Distinguishing independent and shared effects of material/structural conditions and psychosocial resources on educational inequalities in self-rated health: results from structural equation modelling

Objectives: The aim of this study was to distinguish independent and shared effects of material/structural factors and psychosocial resources in explaining educational inequalities in self-rated health (SRH) by using structural equation modelling. Study design: Cross-sectional survey. Methods: Data were derived from a questionnaire sent to a random sample of the population in ﬁ ve counties in Sweden in 2008. The study population (aged 25 e 75 years) included 15,099 men and 17,883 women. Exploratory structural equation modelling was used to analyse the pathways from educational level to SRH. Results: The pathway including both material/structural factors (e.g. ﬁ nancial buffer and unemployment) and psychosocial resources (e.g. sense of coherence and social participation) explained about 40% of educational differences in SRH for both men and women. The pathways including only the independent effects of psychosocial resources (14% in men and 20% in women) or material/ structural factors (9% and 18%, respectively) explained substantial but smaller proportions of the differences. Conclusions: The major pathway explaining educational inequalities in SRH included both material/ structural factors and psychosocial resources. Therefore, to reduce educational inequalities in SRH, interventions need to address both material/structural conditions and psychosocial resources across educational groups.


Introduction
The persistence of social inequalities in health, also in the highly developed 'welfare states' of Western Europe, has been called one of the great disappointments of public health. 1Social inequalities in health have been found for most measures of socioeconomic status (SES), e.g.education, income and occupation, and for most health outcomes, e.g.mortality, morbidity in several diseases and self-rated health (SRH).2e6 In Sweden, low-education groups have a doubled risk of reporting poor SRH compared with high-education groups, 7 and social inequalities in health have widened over time. 3There are two main explanations for the persisting inequalities: the material/structural and psychosocial pathways. 1he materialist/structural explanation focuses on deficiencies in tangible material goods and structural conditions which are more common in low-SES groups. 8,9The psychosocial pathway implies that inequalities in health are due to an imbalance between exposures to stressors and available protecting psychosocial resources.10e12 It has, however, been shown that the contributions of material/structural factors and psychosocial resources to SES inequalities in health are not mutually exclusive.13e15 Instead, a substantial part of these inequalities is a result of complex pathways with shared contributions, although the relative importance of these shared pathways is not fully understood.15e17 In order to design efficient interventions to reduce SES inequalities in health, it is necessary to understand to what extent the main contributors to the inequalities are independent of each other and to what extent they arise from a shared effect. 18,19See Fig. 1 for a conceptual model.
Few previous studies have distinguished the importance of the independent and the shared contributions of material/structural and psychosocial factors.A study of the Arab minority in Israel showed that almost all the effect of psychosocial resources on educational inequalities in SRH disappeared when material conditions (in terms of financial resources) were taken into account, 14 indicating a shared effect.Another study, using cross-national data from 28 countries in Europe, showed that most of the contribution of psychosocial factors to SES inequalities among employed men and women was shared with material/structural factors. 13Both these studies were included in a systematic review by Moor et al. 15 The review showed that the psychosocial pathway is important for SES inequalities in SRH but that the contribution of psychosocial factors, when taking material/structural factors into account, is much smaller than when the psychosocial factors are analysed separately.Therefore, Moor et al. concluded that 'studies aiming to explain social inequalities in health need to consider more than one pathway.' 15Notably, all studies included in the review were based on logistic regression models, with separate models for the independent effects and the shared effects.To assess the extent to which associations can be explained in terms of independent or shared pathways, it is preferable to analyse all pathways of interest in the same model, and more recent methods, such as structural equation modelling (SEM), have therefore been suggested. 20hus, the aim of this study was, by using SEM, to distinguish and quantify the relative importance of the independent and the shared contributions of material/structural factors and psychosocial resources in explaining educational inequalities in SRH.

Study settings and participants
The present study used data from a cross-sectional postal public health survey, the Life and health study 2008, addressed to a random sample from the population aged 18e84 years in five counties in the central part of Sweden.The population of the five counties consists of more than one million inhabitants, and the sample frame was the total population register at Statistics Sweden, covering all inhabitants in the area.The survey was carried out during MarcheMay 2008.The total sample size was 68,710.Data collection was discontinued after two postal reminders failed to elicit a response, resulting in an overall response rate of 59%.Nonresponse bias can often be a problem when estimating levels or  proportions, but when estimating associations, as in the present study, the non-response bias is normally less severe. 21Therefore, no weighting to handle a potential non-response bias was used.As individuals with postsecondary education rarely reach their final level of education before the age of 25 and the national education register is not complete for those older than 75 years, a subsample of the survey containing men and women aged 25e75 years, comprising 32,982 participants, was selected.Participants gave their informed consent that questionnaire data would be linked to Swedish official registries by Statistics Sweden through personal identification numbers.After the record linkage, all identity information was removed before the material was handed over from Statistics Sweden to the county councils.Study data are protected pursuant to the laws of official statistics and secrecy.More details about the survey can be found elsewhere. 22

