; Pampel et al., 2010). However, the extent to which physical inactivity, obesity and low fruit/vegetable intake explain the observed variance in health outcomes continues to be debated. Some authors have argued that the predictive value of these factors is relatively low (cf., for instance, Brunello et al., 2016; Hamplová et al., 2022; Jones and Wildman, 2008), while others have suggested that they play a vital role (cf., for instance, Hämmig et al., 2014; Moor et al., 2017). Smoking, however, stands out as a key factor explaining inequalities in both health and mortality, particularly in countries with high smoking prevalence (Lanari et al., 2022). This observation is in line with Mackenbach et al. (2019b), who identified smoking, low income and high bodyweight as critical targets for policies aimed at reducing educational inequalities in life expectancy across Europe.

This study employed established methods (Mackenbach et al., 2008; Moreno-Betancur et al., 2015) to investigate both relative and absolute inequalities, and offers significant policy insights (Hjorthen et al., 2022; Leão et al., 2018; Mackenbach, 2015; van der Wel et al., 2018). The inclusion of both measures, as recommended by previous studies (Mackenbach and Kunst, 1997; Regidor, 2004), provides a more comprehensive understanding of socio-economic health disparities, and is especially important when the frequency of health problems changes (Regidor, 2004), or in cases in which absolute differences close to zero can be expressed as large relative differences.

Combining survey data from 2015 and 2019 increased the sample size but could have masked year-to-year variations. However, separate year analyses (not shown) revealed no substantial differences in RIIs or SIIs for any behavioural risk factor, except for income-related smoking inequalities among women, where the RII was larger in 2019 than in 2015.

Data on education and income, our primary exposures, were obtained from national registries with objectively derived data for almost the entire population, linked at the individual level to outcome measures. Restricting the sample to adults aged 30–69 ensured that most of the sample had completed their education and were in their prime working years. Household income, representing disposable income after tax and including all income sources, adjusted for household composition, likely provided an accurate reflection of the respondents’ economic situation. Although education and income are linked, they influence health through different mechanisms: education promotes healthier lifestyles as a non-material resource, while income reflects material resources and purchasing power, shaping health through access to health care and social comparisons (Huijts et al., 2010). Accounting for both measures was therefore appropriate.

Unlike previous studies focusing on one measure, we consistently highlighted differences by income and education and incorporated multiple data sources. Inequality was prominent in both the population-wide registry data and the representative sample surveys, reinforcing the value of using surveys in tracking social inequalities in health, especially in the absence of registry data. Despite variations in health measures across data sources, previous studies have shown that self-reported health aligns closely with objective health measures (Miller et al., 2022).

This study also has several limitations. We employed a pooled cross-sectional study design, which only allowed for the investigation of associations between socio-economic characteristics and health-related behaviours and mortality, rather than causal inferences. Moreover, we did not address the possible mechanisms behind the gender differences or the variations between RIIs and SIIs based on education versus income. Given the broad selection of health behaviours that were included, these mechanisms were likely diverse. For mortality estimates, both the RIIs and the SIIs were generally larger among men than women, while behavioural risk factors estimates did not show clear gender differences. Nevertheless, having knowledge about associations provides a solid foundation for further work towards meeting the SDGs. Although our analyses presented a snapshot of the current situation, we acknowledge the importance of examining time trends and believe that our findings can serve as a basis for further evaluations in the years leading up to 2030.

Sample surveys generally exhibit selection biases related to both explanatory and outcome variables, as the most disadvantaged groups tend to be underrepresented due to self-selection (as shown in Table 1). To mitigate this problem, we applied survey weights to account for selection bias. Despite these potential biases, comparing survey data with registry data can be informative, even if the outcome measures differ. Self-reported health has been validated in other studies as a reliable measure of both mental and physical health across European countries, reflecting inequalities in mortality as well as in bodily and mental functions between different social groups (Hamplová et al., 2022; Idler and Benyamini, 1997; Moor et al., 2017). Consequently, self-reported health is considered a useful indicator for assessing socio-economic inequalities in health (Baćak and Ólafsdóttir, 2017), as demonstrated in our study.

Whether our findings apply to other contexts with different levels of affluence and inequality warrants further research. Previous studies have shown that economic hardship can exacerbate health inequalities, especially in countries experiencing both austerity and economic recession (Mackenbach et al., 2018). Additionally, our study was limited to the period following the Great Recession, which had a smaller effect in Norway than in Eastern and Southern European countries due to Norway’s higher level of prosperity and robust welfare schemes (Leão et al., 2018; van der Wel et al., 2018).

