Exploring the role of early-life circumstances, abilities and achievements on well-being at age 50 years: evidence from the 1958 British birth cohort study

Objectives We aim to examine the relative contributions of pathways from middle childhood/adolescence to mid-life well-being, health and cognition, in the context of family socio-economic status (SES) at birth, educational achievement and early-adulthood SES. Our approach is largely exploratory, suspecting that the strongest mediators between childhood circumstances and mid-life physical and emotional well-being may be cognitive performance during school years, material and behavioural difficulties, and educational achievement. We also explore whether the effects of childhood circumstances on mid-life physical and emotional well-being differ between men and women. Setting/participants Data were from the National Child Development Study, a fully-representative British birth cohort sample of 17 415 people born in 1 week in 1958. Primary/secondary outcome measures Our four primary mid-life outcome measures are: cognitive performance, physical and emotional well-being and quality of life. Our intermediate adult outcomes are early-adulthood social class and educational/vocational qualifications. Results Using structural equation modelling, we explore numerous pathways through childhood and early adulthood which are significantly linked to our outcomes. We specifically examine the mediating effects of the following: cognitive ability at ages 7, 11 and 16 years; childhood psychological issues; family material difficulties at age 7 years: housing, unemployment, finance; educational/vocational qualifications and social class position at age 42 years. We find that social class at birth has a strong indirect effect on the age 50 outcomes via its influence on cognitive performance in childhood and adolescence, educational attainment and mid-life social class position, together with small direct effects on qualifications and social class position at age 42 years. Teenage cognitive performance has a strong positive effect on later physical health for women, while educational/vocational qualifications have a stronger positive effect on emotional well-being for men. Conclusion Our findings provide an understanding of the legacy of early life on multiple aspects of mid-life health, well-being, cognition and quality of life, showing stronger mediated links for men from childhood social class position to early adult social class position. The observed effect of qualifications supports those arguing that education is positively associated with subsequent cognitive functioning.

 The holistic nature of the analysis, incorporating four outcomes, three of which are strongly correlated after the inclusion our antecedents;  It makes optimal use of all the available longitudinal data by employing Full Information Maximum Likelihood estimation to compensate for the impact of missing data;  It shows the enduring importance of parental social class for cognitive performance and later life outcomes, controlling for family difficulties in childhood;  A limitation is that the subjects of our longitudinal study are not yet old enough for us to examine whether any decline in cognitive performance at age 50 years will become more influential on emotional well-being in future years.

INTRODUCTION
It is well-known that people with advantaged socioeconomic circumstances or status (SES) tend to have better health, across a range of health measures and across different societies [1,2,3]. Most scholars take the view that the mechanism by which SES differences affect health consists of factors such as living and working conditions, access to health services, social relationships and diversified life styles, being unequally distributed across the social hierarchy [1,4]. SES differentials in health may also result from health selection, whereby people with poor health find themselves occupying lower SES through decreased labour force participation or through withdrawal from paid work, thereby decreasing their wage income and inhibiting wealth accumulation [5,6]. Typically, authors examine the impact of childhood circumstances in adult SES, educational attainment and health by the use of multiple linear regression with a single outcome [7,8,9]. What makes our paper unique is that we model mid-life health, quality of life and cognitive function as a set of four inter-related multivariate outcomes.
Authors since the 1990s have advanced the importance of a life course approach to explore the relationship between early life circumstances and later life health outcomes [5,10]. This approach implies a reciprocal relationship between SES and health, and allows for the possibility that favourable circumstances in later life can compensate for the effects of earlier disadvantage [11]. At the same time, the theory of cumulative advantage/disadvantage suggests that individuals within a cohort increasingly diverge over the life course in terms of income, material resources and health, as large initial endowments lead to large additional gains and initial disadvantage causes further decline, regardless of the source of early advantage/disadvantage [12].
In attempting to unravel how accumulations and interactions of these various conditions affect health, some studies have shown that health variations in adults are attributed to both early and later life circumstances [2,13], suggesting that childhood SES has lasting effects on health beyond its impact on adult SES. Others do not find a significant association between childhood SES and later health once adult SES is taken into account [14], suggesting that childhood SES typically determines the living and working conditions in adult life and it is these circumstances that give rise to social inequalities in health. From a life course perspective, SES mobility may influence health outcomes, with upward mobility decreasing the risk of ill-health and partially compensating for earlier F o r p e e r r e v i e w o n l y disadvantages, and downward mobility increasing its risk [11]. Others have emphasised the need for a fuller assessment of the interplay between material circumstances and psychosocial factors over the life course [15,16].
Researchers in the field of 'cognitive reserve' [17,18] have argued that education can itself lead to higher cognition in mid-life and early old age, through increased efficiency of neural network utilisation [19]. Thus it may be that parents with advantaged backgrounds are able to facilitate and promote social, cultural and educational advantage for their offspring which lead to better cognition, which is linked in turn with better health behaviours. We attempt in this paper to explore the pathways linking mid-life cognition with mid-life physical health.
More recently, interest has also focused on mental and emotional well-being in mid-life and early old age, indicating those with high levels of mental & emotional well-being have increased longevity [20], and that cognitive and socioemotional function can fuse to form skills supporting selfregulation, competence, and quality of life that persist into old age through linked reciprocal processes of genetic influence, nurturing, schooling, work, and lifestyle [21].
Our analysis builds upon the framework adopted by Wood et al[22] on the prediction of mental well-being using the Warwick-Edinburgh Mental Well-Being Scale [23] across four UK birth cohort studies, which showed that childhood SES is directly and indirectly (through adult socioeconomic pathways) linked to adult emotional well-being. We expand their approach by having a fuller set of outcome measures (albeit for a single cohort) which include physical and emotional well-being, together with quality of life. In addition, we include a cognitive performance measure at age 50 as our fourth outcome. Taken together, we examine the pathways from childhood to our multivariate mid-life outcome measures for men and women.
The 1958 British Birth Cohort Study (NCDS) provides a rich source of key aspects of the life course beginning with birth circumstances: birthweight, breastfeeding, maternal smoking during pregnancy and parental SES. In addition we include the effects of family difficulties such as unemployment, finance and housing in conjunction with childhood scores of socio-emotional adjustment and cognitive performance, each measured on three occasions spanning the ages 7, 11 and 16 years. The inclusion of SES at age 42 and lifetime highest qualification in the model enables us to trace which childhood effects persist and which are built upon or attenuated by education and social mobility.
Our conceptual framework (Fig.1) describes a 'pathways to risk model,' where the accumulation of advantage/disadvantage from childhood to mid-life represents a model of the life course where antecedent influences are postulated to have consequences for later life outcomes, notably the four life domains covering cognitive functioning, physical and emotional well-being and quality of life. These outcomes, along with those capturing cognitive performance at early ages and family difficulties, are defined as latent variables, each defined as a measurement model consisting of selected manifest variables represented by individual item responses (indicators) taken from NCDS survey data. The inter-relationships between the latent variables, and in particular the interplay between our outcomes, define a structural equation model (SEM) which underpins our statistical analysis [24].

DATA AND METHODS
Our data source, NCDS [25] provides operational measures regarding birth circumstances, the early and teen years covering ages 7, 11 and 16, and adulthood at ages 42 and 50 respectively. The analysis is based upon 3815 male and 4209 female cohort members present at each occasion who all have at least one genuine response for each of the four separate age 50 life domains, and who also participated in at least one of the three cognitive and socio-emotional assessments in order to permit a reliable strategy for handling missing items.
Our four age-50 life domains are represented by well-established measures: scales of physical and emotional well-being (hereafter PWB, EWB) are taken from the Medical Outcomes Survey (MOS) Short Form (SF-36) developed by RAND [26][27][28][29]. Quality of Life (hereafter QoL) is measured by the 12-point version of the 'Control, Autonomy, Self-Realisation and Pleasure' (CASPv2) QoL Scale [30]. The fourth domain outcome, cognitive ability (hereafter Age50 Cog) consists of four cognitive tests examining memory (word recall & delayed recall), executive function (animal naming), attention and mental speed (letter cancellation) [31].
Appendix 1 lists all of the manifest and latent variables employed in the analysis. Indicators for cognitive function at age 50 years are listed here, while those for physical, emotional and quality of life outcomes are expanded upon more fully in Appendix 2. Childhood behavioural indicators are listed in Appendix 3 following Rutter [32].
In fitting our structural equation models using MPlus [33] we obtain direct and indirect (mediated) effects of our background variables in terms of childhood social origins and social adjustment, on outcomes in early adulthood and mid-life at age 50 years.
The sets of predictors determining our paths differ slightly for each of the four outcomes so as to avoid the problem of 'seemingly unrelated regression' [34] To address the issue of attrition bias caused by differential loss to follow-up in later sweeps, we employ full information maximum likelihood estimation (FIML) [35]. The SEM procedure that follows begins with an analysis of the sample as a whole, followed by separate analyses for men and women, supported by a multigroup analysis to test for model equivalence between men and women [36]. Model 'goodness of fit' is assessed via conventional criteria: 'root mean square error' (RMSEA), 'comparative fit index', (CFI) and 'Tucker Lewis fit index', (TLI) where RMSEA<0.05, CFI>0.90 and TFI> 0.90 would provide evidence of 'good fit' [37].

Patient and Public involvement statement
Participant consent was obtained from parents of cohort members at the initiation of the original birth cohort study in 1958. On attaining adulthood, permission was obtained from cohort members themselves at each subsequent longitudinal sweep.

