Religious service attendance rates throughout the 20th century : a longitudinal life course and cohort study in West Germany

(1)

1

Robbert Molenaar (rob_bert@hotmail.nl) 10772243

Master thesis, (Sociology, general track) Supervisor: dr. T. Leopold

Second reader: dr. T. Bol 9-7-2018

Religious Service Attendance Rates Throughout the 20

th

Century: A Longitudinal Life Course and Cohort Study in West

Germany

Robbert Molenaar University of Amsterdam

(2)

2

Abstract

Religious service attendance rates have been declining for decades in Western society. Interestingly, scientific research on this topic has rarely been executed with both the proper method and data. Utilising longitudinal panel data is the only way in which one can separate age effects from cohort, or social, effects. By using the thirty-third version of the ‘German Socio-Economic Panel’ study (SOEP), which spans from 1984 up through 2016, one can avoid the misinterpretations that come along when relying on cross-sectional data or life course interviews. Through the use of a random effects generalised least squares model one can conclude that aging has a minimal effect on someone their religious service attendance rate. This model also proves that the religious service attendance rate has declined in a mostly linear way across birth cohorts that were born in the 20th century, apart from a small increase in religious service attendance that can be traced back to a brief period after World War II.

(3)

3

1. Introduction

Religion and Western society have been inseparable for a long time. The majority of the European population were religious and frequent attendees of religious services. Somewhere along the beginning of the twentieth century this suddenly changed and religiosity and religious service attendance went down (Bruce, 2016). This apparent downfall of religion has been discussed and studied since the early days of sociology; classic authors like Comte, Durkheim, Marx and Weber were all in on it and for good reason. Religion can still be linked to most pieces of art and music that are considered classics, core values of contemporary democratic society have their roots in Christianity and most if not every municipality in the West has its own church.

The downfall of something that used to be everywhere and affected everyone must have implications on society, which perhaps makes studying religion more interesting than ever. From a scientific point one can also claim that the time to study religion and its downfall is now; longitudinal panel studies are starting to amass several decades of data that can be used to identify individual change versus social change. Research on the decline of religiousness and religious service attendance was and is mostly done by using either cross-sectional data or life course interviews. In 1974 Wingrove & Alston already critiqued the use of these methods when studying variables that are subject to social change. They critiqued earlier cross-sectional research as it assumes that individuals are comparable at a similar age regardless of the generation they belong to (Wingrove & Alston, 1974, p. 324). Cross-sectional research neglects differences in generations; because of this it is not the right analytical tool to study something like change in religious service attendance, which is known to be subject to social change (Ibid.). The use of life course interviews is debatable as well, as humans are not known for their superior recollection and are prone to idealise their past, even more so if this past refers to their religious life (Ibid. p. 325).

Wingrove & Alston (1974) pleaded for the need of a longitudinal panel study that represents a national context, but conclude that this would be a time consuming, costly and thus unlikely reality (p. 325). Now, roughly forty years later, we do have access to these longitudinal panel studies. The ‘German Socio-Economic Panel Study’ (SOEP) is one of them, and it allows one to test change in religious service attendance without the problems that one encounters while using cross-sectional data or life course interviews. Another perk of this study is its focus on German society, which allows one to add a European point of view to a research field that is dominated by studies that focus on the United States. The panel study

(5)

5 holds data that roughly measures if and, if yes, how often respondents attend religious services. This also opens up a way in which one can identify if, and if yes how, the attendance of religious services changes across birth cohorts and during the life course. One becomes able to study whether people are actively leaving religious services during their adult life, or whether people start out less religious than their older peers. Recent studies which are based on longitudinal data and longitudinal panel data conclude that religious attendance rates, in Europe, are stabilising after a century of decline (Kaufmann et al, 2012; Burkimsher, 2014). Germany is not represented in these studies, as its reunification is said to complicate the analysis.

Studying age effects and social or generational effects simultaneously is a necessity if one wants to unravel the separate effects that ageing and social or generational change have on religious service attendance or any other relevant variable. Age and generation are inherently connected, as both concepts are derived from an individual their year of birth. Researching an age effect without integrating generation in the model through a variable for birth cohorts will leave one with a section of the population. As said before, a cross-sectional research design is unable to control for social or generational change. This might lead one to the false conclusion that an effect is connected to personal change, while in reality it is partly or solely dependent on a social or generational effect. Hypothetically, a similar situation might occur when research is only based on a social or generational effect. Here, a social or generational effect might be connected to social or generational change, while in reality it is partly or solely dependent on a more personal age effect.

