Article Text

Download PDFPDF

Psychosocial and behavioural factors in the explanation of socioeconomic inequalities in adolescent health: a multilevel analysis in 28 European and North American countries
  1. Irene Moor1,
  2. Katharina Rathmann1,
  3. Karien Stronks2,
  4. Kate Levin3,
  5. Jacob Spallek4,
  6. Matthias Richter1
  1. 1Institute of Medical Sociology, Medical Faculty, Martin-Luther University Halle-Wittenberg, Halle (Saale), Germany
  2. 2Department of Public Health, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands
  3. 3Child and Adolescent Health Research Unit (CAHRU), University of St Andrews, St Andrews, UK
  4. 4Department of Epidemiology & International Public Health, School of Public Health, University of Bielefeld, Bielefeld, Germany
  1. Correspondence to Irene Moor, Institute of Medical Sociology, Medical Faculty, Martin-Luther-University Halle-Wittenberg, Halle (Saale), 06112, Germany; irene.moor{at}


Background The relative contribution of different pathways leading to health inequalities in adolescence was rarely investigated, especially in a cross-national perspective. The aim of the study is to analyse the contribution of psychosocial and behavioural factors in the explanation of inequalities in adolescent self-rated health (SRH) by family wealth in 28 countries.

Methods This study was based on the international WHO ‘Health Behaviour in School-aged Children’ (HBSC) study carried out in 2005/2006. The total sample included 117 460 adolescents aged 11–15 in 28 European and North American countries. Socioeconomic position was measured using the Family Affluence Scale (FAS). Multilevel logistic regression models were conducted to analyse the direct (independent) and indirect contribution of psychosocial and behavioural factors on SRH.

Results Across all countries, adolescents from low affluent families had a higher risk of reporting fair/poor SRH (OR1.76, CI 1.69 to 1.84). Separate adjustments for psychosocial and behavioural factors reduced the OR of students with low family affluence by 39% (psychosocial) and 22% (behavioural). Together, both approaches explained about 50–60% of inequalities by family affluence in adolescent SRH. Separate analyses showed that relationship to father and academic achievement (psychosocial factors) as well as physical activity and consumption of fruits/vegetables (behavioural factors) were the most important factors in explaining inequalities in SRH.

Conclusions More than half of the inequalities by family affluence in adolescent SRH were explained by an unequal distribution of psychosocial and behavioural factors. Combining both approaches showed that the contribution of psychosocial factors was higher due to their direct (independent) and indirect impact through behavioural factors.

  • Adolescents Cg
  • Health Behaviour
  • HEALTH Inequalities
  • Psychosocial Factors
  • Social Epidemiology

Statistics from


It is well established that social determinants have a profound impact on health.1 ,2 An abundant body of research has shown that individuals from lower socioeconomic groups are faced with an increased risk of morbidity and mortality. Material factors, health-related behaviours and psychosocial determinants are identified as key pathways in the explanation of socioeconomic inequalities in adult health.3 It is assumed that health inequalities can be explained by an unequal distribution of these factors across socioeconomic groups as they appear to both independently influence the social gradient in health, while also being inter-related to one another.4 Most previous studies focused on inequalities in health by educational or occupational level. These studies showed that these health inequalities are the result of a complex interplay between material, behavioural and psychosocial mechanisms. For instance, health behaviour is embedded in social contexts and thus depending on the living conditions and psychosocial resources, whereas psychosocial factors, besides having a direct impact on health, also operate through behavioural factors, for example, coping with psychosocial hazards in social relationships may lead to harmful health behaviour.5 Behavioural and psychosocial factors should therefore be considered simultaneously in analyses to unravel their contribution for health inequalities in low-affluent families. Stronks et al6 were probably the first who empirically assessed the relative importance of different explanatory approaches by combining structural and behavioural factors in order to explain inequalities in health among adults. Other studies followed that shed some light, for instance, in Finland,7 ,8 Germany,9 Israel,10 ,11 Korea12 and the Netherlands.13 ,14 The majority of these studies indicate that material living conditions play an important role for explaining health inequalities, by widely working through psychosocial and behavioural factors.