RESULTS
The results begin with consideration of the inter-correlations between our mid-life outcome measures. Further univariate descriptives are contained in Appendix 1. Figure 2 contains a full SEM diagram for the sample as a whole. Subsequent reference will be made to Appendices 4 & 5 for the details of mediation, total and indirect effects and Appendix 6 contains further information on our multigroup analysis based on gender. Table 1 contains a correlation matrix of the four mid-life outcomes expressed as summative indices for each specific life domain. Interestingly, QoL, PWB and EWB demonstrate moderate associations of inter-correlation. While cognitive ability is positively related to the physical and emotional well-being measures, the strength of the association is modest, with PWB having a higher coefficient for women and EWB & QoL a higher coefficient for men.   0.05). These two 'interim' or 'pivotal' destinations, in turn, have notable effects on our four mid-life outcomes at age 50, which we describe under (d) below.

(c) Indirect effects via pathways through childhood mediators
Inspecting the paths in Fig.2 from birth circumstances through ages 7, 11 and 16, we see firstly, that all four of our birth characteristics have direct effects on cognitive performance at age 7 years, when children are in primary school. Social class at birth has the most notable association with age 7 cognition (. 27). This cognitive variable in turn has a strong direct effect on age 11 cognition (.92) and similarly for age 11 on cog performance at age 16 (.92).
There is a clear path via cognitive ability throughout childhood and adolescence, to adult achievement as indicated by Age 42 social class and educational qualifications (Age 16 Cog to SC 42yrs .50, Age 16 Cog to Quals .59).
This legacy of childhood cognitive performance is unsurprising, but interestingly Age 16 cognition also has a direct association with age 50 physical well-being (.12) besides more naturally, Age 50 cognition (.47).
Observed family difficulties in childhood, themselves associated with birth social class (-0.26), also play their part in understanding a child's trajectory in cognitive performance. Characterised by issues with housing, finance and employment when a child is 7 years old, these family difficulties are seen to have negative consequences for cognitive performance at age 7 years (-. 15) and a lagged negative effect on later performance age 16 years (-0.04).
Additional analyses also show that family difficulties impact on a child's emotional and behavioural scores (based on Rutter) at ages 7, 11 and 16 (.09, .06 and .10 respectively). In turn these scores had small significant negative associations with cognitive performance at each corresponding age (-.13, -.04 and -.05 respectively). But most notably, Rutter scores at these three ages are significantly correlated (.49 age 7 to 11 and .44 age 11 to 16), and there are direct effects from age 16 Rutter scores to three of our four age 50 outcomes: EWB (-0.13), QoL (-0.10) and PWB (-0.07).

(d) Indirect effects via adulthood mediators
Adult social class and lifetime qualifications are, together, influential predictors of each of our four mid-life outcomes: adult social class has significant predictive effects (.08 and .11) on PWB and QoL, and our Bonferroni-corrected criterion of p< .001 narrowly excludes its link to EWB (.05 at p<.004).
In general, the socially-advantaged enjoy better health and report higher levels of quality of life than their less advantaged counterparts. Interestingly, educational qualifications predict EWB (.06) and QoL (.10) but not PWB (.04 at p<.01, so not quite significant). As one would expect, educational qualifications are significantly linked to age 50 cognitive ability (.05), EWB (.06) and QoL (.10).

(e) Gender differences
Our multigroup analysis reveals that certain paths have significantly different effects for men and women. In Appendix 5 we display two separate path diagrams for men and women. For men, the direct effect of social class at birth on social class and qualifications in early adulthood is notably stronger (.09 and .07 respectively) than for women (.05, .05). Both sexes see the birth social class effect remain equally strong for cognitive performance at age 7 years and beyond, but women experience a slightly worse negative effect of social disadvantage at birth on the latent variable 'family difficulties' (-0.27 compared to -0.24). In turn, family difficulties have slightly more influence on Age 7 cognition (.17) for men than for women (.14), and a small lagged effect on Age 16 cognition only evident for women. After age 7 the Rutter social adjustment scores do not appear to impact upon the cognitive performance scores for men, but they do for women at 11 and 16, though the effect is small (.05).
Early adult social class has a slightly stronger influence for men than women on our age 50 outcomes PWB and QoL, and Qualifications have a significant effect on EWB for men but not for women. In contrast, women's age 16 cognition has a significant effect on their later PWB (.17), an effect not significant for men; and the path from age 16 cognition to age 50 Cog has a stronger coefficient for women (.47) than men (.43).
Pursuing the evidence for gender differences via a sensitivity analysis under MPlus which formally tests for statistical difference between paths for each sex. Appendix 6 identifies six paths where the effects for men and women are significantly different, but only three have notable differences in terms of their effect sizes in the multigroup analysis. These are:  stronger effect for women from age 16 cognition to age 50 cognition as noted above (.47 to .43);  direct effect of social class at birth on early social class (much stronger for men); and  small effect of age 42 social class on Age 50 Cognition (.07 at p<.002 for men, no sig effect women)

DISCUSSION
Structural equation modelling has teased out the nuances of the influence of pathways from childhood and early adulthood to our four related outcomes of health, well-being and cognition in mid-life. The work demonstrates that a holistic approach to defining the aspects of life that matter in mid-life can be revealing. Physical and emotional health matter for a person's subjective account of their quality of life, and PWB correlates at .69 with EWB, which in turn correlates .83 with QoL. While cognitive performance at age 50 is relatively weakly related, the other three are closely intertwined. These associations are similar in magnitude to those reported in table 1; however they are now estimated by taking the legacy of the life course into account.
We also show that some of the paths to health and well-being in mid-life are different for men and women [39]. However, important communalities remain.
Social origins, in particular social class at birth, have a major indirect influence on mid-life outcomes, the mediation process beginning with a child's cognitive performance throughout formal schooling and subsequently via cognitive performance at age 16 years. Collectively, these influences are weakened or reinforced by the existence of family difficulties in early childhood (themselves influenced by childhood social class). Whilst cognitive performance is a strong predictor of early adult social class and educational status for both men and women, social class at birth also has a lasting association with these mid-life indicators: more so for men than for women. This could well suggest that women have a looser connection with the legacy of social class at birth, perhaps partly because their path to advantaged SES at age 42 may be impeded by caring responsibilities and gender inequalities in the labour market [40].
Early adult social class influences all four of our mid-life outcomes, but we see the direct effect of social class at birth on social class and qualifications in early adulthood is notably stronger for men (.09 and .07 respectively) than for women (.05, .05).
A relatively small but consistent direct effect of qualifications on Age 50 Cognition supports those who argue that education has a causal effect on subsequent cognitive functioning [41], (but note alternative views on this [42]). Equally, the relationship between QoL and early adult social class supports the work of Blane et al [43].
The strength in the evidence presented here lies in the richness of its longitudinal dimension which conveys the importance of early life circumstances upon early adult and mid-life measures of achievement, health and well-being. It may be premature to identify the impact of any early cognitive decline; however, cognitive performance in the teen years does hold a legacy for cognitive performance in mid-life.
Social class at birth has a strong influence on early cognitive performance and beyond, which in turn has consequences for adult achievement and social class. Clearly, the presence of family difficulties and behavioural issues will impede (mediate) this trajectory. Interestingly, childhood behaviour, in particular at age 16, has a lasting effect on mid-life EWB and Quality of Life.

My health is excellent
Answer (1)- (5) Note that the answers to the 36 questions were recoded on a scale from 0 to 100 with 100 indicating the highest levels of health (i.e. negatively-phrased questions being coded in reverse polarity).
These were then divided into eight sub-domains (four physical health, four emotional well-being, with the recoded variables being summed then divided by the number of questions asked: Answer (1)-(4) 10-12 Pleasure 10 I feel full of energy these days. These are modest direct influences on PWB and QoL, but a relatively stronger negative influence on EWB at age 50 years.
Two direct influences stand out in Table B: from social class at birth to social class at 42 years and Qualifications obtained. The direct influences are shown in our estimated pathways diagram (Fig.2), but we also see in Table C there is an additional indirect effect (via childhood cognition at ages 7/11/16), over twice as large as the direct effect in both cases. An MPlus multigroup run by sex was performed as a sensitivity analysis, in which each of the 50 possible paths in Fig.2 was labelled m1-m50, and (in the female model) f1-f49, then a set of 50 'difference' variables was defined diff(i)=m(i)-f(i).
Testing the significance of these 50 'diff' variables, 6 were found to be significant, and the other 43 non-significant.
The MPlus multigroup model was then re-run, constraining all the 44 non-significant paths to be equal for males and females, but allowing the other six paths to vary, thus increasing the parsimony of the model.
The six paths found to be significantly different between men and women are shown in the The first three paths are notable, but in the case of the last three the difference in the point estimates, though significant, is so small as to be unremarkable.
We see that men's Age 50 cognitive ability at age 50 has a stronger link to age 42 SES than for women, where the path is not significant (strictly speaking, male p-value falls just sort of satisfying Bonferroni-corrected significance level .001, but female p-value is nowhere near).
In contrast, men's Age 50 cognitive ability has a weaker link with Age 16 cognition (.42) than for women (.47).
Finally, for men there is a significant direct link from SC birth to SC 42 years, a result not significant for women.

Contributorship Statement
Brian Dodgeon performed the data management and structural equation modelling using the Mplus software, and took responsibility for completing all aspects of the paper.
Praveetha Patalay contributed to the conceptual planning of the analysis jointly with Richard Wiggins, performed preliminary regression modelling and advised on aspects of the structural equation modelling (SEM) strategy.
George Ploubidis advised on many aspects of the statistical analysis, and provided detailed comments on the drafting of the paper.
Richard Wiggins undertook the conceptual planning of the analysis jointly with Praveetha Patalay, and initiated the SEM approach, continuing to provide advice on the models as they became more complex, and collaborating with Brian Dodgeon on the write-up of the findings.

