Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Multiple measures of socio-economic position and psychosocial health: proximal and distal measures

Multiple measures of socio-economic position and psychosocial health: proximal and distal measures Abstract Background The aim of this paper is to compare three models for exploring the links between different measures of adult socioeconomic position (SEP)—education, occupation, income—and psychosocial health. Model I is a basic univariate regression model with psychosocial health as the outcome and a measure of SEP as the predictor. Model II is a multiple regression model with psychosocial health as the outcome with all three measures of SEP allocated the same temporal position as predictors. Model III treats education, a distal measure of SEP, as antecedent to the proximal measures of SEP in the prediction equations linking SEP to health. Methods Participants were drawn from the Whitehall II study, a prospective cohort study of British civil servants. Data analysed here are from Phase 5 (1997–1999) of data collection, 7830 individuals in all. The measures of SEP and psychosocial health were assessed via a self-administered questionnaire. Results The three models can lead to completely different conclusions. Model III, our preferred model, shows education to have a stronger indirect effect on psychosocial health when compared to its direct effect. The indirect effect is due to the effect of education on proximal measures of social position, occupation, and income in this case. Conclusions Results reported here support the hypothesis that a comparison of the relative importance of the different measures of social position in predicting health is meaningless if the causal relationships among these measures are not accounted for. Socioeconomic position, proximal and distal measures, indirect and direct effects, psychosocial health The inverse association between various measures of socio-economic position (SEP) and both mental and physical health is widely recognized. In recent years there have been attempts to compare different indicators of SEP, using multiple regression analysis, in order to identify which has the strongest association with health outcomes.1–3 A case has been made for occupational position being better than education as a discriminator of socio-economic differentials in health.1,4 However, other studies have found stronger effects for income,2,5 or for education.3 Greater risk of morbidity or mortality in adulthood has principally been related to adult socioeconomic circumstances.6 However, it increasingly appears that ill health in adulthood is linked to an accumulation of risk factors over the course of an individual’s life.7–13 For example, the risk of respiratory disease in adulthood,9 or premature death14 has been found to accumulate over childhood and adolescence. Multivariate analyses aimed at comparing different indicators of SEP in predicting health have led to inconsistent results. This can be attributed to three reasons: Different measures implicate different pathways Different measures of SEP represent different facets of social position and are differently related to health outcomes.15 The standard markers of SEP—education, income, and occupation have been found to have stronger or weaker relationships with various risk behaviours,16 resulting in differential relationship with health outcomes. Outcome/disease specificity Not all ill-health outcomes have the same ‘etiologic period’.17 There is some evidence to suggest that childhood circumstances may have closer links with risk of coronary heart disease,18–20 and not to the same extent with other outcomes like cancer.14 The social gradient is sensitive to the proximal/distal nature of the indicator of socio-economic position employed Another explanation for the variable relationship between indicators of SEP and health is likely to be related to the proximal (measure of SEP closer in time to a health outcome) or distal (SEP measure more distant in time from health outcome measure) nature of the particular measure of social position being employed. Proximal measures of social position may discriminate better as they portray the current and accumulated socio-economic circumstances of the individual more accurately. This third explanation of the variation in the social gradient is the focus of this paper. The idea that there is a valid basis for causal and temporal ordering in the various measures of SEP has been advanced before.21,22 An analysis of the socio-economic status of individuals at several stages of their lives showed that socio-economic origins have enduring effects on adult mortality through their effect on later socio-economic circumstances such as education, occupation, and financial resources.23 Distal measures of social position can be seen to affect health directly; and indirectly through their effects on the proximal measures of social position. This paper considers three different statistical models for analysing the relationship between psychosocial health and three measures of social position: education, occupation, and income. The first model is a simple univariate regression model with one predictor, a measure of SEP, and one health outcome. The second model (Figure 1) is a multiple regression model that ignores temporal relationships between the indicators of SEP. The third model (Figure 2) is derived from the life-course perspective where education is seen to structure occupation and income. In this model education is antecedent to both occupation and income. We hypothesize that the third model will permit a better understanding of the effects of education, a distal measure of SEP here, as it will allow us to explore both the direct and indirect effects of education on psychosocial health. The distal quality attributed to education in relation to occupation and income is based on the assumption that most individuals attain their educational qualifications before being employed or having an income. Therefore, the proximal or distal quality of a measure of SEP depends on its temporal place in the life course. All three models will be explored separately in men and women as there is extensive research evidence showing gender-specific relationships between measures of SEP and health.24–27 Participants and Methods Participants The target population for the Whitehall II study was all the London-based office staff, aged 35–55, working in 20 Civil Service departments.28 With a response rate of 73%, the final cohort consisted of 10 308 participants (6895 men and 3413 women) at the first phase of data collection between 1985 and 1988. The screening at baseline (Phase I) involved a clinical examination, and a self-administered questionnaire containing sections on demographic characteristics, health, lifestyle factors, work characteristics, social support, and life events. Since baseline screening five further data collection rounds have been completed. Successive phases alternate between collecting data by self-administered questionnaire only and collecting data via a clinical screening in addition to questionnaire completion. The most recent phase of data collection (Phase VI) was completed between 2000–2001. The data for these analyses are drawn from Phase V of the Whitehall II study, data for which was collected between 1997– 1999. Response at Phase V was 76% (7830 participants) of those who participated at baseline screening, 12 years previously. In addition to those who failed to respond to invitations to participate, non-responders included participants who had died or those who could not be traced. The loss to follow-up is not influenced by age (P = 0.65) or sex (P = 0.51) but is influenced by employment grade (P < 0.001), with the attrition rate being significantly higher in the lower grades. Measures All measures used in this study were drawn from the questionnaire administered at Phase 5 of data collection. Education Education was measured as the highest level of education achieved, with the respondent choosing one of the 11 categories in the questionnaire. This was regrouped into five standard hierarchic levels: (1) no formal education; (2) lower secondary education; (3) higher secondary education; (4) university degree; and (5) higher university degree. Occupation Occupational position was assessed via civil service employment grade. The 12 non-industrial grade levels where regrouped to lead to 6 employment grades. Income Respondents were asked to pick a category that corresponded most closely with their annual personal income (‘amount received annually from salary or wages, or pensions, benefits and allowances before deduction of tax’). There were eight categories in all ranging from ‘<£9999’ to ‘>£70 000’. For the purposes of analysis the two highest and the two lowest personal income categories were collapsed to leave six categories. These categories are as follows, 1 ⩾ £50 000; 2 = £35 000–£49 999; 3 = £25 000–£34 999; 4 = £20 000–£24 999; 5 = £15 000–£19 999; and 6 ⩽ £14 999. Psychosocial health Psychosocial health was conceptualized as a latent variable measured via four indicators. These were: Hostility, measured using 23 items (Cronbach’s alpha = 0.82) from the Cook-Medley hostility scale.29Hopelessness, measured using six items adapted from Beck’s Hopelessness scale.30 These six items (Cronbach’s alpha = 0.80) were measured on five-point scales. General Health Questionnaire (GHQ) is a 30-item screening questionnaire for minor psychiatric disorder suitable for use in the general population samples.31 The GHQ items were scored on four-point Likert scales (0–3) in order to assign each individual a score reflecting the degree of intensity of psychological distress reported by the individual. Self-rated health was assessed on a five-point scale via the following question: ‘In general, would you say your health is excellent/very good/good/fair/poor’. High scores on all four indicators imply poor psychosocial health. Modelling and statistical methods The first step in our analysis consisted of setting up a measurement model for the elaboration of the latent health construct. This involved specifying the indicators of the latent construct, psychosocial health in this case, and assessing its reliability. Specification of the measurement model resembles confirmatory factor analysis in that the indicators used to define the latent construct