Mental Health and Modern Society: An Age-Period-Cohort Analysis
Name: Jos van Leeuwen Student number: 10190899 Supervisor: Thomas Leopold Second reader: Patrick Brown
Master’s programme: Sociology (General track) Email: fransjevanleeuwen@gmail.com
Publication date: 2018-06-09 Word count: 14685
1 Index Abstract ………... 2 Introduction ……… 2 Theoretical framework ……….. 5 General framework ………... 5
Theories of mental health, mental illness, and society ………. 9
Epidemiology and sociology of depression and anxiety ……….. 16
Method ………. 24
Data and sample ……… 24
Operationalization of variables ………. 27
Method of analysis ……… 34
Results ……….. 36
Discussion ……… 43
2
Abstract
It is often claimed that disorders such as depression and anxiety are on the rise in the past few decades, especially among more recent generations. Such developments have been related to changes in social and economic conditions. However, disagreement exists about whether such developments constitute a real change in the occurrence of mental health issues, or whether they are an artifact of changes in diagnostic practices or cultural norms regarding the reporting of symptoms. Existing studies are often unable to answer these questions, as they do not adequately distinguish between rates of diagnosis and underlying mental health issues, or between changes for the general population and in specific birth year cohorts. This study employs 8 waves of German panel data (SOEP, 2002-2016) to construct three different models describing age and cohort patterns of mental health. On the basis of this data, it is argued that average levels of mental health have not decreased between 2002 and 2016. Furthermore, it appears that mental health has actually increased for more recent birth year cohorts rather than decreased. However, the 1990s cohort constitutes a notable exception to this general pattern. These patterns did not substantially change when differences in economical, medical, and environmental circumstances were taken into account. Though I did not investigate changes in patterns of psychiatric
diagnoses, these findings are in direct contrast to claims of an increase in mental health issues.
Keywords: Mental health, Depression, Modernity, Age, Period, Cohort, APC Analysis, SOEP
Introduction
Mental illness has been present in every society, in every culture, and in every historical period. Two disorders which are especially pervasive are anxiety and depression, which together are often referred to as the “common mental disorders” (cf. Risal 2011; World Health
Organization [WHO] 2017). However, during the past decades, many have worried that the number of people suffering from mental health issues is increasing. More specifically, since at least the 1970s claims have been around that the prevalence of depression and anxiety are increasing (cf. Klerman 1979, Klerman & Weissman 1989, Burke et al. 1991, Compton et al. 2006, Ferrari et al. 2013, Weinberger et al. 2017). Such developments have been characterized as a broad social phenomenon, as the advent of a new ‘age of anxiety’ or ‘age of depression’ which
3
is caused by the conditions of modern life. For instance, Twenge (2006) argues that “modern life is not good for mental health”, while Hidaka (2012) characterizes depression as a “disease of modernity”. In this study, I assess the validity of such claims by investigating a large-scale empirical dataset including measures of mental health from 2002 until 2016.
Though a general view exists that mental health issues are on the rise, others disagree. Some claim that prevalence rates have remained stable (cf. Murphy et al. 2000, van der Does 2009). Others argue that apparent rises in the prevalence of disorders are an artifact of changing definitions of disorders (cf. Horwitz & Wakefield 2006, Dehue 2009) or even merely the reflection of a demographic shift (cf. Ferrari et al. 2013). It is of great significance to ascertain the exact nature of this development and its causes, for mental health issues have high individual and social costs (cf. Insel & Scolnick 2006, Greenberg 2010). They cause personal suffering, impair social, occupational and family functioning, and are in general associated with high degrees of disability and mortality (WHO 2017). Thus, they place a heavy burden on both the individual and society.
However, as Wittchen and Uhmann (2010) argue, there is a lack of epidemiological research about important disorders such as depression, while existing studies suffer from a lack of information about lifespan development, a lack of isolation of temporal factors, variability in definitions and methods, and reliance on cross-section data rather than tracing individuals through time (Wittchen & Uhmann 2010). The use of cross-section data is particularly problematic when investigating shifts in general patterns, as it cannot distinguish between differences due to age and differences due to moment of birth (Glenn 1976, Bell & Jones 2015). Thus, it might appear that mental health improves with age, while it is actually the case that mental health is lower for people born more recently. Because I use a longitudinal dataset which includes the same persons in different years, many of these problems are absent from the present study.
In this study, I investigate two questions. The first is whether population-level patterns of mental health in the population have changed across recent years. The second is whether general levels and life course patterns of mental health have changed for more recent generations or birth year cohorts. In other words, I am interested in the effects of age, period, and cohort on mental health.
4
I investigate these questions on the basis of empirical data tracking the development of mental health patterns through time. The data I use is included in the German Socio-Economic Panel Study (SOEP), a large-scale panel dataset which has been conducted annually since 1984. From 2002 to 2016, the SOEP survey has contained a battery of questions regarding mental health once in every two years. For most participants, data is available for multiple editions of the SOEP. Because this data spans a period of 14 years, age overlaps exist between different birth year cohorts. This means that the SOEP data not only allows tracking general developments in mental health, but also particular shifts in different age groups and cohorts.
The results of the analysis which I conducted contradict much of the existing literature. Average mental health was higher for more recent years, instead of lower. It also increased rather than decreased with age, except for the youngest and the oldest respondents. Finally, mental health was higher for more recent birth year cohorts than for earlier ones at the same age instead of lower, except for the most recent cohort. These patterns did not substantially change when differences in economical, medical, and environmental circumstances were taken into account. Though I did not investigate data on psychiatric diagnoses, these findings are in direct contrast to claims of an increase in mental health issues.
In the following, I first give an outline of the field of social epidemiology to which this study belongs. I also discuss common theories of the relation between mental health and society, as well as substantial claims regarding the development of mental health during the last few decades. Afterwards, I will introduce the SOEP dataset which has been used to investigate the research questions and explain the random effects model which has been used to investigate the data. Finally, I present the findings of this study, and relate these to the existing literature.
5
Theoretical framework General framework
In the following section, I first explain some concepts which are especially important for this study, including mental health and illness, social epidemiology, and disorder prevalence and incidence. Because this study makes use of panel data, I also discuss the distinction between age, period, and cohort effects which can be found in such data. After this, I outline the most
important sociological theories of the relationship between social conditions and mental health. Finally, I give an overview of available data regarding depression and anxiety, focusing on individual life-patterns, social characteristics which are connected to these disorders, and their current and historical presence in society in general. I also relate these findings to the sociological theories which will be discussed.
