trends in health and socio-economic status, proximal and

Human Development, Capability and Poverty International Research Centre (HDCP-IRC) WORKING PAPER SERIES ISSN1974-1952 20/2012

Trends in Health and Socio-Economic Status, proximal and remote determinants in a cohort study Agnese Peruzzi

Human Development, Capability and Poverty International Research Centre (HDCP-IRC)

HDCP-IRC WORKING PAPER № 20/201

Human Development, Capability and Poverty International Research Centre

– A research centre promoted by the Institute for Advanced Study (IUSS-Pavia) – www.iusspavia.it/hdcp

2

About HDCP-IRC Launched in the year 2006 and promoted by the Institute for Advanced Study of Pavia (IUSS) – the Human Development, Capability and Poverty International Research Centre (HDCP-IRC) represents an innovative initiative in the academic field, namely the creation of a notable multidisciplinary forum for academic research and high-quality training activities in the fields of human development, including the study of topics such as the quality of life, poverty and inequality, global justice, human rights, gender issues and sustainable development inter alia. The principal theoretical foundation of the HDCP-IRC’s activities is represented by the capability approach, originally formulated by Amartya Sen, 1998 Nobel Laureate in Economics, and subsequently enriched and developed by numerous scholars from different disciplines, in particular the philosopher Martha Nussbaum. The capability approach offers a holistic theoretical framework for conceptualising well-being, quality of life and development, assuming a multidimensional perspective compared to the traditional approaches based on income and utility metric.

Director: Enrica Chiappero-Martinetti

Website: www.iusspavia.it/hdcp

Contact: [email protected]





3

HDCP-IRC WORKING PAPER 20/2012

Trends in Health and Socio-Economic Status, proximal and remote determinants in a cohort study Agnese Peruzzi November, 2013

Abstract

In this paper, we contribute to discussion on the SES-health gradient in the UK, using a sample of about 10000 children drawn from the 1970 British Cohort Study. We apply latent growth curve modeling techniques to analyse the relationship between the progression of different aspects of socioeconomic disadvantages and physical and psychological health during adulthood, as well as the lasting impact of childhood circumstances on such relationships. We consider three distinct but interconnected domains of adult SES: education, income, and social class and we disentangle the complex mechanisms underlying the association between the domains of SES and health during adulthood. Our results confirm that the progression of the SES-health gradient through adulthood is founded upon complex relationships generated by casual impacts of SES on health and by selection mechanisms, both beginning in early life and continually operating through the life cycle. Indeed, on the one hand we find that childhood health and socioeconomic circumstances play a crucial role in determining SES position and health status later in life, being also significantly related to the trajectory of SES and health over time. On the other hand, we see that feedbacks mechanisms operate in adulthood as well, and we emphasise that the processes through which the education, income and social class affect and are affected by health during adulthood differ according to the specific domain of SES considered.

Keywords SES-health gradient, Childhood circumstances, Latent Growth Curve, British Cohort Study





4

1. Introduction The existence of a strong association (often called the gradient) between socio-economic status (SES) and health in adulthood has been documented by a number of researchers from different fields of study, including economics, demography, sociology, medicine, and epidemiology (Adams et al., 2003; Adler et al., 1994; Deaton and Paxson, 1998; Van Doorslaer et al., 1997; Wilkinson and Marmot, 2003). In a comprehensive review of this literature, Goldman (2001) observes that the SES-health gradient has been found for a wide range of indicators of SES (e.g. income, education, and occupation) and health variables (such as self-rated health status, illnesses, disability, mortality, and psychological health). The robustness and universality of such association has stimulated considerable attention in both academic and policy research, aimed to advance the understanding of the determinants of socioeconomic disparities in health. Social and medical scientists have hypothesized two alternative mechanisms to explain the phenomenon, namely social causation and social selection. The social causation hypothesis postulates that individuals with low SES develop health problems as a consequence of living with adversity, through a complex set of indirect causal processes, including, for example, the access to health care services (Cohen, Farley, and Mason 2003; House 2002; Lantz et al. 2005), unhealthy behaviours (Mirowsky and Ross 1999), and lower knowledge on how to self-manage illness and disease (Goldman and Smith 2002). The social selection hypothesis, also referred to as reverse causation or social drift, posits an inverse causality, whereby individuals are sorted into socioeconomic status groups according to their health. For example, unhealthy individuals may have higher medical care expenditures and a reduced ability to work, which in turn may lead to lower income and worse job position. A large body of literature during the past decades has provided valuable theoretical insights and empirical evidence into either the selection or causation mechanisms accounting for links between different SES dimensions and health. However, there is no common agreement on the direction of causality. On the contrary, there has been a growing recognition that a single theory explaining socioeconomic gradients in health is not reasonable (Cutler et al., 2008). Indeed, not only the size of the social inequalities and the patterns of causality differ according to the specific domain of SES considered, but also there exist potentially continuous feedbacks, operating over individuals’ life course, between different domains of SES and measures of health, so that disentangling causal from selection mechanisms turns out to be very demanding. Hence, different dimensions of SES relate to health in different ways and those specific mechanisms should be assessed precisely. For example, Herd, Goesling, and House (2007) test the social causation hypothesis and examine the effects of two important SES domains, e.g. education and income, on both health during childhood and its progression through adulthood. They find that while education plays a bigger role than income in predicting the baseline level of chronic conditions or functional limitations, income dominates in explaining the progression of health problems over time. Looking at the relevance of reverse mechanisms, social selection is found to be important for understanding the association between health and either income





5

(van Doorslaer and Koolman, 2004) or career opportunities in adulthood (van de Mheen, Stronks and Mackenbach, 1998), since healthier individuals are usually capable of working more hours, gaining higher incomes and taking advantages of new career opportunities. However, those patterns are very unlikely when education is used as indicator of SES, because this domain tends to remain stable after early adulthood. Such intricate scenario is further complicated by recent contributions highlighting the crucial role of childhood health and socioeconomic circumstances in determining the co evolution between SES and health through adulthood. Currie (2009), in a review of this literature, finds robust evidence confirming that parents’ socioeconomic status is linked to child health, which in turn, being related to both educational attainments and abilities to effectively participate into the labour market (Palloni, et al., 2009; Smith, 2009), affects health status and SES position later in life. However, the bulk of this literature investigates the impact of childhood antecedents on adult SES and health in cross-section (see for example Case et al., 2005), while the early life influences on the progression of the SES-health gradient through adulthood have not been deeply investigated in a comprehensive manner so far. There are indeed studies establishing the links either between childhood health and adults’ trajectories of self-rated health (Haas, 2007), functional limitations (Haas, 2008) and depressive symptoms (Elovainio et al., 2011; Quesnel-Valléeand Taylor, 2012), or between childhood socioeconomic circumstances and adult trajectories of SES and health (Halleröd and Gustafsson, 2011). However, these studies either analyse the effect of early life antecedents on single adult trajectories, or, when interested in analyzing the progression of the SES-health gradient during adulthood, they control only for childhood socioeconomic circumstances. Building upon the above discussion, this paper aims to provide a unique framework suitable for investigating the lasting effects of childhood health and socioeconomic circumstances on the co-progression of health outcomes and SES domains during adulthood. Hence, in this paper we contribute to the existing literature on the SES-health gradient in the following ways. We first analyze the initial level and progression of three different SES attainments (education, income, and occupation) and two health domains (physical and psychological) over an 8-year period covering the early adulthood of a nationally representative British cohort. Second, we assess the long-term consequences of childhood health and socioeconomic circumstances on the evolving interaction between SES and health trajectories through adulthood. Third, we investigate the specific mechanisms by which trajectories of SES interact with each other and either affect or are affected by physical and psychological health trajectories in early adulthood. We believe that make progress in this area and improving understanding of the multiplicity of mechanisms linking SES indicators and health outcomes is crucial for designing policies aimed at addressing the SES-health gradient, especially under binding public budget constraints. Indeed, strategies for reducing socioeconomic inequality in health may be unsuccessful if they overlook the multi-faceted contributors of the gradient. For example, if for