Even in adolescence, health inequalities exist and persist over the past decades.15–17 Several studies have argued that psychosocial and behavioural factors may also account for social inequalities in health in young people.18 ,19 Inequalities in adolescent health are shaped by the conditions and the context in which young people grow up.20 For example, a wide range of behavioural factors are associated with health inequalities in adolescence19 ,21 and also in adulthood.6 ,22 There is also some evidence on the health-damaging impact of various psychosocial aspects of the socialising contexts of young people23 such as school,24 peer group15 ,25 and family.26 ,27 However, not much is known about their contribution and relative importance in the explanation of inequalities in adolescent health. An explorative analysis in Germany showed that adolescent health behaviour and psychosocial factors from family, peer and school context largely contributed to explain inequalities in low-affluent families for adolescent health.28

While for adults, previous studies have initiated to identify and quantify the impact of mediating factors on socioeconomic inequalities in health in cross-national studies29; such research efforts have not been directed towards adolescents. Identifying mediating factors during adolescence is of particular relevance as this is a stage in life when people become independent and adopt many behaviours which pose risk to and benefit health, often tracking into adulthood.3 The main objective of this study is to explore how material disadvantage in the family is working through psychosocial and behavioural factors using a cross-national perspective. Thus, the aims of the present study are (1) to investigate the extent of inequalities in self-rated health (SRH) by family wealth in adolescents among 28 European and North American countries, (2) to assess the relative contribution of psychosocial and behavioural factors as well as (3) their shared impact (direct and indirect contribution) to the explanation of inequalities in adolescent SRH by family wealth. The theoretical model is based on Stronks et al6 explanatory approach about how socioeconomic inequalities in health are mediated through behavioural and psychosocial factors taking the direct as well as the shared effects into account (figure 1).

Figure 1

Conceptual model for explaining social inequalities in health (modified after Stronks et al6).

Materials and methods

Study population

The WHO Health Behaviour in School-aged Children (HBSC) study is a cross-sectional survey carried out every 4 years since 1982 in a growing number of countries. The aim of the study is to increase the understanding of adolescent health and health behaviours. The seventh wave of the survey was conducted in 2005/2006 in 41 countries in Europe, North America and Israel. The HBSC study is a school-based survey collecting data from 11, 13 and 15-year-old students through self-completion questionnaires carried out in the classroom based on an internationally agreed protocol. A detailed description of the aims and theoretical framework of the study can be found elsewhere.15 Students were selected using a clustered sampling design, where the initial sampling unit was the school class. In the present study, 117 460 students were included in the analyses. Data from England, Scotland and Wales were merged to represent the UK, as were data from the French-speaking and Flemish-speaking parts of Belgium to represent Belgium. Further, USA, Austria, Malta and Greenland were excluded from the analyses because of high missing values in the key variables, or due to small sample sizes (Greenland and Malta: N<500). From the remaining 35 countries, 28 countries show significant inequalities in SRH by family affluence (OR>1) and were therefore included in the analyses. In total, N=87 540 were excluded, due to missing values (N=44 372) and excluding countries (N=43 168). We found differences in missing values by age (less 11-year-olds), gender (less boys) and Family Affluence Scale (FAS; less low-affluent students). However, the differences in the distribution were small.

Instruments and variables

Self-rated health

SRH was measured by asking the students “Would you say your health is…” with answer categories of (1) excellent; (2) good; (3) fair; (4) poor. The response options were dichotomised into excellent/good versus fair/poor.30 SRH is an indicator of subjective health, which has been widely used in public health research.31 The feasibility and psychometric robustness of this indicator have been demonstrated among adolescents in international studies.32 Furthermore, there is evidence for the relatively stable construct of SRH in longitudinal studies during adolescence and the correlation with a variety of health outcomes in adolescent well-being and health-compromising behaviour.31 ,33 Thus, SRH is considered as a valid measure among adolescents.33

Socioeconomic position

In studies based on self-reports from children and adolescents, many respondents are unable to provide information about parental occupation, education or income, resulting in high levels of missing data.34 ‘FAS’ was developed as a proxy indicator of socioeconomic position for adolescents based on the material assets or conditions of the family household including four simple items.15 FAS items include: family car (0, 1, 2 or more), own bedroom (no=0, yes=1), family holidays during the past 12 months (0, 1, 2, 3 or more) and family computer (0, 1, 2, 3 or more). For holidays and computer, the two highest response categories (2, 3 or more) were combined. The FAS score ranged from 0 (low affluence) to 9 (high affluence). A composite FAS score was calculated by summing up the responses to these four items and recoded into country-specific FAS tertiles (low, medium and high family affluence). Using FAS tertiles instead of a continuous measure gives an indication of relative inequalities between the highest and lowest affluent groups. Using family affluence as a standardised continuous variable may be preferable if the gradient or slope of the relationship is of interest or if a single parameter is preferred. However, this assumes a consistent relationship across the range with a unit change having a similar effect at the outer ends as the middle, when in fact this may not be the case. The validity of FAS has been addressed by several studies. Currie et al34 and Molcho et al35 also showed that FAS revealed a moderate internal reliability. The FAS, a measure of material wealth, was interpreted as a proxy for socioeconomic position for the purposes of this study.