Mental health
Two concepts which are central to this study are mental health and mental illness. These are often defined in relation to each other: mental health is the absence of mental illness, and mental illness is a disturbance of a normally healthy mind. More comprehensively, the World Health Organization (WHO) defines mental health as “a state of well-being in which the individual realizes his or her own abilities, can cope with the normal stresses of life, can work productively and fruitfully, and is able to make a contribution to his or her community” (WHO 2004), and the American Psychological Association (APA) defines mental disorder as “a
syndrome characterized by clinically significant disturbance in an individual's cognition, emotion regulation, or behavior that reflects a dysfunction in the psychological, biological, or
developmental processes underlying mental functioning” (APA 2013). Besides mental health and illness in general, psychologists distinguish many different disorders which affect one or more specific aspects of psychological functioning. Two disorders which occur at an especially high rate are depression and anxiety. Together with somatoform disorder, formerly called unexplained somatic illness, these are referred to as the “common mental disorders”. They are often contrasted with the “severe mental disorders”, which include schizophrenia, manic depression, and epilepsy, as their symptoms are relatively mild (Risal 2011, WHO 2017). Because of their ubiquity and their resemblance to subclinical mental health issues which are present in the general population, the common mental disorders also receive the most attention from sociologists and other social
6
scientists. Thus, a Google Scholar search for “schizophrenia sociology” gives only about 80.000 results, while a search for “depression sociology” gives 900.000.
Social epidemiology
Mental health problems are normally a topic for psychologists and other mental health professionals. At first sight, it might seem that a study of such phenomena is beyond the scope of sociology, which is concerned with social processes and structures rather than individual states of mind. However, both psychologists and sociologists recognize that social conditions play an important role in explaining changes in the number of people suffering from mental health problems. Indeed, Durkheim, one of the first sociologists, famously related suicide rates to social factors, such as religious background and cultural changes (Durkheim 1897). Simmel, another early sociologist, argued that life in the big city has a negative effect on mental health (Simmel 1903). The study of how such social factors affect patterns of disease and health is referred to as social epidemiology (cf. Honjo 2004, Galea & Link 2014). This field can draw on theories derived from both psychology and sociology. In psychology, the diathesis-stress model and the biopsychosocial model take biological, psychological, as well as social conditions in account (cf. Engel 1977, Shanahan & Hofer 2005, Swearer & Hymel 2015). In sociology, explanations of mental health problems have been formulated by proponents of many different schools, which will be considered below.
Still, some claim that social epidemiology is not a legitimate discipline. For instance, Zielhuis and Kiemeney (2001) argue that epidemiology is a medical science, and thus should be based on biomedical theory alone. Psychology and sociology, however, study the “behaviour (including behaviour related to health and disease) of societies and individuals”, which cannot be explained by reference to biological causes (Zielhuis & Kiemeney 2001: 43). For such reasons, social epidemiology has been called “misguided, unscientific, ideological, or too overreaching” (Kaplan 2004: 14). However, Muntaner (2001) questions the intentions of those who criticize social epidemiology from a biomedical perspective, pointing to the private economic interests behind clinical medicine and pharmaceutical companies. According to him, epidemiologists avoid discussion of the health effects of “poverty, income inequality, discrimination, violence, immigration, anti-union activity and patriarchy” because they do not want to jeopardize their status, funding, and influence (Muntaner 2001: 626). From a more theoretical perspective,
7
Krieger has defended the relevance of social science for epidemiology, arguing that it is not primarily concerned with disease causation, but rather with disease distribution. This means that social inequalities are in fact important for epidemiology to the extent that they are reflected in health outcomes (Krieger 2001: 160).
Incidence and prevalence
In psychiatric epidemiology, two measures are commonly used to assess the presence of mental health issues in the general population: incidence and prevalence. The incidence rate of a disorder is the number of new cases of the disorder in a period of time. For instance, the number of new cases of depression in Germany in 2010, i.e. the incidence rate of depression for this year, was approximately 820.000 (Gerste & Roick 2014). The incidence proportion is the number of new cases as a proportion of the total population. As the population size of Germany in 2010 was 81.7 million, the incidence proportion of depression for this year was 0.011, meaning that 1.1% of the total population suffered from a new case of depression. The point prevalence is the proportion of current (rather than new) cases of a condition at a certain point in time. Most often the twelve-month prevalence is given, which includes all persons who suffered from a disorder during the past year. In 2014, the twelve-month prevalence of depression amounted to 7.7% of the population of Germany (Jacobi et al. 2014). One last measure of importance is lifetime prevalence, which includes not only persons who suffered from a disorder during the last year, but also every person who suffered from this disorder at an earlier point in their lives. This amounted to 11.6% of the population of Germany in 2014 (ibid.). Such distributions of mental health issues in the general population and in relation to certain socio-demographic factors are explored in more detail further on in this study.
Age, period, and cohort effects
This study makes use of a specific kind of data called panel data or longitudinal data. Such data consists of observations of multiple variables obtained over multiple time periods for the same individuals. This means that, in contrast to time-series and cross-sectional data, panel data includes both within-subject and between-subject measures. One advantage of such data is that it offers many different data points, thus increasing degrees of freedom and reducing collinearity among variables. Another advantage is that it allows controlling for unobserved
8
differences between individuals, as long as the sources of such heterogeneity remain constant within individuals across different observations (Hsiao 2007: 4).
As I investigate changes in the general distribution and in the lifetime patterns of mental health issues, it is important to distinguish between three sources of variation across time which can be present in panel data: age, cohort and period effects. Age effects are differences which occur within individuals over time, such as an increase or decrease of someone’s mental health over time. Cohort effects are differences between individuals which were born in different time periods, which are due to the unique experiences related to their particular moment of birth. For instance, young people nowadays may experience more stress, leading to higher chances of becoming depressed for someone born in the 1990s in their twenties than for someone born in the 1950s at the same age. Finally, period effects are differences between time periods which affect every individual independently of their age or their moment of birth. Thus, changes in general welfare or in psychiatric practices may influence the mental health of the whole population in the same way (for more detailed expositions of these effects and their nature, cf. Palmore 1978, Suzuki 2012, Bell & Jones 2015).
As will be shown, it has been argued that changes have occurred in the general prevalence of depression and anxiety, in their prevalence in younger cohorts, and in their age patterns across cohorts. These are respectively period effects, cohort effects, and interactions between cohort and age effects, meaning that all three are relevant for this study. In the methods section, I discuss how I distinguish between these effects in my empirical data, and how I deal with a pernicious issue called the “interpretation problem” which such an analysis involves.
9
Theories of mental health, mental illness and society
In the sociology and social epidemiology of mental illness, a wide variety of theories is available. Indeed, almost every substantial theory of society bears upon the nature of mental health and deviations from it. In an influential handbook, Pilgrim and Rogers distinguish five different perspectives which they relate to five famous social theorists: social causation theory (Durkheim), societal reaction theory (Weber), critical theory (Freud), social constructivism (Foucault), and critical realism (Marx) (Pilgrim & Rogers 2014: 8). In another article they also include human ecology, which is connected to the Chicago school of sociology (Pilgrim & Rogers 2005: 233). Krieger instead distinguishes three main approaches: psychosocial theory, social production theory, and ecosocial theory (Krieger 2001: 667). However, the psychosocial and ecosocial models are not separate theories but rather integrative approaches, of which the former includes psychology and sociology, while the latter includes the biology of mental health as well. Social production theory is identified by Krieger with Marxist theory, which means it can be regarded as a synonym for critical realism in Pilgrim and Rogers’ sense. Still others
distinguish only two basic paradigms, namely social causation theory and social selection theory (cf. Dohrenwend et al. 1992, Hudson 2005). When contrasted in this way, the former is a general approach which includes all specific theories awarding an independent role to social factors, while the latter views social differences as a result rather than a cause of mental health issues.