6

a specific SES dimension selection mechanisms explain the health gradient, then policy intervention affecting that SES domain would be less effective in improving health. This paper is structured as follows. In Section 2 we describe our data. The statistical model is presented in Section 3, while Section 4 discusses the results. Conclusions are drawn in Section 5. 1. Data and measures 1.1. Dataset

Analysing the long-term consequences of childhood circumstances on the interrelated

adults’ trajectories of health and socioeconomic status requires longitudinal studies, beginning in early life and repeatedly collecting information through adulthood. Birth cohort surveys, which follow large and nationally representative samples of individuals and collect detailed data tracking their life course experiences, are well suited to this end. This study uses data from a UK birth cohort survey, the 1970 British Cohort Study (BCS70), which has followed all individuals (17,287) born in Great Britain between April 5 and 11, 1970. The cohort members have been contacted eight times since the first initial survey at birth, at 5, 10, 16, 26, 30, 34, 38 and 42 years of age.1 Our analysis is based on six out of the nine waves of information collected to date to gather information from the whole cohort. Information on adults’ physical and psychological health as well as their socioeconomic status—including income, education, and social class—are based on repeated measurements collected in three adulthood waves (the 26-year follow-up, the 30-year follow up and the 34-year follow up). Early socio-economic circumstances and childhood initial health endowments, which are likely to be important factors in determining adult outcomes, are derived from the three childhood waves (the Birth survey, the 5-year and the 10-year follow-up surveys).

As for the sample, we restrict our analysis to the 9779 individuals who responded to at least two of the three adulthood follow-up surveys considered (56.6% of the initial sample). Since we are also interested in examining the early life influences on adult outcomes, the sample is further narrowed by discarding records for individuals (104) who were not interviewed in all of the childhood waves up to age 10. The final sample consists of 9675 individuals of whom 52.7% female and 47.3% male.2

1Access to the data via the UK data archive, University of Essex, is gratefully acknowledged. 2Note that although the BCS70 has inevitably been affected by attrition, the UK Office for National Statistics (1999, p.11) states: “Analysis of differential response comparing achieved samples and target samples for any follow-up,.., shows that the achieved samples are broadly representative of the target sample”.





7

1.2. Measures of socio-economic status (SES) and health in adulthood

This study focuses on three distinct but interconnected domains of SES: education, income, and social class, in order to analyse whether the mechanisms linking different dimensions of SES to health diverge or coincide. Summary statistics for the key variables of interest in adulthood are presented in Table 1.

Adulthood variables Full Sample Women Men

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Mean (S.D.)

Age 26 Age 30 Age 34 Age 26 Age 30 Age 34 Age 26 Age 30 Age 34

Socioeconomic measures

Education 2.185 (1.267)

2.221 (1.222)

2.142 (1.319)

No qualification (%) 11.0 9.3 13.0 NVQ level 1 (%) 16.4 15.5 17.5 NVQ level 2 (%) 39.2 41.6 36.4 NVQ level 3 (%) 10.0 11.0 8.6 NVQ level 4/ above (%) 23.4 22.6 24.5

Income 942 (512)

1157 (606)

1388 (818)

816 (448)

957 (544)

1085 (716)

1082 (541)

1351 (600)

1685 (804)

Social class

2.944 (1.231)

3.009 (1.192)

3.060 (1.220)

3.066 (1.122)

3.092 (1.111)

3.135 (1.164)

2.804 (1.332)

2.932 (1.258)

2.992 (1.265)

Unskilled (%) 2.4 2.0 2.1 1.8 2.2 1.6 3.1 1.8 2.5 Semi-skilled (%) 13.5 10.7 11.1 12.3 10.9 13.5 14.9 10.6 8.9 Manual skilled (%) 17.3 20.1 19.1 7.8 7.5 6.4 28.4 31.7 30.7 Non-manual skilled (%) 27.3 25.1 21.0 38.1 38.6 32.1 14.8 12.6 11.0 Managerial (%) 33.0 35.7 39.9 35.6 36.5 41.0 29.9 35.0 38.8 Professional (%) 6.5 6.4 6.8 4.4 4.3 5.4 8.9 8.3 8.1

Health measures

Self reported health 1.745 (0.641)

1.838 (0.705)

1.945 (0.852)

1.776 (0.637)

1.839 (0.708)

1.972 (0.880)

1.706 (0.644)

1.836 (0.701)

1.914 (0.818)

Excellent (%) 35.6 32.3 32.9 32.9 32.3 32.8 38.9 32.3 33.0 Good (%) 55.3 53.5 46.1 57.5 53.6 44.6 52.6 53.4 47.7 Fair (%) 8.1 12.3 14.7 8.5 12.0 15.2 7.6 12.6 14.2 Poor or very poor (%) 1.0 1.9 6.3 1.1 2.1 7.4 0.9 1.7 5.1

Malaise index

1.588 (1.690)

1.591 (1.820)

1.662 (1.892)

1.941 (1.786)

1.826 (1.851)

1.888 (1.950)

1.204 (1.486)

1.326 (1.746)

1.407 (1.793)

Notes:(1) Sample sizes—9675 observations in total, 5095 women and 4580 men. (2) Income amounts are expressed in 2005 £.





8

Our measure of education, the education score, is the highest educational qualification

achieved by age 26, classified into 5 categories:3 cohort members who had left school with no qualifications, those with qualifications up to NVQ level 1 (CSE obtained and their training equivalents), those with qualification up to NVQ level 2 (GCSE or ordinary secondary qualifications and their equivalents), those with advanced secondary education or NVQ level 3 (‘A’ levels or university entry qualifications and their equivalents), or NVQ level 4 and above (degree level or equivalent). The mean education score is 2.185 (the percentage of cohort members in each response category is reported in Table 1) and women, on average, are better educated than men. The economic aspect of SES is captured by individuals’ net monthly income - after deductions but including overtime and bonuses – collected at age 26, 30 and 34 and expressed in June 2005 pounds.4 As expected, there is an increase in mean net monthly income over time, both for men and women. However, women on average reported lower income than men at each wave considered (see Table 1). The last indicator of SES considered is social class, which is based on occupation and measured by the Registrar General’s measure of Social Class (RGSC). The RGSC is defined according to job status and the associated education and prestige and classified into six ordinal categories: Unskilled; Semi-skilled; Manual skilled; Non-manual skilled; Managerial; Professional. There is a general tendency toward upward social mobility over time, for both men and women. Interestingly, despite women on average reported lower income than men, their distribution is more skewed toward higher social class scores than that of men at each wave.

As for the measures of health, we follow the World Health Organization’s definition, which conceives it as a multi-dimensional phenomenon, encompassing both physical and psychological wellbeing. Our primarily indicator of physical health is the individual’s self-reported health status, which provides a subjective assessment of the respondent’s health at age 26, 30 and 34. Subjective health status has been shown to be an important indicator of health, which significantly predicts future health status, even when controlling for objective indicators such as physician assessed health status and health-related behaviours (Idler and Benyamini, 1997). This variable is measured using a four-point scale:50=excellent; 1=good; 2=fair; 3=poor or very poor. Members of the BCS70 report worse health at higher ages, wherein the mean score rises from 1.745 at age 26 to 1.945 at age 34 and the cumulative percentage of individuals reporting themselves to be in excellent or good health falls from 88.9% at age 26, to 85.8% at age 30, and to 79% at age 34. Both men and women showed similar patterns, although, consistently with results from previous studies (see for example, Case et al., 2005), women on average report worse health than man. As for the assessment of psychological

3The classification is based on a scale related to the National Vocational Qualification (NVQ) levels (Makepeace and al., 2003) 4 The adjustment for changes in the price level is based on the consumer price index – June 2005 - supplied by the UK Office for National Statistics. 5 The original scale in the 34-year follow up survey was a five-item scale, from excellent to very poor. For comparability reasons, we collapse the responses “poor” and “very poor”, into one category.





