The theoretical framework of this research is given by Grossman’s demand for health model (1972). Health is an important part of human capital, with higher levels of health implying higher individual productivity. Everyone is born with a certain innate stock of health, which suffers from amortisation, but can be enhanced with health investments. Health investment is a set of actions, including regular doctor visits, following medical prescriptions, conducting a healthy lifestyle with regular physical activities, healthy nutrition, and avoiding harmful addictions like smoking and drinking alcohol. Although health investments require financial resources and time, they also extend the time horizon, during which an individual receives benefits from other components of human capital. This makes health complementary to other forms of human capital and determines one’s ability to acquire other skills (Currie 2005; Currie and Stabile 2006). Moreover, better health translates into the ability to consume more and brings greater earnings potential due to increased individual productivity. Finally, the duration of one’s life is endogenous: when the health stock falls below a critical threshold, life is over. In this framework, non-cognitive skills can affect individual health investments, eventually increasing the health stock and extending the time horizon. Moreover, health is a self-reproducing system where the state of health at each consequent step depends on the previous health investments. The larger the past investments, the more productive it becomes to invest in health in the forthcoming periods (Hai and Heckman 2017).

From the psychological perspective, there are several theories explaining the link between personality and health inequalities. One of them is the Health belief model (Rosenstock 1974), suggesting that if individuals rate themselves as susceptible to a health condition that, in their view, can bring serious consequences, they are more likely to take actions to reduce their risks if these actions are believed to be beneficial. Personality can shape individual beliefs about the potential susceptibility to health risks and benefits from interventions.

Empirical literature investigating the relationship between non-cognitive skills and health generally relies on the well-known psychological concepts. The first concept is the Big Five, suggesting that everyone can be described from the viewpoint of five personality categories: openness to experience (defined as being imaginative and inventive), conscientiousness (comprised of diligence and hard work), extraversion (defined as being sociable and assertive), agreeableness (includes altruism, cooperation, and friendliness), and neuroticism (or the reverse of emotional stability). An alternative measure, frequently used in literature, is locus of control, which reflects the individual tendency to explain various life events with external forces (external locus of control) or with one’s own actions and behaviours (internal locus of control).

The particular attention to the Big Five and locus of control as proxies for non-cognitive skills is motivated by their stability over time. Psychological literature reports a trend increase in neuroticism and a decrease in conscientiousness in older cohorts due to ageing, as well as a decrease in neuroticism, extraversion, openness and an increase in agreeableness and conscientiousness in younger cohorts due to their maturing (Terracciano et al. 2005). Nevertheless, there is a large body of evidence of the predictive power of childhood, along with adolescent measures of personality, on adult outcomes (Conti and Hansman 2013). Personality measured at the age of 30 can be considered as fixed (Terracciano et al. 2006). In economic research, it has been shown that the Big Five and locus of control, measured with short survey instruments, remain stable across one’s working life. Cobb-Clark and Schurer (2012, 2013), using Australian longitudinal household data, showed that the Big Five and locus of control are consistent, at least in the short- and medium-term, and remain unresponsive to major life events. Health shocks have also been shown to have a relatively small and mostly insignificant effect on non-cognitive skills measured by locus of control; more pronounced reverse effects may only appear in older cohorts. (Marsaudon 2022; Cobb-Clark and Schurer 2012).

Although economic research has been mostly relying on the Big Five and locus of control as the core measures of non-cognitive skills, the choice of measure is usually motivated by data availability. For instance, Chiteji (2010) explores the association between future orientation and self-efficacy as measures of non-cognitive skills and health behaviours. Apart from locus of control, Mendolia and Walker (2014) used self-esteem and work ethic to proxy non-cognitive skills. However, such concepts as locus of control, self-efficacy, neuroticism, and self-esteem are closely related and may be measuring the same higher order concept (Judge et al. 2002).

Psychological literature consistently reports conscientiousness to be the most predictive personality trait for health-related outcomes (Friedman et al. 1995; Jokela et al. 2013; Kern and Friedman 2008). Being at the lowest 25% of conscientiousness distribution is associated with a 30 percent increase in preliminary mortality risk compared to the upper 25%, which is equal to a 2 year longer life expectancy (Friedman et al. 1995). However, the effect of traits, particularly conscientiousness, on health behaviours may be stronger for younger adults than for older ones (Bogg and Roberts 2004).

In contrast to the health benefits of conscientiousness, neuroticism is generally associated with the reduced health and higher mortality risks (Savelyev and Tan 2019). Neuroticism is correlated with worse health, especially when it is measured as a construct of psychosocial and subjective aspects (Friedman 2019). High neuroticism combined with high conscientiousness may create a phenomenon of healthy neuroticism, predicting lower mortality risks (Friedman 2000). However, recent empirical findings did not support the existence of this link (Turiano et al. 2020). Evidence concerning the remaining Big Five constructs (i.e., extraversion, openness, agreeableness) is less pronounced.