are theoretically driven. Psychosocial health has been assessed via four observed variables: hostility, hopelessness, GHQ, and self-rated health. The appropriateness of the measurement model involves examination of the statistical significance of each factor loading and calculation of the composite reliability. Composite reliability is a measure of the internal consistency of the indicators, depicting the extent to which they indicate the common latent construct.32 Composite reliability of around 0.70 is seen to be an acceptable level.32 Once the psychosocial health measure was established the effects of SEP on health were estimated using Models I, II, and III. All three models treat the different indicators of SEP as continuous variables so the interpretation of the regression coefficients in all three models is similar. Model I is a univariate regression model; three such models are specified, each with psychosocial health as a dependent variable and a measure of SEP as the independent variable. Model II is a multiple linear regression model with the three measures of SEP as independent variables. A path diagram for Model II is depicted in Figure 1. All observed variables in the figures are denoted by rectangular boxes and unobserved variables in ovals. The unobserved variables are latent constructs and error terms. Error terms are associated with all endogenous variables and represent measurement error along with effects of variables not measured in the study. Correlations between variables are denoted by double-headed arrows and causal paths by single-headed arrows. Model II (Figure 1) compares the relative importance of the three measures of SEP in predicting psychosocial health. Paths a, b, and c, denoted by single headed arrows, link all three measures of social position to psychosocial health. These paths estimate the effect of a measure of social position on psychosocial health while adjusting for the effects of the other measures of SEP. Model III is depicted in Figure 2. It reflects temporal ordering among the different measures of SEP. Education is the distal measure of SEP, while occupation and income are proximal measures of SEP. Therefore, education is theorized to have both a direct effect (path d, denoted by single headed arrow between education and psychosocial health) and indirect effect on psychosocial health. The indirect effect of education on psychosocial health involves three pathways: the first is path ef, the second is path egi, and the third is path hi (Figure 2). Structural equation models33 (SEM) were used to fit all three models, with a measurement model used to elaborate the latent psychosocial health construct. It was convenient to fit all three models using a SEM software package, but equally Models I and II could have been fitted using a standard regression package with the latent health outcome calculated separately. Indeed, the health outcome could be a manifest variable, in which case the measurement model would be unnecessary. It should therefore be emphasized that the focus here is on using SEM to model the temporal ordering between the three measures of SEP and psychosocial health in Model III, and not specifying the measurement model for psychosocial health. The indirect effects of a variable are mediated through intervening variables.34 To assess model fit, root mean square error of approximation (RMSEA) and comparative fit index (CFI) were used.33,35 An RMSEA value of below 0.05 and a CFI value close to 1 indicates a good fitting model. The analyses were carried out using AMOS version 4.01.36 The AMOS program allows maximum likelihood estimation based on incomplete data, known as full-information maximum likelihood (FIML). This approach is based on the direct maximization of the likelihood of all observed data, not just from cases with complete data. Full-information maximum likelihood is preferable to estimation based on complete data (the listwise deletion approach) as FIML estimates will show less bias and be more reliable than the listwise deletion approach even when data deviate from missing at random and are non-ignorable.37 The results were checked using asymptotically distribution free methods (as some of the data are not normally distributed) and similar results to FIML were found. Bollen’s incremental fit-index values were also examined as these are least biased due to non-normality of variables and they were all above 0.99. Results Table 1 shows the breakdown of the sample by measures of SEP used in this study. Initial examination of the data consisted of an examination of the bivariate relationships between the three measures of SEP and the four indicators of psychosocial health used in this paper (Table 1). High scores on all four measures of psychosocial health represented poorer well-being. As expected, there was an inverse association between SEP and indicators of psychosocial health. This implies that irrespective of the measure of SEP used, low social position was associated with higher levels of hostility, hopelessness, and poorer self-rated health. The test for trend shown in Table 1 was done using a one-way anova. Minor psychiatric disorder as assessed by the GHQ showed a significant inverse relationship only with the measure of occupational position. The measurement model was specified separately for men and women. Among men the factor loadings of the different indicators on psychosocial health were as follows: hostility: 0.51 (P < 0.001), hopelessness: 0.75 (P < 0.001), GHQ: 0.67 (P < 0.001), self-rated health: 0.45 (P < 0.001). The factor loadings in women were hostility: 0.51 (P < 0.001), hopelessness: 0.75 (P < 0.001), GHQ: 0.68 (P < 0.001), self-rated health: 0.42 (P < 0.001). All four indicator variables had significant loadings on psychosocial health among both men and women. The composite reliability of the latent construct psychosocial health was 0.69 in both men and women, implying that the four observed variables reliably measure the latent construct psychosocial health. The results reported in Table 1 are further explored using three separate univariate regressions with the latent construct psychosocial health as the outcome (Table 2, Model I). In these analyses all three measures of SEP were used, individually and one at a time, to predict psychosocial health. All measures of social position have a significant effect on the outcome measure. Furthermore, all regression coefficients were positive indicating the high scores on the measures of social position (implying lower social position) are associated with high score on psychosocial health (implying poorer health). Results for all three models explored in this study are reported using standardized regression coefficients. These coefficients are calculated from standardized data and reflect the impact on the outcome variable of a change of one standard deviation in the predictor variables. For example, according to Model I (Table 2, Model I), an increase of one standard deviation in occupational position in men (indicating lower SEP) increases 0.33 standard deviation in psychosocial health (indicating poorer health). The advantage of standardized regression coefficient is that it eliminates the problem of dealing with different units of measurement, allowing assessment of the relative importance of predictor variables to be made. Model II (Figure 1) is essentially a multiple regression of health on the three SEP measures, allowing estimation of the effect of each measure of SEP on psychosocial health while controlling for the effects of the other two measures. Model III (Figure 2) is an SEM model allowing estimation of both direct and indirect effects of education, a distal measure of SEP in this analysis. The purpose of the analysis is not to compare Models II and III in terms of how well they fit the data, rather it is to demonstrate how treating a distal predictor as a proximal predictor can give misleading results. In fact, both these models fit the data equally well; for example the combined fit for men and women is: χ2 = 514.96, d.f. = 22, MSEA = 0.047, CFI = 0.995. This is because they are equivalent models,38 the difference being that Models II and III provide alternative ways of looking at the data. As is clear from Figure 1, Model II treats the three measures of SEP as being situated at the same point in temporal space. It allowed us to examine the direct effects of each measure of social position on psychosocial health. In men, the direct effect is significant for education and occupation but not for income. The negative coefficient (Table 2) for education indicates that a low score of education (implying higher education) is associated with a high score on psychosocial health (implying poorer health). The positive direct effect of occupation on psychosocial health implies that higher score on occupation (implying lower position) is associated with high score on psychosocial health (implying poorer health). The results obtained here (for both education and income) are different from the univariate model, essentially because the effect of each indicator of SEP on health in Model II is analysed while controlling for the effects of the other two measures. For women, the direct effects of SEP on psychosocial health (Table 2) are significant for occupation and income. A high score on occupation (implying lower SEP) is related to a higher score on psychosocial health (implying poorer health), whereas a low score on income (implying higher income) is associated with a high score on psychosocial health (implying poorer health). The size of the direct effect of the three measures of SEP used on psychosocial health is strongest for occupation, in both men and women. As depicted in Figure 2, Model III takes into account hypothesized temporal relationship between the measures of SEP. Education was seen to have direct effect d on psychosocial health and indirect effects ef, egi, and hi on psychosocial health. From Table 2, it can be seen that the direct effects of all three measures of SEP remained the same in Model III. The difference between Model II and III is that the latter permits the estimation of indirect effect. The indirect effect of education in men was calculated as following: path ef (0.50*0.42 = 0.21) + path egi (0.50*0.57*−0.04 = −0.01) + path hi (0.15*−0.04 = −0.01); leading to a total indirect effect of 0.19 (Table 2). The indirect effect implies