ABSTRACT 1 Objectives
We aim to examine the relative contributions of pathways from early childhood/adolescence to midlife well-being, health and cognition, in the context of family socio-economic status (SES) at birth, educational achievement and early-adulthood SES. We hypothesise that effects of childhood circumstances on mid-life physical & emotional well-being are strongly mediated by cognitive

Results
Using structural equation modelling, we explore numerous pathways through childhood and early adulthood which are significantly linked to our outcomes. We specifically examine the mediating effects of the following: cognitive ability at ages 7, 11 and 16 years; childhood psychological issues ; family material difficulties at age 7: housing, unemployment, finance; educational/vocational qualifications and social class position at age 42 years.
We find that social class at birth has a strong indirect effect on the age 50 outcomes via its influence on cognitive ability, educational attainment and mid-life social class position, together with small direct effects on qualifications and social class position at age 42 years . Teenage cognitive performance has a strong positive effect on later physical health for women, whilst educational/vocational qualifications have a stronger positive effect on emotional well-being for men.

Conclusion
Our findings provide an understanding of the legacy of early life on multiple aspects of mid-life health, well-being, cognition and quality of life, showing stronger mediated links for men from childhood social class position to early adult social class position. The observed effect of qualifications supports those arguing that education is positively associated with subsequent cognitive functioning.

Article Summary
Strengths and limitations of this study:  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o n l y  It shows the enduring importance of parental social class for cognitive performance and later life outcomes, controlling for family material difficulties in childhood  A potential limitation is that our model does not include a broader set of demographic and psychosocial factors: a result of our decision to restrict the number of early-life influences to avoid over-complexity in the statistical modelling. There is also a small amount of attrition bias between birth and age 50.

INTRODUCTION
Authors since the 1990s have advanced the importance of a life course approach to explore the relationship between early life circumstances and later life health outcomes [1,2]. Carr [3] emphasises how diverse early life experiences affect physical, emotional and cognitive health in later life by drawing upon a wealth of longitudinal data across the United States, Australia and Europe. Our empirical study is based on a single British Birth Cohort which covers several waves of data collection from birth (1958), middle childhood (age 7 & 11), adolescence (age 16) and mid-life at ages 42 and 50 years for over 8,000 sample members. Age 50 years represents a specific mid-point of mid-life (40-60) as defined by Midlife Development in the United States (MIDUS): Brim et al [4] suggest "Midlife has been described as the last uncharted territory of the life course". To this extent our conceptual framework is largely informed by Elder [5]: it may be described as a cumulative (dis-) advantage model [6] where we examine the legacy of early life circumstances, notably the British Registrar General's classification of occupations (RG social class) [7] as having lasting effects on health and well-being at age 50 years. Our approach implies a reciprocal relationship between socio-economic status (SES) and health, and allows for the possibility that favourable circumstances in later life can act counter to the effects of earlier disadvantage [8].
To this end we select outcomes at age 50 years covering four distinct areas: self-assessed quality of life, physical and emotional well-being, and a cognitive ability test.
Our research interest is in exploring the extent to which these outcomes inter-relate, adopting an empirical view that, taken together, these measures would tell us more about the consequences of the past than studied individually [9]. There is a growing body of evidence based on United Kingdom (UK) longitudinal studies [10,11,12] that provides a strong argument for combining these outcomes in order to provide a holistic perspective: in particular, Cooper et al [13] provided both substantive and empirical evidence for examining gender differences in this analysis.
Drawn from the same rich longitudinal resource, our predictors and mediators begin with birth circumstances: birthweight, breastfeeding, maternal smoking during pregnancy and parental SES. From the age 7 sweep of the study we include the effects of family material difficulties such as unemployment, finance and housing, in conjunction with childhood scores of socio-emotional adjustment and cognitive performance, each measured on three occasions spanning the ages 7, 11 and 16 years. The inclusion of SES at age 42 and lifetime highest qualification in the model enables us to trace which childhood effects persist, and which are built upon or attenuated by education and social mobility.
Our analysis builds upon the framework adopted by Wood et al [14] on the prediction of mental well-being as a single outcome [15] across four UK birth cohort studies, which showed that  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y childhood SES is directly and indirectly (through adult socioeconomic pathways) linked to adult emotional well-being. We expand their approach by having a fuller set of four outcome measures, albeit for a single birth cohort. Taken together, we examine the pathways from childhood to our multivariate mid-life outcome measures for men and women.
Our conceptual framework is summarised in Fig.1. The inter-relationships between the latent variables, and in particular the interplay between our outcomes, define a structural equation model (SEM) which underpins our statistical analysis [16].

DATA AND METHODS
Our data source, the British National Child Development Study (NCDS) is fully representative of the population of Great Britain (GB), having interviewed the parents of 98.5% of all babies born in the same week of 1958 [17], and followed them up longitudinally with ten subsequent interviews throughout life. It provides operational measures regarding birth circumstances, the early and teen years covering ages 7, 11 and 16, and adulthood at ages 42 and 50 respectively.
The analysis is based upon 3815 male and 4209 female cohort members present at each occasion who all have at least one genuine response for each of the four separate age 50 life domains, and who also participated in at least one of the three cognitive and socio-emotional assessments, in order to permit a reliable strategy for handling missing items.
We are aware there is a certain amount of attrition bias as a result of differential loss-to follow-up from the 17,415 cohort members present at the birth study in 1958. As a check on the impact of sample loss, we were reassured by the close match of the distribution of the birth social class variable within our sample of 8,024 against that of the same variable within the full 17,415 birth sample (chi-square=1.29, 5d.f., non-significant) (see Appendix 1).
Our outcomes at age 50 years cover four distinct measures, the first two being self-reported physical and emotional well-being (PWB and EWB respectively) measured using sub-scales [18] of the RAND 36-Item Short-Form Health Survey [19][20][21][22][23] (hereafter SF-36). Third, a self-reported measure of quality of life (QoL) covering the life domains of 'Control, Autonomy, Self-Realisation and Pleasure (CASP-12 v2) [24]: a theoretically-informed measure of subjective well-being, closely following the approach adopted by Ed Diener and his colleagues [25]. The fourth domain outcome, cognitive ability (hereafter Age50Cog) consists of four cognitive tests examining memory (word recall & delayed recall), executive function (animal naming), attention and mental speed (letter cancellation) [26].
Appendix 2 lists all of the manifest and latent variables employed in the analysis. Indicators for cognitive function at age 50 years are listed here, while those for physical, emotional and quality of life outcomes are expanded upon more fully in Appendix 3. Childhood behavioural indicators are listed in Appendix 4 following Rutter [27]. We recognise of course that Rutter scores may be sub- divided into constituent dimensions of problem behaviour (withdrawn/ anxious / oppositional/ inattentive [28], but for this paper we prefer to use Rutter total scores, following Parsons et al [29]. In fitting our structural equation models using the MPlus software [30] we obtain direct and indirect (mediated) effects of our background variables in terms of childhood social origins and social adjustment, on outcomes in early adulthood, and mid-life at age 50 years. We regard all four of our birth indicators as primary predictors, valuing the opportunity to assess the relative impact of each for policy interventions [31]. Our birth social class variable is based on father's occupation, since in 1958 when the cohort were born, only 38% of mothers were economically active.
The sets of predictors determining our paths differ slightly for each of the four outcomes so as to avoid the problem of 'seemingly unrelated regression' [32] We employ full information maximum likelihood estimation (FIML) [33]. This addresses the issue of item non-response arising where questions are avoided or not answered by certain participants.
The SEM procedure that follows begins with an analysis of the sample as a whole, followed by separate analyses for men and women, supported by a multigroup analysis [34] to test for the structural equivalence of the models for men and women. Model 'goodness of fit' is assessed via conventional criteria: 'root mean square error' (RMSEA), 'comparative fit index', (CFI) and 'Tucker Lewis fit index', (TLI) where RMSEA<0.05, CFI>0.90 and TFI> 0.90 would provide evidence of 'good fit' [35].
Having a large number of SEM pathways involving 261 free parameters, we had regard to the n:q rule [36] and were satisfied that our sample-size-to-parameters ratio was comfortably within Jackson's guideline of 10:1. Or application of SEM involved the use of modification indices, which led to our running correlations on suggested variables. All our final models were identified, all manifest variables loading onto the respective latent variables with satisfactory fit. The details of the individual assessments of our measurement models can be made available on request.

Patient and Public involvement statement
Participant consent was obtained from parents of cohort members at the initiation of the original birth cohort study in 1958. On attaining adulthood, permission was obtained from cohort members themselves at each subsequent longitudinal sweep.

RESULTS
The results begin with consideration of the inter-correlations between our mid-life outcome measures. Further univariate descriptives are contained in Appendix 2. Figure 2 contains a full SEM diagram for the sample as a whole. Subsequent reference will be made to Appendix 5 for the details of mediation, total and indirect effects. Appendix 6 contains separate analyses by sex and Appendix 7 contains further information on our multigroup analysis based on gender. As a preliminary inspection of the inter-relationships between our four outcome measures, Table 1 contains a correlation matrix of the four summative indices based on these measures for both men and women.   Whilst there are various pathways that connect early life circumstances with cognitive performance, family material difficulties and social adjustment, there remains a small direct effect of social class at birth on each of our two early adult outcomes: Age 42 Social class and educational/vocational qualifications (both path coefficients equal 0.05). These two 'interim' or 'pivotal' destinations, in turn, have notable effects on our four mid-life outcomes at age 50, which we describe under (d) below.