9

health, our indicator is based on the short-version of the Malaise Inventory (Rutter et al., 1970), which is used to measure the prevalence of depressive symptoms across cohort members during their life. The Malaise inventory in its standard format comprises a 24-item list of symptoms from Cornell Medical Index, covering emotional disturbance and associated physical symptoms. The full list was collected at age 26 and 30, while, at age 34, only 9 out of the 24 questions were asked. Thus, for comparability, we use the 9-item sub-scale that is common to all the waves of interest. More specifically, respondents were asked whether (1) they felt tired most of the time, (2) they often felt depressed, (3) they often got worried about things, (4) they often got into a violent rage, (5) they suddenly became scared for no good reason, (6) they were easily upset of irritated, (7) they were constantly keyed up and jittery, (8) every little thing got on their nerves and warn them out, and (9) their heart often raced like mad. For each questions, 1 point is awarded for every ‘yes’ response, 0 points for every ‘no’ response. An overall Malaise index for a cohort member is the sum across the individual variables and thus ranges between a minimum score of 0 (no symptoms) and a maximum of 9 (9 symptoms). At first sight, there is a slight tendency toward an increase in the prevalence of depressive symptoms across cohort members, with a mean score rising from 1.59 at age 26 to 1.66 at age 34. This increasing pattern is confirmed among men but not among women, who report the highest average number of depressive symptoms (1.94) at age 26. In general, women report more depressive symptoms than man (see Table 1), but the gender gap decreases with age. 1.3. Measures of the early life antecedents

The influence of childhood disadvantages on adult outcomes in terms of social status and

health is assessed using a broad set of early life indicators related to both childhood health and socioeconomic circumstances. Summary statistics for key variables are presented in Table 2. Childhood initial health endowments are measured by means of three indicators: birth weight, mother’s smoking habits during pregnancy, and early mental health problems. Low birth weight (less than 2500 grams) is a strong indicator of poor general health of newborns and a key determinant of infant survival, health, and adult outcomes (Black et al., 2007). In our study, the incidence of low birth weight amongst the population is 5.8%. Mothers’ smoking habits during pregnancies are associated with early foetal growth, which in turn increases the risk of adverse birth outcomes and growth rate in early childhood (Sydsjö, 2011). We observe that 39% of cohort members’ mothers reported that they smoked during pregnancy, among which 12% reported heavy smoking (more than 14 cigarettes per day). As for the measure of early mental health problems, the same set of questions identifying the prevalence of depressive symptoms during adulthood (the short-version of the Malaise Inventory) was administered to the mother in relation to her 5-years-old child. In order to facilitate results interpretation, we follow Thompson et al. (2003) and classify children’s total Malaise score





10

(previously ranging from 0 to 9) into three levels of severity: “normal” scores for those up to the 80th percentile (82.5% of our sample), “moderate” problem scores, between the 80th and 95th percentile (13.5%) and “severe” problem scores, above the 95th percentile (4%).

Childhood socioeconomic circumstances are captured through multiple indicators. The economic aspects of socioeconomic status are measured by a synthetic indicator of household endowment of appliances at age 5 (family had fewer than four appliances out of phone, fridge, colour tv, washing machine, dryer and car) and of being at risk of poverty at age 10 (equivalised household income falling below 60% of median income). In our study, 18.4 % of children suffered from appliances deprivation at age 5 and 20.3% were at risk of poverty at age 10. Also included in the analysis are measures of parental highest level of educational qualification and social class. Mother’s and father’s qualification is classified into 4 categories ranging from no qualification to university degree. Parental social class is derived from fathers’ occupation (or mothers’ if missing) and is classified in the same way as social status during adulthood (see above).





11

Table 1 Summary statistics for the early life antecedents

Early life antecedents Wave (age) %

Childhood initial health endowments

Birth weight 1970 (Birth) More than 2.5 grams 94.2 Less than 2.5 grams 5.8

Mother Smoking habits during pregnancy 1970 (Birth) Non smoker 61.1 Moderate smoker 27.0 Heavy smoker 11.9

Early mental health 1975 (Age 5) Normal score 82.5 Moderate problems 13.5 Severe problems 4.0

Childhood socio-economic circumstances

Household deprivation in appliances (phone, fridge, colour TV, washing machine, dryer and car)

1975 (Age 5)

More than four appliances 81.6 Fewer than four appliances 18.4

Income poverty (equivalised income below 60% of the median income) 1980 (Age 10)

Not at risk of poverty 79.7 At risk of poverty 20.3

Mother’s highest qualification 1975 (Age 5) A levels/ Degree 12.1 O levels 19.7 Vocational 15.8 No qualification 52.4

Father’s highest educational qualification 1975 (Age 5) A levels/ Degree 24.9 O levels 17.6 Vocational 12.6 No qualification 44.9

Parental social class 1975 (Age 5) Professional/Managerial 27.6 Non-manual skilled 10.5 Manual skilled 44.1 Semi-skilled/ Unskilled 17.8

Control variables

Child’s sex 1970 (Birth) Female 52.7 Male 47.3





12

2. Disentangling the relationship between health and SES in adulthood

2.1. Method

In order to assess how changes in SES status affect changes in physical and mental health throughout adulthood as well as the role that circumstances in early life play in determining such coevolution, we adopt Latent Growth Curve (LGC) model, in virtue of its flexibility in handling a variety of different issues involving growth processes (Kaplan, 2008). LGC is a statistical technique used in the structural equation modeling framework designed to describe developmental trajectories in terms of parsimonious models, allowing for the estimation of inter-individual variability in intra-individual patterns of change over time (Willett and Sayer, 1994). The basic growth model has two components: a fixed effect representing a single value that exists in the population (e.g. the mean of the trajectory pooling of all the individuals within the sample), and a random effect, which represents the random probability distribution around that fixed effect (e.g., the variance of the individual trajectories around the group means). In analytical terms, in a LGC model the unobserved trajectory of an an outcome for individual at time is modeled by two latent variables, one for the intercept of the growth curve and one for its slope, and may be expressed as:

(1) where is a vector representing the repeated measures for each person , is a

vector of factor loadings representing time, which are fixed under the assumption that the rate of change is constant across time (linear trajectory); and are respectively, random intercepts and slopes for the individual , and is a vector of individual measurement errors, which are assumed to be random with respect to time. In an unconditional growth trajectory (i.e. trajectory with no predictors), the individual latent intercept can be expressed as a function of a fixed component, namely, the overall mean group intercept or starting point ( ) and a random component, e.g. the individual deviation from the mean group intercept ( ):

(1a) In the same vein, the individual latent slope can be expressed as a function of the mean slope, or rate of change within the general population ( ) and the individual deviation ( ):

(1b)





13

For example, let us consider that we have repeated observations of income collected at 3 time periods ( ) and that we are interested in analyzing the progression of income and the shape of its trajectory over time. We further assume that the change in income is a linear function of time and, coherently, we fix the factor loadings for the slope to predetermined values ( ). Figure 1 presents a graphical representation of the model for income.

Figure 1. A visual exemplification of the latent growth trajectory of income

INCOME at age 26 (t0)



Intercept (I) of the Income trajectory

Slope (S) of the Income

trajectory

λO==0% λ1==1% λ2==2%

Notes: (1) The ellipses represent latent variables, while the rectangles represent observed indicators. (2) Loading on the intercepts are all fixed to 1, while loadings on the slope are fixed to the lambdas coefficients shown in the figure, in order to model a linear trajectory of income. (3) The figure can be generalized to represent the other unconditional trajectories, such as the one estimated for social class.