Measures
Educational level.Educational level was obtained from the national education register and was classified into low (elementary school), medium (upper secondary school) and high education (at least two years of university or corresponding education).
Self-rated health.SRH was measured on a five-grade scale with the options 'very good', 'good', 'neither good nor poor', 'poor' and 'very poor'.In the present study, the first two options were classified as good, and the rest as poor.
Material/structural factors.Among material factors, poverty and economic hardship are major contributors to SES inequalities in health. 23In the present study, two indicators of economic hardship were included: financial buffer, assessed by a question about whether the respondent in the course of a week was able to raise an amount of 20,000 SEK (approx.EUR 1900), and inability to meet expenses, assessed by a question on difficulties in managing current expenditures for food, rent, bills and so on during the past 12 months.Both these measures have been shown to be associated with SRH in previous studies. 7,24tructural factors extend the material approach to include also social concomitants. 9This concept can relate to the labour market, to the family situation as well as to civic participation.In the present study, three measures of structural factors were included: membership in associations, living in a single household and unemployment.Membership in associations was assessed by the question: "Do you participate in activities or attend meetings in any group, organisation, association or communion?",which was dichotomized (yes/no).Single household was defined as a person living alone and was derived from two questions on family structure.Unemployment (yes/no) was derived from a question on current employment status.As the usual age of retirement in Sweden is 65 years, respondents older than 64 years in most cases reported their main occupation to be retired because of old age.These were given the unemployment status 'no'.Psychosocial resources.A major argument for the hypothesis of psychosocial pathway to SES inequalities in health is that psychosocial resources are, in general, not equally distributed over SES groups. 11External psychosocial resources comprise protective factors in the social environment.Two major domains are social support, 25 comprising emotional and practical support, and social participation, 26 describing being integrated in and interacting with the wider social community.Both social support and social participation are associated with SRH. 24,27In the present study, both social support and social participation were included as measures of external psychosocial resources.Social support was assessed by three questions: 'Do you have any persons in your surroundings from whom you can get support in emotional crises or problems?','… from whom you can get help with shopping and cooking in case of disease?' and '… who could help you if you were moving to a new place?',with response options 'Yes, certainly' (3), 'Yes, probably' (2), 'Probably not' and 'No'.The two latter groups were small and had similar characteristic and were therefore combined into one group (1).Using factor analysis, a variable capturing the common feature of these three questions was created, representing overall social support.Social participation was assessed by the question, "How often do you usually do the following things in your leisure time?" with six items included: 'Watching sports events live', 'Attend a music concert', 'Go to the theatre', 'Go to the cinema', 'Visit exhibition/museum' and 'Take part in study circle or course', all with response options 'Every day', 'Every week', 'Every month', 'Every year', 'More seldom' and 'Never'.
Responding every year or more often on more than two items was considered as social participation.
Internal psychosocial resources are psychological factors representing perceived ability to handle external life strain. 28One of the most studied resources is the concept of sense of coherence, developed by Antonovsky. 29The concept origins from his studies of salutogenesis, i.e. determinants of health and well-being, also in difficult life situations, and comprises feeling of comprehensibility, manageability and meaningfulness.In the present study, sense of coherence was measured using the 13-item questionnaire (SOC-13).Another aspect of psychosocial resources is related to expectancies. 28,30A common measure is optimism, conceptualised as expecting good things rather than bad things to happen in the future, shown to be a good predictor of positive health outcomes. 31,32ptimism has been measured in various ways, including a single question about the way the respondent feels about the future. 33In a meta-review, associations between optimism and physical health were shown to be robust irrespective of type of measure used. 31In addition, the level of optimism has been found to vary over educational levels. 7In the present study, optimism was measured by the question: 'How do you personally look upon the future?' with options: 'Very optimistic' (5), 'Rather optimistic' (4), 'Neither optimistic nor pessimistic' (3), 'Rather pessimistic' (2) and 'Very pessimistic' (1).The variable was used in its original 5-scale form in the analyses.
Present disease or ill-health.As SRH is influenced by manifesting physical and mental health conditions, 34 three measures of disease and ill health were included in the analyses.Present longstanding disease was assessed by the question, 'Do you have any longstanding disease (more than 6 months), permanent ailment from accident, impaired function or other medical disorder?' (No/Yes).Musculoskeletal complaints were derived from the survey question, 'During the last three months, have you experienced any of following complaints or symptoms?' 'Pain in neck or shoulders', 'Pain in back or hips' and 'Pain in hands, arms, legs, knees or feet'.Anxiety/depression was measured by the question in the EQ-5D instrument, asking if the respondent felt 'anxious or depressed' (not/moderately/extremely), coded as 1, 2 and 3, respectively. 35he choice of explanatory factors was mainly based on theoretical and evidence-based considerations but was also restricted to the items covered by the questionnaire.In addition to the included variables, we also tested to include physical living environment, receiving social security benefits, housing tenure and psychosocial living environment in the exploratory process.These variables were however found to be either too closely correlated with other variables or to lack significant exploratory value.