In the following, I first outline seven important approaches to the sociology of mental health: social causation, social reaction, social production, social construction, critical theory, human ecology, and social selection. After this, I discuss how psychosocial and ecosocial models propose to integrate sociological, psychological and biological explanations. Many of these ideas can explain the same phenomena, and given the limited availability of data, I am not able to empirically distinguish between them, However, they do all point to the importance of
sociological theory for explaining historical and socio-demographic differences in mental health issues. Furthermore, they serve as a theoretical foundation for the concrete expectations regarding the age and cohort patterns in mental health issues which are discussed in the next section. In this sense, they provide an important framework for the approach chosen in this study.
10
Social causation
Social causation theory is a classical sociological approach to mental health which is based on the works of Emile Durkheim. It accepts the reality of psychiatric diagnoses, bracketing out questions about their legitimacy. Instead, it focuses on the role of society in causing mental health issues. It regards social conditions as separately existing phenomena which can have real effects on individuals (Pilgrim & Rogers 2014: 11). Some of its stricter adherents even hold that sociological explanations in general do not require a theory of individual agency, as “social structure may constrain individuals, apart from any consideration of the subjective states of actors” (Sawyer 2005: 114).
Social causation theories of mental illness typically focus on inequalities related to class, race, gender and age, viewing these as disadvantages which cause psychological distress and thus lead to mental problems (Pilgrim & Rogers 2014: 14). For instance, Saraceno, Levav, and Kohn (2005) identify various sources of “social stress”, such as financial hardships and employment insecurity, while Prins and colleagues (2015) explain findings of higher rates of depression and anxiety among those occupying a place in the middle of the economic hierarchy, such as managers and small employers, by arguing that their contradictory class location generate
psychological conflict. Such theories mainly focus on differences in mental health issues between different groups in one society. However, they can also be applied to broad developments in such phenomena by connecting these to changes in social structure or social hierarchy.
Social reaction
The second approach discussed by Pilgrim and Rogers (2014) is social reaction theory or labelling theory, which they trace back to Weber’s interpretive sociology. According to this theory, the influence of society on mental illness consists in the interpretation rather than the causation of mental health issues. People continuously engage in the classification and
simplification of the multifarious and complex reality which surrounds them, and this also occurs in human interactions. This process is called stereotyping and consists in the attribution of fixed group characteristics to individuals (Pilgrim & Rogers 2014: 24).
Social reaction theorists point to the negative effects of stereotyping on the experience of mental health issues. For instance, depression is typically associated with having suicidal
11
certain kind of mental health problems are treated as if they exhibit such behaviors, regardless of their applicability to their particular cases. Furthermore, the process of stereotyping includes emotional and moral reactions, such as anxiety, hostility, and rejection (Pilgrim & Rogers 2014: 24). In this way, stigmatized people become isolated and neglected, which leads to a further decrease in their wellbeing. Thus, Bury has argued that the stigma of mental illness does not only influence the experiences of ‘patients’, but also the severity of their symptoms, as "once labelled, the individual is likely to take on the characteristics of the label, thus confirming the original social response" (Bury 2005: 19). From this point of view, prevalence change can be explained as a result of changes in the interpretation and assessment of mental health issues. If anxiety or sadness are perceived as serious issues or deviations, the chance increases that they are perceived as disorders rather than normal mental states. Conversely, if the stigma associated with such issues decreases, people suffering from them might be more willing to acknowledge them and share their experiences with others.
Social production
A third approach is social production theory (Krieger 2001), which is sometimes also referred to as critical realism (Pilgrim & Rogers 2004). This approach is associated with Marxist points of view. As Krieger (2001) notes, this theory shares with social causation theory a focus on social conditions as causes of psychological distress. Most importantly, it posits a relation between social inequality and health inequality. However, unlike social causation theory, it does not view society as an objective, independent structure, which operates independently of human intentions. Instead, it holds that the goal of economic and political institutions is to produce and maintain social inequality. Thus, inequality is a product of powerful groups in society which want to maintain their privileged positions rather than a sociological ‘given’ (Krieger 2001: 670). In this sense, one can talk about the “political economy” (Burns 2015) or “social production” (Conrad & Kern 1981) of disease and illness. This theory can explain prevalence change in the same way as social causation theory. However, it leaves room for identifying the actors which profit from social and economic developments, as it invokes not only “downstream” effects of social structures on individuals, but also “upstream” effects of political decisions on social structures (Krieger 2001: 670).
12
Critical theory
Like social production theory, critical theory regards political and economic inequality as an important source of mental health issues. However, besides such a focus on external social structures it also considers the way in which these affect the inner mental life of the members of a society. In this sense, it may be regarded as a fusion of Marxist and Freudian ideas (Pilgrim & Rogers 2014: 13). Such an approach was first proposed by members of the Frankfurt School of social research in the 1930s. The main difference between social production theory and critical theory, then, lies in the importance these two approaches award to cultural and psychological phenomena. Whereas the former considers these simply as products of economic and political circumstances, the latter regards psyche and culture as factors which are important in their own regard. Thus, psyche and society are not related in a linear fashion but mutually influence each other.
An influential example of such an application of critical theory to mental health issues is Frankl’s (1946) theory of “existential vacuum”, according to which the loss of traditional values and the breakdown of stable communities has led to a widespread feeling of meaninglessness which constitutes “the mass neurosis of the present time”. Critical theory also includes
approaches which are very critical of mainstream psychiatry, such as antipsychiatry (cf. Szasz 1961, Laing 1967), critical psychiatry (cf. Ingleby 1980), and postpsychiatry (cf. Bracken & Thomas 2001). Such theories argue that mainstream psychiatry is a pseudoscience which does not offer real explanations but is only based on the registration of symptoms. Furthermore, instead of recognizing the psychosocial causes of disorder, it only offers medication for symptoms, while underlying problems remain ignored. This has been related to the special interests existing within psychiatry, such as inappropriate relations between psychiatrists and the pharmaceutical industry (Angell 2011). Such theories may be used to argue that apparent
increases in the occurrence of mental illness only reflect the pathologization or medicalization of normal mental states, or even that the way in which people with mental problems are treated by the medical system only exacerbate their issues.