(c) Indirect effects via pathways through childhood mediators
Inspecting the paths in Fig.2 from birth circumstances through ages 7, 11 and 16, we see firstly, that all four of our birth characteristics have direct effects on cognitive performance at age 7 years, when children are in primary school. Social class at birth has the most notable association with age 7 This legacy of childhood cognitive performance is unsurprising, but interestingly Age 16 cognition also has a direct association with age 50 physical well-being (.12) besides more naturally, Age 50 cognition (.47).
Family material difficulties in childhood, themselves associated with birth social class (-0.26), also play their part in understanding a child's trajectory in cognitive performance. Characterised by issues with housing, finance and employment when a child is 7 years old, these family material difficulties are seen to have negative consequences for cognitive performance at age 7 years (-.15) and a lagged negative effect on later performance age 16 years (-0.04).
Additional analyses also show that family material difficulties impact on a child's emotional and behavioural scores (based on Rutter) at ages 7, 11 and 16 (.09, .06 and .10 respectively). In turn these scores had small significant negative associations with cognitive performance at each corresponding age (-.13, -.04 and -.05 respectively). But most notably, Rutter scores at these three ages are significantly correlated (.49 for age 7 to 11 and .44 for age 11 to 16), and there are direct effects from age 16 Rutter scores to three of our four age 50 outcomes: EWB (-0.13), QoL (-0.10) and PWB (-0.07).

(d) Indirect effects via adulthood mediators
Adult social class and lifetime qualifications are, together, influential predictors of each of our four mid-life outcomes: adult social class has significant predictive effects (.08 and .11) on PWB and QoL, and our Bonferroni-corrected criterion of p< .001 narrowly excludes its link to EWB (.05 at p<.004).
In general, the socially-advantaged enjoy better health and report higher levels of quality of life than their less advantaged counterparts. Interestingly, qualifications predict age 50 cognitive ability (.05), EWB (.06) and QoL (.10) but not PWB (.04 at p<.01).

(e) Gender differences
Our multigroup analysis reveals that certain paths have significantly different effects for men and women. In Appendix 6 we display two separate path diagrams for men and women. For men, the direct effect of social class at birth on social class and qualifications in early adulthood is notably stronger (.09 and .07 respectively) than for women (.05, .05).
Both sexes see the birth social class effect remain equally strong for cognitive performance at age 7 years and beyond, but women experience a slightly worse negative effect of social disadvantage at birth on the latent variable 'family material difficulties' (-0.27 compared to -0.24 for men). In turn, family material difficulties have slightly more influence on Age 7 cognition (-.17) for men than for women (-. 14), and a small lagged effect on Age 16 cognition only evident for women. After age 7 Early adult social class has a slightly stronger influence for men than women on our age 50 outcomes PWB and QoL, and Qualifications have a significant effect on EWB for men but not for women. In contrast, women's age 16 cognition has a significant effect on their later PWB (.17), an effect not significant for men; and the path from age 16 cognition to age 50 Cog has a stronger coefficient for women (.47) than men (.43).
We pursue the evidence for gender differences via a sensitivity analysis under MPlus which formally tests for statistical difference between paths for each sex. Appendix 7 identifies six paths where the effects for men and women are significantly different, but only three have notable differences in terms of their effect sizes in the multigroup analysis. These are:  stronger effect for women from age 16 cognition to age 50 cognition as noted above (.47 to .43);  direct effect of social class at birth on early social class (much stronger for men); and  small effect of age 42 social class on Age 50 Cognition (.07 at p<.002 for men, no significant effect women)

DISCUSSION
Structural equation modelling has elucidated the nuances of the influence of pathways from childhood and early adulthood to our four related outcomes of health, well-being and cognition in mid-life. The work demonstrates the benefits of a holistic approach to defining the aspects of the life course that affect well-being in mid-life. Physical and emotional health are important factors affecting a person's subjective account of their quality of life, and PWB correlates at .69 with EWB, which in turn correlates .83 with QoL. While cognitive performance at age 50 is relatively weakly correlated, the other three are closely intertwined. These associations are similar in magnitude to those reported in table 1; however they are now estimated by taking the legacy of the life course into account.
We also show that some of the paths to health and well-being in mid-life are different for men and women [38]. However, important communalities remain.
Social origins, in particular social class at birth, have a major indirect influence on mid-life outcomes, the mediation process beginning with a child's cognitive performance throughout formal schooling and subsequently via cognitive performance at age 16 years. Collectively, these influences are weakened or reinforced by the existence of family material difficulties in early childhood (themselves influenced by childhood social class). Whilst cognitive performance is a strong predictor of early adult social class and educational status for both men and women, social class at birth also has a lasting association with these mid-life indicators: more so for men than for women.
This could well suggest that women have a looser connection with the legacy of social class at birth, perhaps partly because their path to advantaged SES at age 42 may be impeded by caring responsibilities and gender inequalities in the labour market [39]. Early adult social class influences all four of our mid-life outcomes, but we see the direct effect of social class at birth on social class and qualifications in early adulthood is notably stronger for men (.09 and .07 respectively) than for women (.05, .05).
A relatively small but consistent direct effect of qualifications on Age 50 Cognition supports those arguing that education is positively associated with subsequent cognitive functioning [40], although we are aware that there are counter-arguments to this view [41]). Equally, the relationship between QoL and early adult social class finds support in the work of Blane et al [42].
The strength in the evidence presented here lies in the richness of its longitudinal dimension which conveys the importance of early life circumstances upon early adult and mid-life measures of achievement, health and well-being.
One limitation of our study is that our model does not fully take into account all of the potential demographic and psychosocial factors other researchers have studied as predictors of satisfactory midlife development [28]. Nevertheless, it can be argued that our birth social class variable acts as a 'proxy' for certain parental beliefs and behaviours [43]. We are embarking on developing a conceptual framework and modelling strategy to extend our empirical analyses in the future as our cohort members age: in particular, our subjects are not yet old enough for us to assess whether any decline in cognitive performance at age 50 years will lead to subsequent decline and become more influential on emotional well-being and quality of life. However, we were able to show that cognitive performance in the teen years holds a legacy for cognitive performance in mid-life. A final limitation is a certain amount of attrition bias between birth and age 50 (differential non-response) which was addressed earlier in the paper and in Appendix 1.
Social class at birth has a strong influence on early cognitive performance and beyond, which in turn has consequences for adult achievement and social class. Clearly, the presence of family material difficulties and behavioural issues will impede (mediate) this trajectory. Interestingly, childhood behaviour, in particular at age 16, has a lasting effect on mid-life EWB and Quality of Life.

Consent of authors
I, Brian Dodgeon, the Submitting Author, have the right to grant and do grant on behalf of all authors of the Work (as defined in the below author licence), an exclusive licence and/or a non-exclusive licence for contributions from authors who are: i) UK Crown employees; ii) where BMJ has agreed a CC-BY licence shall apply, and/or iii) in accordance with the terms applicable for US Federal Government officers or employees acting as part of their official duties; on a worldwide, perpetual, irrevocable, royalty-free basis to BMJ Publishing Group Ltd ("BMJ") its licensees and where the relevant Journal is co-owned by BMJ to the co-owners of the Journal, to publish the Work in BMJ Open and any other BMJ products and to exploit all rights, as set out in our licence. The Submitting Author accepts and understands that any supply made under these terms is made by BMJ to the Submitting Author unless you are acting as an employee on behalf of your employer or a postgraduate student of an affiliated institution which is paying any applicable article publishing charge ("APC") for Open Access articles. Where the Submitting Author wishes to make the Work available on an Open Access basis (and intends to pay the relevant APC), the terms of reuse of such Open Access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this Work are set out in our licence referred to above.

Competing interests
All four authors declare they have no competing interests and no financial relationships with any organisation that might have an interest in the submitted work in the previous three years, and no other relationships or activities that could appear to have influenced the submitted work.

Data Sharing statement
The analysis is based on secondary data available for download from the UK Data Archive at https://beta.ukdataservice.ac.uk/datacatalogue/series/series?id=2000032. Syntax used to derive variables for this analysis can be obtained by e-mailing the corresponding author, Brian Dodgeon.

Ethics
Ethical approval was not required, since the study used the secondary data of the National Child Development Study, which itself had received ethics consent from the London Research Ethics Committee, ref 08/H0718/29 for its age 50 survey, the source of our outcome data. Multicentre Research Ethics Committee (MREC) approval was received for earlier NCDS surveys.    36. My health is excellent Answer (1)- (5) Note that the answers to the 36 questions were recoded on a scale from 0 to 100 with 100 indicating the highest levels of health (ie negatively-phrased questions being coded in reverse polarity).
These were then divided into eight sub-domains (four physical health, four emotional well-being, with the recoded variables being summed then divided by the number of questions asked: These are modest direct influences on PWB and QoL, but a relatively stronger negative influence on EWB at age 50 years. Two direct influences stand out in Table B: from social class at birth to social class at 42 years and Qualifications obtained. The direct influences are shown in our estimated pathways diagram (Fig.2), but we also see in Table C there is an additional indirect effect (via childhood cognition at ages 7/11/16), over twice as large as the direct effect in both cases. An MPlus multigroup run by sex was performed as a sensitivity analysis, in which each of the 50 possible paths in Fig.2 was labelled m1-m50, and (in the female model) f1-f49, then a set of 50 'difference' variables was defined diff(i)=m(i)-f(i).
Testing the significance of these 50 'diff' variables, 6 were found to be significant, and the other 43 non-significant.
The MPlus multigroup model was then re-run, constraining all the 44 non-significant paths to be equal for males and females, but allowing the other six paths to vary, thus increasing the parsimony of the model.
The six paths found to be significantly different between men and women are shown in the The first three paths are notable, but in the case of the last three the difference in the point estimates, though significant, is so small as to be unremarkable.
We see that men's Age 50 cognitive ability at age 50 has a stronger link to age 42 SES than for women, where the path is not significant (strictly speaking, male p-value falls just sort of satisfying Bonferroni-corrected significance level .001, but female p-value is nowhere near).
In contrast, men's Age 50 cognitive ability has a weaker link with Age 16 cognition (.42) than for women (.47).
Finally, for men there is a significant direct link from SC birth to SC 42 years, a result not significant for women.

