Socioeconomic Status and Health
Socioeconomic status is a multifaceted concept that is often measured by education, income, or occupation. A vast body of empirical research on contemporary developed countries has established a strong and robust relationship between each of these measures and health. Higher education is associated with lower adult mortality in Britain (Kunst and Mackenbach
1994; Mackenbach et al.
2003), the United States (Cutler et al.
2012; Halpern-Manners et al.
forthcoming; Hayward et al.
2015; Kunst and Mackenbach
1994; Masters et al.
2012), continental Europe (Kunst and Mackenbach
1994), and Scandinavia (Brønnum-Hansen and Baadsgaard
2007; Kravdal
2017; Olausson
1991; Shkolnikov et al.
2012; Steingrímsdóttir et al.
2012; Torssander and Erikson
2010; Vågerö and Norell
1989).
Similarly, occupational rank is related to adult mortality. For example, in the Whitehall studies of British civil servants, a clear health gradient was found in terms of job status (Marmot
2004:27; Marmot et al.
1991). In Sweden, there also appears to be a clear social class gradient in mortality: higher social class is associated with lower mortality overall (Burström et al.
2005b; Hartman and Sjögren
2018; Olausson
1991; Torssander and Erikson
2010; Vågerö and Norell
1989).
There are also differences in mortality by income, wealth, or poverty (e.g., Case and Deaton
2017; Chetty et al.
2016; Elo
2009; Hederos et al.
2018; Smith
1999; Torssander and Erikson
2010). Higher income and/or more wealth is consistently related to lower mortality, even though the strength of the relationship depends on age and is often attenuated when education is controlled for.
Lifestyle factors are often mentioned as important reasons why low socioeconomic status is related to worse health and higher mortality in contemporary settings, through higher smoking prevalence, higher alcohol consumption, greater inactivity, and higher obesity rates (Adler and Stewart
2010; Cavelaars et al.
2000; Elo
2009; Marmot
2004; Norström and Romelsjö
1998; Razzell and Spence
2006; Smith
1999; Vågerö and Norell
1989). Such factors may also indirectly contribute to health given that the success of medical treatment partly depends on lifestyle (Mills et al.
2011).
Differences in access to health care could also be an important explanation for the health gradient. This is especially true in contexts lacking universal provision of health care at a low cost (Adler and Stewart
2010) but also where health care is universal and affordable (van Doorslaer et al.
2000). In the latter case, the health gradient could be related to the underutilization of health services by groups of lower socioeconomic status (Steingrímsdóttir et al.
2012). This could also explain why there is no strong evidence that increased provision of health care actually reduces the health gradient (Smith
1999). Thus, it is not certain that increasing access to affordable health care would eliminate the health gradient in contemporary societies (Adler and Stewart
2010).
Environmental factors could also contribute to the health gradient if different socioeconomic groups systematically were to become exposed to various forms of air and water pollution as well as to other factors influencing the quality of life, such as crime. Adler and Stewart (
2010), however, argued that such environmental factors are not of major importance for the modern health gradient.
Marmot (e.g., 2004) stressed the role of psychosocial factors in generating socioeconomic differences in health. Theoretically,
psychosocial factors refer to the degree to which individuals can control their life situation, focusing in particular on the work situation. Lack of control leads to stress, which negatively affects health through different physiological mechanisms, including increased blood pressure and susceptibility to infection as well as the clogging of blood vessels. What matters is not only
exposure to stress but also the individual’s ability to
cope with it. In particular, the combination of high exposure to stress and low levels of coping leads to negative health effects (Adler and Stewart
2010). Typically, low socioeconomic status is associated with both more stress and less ability to cope because of weaker social networks and a lack of other resources (Marmot
2004; Steptoe and Kivimaki
2013). Even though the health penalty linked to low socioeconomic status to a certain extent is context-specific and attenuated in more equal societies, Marmot (
2004:27–31) asserted that the links between social position, stress, and health has been present at least since the nineteenth century. Others have argued that the gradient in the Whitehall study is due to health selection into higher classes and not due to stress or the ability to cope with stress (Case and Paxson
2011).