Social construction
Social constructivism is philosophically the most radical of the approaches which are discussed here. Central to social constructivism is the view that “reality is not fixed, stable,
self-13
evident and waiting to be revealed, but rather a product of human activity and in this sense, constructed by humans” (Page 2014: 4). This means that constructivist theories of mental illness focus on the activities of psychiatrists and mental health professionals rather than objective social structures or subjective internal states. In contrast to the other positions, such theories deny the reality of not only the psychiatric diagnoses, but also of the underlying mental health problems they are supposed to designate. At their most extreme, they regard mental illness as a myth or metaphor which is nothing more than a product of psychiatric activity (Pilgrim & Rogers 2014: 15)
Within this framework, Pilgrim and Rogers (2014) distinguish three different approaches: the social forces approach, the linguistic approach, and the scientific production approach. The first, which resembles social reaction theory, accepts phenomena such as mental illness as given. However, it is concerned with individual experience rather than objective causes. The second, like social production theory, focuses on the reproduction of power structures which perpetuate social inequalities. However, it also investigates how language and symbols are used to create a certain vision of society which facilitates such social reproduction (Pilgrim & Rogers 2014: 15). Thus, Foucault (1961) argues that the eighteenth and nineteenth century witnessed the
transformation of “unreason” into “madness”, which allowed the state to repress deviant behavior by excluding those suffering from mental problems from general society. The third instead focuses on scientific production. Like critical theory, it argues that "science-in-action" is a messy process which involves individual interests and questionable methodological choices (Pilgrim & Rogers 2014: 15), and not an objective procedure for uncovering “natural kinds” which exist independently of their conceptualization by scientists (Zachar 2000).
Human ecology
Human ecology is an approach which derives from the Chicago School of Sociology (Pilgrim & Rogers 2005). Its distinctive feature is that it focuses solely on the “attributes of organized populations”, regarding communities, regions, and societies as really existing macroscopic entities (Schnore 1961: 137). Like social causation theory and social production theory, human ecology views social conditions as objective causes of individual mental health issues. However, unlike these theories it conceptualizes social forces from an ecological perspective. Individuals live together in a society, and their mental health status is affected
14
primarily by their interpersonal relationships, rather than by independent social structures or partisan political decisions. Thus, human ecology argues that inequalities in interpersonal resources are only one property of a population which influences mental health, rather than its primary social determinant (Pilgrim & Rogers 2005: 232). Within such a framework, changes in the prevalence of mental health issues could be explained through reference to changes in the structure of social networks and associated social support.
Social selection
In contrast to the other theories which have been discussed, social selection theory argues that mental health issues affect social status, rather than the other way around. Thus, according to this theory, low income and a marginalized social position are an effect and not a source of mental illness (cf. Dohrenwend et al. 1992, Mossakowski 2014). Examples of such approaches are downward mobility or drift theories and genetic predisposition theories of mental health. The downward mobility or drift theory posits that mental illness inhibits one’s abilities to work or study, and thus leads people to drift into a lower social class or to be unable to escape such a position (Mossakowski 2014). Genetic predisposition theories further posit that the original source of mental illness is the genetic makeup of an individual, thus fully reducing social circumstances to effects rather than causes (Dohrenwend et al. 1992: 946).
An important problem for social selection theory is that it cannot explain real changes in mental health issues, except through positing a change in the makeup of individuals. However, it is very unlikely that individuals nowadays possess different genes or cognitive structures than in the recent past. Dohrenwend and colleagues (1992: 955) provide empirical evidence against this theory by pointing to the relation between ethnic status and mental health. As ethnic status is not dependent upon someone's attainments but determined at birth, it cannot be affected by a
personal predisposition towards mental illness. This means that in this case, the only explanation which is available is that a social characteristic does actually influence a health outcome.
Integrative approaches
Though some of these positions are skeptical of mainstream psychiatric theory, all of them can in principle be coupled to theories of individual mental health. Such an integrative approach is called a psychosocial theory. As such an approach can take many different forms, it
15
does not include much substantial points, except general ideas such as that the boundaries of illness are socially determined (Engel 1977), that psychosocial dimensions cause and shape mental illness (Richter 1999:28), that social circumstances lead to stress and thus cause illness (Krieger 2001), or that illness is affected by multiple levels of organization (Borrell-Carrió, Suchman & Epstein 2004). Some frameworks are even more wide-ranging, also encompassing biological and ecological theories. Examples are biopsychosocial theory (Engel 1977), ecosocial theory (Krieger 2001), and the social-ecological systems perspective (McMichael 1999). As this study is primarily about the social aspect of mental health, integrative approaches are less relevant than the specific sociological theories discussed above. However, it is important to realize that this focus does not imply that mental health is completely a social phenomenon.
16
Epidemiology and sociology of depression and anxiety General prevalence
Depression and anxiety are sometimes referred to as the “common mental disorders”, as they have the highest prevalence of all mental illnesses. As Layard and colleagues (2013)
observe, together these disorders occur at any one time in nearly one in 10 persons on the planet. In 2013, this amounted to no less than 670 million people, of which 400 million suffer from depression and 270 million from anxiety. It was long thought that mental disorders primarily occur in rich countries. However, studies have shown that most persons with a mental disorder actually reside in low- and middle-income countries. Prevalence rates differ but are broadly the same for countries with similar income levels (Layard, Chisholm, Patel & Saxena 2013).
Various twelve-month prevalence rates of depression have been reported for Germany, ranging from 6% (Busch, Hapke & Mensink 2011) to 10.2% (Erhart & Stillfried 2012). Below, I have included a table in which some of these estimates are set next to each other (Table 1). Less data is available on anxiety disorders, but Jacobi and colleagues (2014) report a twelve-month prevalence of 14.5% in Germany. One explanation for differences in reported prevalence rates of depression is that they reflect differences in the ways which the various studies operationalize mental illness. Some measure symptoms, while others ask for previous diagnosis of depression (cf. Maske et al. 2015). Furthermore, some of these studies use different age limits, and
vulnerable groups may be underrepresented because of research designs (cf. Busch et al. 2013).
Twelve-month prevalence estimates for depression in Germany
Author Year Prevalence Source Measure
Jacobi et al. (2004) 1999 8.5% GHS CIDI
Erhart & Stillfried (2012) 2007 10.2% Medical information Diagnosis Busch, Hapke & Mensink (2011) 2009 6.0% GEDA Diagnosis Busch et al. (2013) 2010 6.0% DEGS1 Diagnosis Busch et al. (2013) 2010 8.1% DEGS1 PHQ-9 > 10 Jacobi et al. (2014) 2010 7.7% DEGS1 CIDI Robert Koch-Institut (2014) 2012 8.0% GEDA Diagnosis Source: Authors cited above
17
Age patterns
Besides general prevalence, it is also important to consider distributions of disorders within society. A first factor associated with differences in mental health is age. A study by Kessler and colleagues (2005), which made use of data from the National Comorbidity Survey (NCS), showed that 25% of all mood disorders start before age 18 years and 75% before age 43 years, while the median age of onset was age 30 years. For anxiety disorders, 25% of cases occurred before age 6 and 75% before age 21, the median age of onset being age 11 years. Prevalence appeared to be associated with age as well and was characterized by an increase from 18 to 40 years of age, after which it declined again. Similar results were found by Patten,
Meadows, and Brown (2005), who made use of cross-sectional data included in the 2002 Canadian Community Health Survey (CHS).