Primary/secondary outcome measures
Four primary mid-life outcome measures: cognitive performance; physical and emotional well-being and quality of life. Our intermediate adult outcomes are early-adulthood social class and educational/vocational qualifications.

Results
Using structural equation modelling, we explore pathways through childhood and early adulthood which link significantly to our outcomes. We examine the mediating effects of: cognitive ability at ages 7, 11 and 16 years; childhood psychological issues; family material difficulties at age 7: housing, unemployment, finance; educational/vocational qualifications, social class position age 42 years.
We find social class at birth has a strong indirect effect on the age 50 outcomes via its influence on cognitive performance in childhood/adolescence, educational attainment and mid-life social class position, together with small direct effects on qualifications and social class position at 42 years.
Teenage cognitive performance has a strong positive effect on later physical health for women; qualifications have a stronger positive effect on emotional well-being for men.

Conclusion
We show the legacy of early life on multiple aspects of mid-life health, well-being, cognition and quality of life: stronger links for men from childhood social class position to adult social class position. The observed effect of qualifications supports those arguing that education associates positively with subsequent cognitive functioning.

Article Summary
Strengths and limitations of this study:  The richness of its longitudinal dimension, conveying the importance of early life circumstances upon early adult and mid-life measures of achievement, health and wellbeing;  The integrated nature of the analysis simultaneously models pathways to four mid-life outcomes, three of which are strongly correlated after the inclusion of several key early-life antecedents;  It makes optimal use of all the available longitudinal data by employing Full Information Maximum Likelihood estimation to take account of item non-response;  It shows the enduring importance of parental social class for cognitive performance and later life outcomes, controlling for family material difficulties in childhood.  A potential limitation is that our model does not include a broader set of demographic and psychosocial factors: a result of our decision to restrict the number of early-life influences to avoid over-complexity in the statistical modelling. There is also a small amount of attrition bias between birth and age 50.

INTRODUCTION
Authors since the 1990s have advanced the importance of a life course approach to explore the relationship between early life circumstances and later life health outcomes [1,2]. Carr [3] emphasises how diverse early life experiences affect physical, emotional and cognitive health in later life by drawing upon a wealth of longitudinal data across the United States, Australia and Europe. Our empirical study is based on a single British Birth Cohort which covers several waves of data collection from birth (1958), middle childhood (age 7 & 11), adolescence (age 16) and mid-life at ages 42 and 50 years for over 8,000 sample members. Age 50 years represents a specific mid-point of mid-life (40-60) as defined by Midlife Development in the United States (MIDUS): Brim et al [4] suggest "Midlife has been described as the last uncharted territory of the life course". To this extent our conceptual framework is largely informed by Elder [5]: it may be described as a cumulative (dis-) advantage model [6] where we examine the legacy of early life circumstances, notably the British Registrar General's classification of occupations (RG social class) [7] as having lasting effects on health and well-being at age 50 years. Our approach implies a reciprocal relationship between socio-economic status (SES) and health, and allows for the possibility that favourable circumstances in later life can act counter to the effects of earlier disadvantage [8].
To this end we select outcomes at age 50 years covering four distinct areas: self-assessed quality of life, physical and emotional well-being, and a cognitive ability test.
Our research interest is in exploring the extent to which these outcomes inter-relate, adopting an empirical view that, taken together, these measures would tell us more about the consequences of the past than studied individually [9]. There is a growing body of evidence based on United Kingdom (UK) longitudinal studies [10,11,12] that provides a strong argument for looking at these outcomes together within the same model. Cooper et al [13] provided both substantive and empirical evidence for examining gender differences in this analysis.
Drawn from the same rich longitudinal resource, our predictors and mediators begin with birth circumstances: birthweight, breastfeeding, maternal smoking during pregnancy and parental SES. From the age 7 sweep of the study we include the effects of family material difficulties such as unemployment, finance and housing, in conjunction with childhood scores of socio-emotional adjustment and cognitive performance, each measured on three occasions spanning the ages 7, 11 and 16 years. The inclusion of highest qualification and SES at age 42 in the model enables us to trace which childhood effects persist, and which are built upon or attenuated by education and social mobility.
Our analysis builds upon the framework adopted by Wood et al [14] on the prediction of mental well-being as a single outcome [15] across four UK birth cohort studies, which showed that childhood SES is directly and indirectly (through adult socioeconomic pathways) linked to adult emotional well-being. We expand upon their approach by including four outcome measures, albeit for a single birth cohort.
Our analysis adopts a structural equation modelling (SEM) approach [16], in order to explore the pathways from childhood and adolescence to our multivariate mid-life outcomes for both men and women.

DATA AND METHODS
Our data source, the British National Child Development Study (NCDS) is fully representative of the population of Great Britain (GB), having interviewed the parents of 98.5% of all babies born in the same week of 1958 [17], and followed them up longitudinally with ten subsequent interviews throughout life. It provides operational measures regarding birth circumstances, the early and teen years covering ages 7, 11 and 16, and adulthood at ages 42 and 50 respectively.
The analysis is based upon 3815 male and 4209 female cohort members present at each occasion who all have at least one genuine response for each of the four separate age 50 life domains, and who also participated in at least one of the three cognitive and socio-emotional assessments, in order to permit a reliable strategy for handling missing items.
We are aware there is a certain amount of attrition bias as a result of differential loss-to follow-up from the 17,415 cohort members present at the birth study in 1958. As a check on the impact of sample loss, we were reassured by the close match of the distribution of the birth social class variable within our sample of 8,024 against that of the same variable within the full 17,415 birth sample (chi-square=1.29, 5d.f., non-significant) (see Appendix 1).
Our outcomes at age 50 years cover four distinct measures, the first two being self-reported physical and emotional well-being (PWB and EWB respectively) measured using sub-scales [18] of the RAND 36-Item Short-Form Health Survey [19][20][21][22][23] (hereafter SF-36). Third, a self-reported measure of quality of life (QoL) covering the life domains of 'Control, Autonomy, Self-Realisation and Pleasure (CASP-12 v2) [24]: a theoretically-informed measure of subjective well-being, closely following the approach adopted by Ed Diener and his colleagues [25]. The fourth domain outcome, cognitive ability (hereafter Age50Cog) consists of four cognitive tests examining memory (word recall & delayed recall), executive function (animal naming), attention and mental speed (letter cancellation) [26].
Appendix 2 lists all of the manifest and latent variables employed in the analysis. Indicators for cognitive function at age 50 years are listed here, while those for physical, emotional and quality of life outcomes are expanded upon more fully in Appendix 3. Childhood behavioural indicators are listed in Appendix 4 following Rutter [27]. We recognise of course that Rutter scores may be subdivided into constituent dimensions of problem behaviour (withdrawn/ anxious / oppositional/ inattentive [28], but for this paper we prefer to use Rutter total scores, following Parsons et al [29]. In fitting our structural equation models using the MPlus software [30] we obtain direct and indirect (mediated) effects of our background variables in terms of childhood social origins and social adjustment, on outcomes in early adulthood, and mid-life at age 50 years. We regard all four of our birth indicators as primary predictors, valuing the opportunity to assess the relative impact of each for policy interventions [31]. Our birth social class variable is based on father's occupation, since in 1958 when the cohort were born, only 38% of mothers were economically active.
The sets of predictors determining our paths differ slightly for each of the four outcomes so as to avoid the problem of 'seemingly unrelated regression' [32]  We employ full information maximum likelihood estimation (FIML) [33]. This addresses the issue of item non-response arising where questions are avoided or not answered by certain participants.
The SEM procedure that follows begins with an analysis of the sample as a whole, followed by separate analyses for men and women, supported by a multigroup analysis [34] to test for the structural equivalence of the models for men and women. Model 'goodness of fit' is assessed via conventional criteria: 'root mean square error' (RMSEA), 'comparative fit index', (CFI) and 'Tucker Lewis fit index', (TLI) where RMSEA<0.05, CFI>0.90 and TFI> 0.90 would provide evidence of 'good fit' [35].
Having a large number of SEM pathways involving 261 free parameters, we had regard to the n:q rule [36] and were satisfied that our sample-size-to-parameters ratio was comfortably within Jackson's guideline of 10:1. Our application of SEM involved the use of modification indices, which led to our running correlations on suggested variables. All our final models were identified, all manifest variables loading onto the respective latent variables with satisfactory fit. The details of the individual assessments of our measurement models appear in Appendix 5.

Patient and Public involvement statement
Participant consent was obtained from parents of cohort members at the initiation of the original birth cohort study in 1958. On attaining adulthood, permission was obtained from cohort members themselves at each subsequent longitudinal sweep.

RESULTS
The results begin with consideration of the inter-correlations between our mid-life outcome measures. Further univariate descriptives are contained in Appendix 2. Figure 1 contains a final SEM diagram for the sample as a whole. Subsequent reference will be made to Appendix 6 for the details of mediation, total and indirect effects. Appendix 7 contains separate analyses by sex, and Appendix 8 contains further information on our multigroup analysis based on gender.

(a) Descriptives & correlations
As a preliminary inspection of the inter-relationships between our four outcome measures, Table 1 contains a correlation matrix of the four summative indices based on these measures for both men and women. Interestingly, QoL, PWB and EWB demonstrate moderate associations of inter-correlation. While cognitive ability is positively related to the physical and emotional well-being measures, the strength of the association is modest, with PWB having a higher coefficient for women and EWB & QoL a higher coefficient for men. Fig. 1 contains estimates of association amongst these outcomes as defined by measurement models under a full SEM analysis. As a check on the robustness of our statistical findings the SEM analyses were subject to bootstrap analysis under 1,000 repetitions. We also carried out a check on the influence of outliers (results available on request).