Differences in diet and nutrition are important for socioeconomic differences in mortality. In today’s developed countries, an unhealthy diet is strongly associated with social class (Darmon and Drewnowski
2008), promoting obesity and leading to conditions such as diabetes, hypertension, and heart disease. In the past, malnourished individuals were more likely to die from a number of low-virulent diseases, including measles, diarrhea, tuberculosis, respiratory diseases, pertussis, other intestinal diseases, cholera, leprosy, and herpes (Rotberg and Rabb
1985:305–308). During the first part of the nineteenth century, a pronounced mortality response, particularly among the lower classes, to changes in food prices in Sweden indicates that segments of the population overall were very vulnerable before the onset of modern economic growth (Bengtsson
2004; Bengtsson and Dribe
2005; Bengtsson and Ohlsson
1985). Indeed, improvements in nutritional status have been the leading explanation for the decline of mortality in the nineteenth century (Floud et al.
2011; see also Fogel
2004; McKeown
1976,
1983), but there are also dissenting views (e.g., Easterlin
1996; Livi-Bacci
1991; Szreter
1988).
Finally, it has been suggested that the relationship between socioeconomic status and health in adulthood may have its origin earlier in life (Elo
2009; Smith
1999). Conditions in early life (during the fetal stage and infancy) may have long-lasting impacts on health (e.g., Almond
2006; Barker
1998; Bengtsson and Lindström
2003; Bleakley
2007; Case and Paxson
2008; Elo and Preston
1992; Finch and Crimmins
2004; Quaranta
2013), and exposure to poor nutrition or disease during early life can affect organ development and program the onset of disease in adulthood as well as influence an individual’s cognitive ability. This could in turn affect both health and socioeconomic attainment and thereby explain part of the association between socioeconomic status and health in adulthood that has been observed (e.g., Chandra and Vogl
2010; Cutler et al.
2012).
Although low socioeconomic status is often assumed to have a causal effect on health, some have argued that the direction of causality is more likely to be reversed (e.g., Chandra and Vogl
2010; Cutler et al.
2012; Deaton
2003; Montez and Friedman
2015; O’Donnell et al.
2015; Smith
1999,
2004). Despite this, several empirical studies using quasi-experimental designs have identified a causal effect of socioeconomic status on health or mortality (e.g., Lindahl
2005; Lleras-Muney
2005; Lundborg et al.
2016; Spasojevic
2010).
A Changing Gradient Over Time?
The socioeconomic mortality gradient in many contemporary settings has been comprehensively mapped, but far less is known about mortality differences by socioeconomic status before the 1970s because of the lack of sufficiently detailed data. Especially glaring is the lack of longitudinal data; most studies for this period are based on cross-sectional data, such as those of Willermé on Paris in 1817, Engels on Manchester in 1850, and Virchow on Upper Silesia in 1847–1848 (see Deaton
2016:1703). In another example, Chapin (
1924) used cross-sectional data to examine the mortality differences between taxpayers and nontaxpayers in 1865 in Providence, Rhode Island, indicating the important role of poverty in health and mortality. The findings revealed higher overall mortality as well as mortality by several important causes of death, such as pulmonary tuberculosis, heart disease, and respiratory diseases among the nontaxpayers but only small differences for contagious diseases. In addition, Blum et al. (
1990) found substantial socioeconomic differences in remaining life expectancy at age 40 in a study of marriage certificates in Paris in the 1860s, which also included information on age at death of the deceased parents of the bride and groom. A problem found in many of the cross-sectional studies is that the population at risk was not well measured, which results in biased estimates of mortality differences (see Bengtsson and van Poppel
2011).
The ambiguous results from previous research using cross-sectional data for the period before the 1970s have also led to contradictory views in the literature as to whether the socioeconomic gradient in mortality has widened, narrowed, or remained constant since the early phases of the mortality transition. The fact that public health measures, as well as subsidized health care, reached an increasing share of the population during the course of the twentieth century could be expected to have led to a convergence in mortality across social strata (Antonovsky
1967). Because of the redistribution of resources between individuals through the tax system, the level of economic well-being enjoyed by the societies’ poorest people has increased, which should promote a reduction in mortality differences.
This view was challenged by the
fundamental causes theory, arguing that mortality differences have remained more or less constant over the past 200 years (Link and Phelan
1996; Phelan et al.