The ‘German Socio-Economic Panel Study’ is unbeaten for its consistency and Germany has a similar religious heritage to other Western Europe countries. It is thus worthwhile and necessary to investigate how the religious service attendance rate changes over someone their life course, while simultaneously testing how it changed over generations. The purpose of this thesis is to formulate an answer to the following research questions: ‘How did the religious service attendance rates of West German individuals change during the life course?’ and ‘How did the West German religious service attendance rates change across birth cohorts in the 20th century?’.

The paper is organised as follows. First, the relevant theory surrounding the effect of aging on religious service attendance is used to formulate a hypothesis. What follows is an assessment of various secularisation theories, sociological papers and historical papers to formulate a hypothesis concerning the effect that social change, through birth cohorts, has on the attendance of religious services. After this theoretical section the dataset and analytical

(6)

6 approach will be discussed. Finally, the results will be presented and discussed along the lines of the literature which will lead up to answering the research questions in the conclusion.

(7)

7

2. Theory

Religion, as a part of individual life, can be seen divided into five dimensions (Pearce et al., 2017). Attending religious services is a subcategory of ‘external practice’; this dimension can be seen as the social dimension of religion as it contains membership of a religious organisation and the social activities that someone their religion offers (Ibid., p. 370). The remaining four dimensions of religion are: ‘religious beliefs’ which is the way in which the individual relates themselves to a standard set of beliefs, ‘religious exclusivity’ which illustrates the magnitude in which one experiences their religion as the definite truth, personal practice should be seen as the way in which an individual makes religion one their ‘own’ and ‘religious salience’ is the importance of religion to life and identity (Ibid., pp. 369-371). External practice, and thus the attendance of religious services, is the only dimension that is mainly social, others are of a more personal nature with either a cultural or psychological notion.

2.1 The relation between age and religious service attendance

Theory within the sociology of religion suggests that one can distinguish four models concerning the relationship between age and the attendance of religious services. These four models are the ‘traditional’, ‘disengagement’, ‘family-cycle’ and the ‘stability model’ (Bahr, 1970).

The ‘traditional model’ argues that there is a strong decline in religiousness and the attendance of religious services between the ages of 18 and 35, this gradually increases again after the age of 35. Within this model the formation of a career and family are seen as strong points of disenchantment with religion (Bahr, 1970; Chaves, 1987). Some cross-sectional research supports this model. Uecker et al. (2007) state that; leaving the parental home will lower the attendance of religious services as one is able to decide for themselves what one likes to do in their life (p. 1668). The changes in habits that are the consequence of moving out of the parental home do generally not correspondent with religious life (Ibid., p. 1684). Uecker et al. see an increase in (excessive) alcohol consumption, use of drugs, non-marital sex and cohabitation before marriage as reasons for young adults to stop attending religious services (Ibid., pp. 1670-1685). Roozen (1980) reinforces this idea by arguing that going to

(8)

8 church has little to offer and is seen as irrelevant by younger individuals. However Roozen states that all this happens earlier, in the teenage years, and that most, American, ‘church-dropouts’ are back in church before they turn 35 (Ibid.).

The ‘family-cycle model’ states that whether someone attends religious services depends on their stage in life; introducing your own children to religion enhances the attendance rate while the attendance rates before and after this moment are considerably lower (Bahr, 1970). This can be seen as a counter model to the traditional model, arguing that going through the formation of a family is in fact a re-enchantment with religion and that this does not change until the children leave their parents’ home (Chaves 1987). In their earlier discussed research, Uecker et al. (2007) conclude that marriage enhances the religious service attendance and general religiousness, but they themselves state that this might only prove that religious individuals marry more often than nonreligious individuals which leaves any evidence for the traditional model rather scarce (p. 1684).