Data analysis
Prevalence poor SRH and categorical measures of material/structural factors and psychosocial resources were calculated by educational level, and differences over educational level groups were tested by Chi-squared test.Sense of coherence was measured on a continuous scale, but the distribution was skewed.Therefore, median and interquartile range were calculated by educational level, and differences were tested by the non-parametric one-way Kruskal-Wallis test.
The pathways from educational level to SRH were modelled in an exploratory manner, using SEM, 36 by including all available measures of material/structural factors and psychosocial resources.The models were adjusted by age, longstanding disease, musculoskeletal complaints and anxiety/depression.All possible associations between the included variables, i.e. educational level, all psychosocial resources, all material/structural factors, adjustment variables described previously and SRH, were evaluated using Pvalues for the standardised path coefficient estimates.Associations with P values > 0.05 were excluded from the models, except for associations involving educational level or age which were kept regardless of P values.The latter was because educational level was the primary factor of interest and because we wanted all analyses to be age-adjusted.
All analyses were stratified by sex, to find out if the relative importance of independent and shared contributions of material/ structural factors and psychosocial resources differs between men and women.Illustrations of the full models, including standardised estimates, can be found in the supplementary material.
The single pathways from educational level to SRH were grouped into one of the four path types illustrated in Fig. 1, with one direct path from educational level to SRH (A) and three indirect path types from educational level to SRH (BeD).The indirect path types were (B) including only material/structural factors, (C) including only psychosocial resources and (D) including both material/structural factors and psychosocial resources (shared effect).For comparability reasons, all effects were standardised.The standardised coefficients (estimated effects) can be interpreted that the more the coefficient deviates from 0, the stronger the effect is.The total standardised effects for all individual pathways that fall into each of the path types (A, B, C and D) were estimated.The relative contributions of each of the four path types were estimated as proportions of the total effect of educational level on SRH, where the educational levels were weighted proportional to size (i.e. the number of respondents in the sample, in the low-education group and in the medium-education group, respectively).
In the models, some of the dependent variables, e.g.SRH, were dichotomous or categorical, which require the use of a probit link in the estimations.As has been recommended in such cases, the weighted least squares mean and variance adjusted estimator was used to estimate the effects in the models. 37s there is no definitive measure of fit for SEM models, a panel of indices have been used to evaluate how well the model reproduces the observed correlation between variables. 38The most common measure is the Chi-squared statistics, where a significant Chi-squared value indicates poor fit.The Chi-squared statistic is, however, highly sample size sensitive, 39 so other measures, including root mean square error approximation (RSMEA), comparative fix index (CFI) and Tucker-Lewis Index (TLI), sometimes also called non normed fit index, were used as complements.Proposed cutoff values for acceptable fit are 0.06 for RSMEA (smaller values indicate good fit) and 0.95 for CFI and TLI (the closer to one, the better). 38The SEM analyses were performed using Mplus, version 7.

Sample characteristics
The proportions of poor SRH in low-education groups were twice those in high-education groups, for both men and women (Table 1).
Material/structural risk factors were more prevalent in loweducation groups, particularly lack of financial buffer.Also, practically all the psychosocial resources showed educational gradients unfavourable to respondents with low education, most pronounced for optimism and social participation.Prevalence of present disease and ill health was higher in groups with low educational levels (Table 1).