However, as the NCS and the CHS are cross-sectional surveys, differences between age groups had to be estimated on the basis of between-subject rather than within-subject data, while information on age-of-onset relied on self-report. As will be discussed in the methods section, such data cannot distinguish between age and cohort effects. Furthermore, it has been questioned whether an apparent decrease in prevalence of depression among the elderly reflects a real decrease in depressive symptoms. Alternative explanations which have been offered include unwillingness to acknowledge mental health issues, or misattribution of depressive symptoms to physical problems (Christensen et al. 1999: 325). It may also be argued that low depression prevalence rates among younger age groups reflect the existence of a substantial amount of untreated cases of depression rather than a lack of mental health issues.
Furthermore, disagreement exists about the actual age patterns of depression. Stordal and colleagues (2001) found that dimensional depression scores and prevalence rates of depression increase continuously with age, without decreasing again after a certain age. Others have instead claimed that depression decreases from adolescence to middle age, after which it increases again until it reaches its highest level at age 80 (Mirowski & Ross 1992; Kessler, Foster, Webster & House 1992). This pattern has been related to changing perceptions of self-efficacy: while people in their 40s regard themselves as competent and in control, the elderly lose their independence and have to give up more and more activities (cf. Seligman 1975; Bandura 1986). Christensen and colleagues (1999), who conducted a cross-sectional study in Australia among respondents
18
ranging from 18 to 79 years, instead argue that age correlates negatively both with depression and anxiety disorders: as age increases, average mental health improves and the prevalence of
disorders decreases. The same pattern was also found by Busch and colleagues (2013), who used data from the 2011 German Health Interview and Examination Survey for Adults [DEGS1], a large-scale cross-sectional survey conducted in Germany. These authors found that in the German general population, the prevalence of current depressive symptoms is highest among young adults, and then decreases. However, like the studies based on the NCS and CHS, they found that rates of diagnosed depression actually increased until the age of 50, after which they decreased again.
Group differences
Besides age, the prevalence of depression and anxiety in the population is also associated with other personal characteristics, including sex, socio-and economic status, and ethnic
affiliation. Though such characteristics are not included in the final model, their effect on mental health is often explained through reference to social factors. Thus, they provide examples of how the sociological theories discussed previously can be applied.
Sex
First, it has long been known that there are significant gender differences in the
prevalence of disorders. Both depression (Karger 2014: 1092) and anxiety disorders (McLean, Asnaani, Litz & Hofmann 2011) occur twice as much among women than among men. For Germany, lifetime prevalence rates of depression have been reported of 10.2% for women and 6.1% for men (Busch et al. 2013), while twelve-month prevalence rates of depression for men and women were respectively 6.1% and 12.4% (ibid.), and of anxiety disorders respectively 9.3% and 21.3% (Jacobi et al. 2014).
Such gender differences have been explained by reference to genetic, hormonal, psychological factors, and psychosocially as a result of gender roles (Kuehner 2003). Thus, Karger (2014) argues that women may experience more stress because they are expected
nowadays both to pursue a career and to be caring mothers and housewives. Alternatively, it may be the case that men suffer as much from depression as women but are less likely to report mental health issues and seek help due to different cultural expectations (Karger 2014: 1095). Some of
19
the differences between men and women may also be mediated by other factors. For instance, women often have a lower socioeconomic status and educational level than men, which may in part explain the higher risk of mental health issues for women (Muntaner, Eaton, Miech & O’Campo 2004).
Socio-economic status
Besides gender and age, socio-economic status and education level are also associated with mental health issues. Thus, it has been shown that the risk of poor mental health increases with the level of economic disadvantage (Everson et al. 2002), that higher levels of depression exist among those with a lower level of education (Miech & Shanahan 2000). Similar findings have been reported for anxiety (cf. Najman et al. 2009). Thus, Fryers and colleagues conclude that both depression and anxiety are associated with unemployment, material disadvantage, and poor education (Fryers, Melzer, Jenkins & Brugha 2005).
Various explanations have been offered for such findings. Some authors, proceeding from social causation theory, argue that the hardships caused by economic disadvantage, such as problems paying bills and phones being turned off, are a significant source of psychological distress (Heflin & Iceland 2009: 1051, Mossakowski 2014). From the point of view of human ecology, it has also been noted that poverty is spatially concentrated in neighborhoods in which social order has broken down, and crime, trouble and vandalism are rampant (Ross 2000: 177). Besides such real effects of poverty, social reaction theory can explain this association as a consequence of lower self-esteem existing among the poor, resulting from a feeling of personal failure in comparison with others who are more affluent (cf. McLeod & Shanahan 1993).
Ethnicity and race
A third factor which is often associated with higher levels of depression is ethnic and racial identity. Dohrenwend and colleagues (1992) mention a study from 1850, in which Jarvis reported high rates of "insanity" among Irish immigrants in Massachusetts. Furthermore, they cite multiple twentieth century examples of high rates of mental health issues among ethnic
minorities, including “blacks and Hispanics in New York, Indians and Pakistanis in London, and North African Jews in Israel” (Dohrenwend et al. 1992: 946). More recently, Riolo and
20
African Americans or Mexican Americans, lifetime prevalence rates being respectively 10.4, 7.5, and 8%. However, for dysthymic disorder, a less serious form of mood disorder, results were reversed: lifetime prevalence for white Americans was 5.7%, for African Americans 7.5% and for Mexican Americans 7.4%. The authors primarily focus on the higher rate of dysphoria among non-whites, pointing to factors such as poverty, lack of education, and cultural differences as possible explanations (Riolo et al. 2005). However, in another study by Dunlop and colleagues (2003), African Americans and Hispanics were actually found to exhibit elevated rates of major depression in comparison to white Americans, though after controlling for confounding factors this pattern reversed again and African Americans exhibited the lowest rates of depression (Dunlop et al. 2003).
Such differences in prevalence rates have also been shown to exist in the case of anxiety disorders. For instance, social anxiety occurs at different rates in people of African American and European American descent (Grant et al. 2005), while Neal and Turner (1991) also found limited evidence for a more general relation between anxiety and race. The latter suggest that such anxiety is often the reflection of really existing racial hostility and stereotyping, while
Dohrenwend and colleagues (1992) point to the "harsh discrimination" faced by ethnic minorities which are assimilated into "the structures of a relatively open-class, urban society" (Dohrenwend et al. 1992: 946).