Fitting a linear latent growth curve means testing whether the measures of income

collected at the 3 points in time can be modeled as a latent trajectory synthetized by two factors: intercept and slope. For both factors, a mean and a variance are estimated according to equations (1a) and (1b). The mean of the latent intercept describes the average initial level of income in the general population6 and the mean of the latent slope describes its rate of change over time: mean intercept and mean slope therefore summarize the underlying trajectory pooling of the entire sample, and provide an estimate of the fixed effects within the population. In contrast, the variances of the intercept and slope represent the between-person variability in, respectively, the individual initial income, and its rate of growth, and represent therefore the random effects component. Smaller random effects (smaller variances of intercepts and slopes) imply that the parameters describing the general trajectory of income are more similar across the individuals, while larger random effects (larger variances) indicate that there are greater individual differences in the magnitude of the trajectory parameters around the mean values.

The LGC approach has several features that make it appropriate in the context of the present study. First, it can be extended to model curvilinear trajectories, wherein the rate of change is not constant across time. Indeed, the assumption of linear change is not always 6Note that in general, the intercept represents the estimated initial level and its value may slightly differ from the actual mean for the first measure of income.





14

realistic, since some processes may accelerate or decelerate, as people get older. Meredith & Tisak (1990) propose an approach for identifying the correct functional form of the trajectory, which entails freely estimating a subset of the factor loadings associated with the slope, so that the change optimally corresponds to the unique characteristics of the data under study. Considering again the income trajectory over 3 time periods, the nonlinear growth curve approach requires, for purposes of model identification, fixing the first and second loadings respectively to zero and to one, while freeing the loading associated to the third wave so that time metrics are empirically determined ( ). Second, LGC can be extended to simultaneously model multiple outcome growth processes, either parallel or sequential, and allows the developmental process of one domain to interact (parallel outcomes) or predict (sequential outcomes) the developmental process of a different domain (see, e.g., B. Muthén and Curran, 1997). Thus, for example, we can assess the degree to which the initial level of physical health in early adulthood is correlated with the initial level of income and also the extent to which initial health affects the rates of growth of income, and vice-versa, through a joint analysis of the two processes. Third, the model can be expanded to include exogenous predictors of initial status and growth (conditional growth trajectory), with the goal of determining what variables are associated with individuals who report higher versus lower intercepts or steeper versus flatter slopes. The conditional growth trajectory is still modeled as in Eq. (1). However, the individual latent intercept and slope are now treated as random variables that are explained also by a vector of covariates ( ):

(2a) (2b)

wherein the matrixes and respectively describe the effect of covariates on the latent

intercept and slope. 2.2. The empirical framework

The empirical analysis proceeds as follows. We first fit a series of unconditional growth models, in order to establish whether income, social class, physical health, and psychological health could be measured as two-factor latent growth processes for the adults in the cohort study. Since education tends to remain stable throughout adulthood, we do not model this domain of SES as a latent growth process. We fit separate models for women and men, in order to control for potential gender differences, and we freely estimate the factor loading associated to the third wave of each process, thus allowing for nonlinear growth trajectories. Once having established the correct functional form of each trajectory separately, we fit a unique conditional model, through which we analyse the lasting long-term influences of early life antecedents on





15

the jointly estimated trajectories of socio-economic status and health in adulthood. At this stage, we also include cohort members’ highest educational qualification achieved by age 26, in order to explore the mechanisms linking this important domain with both childhood circumstances and the four adult trajectories described. The model, specified in accordance with Figure 2, has a twofold purpose. First, it aims at testing whether the initial health endowments and childhood socioeconomic circumstances significantly predict the highest qualification achieved by age 26 and the initial level and rate of change of the four growth processes estimated. Second, by modeling simultaneously multiple growth processes, it allows us to follow the progression of the SES-health gradient over time and to assess whether changes in SES status are associated with changes in physical and mental health throughout early adulthood. In such a way, we account for possible interdependencies and cross-domain influences between SES and health domains and we analyse the extent through which different socioeconomic disadvantages either cause health to vary, or are caused by health changes.

Figure 2. Early life antecedents and growth trajectories of socio-economic status and health

EARLY LIFE ANTECEDENTS

Initial health endowments •  Birth weight •  Prenatal smoking •  Early mental problems Socio-economic circumstances •  Household deprivations •  Income poverty •  Mother’s qualification •  Father’s qualification •  Parental social class

INCOME INTERCEPT

PHISICAL HEALTH INTERCEPT

DEPRESSION INTERCEPT

SOCIAL CLASS INTERCEPT

INCOME SLOPE

PHISICAL HEALTH SLOPE

DEPRESSION SLOPE

SOCIAL CLASS SLOPE

HIGHEST QUALIFICATION





16

All analyses are carried out using STATA v. 12.1 and MPlus v. 6 and are tested using the maximum likelihood estimator with robust standard errors and chi-square (MLR). In assessing models fit, the following criteria (Hu and Bentler, 1999) are used: (i) the Comparative Fit Index (CFI), where values ≥0.95 indicate an excellent fit and values >0.90 an adequate fit, (ii) the Root Mean Square Error of Approximation (RMSEA) where values ≤0.06 are considered as indication of good fit and below 0.08 of adequate fit, (iii) the Standard Root Mean Square Residual (SRMR) where values ≤0.05 indicate a good fit and below 0.10 an adequate fit. We do not rely on the chi-square test since is directly affected by sample size: for big samples, even small differences may become significant (MacCallun, Browne and Sugawara, 1996). 3. Results 3.1. Unconditional growth trajectories

In the first step of the analysis, four unconditional growth models are fitted to data in order to establish separately the correct shape of each trajectory, both among men and women. The findings suggest that all but the trajectory of psychological health can be modeled as linear processes. Therefore, in the successive analyses we set the factor loadings for the trajectories of income, social class and physical health to predetermined values, while we freely estimate the loading associated to the third wave in the trajectory of psychological health. Table 3 reports the parameter estimates for the unconditional growth trajectories of socio-economic status and health7. The descriptive goodness-of-fit indices show that the four unconditional trajectories fit the data extremely well, both within the general population (CFI=0.998; RMSEA=0.012; SRMR=0.009) and in the separate models for women (CFI=0.995; RMSEA=0.017; SRMR=0.014) and men (CFI=0.994; RMSEA=0.020; SRMR=0.014). Looking first at the linear trajectory of income within the full sample, the mean slope is significant and positive, thus suggesting that individuals’ income tends to increase by 227 pounds (2.270*100) per time point of measurement as individuals age from their 26s into their early 34s, from an average initial level of about 900 pounds (9.050*100). Different fixed effects parameters characterize the progression of income among men and women. In general, female have a lower income at age 26 and a flatter trajectory of growth compared to male. As far as the random effects are concerned, we found statistically significant differences in the initial level and the rate of change and therefore some individuals have higher (or lower) starting income and increase (or decrease) their income at a different rate than the average growth rate. More specifically, compared to women, men are characterized by a higher between-person variability in the individual initial income (12.751 vs 9.661) and lower

7The results obtained by fitting the four growth trajectories separately and jointly were similar and did not reveal any substantial difference regarding either the structure of the relationships or the fit of the model. Therefore, for the sake of parsimony, we report only the results of the trajectories for the unique measurement model.





17

between-person variability in the rate of growth (6.586 vs 4.789). Finally, we found a significant and positive covariance between intercept and slope, both in the full sample and in the separate models for female and male, which suggests that the higher the initial levels of income, the higher its growth rates over time and vice-versa: the lower the initial level, the flatter the trajectory of growth.