(b) Birth social class effects on early mid-life outcomes
The path diagram shown in Fig 1 provides   Whilst there are various pathways that connect early life circumstances with cognitive performance, family material difficulties and social adjustment, there remains a small direct effect of social class at birth on each of our two early adult outcomes: Age 42 Social class and educational/vocational qualifications (β=0.05 for both paths). These two 'interim' or 'pivotal' destinations, in turn, have notable effects on our four mid-life outcomes at age 50, which we describe under (d) below.

(c) Indirect effects via pathways through childhood mediators
Inspecting the paths in Fig.1  This legacy of childhood cognitive performance is unsurprising, but interestingly Age 16 cognition also has a direct association with age 50 physical well-being (β=.12) besides more naturally, Age 50 cognition (β=.47).
Family material difficulties in childhood, themselves associated with birth social class (β=-0.26), also play their part in understanding a child's trajectory in cognitive performance. Characterised by issues with housing, finance and employment when a child is 7 years old, these family material difficulties are seen to have negative consequences for cognitive performance at age 7 years (β=-.15) and a lagged negative effect on later performance age 16 years (β=-0.04).
Additional analyses also show that family material difficulties impact on a child's emotional and behavioural scores (based on Rutter) at ages 7, 11 and 16 (β=.09, .06 and .10 respectively). In turn these scores had small significant negative correlations with cognitive performance at each corresponding age (-.15, -.11 and -.11 respectively). But most notably, Rutter scores at these three ages are significantly correlated (.49 for age 7 to 11 and .44 for age 11 to 16), and there are direct effects from age 16 Rutter scores to three of our four age 50 outcomes: EWB (β=-0.13), QoL (β= -0.10) and PWB (β=-0.07).

(d) Indirect effects via adulthood mediators
Adult social class and lifetime qualifications are, together, influential predictors of each of our four mid-life outcomes: adult social class has significant predictive effects (β=.08 and .11) on PWB and QoL.
In general, the socially-advantaged enjoy better health and report higher levels of quality of life than their less advantaged counterparts. Interestingly, qualifications predict age 50 cognitive ability (β=.05), EWB (β=.06) and QoL (β=.10) but not PWB.

(e) Gender differences
Our multigroup analysis reveals that certain paths have significantly different effects for men and women. In Appendix 7 we display two separate path diagrams for men and women. For men, there is a direct effect of social class at birth on Age 42 social class and qualifications (β=.09 and .07 respectively), but for women there is no significant effect.
Both sexes see the birth social class effect remain equally strong for cognitive performance at age 7 years and beyond, but women experience a slightly worse negative effect of social disadvantage at birth on the latent variable 'family material difficulties' (β=-0.27 compared to -0.24 for men). In turn, family material difficulties have slightly more influence on Age 7 cognition (β=-.18) for men than for women (β=-.16), and a small lagged effect on Age 16 cognition only evident for women (β=-.05). The Rutter social adjustment scores at age 11 do not significantly correlate with the cognitive performance scores for men, but they do for women at ages 7, 11 and 16 (-.15, -.11 and -.13 compared with men's -.14 at age 7, -.11 at age 16).
Early adult social class has a slightly stronger influence for men than women on our age 50 outcomes PWB and QoL, and Qualifications have a significant effect on EWB for men, but not for women. In contrast, women's age 16 cognition has a significant effect on their later PWB (β=.17), an effect not significant for men; and the path from age 16 cognition to age 50 Cog has a stronger coefficient for women (β=.49) than men (β=.43).
We pursue the evidence for gender differences via a sensitivity analysis under MPlus which formally tests for statistical difference between paths for each sex. Appendix 8 identifies six paths where the

DISCUSSION
Structural equation modelling has elucidated the nuances of the influence of pathways from childhood and early adulthood to our four related outcomes of health, well-being and cognition in mid-life. The work demonstrates the benefits of an integrated approach to defining the aspects of the life course that affect well-being in mid-life. Physical and emotional health are important factors affecting a person's subjective account of their quality of life, and PWB correlates at .69 with EWB, which in turn correlates .83 with QoL. While cognitive performance at age 50 is relatively weakly correlated, the other three are closely intertwined. These associations are similar in magnitude to those reported in table 1; however they are now estimated by taking measurement error and the legacy of the life course into account.
We also show that some of the paths to health and well-being in mid-life are different for men and women [38]. However, important communalities remain.
Social origins, in particular social class at birth, have a major indirect influence on mid-life outcomes, the mediation process beginning with a child's cognitive performance throughout formal schooling and subsequently via cognitive performance at age 16 years. Collectively, these influences are weakened or reinforced by the existence of family material difficulties in early childhood (themselves influenced by childhood social class). Whilst cognitive performance is a strong predictor of early adult social class and educational status for both men and women, social class at birth also has a lasting association with these mid-life indicators for men but not for women.
This could well suggest that women have a looser connection with the legacy of social class at birth, perhaps partly because their path to advantaged SES at age 42 may be impeded by caring responsibilities and gender inequalities in the labour market [39].
Early adult social class influences all four of our mid-life outcomes, but we see the direct effect of social class at birth on social class and qualifications in early adulthood is pronounced for men (β=.09 and .07 respectively), but not significant for women.
A relatively small but consistent direct effect of qualifications on Age 50 Cognition supports those arguing that education is positively associated with subsequent cognitive functioning [40], although we are aware that there are counter-arguments to this view [41]). Equally, the relationship between QoL and early adult social class finds support in the work of Blane et al [42].
The strength in the evidence presented here lies in the richness of its longitudinal dimension which conveys the importance of early life circumstances upon early adult and mid-life measures of achievement, health and well-being. One limitation of our study is that our model does not fully take into account all of the potential demographic and psychosocial factors other researchers have studied as predictors of satisfactory midlife development [28]. Nevertheless, it can be argued that our birth social class variable acts as a 'proxy' for certain parental beliefs and behaviours [43]. We are embarking on developing a conceptual framework and modelling strategy to extend our empirical analyses in the future as our cohort members age: in particular, our subjects are not yet old enough for us to assess whether any decline in cognitive performance at age 50 years will lead to subsequent decline and become more influential on emotional well-being and quality of life. However, we were able to show that cognitive performance in the teen years holds a legacy for cognitive performance in mid-life. A final limitation is a certain amount of attrition bias between birth and age 50 (differential non-response) which was addressed earlier in the paper and in Appendix 1.
Social class at birth has a strong influence on early cognitive performance and beyond, which in turn has consequences for adult achievement and social class. Clearly, the presence of family material difficulties and behavioural issues will impede (mediate) this trajectory. Interestingly, childhood behaviour, in particular at age 16, has a lasting effect on mid-life PWB, EWB and Quality of Life.
In sum, health, emotion, quality of life and to a lesser extent cognitive function do not stand alone as markers of well-being in mid-life. They represent the constituents of an inter-related whole which is shaped by early life circumstances, family material difficulties, social adjustment and cognitive performance in childhood. 7 Szreter, S.R.S (1984). The Genesis of the Registrar-General's Social Classification of Occupations. British Journal of Sociology 35(4), 522-54  9 Marmot, M. G, Fuhrer, R, Ettner, S.L, Marks, N.F, Bumpass, L.L and Ryff, C.D (1998) Contribution of psychosocial factors to socioeconomic differences in health. Millbank Quarterly 76, 403-48. 10 Gale     23 Mishra  32 Zellner, A. 1962. An efficient method of estimating seemingly unrelated regression equations and tests for aggregation bias. Journal of the American Statistical Association 57: 348-368. 33 Arbuckle, J.L. (1996) 37 Weisstein

Consent of authors
I, Brian Dodgeon, the Submitting Author, have the right to grant and do grant on behalf of all authors of the Work (as defined in the below author licence), an exclusive licence and/or a non-exclusive licence for contributions from authors who are: i) UK Crown employees; ii) where BMJ has agreed a CC-BY licence shall apply, and/or iii) in accordance with the terms applicable for US Federal Government officers or employees acting as part of their official duties; on a worldwide, perpetual, irrevocable, royalty-free basis to BMJ Publishing Group Ltd ("BMJ") its licensees and where the relevant Journal is co-owned by BMJ to the co-owners of the Journal, to publish the Work in BMJ Open and any other BMJ products and to exploit all rights, as set out in our licence.
The Submitting Author accepts and understands that any supply made under these terms is made by BMJ to the Submitting Author unless you are acting as an employee on behalf of your employer or a postgraduate student of an affiliated institution which is paying any applicable article publishing charge ("APC") for Open Access articles. Where the Submitting Author wishes to make the Work available on an Open Access basis (and intends to pay the relevant APC), the terms of reuse of such  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Open Access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this Work are set out in our licence referred to above.

Competing interests
All four authors declare they have no competing interests and no financial relationships with any organisation that might have an interest in the submitted work in the previous three years, and no other relationships or activities that could appear to have influenced the submitted work.

Data Sharing statement
The analysis is based on secondary data available for download from the UK Data Archive at https://beta.ukdataservice.ac.uk/datacatalogue/series/series?id=2000032. Syntax used to derive variables for this analysis can be obtained by e-mailing the corresponding author, Brian Dodgeon.

3.
Vigorous activities, such as running, lifting heavy objects, participating in strenuous sports 4.
Moderate activities, such as moving a table, pushing a vacuum cleaner, bowling, playing golf? 5.
Climbing several flights of stairs? 7.
Climbing one flight of stairs? 8.