Within the ‘disengagement model’ one considers that the older someone is the less someone is engaged with society. Following this logic people who are over the age of fifty will attend religious services less frequently because they are losing touch with their religious organisation and society in general (Bahr, 1970). Thus, within this model older people seem to participate less often in social interactions, simultaneously their social circle is also inviting them less often to participate in these social interactions (Cumming et al, 1960). It is often seen as a more self-centred way of life (Ibid.). One should distinguish two sides of this disengagement theory; it seems to exist out of a social and a normative aspect (Hochschild, 1975). The social aspect refers to social interactions related to work, family and leisure while the normative aspect consists of the meaning that one gives to these social interactions (Ibid.). While Ainlay et al. (1992) do find some support for the disengagement model, as their life course interviews suggest that older individuals attend church less often, they are not convinced of the theory itself as their findings suggest that older individuals go to church less often because it gets harder for them to move around town due to the illnesses that come with old age and not due to social isolation (p. 184).

Finally, the ‘stability model’ states that aging and the attendance of religious services are not related; the attendance of religious services seems to be established at a young age (Bahr, 1970). Any changes that seem to correlate with age are in fact generational changes influenced by social or cultural changes; there is no individual change in the religious service attendance rate. As the paragraph below will demonstrate, most findings will verify this stability model.

(9)

9 However convincing Bahr his theoretical models seem to be, most studies have yet to find empirical evidence to support anything but the stability model. Wingrove & Alston (1974) were an early example of those who debunked most of Bahr his theoretical models and found that different birth cohorts, of individuals born between 1885 and 1934 in the United States, correlate with different theoretical models but in the end none of the models could give a satisfying explanation to why people decide to stop attending religious services. However, Wingrove & Alston were unable to test Bahr his theoretical models with longitudinal panel data. More recent and relevant research, that does use longitudinal data, only seems to fit into Bahr his stability model: Schwadel (2010; 2011) states that people of older cohorts, born before 1940, in the United States have a slightly higher chance to attend religious services and that this might give the false assumption that older individuals attend religious services more often. Schwadel could not test this assumption as the datasets that he used were not panel studies. Elder & George (2016) state that: “none of the studies of cohort differences in religious service attendance reported significant age effects” (p. 68). Elder and George refer to Schwadel his longitudinal studies and to a panel data study by Crockett & Voas (2006), who found that there were no age effects between 1983 and 2002 on the attendance of religious services in Britain. Neither could they find any significant age effects earlier in the 20th century (Ibid.).

Concluding the theoretical frame of the effect that aging has on religious service attendance, one can state that only some cross-sectional research claims to find such a thing. Most if not all longitudinal studies argue that every apparent effect that age has the religious service attendance rate is under the influence of some sort of a cohort or generational effect. Even cross-sectional research only seems to partially support Bahr his other theoretical models. This leads to the hypothesis that the attendance of religious services is not influenced by an age effect. Nonetheless if this proves to be true, differences in the absence of an age effect should be tested across birth cohort.

2.2 Secularisation among different birth cohorts in the 20th_century

Whichever sociological theory on the decline of traditional religiousness, or ‘secularisation’, and thus attendance of religious services one consults, they all lead to the conclusion that every younger generation spends less time on religious practices than their predecessors. However, it is important to consider that there is no such thing as an overarching ‘secularisation theory’. The concept ‘secularisation’ is mostly used to describe a

(10)

10 process, that is mainly relevant to the West, whereby modernisation processes influence religion and religiousness in a negative way (Pickel, 2011). These theories do differ in their explanation of which process is exactly influencing which part of religious life (Ibid.). Thus, secularisation is often seen as a process, alongside industrialisation and urbanisation, that is part of a transition from a pre-modern to a modern society (Elder & George, 2016).

With the help of classical sociological theory one can distinguish two theoretical perspectives of secularisation; the ‘rational’ and the ‘functional’ (Norris & Inglehart, 2004). The ‘rational perspective’ has its roots in Max Weber his work on the capitalist spirit of Protestantism; the base assumption of this perspective is that the Enlightenment and the industrialisation of society forced a rational stance based on empirical proof, science and technology upon everyday life (Ibid., p. 7). This newfound rational way of thinking is believed to be the cause of religion its loss of credibility (Ibid.). When one follows the path of rational thinking it is hard to believe in something that cannot be proved through science. Literal biblical teachings became an unlikely reality because a god who purposely created ‘everything’ seemed empirically impossible. However, the emerging ‘religion of rationality’ was only the beginning, as secularisation did not really ‘kick off’ until the arrival of mass-education. Within the rational perspective, mass-education is seen as the social force that pushed Western society towards cultural change and thus towards widespread secularisation through the rational teachings of education (Ibid. p. 8).