Interactions
Besides these general effects of demographic and socio-economic factors on mental health, there are also interactions between them. Thus, Jorm (1987) has shown that differences in mental health issues between men and women are age-specific. In childhood, there is little
difference, but prevalence rates diverge in middle life. When people reach old age, this difference again becomes smaller. The association between depression and educational level is also age-dependent and was specifically found to increase with age (Miech & Shanahan 2000). For anxiety disorders, interactions were also found between race on the one hand, and sex and age and the other, but not between age and sex (Ortega & Myles 1987). In both cases, differences in mental health issues were larger for blacks than for whites. Race and sex also interact in the case of depressive symptoms: black men experienced more symptoms than white men, but the
21
differential access to mental health services, arguing that more barriers to receiving help exist for black men than for the other categories.
Historical changes
Until now, I have discussed patterns in the distribution of depression and anxiety as if they are stable and historically unchanging. However, as has been mentioned, it has been argued that historical changes have occurred in the general prevalence and life-course patterns of mental disorders. Already in the 70s, Klerman argued that Western societies were entering an "age of melancholy" (Klerman 1979), based on observations of increasing prevalence of clinical depression among adolescents and young adults. Since then, many have claimed that the prevalence of mental health issues is increasing. For instance, Fombonne (1994) argues for an increase over time of depressive phenomena, connecting findings from a broad range of statistics including prospective studies, community surveys, repeated cross‐sectional surveys, admission data and suicide statistics. Bandelow (2005) discusses similar views that “each year more and more people are suffering from anxiety disorders". More recently, Weinberger and colleagues (2017) have found that in the United States, the prevalence of depression has increased between 2005 to 2015 from 6.6% to 7.2%. Wittchen and Uhmann (2010) also found a difference in the cumulative risk of suffering from a major depressive episode and the age of onset between birth cohorts. Among more recent birth cohorts, the age of first onset is decreasing, while the
prevalence of depressive episodes is increasing.
The most important question of the present study is whether the prevalence of mental health issues, and most importantly depression and anxiety, has really changed during the last decades. In the scholarly debate about historical changes in the prevalence of disorders, four broad positions can be discerned. Each of these has been argued for from a variety of different perspectives, of which I provide an overview in the following section. These positions are:
1. Realism: Change in prevalence because of an actual change in mental states
2. Diagnostic artifact: Change in prevalence because of change in diagnostic practices 3. Demographic artifact: Change in prevalence because of change in demographics 4. Stability: No change in prevalence
22 Realism
Theoretical explanations for such an increase in reported symptoms of depression and anxiety have been drawn from a variety of sources. For instance, Klerman and Weissman (1989) point to increased urbanization, geographic mobility and associated loss of attachment, social anomie, changes in family structure, and changes in the gender-roles of women. All these social changes are typical for modern Western society. Baxter and colleagues argue that such
phenomena have led to an increase in psychosocial stressors, which in their turn cause mental health issues (Baxter et al. 2014). Thus, Twenge (2006) asserts that “modern life is not good for mental health”, pointing to such factors as loneliness and isolation (ibid.: 149), job market
insecurity (ibid.: 157), increasing income inequality (ibid.: 160), unrealistically high expectations of career success and life experiences (ibid.: 171), and increased media reporting of dangers and threats such as terroristic attacks (ibid.: 178). Hidaka (2012) argues that proof for a relation between modern social conditions and anxiety and depression rates is constituted by the
correlation between a country’s GDP per capita as a measure of modernity, and lifetime risk of mood disorders. More specific characteristics of modern society which contribute to this relation are a decline in physical well-being, and a “toxic social environment”, including once again increasing competition, inequality, and social isolation as important depressogenic factors (Hidaka 2012). Taking a lead from Hidaka, such assessments of modern social conditions could be referred to as “toxic modernity” theory.
Artifact
Others argue that changes in the prevalence of depression and anxiety are not real phenomena which are associated with social changes, but rather an artifact of some other development. According to Horwitz and Wakefield (2006), the current epidemic of mood disorders is nothing more than a “fiction”, an artifact of the way in which psychiatry has turned normal sadness into pathology. Another reason for such inflation of perceived disorder rates is the inclusion in general population studies of standard psychiatric symptom checklist, as these are meant for assigning a specific diagnosis to someone who already has mental health issues, and not to assess the occurrence of mental illness in the general population. Such a use of standard questions about symptoms without providing information about personal context makes it impossible to distinguish between “the normal distress experienced in life” and “genuinely
23
pathological conditions that indicate an underlying mental illness” (Horwitz & Wakefield 2006.: 19). Mulder (2008) lists a variety of other ways in which psychological issues have been
pathologized or medicalized, such as confusing illness with sharing symptoms, surveying
symptoms out of context, the benefits of diagnosis to the pharmacological industry; and changing social constructions around sadness and distress.
From a slightly different angle, Dehue (2009) has argued that the current "depression epidemic" reflects a cultural shift in the definition of depression from melancholy to lack of motivation. Instead of locating its source in real shifts in social conditions or economic
conditions, Dehue argues that the apparent rise of depression rates is a result of this new, more inclusive definition of depression, which applies to many more people than the old definition.
Another approach, which is even more deflationary, is that though there is a real increase in depression prevalence, this is mainly due to a shift in the composition of the population. For instance, Ferrari et al. (2013) argue that though the total number of people with depression has increased, this was mainly due to population growth and increasing average age. A similar view is shared by other researchers concerned with the consequences of an ageing population for the health system (cf. Snowdon 2002; Cong, Dou, Chen & Cai 2015; Padayachey, Ramlall & Chipps 2016). Such studies locate changes in prevalence not in changes in social structures or cultural norms, but solely in demographic factors.
Stability
Finally, some argue that no change has occurred in mental illness patterns at all. Murphy and colleagues (2000) have shown that over a 40-year period, prevalence rates of depression in Canada have remained stable, and that historical change occurs rather in sex and age patterns rather than in the population as a whole. Similarly, Hawthorne, Goldney, and Taylor (2008) show that prevalence rates have not increased in Australia, and similar findings have been reported for the Netherlands (cf. van der Does 2009, de Graaf, ten Have, van Gool & van Dorsselaer 2012). Baxter and colleagues explain such a lack of increasing prevalence by pointing out that though some risk factors for mental illness have increased, these have been offset by decreases in others. Thus, though inequality may have risen and social cohesion has decreased, living standards and education levels have improved (Baxter et al. 2014). In contrast to “toxic modernity” theory, this positive assessment of modern society could be termed “positive modernity” theory
24
Method Data and sample
In this study, I investigate changes in the prevalence and life course patterns of mental health issues such as depression and anxiety. To this end, I use the German Socio-Economic Panel Study (SOEP) dataset, a large-scale longitudinal panel data study conducted in Germany every year since 1984. The SOEP questionnaires include a large range of questions about individual life experiences, social position, and economic participation, ranging from religion to employment, and from family structure to life satisfaction. The SOEP survey also contains a measure of mental health called the Mental Health Component Scale or MCS. This measure constitutes the main dependent variable of this study and is described more into detail below.