The results of the social class trajectory for the full sample confirm the existence of a general tendency toward upward mobility over time, where individuals begin the trajectory at age 26 with a mean of approximately 2.86 on the social class scale and their position improves on average by around 0.09 units at each wave. The variance of both the intercept and slope is significant and appreciable, and therefore there exist differences in the magnitude of the trajectory parameters around the mean values. Finally, the negative covariance between the average initial social class and its rate of change (-0.067, p<0.001) shows ceiling or floor effects and reflects the fact that individuals who already have the most prestigious position at

cannot advance further. As for the separate models for male and female, the results show that, compared to male, female belong in general to higher social classes at age 26 ( ; while ), but have a flatter rate of growth over time ( ; while ).

As for the linear trajectory of physical health, the mean of latent intercept within the full sample is 1.751 (p < 0.001) and the mean of the latent slope is 0.096 (p < 0.001). Hence, individuals begin the trajectory at age 26 with a mean of approximately 1.75 on the physical health scale and their health declines on average by around 0.10 units at each wave. Again, the significant variance of both the intercept and slope suggests that there are appreciable individual differences in the magnitude of the trajectory parameters around the mean values. Interestingly, the health status at the initial time of measurement appeared to be unrelated to health changes over time, as illustrated by the non-significant covariance of -0.08 between the latent slope and intercept. Similar random effects characterize the progression of physical health among men and women, while, with regards to the fixed effects, the results show that female, compared to male, report on average lower initial level of health at age 26 (1.774 vs 1.726) but have a flatter rate of decline (0.091 vs 0.102).

Looking at the trajectory of psychological health, we find important gender differences in the progression of depressive symptoms over time, which would be masked if we considered the population as a whole. Indeed, the results from the separate models for males and females reveal a non-linear deterioration of mean psychological health over time, as illustrated by the estimated factor loadings associated to the latent slope at , which are significantly different from 2 (the latter would be the third loadings in the linear model). However, the estimated loadings differ across gender, being equal to 0.64 in the model for female and to 2.94 in the model for men. These loadings represent the proportion of the total amount of change that has occurred up to and indicate that the change observed in women psychological health between 1996 and 2004 was 0.64 times the change observed between 1996 and 2000. On the





18

contrary, the change observed in men psychological health between 1996 and 2004 was almost 3 times the change observed between 1996 and 2000. Among women, the baseline number of depressive symptoms at age 26 is 2.052 and the number of depressive symptoms significantly declines by 0.227 between and , where the magnitude of this decrease drastically diminishes with the passage of time (as suggested by the value of 0.64 of the estimated factor loading at ). Male instead suffer in general from lower depressive symptoms at age 26 (1.326), the number of depressive symptoms increases by 0.035 between and , and the magnitude of this increase rises as men got older (as indicated by the value of 2.94 of the estimated factor loading at ). Significant variance is found in the random intercept but not in the random slope for both women and men, which suggests that, within each group, there exists between-person variability around the mean value of the intercept but not in its rate of change. As such, individuals have different initial levels of psychological health but progress over time at approximately the same rate. Interestingly, the covariance between baseline scores and slopes is positive and significant (0.150, p<0.005) in the model for the general population, thus suggesting that those with higher initial depressive symptoms (intercept) were most likely to experience a steeper trajectory (slope) during the follow-up. However, in the separate models for male and female this relationship is no longer either significant or univocal.





19

Table 2 Unconditional growth trajectories of socio-economic status and health: full sample and by gender

Full sample Women Men

Intercept

(I) Slope

(S)

Intercept (I)

Slope (S)

Intercept

(I) Slope

(S)

Income trajectory (Income)

Loadings ( ) [0, 1, 2] [0, 1, 2] [0, 1, 2] Mean 9.056***

(0.060) 2.270*** (0.051)

7.775*** (0.077)

1.447*** (0.065)

10.432*** (0.088)

3.104** (0.073)

Variance 12.658*** (1.293)

6.478*** (0.729)

9.661*** (0.869)

6.586*** (0.568)

12.751*** (2.417)

4.789*** (1.328)

Covariance (I, S) 3.316*** (0.768)

1.535*** (0.534)

2.890* (1.547)

Social class trajectory (Class)

Loadings ( ) [0, 1, 2] [0, 1, 2] [0, 1, 2] Mean 2.864***

(0.015) 0.091*** (0.008)

2.964*** (0.019)

0.070*** (0.011)

2.754*** (0.022)

0.118*** (0.011)

Variance 1.040*** (0.036)

0.060*** (0.017)

0.870*** (0.050)

0.080*** (0.022)

1.200*** (0.052)

0.056*** (0.026)

Covariance (I, S) -0.066*** 0.021

-0.057** (0.028)

-0.085** (0.030)

Physical health trajectory (Health)

Loadings ( ) [0, 1, 2] [0, 1, 2] [0, 1, 2] Mean 1.751***

(0.007) 0.096*** (0.005)

1.774*** (0.009)

0.091*** (0.007)

1.726*** (0.010)

0.102*** (0.007)

Variance 0.209*** (0.011)

0.030*** (0.006)

0.206*** (0.015)

0.028*** (0.008)

0.212*** (0.016)

0.036*** (0.010)

Covariance (I, S) -0.005 (0.007)

0.003 (0.009)

-0.016 (0.007)

Psychological health trajectory (Depression)

Loadings ( ) [0, 1, 0.38] [0, 1, 0.64] [0, 1, 2.94] Mean 1.726***

(0.018) -0.136*** (0.017)

2.052*** (0.027)

-0.227*** (0.026)

1.326*** (0.024)

0.035** (0.015)

Variance 1.791*** (0.074)

0.146 (0.186)

1.789*** (0.163)

-0.033 (0.240)

1.711*** (0.101)

0.129 (0.090)

Covariance (I, S) 0.150** (0.071)

0.182 (0.172)

-0.043 (0.147)

Model Fit CFI 0.998

0.012 0.009

0.995 0.017 0.014

0.994 0.020 0.014

RMSEA SRMR

Notes: (1) Since the results for the four separate growth models suggest that all but the trajectory of psychological health can be modeled as linear processes, the factor loadings for the trajectories of income, social class and physical health are set to the predetermined values: [0, 1, 2], while the loading associated to the third wave in the trajectory of psychological health are freely estimated. (2) In the estimation of the income trajectory, the incomes





20

are transformed by dividing them by 100. (3) Physical health status is coded as 0= excellent, 1=good, 2= fair, 3= poor or very poor and therefore higher values indicate worse health. (4) *** indicates statistical significance at 1% level, ** indicates significance at 5% level and * indicates significance at 10% level. Standard errors are in parentheses.

3.2. Conditional growth trajectories

Table 4 presents the estimates and the descriptive goodness-of-fit indices of the conditional

trajectories of socio-economic status and health in adulthood. The results of fitting the models to the male and female data separately show that, in general, the mechanisms by which early circumstances affect the domains of SES and physical health in adulthood are not gender specific (sign and significance of the coefficients are not substantially different). Hence, for parsimony, we report and comment only the estimates from the model fitted in the pooled sample and we include the gender indicator among the control variables8. However, since the progression of psychological health markedly differs across gender, in relation to this domain the estimates from the separate models for women and men are reported. In general, covariates with negative effects on the intercepts are associated with lower baseline levels, while covariates with negative effects on the slopes are associated with flatter rate of change over time. For example, in relation to the income trajectory, we see that in general men, compared to women, begin their trajectory at age 26 with a higher income (approximately 274 pounds more) and have a faster rate of growth over time (1.654, p<0.001).