that a higher score on education (implying lower level of education) is associated with a higher score on psychosocial health (implying poorer health). The same pattern of results holds for the indirect effects of education on psychosocial health in women. In our conceptualization of the direct and indirect effects model, occupation was allowed to have an effect on personal income (Figure 2). Therefore, occupation had a direct effect f and an indirect effect gi on psychosocial health. Income was hypothesized only to have a direct effect on psychosocial health. The indirect effect associated with education was relatively large for both men and women. Occupation was the strongest predictor of psychosocial health in both men and women, poorer occupational position was associated with poor psychosocial health. High personal income was not found to be protective of psychosocial health in this sample. Discussion This paper set out to show that distal measures of SEP affect health both directly and indirectly, and that ignoring the indirect effects will produce misleading conclusions concerning the importance of measures of SEP to health. The indirect effect of distal measures of SEP is mediated by the effect of distal measures of SEP on the proximal measures. In other words, education influences health outcomes both directly; and indirectly through its effect on occupation and income. Here we have illustrated this using psychosocial health but conclusions drawn here potentially apply to any other health outcome. The direct effect of education was assessed by adjusting for the effects of occupation and income (Figure 1) in multiple regression analysis (Model II). This is standard practice in analysis where the aim is to identify ‘independent’ effects of predictors, measures of SEP in this case. The direct effect of education suggests that high educational achievement is predictive of poorer psychosocial health. However, this conclusion, based on multiple regression where the temporal aspect of different measures of SEP is ignored, is correct but misleading. From Table 2, the direct effect of education on psychosocial health for men is −0.13. This is not erroneous; it is plausible to believe that the better-educated have poorer psychosocial health than those less educated given that they have achieved the same income and occupational status. However, the indirect effect of education on psychosocial health is positive, 0.19 (from Table 2). We would suggest that this is a better indication of the effect of education on psychosocial health through the life course. In fact, this was seen in the simple analysis provided by Model I, with the effect of education on psychosocial health estimated without adjusting for the effects of the proximal measures of SEP (occupation and income, in this paper). Different indicators of SEP—parental social class, educational, occupational position, income, etc.—are routinely compared for their effects on ill health and mortality. These comparisons ignore the fact that different measures of SEP are linked to different phases of the life course, and that they are also associated with each other. Taking into account the links between different measures of SEP changes the conclusions drawn from comparative analyses. The results reported here show different signs for direct and indirect effects, with the direct effects model (Model II) not showing higher educational level to be predictive of good psychosocial health. It is only when the indirect effects of education are considered, does the relationship between education and psychosocial health become clear. The existence of socio-economic inequality in health is readily accepted. However, the causal pathways linking SEP to poor health are only beginning to be explored. It is likely that education informs lifestyles and attitudes towards and knowledge of health behaviours. Similarly, occupational position is likely to be important due to the effects of the work environment, both physical and psychological, on health. Income has theoretically been linked to the material deprivation pathway, the lack of income preventing basic needs from being met. In fact, theorists propose a wider effect of education on health. Ross and Wu22 have advanced three possible pathways between education and health: Work and economic conditions Higher education leads to better jobs, lower unemployment and thus better work conditions. Social-psychological resources Higher education arms individuals with a sense of personal control, also allowing them to establish better social support. Health lifestyle Individuals who have a better education have a healthier lifestyle—they exercise more, drink moderately and smoke less. It is likely that the recent trend in social inequalities research to compare the various indicators of SEP is driven by a need to understand varying results. American studies have traditionally used education to measure SEP,39 while the British studies have used occupational position.28 The use of education has been championed on the grounds that it is less liable to reverse causation, is stable over the life course, and can be applied to those outside work.40 The disadvantage with education is that it does not capture the changes in adult socio-economic circumstances or accumulated SEP. In this paper we have chosen to ignore childhood socio-economic circumstances for the sake of simplicity. Early childhood circumstances are likely to influence adult health through both direct and indirect effects. The direct effects may involve susceptibility to disease through biological mechanisms or through the socialization of unhealthy behavioural practices or psychological response strategies. The indirect effects are likely to be mediated by adult SEP achieved by individuals. There is some evidence to suggest that childhood socio-economic circumstance shows an independent effect on both adult health and on health related behaviour.40 The results presented here attempt to show that when different measures of SEP are being compared for their effect on health, the interrelationships between them should be accounted for. Conclusions drawn from studies where the various indicators of SEP are allocated the same temporal space will almost certainly be misleading. We have used a simple example to show that proximal and distal measures of SEP operate differently due to the interrelationship among them. These associations among the various measures of SEP will be critically important in models where pathways linking SEP to health along with possible confounders are explicated. Conclusions similar to the ones drawn in this paper were recently published by Weitkunat and Wildner.41 In their paper, unequal proximity of different variables is considered in relation to predicting health. Psychological factors are hypothesized as being more distal than biological factors and it is hypothesized that assigning them equal proximity will produce misleading results. The authors used a simulation study to show that multiple regression analysis can give misleading results when analysing ‘sequentially caused relationships’ because it treats a ‘distant causal factor’ as being equally distant to the outcome as a ‘proximal causal factor’. The authors used this artificial example to show that pathway analysis is correctly able to identify the true causal sequence. The variables we examined were all measures of adult SEP, and we have shown that ‘unequal proximity’ is an issue even when predictor variables are similar in nature. In our paper, proximity is considered in terms of temporal place of a measure of SEP in relation to a health outcome. However, the analogy with Weitkunat and Wildner is clear, offering support for our conclusions. A recent paper by Didelez and colleagues42 also focuses on the necessity of choosing appropriate statistical models as being key to obtaining valid and meaningful results in epidemiological studies. The limitations of restricting analyses to regression models is discussed, with the authors emphasizing the need for quantitative data analyses to reflect advances in conceptual and theoretical models. Didelez et al.42 discuss the relationship between research and theory concluding that ‘... oscillation, refinement, and replication is fundamental to providing the rationale for inclusion and modelling of variables’. There are some important areas of further work that need to be addressed before causal interpretations can be made. The data reported here are cross-sectional and these models need to be replicated with longitudinal data before firm conclusions about the causal effect of each variable on health can be drawn. Also, other health outcomes need to be analysed in order to examine the direct and indirect effects of proximal and distal measures on different health outcomes. Despite these shortcomings, this paper offers an example of the way in which specific theoretical relationships between variables can be examined. It should also be noted that SEM is obviously not being advocated as a panacea, but only as one of the ways in which theoretically driven analyses can be specified. We have discussed recent papers that have used other techniques that address the need for a close relationship between theory and analysis: path analysis,41 and graphical chain models.42 Acknowledgements The Whitehall II study has been supported by grants from the Medical Research Council; British Heart Foundation; Health and Safety Executive; Department of Health; National Heart Lung and Blood Institute (HL36310), US, NIH: National Institute on Aging (AG13196), US, NIH; Agency for Health Care Policy Research (HS06516); and the John D and Catherine T MacArthur Foundation Research Networks on Successful Midlife Development and Socio-economic Status and Health. MM is supported by an MRC Research Professorship. We also thank all participating civil service departments and their welfare, personnel, and establishment officers; the Occupational Health and Safety Agency; the Council of Civil Service Unions; all participating civil servants in the Whitehall II study; and all members of the Whitehall II study team. Table 1 Means of psychosocial health indicatorsa for categories of socio-economic position measures . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 Open in new tab Table 1 Means of psychosocial health indicatorsa for categories of socio-economic position measures . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 Open in new tab Table 2 Effects (standardized regression coefficients) of indicators of socio-economic positiona on psychosocial healthb . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA Open in new tab Table 2 Effects (standardized regression coefficients) of indicators of socio-economic positiona on psychosocial healthb . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA Open in new tab Figure 1 Open in new tabDownload slide Direct effects of education, income, and occupation on psychosocial health (Model II), path coefficients shown for men in the Figure Notes: a, b, and c are standardized regression coefficients reflecting the effect of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 5) represent the error terms associated with endogenous variables in the model. Figure 1 Open in new tabDownload slide Direct effects of education, income, and occupation on psychosocial health (Model II), path coefficients shown for men in the Figure Notes: a, b, and c are standardized regression coefficients reflecting the effect of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 5) represent the error terms associated with endogenous variables in the model. Figure 2 Open in new tabDownload slide Direct and indirect effects of education, income, and occupation on psychosocial health (Model III), path coefficients shown for men in the Figure Notes: d, e, f, g, h and i are standardized regression coefficients reflecting the effects (direct and indirect) of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 7) represent error associated with endogenous variables in the model. Figure 2 Open in new tabDownload slide Direct and indirect effects of education, income, and occupation on psychosocial health (Model III), path coefficients shown for men in the Figure Notes: d, e, f, g, h and i are standardized regression coefficients reflecting the effects (direct and indirect) of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 7) represent error associated with endogenous variables in the model. References 1 Davey Smith G, Hart C, Hole D et al. Education and occupation social class: which is the more important indicator of mortality risk? J Epidemiol Community Health 1998 ; 52: 153 –60. 2 Geyer S, Peter R. Income, occupational position, qualification and health inequalities—competing risks? (Comparing indicators of social status). J Epidemiol Community Health 2000 ; 54: 299 –305. 3 Winkleby MA, Jatulis DE, Frank E, Fortmann SP. Socioeconomic status and health: How education, income, and education contribute to risk factors for cardiovascular disease. Am J Public Health 1992 ; 82: 816 –20. 4 Dahl E. Social inequalities in ill-health: the significance of occupational position, education and income results from a Norwegian survey. Sociol Health Illness 1994 ; 16: 644 –67. 5 Blaxter M. Health and Lifestyles. London: Routledge, 1990 . 6 Marmot M, Shipley M, Brunner E, Hemingway H. Relative contribution of early life and adult socioeconomic factors to adult morbidity in the Whitehall II study. J Epidemiol Community Health 2001 ; 55: 301 –07. 7 Ben-Shlomo Y, Davey Smith G. Deprivation in infancy or in adult life: which is more important for mortality risk? Lancet 1991 ; 337: 530 –34. 8 Department of Health. Variation in Health: what can the Department of Health and the NHS do? London: Department of Health, 1995 . 9 Mann SL, Wadsworth ME, Colley JR. Accumulation of factors influencing respiratory illness in members of a national birth cohort and their offspring. J Epidemiol Community Health 1992 ; 46: 286 –92. 10 Marmot M. Aetiology of coronary heart disease. BMJ 2001 ; 323: 1261 –62. 11 Power C, Manor O, Fox AJ, Fogelman K. Health in childhood and social inequalities in health in young adults. J R Statist Soc 1990 ; 153: 17 –28. 12 Power C, Mathews S, Manor O. Inequalities in self rated health in the 1958 birth cohort: lifetime social circumstances or social mobility? BMJ 1996 ; 313: 449 –53. 13 Power C, Mathews S, Manor O. Inequalities in self-rated health: explanations from different stages of life. Lancet 1998 ; 351: 1009 –14. 14 Davey Smith G, Hart C, Blane D, Gillis C, Hawthorne V. Lifetime socioeconomic position and mortality: prospective observational study. BMJ 1997 ; 314: 547 –52. 15 Bartley M, Sacker A, Firth D, Fitzpatrick R. Understanding social variation in cardiovascular risk factors in women and men: the advantage of theoretically based measures. Soc Sci Med 1999 ; 49: 831 –45. 16 Lynch JW, Kaplan GA, Salonen JT. Why do poor people behave poorly? Variation in adult health behaviours and psychosocial characteristics by stages of the socioeconomic lifecourse. Soc Sci Med 1997 ; 44: 809 –19. 17 Krieger N, Williams DR, Moss NE. Measuring social class in US public health research: Concepts, methodologies, and guidelines. Annu Rev Public Health 1997 ; 18: 341 –78. 18 Kaplan GA, Salonen JT. Socioeconomic conditions in childhood and ischaemic heart disease during middle age. BMJ 1990 ; 301: 1121 –23. 19 Vagero D, Leon D. Effect of social class in childhood and adulthood on adult mortality. Lancet 1994 ; 343: 1224 –25. 20 Wannamethee SG, Whincup PH, Shaper G, Walker M. Influence of father’s social class on cardiovascular disease in middle-aged men. Lancet 1996 ; 348: 1259 –63. 21 Duncan OD. How destination depends on origin in the occupational mobility table. Am J Sociol 1979 ; 84: 793 –803. 22 Ross CE, Wu C-L. The links between education and health. Am Sociol Rev 1995 ; 60: 719 –45. 23 Mare RD. Socio-economic careers and differential mortality among older men in the United States. In: Vallin J, D’Souza S, Polloni A (eds). Comparative Studies of Mortality and Morbidity: Old and New Approaches to Measurement and Analysis. Oxford: Oxford University Press, 1990 , pp.362–87. 24 Arber S. Comparing inequalities in women’s and men’s health: Britain in the 1990s. Soc Sci Med 1997 ; 44: 773 –87. 25 Arber S, Cooper H. Gender differences in health in later life: the new paradox? Soc Sci Med 1999 ; 48: 61 –76. 26 Sacker A, Bartley M, Firth D, Fitzpatrick R. Dimensions of social inequality in the health of women in England: occupational, material, and behavioral pathways. Soc Sci Med 2001 ; 52: 763 –81. 27 Sacker A, Firth D, Fitzpatrick R, Lynch K, Bartley M. Comparing health inequality in men and women: prospective study of mortality 1986–96. BMJ 2000 ; 320: 1303 –07. 28 Marmot MG, Davey Smith G, Stansfeld S et al. Health inequalities among British civil servants: the Whitehall II study. Lancet 1991 ; 337: 1387 –93. 29 Cook WW, Medley DM. Proposed hostility and Pharisaic values for the MMPI. J Appl Psychol 1954 ; 38: 414 –18. 30 Beck AT, Weissman A, Lester D, Trexler L. The measurement of pessimism: The hopelessness scale. J Consult Clin Psychol 1974 ; 42: 861 –65. 31 Goldberg DP. The Detection of Psychiatric Illness by Questionnaire. (Institute of Psychiatry Maudsley mongraphs No. 21). London: Oxford University Press, 1972 . 32 Hair JF, Anderson RE, Tatham RL, Black WC. Multivariate Data Analysis. NJ: Prentice-Hall, Inc., 1998 . 33 Mueller RO. Basic Principles of Structural Equation Modeling. New York: Springer Verlag, 1996 . 34 Vassend O, Skrondal A. The role of negative affectivity in self-assessment of health. J Health Psychol 1999 ; 4: 465 –82. 35 Sacker A, Bartley M, Firth D, Fitzpatrick R. Dimensions of social inequality in the health of women in England: occupational, material, and behavioral pathways. Soc Sci Med 2001 ; 52: 763 –81. 36 Arbuckle JL, Wothke W. Amos 4.0 User’s Guide. Chicago: Small Waters Corp., 1995 . 37 Arbuckle JL. Full-information estimation in the presence of incomplete data. In: Marcoulides GA, Schumacker RE (eds). Advanced Structural Equation Modeling Techniques. Mahwah, NJ: Lawrence Erlbaum Associates, 1996 , pp.243–77. 38 Lee S, Hershberger S. A simple rule for generating equivalent models in covariance structure modeling. Multivariate Behav Res 1990 ; 25: 313 –34. 39 Krieger N, Fee E. Social class: the missing link in US health data. Int J Health Serv 1994 ; 24: 25 –44. 40 Valkonen T. Adult mortality and level of education: a comparison of six countries. In: Fox J (ed.). Health Inequalities in European Countries. Aldershot: Gower, 1989 , pp.142–72. 41 Weitkunat R, Wildner M. Exploratory causal modeling in epidemiology: Are all factors created equal? J Clin Epidemiol 2002 ; 55: 436 –44. 42 Didelez V, Pigeot I, Dean K, Wister A. A comparative analysis of graphical interaction and logistic regression modelling: self-care and coping with a chronic illness in later life. Biom J 2002 ; 44: 410 –32. © International Epidemiological Association 2002 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Journal of Epidemiology Oxford University Press

Multiple measures of socio-economic position and psychosocial health: proximal and distal measures

Loading next page...
 
/lp/oxford-university-press/multiple-measures-of-socio-economic-position-and-psychosocial-health-Rl1DZLCzRF

References (49)

Publisher
Oxford University Press
Copyright
© International Epidemiological Association 2002
ISSN
0300-5771
eISSN
1464-3685
DOI
10.1093/ije/31.6.1192
Publisher site
See Article on Publisher Site

Abstract