2010). Although the specific mechanisms have varied over time, the higher-status groups have always had a mortality advantage because of their greater resources. A recent version of the fundamental causes theory, which is closer to Antonovsky’s view, attempts to take aspects of both the demographic transition and the epidemiological transition theories into account (Clouston et al.
2016). The argument is that as mortality declined, new diseases came to dominate total mortality but with each new disease going through similar phases. Early on, diseases were largely nonpreventable because of a lack of knowledge regarding the causal agents and treatment. In this stage of “natural mortality,” socioeconomic differences in mortality from the disease were usually small, and they could even be in favor of groups with lower socioeconomic status. Social differences arose in the following stage, mainly because of new knowledge on how to prevent disease, which favored the high-status groups, who were quicker to acquire the new information and change their behavior. With a lag, mortality from the disease among the groups with lower socioeconomic status also started to decline; after a while, the rate of improvement was faster among the low-status groups, and inequalities were reduced. This process was repeated disease by disease and in all stages of the mortality transition, except before it began, which is why high-status groups had lower all-cause mortality. In this sense, socioeconomic status is a fundamental cause, even though the precise mechanisms might be different for each disease and in each specific period.
The number of studies using longitudinal data to explore changes in mortality differences by social class over time is low. Van Poppel et al. (
2009) found a class gradient in adult mortality in parts of the Netherlands for much of the nineteenth century, which narrowed over time up to 1920, when their study ended. Before then, both the elite and artists lived longer than other groups (van Poppel et al.
2009; van Poppel et al.
2013), but medical professionals did not have this advantage (van Poppel et al.
2016). In fact, medical professionals in Britain also had a shorter life expectancy until approximately 1900 (Woods
2004). British and Russian academics had an advantage in life expectancy at age 50 far back in time, and the gap widened after the 1950s (Andreev et al.
2011). There is evidence of an emerging class gradient in adult mortality in the 1930s in Britain (see also Pamuk
1985; Woods
2000:207).
Recent research on the United States, however, has reported only modest mortality differences by educational attainment for cohorts born during the end of the 1800s and early 1900s (Masters et al.
2012). Additionally, other studies of different historical contexts before the modern period found only minor social differences in adult mortality for men or for both sexes combined (Alfani and Bonetti
2019; Bengtsson and Dribe
2011; Edvinsson
1992; Edvinsson and Broström
2012; Edvinsson and Lindkvist
2011; Smith
1983). Two studies investigating gender differences in mortality by socioeconomic status in Sweden and Estonia found a mortality advantage for higher-status women but not for higher-status men around the turn of the twentieth century (Dribe and Eriksson
2018; Jaadla et al.
2017). In some cases, higher-status men even have a higher mortality than lower-status men, most likely as a result of adverse lifestyles (Dribe and Eriksson
2018; Razzell and Spence
2006).
Other studies using longitudinal data argued that mortality differences in the past were small, or possibly even reversed, because mortality was mainly due to communicable and often highly virulent diseases (e.g., Bengtsson and Dribe
2011; Smith
1983; Woods
2004). Because of the nature of the predominant diseases, the upper classes were possibly even more exposed, which in combination with the lack of effective treatment resulted in higher mortality. Some researchers have also noted that spatial differences in mortality were often much larger than socioeconomic differences in the past (Edvinsson and Lindkvist
2011; Garrett et al.
2001; Reid
1997; Smith
1983; van Poppel et al.
2005; Woods
2000,
2004; Woods et al.
1993).
Context and Data
We use data from the Scanian Economic Demographic Database (SEDD) for the period 1813–2015. These data consist of individual-level longitudinal information from five rural and semi-urban parishes and, after 1922, the port town of Landskrona in southern Sweden (Bengtsson et al.
2018). The five parishes have a combined population of 4,500 in 1830; 5,500 in 1900; and, together with Landskrona, 37,500 in 2000. The database is one of the very few that can follow individuals across multiple generations from preindustrial times up to the present, with detailed and frequently updated information on occupation and on different demographic outcomes, including migration. The latter is very important because it provides a precise measure of the population at risk. The study population is not a random sample of Sweden but is broadly representative by reflecting conditions shared by populations in similar areas during the time studied (see Bengtsson
2004; Dribe and Helgertz
2016; Dribe et al.