Psychosocial and behavioural factors

In table 1, psychosocial and behavioural factors were listed in order to present the operationalisation of the variables included.

Table 1

Operationalisation of study variables on psychosocial and behavioural factors

Statistical analyses

Multilevel analyses were performed to account for the hierarchically clustered structure of the data (students nested in countries).36 The level 1 units are the individual students, and the level 2 units are the 28 countries included in the study. The multilevel analyses allowed the regression constant (intercept) to vary between level 2 units (here: countries). Cases with missing values were excluded from the analyses. There was no significant interaction between family affluence and gender as well as between family affluence and age in relation to SRH. Therefore, we conducted the analyses for the pooled sample, by controlling for gender and age. To analyse the association between FAS and SRH, logistic regression models were conducted for each country separately. ORs for family affluence with 95% CIs were calculated, using the highest tertile of family affluence as the reference category. Furthermore, those mediating factors which were significantly associated with SRH (logistic regression), and had a negative association with family affluence (bivariate analysis, χ2 test) were selected for explanatory analyses. We set up the significance level to <0.00 001 due to our large-N data set.29 Drawing on prior analyses among adults14 ,29 ,37 and recently for adolescents,28 different models—with identical sample size—were calculated. The reference model (model 1) consists of ORs for fair/poor SRH by family affluence, adjusted for age and gender only. Single explanatory factors from each psychosocial and behavioural variable set were first added separately (model 1+single variable) and in a second step as one block for psychosocial factors (model 2) and behavioural factors (model 3). In the third step, we included psychosocial and behavioural factors in the model, simultaneously (model 4). For each model, we calculated the percentage change in ORs for SRH in the different family affluence groups due to the addition of the correlates (((OR(model 1)−OR(model 2–4))/(OR(model 1)−1))×100)6 ,12 ,14 ,28 ,37 Although some variables contributed only marginally to the explanation of health inequalities (eg, only 1%) as a single variable, as a group they accounted for about 5–10% of the observed inequalities (behaviour or psychosocial factor block). Therefore, we decided to include these variables in the models.19 ,38 Single variables that did not contribute to the reduction of the OR were excluded from the analysis in models 2–4. By comparing models 2–4, we were able to quantify the direct and indirect contributions of psychosocial and behavioural factors to the explanation of social inequalities in SRH. According to a power analysis with the software G*Power,39 we calculated effect sizes of the expected ORs. Based on the sample of 117 640 adolescents, ORs of around 0.974 can be expected in logistic regression models, when a fixed α level of 0.05, R² of 0.2 and a desired power (1–β) of 0.9 are considered. The statistical analysis was conducted using the software STATA V.12.


Table 2 shows the sample size of the study population for family affluence and SRH. Across all countries, the majority of adolescents assessed their health as excellent/good, whereas on average 16% reported fair/poor health. Owing to the operationalisation of FAS into tertiles, 27% of students were categorised as low-affluent, 38% as medium-affluent and about 34% as high-affluent students. Significant inequalities in SRH by family affluence were found with an OR of 1.23 (medium FAS) and an OR of 1.76 (low FAS) compared with high-affluent students based on multilevel regression models (average). The intraclass correlation coefficient in the null model (not presented) indicated that about 4.6% of the variation within adolescents’ SRH is attributable to differences between countries.

Table 2

Characteristics of the study population by SRH, family affluence (FAS) and age for boys (n=54 907) and girls (n=62 553), as well as (multilevel) logistic regression models for family affluence and SRH, separately by country and in total (n=117 460)

In table 3, the associations between fair/poor SRH and psychosocial and behavioural factors, respectively, as well as with family affluence are shown. Almost all psychosocial and behavioural factors were unequally distributed with FAS and were also significantly related to SRH. In total, 11 of the behavioural factors were therefore included in the (multilevel) logistic regression models, with the exception of consumption of sweets (Table 4). Nine of the 10 psychosocial factors (excluded: school-related stress) met the requirements of a significant association with SRH as well as with family affluence and were also included in the multivariate analysis.