The ‘functional perspective’ perceives religion as something that is more than a mere set of beliefs and values; it is also a system of actions, rituals and symbols that structure everyday life. This way of thinking can be traced back to Durkheim his work on the forms of religious life (Ibid., pp. 8-9). Thus, within Durkheim his perception of religion, religion has a function in society; its rituals create and preserve social solidarity and cohesion: “creating collective benefits” (Ibid., pp. 9). Due to the rise of specialised professionals and organisations, again a process that originates from the era of industrialisation, religion gradually lost its practical functions to specialised forms of healthcare, education, control, politics, and welfare (Ibid.). As an institution, religion lost its importance in society because new institutions were capable of structuring everyday life in a ‘better’ way due to their specialised nature. The need for a religion and religion itself seems to simply fade as society becomes more differentiated through specialised organisations.

Two of the more influential contemporary understandings of secularisation were written by Dobbelaere and Bruce. Dobbelaere identifies three dimensions of secularisation (Chaves, 1989; Pickel, 2011). The first dimension is ‘laicisation’ and indicates a process

(11)

11 wherein ‘other’ institutions gain power alongside religious institutions (Ibid.). ‘Religious change’ is the second dimension, this refers to an internal change of religious organisations whereby the religious organisations are conforming themselves to the new secularised world (Ibid.). Dobbelaere his last dimension is ‘religious (dis-)involvement’ and it refers to a decline in religious practices and beliefs (Ibid.). Bruce identifies three aspects as well and starts with a decline of faith due to an absence of religious importance in the life of individuals (Pickel, 2011). His second aspect is religion its loss of social significance in everyday life and the public debate, while the third aspect covers the decreasing importance of religion in the non-religious part of society (Ibid.). All contemporary secularisation theories seem to share similar core concepts, ‘differentiation’, ‘rationalisation’ and ‘worldliness’, which indicate a change over time resulting in a decline of all traditional forms of religion (Tschannen, 1991).

The secularisation thesis has been critiqued since the 1960s, most notably by Beck and Luckmann. Beck argues that a decline in religious service attendance goes alongside an increase of individual religiosity; people are not becoming less religious, it is only the social dimension of religion that is losing ground (Speck, 2012). This individualisation of religion builds upon Luckmann his privatisation of religion. Privatised forms of religion are the result of centuries of modernisation; it is most notably a product of transformations within the economy, changes in the church-state relation, transformations in family structures and various urbanisation processes (Luckmann, 1999; Luckmann, 2003). Nowadays different worldviews are available to everyone; the idea that one their own world-view is superior is old-fashioned, because of this it is hard for religions to hold up to their single world-view: “the Church cannot interpret the world as a whole anymore” (Pollack & Pickel, 2007, p. 606). Within this theoretical perspective, church and religion are seen as two different things (Ibid., 2007). All of this gives individuals the opportunity to practise their ‘own’ form of religion, while being disconnected from traditional religious dogmas. While Beck and Luckmann originally posed their theories as a critique towards secularisation, they are now often incorporated as an aspect of secularisation. Casanova, most notably, distinguishes the ‘privatisation of religion’ as a separate aspect of secularisation alongside personal loss of religiosity and an increasing separation of religion and state (Pickel, 2011).

A fourth, and last, approach to the decline of traditional religion comes from a ‘demand side’ perspective. It originates from the early 1990s and functions as another critique on the classical secularisation theories (Norris & Inglehart, 2004, p. 11). This theory focusses on competition and assumes that there is an unconscious individual need for religion; this need has to be stimulated by religions in order to awake religiosity (Müller, 2011). Within this

(12)

12 perspective the decline of religion is due to a mismatch of what contemporary religion can supply as opposed to what contemporary society demands; a market analogy is applied to religion and religion cannot satisfy its ‘customers’ (Sherkat & Ellison, 1999). Treating religion as a producer in a market is not popular in the sociology of religion. Norris & Inglehart (2004) state that the demand side perspective of religion is in itself solely based on ‘faith’ instead of empirical evidence (p. 11). Its basic assumption, that there is an individual demand for religion, does not seem to exist as of now in Western society, as former East-Germany shows that religious demand can indeed vanish (Froese & Pfaff, 2005).