2015; Lazuka
2017:56–60). More specifically, for the period 1813–1920, for which we have data for only the five rural and semi-urban parishes, the area reflects the population density, age, and occupational structure in Sweden outside of Stockholm. In fact, until about 1930, more than half of the Swedish population lived in rural areas. From 1922, the addition of Landskrona allows for the examination of a quintessential industrial town, again reflecting Sweden outside of Stockholm as a whole.
Life expectancy increased throughout the study period. During the twentieth century alone, life expectancy at birth in Sweden rose from 52 to 77 years for men and from 55 to 82 years for women. Mortality rates for men aged 30–34 years fell from 6 to under 1 per thousand between 1900 and 1997 and from 57 to 34 per thousand at ages 70–74 years. Over the same period, mortality for women declined from 6 to 0.4 per thousand at ages 30–34 and from 50 to 19 per thousand at ages 70–74 (Statistics Sweden
1999: table 5.3) As mortality declined, disease patterns also changed, from a predominance of infectious diseases to chronic diseases, such as cardiovascular disease and cancer (Preston
1976). Life expectancy in the study area was very similar to that for Sweden as a whole, although it was slightly higher for cohorts born between 1850 and 1900. Likewise, causes of death followed a pattern similar to that in Sweden as a whole (Lazuka
2017: figs. 6 and 7). The share of mortality due to influenza, pneumonia, and diarrhea fell from more than 30% to 10% in the period 1920–1950. Meanwhile, as in other parts of Sweden, the share of mortality in chronic diseases, cardiovascular diabetes, and cancer increased.
The study period is broken down into six subperiods: 1813–1864, 1865–1919, 1920–1949, 1950–1969, 1970–1989, and 1990–2015. The first period (1813–1864) corresponds to the preindustrial or early industrial phase, when adult mortality was at pretransitional levels and 90% of the population lived in rural areas. For this period in this area, we know that there was a socioeconomic gradient in child and adult mortality between peasants and agricultural laborers in the response to changes in food prices, indicating differences in nutritional status (Bengtsson
2004), which disappeared in subsequent periods (Bengtsson and Dribe
2005). The major health intervention in this period was smallpox vaccination, which started in 1801 and became compulsory in 1816 (Sköld
1996), but there were no pronounced class differences in vaccination rates (Dribe and Nystedt
2003).
Starting in the 1860s, Sweden experienced its industrial breakthrough, with rapid mechanization of agriculture and increasing urbanization and real wages for workers (Schön
2010). From about 1850, adult mortality started its continuous decline (von Hofsten and Lundström
1976). New public health measures, such as improved education for midwives and the establishment of isolation hospitals, were implemented (Lazuka
2018). Investments in improved water and sanitation were also made in urban areas; and in 1900, 50% of the towns had new water systems, and 60% had sewage treatment, both of which contributed to the eradication of the urban mortality penalty by 1930 (Helgertz and Önnerfors
2019). Nevertheless, there were socioeconomic differences in child mortality in this period, particularly in urban areas (Burström et al.
2005a; Edvinsson
1992; Molitoris and Dribe
2016). There were also class differences in marital fertility (Bengtsson and Dribe
2014) as well as strong persistence in socioeconomic status across generations (Dribe et al.
2015), which shows that social class measured by occupation is an important dimension of social stratification in the nineteenth and early twentieth centuries.
In the period 1920–1949, Sweden was in the middle of its industrial transition and showed higher rates of economic growth than most other Western countries (Schön
2010:191) The proportion of men employed in industry in the five rural/semi-urban parishes was 28% in 1930 compared with 32% for the entire country (Statistics Sweden
1936: tables 1 and 2). This was also a period when public welfare institutions were greatly expanded, including pensions, housing allowances, and income compensation during sickness and for work injuries. Nevertheless, the degree of compensation was modest (Edebalk
1996). It was not until 1948, when a new pension system was introduced, that retirees could expect to be able to live on their pension (Edebalk and Olsson
2011). Sulfa, introduced on a large scale in the beginning of the 1940s, instantly reduced pneumonia mortality. The drug was very inexpensive, and there is no evidence of class differences in its use or effects (Lazuka
2020). Regarding lifestyles, most evidence indicates a clear social difference in tobacco use in the period up until the 1940s, with the middle and upper classes smoking more, leading to potentially important health consequences (Dribe and Eriksson
2018; Nordlund
2005).