Table 3

Separate logistic models for fair/poor SRH (ORs, 99% CI)* and prevalence rates by family affluence for psychosocial and behavioural factors (n=117 460)

Table 4

Multilevel regression models (ORs and 99% CIs) of fair/poor self-rated health by family affluence, crude and adjusted for psychosocial and behavioural factors* (n=117 460)

Contribution of behavioural and psychosocial factors in explaining inequalities in SRH

Table 4 shows the multilevel logistic regression models for the separate and combined analyses of psychosocial and behavioural factors in order to explain inequalities in SRH. The analysis of each single psychosocial variable revealed that relationship to father and academic achievement had the highest contribution (about 14–17% reduction in OR) to mediate the association between family affluence and SRH. With regard to behavioural factors, physical activities as well as consumption of fruits and vegetables were the strongest determinants of health inequalities (7–13%). The contribution of all other single variables was smaller. In contrast, school-related stress (psychosocial factor), consumption of soft drinks, playing computer games or PC using (behavioural factors) did not contribute to social inequalities in health and were therefore not considered in models 2–4.

In model 2, the whole group of psychosocial factors was taken into account and resulted in a reduction of the OR from 1.76 to 1.46 (low FAS), corresponding to a relative contribution of 39% (model 2) in explaining health inequalities. For medium FAS, the same reduction was found (39%). Behavioural factors contribute to 22% (low FAS) and to 26% (medium FAS) to the explanation of inequalities in SRH. Adjustment for psychosocial and behavioural factors together explained more than half of social inequalities in adolescent SRH with 53% for low FAS (OR 1.76 to 1.36) and 57% for medium FAS (OR 1.23 to 1.10). However, approximately 40% of health inequalities could not be explained by the factors under study.

Direct and indirect contribution of behavioural and psychosocial factors

However, the separate analyses do not provide the ‘real’ contribution of each group of factors as they are likely to be inter-related. Therefore, we estimated the direct (independent) and indirect effects of psychosocial and behavioural factors by separate and joint adjustment (model 2–4). Combining psychosocial factors (model 2) and behavioural factors (model 3), the OR for low FAS decreased to 53%, indicating a 14% additional reduction of the OR compared with the contribution of psychosocial determinants only (model 4−model 2: 53%−39%=14%). This additional reduction can be defined as the independent effect of behavioural factors (see also figure 2). The remaining contribution of behavioural factors (model 3) reflects the impact of psychosocial factors which mediate the relationship through behavioural factors (model 3−behavioural factors independent: 22%−14%=8%). The independent impact of psychosocial factors can be calculated congruently (model 2−indirect contribution of psychosocial factors: 39%−8%=31%), representing a direct reduction of 31% for students with low FAS. For medium FAS, the results were similar (figure 2).

Figure 2

Independent and direct effects of psychosocial and behavioural factors to inequalities of fair/poor self-rated health (n=117 460; FAS, Family Affluence Scale).


So far, only few studies investigated the relative contribution of different pathways and mechanisms leading to health inequalities in adolescence. This is to our knowledge the first study investigating the role of psychosocial and behavioural factors for explaining inequalities in adolescent health by family wealth using a cross-national perspective. Our findings identified clear inequalities in adolescent SRH in 28 countries. Furthermore, we could show that psychosocial and behavioural factors are unequally distributed across different family affluence groups and have a substantial impact on health, already in this young age group. Our results indicate that psychosocial and behavioural factors account for more than half of inequalities in health by family affluence. This study contributes to the current literature explaining inequalities in adolescent health. There have been several studies which attempt to explain health inequalities in adulthood7 ,10 ,14 and few which consider adolescence.19 ,28 However, relatively little empirical evidence was provided regarding the main pathways leading to socioeconomic inequalities in adolescent health in a cross-national perspective while also considering the shared contribution of these factors. Furthermore, we could also provide a deeper insight into which single factors were of greatest importance. Assessing the contribution of both factors simultaneously, it was possible to unravel how psychosocial factors operate through behavioural factors. Therefore, the contribution of psychosocial factors was higher due to their independent (direct) and indirect impact through behavioural factors.