Survey data directly capture risk factors like diet and smoking, offering insights that registry data cannot. While we grouped causes of death as smoking-related or amenable to medical intervention, following Mackenbach’s classification (2018), misclassification was unavoidable. For example, some intervention-amenable diseases are behaviour-linked, and smoking-related causes exclude conditions like ischaemic heart disease despite its connection to smoking. Thus, while registry data offer unbiased population-level insights without selection bias, we prioritised survey findings in our assessments of risk factors.

Norway, as a welfare state, provides its citizens with health and welfare services, affordable higher education and income security (Molven and Ferkis, 2011). The Norwegian Public Health Act (Norwegian Ministry of Health and Care, 2011) further strengthened public health regulations. Multiple universal programs, such as vaccination, screening and pre- and postnatal care initiatives, are in place, in addition to regulations aimed at limiting consumption of tobacco and alcohol. Similar egalitarian programs have been implemented in countries like the UK, the Netherlands, Sweden and Finland (Lahelma et al., 2019). Reductions in social inequalities in health have, however, also been observed in countries without such programs (Hu et al., 2016).

Despite Norway’s history of generous health and welfare policies, our findings suggest that these efforts alone do not mitigate avoidable health inequalities. While social security and public services are essential, these universal measures appear insufficient for eliminating socio-economic inequalities in health, particularly in dental health, which is largely unfunded (Hu et al., 2016). Future research should assess inequities in the use of both health and dental services, areas that were not addressed here. As the population ages, ensuring equitable health services across socio-economic groups will become increasingly important for stabilising or reducing socio-economic inequalities in health (Mackenbach et al., 2008; Rydland, 2021; Rydland et al., 2020).

Governments and other stakeholders have recently recognised that while universal public health measures improve overall population health, they may inadvertently increase social inequality in health (cf., for instance, McGill et al., 2015 for systematic review results related to eating habits). This paradox is partly attributed to health literacy, as individuals with higher health literacy are more likely to benefit from universal measures (Berkman et al., 2011). Based on our findings, this appears to be the case in Norway as well. Consequently, the focus is gradually shifting towards employing proportionate approaches or specifically targeting at-risk groups. Proportionate universalism implies designing policies that “act across the whole social gradient, to ‘level up the gradient’ while addressing the needs of those in the most vulnerable and excluded situations” (Goldblatt et al., 2023). This strategy ensures that interventions are proportional to the needs of different groups. In Norway, for example, the national cervical screening program has recently been extended to offer in-home tests for individuals who have not attended screenings in the past 10 years. Targeting at-risk groups means focusing on individuals with higher risk exposure, who are often members of vulnerable groups with shared social characteristics (Frohlich and Potvin, 2008). Consequently, public health interventions that take a proportionate or more targeted approach to reaching at-risk groups might be a valuable addition to efforts to reduce socio-economic inequalities in health in welfare societies where universal measures have been the “gold standard”. Further research into such interventions and their effectiveness for various subgroups is, however, warranted.

Our findings can support the endeavours of the UN and European and national authorities aimed at reducing socio-economic inequalities in health, particularly in relation to the UN’s Sustainable Development Goal 3 (Good Health and Well-being) and Goal 10 (Reduced Inequalities) (United Nations, 2015). Our results suggest that both sample surveys and registry data are useful in the continued effort, as the results are largely comparable across the data sources. Furthermore, the strong agreement between findings across various health outcomes, regardless of whether they are examined in absolute or in relative terms, suggests that not all measures or methods need always be applied in ongoing policy evaluation work.

Evaluations of how current national, supra-national and sub-national policies mitigate health inequalities are scarce. The optimal design and content of such actions remain unclear, as is the counterfactual scenario of how the situation would develop in the absence of such policies. More evidence is needed to determine which policies work – and for whom, when and where they work – to improve population health, reduce inequalities and limit the unnecessary use of public economic and non-economic (e.g., personnel) resources for ineffective interventions. A recently launched large-scale European Joint Action, JA PreventNCD, aims to address inequalities in cancer and other non-communicable diseases by evaluating policies, actions and interventions with a strong focus on inequality (Klepp, 2023).