Abstract Background The aim of this paper is to compare three models for exploring the links between different measures of adult socioeconomic position (SEP)—education, occupation, income—and psychosocial health. Model I is a basic univariate regression model with psychosocial health as the outcome and a measure of SEP as the predictor. Model II is a multiple regression model with psychosocial health as the outcome with all three measures of SEP allocated the same temporal position as predictors. Model III treats education, a distal measure of SEP, as antecedent to the proximal measures of SEP in the prediction equations linking SEP to health. Methods Participants were drawn from the Whitehall II study, a prospective cohort study of British civil servants. Data analysed here are from Phase 5 (1997–1999) of data collection, 7830 individuals in all. The measures of SEP and psychosocial health were assessed via a self-administered questionnaire. Results The three models can lead to completely different conclusions. Model III, our preferred model, shows education to have a stronger indirect effect on psychosocial health when compared to its direct effect. The indirect effect is due to the effect of education on proximal measures of social position, occupation, and income in this case. Conclusions Results reported here support the hypothesis that a comparison of the relative importance of the different measures of social position in predicting health is meaningless if the causal relationships among these measures are not accounted for. Socioeconomic position, proximal and distal measures, indirect and direct effects, psychosocial health The inverse association between various measures of socio-economic position (SEP) and both mental and physical health is widely recognized. In recent years there have been attempts to compare different indicators of SEP, using multiple regression analysis, in order to identify which has the strongest association with health outcomes.1–3 A case has been made for occupational position being better than education as a discriminator of socio-economic differentials in health.1,4 However, other studies have found stronger effects for income,2,5 or for education.3 Greater risk of morbidity or mortality in adulthood has principally been related to adult socioeconomic circumstances.6 However, it increasingly appears that ill health in adulthood is linked to an accumulation of risk factors over the course of an individual’s life.7–13 For example, the risk of respiratory disease in adulthood,9 or premature death14 has been found to accumulate over childhood and adolescence. Multivariate analyses aimed at comparing different indicators of SEP in predicting health have led to inconsistent results. This can be attributed to three reasons: Different measures implicate different pathways Different measures of SEP represent different facets of social position and are differently related to health outcomes.15 The standard markers of SEP—education, income, and occupation have been found to have stronger or weaker relationships with various risk behaviours,16 resulting in differential relationship with health outcomes. Outcome/disease specificity Not all ill-health outcomes have the same ‘etiologic period’.17 There is some evidence to suggest that childhood circumstances may have closer links with risk of coronary heart disease,18–20 and not to the same extent with other outcomes like cancer.14 The social gradient is sensitive to the proximal/distal nature of the indicator of socio-economic position employed Another explanation for the variable relationship between indicators of SEP and health is likely to be related to the proximal (measure of SEP closer in time to a health outcome) or distal (SEP measure more distant in time from health outcome measure) nature of the particular measure of social position being employed. Proximal measures of social position may discriminate better as they portray the current and accumulated socio-economic circumstances of the individual more accurately. This third explanation of the variation in the social gradient is the focus of this paper. The idea that there is a valid basis for causal and temporal ordering in the various measures of SEP has been advanced before.21,22 An analysis of the socio-economic status of individuals at several stages of their lives showed that socio-economic origins have enduring effects on adult mortality through their effect on later socio-economic circumstances such as education, occupation, and financial resources.23 Distal measures of social position can be seen to affect health directly; and indirectly through their effects on the proximal measures of social position. This paper considers three different statistical models for analysing the relationship between psychosocial health and three measures of social position: education, occupation, and income. The first model is a simple univariate regression model with one predictor, a measure of SEP, and one health outcome. The second model (Figure 1) is a multiple regression model that ignores temporal relationships between the indicators of SEP. The third model (Figure 2) is derived from the life-course perspective where education is seen to structure occupation and income. In this model education is antecedent to both occupation and income. We hypothesize that the third model will permit a better understanding of the effects of education, a distal measure of SEP here, as it will allow us to explore both the direct and indirect effects of education on psychosocial health. The distal quality attributed to education in relation to occupation and income is based on the assumption that most individuals attain their educational qualifications before being employed or having an income. Therefore, the proximal or distal quality of a measure of SEP depends on its temporal place in the life course. All three models will be explored separately in men and women as there is extensive research evidence showing gender-specific relationships between measures of SEP and health.24–27 Participants and Methods Participants The target population for the Whitehall II study was all the London-based office staff, aged 35–55, working in 20 Civil Service departments.28 With a response rate of 73%, the final cohort consisted of 10 308 participants (6895 men and 3413 women) at the first phase of data collection between 1985 and 1988. The screening at baseline (Phase I) involved a clinical examination, and a self-administered questionnaire containing sections on demographic characteristics, health, lifestyle factors, work characteristics, social support, and life events. Since baseline screening five further data collection rounds have been completed. Successive phases alternate between collecting data by self-administered questionnaire only and collecting data via a clinical screening in addition to questionnaire completion. The most recent phase of data collection (Phase VI) was completed between 2000–2001. The data for these analyses are drawn from Phase V of the Whitehall II study, data for which was collected between 1997– 1999. Response at Phase V was 76% (7830 participants) of those who participated at baseline screening, 12 years previously. In addition to those who failed to respond to invitations to participate, non-responders included participants who had died or those who could not be traced. The loss to follow-up is not influenced by age (P = 0.65) or sex (P = 0.51) but is influenced by employment grade (P < 0.001), with the attrition rate being significantly higher in the lower grades. Measures All measures used in this study were drawn from the questionnaire administered at Phase 5 of data collection. Education Education was measured as the highest level of education achieved, with the respondent choosing one of the 11 categories in the questionnaire. This was regrouped into five standard hierarchic levels: (1) no formal education; (2) lower secondary education; (3) higher secondary education; (4) university degree; and (5) higher university degree. Occupation Occupational position was assessed via civil service employment grade. The 12 non-industrial grade levels where regrouped to lead to 6 employment grades. Income Respondents were asked to pick a category that corresponded most closely with their annual personal income (‘amount received annually from salary or wages, or pensions, benefits and allowances before deduction of tax’). There were eight categories in all ranging from ‘<£9999’ to ‘>£70 000’. For the purposes of analysis the two highest and the two lowest personal income categories were collapsed to leave six categories. These categories are as follows, 1 ⩾ £50 000; 2 = £35 000–£49 999; 3 = £25 000–£34 999; 4 = £20 000–£24 999; 5 = £15 000–£19 999; and 6 ⩽ £14 999. Psychosocial health Psychosocial health was conceptualized as a latent variable measured via four indicators. These were: Hostility, measured using 23 items (Cronbach’s alpha = 0.82) from the Cook-Medley hostility scale.29Hopelessness, measured using six items adapted from Beck’s Hopelessness scale.30 These six items (Cronbach’s alpha = 0.80) were measured on five-point scales. General Health Questionnaire (GHQ) is a 30-item screening questionnaire for minor psychiatric disorder suitable for use in the general population samples.31 The GHQ items were scored on four-point Likert scales (0–3) in order to assign each individual a score reflecting the degree of intensity of psychological distress reported by the individual. Self-rated health was assessed on a five-point scale via the following question: ‘In general, would you say your health is excellent/very good/good/fair/poor’. High scores on all four indicators imply poor psychosocial health. Modelling and statistical methods The first step in our analysis consisted of setting up a measurement model for the elaboration of the latent health construct. This involved specifying the indicators of the latent construct, psychosocial health in this case, and assessing its reliability. Specification of the measurement model resembles confirmatory factor analysis in that the indicators used to define the latent construct are theoretically driven. Psychosocial health has been assessed via four observed variables: hostility, hopelessness, GHQ, and self-rated health. The appropriateness of the measurement model involves examination of the statistical significance of each factor loading and calculation of the composite reliability. Composite reliability is a measure of the internal consistency of the indicators, depicting the extent to which they indicate the common latent construct.32 Composite reliability of around 0.70 is seen to be an acceptable level.32 Once the psychosocial health measure was established the effects of SEP on health were estimated using Models I, II, and III. All three models treat the different indicators of SEP as continuous variables so the interpretation of the regression coefficients in all three models is similar. Model I is a univariate regression model; three such models are specified, each with psychosocial health as a dependent variable and a measure of SEP as the independent variable. Model II is a multiple linear regression model with the three measures of SEP as independent variables. A path diagram for Model II is depicted in Figure 1. All observed variables in the figures are denoted by rectangular boxes and unobserved variables in ovals. The unobserved variables are latent constructs and error terms. Error terms are associated with all endogenous variables and represent measurement error along with effects of variables not measured in the