Comparison with previous research

Our study adds further evidence to previous research which identified socioeconomic inequalities in adolescent health.15 ,40 ,41 In addition, we found social inequalities for almost all mediating factors. This association is known from other studies among adults 5 ,14 ,29 and also in adolescence 15 ,21 ,42–44 showing that people with lower socioeconomic position are more likely faced with less psychosocial resources and show more often detrimental health behaviour. These mechanisms are identified as key pathways for the link between social status and health as prior studies have demonstrated in adulthood.10–12 ,14 ,37 In regard to the life course perspective, increased attention should be paid to the impact of social determinants on health in adolescence not least because it is a key developmental life stage,20 where healthy and unhealthy habits are formed, risk behaviours are adopted and independent choices are made.22 From a life course perspective it is assumed that socioeconomic determinants have an essential impact on health throughout the life course 45 and that socioeconomic position in early life has an independent and sometimes larger effect on health than adult socioeconomic position.22 ,46 ,47 Health and health behaviour established in adolescence are strongly linked into adult life, therefore identifying underlying determinants and mechanisms is important.20 So far, there has been little research on how socioeconomic status or socioeconomic disadvantage works through psychosocial and behavioural factors and thus contribute to the explanation of health inequalities.

Pathways linking family affluence to adolescent health

The pathways through which socioeconomic status results in health inequalities in adolescence include differential exposure to risk factors and health behaviours, as well as differential individual psychosocial resources.18 The question of how family affluence gets ‘under the skin’ and operates through psychosocial and behavioural factors is complex. The psychosocial and behavioural factors examined in this study reveal that those single factors related to the family background which was relationship to parents, especially to the father as well as academic achievement, have the largest mediating effect. Parents in less-affluent strata are more often faced with fewer material and psychological resources, which can have a negative impact on the quality of the marital, as well as the parent–child relationship.48 Moreover, fewer resources available to low-affluent parents may limit their ability to support their children in schoolwork, thus might have negative consequences for their academic achievement and satisfaction with school.49 Academic achievement or educational aspirations could be interpreted as an indicator of adolescents’ (future) socioeconomic position. Adolescents own educational aspirations or educational track have been shown to have a strong relationship to health and health behaviour.19 ,50 Of the behavioural factors examined, it seems that psychosocial hazards only marginally lead to harmful behaviour, so that the hypotheses of Harwood et al5 need to be reconsidered for explaining inequalities by family affluence in adolescent SRH. Instead those health-related behaviours learned early in life in the family context, such as healthy nutrition (consumption of fruit and vegetables) and physical activity, appear to be more important than peer-related factors such as smoking or alcohol consumption. This is in line with the equalisation hypothesis of West51 suggesting that effects of the peer group and youth culture cut across the influence of family socioeconomic background, leading to less inequalities in health in young people. This seems to play a role for substance behaviour which is highly influenced by peers. In summary, our results support the idea that material disadvantage affects the health of adolescents through a mechanism of limited resources.52

Comparable studies among adolescents are still limited. Some few studies have focused on single pathways, such as on the contribution of psychosocial factors53 or behavioural factors in the relationship of social inequalities in adolescent health; Richter et al42 showed that behavioural factors contributed to 24% of inequalities in SRH among school children in 33 European and North American countries, although the strength of this contribution varied across countries.42 Regarding the importance of behavioural factors for explaining health inequalities throughout the life course, Van de Mheen et al22 found that behavioural factors explain the relationship between socioeconomic status and adult health for approximately 10%. However, to the best of our knowledge, only one German study has investigated the relative contribution of multiple pathways including psychosocial and behavioural factors in socioeconomic differences in adolescent health. The authors identified that separate analyses of different key explanatory factors could explain about 30–50% in adolescent SRH and 80–100% taking all pathways into account.28 Although the current study explained a large part of social inequalities in health, about 40–50% of the inequalities remained unexplained. We encourage further studies to investigate the impact of multiple pathways using different variables to shed more light on this complex association.

Methodological considerations

The main strength of the study is the international data set for adolescents aged 11–15 with large sample sizes (>110 000) among 28 European and North American countries. In addition, a wide range of behavioural and psychosocial factors of high relevance in adolescent life could be taken into account.