Appendix 7: Separate SEM path diagrams for men and women
Running the analyses separately by sex, we see that certain pathways are significant for one sex but not the other (p<.001) and certain path coefficients are notably different:

Men only (N=3815)
Women only (N=4209) An MPlus multigroup run by sex was performed as a sensitivity analysis, in which each of the 50 possible paths in Fig.1 was labelled m1-m50, and (in the female model) f1-f50, then a set of 50 'difference' variable was defined diff(i)=m(i)-f(i).
Testing the significance of these 50 'diff' variables, 6 were found to be significant, and the other 43 non-significant.
The MPlus multigroup model was then re-run, constraining all the 44 non-significant paths to be equal for males and females, but allowing all the other six paths to vary, thus increasing the parsimony of the model.
The six paths found to be significantly different between men and women are shown in this

Results
Using structural equation modelling, we explore numerous pathways through childhood and early adulthood which are significantly linked to our outcomes. We specifically examine the mediating effects of the following: cognitive ability at ages 7, 11 and 16 years; childhood psychological issues ; family material difficulties at age 7: housing, unemployment, finance; educational/vocational qualifications and social class position at age 42 years.
We find that social class at birth has a strong indirect effect on the age 50 outcomes via its influence on cognitive performance in childhood and adolescence, educational attainment and mid-life social class position, together with small direct effects on qualifications and social class position at age 42 years. Teenage cognitive performance has a strong positive effect on later physical health for women, whilst educational/vocational qualifications have a stronger positive effect on emotional well-being for men.

Conclusion
Our findings provide an understanding of the legacy of early life on multiple aspects of mid-life health, well-being, cognition and quality of life, showing stronger mediated links for men from childhood social class position to early adult social class position. The observed effect of qualifications supports those arguing that education is positively associated with subsequent cognitive functioning.

Article Summary
Strengths and limitations of this study:  The richness of its longitudinal dimension, conveying the importance of early life circumstances upon early adult and mid-life measures of achievement, health and wellbeing;  The integrated nature of the analysis simultaneously models pathways to four mid-life outcomes, three of which are strongly correlated after the inclusion of several key early-life antecedents;  It makes optimal use of all the available longitudinal data by employing Full Information Maximum Likelihood estimation to take account of item non-response;  It shows the enduring importance of parental social class for cognitive performance and later life outcomes, controlling for family material difficulties in childhood.  A potential limitation is that our model does not include a broader set of demographic and psychosocial factors: a result of our decision to restrict the number of early-life influences to avoid over-complexity in the statistical modelling. There is also a small amount of attrition bias between birth and age 50.

INTRODUCTION
Authors since the 1990s have advanced the importance of a life course approach to explore the relationship between early life circumstances and later life health outcomes [1,2]. Carr [3] emphasises how diverse early life experiences affect physical, emotional and cognitive health in later life by drawing upon a wealth of longitudinal data across the United States, Australia and Europe. Our empirical study is based on a single British Birth Cohort which covers several waves of data collection from birth (1958), middle childhood (age 7 & 11), adolescence (age 16) and mid-life at ages 42 and 50 years for over 8,000 sample members. Age 50 years represents a specific mid-point of mid-life (40-60) as defined by Midlife Development in the United States (MIDUS): Brim et al [4] suggest "Midlife has been described as the last uncharted territory of the life course". To this extent our conceptual framework is largely informed by Elder [5]: it may be described as a cumulative (dis-) advantage model [6] where we examine the legacy of early life circumstances, notably the British Registrar General's classification of occupations (RG social class) [7] as having lasting effects on health and well-being at age 50 years. Our approach implies a reciprocal relationship between socio-economic status (SES) and health, and allows for the possibility that favourable circumstances in later life can act counter to the effects of earlier disadvantage [8].
To this end we select outcomes at age 50 years covering four distinct areas: self-assessed quality of life, physical and emotional well-being, and a cognitive ability test.
Our research interest is in exploring the extent to which these outcomes inter-relate, adopting an empirical view that, taken together, these measures would tell us more about the consequences of the past than studied individually [9]. There is a growing body of evidence based on United Kingdom (UK) longitudinal studies [10,11,12] that provides a strong argument for looking at these outcomes together within the same model. Cooper et al [13] provided both substantive and empirical evidence for examining gender differences in this analysis.
Drawn from the same rich longitudinal resource, our predictors and mediators begin with birth circumstances: birthweight, breastfeeding, maternal smoking during pregnancy and parental SES. From the age 7 sweep of the study we include the effects of family material difficulties such as unemployment, finance and housing, in conjunction with childhood scores of socio-emotional adjustment and cognitive performance, each measured on three occasions spanning the ages 7, 11 and 16 years. The inclusion of highest qualification and SES at age 42 in the model enables us to trace which childhood effects persist, and which are built upon or attenuated by education and social mobility.
Our analysis builds upon the framework adopted by Wood et al [14] on the prediction of mental well-being as a single outcome [15] across four UK birth cohort studies, which showed that childhood SES is directly and indirectly (through adult socioeconomic pathways) linked to adult Our analysis adopts a structural equation modelling (SEM) approach [16], in order to explore the pathways from childhood and adolescence to our multivariate mid-life outcomes for both men and women.

DATA AND METHODS
Our data source, the British National Child Development Study (NCDS) is fully representative of the population of Great Britain (GB), having interviewed the parents of 98.5% of all babies born in the same week of 1958 [17], and followed them up longitudinally with ten subsequent interviews throughout life. It provides operational measures regarding birth circumstances, the early and teen years covering ages 7, 11 and 16, and adulthood at ages 42 and 50 respectively.
The analysis is based upon 3815 male and 4209 female cohort members present at each occasion who all have at least one genuine response for each of the four separate age 50 life domains, and who also participated in at least one of the three cognitive and socio-emotional assessments, in order to permit a reliable strategy for handling missing items.
We are aware there is a certain amount of attrition bias as a result of differential loss-to follow-up from the 17,415 cohort members present at the birth study in 1958. As a check on the impact of sample loss, we were reassured by the close match of the distribution of the birth social class variable within our sample of 8,024 against that of the same variable within the full 17,415 birth sample (chi-square=1.29, 5d.f., non-significant) (see Appendix 1).
Our outcomes at age 50 years cover four distinct measures, the first two being self-reported physical and emotional well-being (PWB and EWB respectively) measured using sub-scales [18] of the RAND 36-Item Short-Form Health Survey [19][20][21][22][23] (hereafter SF-36). Third, a self-reported measure of quality of life (QoL) covering the life domains of 'Control, Autonomy, Self-Realisation and Pleasure (CASP-12 v2) [24]: a theoretically-informed measure of subjective well-being, closely following the approach adopted by Ed Diener and his colleagues [25]. The fourth domain outcome, cognitive ability (hereafter Age50Cog) consists of four cognitive tests examining memory (word recall & delayed recall), executive function (animal naming), attention and mental speed (letter cancellation) [26].
Appendix 2 lists all of the manifest and latent variables employed in the analysis. Indicators for cognitive function at age 50 years are listed here, while those for physical, emotional and quality of life outcomes are expanded upon more fully in Appendix 3. Childhood behavioural indicators are listed in Appendix 4 following Rutter [27]. We recognise of course that Rutter scores may be subdivided into constituent dimensions of problem behaviour (withdrawn/ anxious / oppositional/ inattentive [28], but for this paper we prefer to use Rutter total scores, following Parsons et al [29]. In fitting our structural equation models using the MPlus software [30] we obtain direct and indirect (mediated) effects of our background variables in terms of childhood social origins and social adjustment, on outcomes in early adulthood, and mid-life at age 50 years. We regard all four of our birth indicators as primary predictors, valuing the opportunity to assess the relative impact of each for policy interventions [31]. Our birth social class variable is based on father's occupation, since in 1958 when the cohort were born, only 38% of mothers were economically active.
The sets of predictors determining our paths differ slightly for each of the four outcomes so as to avoid the problem of 'seemingly unrelated regression' [32] We employ full information maximum likelihood estimation (FIML) [33]. This addresses the issue of item non-response arising where questions are avoided or not answered by certain participants.
The SEM procedure that follows begins with an analysis of the sample as a whole, followed by separate analyses for men and women, supported by a multigroup analysis [34] to test for the structural equivalence of the models for men and women. Model 'goodness of fit' is assessed via a combination of conventional criteria: 'root mean square error' (RMSEA), 'comparative fit index', (CFI) and 'Tucker Lewis fit index', (TLI) where RMSEA<0.05, CFI>0.90 and TFI> 0.90 would typically provide evidence of 'good fit,' though each of these criteria individually are regarded as indicative, and are not always strictly adhered to as arbiters of model validity [35].
Having a large number of SEM pathways involving 261 free parameters, we had regard to the n:q rule [36] and were satisfied that our sample-size-to-parameters ratio was comfortably within Jackson's guideline of 10:1. Our application of SEM involved the use of modification indices, which led to our running correlations on suggested variables. All our final models were identified, all manifest variables loading onto the respective latent variables with satisfactory fit. The details of the individual assessments of our measurement models appear in Appendix 5.

Patient and Public involvement statement
Participant consent was obtained from parents of cohort members at the initiation of the original birth cohort study in 1958. On attaining adulthood, permission was obtained from cohort members themselves at each subsequent longitudinal sweep.

RESULTS
The results begin with consideration of the inter-correlations between our mid-life outcome measures. Further univariate descriptives are contained in Appendix 2. Figure 1 contains a final SEM diagram for the sample as a whole. Subsequent reference will be made to Appendix 6 for the details of mediation, total and indirect effects. Appendix 7 contains separate analyses by sex, and Appendix 8 contains further information on our multigroup analysis based on gender.
Note that all coefficients reported in the Results section are standardised.