study. Correlations between variables are denoted by double-headed arrows and causal paths by single-headed arrows. Model II (Figure 1) compares the relative importance of the three measures of SEP in predicting psychosocial health. Paths a, b, and c, denoted by single headed arrows, link all three measures of social position to psychosocial health. These paths estimate the effect of a measure of social position on psychosocial health while adjusting for the effects of the other measures of SEP. Model III is depicted in Figure 2. It reflects temporal ordering among the different measures of SEP. Education is the distal measure of SEP, while occupation and income are proximal measures of SEP. Therefore, education is theorized to have both a direct effect (path d, denoted by single headed arrow between education and psychosocial health) and indirect effect on psychosocial health. The indirect effect of education on psychosocial health involves three pathways: the first is path ef, the second is path egi, and the third is path hi (Figure 2). Structural equation models33 (SEM) were used to fit all three models, with a measurement model used to elaborate the latent psychosocial health construct. It was convenient to fit all three models using a SEM software package, but equally Models I and II could have been fitted using a standard regression package with the latent health outcome calculated separately. Indeed, the health outcome could be a manifest variable, in which case the measurement model would be unnecessary. It should therefore be emphasized that the focus here is on using SEM to model the temporal ordering between the three measures of SEP and psychosocial health in Model III, and not specifying the measurement model for psychosocial health. The indirect effects of a variable are mediated through intervening variables.34 To assess model fit, root mean square error of approximation (RMSEA) and comparative fit index (CFI) were used.33,35 An RMSEA value of below 0.05 and a CFI value close to 1 indicates a good fitting model. The analyses were carried out using AMOS version 4.01.36 The AMOS program allows maximum likelihood estimation based on incomplete data, known as full-information maximum likelihood (FIML). This approach is based on the direct maximization of the likelihood of all observed data, not just from cases with complete data. Full-information maximum likelihood is preferable to estimation based on complete data (the listwise deletion approach) as FIML estimates will show less bias and be more reliable than the listwise deletion approach even when data deviate from missing at random and are non-ignorable.37 The results were checked using asymptotically distribution free methods (as some of the data are not normally distributed) and similar results to FIML were found. Bollen’s incremental fit-index values were also examined as these are least biased due to non-normality of variables and they were all above 0.99. Results Table 1 shows the breakdown of the sample by measures of SEP used in this study. Initial examination of the data consisted of an examination of the bivariate relationships between the three measures of SEP and the four indicators of psychosocial health used in this paper (Table 1). High scores on all four measures of psychosocial health represented poorer well-being. As expected, there was an inverse association between SEP and indicators of psychosocial health. This implies that irrespective of the measure of SEP used, low social position was associated with higher levels of hostility, hopelessness, and poorer self-rated health. The test for trend shown in Table 1 was done using a one-way anova. Minor psychiatric disorder as assessed by the GHQ showed a significant inverse relationship only with the measure of occupational position. The measurement model was specified separately for men and women. Among men the factor loadings of the different indicators on psychosocial health were as follows: hostility: 0.51 (P < 0.001), hopelessness: 0.75 (P < 0.001), GHQ: 0.67 (P < 0.001), self-rated health: 0.45 (P < 0.001). The factor loadings in women were hostility: 0.51 (P < 0.001), hopelessness: 0.75 (P < 0.001), GHQ: 0.68 (P < 0.001), self-rated health: 0.42 (P < 0.001). All four indicator variables had significant loadings on psychosocial health among both men and women. The composite reliability of the latent construct psychosocial health was 0.69 in both men and women, implying that the four observed variables reliably measure the latent construct psychosocial health. The results reported in Table 1 are further explored using three separate univariate regressions with the latent construct psychosocial health as the outcome (Table 2, Model I). In these analyses all three measures of SEP were used, individually and one at a time, to predict psychosocial health. All measures of social position have a significant effect on the outcome measure. Furthermore, all regression coefficients were positive indicating the high scores on the measures of social position (implying lower social position) are associated with high score on psychosocial health (implying poorer health). Results for all three models explored in this study are reported using standardized regression coefficients. These coefficients are calculated from standardized data and reflect the impact on the outcome variable of a change of one standard deviation in the predictor variables. For example, according to Model I (Table 2, Model I), an increase of one standard deviation in occupational position in men (indicating lower SEP) increases 0.33 standard deviation in psychosocial health (indicating poorer health). The advantage of standardized regression coefficient is that it eliminates the problem of dealing with different units of measurement, allowing assessment of the relative importance of predictor variables to be made. Model II (Figure 1) is essentially a multiple regression of health on the three SEP measures, allowing estimation of the effect of each measure of SEP on psychosocial health while controlling for the effects of the other two measures. Model III (Figure 2) is an SEM model allowing estimation of both direct and indirect effects of education, a distal measure of SEP in this analysis. The purpose of the analysis is not to compare Models II and III in terms of how well they fit the data, rather it is to demonstrate how treating a distal predictor as a proximal predictor can give misleading results. In fact, both these models fit the data equally well; for example the combined fit for men and women is: χ2 = 514.96, d.f. = 22, MSEA = 0.047, CFI = 0.995. This is because they are equivalent models,38 the difference being that Models II and III provide alternative ways of looking at the data. As is clear from Figure 1, Model II treats the three measures of SEP as being situated at the same point in temporal space. It allowed us to examine the direct effects of each measure of social position on psychosocial health. In men, the direct effect is significant for education and occupation but not for income. The negative coefficient (Table 2) for education indicates that a low score of education (implying higher education) is associated with a high score on psychosocial health (implying poorer health). The positive direct effect of occupation on psychosocial health implies that higher score on occupation (implying lower position) is associated with high score on psychosocial health (implying poorer health). The results obtained here (for both education and income) are different from the univariate model, essentially because the effect of each indicator of SEP on health in Model II is analysed while controlling for the effects of the other two measures. For women, the direct effects of SEP on psychosocial health (Table 2) are significant for occupation and income. A high score on occupation (implying lower SEP) is related to a higher score on psychosocial health (implying poorer health), whereas a low score on income (implying higher income) is associated with a high score on psychosocial health (implying poorer health). The size of the direct effect of the three measures of SEP used on psychosocial health is strongest for occupation, in both men and women. As depicted in Figure 2, Model III takes into account hypothesized temporal relationship between the measures of SEP. Education was seen to have direct effect d on psychosocial health and indirect effects ef, egi, and hi on psychosocial health. From Table 2, it can be seen that the direct effects of all three measures of SEP remained the same in Model III. The difference between Model II and III is that the latter permits the estimation of indirect effect. The indirect effect of education in men was calculated as following: path ef (0.50*0.42 = 0.21) + path egi (0.50*0.57*−0.04 = −0.01) + path hi (0.15*−0.04 = −0.01); leading to a total indirect effect of 0.19 (Table 2). The indirect effect implies that a higher score on education (implying lower level of education) is associated with a higher score on psychosocial health (implying poorer health). The same pattern of results holds for the indirect effects of education on psychosocial health in women. In our conceptualization of the direct and indirect effects model, occupation was allowed to have an effect on personal income (Figure 2). Therefore, occupation had a direct effect f and an indirect effect gi on psychosocial health. Income was hypothesized only to have a direct effect on psychosocial health. The indirect effect associated with education was relatively large for both men and women. Occupation was the strongest predictor of psychosocial health in both men and women, poorer occupational position was associated with poor psychosocial health. High personal income was not found to be protective of psychosocial health in this sample. Discussion This paper set out to show that distal measures of SEP affect health both directly and indirectly, and that ignoring the indirect effects will produce misleading conclusions concerning the importance of measures of SEP to health. The indirect effect of distal measures of SEP is mediated by the effect of distal measures of SEP on the proximal measures. In other words, education influences health outcomes both directly; and indirectly through its effect on occupation and income. Here we have illustrated this using psychosocial health but conclusions drawn here potentially apply to any other health outcome. The direct effect of education was assessed by adjusting for the effects of occupation and income (Figure 1) in multiple regression analysis (Model II). This is standard practice in analysis where the aim is to identify ‘independent’ effects of predictors, measures of SEP in this case. The direct effect of education suggests that high educational achievement is predictive of poorer psychosocial health. However, this conclusion, based on multiple regression where the temporal aspect of different measures of SEP is ignored, is correct but misleading. From Table 2, the direct effect of education on psychosocial health for men is −0.13. This is not erroneous; it is plausible to believe that the better-educated have poorer psychosocial health than those less educated given that they have achieved the same income and occupational status. However, the indirect effect of education on psychosocial health is positive, 0.19 (from Table 2). We would suggest that this is a better indication of the effect of education on psychosocial health through the life course. In fact, this was seen in the simple analysis provided by Model I, with the effect of education on psychosocial health estimated without adjusting for the effects of the proximal measures of SEP (occupation and income, in this paper). Different indicators of SEP—parental social class, educational, occupational position, income, etc.