Looking first at the domains of SES measured in early adulthood, the evidence confirms that childhood health endowments and socio-economic circumstances affect individuals’ SES position at age 26. More specifically, similar patterns are observed in relation to the domains of education and social class, where we see that all the childhood indicators have significant and negative effects on their baseline level at age 26. In relation to both education and social class, mother’s qualification is more detrimental than father’s qualification, while suffering from severe mental problems at age 5 is the strongest predictor among the childhood health indicators considered. As for the income domain, only a subset of the childhood disadvantages considered significantly predicts its baseline level at age 26, namely mother prenatal smoking habits, parents’ education and the indicators of resources deprivation at age 5 and age 10. Social class of origin represents a significant risk factor only for children of manual skilled parents. We also find that father’s qualification is the strongest predictor of individuals’ income at age 26, thus overcoming the effect of mother’s qualification in relation to this specific domain. Basically, these results suggest the existence of scarce social mobility across generations, whereby children from deprived socio-economic contexts turn out to be more deprived in the same domains in early adulthood. Looking at the predictors of adult physical health, we see that only early mental problems significantly increase the intercept of the health

8The estimates for the separate models are available upon request.





21

trajectory, while, quite surprisingly, we do not find any significant effects stemming from the indicators of childhood physical health endowments. However, since the indicators of born at low-birth weight and prenatal smoking exert significant effects on education, it may be that the relevance of these conditions is entirely mediated by earlier experiences and disappear when controlling for a wide range of different risk factors. In addition, among the childhood socio-economic circumstances considered, only mothers’ qualification and resources available during childhood affect adult physical health, while neither parental social class nor father’s qualification, once controlled for the other factors, are significantly related to it.

As for the effects of early life antecedents on the baseline level of psychological health, the results suggest that children with early mental problems, compared to the others, suffer from a higher number of depressive symptoms as adults, being that effect stronger for women that for men. Moreover, the childhood risk factors relevant in predicting adults’ psychological wellbeing are related to the physical health domain in the sample of men and to early socio-economic circumstances in the sample of women. Indeed, higher numbers of depressive symptoms at age 26 are found among males either born at low birth weight or whose mothers smoke during pregnancy, and among females who either experienced deprivation in appliances, or with parents having no qualification. Summing up, the results presented so far confirm that childhood socioeconomic circumstances have a pervasive influence on adult outcomes, playing an important role in determining both SES position and health status at age 26. The only exception is provided by men psychological health status, which does not seem to be affected by early socioeconomic disadvantages.

As for the long-lasting consequences of childhood health, suffering from moderate or severe mental problems at age 5 predicts both physical and psychological adult health, thus suggesting that the existence of a direct pathway stemming from early mental problems to future health, even after controlling for past and current SES. On the contrary, early physical health disadvantages directly affect only men psychological well-being, being their effect on future physical health not significant. Moreover, we find evidence of selection mechanisms, operating between childhood health and adult SES. Indeed, on the one hand we find that children with mental problems, compared to the others, ended up with lower education and in less prestigious social class, even when controlling with childhood socioeconomic circumstances. On the other hand, our results indicate that the adverse birth outcomes associated with smoking during pregnancies have long-term consequences on future SES, being the relation robust across all the adult SES domains considered.

Finally, in order to assess the early life influences on the progression of the SES-health gradient through adulthood, we move to the second part of Table 4 and look at the impact of childhood circumstances on the latent slopes. Here we see that both deprivations in appliances and parents’ education significantly affect the slope of the income trajectories. As such, compared to children from more affluent contexts, those grown up in deprived socio-economic circumstances not only start with a lower income at age 26, but they also present a flatter rate





22

of growth over early adulthood. As for the progression of social class, we see that childhood disadvantages, although representing important predictors of social position at age 26, do not have significant effects on individual career advancement. On the contrary, looking at the deterioration of physical health over time, we see that the health of children born to a low birth weight or suffering from early mental problems declines faster as they become adults, compared to those who did not experience any health problem in childhood. Interestingly, among the indicators of socio-economic circumstances, only parental social class seems to have a positive effect on the slope of the physical health trajectory, while we do not find any significant effect stemming from parents’ education or poverty indicators. Again, gender differences are found in the way by which covariates affect the rate of change in depressive symptoms. Indeed, the progression of depressive symptoms among men is not significantly predicted by childhood circumstances while the number of depressive symptoms among women declines slower for those who had suffered from mental problems during childhood, compared to those who had not.





23

Table 3 Conditional trajectories of SES and health

Latent intercepts: Latent slopes:

Education Income Class Health Depression Income Class Health Depression

Full Full Full Full Women Men Full Full Full Women Men

Childhood initial health endowments

Born at low birth weight -0.103* -0.527 -0.020 -0.038 0.082 0.206* 0.001 -0.031 0.063** 0.073 0.002 Mother Prenatal Smoking habits [Non smoker]

Indicator: moderate smoking -0.169*** -0.388** -0.183*** 0.028 0.072 0.131** 0.125 0.022 0.007 -0.011 -0.001 Indicator: heavily smoking -0.260*** 0.023 -0.128** 0.032 0.128 0.003 -0.066 0.029 0.033* 0.154 0.009

Early mental health [Normal score] Indicator: moderate problems -0.142*** -0.111 -0.143*** 0.061*** 0.332*** 0.188*** -0.179 -0.019 -0.003 0.149* 0.003 Indicator: severe problems -0.459*** -0.354 -0.267*** 0.091* 0.662*** 0.454*** 0.201 0.037 0.044 0.063 0.024

Childhood socio-economic circumstances

At risk of income poverty -0.246*** -0.891*** -0.240*** 0.057** 0.076 0.093 -0.145 0.014 -0.001 -0.046 0.003 Deprivation in appliances -0.181*** -0.447* -0.238*** 0.070*** 0.205** 0.084 -0.459*** 0.040 0.018 -0.015 0.008 Mother’s highest qualification [Degree]

Indicator: O-level -0.311*** -0.375 -0.230*** 0.051* 0.163 -0.069 -0.629*** -0.012 -0.022 -0.151 0.010 Indicator: vocational qualification. -0.568*** -0.647** -0.304*** 0.054* 0.262** 0.006 -0.683*** 0.001 -0.0010 -0.133 -0.002 Indicator: no qualification -0.801*** -1.123*** -0.521*** 0.092*** 0.294*** 0.072 -0.903*** 0.013 -0.019 -0.170* -0.008

Father’s highest qualification [Degree] Indicator: O-level -0.298*** -0.765*** -0.191*** 0.011 -0.049 0.021 -0.251 0.014 -0.030 -0.163* -0.006 Indicator: vocational qualification -0.421*** -0.830*** -0.313*** 0.009 0.042 0.103 -0.542*** 0.028 0.002 -0.057 0.003 Indicator: no qualification -0.562*** -1.211*** -0.389*** 0.031 0.165* 0.108 -0.798*** -0.020 0.015 -0.110 0.012

Parental social class [Professional/Managerial] Indicator: non-manual skilled -0.036 0.279 0.080 -0.046 0.181 -0.061 0.188 0.005 0.037* -0.066 0.002 Indicator: manual skilled -0.223*** -0.284* -0.184*** -0.014 -0.023 0.016 -0.186 -0.006 0.032** 0.076 0.004 Indicator: semi-skilled/ unskilled -0.467*** -0.200 -0.407*** -0.051 -0.225 -0.214 -0.412 0.061 0.051 -0.065 -0.006

Control variables

Indicator: male -0.084*** 2.737*** -0.189*** -0.040*** 1.654*** -0.051** 0.002

Model fit:

CFI 0.997 0.995 0.993 RMSEA 0.010 0.012 0.014 SRMR 0.007 0.009 0.009

Notes: (1) The reference categories for the ordinal indicators are reported in squared parentheses. (2) The coefficients in the second column, summarizing the impact of early life antecedents on the highest level of qualification achieved by age 26, have to be interpreted as ordered probit coefficients. (3)The coefficients reported in the columns under the label ‘Latent Intercepts’ and ‘Latent Slopes’ are interpreted as deviations from the average underlying intercept and slope describing the unconditional trajectories in Table 3. (4) Separate estimates for men and women are reported in relation to the psychological health trajectory, while in the other cases the results reported come from the model fitted in the pooled sample. (5) *** indicates statistical significance at 1% level, ** indicates significance at 5% level and * indicates significance at 10% level. Standard errors are not reported.