The SOEP has been conducted every year since 1984 until today. The latest edition which has been made available for research was conducted in 2016. From 2002 onwards, the SOEP has included data on mental health once in every two years. This means that relevant data is available for eight different SOEP editions. Each of these editions included between 20,000 and 30,000 respondents. The SOEP researchers aim to include the same respondents in every edition but respondents may still drop out for various reasons. To keep the size of the research population stable, new refreshment samples are included once every few years.
The SOEP dataset is particularly well-suited for answering the main questions of this study. First of all, it includes multiple observations for the same individuals across different years. Thus, using the SOEP it is not only possible to compare average rates of mental health between years, but also to model patterns of change within individuals across their lifespan. This greatly enhances the power of the statistical analyses I conduct, compared to studies based on surveys including different respondents for each edition. Furthermore, because the measurement of interest was observed over a period of 14 years, there are age overlaps between different cohorts. For instance, participants from the 1980s cohort have been measured when they were 20 in 2002, while participants from the 1990s cohort were measured when they were 20 in 2012. Such overlaps facilitate disentangling the effects of age and cohort, both of which are relevant for this study.
The total number of unique respondents included in one or more of the editions of the SOEP relevant for this study was 62,144 respondents. This was the starting sample for this study. Of these respondents, the number who completed the mental health questionnaire included in the
25
SOEP at least one time was 56,974. All these respondents can in principle be included in the analysis. However, due to its sampling design the SOEP dataset contains population structures which are not included in the model but which may distort findings and affect their
interpretability. Thus, I have chosen to create a more homogeneous research population by imposing several restrictions on the dataset.
First, I have excluded respondents with an immigration background, so as to control for cultural and social-structural differences between immigrants and native Germans. For the same reason, I have excluded respondents who lived in East Germany or abroad in 1989 as well. Thirdly, I have included only respondents who were part of the basic and refreshment samples contained in the SOEP, while all respondents belonging to other samples are excluded. Finally, all respondents were dropped who were younger than 18 or older than 79 during their
participation in the survey, in order to facilitate comparison of results with other epidemiological studies of mental health such as the DEGS1 and the NCS. After these sample cuts, the final sample which remains for analysis contains 20,031 respondents, which is 32.2% of the starting sample. More detailed information about the effects of these successive sample cuts on the sample size can be found in the table below (Table 2).
The final sample consists of 20,031 respondents, of which 9,736 are males and are 10,295 females. They were born between 1923 and 1998, and range in age from 18 to 79 years old. The mean age of a subject during an observation was 49.12 years. The sample includes 75,677 separate observations of their mental health. The minimum number of observations per person is 1, and the maximum 8. The mean number of observations per person is 3.8, which means that for most participants no data is available for every survey year. One reason is that some participants only started participating in the SOEP during later editions, because they were too young to participate before or because they were included in a refreshment sample. Another reason is that some participants dropped out. Of the 20,031 subjects who were included in the SOEP survey during the research period, a total of 7,745, or 38.6% of the total sample, still participated in the 2016 survey (Table 3).
26
Sample constraints and sample sizes
Sample constraints Total N Total N % All respondents 2002-2016 62,144 100 MCS measured at least once 56,974 91.7 No migration background 39,558 63.7 Not in East Germany or abroad in 1989 29,265 47.1 Included in basic and refreshment samples 20,819 33.5
Aged 18-79 20,031 32.2
Source: German Socio-Economic Panel Study (SOEP), 1984-2016, version 33
Table 2. Sample sizes after each successive sample cut.
Sample development and attrition
Year N (New) N (2016) % (2016) 2002 10,646 3,128 29.4% 2004 995 255 25.6% 2006 2,218 660 29.7% 2008 612 215 35.1% 2010 438 182 41.5% 2012 4,062 2,450 60.3% 2014 636 442 69.4% 2016 422 422 100.0% Total 20,031 7,745 38.6% Source: German Socio-Economic Panel Study (SOEP), 1984-2016, version 33
27
Operationalization of variables Age and cohort
The specific topic I investigate is developments in the prevalence and age patterns of mental health. This requires only age, birth cohort, and period as independent variables, while other differences between participants are irrelevant. Information on the age of subjects is directly available. Birth year cohorts, each spanning a decade, were created on the basis of
information on the birth year. Because of the sample cuts, the first birth year cohort present in the final sample is the 1920s cohort, which contains subjects born between 1923 and 1929. The final birth year cohort is the 1990s cohort, which contains subjects born between 1990 and 1998.
For cohorts which include observations for every edition of the SOEP used in this study, this amounts to eight separate periods of measurement, spanning a period of 14 years. However, as some subjects were born in the first years included in the cohorts, and others in the final years, some cohorts include measures for more than 14 years. For instance, for someone who is born in 1959, and measured from 2002 to 2016, information is available from the ages of 43 until 57, while for someone born in 1950 and measured during the same period, information is available from the ages of 52 until 66. Thus, the 1940s, 1950s, 1960s, and 1970s cohorts have a total observed age span of 24 years, the 1980s cohort has an observed age-span of 19 years, and the 1930s cohort an observed age-span of 17 years. However, the 1990s cohort includes only five distinct measures and the 1920s cohort only four, which means that these cohorts have been measured over a period of respectively eight and six years. These cohorts have observed age-spans of respectively 9 years and 7 years.
Period
Information on period is directly available in the form of the year in which a particular observation was recorded, However, it is impossible to include period terms next to age and cohort terms in one statistical model. The reason is that these three effects are defined in terms of each other: period (survey year) minus age is cohort (birth year), period minus cohort is age, and age plus cohort is period. For instance, if a participant in a study is aged 25, and his birth year is 1980, it follows that the period is 2005. This means that it is impossible for one of these three factors to vary independently of the others, which makes them perfectly collinear. Consequently, in cross-sectional data, it is impossible to know whether a difference between subjects born in
28
different years, and measured at the same time, is caused by their age or by their moment of birth. Similarly, in time-series data, it is impossible to know whether a difference between subjects born in the same year, and measured at different moments, is caused by their cohort or by the period. This issue is referred to as the “identification problem” (cf. Blalock 1966, Glenn 1976, Palmore 1978). The table below sets out the different forms this problem takes, depending on which kind of data is available (Table 4).
The identification problem in different types of data
Data type Data structure Identification problem Age Cohort Period
Cross-sectional Different Different Same Age or cohort? Time-series / Intra-cohort Different Same Different Age or period? Intra-age Same Different Different Cohort or period? Panel Different Different Different Age, period, or cohort? Source: Own creation
Table 4. The identification problem in different types of data.
Though many solutions have been offered for the identification problem, it has not yet been solved. One common approach is the “constraints approach”, which consists in dropping out one of the three factors which can be supposed to have no effect on the outcome. Another is the “proxy variables approach”, which involves the use of one or more control variables to capture the source of one of the effects (Browning, Crawford & Knoef 2012). Such approaches must be justified on empirical or theoretical grounds, for if the dropped term actually does have an effect on the outcome, or if the control variables do not adequately capture it, the other two terms will be biased. Empirically, one may first check separately for age, period, and cohort effects, to see if one or more of them are operative. If it appears that a dimension is not related to the outcome, it can be dropped from the model, and the collinearity problem disappears. However, if all three dimensions appear to be important, one can be replaced by proxy variables to avoid collinearity, while still accounting for its effect (Yang & Land 2013). Theoretically, one can base such
decisions on earlier findings and theories and use these to decide whether it is to be expected that one of the dimensions has an effect on the outcome.