—are routinely compared for their effects on ill health and mortality. These comparisons ignore the fact that different measures of SEP are linked to different phases of the life course, and that they are also associated with each other. Taking into account the links between different measures of SEP changes the conclusions drawn from comparative analyses. The results reported here show different signs for direct and indirect effects, with the direct effects model (Model II) not showing higher educational level to be predictive of good psychosocial health. It is only when the indirect effects of education are considered, does the relationship between education and psychosocial health become clear. The existence of socio-economic inequality in health is readily accepted. However, the causal pathways linking SEP to poor health are only beginning to be explored. It is likely that education informs lifestyles and attitudes towards and knowledge of health behaviours. Similarly, occupational position is likely to be important due to the effects of the work environment, both physical and psychological, on health. Income has theoretically been linked to the material deprivation pathway, the lack of income preventing basic needs from being met. In fact, theorists propose a wider effect of education on health. Ross and Wu22 have advanced three possible pathways between education and health: Work and economic conditions Higher education leads to better jobs, lower unemployment and thus better work conditions. Social-psychological resources Higher education arms individuals with a sense of personal control, also allowing them to establish better social support. Health lifestyle Individuals who have a better education have a healthier lifestyle—they exercise more, drink moderately and smoke less. It is likely that the recent trend in social inequalities research to compare the various indicators of SEP is driven by a need to understand varying results. American studies have traditionally used education to measure SEP,39 while the British studies have used occupational position.28 The use of education has been championed on the grounds that it is less liable to reverse causation, is stable over the life course, and can be applied to those outside work.40 The disadvantage with education is that it does not capture the changes in adult socio-economic circumstances or accumulated SEP. In this paper we have chosen to ignore childhood socio-economic circumstances for the sake of simplicity. Early childhood circumstances are likely to influence adult health through both direct and indirect effects. The direct effects may involve susceptibility to disease through biological mechanisms or through the socialization of unhealthy behavioural practices or psychological response strategies. The indirect effects are likely to be mediated by adult SEP achieved by individuals. There is some evidence to suggest that childhood socio-economic circumstance shows an independent effect on both adult health and on health related behaviour.40 The results presented here attempt to show that when different measures of SEP are being compared for their effect on health, the interrelationships between them should be accounted for. Conclusions drawn from studies where the various indicators of SEP are allocated the same temporal space will almost certainly be misleading. We have used a simple example to show that proximal and distal measures of SEP operate differently due to the interrelationship among them. These associations among the various measures of SEP will be critically important in models where pathways linking SEP to health along with possible confounders are explicated. Conclusions similar to the ones drawn in this paper were recently published by Weitkunat and Wildner.41 In their paper, unequal proximity of different variables is considered in relation to predicting health. Psychological factors are hypothesized as being more distal than biological factors and it is hypothesized that assigning them equal proximity will produce misleading results. The authors used a simulation study to show that multiple regression analysis can give misleading results when analysing ‘sequentially caused relationships’ because it treats a ‘distant causal factor’ as being equally distant to the outcome as a ‘proximal causal factor’. The authors used this artificial example to show that pathway analysis is correctly able to identify the true causal sequence. The variables we examined were all measures of adult SEP, and we have shown that ‘unequal proximity’ is an issue even when predictor variables are similar in nature. In our paper, proximity is considered in terms of temporal place of a measure of SEP in relation to a health outcome. However, the analogy with Weitkunat and Wildner is clear, offering support for our conclusions. A recent paper by Didelez and colleagues42 also focuses on the necessity of choosing appropriate statistical models as being key to obtaining valid and meaningful results in epidemiological studies. The limitations of restricting analyses to regression models is discussed, with the authors emphasizing the need for quantitative data analyses to reflect advances in conceptual and theoretical models. Didelez et al.42 discuss the relationship between research and theory concluding that ‘... oscillation, refinement, and replication is fundamental to providing the rationale for inclusion and modelling of variables’. There are some important areas of further work that need to be addressed before causal interpretations can be made. The data reported here are cross-sectional and these models need to be replicated with longitudinal data before firm conclusions about the causal effect of each variable on health can be drawn. Also, other health outcomes need to be analysed in order to examine the direct and indirect effects of proximal and distal measures on different health outcomes. Despite these shortcomings, this paper offers an example of the way in which specific theoretical relationships between variables can be examined. It should also be noted that SEM is obviously not being advocated as a panacea, but only as one of the ways in which theoretically driven analyses can be specified. We have discussed recent papers that have used other techniques that address the need for a close relationship between theory and analysis: path analysis,41 and graphical chain models.42 Acknowledgements The Whitehall II study has been supported by grants from the Medical Research Council; British Heart Foundation; Health and Safety Executive; Department of Health; National Heart Lung and Blood Institute (HL36310), US, NIH: National Institute on Aging (AG13196), US, NIH; Agency for Health Care Policy Research (HS06516); and the John D and Catherine T MacArthur Foundation Research Networks on Successful Midlife Development and Socio-economic Status and Health. MM is supported by an MRC Research Professorship. We also thank all participating civil service departments and their welfare, personnel, and establishment officers; the Occupational Health and Safety Agency; the Council of Civil Service Unions; all participating civil servants in the Whitehall II study; and all members of the Whitehall II study team. Table 1 Means of psychosocial health indicatorsa for categories of socio-economic position measures . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 Open in new tab Table 1 Means of psychosocial health indicatorsa for categories of socio-economic position measures . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 . . . . Hostility . Hopelessness . GHQb . Self-rated health . . . . f (%) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . Male
 M (SD) . Female