24

3.3. Analysing parallel processes through adulthood: cross-domain interdependencies

The mechanisms linking different domains of SES and health during adulthood are shown in Table 5, which presents the bivariate associations between education, latent intercepts, and slopes of the four trajectories estimated. In the table, the interaction between domains of SES and health at a specific point in time (age 26) is captured by the pairwise correlation coefficients between the latent intercepts and education and between the latent intercepts themselves. Table 4 Analysing parallel processes: bivariate correlations between highest qualification, latent intercepts and slopes

Latent intercepts: Latent slopes:

Education

Income Class Health Depression Income Class Health

Latent intercepts:

Income 0.884*** (0.080)

Class 0.545*** (0.020)

1.149*** (0.082)

Health -0.089*** (0.010)

-0.281*** (0.048)

-0.079*** (0.011)

Depression -0.165*** (0.024)

-0.804*** (0.114)

-0.151*** (0.027)

0.262*** (0.022)

Latent slopes:

Income 0.911 *** (0.043)

2.072** (0.927)

0.893*** (0.064)

-0.120*** (0.038)

-0.137 (0.090)

Class -0.002 (0.010)

-0.145*** (0.046)

-0.080*** 0.023

0.006 (0.006)

-0.033** (0.015)

0.189*** (0.035)

Health -0.010* (0.007)

0.014 (0.029)

0.000 (0.007)

-0.010 (0.008)

0.014 (0.014)

-0.016 (0.025)

0.003 (0.004)

Depression -0.029 (0.013)

0.258** (0.106)

-0.019 (0.026)

0.040* (0.024)

0.232* (0.126)

-0.145 (0.094)

-0.004 (0.024)

0.025* (0.015)

Notes: (1) *** indicates statistical significance at 1% level, ** indicates significance at 5% level and * indicates significance at 10% level. Standard errors are in parentheses.

The different domains of SES identified are positively correlated with each other ( ; ). Thus, as expected, more educated people at age 26 tend to have also higher levels of income and more prestigious job position. Similarly, a significantly positive association is observed between the two dimensions of health identified ( ), which suggests that psychological and physical health move together and people who have higher level of depressive symptoms show also worse physical health conditions. As for the associations between health and SES domains, our results confirm the existence of a strong SES-health gradient, which is robust across the socioeconomic dimensions and health outcomes. Indeed, the intercept of each SES domain has a significant negative association with the intercept of





25

both physical health and depressive symptoms9 at age 26. In other words, people who have better education, higher incomes, and more prestigious job positions enjoy also better physical health conditions and suffer from lower numbers of depressive symptoms in early adulthood.

The pairwise correlations between latent intercepts and slopes help us to gain a better understanding of the dynamic of the mechanisms by which different domains of SES are associated to health.

We first observed that both education and good initial job position promote a faster income development over time ( ), while a high initial income is associated with slower occupational advancements ( ). This result can be interpreted again in terms of ceiling effects and reflects the fact that individuals who have at the higher income are also in the most prestigious positions, and thus cannot advance further. We also find a significant positive association between income slope and social class slope ( ), suggesting that people that move into more prestigious occupations tend to have a more rapid income development. As for the relationships over time between the two health domains, we do not find any significant association between initial psychological health and the rate of physical health deterioration. However, the two domains of health appear strongly interlinked, being the rate of growth of depression symptoms positively associated with both the initial physical health status ( ), and its rate of change over time ( ). In other worlds, individuals suffering from worse physical health at age 26 tend also to faster accumulate depressive symptoms over time and are in turn associated with a more rapid deterioration of physical health.

Finally, we find that different domains of SES exert an effect on the progression of health through diverse mechanisms. Indeed, the education achieved by age 26 is associated with the rate of growth of physical health; the initial level of income is significant in relation to the deterioration of psychological health; and initial occupational position does not show any significant effect on the health trajectories. Looking at the signs of the coefficients, we find that while a better education at age 26 promotes slower rates of physical health deterioration ( ), higher initial levels of income are associated with faster rates of accumulation on the number of depressive symptoms over time ( ). At the same time, the analysis suggests the existence of selection mechanisms operating also though early adulthood, whereby initial health status is associated with the progression of socio-economic advantages. Indeed, the negative correlations between initial physical health and the rate of growth of income ( ) suggest that poor health at age 26 is associated with a slower income development over time. Moreover, the significant correlation between initial psychological health and social class slope 9Note that the lower the physical and psychological health scores, the better is the health status.





26

( ) confirms a health selection into less prestigious occupation: people suffering from higher depressive symptoms at age 26 experience slower occupational advancements over time compared to the others. .

4. Conclusions

This paper contributes to the existing literature on the SES-health gradient by investigating the scaring effects of childhood circumstances on the co-evolution of SES and health trajectories through adulthood in the UK. Two aspects of our findings are particularly noteworthy.

First, the results show that childhood health and socioeconomic circumstances play a crucial role in determining SES position and health status later in life, being also significantly related to the trajectory of SES and health over time. However, the different indicators of early disadvantages trigger off different trajectories in adulthood. With regards to the relevance of early health conditions, the evidence suggests the existence of selection mechanisms beginning in early life, whereby children suffering from mental or physical health problems are selected into lower social positions as adults, and the relation is consistent across the three adult domains of SES considered, also when controlling for childhood socio-economic circumstances. With regards to childhood socioeconomic conditions, our findings suggest that children growing up in poverty and from less educated parents (especially mothers) are more likely to end up with inferior education, less prestigious occupations, lower earnings and to suffer from worse physical and psychological health. The long lasting consequences of those early conditions extend also to the progression of income over time: disadvantaged children, compare to the others, have flatter rates of growth in early adulthood. Instead, parental social class operates through different mechanisms; it tends to both transmit across generations and influence educational achievements and income. However, parental social class does not affect the onset of physical or psychological health at age 26, but it represents the only significant determinant, among the socioeconomic circumstances considered, of faster trajectories of health deterioration through early adulthood.

Second, feedbacks mechanisms are found operating in adulthood as well, despite the processes through which education, income and social class affect and are affected by health during adulthood differ according to the specific domains of SES and health considered. More specifically, social selection has found to be important for understanding the association between health and either income progression or career opportunities. However, different selection mechanisms are distinguishable: adults with poor physical health are characterized by flatter trajectories of income growth, while those with higher numbers of depressive symptoms are characterized by less rapid career advancements. As for the relevance of causal mechanisms, the results suggest that while education plays a bigger role that income in predicting the deterioration of physical health over early adulthood, income dominates in





27

explaining the progression of depressive problems over time. On the contrary, the initial social class does not result to be significantly associated with either physical or psychological health deterioration.

From a policy perspective, our findings have several implications. First, the fact that early life circumstances have long-lasting consequences on the trajectories of SES and health during adulthood suggests that policies aimed to tackle the SES-health gradient in adulthood should be primary concerned with children disadvantages. More specifically, according to the findings of this study, interventions either fostering mother’s education or promoting mental health in the early years of a child’s life, would be the most effective policies for preventing successive socio-economic inequalities in health. Second, having recognized that the progression of the SES-health gradient through adulthood is founded upon complex relationships generated by mutual impacts between SES domains and health, this study suggests that reducing health inequalities requires targeting different spheres of adult deprivation through multidimensional interventions. As such, an active engagement of the health care sector, as well as of many other policy areas (including education, social security, working life), is crucial.