Combining the classical interpretations of secularisation together with its critiques leads to the conclusion that changes in religiosity occurred on the social, intuitional and personal level. A loss of religion on the personal level indicates that someone their personal relationship with a ‘god’ fades and that the rates of prayer and attendance of religious services go down.

Solving the problem of the origin of secularisation is complex if not impossible, but its implications on society can and have been measured; Crockett & Voas (2006) report through a British panel study that a decline of religious service attendance can be found within every generation from 1914 up until 1973. Religious service attendance declined across several birth cohorts in the 20th century (Ibid.). This process is not unique for the United Kingdom as Schwadel (2010; 2011) states that people of older cohorts, born before 1940 in the United States, have a slightly higher chance to attend religious services than individuals born after 1940. Kaufmann et al. (2012) find a similar process in Spain, Belgium, Ireland, France, the Netherlands, Britain, Denmark, Norway, Iceland, and Sweden, through their use of the European Value Study. Their longitudinal study consisted of respondents who were born between 1915 and 1994, individuals born before 1945 belong to birth cohorts that show a strong decline in the attendance of religious services while younger cohorts show a less dramatic decline (Ibid.). However, they also find that the cohort born before 1920 seems less religious than cohort born during the 1920s. Interestingly, Kaufmann et al (2012) do only elaborate on this by including a footnote which states that the cohort born before 1920 does not follow the general trend (pp. 75-78). They fail to explain this deviation from the trend. Furthermore, Kaufmann et al. note that the attendance of religious services in the youngest cohorts, individuals born in the 1980s and 1990s, is stabilising (Ibid.). More recent data from the European Value Study suggests that young adults in Northern and Eastern Europe are even showing more affection towards religion (Burkimsher, 2014). Research in the United States suggests that the religious habits of the more devoted religious follower have not

(13)

13 changed that much, while the more moderate religious following show a decline within every facet of religious life (Schnabel & Bock, 2017).

Along the lines of the various secularisation theories and several studies one can hypothesise that the attendance rates of religious services have declined in West Germany during the 20th century. This decline is known to be more dramatic for cohorts born before 1945 since these cohorts have consciously experienced the cultural changes that occurred during the 1960. It is known that, during the 1960s, a relative large amount of people, often adolescents and young adults, became alienated with religion (Brown, 2010, pp. 474-475). This happened for two reasons: on the one hand the popular youth culture that emerged in the 1960s did rarely concern itself with religion; topics of importance during this cultural revolution were equal rights, sex, peace and anything else that could be used as a distinction from earlier generations (Ibid.). On the other hand lays a reformation within Christianity; as it became more conservative and fundamentalist it alienated the following who were only mildly interested in religious life (Ibid.). Moreover, it seems that most social scientists seem to ignore the fact that religion, in at least Britain, Germany and Scandinavia, experienced some kind of rebirth during the late 1940s and 1950s (Ibid., p. 471). This religious rebirth was preceded by a movement towards secularisation in the early 1900s due the influence of a relatively progressive, liberal elite during the interwar years (Ibid., p. 472). This trend was reversed after World War II up until the 1960s, when progressive liberal ideas re-entered society (Ibid.). This knowledge leads one to believe that individuals born in the early 20th century, who came of age during the interwar years, will show lower rates of religiosity and religious service attendance than those who were born during the 1930s and came of age during the religious revival after World War II.

Birth cohorts that were born after 1945 seem to show quite a dramatic decline in their religious service attendance. This dramatic decline gradually decreased throughout the ‘younger’ birth cohorts and for this reason the youngest birth cohorts will show a movement towards stability. A movement towards stability can either mean one of two things: first it might just be a logical consequence of the more drastic decline that happened earlier in the 20th century. The decline appears to be less dramatic because the majority of individuals already stopped attending religious services; attendance rates are at such a low level that they can barely drop any more resulting in what is called a floor-effect. Second, as Schnabel & Bock (2017) proved in the United States, the most devoted religious following stays as religious as they always were, meaning that there will always be a percentage of the

(14)

14 population who attends religious services which statistically translates itself into a movement towards stability.