The subsequent period, 1950–1969, saw rapid economic growth and further development of the Swedish welfare state. In this period, as cigarette smoking grew rapidly, the social differences started to disappear. Then, in the subsequent period, when the adverse health effects of smoking became universally appreciated, the middle and upper classes were the first to stop. This gave rise to the now familiar pattern of smoking being highly correlated with a low socioeconomic status (Nordlund
2005). In this period, smoking was still much more prevalent among men than among women.
The early 1970s, again, saw a continuing expansion of the welfare state, covering almost all aspects of childcare to old-age care, and from income compensation to health care. Meanwhile, levels of education increased, and manual work declined in importance, which implied increased upward social mobility (Dribe et al.
2015).
We use data for the five rural/semi-urban parishes from 1813 to 2015 and for Landskrona from 1922 onward. Information is provided from continuous population registers (a household-based register where information at the individual level is continuously updated), with information on demographic events, including migration to and from households for all individuals in the area. Birth and death registers have been used to add events not recorded in the population registers. Another important characteristic of the data is that migration, into and out of the study area, is comprehensively recorded, meaning that the population at risk is well defined.
From 1968, longitudinal individual-level information covering the entire country is available in administrative registers at Statistics Sweden. Data from these registers have been linked to the historical sample, which has allowed an extension of the database along several dimensions. First, individuals who had ever lived in the study area prior to 1968 but lived elsewhere in the country were followed until 2015 or until death or emigration. Additionally, spouses, parents, grandparents, children, and siblings of individuals belonging to the original population were added to the database if they were alive and living in Sweden sometime after 1967. All individuals added to the sample population were similarly followed until 2015, death, or emigration from Sweden.
The main analysis focuses separately on men and women residing in the study area, of ages 30–89 as well as of ages 30–59 and 60–89 separately. Sensitivity analyses are conducted, comparing different measures of occupation and including only currently married individuals in the sample. In addition, for the two periods after 1970, we compare the sample with a sample excluding foreign-born people and with those people who had ever lived in the area and their relatives regardless of where they lived in Sweden.
We measure social class based on the individual’s and his or her spouse’s occupation for the currently married. Occupational notations have been coded in an internationally comparable coding scheme for historical occupations: the Historical International Standard Classification of Occupations (HISCO) (van Leeuwen et al.
2002). These standardized occupations have subsequently been coded into the Historical International Social Class Scheme (HISCLASS), a 12-category occupational classification scheme based on skill level, degree of supervision, whether manual or nonmanual, and whether urban or rural (van Leeuwen and Maas
2011). In the analysis, we use a six-class version of the scheme, which includes the following classes: higher white-collar workers (HISCLASS 1–2), lower white-collar workers (HISCLASS 3–5), medium-skilled workers (HISCLASS 6–7), lower-skilled workers (HISCLASS 9–10), unskilled workers (HISCLASS 11–12), and farmers (HISCLASS 8). In all analyses, we also include individuals without a registered occupation as a separate category (N/A), which is a very heterogeneous group varying greatly over time. Farmers are also a heterogeneous group, which includes both large-scale farmers who had workers employed at their farms and small-scale farmers who worked on other farms to make their living. This heterogeneity is why it is problematic to fit farmers into the class scheme at any time, a problem exacerbated by the data set encompassing such a long period. Furthermore, the group was already very small by the 1950s.
The remaining five classes, which we focus on, broadly reflect a status hierarchy from the lowest status (unskilled workers) to the highest status (higher white-collar workers). The class scheme is frequently used in historical studies of social stratification and is very similar to other commonly used class schemes in the stratification literature, such as the EGP scheme (see Erikson and Goldthorpe
1992).
For married women, their own social class is unlikely to be a valid indicator of their actual social position given that the share of women with gainful employment was very low well into the twentieth century. Consequently, in the main analysis, we use the highest class within the couple to indicate social class. In the sensitivity analysis, we also show results using individual occupation to measure class.