Conscientiousness, agreeableness, and emotional stability are positively correlated with the overall good self-assessed health, although only neuroticism is significantly associated with an objective measure of the number of sick days (Bellman and Hübler 2022). On PIAAC data (Programme for the International Assessment of Adult Competencies) from Poland, The Big Five categories explain a larger portion of additional variance in self-assessed health (2.6%) than cognitive abilities (0.3%). While conscientiousness and extraversion are positively related to health, neuroticism implies a negative effect (Palczyńska and Świst 2018).

Locus of control is also consistently associated with health outcomes. Externality implies a significantly higher hazard rate ratio for mortality, although for women the effect is either small or insignificant after the adjustment for health behaviours (Lindström and Rosvall 2020; Lindström et al. 2022). Externality is associated with higher chances of dropping out of the labour force after a health shock among men, unrelated to early retirement, and fewer working hours per week, especially among men with lower socioeconomic status (Schurer 2017).

An open question is why non-cognitive skills predict health-related outcomes and whether fostering positive skills will help the improvement of health. The main channels linking personality with longevity and other health outcomes are health behaviours such as nutrition, sleep, smoking, and drinking alcohol. Heckman et al. (2006) showed that locus of control and self-esteem causally affect the probability of smoking and risky behaviours. Physical activity is correlated with extraversion, neuroticism, conscientiousness, and openness (Wilson and Dishman 2015), as well as internal locus of control (Cobb-Clark et al. 2014). External locus of control is associated with higher chances of risky health behaviours (Mendolia and Walker 2014). A positive effect of traits related to emotional stability is observed for the reduction of alcohol consumption and an increase in physical activity (Chiteji 2010). Childhood skills predict adolescent health with external locus of control being negatively associated with smoking (Attanasio et al. 2020). When it comes to unhealthy habits, personality determines individual susceptibility to peer pressure, making less emotionally stable adolescents more likely to start smoking (Hsieh and van Kippersluis 2018). The positive effect of higher conscientiousness on longevity is by 42% mediated by risky habits such as drinking patterns and smoking (Turiano et al. 2015). In a recent context, personality is often associated with COVID-19 precautionary behaviours (higher openness, conscientiousness, neuroticism) (Airaksinen et al. 2021), including vaccination (Roshchina et al. 2022).

Gender differences in the effects also exist. While for men openness is associated with health-harming behaviours, for women the relationship is reversed (Savelyev and Tan 2019). Different explanations can be applied. For instance, while men with internal locus of control expect to have higher rewards from a healthy diet and exercise, internal women get more satisfaction from healthy activities compared to their external counterparts (Cobb-Clark et al. 2014). Programs promoting positive non-cognitive skills in childhood and adolescence may have gender-specific effects, with more pronounced results for boys rather than girls (Conti et al. 2016). Due to gender differences in the effects for various skills, some caution should be taken when choosing the skills to foster (Savelyev 2022).

An alternative pathway from non-cognitive skills to health is education. Conti and Hansman (2013) suggest that non-cognitive skills contribute to the education-health gradient almost as much as cognitive abilities. Non-cognitive skills are a stronger predictor of mortality than cognition (Öhman 2015). On a longitudinal sample of high-ability individuals, Savelyev (2022) shows a strong effect of conscientiousness and extraversion on longevity for men and no effect for women, although little of this relationship can be attributed to education. Approximately one third of the education-health gradient can be explained by non-cognitive skills (Carter et al. 2019). Recent literature suggests that in most health outcomes, non-cognitive skills reduce the education-health gradient by 20-30%, even after taking into consideration sibling-fixed effects (Gørtz and Gensowski 2023).

Part of the relationship also runs through socioeconomic status and income (Öhman, 2015). Stronger associations between non-cognitive skills (both the Big Five and LOC) and health outcomes are observed for low-income earners (Öhman 2015; Musich et al. 2020). Locus of control may as well have a modest but significant effect on the formation of socioeconomic inequality in adult health (Pedron et al. 2021). Policies improving non-cognitive skills may be more important for those with the poorest health (Atkins et al. 2020). However, even after controlling for health behaviours, education, and socioeconomic characteristics, research reports the existence of a direct effect of non-cognitive skills (Hampson et al. 2007; Lodi-Smith et al. 2010).

For Russia, there exists only limited economic research on the determinants of health and longevity. Roshchina (2009), on RLMS-HSE data for years 2000-2005, shows positive returns on private health investment, such as physical activity, avoiding smoking and abstinence, although the analysis does not cover any psychological factors. Denisova (2010), conducting a survival analysis on the same dataset but during a different time period (1994-2007), suggests that excessive alcohol consumption and smoking were prominent health behaviours reducing longevity in the Russian population during the economic transition and early 2000‘s. A recent study, using survival modelling on the same data for years 2004-2019, showed a non-existent effect of marriage for men’s health, a singlehood health penalty for younger women, and a premium for older cohorts (Akhtemzyanov 2023).