29
As was shown in the theoretical and empirical discussion, it has been argued that certain society-level developments have occurred which have influenced the mental health of the population as a whole. These include changes in unemployment rates, income, healthcare, cultural norms, living conditions, and so on. Furthermore, an explorative analysis showed that if the different effects are tested separately, they are all significant (Table 5). To take this into account, I have opted for the “proxy variables approach”. Because of the lack of data on many of these developments, I have focused on five broad phenomena for which reliable data exists: unemployment rates, household income, health expenditure, obesity rates, and urbanization levels. Using these proxy variables, I attempt to account for economic circumstances, quality of the healthcare system, physical wellbeing, and physical environment.
The inclusion of inequality rates, unemployment rates, and average household income is meant to capture the effects of economic circumstances on mental health. Inequality rates are provided by the OECD, an international organization for economic cooperation and development (OECD 2018). Annual unemployment rates for Germany are available from Eurostat, the
statistical bureau of the European Union (Eurostat 2018). Information about household income is directly available in the SOEP dataset, and on the basis of this data I computed average
household incomes for every year. Information about health expenditure as a share of the total GDP of Germany is available through the Federal Health Monitoring System (FHMS 2018). This service is provided by Destatis, the federal statistical office of Germany. Destatis also provides information about urbanization rates for each year (Destatis 2018). Finally, obesity rates for Germany are provided by the World Health Organization (WHO 2018). Originally, I intended to include information on changes in stigmatization, stereotyping, and other cultural norms and practices regarding mental health issues as well. Unfortunately, no adequate data was available to capture such phenomena.
30 z p * Model 1 Age 18.21 <0.001 *** Model 2 Period (year) 8.92 <0.001 *** Model 3 Cohort 1930 3.23 0.001 *** 1940 4.8 <0.001 *** 1950 -1.23 0.220 1960 -3.84 <0.001 *** 1970 -4.41 <0.001 *** 1980 -2.18 0.029 ** 1990 -3.09 0.002 ***
Table 5. Exploration of single age, period, and cohort effects.
Mental health
Since 2002, the SOEP contains a measure of mental health called the Mental Health Component Score (MCS). This measure is based on a set of 12 questions on health-related
quality of life which are included in the survey. These questions are closely modeled on version 2 of the 12-Item Short Form Health Survey (SF-12v2), a short survey which provides a general indication of mental and physical health (Andersen 2007). For instance, respondents are asked how often during the last four weeks they "felt run-down and melancholy", or that due to mental health they were "limited socially, i.e. in contact with friends, acquaintances or relatives". Answers can be given on a scale ranging from 1 to 5 indicating the severity of the mental and physical health problems faced by the respondent. These questions can be grouped into two broad
31
dimensions, the Mental Health Component (MHC) and the Physical Health Component (PHC), of which the former corresponds to the MCS score included in the SOEP. Each of these
dimensions includes four more specific subscales focusing on a specific aspect of mental or physical health, such as bodily pain, vitality or social functioning.
The answers to the specific questions can be recoded on a scale from 0 to 100, and then standardized to obtain summary scale scores with a mean 50 and a standard deviation of 10. On this recoded scale, a high score indicates good mental or physical health, and a low score indicates poor mental or physical health. Such norm-based scores facilitate the comparison of measures across different groups and different years. Though these summary scales were originally based only on items belonging to the specific dimension, the more advanced bidimensional response process (BRP) model uses items belonging to both dimensions to
compute the summary scale scores (Forero, Vilagut, Adroher & Alonso 2013). As Andersen and colleagues (2007) describe, the SOEP researchers followed the same approach when computing MCS scores. These researchers used the 2004 edition sample as a baseline to determine norm-based scores, after which they applied the same scoring norms to all other editions as well. The goal of this procedure was to obtain scores which can be compared across different editions of the SF-12 to investigate developments in mental health (Andersen, Mühlbacher, Nübling, Schupp & Wagner 2007).
Besides serving as an indication of general mental health, the SF-12 has also been used as a screening measure indicating the prevalence of depression and anxiety in the general
population. Normally, clinical instruments include at least 10 questions, and they focus
specifically on one disorder. In contrast, the SF-12 consists of only 12 very broad questions about general mental and physical health. Thus, it might seem that it is not particularly well suited to indicate the presence of a specific disorder. Still, the use of such brief questionnaires is not uncommon in clinical psychology. For instance, the PHQ-2, a popular screening instrument for depression, includes only 2 items (Löwe, Kroenke & Gräfe 2005).
Furthermore, it has been shown a high correlation exists between the items included in the MHC and traditional clinical instruments used for the diagnosis of depression and anxiety
disorders (Vilagut et al. 2013, Vera-Villaroel et al. 2014). Thus, it has been argued that it can be used as a valid and reliable instrument to measure the prevalence of depression and anxiety disorders in the general population. For the standardized scores with a mean of 50 and a standard
32
deviation of 10, cutoff rates of 45.6 and 40.2 have been proposed as indicators of a depressive disorder, depending on whether one uses respectively the traditional scoring algorithm or the more advanced BRP algorithm (Vilagut et al. 2013). In a Chinese sample with a mean score of 54.3, optimal cutoff rates instead ranged from 48.1 to 50.2 for various age cohorts (Yu, Yan & Chow 2015).
The reported cut-off rates are based on the extent to which SF-12 scores accord with a separate diagnosis of depression. While such information is included in the SOEP, I have chosen not to focus on prevalence scores, but rather on mental health in general. One reason is that more data is available about mental health than about diagnosed depression: the former was included in the SOEP from 2002 until 2016, while information about the latter is only present between 2009 and 2015. It was also not possible to integrate these two sources of data, as questions about diagnosed depression were only included in years in which the questions on which the MCS is based was not administered. A second reason is that the I am not only interested in depression, but also in other common mental health issues, such as anxiety. Finally, changes in diagnosis may be artifacts of changes in psychiatric practices, while I am primarily interested in real changes in mental health. Because of these reasons, I have opted to look instead at mean
summary scores of the MCS, regarding deviations from this score as an indication of changes in mental health. However, after lagging information about diagnosed depression for one year, a comparison between respondents with a diagnosed depression and those without a diagnosis showed that the mean MCS scores of these groups strongly diverge: mean scores for those without a diagnosis were 51.4, while mean scores for those with a diagnosis were 40.99. This means that it can be assumed that MCS scores are related to depression in the SOEP dataset. In the table presented below, mean MCS scores are presented for different age groups, birth year cohorts, and survey periods (Table 6).
