Childhood economic conditions and length of life: Evidence from the UK Boyd Orr cohort, 1937–2005

Abstract

We study the importance of childhood socioeconomic conditions in predicting differences in life expectancy using data from a large sample of children collected in 16 locations in England and Scotland in 1937–39, who have been traced through official death records up to 2005. We estimate a number of duration of life models that control for unobserved family heterogeneity. Our results confirm that childhood conditions such as household income and the quality of the home environment are significant predictors of longevity. Importantly, however, the role of socioeconomic status appears to differ across cause of death, with household income being a significant predictor of death from smoking-related cancer. Moreover, we find that (1) poor housing conditions in childhood is associated with reduced longevity, that (2) early doctor-assessed childhood health conditions significantly predict a reduced length of life, that (3) children born in a location with relatively high infant mortality rates live significantly fewer years, and that (4) there is a high correlation in longevity across children from the same family across all causes of death. We estimate that the difference in life expectancy between those with the ‘best’ and ‘worst’ observable characteristics is about 9 years, which increases to 20 years when we take into account the ‘best’ and ‘worst’ observable and unobservable household characteristics.

Introduction

There has been much recent interest both among epidemiologists and economists in how socioeconomic conditions during childhood influence health outcomes in later life. Here we examine longevity, arguably the most important outcome. We use a dataset for children originally surveyed in Britain during the late 1930s combined with follow-up data on the date and the cause of death. This data offers a unique combination of a rich array of variables recorded during childhood and a long subsequent timespan. We use a Mixed Proportional Hazard Model to estimate the effects of a variety of individual, household and local characteristics to estimate the net effects of a variety of different childhood conditions on longevity.

Lifecourse epidemiology has been studied in a range of different settings. In their review of the literature Galobardes et al. (2004) nevertheless comment that there has been ‘relatively little investigation of how early life circumstances influence adult health’ (p. 8). Adverse conditions in infancy and childhood can influence health in later life either directly (latent effects), or cumulatively, or by their influence on subsequent lifecourse pathways. These effects can operate through continuities either in physiological or metabolic risk factors or in socioeconomic status, which in turn links to behavioural risk factors (Ben Shlomo and Kuh, 2002, Kuh et al., 2003). While there has been considerable interest in the pathways and mechanisms that lead from childhood circumstances to adult health status, this has sometimes obscured the net (or reduced form) effects on chronic disease and death.

A range of studies have examined the effects of socioeconomic status on health at different stages of the lifecourse. Analysis of the Whitehall II study of British civil servants in 1985–88 shows that mortality risk factors such as chronic heart disease, chronic bronchitis and depression are strongly related to father's social class, but these effects are severely attenuated when current employment grade is included as an explanatory variable (Brunner et al., 1999, Marmot et al., 2001). However, adult socioeconomic status is closely related to that in childhood, suggesting strong socioeconomic pathways.¹ Power et al. (2005) provide similar evidence on more representative data for six countries linking child and adult socioeconomic status to health risk behaviours such as smoking and obesity. In an analysis of the National Child Development Study (NCDS), a 1958 UK Birth Cohort, Case et al. (2005) found that individual health and socioeconomic background were strongly related to self-reported health status in adulthood. Exploring the intervening mechanisms, they conclude that health outcomes in adulthood are largely carried forward from childhood. Using the same birth cohort, Power et al. (2007) find similar evidence for clinical health conditions. Smith (2009) points to the importance of controlling for family unobservable heterogeneity in establishing the link between childhood conditions and adult outcomes. Taken together studies such as these suggest that childhood conditions influence subsequent health risk directly as well as through their influence on status and behaviour later in life.

A number of studies have focused on the overall relationship between childhood socioeconomic status and mortality. In a study of a Norwegian cohort born between 1955 and 1965 Strand and Kunst (2007) found that childhood socioeconomic conditions such as parents’ income and education and occupational class are strongly associated with early cardiovascular mortality but that this is severely attenuated when the individual's own socioeconomic status was included.² In another recent study Hamil-Luker and O’Rand (2007) examined the incidence of heart attack in a US sample of over-50s. They found that the latent effects of socioeconomic deprivation were more important for women while path-specific effects were more important for men.

Studies of mortality in cohorts born before 1940 are rare because there are few contemporary surveys for which subsequent life events can be traced; and those based on data collected retrospectively are subject to survivor bias. However, a recent study by Van den Berg et al. (2006) examined deaths in three Dutch provinces from 1812 to 1992 to explore the effects of early life conditions on mortality. Using a Mixed Proportional Hazard (MPH) Model they found significant positive relationships between longevity and father's social class and literacy as well as a negative effect of being born in a recession. While this approach focuses on the reduced form effects of childhood conditions (particularly macroeconomic conditions) on mortality, it is limited by the lack of detailed household-level information on income, consumption and family structure.

In this paper we contribute to this literature by exploring the role of childhood socioeconomic conditions of children living in England and Scotland in the 1930s in predicting their length of life. The data collected between 1937 and 1939 are from the Carnegie U.K. Trust's Study of Family Diet and Health in Pre-War Britain (referred to as the Boyd Orr cohort after Sir John Boyd Orr, the Director of the Rowett Institute, where the study was undertaken). This was one of the very first studies to focus specifically on children and it recorded a range of information on the living conditions and household characteristics of 4999 children in 1343 families in 16 different localities. Data on mortality for these individuals from the time of the original survey up to 2005 has been collected from official death registers by researchers at the Department of Social Medicine at the University of Bristol. Our analysis builds on several papers by medical researchers who have used these data to study mortality (see Frankel et al., 1999, Dedman et al., 2001).³ We use this data to estimate multivariate single and competing risks duration models, focusing on the effects of variables such as income and food consumption, as well as demographic variables and housing conditions. We also evaluate neighbourhood effects using data on local unemployment and infant mortality.

Our data has a number of advantages as compared with much of the literature. First, as the individuals in the sample were directly observed as children, we avoid the selectivity associated with sampling a population that had already survived into middle age. Secondly, the survey provides a rich array of variables on the living conditions and the socioeconomic and demographic structure of the households in which they grew up. Thus our explanatory variables are not simply restricted to a few indicators such as social class or parents’ education which are collected retrospectively and may be subject to recall bias. Third, unlike studies based on birth cohorts, we observe multiple children in each household, and we use this information to improve the identification of unobserved heterogeneity in our models. Fourthly, because these individuals were observed in the 1930s a significant proportion of them have died and hence we are not restricted only to the study of premature mortality. Finally, we have information on disease-specific mortality that allows us to test whether childhood living conditions have differential effects on the probability of dying from heart disease, all cancers, and all other causes at any time during the life cycle.

Against this we must set three disadvantages. Firstly, because the individuals in the sample were observed as children and only in one period, we cannot assess the effects of changing macroeconomic conditions (although we do observe the differences across localities). Secondly, the individuals in the survey were observed once at ages up to 18 and so we do not follow them from birth. Thirdly, we do not have information on the entire subsequent histories of those in the sample and so we do not examine the relative effects of socioeconomic status at different stages of the lifecourse. Thus we are not able to explore the different pathways through which conditions in childhood influence behavioural or metabolic risk factors later in life and so our results must be interpreted as reduced form effects.

The rest of the paper is set out as follows. In Section 2, we describe the historical context of the period in which the children in the sample were first observed. Section 3 provides details of our modelling strategy for the duration of life. The results of the estimation are presented in Section 4 and the interpretation and significance of these results is discussed in the concluding Section 5.