The link between non-cognitive skills, health behaviours and health outcomes, including longevity for Russia, remains largely unexplored. The probability of alcohol consumption and the amount of alcohol consumed are significantly affected by the Big Five. While only conscientiousness (a negative relationship) and extraversion (a positive relationship) are related to the probability of alcohol consumption, agreeableness and neuroticism are only related to the volume of consumption. Moreover, non-cognitive skills may mediate the relationship between education and alcohol consumption (Rozhkova et al. 2023). Finally, the Big Five and risk attitudes are demonstrated to significantly affect vaccination hesitancy and refusal during the COVID-19 pandemic (Roshchina et al. 2022). Conscientiousness, openness to experience, internal locus of control, and emotional stability are positively correlated with individual intentions to pursue higher education and the probability to graduate with a university degree (Rozhkova and Roshchin 2021). In the labour market context, a highly significant positive correlation is observed between conscientiousness, openness, emotional stability, and wages (Rozhkova 2019; Gimpelson et al. 2020).

For this study, data from the Russian Longitudinal Monitoring Survey (RLMS-HSE) is used. The survey is conducted annually since 1994 on a large nationally representative sample of approximately 14,000 respondents from 5,000 households, and comprises a wide array of socioeconomic, psychological, and health-related questions. The data are retrieved from rounds 20-30 of the survey which were conducted in 2011-2021. The analysis is performed on a sample for all individuals aged 17-100 years. Additional models calculated on a limited sample aged 18-65 provide similar results. As health behaviours and the effect of non-cognitive skills on longevity might have gender specifics, analysis is performed for men and women separately.

The RLMS-HSE dataset contains several measures of non-cognitive skills. Wave 20, for 2011, provides questions related to individual locus of control. To construct a measure of internal locus of control, an average of 7 behavioural questions, assessed on a scale from 1 to 4, is used. Given the personality stability assumption, the 2011 measure is extrapolated to future rounds if the respondent participated in them. The final measure is standardised with a mean of zero and a standard deviation of one. The full list of questions and a basic distribution of responses are available in Table 1A in the Supplementary materials.

In 2016 (round 24), a new block of 24 behavioural questions was introduced to the survey. These questions can be mapped into 5 categories: 3 questions are related to openness to experience, 7 to conscientiousness, 3 to extraversion, 4 to agreeableness, and 7 to neuroticism. The questions are assessed on a scale from 1 to 4. Each category is constructed as an average of the attributed questions and standardised with a mean of 0 and a standard deviation of 1. The Big Five is used separately from locus of control due to the time mismatch and loss of observations. Also, the longevity analysis with the Big Five only covers 5 years (from 2016 to 2021), while that with locus of control covers 10 years. Although this can be considered as a limitation, the results provide some interesting insights about the relationship between health and personality.

Three different approaches are used to assess the association between non-cognitive skills and health, given the benefits of longitudinal data. We describe them below separately.

Tables 2–5 provide evidence on the association between non-cognitive skills and health self-assessments, controlling and not controlling for health behaviours (full results are available in the Supplementary materials). First, there is a clear association between self-assessed health and health habits in both genders and in both the Big Five and locus of control models. Quantitatively, the most pronounced effects arise between alcohol consumption and self-assessed health, although it is controversial. Alcohol consumption is negatively associated with very bad and bad health self-assessments, but positively with average health. Among men, the predicted probability of bad and very bad health is 9 percent for those drinking alcohol and 16 percent for abstainers. In contrast, probability of average health is 55 percent among non-abstainers and 47 percent among abstainers. It is likely that excessive drinking might have negative effects on health, while moderate consumers, which comprise most of the sample, do not experience any negative influences on health. The observed associations are consistent in both genders. In contrast, smoking is positively associated with the probability of bad health in the female sample, increasing it by 2.7 percentage points, but slightly negatively with very good health for males. In the locus of control models, the effect is more notable with smoking reducing the probability of good health by almost 4 percentage points among males and by 2 percentage points among females. Although it may seem that small size effects contradict the common knowledge about negative health effects of smoking and alcohol consumption, it is important to keep in mind that, first, the observed effects are statistically significant after controlling for a large set of variables, and second, this is subjective health measure rather than objective. When it comes to objective longevity, the size effects are significantly more pronounced (see section “Non-cognitive skills and longevity” of the Results).