 M (SD) . a High scores on all indicators denotes poor psychosocial health. bGeneral Health Questionnaire. Education high 1 13.7% 5.57 (4.03) 5.28 (3.89) 7.43 (3.27) 7.23 (2.98) 24.35 (10.33) 23.42 (9.49) 2.60 (0.89) 2.28 (0.84) 2 21.9% 6.09 (4.29) 5.60 (4.29) 7.89 (3.50) 7.62 (3.22) 24.60 (10.43) 23.90 (10.10) 2.55 (0.89) 2.50 (0.89) 3 26.7% 6.07 (4.37) 6.04 (4.27) 7.54 (3.42) 7.84 (3.42) 23.53 (10.66) 24.32 (11.09) 2.52 (0.90) 2.51 (0.88) 4 26.7% 6.29 (4.52) 6.13 (4.35) 7.76 (3.41) 7.94 (3.56) 23.48 (10.63) 24.23 (10.28) 2.59 (0.87) 2.58 (0.88) low 5 11.0% 6.56 (4.41) 6.89 (4.44) 8.52 (3.80) 8.05 (3.65) 25.00 (11.27) 23.77 (10.35) 2.73 (0.89) 2.64 (0.87) Test for Linearity P < 0.0005 P < 0.0005 P = 0.001 P = 0.001 P = 0.49 P = 0.43 P = 0.02 P < 0.0005 Occupation high 1 20.1% 4.92 (3.72) 4.88 (3.78) 6.74 (2.99) 6.93 (3.07) 22.30 (9.35) 22.67 (9.20) 2.36 (0.85) 2.31 (0.88) 2 21.4% 5.59 (3.92) 5.54 (3.97) 7.25 (3.07) 7.44 (3.15) 23.51 (9.86) 23.55 (10.68) 2.47 (0.84) 2.44 (0.86) 3 13.6% 5.79 (4.13) 6.04 (4.09) 7.65 (3.31) 7.81 (3.14) 24.12 (11.16) 23.92 (10.06) 2.54 (0.87) 2.49 (0.85) 4 16.5% 6.34 (4.17) 6.03 (4.28) 7.89 (3.42) 7.86 (3.41) 24.12 (11.16) 24.54 (10.15) 2.67 (0.86) 2.57 (0.86) 5 13.6% 7.44 (4.89) 6.92 (4.65) 8.97 (3.99) 8.44 (3.52) 26.44 (12.97) 24.95 (10.96) 2.79 (0.96) 2.61 (0.88) low 6 14.8% 8.04 (5.24) 7.35 (4.90) 9.33 (3.88) 8.83 (3.91) 26.10 (11.13) 25.53 (11.58) 2.92 (0.90) 2.86 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 Income high 1 9.4% 5.16 (3.67) 5.11 (4.03) 6.89 (3.02) 7.08 (3.10) 24.31 (9.76) 23.18 (9.15) 2.42 (0.86) 2.32 (0.83) 2 15.2% 5.47 (3.79) 5.76 (3.99) 7.29 (2.99) 7.48 (3.24) 24.20 (9.83) 24.76 (10.45) 2.48 (0.85) 2.44 (0.91) 2 16.6% 5.62 (4.09) 5.71 (4.02) 7.45 (3.46) 8.00 (3.36) 24.07 (10.76) 24.48 (10.47) 2.55 (0.86) 2.45 (0.86) 4 14.3% 5.95 (4.33) 5.72 (4.39) 7.70 (3.52) 7.46 (3.22) 23.98 (11.18) 24.51 (11.02) 2.53 (0.87) 2.54 (0.88) 5 14.5% 6.58 (4.71) 6.21 (4.49) 8.19 (3.53) 8.10 (3.65) 24.21 (11.27) 23.23 (10.53) 2.58 (0.89) 2.62 (0.85) low 6 30.0% 7.01 (4.75) 6.61 (4.51) 8.37 (3.73) 8.06 (3.49) 24.32 (11.04) 23.74 (10.28) 2.76 (0.93) 2.65 (0.92) Test for Linearity P < 0.0005 P < 0.0005 P < 0.0005 P < 0.0005 P = 0.79 P = 0.46 P < 0.0005 P < 0.0005 Open in new tab Table 2 Effects (standardized regression coefficients) of indicators of socio-economic positiona on psychosocial healthb . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA Open in new tab Table 2 Effects (standardized regression coefficients) of indicators of socio-economic positiona on psychosocial healthb . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA . . Model I . Model II . Model III . Sex . Indicator of socio-economic position . Univariate effect . Direct effect . Direct effect . Indirect effect . a High scores on all measures of social position denote low social position. bHigh score on psychosocial latent construct denotes poorer health. M Education 0.05 (P = 0.01) −0.13 (P < 0.001) −0.13 (P < 0.001) 0.19 (P < 0.001) Occupation 0.33 (P < 0.001) 0.42 (P < 0.001) 0.42 (P < 0.001) −0.02 (P = 0.10) Income 0.17 (P < 0.001) −0.04 (P = 0.10) −0.04 (P = 0.10) NA F Education 0.10 (P < 0.001) −0.05 (P = 0.12) −0.05 (P = 0.12) 0.16 (P < 0.001) Occupation 0.25 (P < 0.001) 0.36 (P < 0.001) 0.36 (P < 0.001) −0.07 (P < 0.001) Income 0.10 (P < 0.001) −0.12 (P < 0.001) −0.12 (P < 0.001) NA Open in new tab Figure 1 Open in new tabDownload slide Direct effects of education, income, and occupation on psychosocial health (Model II), path coefficients shown for men in the Figure Notes: a, b, and c are standardized regression coefficients reflecting the effect of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 5) represent the error terms associated with endogenous variables in the model. Figure 1 Open in new tabDownload slide Direct effects of education, income, and occupation on psychosocial health (Model II), path coefficients shown for men in the Figure Notes: a, b, and c are standardized regression coefficients reflecting the effect of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 5) represent the error terms associated with endogenous variables in the model. Figure 2 Open in new tabDownload slide Direct and indirect effects of education, income, and occupation on psychosocial health (Model III), path coefficients shown for men in the Figure Notes: d, e, f, g, h and i are standardized regression coefficients reflecting the effects (direct and indirect) of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 7) represent error associated with endogenous variables in the model. Figure 2 Open in new tabDownload slide Direct and indirect effects of education, income, and occupation on psychosocial health (Model III), path coefficients shown for men in the Figure Notes: d, e, f, g, h and i are standardized regression coefficients reflecting the effects (direct and indirect) of education, occupation, and income on psychosocial health. ‘e’ terms (1 to 7) represent error associated with endogenous variables in the model. References 1 Davey Smith G, Hart C, Hole D et al. Education and occupation social class: which is the more important indicator of mortality risk? J Epidemiol Community Health 1998 ; 52: 153 –60. 2 Geyer S, Peter R. Income, occupational position, qualification and health inequalities—competing risks? (Comparing indicators of social status). J Epidemiol Community Health 2000 ; 54: 299 –305. 3 Winkleby MA, Jatulis DE, Frank E, Fortmann SP. Socioeconomic status and health: How education, income, and education contribute to risk factors for cardiovascular disease. Am J Public Health 1992 ; 82: 816 –20. 4 Dahl E. Social inequalities in ill-health: the significance of occupational position, education and income results from a Norwegian survey. Sociol Health Illness 1994 ; 16: 644 –67. 5 Blaxter M. Health and Lifestyles. London: Routledge, 1990 . 6 Marmot M, Shipley M, Brunner E, Hemingway H. Relative contribution of early life and adult socioeconomic factors to adult morbidity in the Whitehall II study. J Epidemiol Community Health 2001 ; 55: 301 –07. 7 Ben-Shlomo Y, Davey Smith G. Deprivation in infancy or in adult life: which is more important for mortality risk? Lancet 1991 ; 337: 530 –34. 8 Department of Health. Variation in Health: what can the Department of Health and the NHS do? London: Department of Health, 1995 . 9 Mann SL, Wadsworth ME, Colley JR. Accumulation of factors influencing respiratory illness in members of a national birth cohort and their offspring. J Epidemiol Community Health 1992 ; 46: 286 –92. 10 Marmot M. Aetiology of coronary heart disease. BMJ 2001 ; 323: 1261 –62. 11 Power C, Manor O, Fox AJ, Fogelman K. Health in childhood and social inequalities in health in young adults. J R Statist Soc 1990 ; 153: 17 –28. 12 Power C, Mathews S, Manor O. Inequalities in self rated health in the 1958 birth cohort: lifetime social circumstances or social mobility? BMJ 1996 ; 313: 449 –53. 13 Power C, Mathews S, Manor O. Inequalities in self-rated health: explanations from different stages of life. Lancet 1998 ; 351: 1009 –14. 14 Davey Smith G, Hart C, Blane D, Gillis C, Hawthorne V. Lifetime socioeconomic position and mortality: prospective observational study. BMJ 1997 ; 314: 547 –52. 15 Bartley M, Sacker A, Firth D, Fitzpatrick R. Understanding social variation in cardiovascular risk factors in women and men: the advantage of theoretically based measures. Soc Sci Med 1999 ; 49: 831 –45. 16 Lynch JW, Kaplan GA, Salonen JT. Why do poor people behave poorly? Variation in adult health behaviours and psychosocial characteristics by stages of the socioeconomic lifecourse. Soc Sci Med 1997 ; 44: 809 –19. 17 Krieger N, Williams DR, Moss NE. Measuring social class in US public health research: Concepts, methodologies, and guidelines. Annu Rev Public Health 1997 ; 18: 341 –78. 18 Kaplan GA, Salonen JT. Socioeconomic conditions in childhood and ischaemic heart disease during middle age. BMJ 1990 ; 301: 1121 –23. 19 Vagero D, Leon D. Effect of social class in childhood and adulthood on adult mortality. Lancet 1994 ; 343: 1224 –25. 20 Wannamethee SG, Whincup PH, Shaper G, Walker M. Influence of father’s social class on cardiovascular disease in middle-aged men. Lancet 1996 ; 348: 1259 –63. 21 Duncan OD. How destination depends on origin in the occupational mobility table. Am J Sociol 1979 ; 84: 793 –803. 22 Ross CE, Wu C-L. The links between education and health. Am Sociol Rev 1995 ; 60: 719 –45. 23 Mare RD. Socio-economic careers and differential mortality among older men in the United States. In: Vallin J, D’Souza S, Polloni A (eds). Comparative Studies of Mortality and Morbidity: Old and New Approaches to Measurement and Analysis. Oxford: Oxford University Press, 1990 , pp.362–87. 24 Arber S. Comparing inequalities in women’s and men’s health: Britain in the 1990s. Soc Sci Med 1997 ; 44: 773 –87. 25 Arber S, Cooper H. Gender differences in health in later life: the new paradox? Soc Sci Med 1999 ; 48: 61 –76. 26 Sacker A, Bartley M, Firth D, Fitzpatrick R. Dimensions of social inequality in the health of women in England: occupational, material, and behavioral pathways. Soc Sci Med 2001 ; 52: 763 –81. 27 Sacker A, Firth D, Fitzpatrick R, Lynch K, Bartley M. Comparing health inequality in men and women: prospective study of mortality 1986–96. BMJ 2000 ; 320: 1303 –07. 28 Marmot MG, Davey Smith G, Stansfeld S et al. Health inequalities among British civil servants: the Whitehall II study. Lancet 1991 ; 337: 1387 –93. 29 Cook WW, Medley DM. Proposed hostility and Pharisaic values for the MMPI. J Appl Psychol 1954 ; 38: 414 –18. 30 Beck AT, Weissman A, Lester D, Trexler L. The measurement of pessimism: The hopelessness scale. J Consult Clin Psychol 1974 ; 42: 861 –65. 31 Goldberg DP. The Detection of Psychiatric Illness by Questionnaire. (Institute of Psychiatry Maudsley mongraphs No. 21). London: Oxford University Press, 1972 . 32 Hair JF, Anderson RE, Tatham RL, Black WC. Multivariate Data Analysis. NJ: Prentice-Hall, Inc., 1998 . 33 Mueller RO. Basic Principles of Structural Equation Modeling. New York: Springer Verlag, 1996 . 34 Vassend O, Skrondal A. The role of negative affectivity in self-assessment of health. J Health Psychol 1999 ; 4: 465 –82. 35 Sacker A, Bartley M, Firth D, Fitzpatrick R. Dimensions of social inequality in the health of women in England: occupational, material, and behavioral pathways. Soc Sci Med 2001 ; 52: 763 –81. 36 Arbuckle JL, Wothke W. Amos 4.0 User’s Guide. Chicago: Small Waters Corp., 1995 . 37 Arbuckle JL. Full-information estimation in the presence of incomplete data. In: Marcoulides GA, Schumacker RE (eds). Advanced Structural Equation Modeling Techniques. Mahwah, NJ: Lawrence Erlbaum Associates, 1996 , pp.243–77. 38 Lee S, Hershberger S. A simple rule for generating equivalent models in covariance structure modeling. Multivariate Behav Res 1990 ; 25: 313 –34. 39 Krieger N, Fee E. Social class: the missing link in US health data. Int J Health Serv 1994 ; 24: 25 –44. 40 Valkonen T. Adult mortality and level of education: a comparison of six countries. In: Fox J (ed.). Health Inequalities in European Countries. Aldershot: Gower, 1989 , pp.142–72. 41 Weitkunat R, Wildner M. Exploratory causal modeling in epidemiology: Are all factors created equal? J Clin Epidemiol 2002 ; 55: 436 –44. 42 Didelez V, Pigeot I, Dean K, Wister A. A comparative analysis of graphical interaction and logistic regression modelling: self-care and coping with a chronic illness in later life. Biom J 2002 ; 44: 410 –32. © International Epidemiological Association 2002

Journal

International Journal of EpidemiologyOxford University Press

Published: Dec 1, 2002

There are no references for this article.