However, there are several limitations. Because of the cross-sectional design of HBSC, our study is limited in terms of its potential to establish causal relationships. We assumed that psychosocial and behavioural factors mediate the relationship between family affluence and SRH. The causal direction could also be vice versa. For example, low SRH may lead to higher rates of risk behaviour. It is, however, rather unusual that adolescent SRH might influence material conditions of the family. There may however be other confounding factors not adjusted for here that could influence both. Longitudinal studies are needed in order to gain more information about the relevance of these assumptions.

Moreover, it should be acknowledged that the FAS is only a proxy for adolescents’ socioeconomic status. In contrast to parental education and occupation, family affluence is much more related to material wealth. Interpretation was therefore based on how socioeconomic situation, or rather material deprivation, impacted SRH through psychosocial and behavioural factors. However, we are confident of the robustness of the study findings as similar associations have been observed using other socioeconomic indicators such as parental occupation.28 Furthermore, by excluding missing values, we found slight differences in missing values by age, gender and FAS, including slightly more girls, 15-year-olds and students with higher FAS in the analyses. However, we do not expect that the main conclusions of the study are strongly biased. As we included less low affluent students, we rather underestimate the extent of health inequalities.

The relative contribution of psychosocial and behavioural factors also depends on the pathways and variables selected for the analysis. Although taking into account important approaches based on empirical evidence for adulthood and adolescence, the inclusion of other explanatory mechanisms such as biomedical or cultural factors may yield different estimates for the contribution of psychosocial and behavioural determinants. However, as previous studies indicate, biomedical37 or cultural10 factors tend to be less important in explaining socioeconomic inequalities in health, we do not expect great differences in the contribution. Furthermore, it is important to acknowledge that the relative contribution of psychosocial and behavioural factors should be considered as approximate measures of importance rather than ‘absolute parameters’.14


Our study adds new evidence on behavioural and psychosocial factors that mediate the relationship between social position and adolescent health across European and North American countries. Understanding the mechanisms responsible for socioeconomic differences in adolescent health gives us the opportunity to develop effective strategies to tackle these inequalities. Policies intending to reduce health inequalities should take multiple pathways into account, as the explanatory factors are not mutually exclusive. So far, priority has been given to address health inequalities by trying to change health behaviours.54 According to our finding, interventions which emphasise behavioural factors can only be partly successful. While investment in health behaviour is of course important, the wider social, economic and psychosocial determinants should not be neglected. Since health behaviours reflect underlying inequalities in material and psychosocial resources, tackling these inequalities could only be effective by taking the wider determinants of health inequalities into account.55

What is already known on this subject

  • Social inequalities in health are found for various European countries in adulthood and adolescence.

  • Studies investigating explanations for social inequalities in health identified psychosocial and behavioural factors as key pathways in adulthood.

  • So far, still little evidence is available on the pathways for explaining social inequalities in adolescent health, especially using cross-national data.

What this paper adds

  • The study is among the first to systematically examine the contribution of psychosocial and behavioural factors for explaining health inequalities in adolescence by family wealth using a cross-national perspective.

  • Socioeconomic inequalities in adolescent self-rated health were found in 28 European and North American countries.

  • Psychosocial and behavioural factors explained approximately 20–40% of health inequalities by family affluence for each factor group separately and about 50–60% altogether.

  • Our study highlights the need for improving the wider determinants of health in adolescence in order to reduce current and future health inequalities among young people across European and North American countries.


The Health Behaviour in School-aged Children (HBSC) study is an international survey conducted in collaboration with the WHO Regional Office for Europe. The current International Coordinator of the study is Candace Currie, CAHRU, University of St Andrews, Scotland. The data bank manager is Oddrun Samdal, University of Bergen, Norway. The data collection in each country was funded at the national level. We are grateful for the financial support offered by the various government ministries, research foundations and other funding bodies in the participating countries and regions.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Contributors IM developed the study idea, led the writing and the interpretation of the data and wrote the first draft of the article. KR performed the statistical analyses and provided critical revisions on several drafts. MR made substantial contributions to the conception of the study, supervised the data analyses and provided critical comments on the manuscript. KR, KS, KL, JS and MR assisted with data interpretation and participated in reviewing and revising the article. All authors read and approved the final manuscript.

  • Ethics approval Ethical approval for each national survey was obtained according to the national guidance and regulation in place at the time of data collection.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.