Second, there is a positive association between health self-assessment and education on Russian data before and after controlling for health habits, which is observed only on the female sample. Higher education implies a higher probability of good health (approximately 2 percentage points in both the Big Five and locus of control before and after controlling for habits) and a reduced probability of bad (approximately 3 percentage points) and very bad health (approximately 1 percentage point). The size of the effect is slightly reduced after the introduction of health habits.

Third, we observe a significant association between health and non-cognitive skills, proxied with different psychological instruments. From the Big Five, there is a beneficial association between health and conscientiousness, both before and after controlling for habits. Higher conscientiousness is positively associated with a better health assessment and negatively with worse. The size of the effect is only slightly reduced after controlling for physical activity, smoking, and drinking. In contrast, neuroticism is associated with the reduced probability of good health (for the male sample, one standard deviation increase decreases the probability of good health by 4 percentage point, respectively; for the female sample, by 3.7 percentage point) and the increase of bad and very bad health, which is also in line with the literature (Friedman et al. 1995).

The other Big Five categories demonstrate fewer stable results and are less consistent across genders. Extraversion demonstrates an increased probability of good health, a reduced probability of average health, and mixed results for bad and very bad health in the female sample. Openness demonstrates a reduced probability of good health in both genders, but a higher probability of average health. Agreeableness shows a very small positive association with very good health and low significance on the female sample. Finally, internal locus of control is associated with better self-assessed health both before and after controlling for health-related habits. For both genders, a standard deviation increase in internality is associated with a 3-percentage point lower probability of bad health and 4 percentage point higher probability of good health. Figures 1–2 graphically illustrate the observed direct effects of non-cognitive skills in terms of predicted probabilities of various health outcomes.

Figure 1.

Marginal effects, male sample

Figure 2.

Marginal effects, female sample

Tables 6–7 shed further light on health behaviours as possible mechanisms, mediating the effect of non-cognitive skills on health (full results are available in the Supplementary materials). Openness to experience is positively associated with the probability of physical activity for both males (one standard deviation increase in openness raises the probability by 3 percentage points) and females (by 5 percentage points), while extraversion is positively significant only for the male sample. The probability of smoking is reduced by 3 percentage points by conscientiousness in the male sample but increased by 2 percentage point by extraversion in the female sample, and by neuroticism in both samples (by 2 percentage points for males and 1 percentage points for females). With more neurotic individuals, smoking may serve as a coping mechanism for stress, while more extraverted individuals may use smoking as a social activity. Probability of drinking alcohol is reduced by conscientiousness by 2 percentage points and 1 percentage point with agreeableness in the male sample, although the effect is of low statistical significance. In both genders, drinking is positively related to extraversion. Internal locus of control is positively associated with physical activity in both genders (2-3 percentage point increase in the probability) and negatively with smoking in the male sample (4 percentage point reduction), while positively with drinking only for the female sample (3 percentage point increase, respectively).

Finally, Table 8 shows the results of the duration analysis, which accounts for the problem of reversed causality (full results are available in the Supplementary materials). First, health habits demonstrate a statistically significant association with longevity. Although the observed results do not contradict the models with health self-assessment, there are some surprising differences. Physical activity, which was previously statistically significant for health self-assessment both in male and female samples, reduced the risks of mortality for women by 61 percentage points in the Big Five model, and by 43 percentage points in the locus of control model. At the same time, smoking, which did not show large positive correlations with worse health assessments, is highly predictive of the actual mortality. Smoking increases individual risks of mortality by 61 and 87 percentage points for males in the Big Five and locus of control models, respectively, and by 76 and 153 percentage points for females. Smoking was also considered one of the main contributors to adult mortality during the economic transition in Russia (Denisova 2010). Alcohol consumption reduces the mortality risks in the Big Five model for the male sample, but the result is highly unstable. As suggested before, the used measure of alcohol consumption does not differentiate between consumption insensitivity, bringing mixed results.

Second, higher education reduces the risks of mortality in both genders, although in self-assessed health models it was only statistically significant for the female sample. In the locus of control models, higher education reduces the risks by 40 percentage points for males and by 32 percentage points for females before controlling for health behaviours. The effect is slightly reduced after their introduction. In the Big Five models, the size of the effect is similar but higher education loses its significance in the model with health habits. The difference in findings can be explained by a shorter period covered by the Big Five models.

Third, non-cognitive skills demonstrate a significant association with longevity. Conscientiousness from the Big Five appears to be the most consistent characteristic, reducing the probability of health-related mortality in both male and female samples. One standard deviation increase in conscientiousness reduces the risk of death by 20 percentage points for males and by 12 percentage points for females after controlling for health habits. One standard deviation increase of neuroticism increases the risks of mortality by 12 percentage points. At the same time, there is a positive effect of openness to experience for females: one standard deviation increase in openness increases the mortality risk by 13 percentage points. Finally, internal locus of control is associated with the reduced risks of mortality for both men (by 11 percentage points) and women (10 percentage points).