(a) Descriptives & correlations
As a preliminary inspection of the inter-relationships between our four outcome measures, Table 1 contains a correlation matrix of the four summative indices based on these measures for both men and women.  1 contains estimates of association amongst these outcomes as defined by measurement models under a full SEM analysis. As a check on the robustness of our statistical findings the SEM analyses were subject to bootstrap analysis under 1,000 repetitions. We also carried out a check on the influence of outliers (results available on request).

(b) Birth social class effects on early mid-life outcomes
The path diagram shown in Fig 1 provides
Family material difficulties in childhood, themselves associated with birth social class (-0.26), also play their part in understanding a child's trajectory in cognitive performance. Characterised by issues with housing, finance and employment when a child is 7 years old, these family material difficulties are seen to have negative consequences for cognitive performance at age 7 years (-.15) and a lagged negative effect on later performance age 16 years (-0.04).
Additional analyses also show that family material difficulties impact on a child's emotional and behavioural scores (based on Rutter) at ages 7, 11 and 16 (.09, .06 and .10 respectively). In turn these scores had small significant negative correlations with cognitive performance at each corresponding age (-.15, -.11 and -.11 respectively). But most notably, Rutter scores at these three ages are significantly correlated (.49 for age 7 to 11 and .44 for age 11 to 16), and there are direct effects from age 16 Rutter scores to three of our four age 50 outcomes: EWB (-0.13), QoL (-0.10) and PWB (-0.07).

(d) Indirect effects via adulthood mediators
Adult social class and lifetime qualifications are, together, influential predictors of each of our four mid-life outcomes: adult social class has significant predictive effects (.08 and .11) on PWB and QoL.
In general, the socially-advantaged enjoy better health and report higher levels of quality of life than their less advantaged counterparts. Interestingly, qualifications predict age 50 cognitive ability (.05), EWB (.06) and QoL (.10) but not PWB.

(e) Gender differences
Our multigroup analysis reveals that certain paths have significantly different effects for men and women. In Appendix 7 we display two separate path diagrams for men and women. For men, there is a direct effect of social class at birth on Age 42 social class and qualifications (.09 and .07 respectively), but for women there is no significant effect.
Both sexes see the birth social class effect remain equally strong for cognitive performance at age 7 years and beyond, but women experience a slightly worse negative effect of social disadvantage at birth on the latent variable 'family material difficulties' (-0.27 compared to -0.24 for men). In turn, family material difficulties have slightly more influence on Age 7 cognition (-.18) for men than for women (-.16), and a small lagged effect on Age 16 cognition only evident for women (-.05). The Rutter social adjustment scores at age 11 do not significantly correlate with the cognitive performance scores for men, but they do for women at ages 7, 11 and 16 (-.15, -.11 and -.13 compared with men's -.14 at age 7, -.11 at age 16).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Early adult social class has a slightly stronger influence for men than women on our age 50 outcomes PWB and QoL, and Qualifications have a significant effect on EWB for men, but not for women. In contrast, women's age 16 cognition has a significant effect on their later PWB (.17), an effect not significant for men; and the path from age 16 cognition to age 50 Cog has a stronger coefficient for women (.49) than men (.43).
We pursue the evidence for gender differences via a sensitivity analysis under MPlus which formally tests for statistical difference between paths for each sex. Appendix 8 identifies six paths where the effects for men and women are significantly different, but only three have notable differences in terms of their effect sizes in the multigroup analysis. These are:  stronger effect for women from age 16 cognition to age 50 cognition as noted above (.49 to .43);  direct effect of social class at birth on Age 42 social class (.09 for men, no significant effect for women); and  link for women from Age16 Cognition to Phys WBeing (.17)(no significant effect for men)

DISCUSSION
Structural equation modelling has elucidated the nuances of the influence of pathways from childhood and early adulthood to our four related outcomes of health, well-being and cognition in mid-life. The work demonstrates the benefits of an integrated approach to defining the aspects of the life course that affect well-being in mid-life. Physical and emotional health are important factors affecting a person's subjective account of their quality of life, and PWB correlates at .69 with EWB, which in turn correlates .83 with QoL. While cognitive performance at age 50 is relatively weakly correlated, the other three are closely intertwined. These associations are similar in magnitude to those reported in table 1; however they are now estimated by taking measurement error and the legacy of the life course into account.
We also show that some of the paths to health and well-being in mid-life are different for men and women [38]. However, important communalities remain.
Social origins, in particular social class at birth, have a major indirect influence on mid-life outcomes, the mediation process beginning with a child's cognitive performance throughout formal schooling and subsequently via cognitive performance at age 16 years. Collectively, these influences are weakened or reinforced by the existence of family material difficulties in early childhood (themselves influenced by childhood social class). Whilst cognitive performance is a strong predictor of early adult social class and educational status for both men and women, social class at birth also has a lasting association with these mid-life indicators for men but not for women.
This could well suggest that women have a looser connection with the legacy of social class at birth, perhaps partly because their path to advantaged SES at age 42 may be impeded by caring responsibilities and gender inequalities in the labour market [39].
Early adult social class influences all four of our mid-life outcomes, but we see the direct effect of social class at birth on social class and qualifications in early adulthood is pronounced for men (.09 and .07 respectively), but not significant for women.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y A relatively small but consistent direct effect of qualifications on Age 50 Cognition supports those arguing that education is positively associated with subsequent cognitive functioning [40], although we are aware that there are counter-arguments to this view [41]). Equally, the relationship between QoL and early adult social class finds support in the work of Blane et al [42].
The strength in the evidence presented here lies in the richness of its longitudinal dimension which conveys the importance of early life circumstances upon early adult and mid-life measures of achievement, health and well-being.
One limitation of our study is that our model does not fully take into account all of the potential demographic and psychosocial factors other researchers have studied as predictors of satisfactory midlife development [28]. Nevertheless, it can be argued that our birth social class variable acts as a 'proxy' for certain parental beliefs and behaviours [43]. We are embarking on developing a conceptual framework and modelling strategy to extend our empirical analyses in the future as our cohort members age: in particular, our subjects are not yet old enough for us to assess whether any decline in cognitive performance at age 50 years will lead to subsequent decline and become more influential on emotional well-being and quality of life. However, we were able to show that cognitive performance in the teen years holds a legacy for cognitive performance in mid-life. A final limitation is a certain amount of attrition bias between birth and age 50 (differential non-response) which was addressed earlier in the paper and in Appendix 1.
Social class at birth has a strong influence on early cognitive performance and beyond, which in turn has consequences for adult achievement and social class. Clearly, the presence of family material difficulties and behavioural issues will impede (mediate) this trajectory. Interestingly, childhood behaviour, in particular at age 16, has a lasting effect on mid-life PWB, EWB and Quality of Life.

Consent of authors
I, Brian Dodgeon, the Submitting Author, have the right to grant and do grant on behalf of all authors of the Work (as defined in the below author licence), an exclusive licence and/or a non-exclusive licence for contributions from authors who are: i) UK Crown employees; ii) where BMJ has agreed a CC-BY licence shall apply, and/or iii) in accordance with the terms applicable for US Federal Government officers or employees acting as part of their official duties; on a worldwide, perpetual, irrevocable, royalty-free basis to BMJ Publishing Group Ltd ("BMJ") its licensees and where the relevant Journal is co-owned by BMJ to the co-owners of the Journal, to publish the Work in BMJ Open and any other BMJ products and to exploit all rights, as set out in our licence.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y The Submitting Author accepts and understands that any supply made under these terms is made by BMJ to the Submitting Author unless you are acting as an employee on behalf of your employer or a postgraduate student of an affiliated institution which is paying any applicable article publishing charge ("APC") for Open Access articles. Where the Submitting Author wishes to make the Work available on an Open Access basis (and intends to pay the relevant APC), the terms of reuse of such Open Access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this Work are set out in our licence referred to above.

Competing interests
All four authors declare they have no competing interests and no financial relationships with any organisation that might have an interest in the submitted work in the previous three years, and no other relationships or activities that could appear to have influenced the submitted work.

Data Sharing statement
The analysis is based on secondary data available for download from the UK Data Archive at https://beta.ukdataservice.ac.uk/datacatalogue/series/series?id=2000032. Syntax used to derive variables for this analysis can be obtained by e-mailing the corresponding author, Brian Dodgeon.
These were then divided into eight sub-domains (four physical health, four emotional well-being, with the recoded variables being summed then divided by the number of questions asked: These are modest direct influences on PWB and QoL, but a relatively stronger negative influence on EWB at age 50 years. Two direct influences stand out in Table B: from social class at birth to social class at 42 years and Qualifications obtained. The direct influences are shown in our estimated pathways diagram (Fig.2), but we also see in Table C there is an additional indirect effect (via childhood cognition at ages 7/11/16), over twice as large as the direct effect in both cases. An MPlus multigroup run by sex was performed as a sensitivity analysis, in which each of the 50 possible paths in Fig.1 was labelled m1-m50, and (in the female model) f1-f50, then a set of 50 'difference' variable was defined diff(i)=m(i)-f(i).
Testing the significance of these 50 'diff' variables, 6 were found to be significant, and the other 43 non-significant.
The MPlus multigroup model was then re-run, constraining all the 44 non-significant paths to be equal for males and females, but allowing all the other six paths to vary, thus increasing the parsimony of the model.
The six paths found to be significantly different between men and women are shown in this The first three paths are notable. The fourth is only significant at the .002 level, and in the case of the last two, the difference in the point estimates, though significant, is so small as to be unremarkable.
We see that women's Age 50 cognitive ability has a stronger link with Age 16 cognition (.49) than men's (. 43), and that for men there is a direct link from SC Birth to SC 42 yrs (.09), a result not significant for women.