SEM analyses
Table 2 presents standardised SEM estimates of effects of educational level on SRH for men and women.The model provided an excellent fit to the data according to the RSMEA, CFI and TLI values.The total standardised effects of low education on poor SRH were b ¼ 0.186 for men and b ¼ 0.193 for women.The total effects of medium education were b ¼ 0.151 and b ¼ 0.133, respectively, all P < 0.001 when compared to high education.For both men and women, the indirect effects of educational level on SRH, explained by psychosocial resources and/or material/structural factors, were larger than the direct effect (Table 2).
Grouping the effects according to the different path types (A, B, C and D) in men, the direct path (A) (b ¼ 0.075, P < 0.001, for low education and b ¼ 0.058, P < 0.001, for medium education) and the indirect pathways with shared effects of both material/structural factors and psychosocial resources (D) (b ¼ 0.080, P < 0.001, for low education and b ¼ 0.053, P < 0.001, for medium education) had the largest standardised effects.Summarising over all educational levels, the direct pathways from educational level to SRH corresponded to 39% (A) and the indirect pathways with shared effects of both material/structural factors and psychosocial resources corresponded to 38% (D), of the total effect for men.The indirect path type involving independent effects of psychosocial resources had a slightly larger effect than the path type only involving independent effects of material/structural factors (C ¼ 14% vs B ¼ 9%) among men (Fig. 2).
In women, the indirect path type involving shared effects of both material/structural factors and psychosocial resources (D) had the largest effect on SRH (b ¼ 0.069, P < 0.001, for low education and b ¼ 0.058, P < 0.001, for medium education) (Table 2), corresponding to 41% of the total effect when summarising over educational groups (Fig. 2).The direct effect of education (A) and the independent indirect effects of material/structural factors (B) and psychosocial resources (C) all explained around 20% each of the total educational effect.