28

References Adams, P., Hurd, M., Mcfadden, D., Merrill, A., Ribeiro, T., 2003. Healthy, Wealthy, and

Wise? Tests for Direct Causal Paths between Health and Socioeconomic Status. Journal of Econometrics 112, 3–56.

Adler, N. E., Boyce, T., Chesney, M. A., Cohen, S., Folkman, S., Kahn, R. L., Syme, S. L., 1994. Socioeconomic Status and Health: the Challenge of the Gradient. American Psychologist 49(1), 15–24.

Black, S., Devereux, P.J., Salvanes., K.G., 2007. From the Cradle to the Labour Market?The Effect of Birthweight on Adult Outcomes. Quarterly Journal of Economics 122 (1), 409–439.

Bollen, K.A., Curran, P.J., 2006. Latent Curve Models: A Structural Equation Perspective. New York: Wiley.

Case, A., Fertig, A., Paxson, C., 2005. The Lasting Impact of Childhood Health and Circumstances.Journal of Health Economics 24, 365–389.

Cohen, Deborah A., Thomas A. Farley, and Karen Mason. 2003. Why Is Poverty Unhealthy? Social and Physical Mediators. Social Science and Medicine 57, 1631–1641.

Currie, J., 2009. Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and Human Capital Development. Journal of Economic Literature 47(1), 87–122.

Cutler, D.M., Lleras-Muney, A., Vogl, T., 2008. Socioeconomic Status and Health: Dimensions and Mechanisms. NBER Working Papers 14333, National Bureau of Economic Research, Inc. http://www.nber.org/papers/w14333.pdf

Deaton, A., Paxson, C., 1998. Aging and Inequality in Income and Health. American Economic Review Paper and Proceedings 88, 248253.

Elovainio, M., Ferrie, J. E., Singh-Manoux, A., Martin Shipely, G., Batty, D., Head, J., et al., 2011. Socioeconomic Differences in Cardiometabolic Factors: Social Causation or Health-related Selection? Evidence from the Whitehall II cohort study, 1991–2004. American Journal of Epidemiology 174(7), 779–789.

Goldman, N., 2001.Social Inequality in Health, Disentangling the Underlying Mechanisms. Annals of the New York Academy of Sciences 954, 118–39.

Goldman, D., Smith, J. P., 2002. Can Patient Self-Management Help Explain the SES Health Gradient? Proceedings of the National Academy of Sciences of the United States of America 99, 109–129.

Haas, S. A., 2007. The long-term Effects of Poor Childhood Health: an Assessment and Application of Retrospective Reports. Demography 44(1), 113–135.

Haas, S., 2008. Trajectories of Functional Health: the ‘Long arm’ of Childhood Health and Socioeconomic Factors. Social Science and Medicine 66, 849–861.

Halleröd, B., Gustafsson, J.-E., 2011. A Longitudinal Analysis of the Relationship between Changes in Socio-economic Status and Changes in Health. Social Science and Medicine 72(1), 116–123.





29

Herd, P., Goesling B., House, J. S., 2007. Socioeconomic Position and Health: The Differential Effects of Education versus Income on the Onset versus Progression of Health Problems. Journal of Health and Social Behavior 48, 223–238.

House, J. S., 2002. Understanding Social Factors and Inequalities in Health: 20th Century Progress and 21st Century Prospects. Journal of Health and Social Behavior 43:125–142.

Hu, L., Bentler, P.M., 1999. Cutoff Criteria for Fit Indexes in Covariance Structure Analysis: Conventional Criteria versus New Alternatives. Structural Equation Modelling 6(1), 1–55.

Idler E, Benyamini Y. 1997. Self-rated Health and Mortality: a Review of Twenty-seven Community Studies. Journal of Health and Social Behavior 38(1), 21–37.

Kaplan, D., 2008. Structural Equation Modeling: Foundations and Extensions. Newbury park, Calif.: Sage Publications.

Kruk, K.E., 2012. Parental Income and the Dynamics of Health Inequality in Early Childhood—Evidence from the UK. Health Economics. doi: 10.1002/hec.2876

Lantz, P.M., House, J.S., Mero R.P., Williams D.R., 2005. Stress, Life Events, and Socioeconomic Disparities in Health: Results from the American’s Changing Lives Study. Journal of Health and Social Behavior 46, 274–288.

MacCallum, R.C., Browne, M.W., Sugawara, H.M., 1996. Power Analysis and Determination of Sample Size for Covariance Structure Modeling. Psychological Methods 1, 130–149.

Makepeace G., Dolton P., Woods L., Joshi H., Galinda-Rueda F., 2003.From School to the Labour Market. In E Ferri, J Bynner and MEJ Wadsworth. eds. Changing Britain, Changing Lives: Three Generations at the Turn of the Century,pp. 29-70. Institute of Education, London.

Meredith W., Tisak J., 1990. Latent Curve Analysis.Psychometrika 55, 107–122. Mirowsky, J., Ross, C.E., 1999. Economic Hardship across the Life Course. American

Sociological Review 64, 548–569. Muthén, B. O., Curran, P. I., 1997. General Longitudinal Modeling of Individual Differences

in Experimental Designs: A latent Variable Framework for Analysis and Power Estimation. Psychological Methods 2, 371–402.

Office for National Statistics, 1999. Tracking People: A guide to Longitudinal Social Sources. London: Office for Statistics.

Palloni, A., Milesi, C., White, R. G., Turner, A., 2009. Early Childhood Health, Reproduction of Economic Inequalities and the Persistence of Health and Mortality Differentials. Social Science and Medicine 68, 1574–1582.

Quesnel-Vallée A., Miles, T., 2012. Socioeconomic Pathways to Depressive Symptmoms in Adulthood: Evidence from the National Longitudinal Survey of Youth 1979. Social Sciences and Medicine 74, 734–743.

Rutter, M., Tizard, J., Whitmore, K., 1970.Education, Health and Behaviour. Longmans, London.

Smith, J. P., 2009. The Impact of Childhood Health on Adult Labour Market





30

Outcomes.Review of Economics and Statistics 91, 478–489. Sydsjö G., 2011. Long-term Consequences of Non-Optimal Birth Characteristics. American

Journal of Reproductive Immunology 66(1), 81–87. doi: 10.1111/j.1600-0897.2011.01035. Thompson, A., Hollis, C., Richards, D., 2003. Authoritarian Parenting Attitudes as a Risk for

Conduct Problems – results from a British National Cohort Study. European Child and Adolescent Psychiatry 12(2), 84–91.

Van deMheen, H. D., Stronks, K., Mackenbach, J. P., 1998. A Lifecourse Perspective on Socio-economic Inequalities in Health: the Influence of Childhood Socio-economic Conditions and Selection Processes. Sociology of Health and Illness 20, 754–777.

Van Doorslaer, E., Koolman, X., 2004. Explaining the Differences in Incomerelated Health Inequalities across European Countries. Health Economics 13, 609–628.

Van Doorslaer, E., Wagstaff, A., Bleichrodt, H., Calonge, S., Gerdtham, U., Gerfin, M.,Geurts, J., Gross, L., Hakkinen, U., Leu, R.E., O’Donell, O., Propper, C., Puffer, F., Rodriguez, M., Sundberg, G., Winkelhake, O., 1997. Income-related Inequalities in Health: some International Comparisons. Journal of Health Economics 16(1), 93–112.

Wilkinson, R., Marmot, M. (Eds.), 2003. Social Determinants of Health: the Solid Facts, 2nd ed. World Health Organization.

Willett, J. B., Sayer, A. G., 1994. Using Covariance Structure Analysis to Detect Correlates and Predictors of Individual Change over Time. Psychological Bulletin 116, 363–381.

trends in health and socio-economic status, proximal and

Documents