Discussion
The results of the present study show that the shared path type including both material/structural factors and psychosocial resources explained the largest proportion of the educational inequalities in SRH, about 40%, among both men and women.In addition, among men, the independent effect of psychosocial resources explained 14% and the independent effect of material/ structural factors explained 9%, whereas in women, these two path types explained around 20% each of the total educational effect.The remainder of the educational effect on SRH (the direct effect) was large (39%) in men but smaller (22%) in women.
The finding shared effect of material/structural factors and psychosocial resources was more important than the independent effects is in line with the conclusion of the review by Moor et al., 15 where the interdependence of these factors was emphasised.Our finding supports and elaborates their conclusion, as we were able to quantify the relative importance of the independent and the shared contributions.However, among the independent effects, Moor et al. found that material factors had the strongest effect. 15In the present study, the independent effects of material/ structural factors and psychosocial resources were about the same size in women, whereas the psychosocial resources had a somewhat stronger effect among men.There are several possible explanations for these diverging findings.First, of the studies included in the review, which involved both material/structural and psychosocial factors, only one study included measures of both external and internal psychosocial resources.The other studies included only external psychosocial resources.Both internal resources, such as sense of coherence and optimism, and external resources, such as social support, are important buffers to external stressors. 40In the present study, we examined several measures of both internal and external psychosocial resources.Our finding that the effects of psychosocial resources were strong is in line with the literature, e.g. the study by Marmot and Wilkinson 41 on the fundamental importance of psychosocial factors for SES differences in health and their relationship with both mental and somatic disease and ill health.The effects of psychosocial resources can also be traced by psychobiological mechanisms and disease vulnerability. 11lso, the different results may be due to the fact that the review was based on studies using logistic regressions, where a series of regression models were compared. 42Such standard regression approach can be problematic and has been shown to be inferior to SEM when performing analysis of relative contribution of different pathways. 43,44The problem is even aggravated when logistic regression models are used because odds ratios from different logistic regression models are not fully comparable. 45Accordingly, the authors of the review concluded that 'there are more recent methods regarding mediation analysis which none of the included studies used'. 15hus, a particular strength of the present study is the use of SEM, especially the path analysis component.This analytical method has been suggested as a useful tool in modelling the different pathways linking social structure to health because the method is able to structure several relations sequential system, which goes beyond the use of interaction terms in common logistic regression models. 20Furthermore, SEM can handle multiple pathways with both independent and shared effects in a single model. 46To our knowledge, no other studies have applied this tool for distinguishing independent and shared effects of psychosocial and material/structural factors on educational inequalities in health.A further strength of the study is that material/structural factors were not confined to financial resources only but also included structural living conditions, which are important when studying SES inequalities in health. 15lthough it has merits, we are aware of several limitations.The survey data used are from 2008.However, previous studies have shown that educational health inequalities in Sweden are persistent or even widening 3,47 and that the impact of the determinants of the inequalities has been stable over time. 7The non-response rate was about 40%.As people with poor health and/or low education often have lower response rates, this may lead to underestimation of health inequalities in our findings.Also, as the results may depend on the measures used, further studies are needed to examine the robustness of our findings. 48However, the fact that the measures used in the present study explained the majority of the educational differences in SRH, especially among women, indicates that the set of variables used in the present study catches a large part of important material/structural factors and psychosocial resources.
Levels of financial security among disadvantaged groups are higher in Sweden than in many other countries, and thus, the results may not be generalisable to an international context.However, as can be seen from the present data, also in this Swedish population, a substantial proportion was lacking financial buffer and/or had problems with meeting the expenses.Moreover, the results concern SRH, and generalisation to other health outcomes may not be appropriate.As a consequence of the cross-sectional design, it was not possible to conduct a straight-forward mediation analysis, i.e. to determine how one factor influences the  outcome through variable.Thus, we cannot fully assess whether the contribution of psychosocial resources was a function of material/structural factors or if the effects of material/structural factors were functions of psychosocial resources.Therefore, future longitudinal studies would be a valuable complement to existing knowledge.However, we can conclude that the shared effect had the largest contribution to the inequalities in the present study.
The results of the present study reflect the common pattern for all respondents aged 25e75 years.An interesting topic for future studies would be to investigate whether the contributions of material/structural factors and psychosocial resources to health inequalities vary by age groups.Another interesting topic might be to investigate the working population using SEM analysis, which would enable to add physical and psychosocial working conditions to the list of explanatory variables.This would further elucidate the contributions of material/structural factors and psychosocial factors to educational differences in SRH in this subgroup.Finally, as our analysis only included material/structural and psychosocial factors, the contribution of other types of factors, e.g.health behaviours, was implicitly included in either the direct effect from educational level to SRH or in the effects involving psychosocial resources. 49This may explain why the direct effect of education was substantial, especially among men.Behavioural factors are important to explain SES inequalities in health, but as the primary purpose of the present study was to compare the relative importance of the independent and the shared contributions of material/ structural factors and psychosocial resources in explaining educational inequalities in SRH, we chose to focus on these two types of factors to simplify an already complex network of associations.Our finding of the large size of the shared effect of material/structural factors and psychosocial resources is in line with policy strategies launched decades ago by WHO. 50Thus, interventions to reduce inequalities in health need to include policies to improve not only economic resources for the poorest 51 but also interventions supporting psychosocial resources.Such interventions should include empowerment strategies and development of health-promoting living arenas in neighbourhoods, at workplaces and in schools, to enable social support, trust, hope and optimism.
In summary, we found that the major contribution of material/ structural factors and psychosocial resources to educational inequalities in SRH was in the form of shared effects of these factors.Measures to reduce educational inequalities in SRH need therefore to address both material/structural conditions and psychosocial resources across educational groups.

Fig. 1 .
Fig.1.Conceptual model of pathways from educational level to self-rated health, including the independent effects of material/structural factors and psychosocial resources, respectively, and the shared effects of material/structural factors and psychosocial resources.

Fig. 2 .
Fig. 2. Standardised total effect of educational level on self-rated health by different pathways (%).A, direct effect (involving neither material/structural factors nor psychosocial resources); B, indirect effect by material/structural factors; C, indirect effect by psychosocial resources; D, shared indirect effects by both material/structural factors and psychosocial resources.

Table 1
Descriptive statistics of age, educational level, self-rated health (SRH) and prevalence of poor SRH, material/structural factors and psychosocial resources by educational level (Life & health study 2008).

Table 2
Standardised effects of educational level on poor self-rated health and effects of material/structural factors and psychosocial resources based on structural equation modelling, men (N ¼ 15,099) and women (N ¼ 17,883) aged 25e75 years.
RSMEA, root mean square error approximation; CFI, comparative fix index; TLI, Tucker-Lewis Index.Note: Indirect effects are calculated by taking the product of all paths leading from the predictor to the outcome.Total effects are the sum of direct and all indirect effects.The models are adjusted for age, longstanding disease, musculoskeletal complaints and anxiety/depression.aThe reference category for educational level is high education.b Material/structural factors include financial buffer, inability to meet expenses, membership in associations, single households and unemployment.c Psychosocial resources include sense of coherence, optimism, social support and social participation.d Values smaller than 0.06 indicate acceptable fit (Hu and Bentler, 1999).e Values larger than 0.95 indicate acceptable fit (Hu and Bentler, 1999).