(15)

15

3. Method

3.1 Data and sample

Within this paper the 2017 version of the ‘German Socio-Economic Panel Study’ (SOEP, version 33;Wagner, Frick, and Schupp 2007) is used to answer the research questions. This German panel study started in 1984 and is repeated every year (DIW Berlin, 2018). Nowadays it is a longitudinal study that interviews almost 25 000 respondents in 15 000 households in each successive wave. Respondents are randomly selected and form a representative sample of the German population (Wagner, Frick, & Schupp, 2007). People are eligible to be included in the survey from the moment they turn seventeen (Ibid.). From 1990, the former German Democratic Republic, or East Germany, is also included in the dataset (Ibid.). Due to its extensive life-span the SOEP contains overlaps in age and generation; respondents from different birth cohorts are observed at the same age. The presence of these overlaps is essential to the analysis. It allows one to compare successive birth cohorts at a set age; for example, one becomes able to compare the predicted religious service attendance rate of 40 year olds who were born in the 1950s, 1960s and 1970s. Due to this, birth cohorts can be compared regardless of possible age effects.

As of 2016 the SOEP has a starting sample of 84 862 respondents measured over 33 waves which results in 619 718 observations. Those 84 862 respondents are all the individuals that were observed at least once between 1984 and 2016. Several sample cuts are necessary in order to prepare the data in such a way that it can be evaluated along the lines of the theoretical background. The sizes of these cuts and their corresponding exclusion criteria are presented in ‘Table 1’. First, all panel waves that did not observe the outcome variable, are excluded from the sample. This sample cut resulted into the exclusion of 21 252 respondents, which is 25.1% of the starting sample. The starting sample of the SOEP contains several oversamples for high-income groups, households and children. Including these oversamples into the sample results into an overrepresentation of these groups, because of this it is necessary to cut them. Cutting these oversamples results into a loss of 18 206 respondents and thus 21.1% of the starting sample. It is also fruitful to exclude the respondents that were socialised in the German Democratic Republic and those that lived in this region after the reunification of Germany. Extensive secularisation occurred relatively early in this region, during the late 19th century, which is why modern panel studies are incapable of studying this

(16)

16

phenomena in East Germany1. Excluding every East German respondent results into another

loss of 12 217 respondents, this represents 14.4% of the initial sample. Respondents with an immigration background are excluded for another reason; Europe is currently experiencing religious revival through the Islam. Children of non-Western immigrants are as religious as their parents and sometimes even more religious, which leads many researchers to believe that the secularisation process is something that exclusively occurs among Western religions (Fleischmann & Phalet, 2012; Voas & Fleischmann, 2012). Including respondents with an immigration background into the sample would intervene with the readability of the results as there exists a clear theoretical difference between Western and non-Western religions. 9 332 respondents, which represents 11% of the initial sample, were lost due to this sample cut. Next, respondents who are younger than 17 years or older than 86 years are excluded from the data set. Respondents under the age of 17 are not supposed to be in the initial sample, and the respondents over the age of 86 are in such small numbers that they could potentially harm the statistical model. Excluding them translates into a loss of 183 respondents and thus 0.22% of the starting sample. Furthermore, the respondents who were born from 1900 through 1909 are excluded, because they only consist out of 268 observations. It is necessary, due to this small amount of observations, to exclude these respondents. This sample cut excludes 147 respondents from the analysis, which is 0.17% of the initial starting sample. Finally, respondents who do not have values for the outcome variable while they were observed in the relevant waves were excluded. This sample cut resulted into the exclusion of 556 respondents, which is 0.66% of the starting sample.

After these sample selections the final sample, which will be used in the analysis, consists out of 22 969 respondents and 90 827 observations, measured over eleven waves

1

The former German Democratic Republic, along with the Czech Republic, can be seen as an outlier among the other former Soviet states. After the fall of the Soviet Union the secularisation process in these countries became even more advanced than that it already was under communist regime, while most other post-communist societies experienced high levels of religious revival (Müller, 2011; Froese & Pfaff, 2005). In a cross-sectional study Grautier (1997) concluded that only the very oldest East German birth cohort showed the same levels of attendance to religious services as the West German society in 1991. East Germans show some of the lowest mean church attendance rates in Europe (Müller, 2011). Froese & Pfaff (2005) give two reasons for East Germany its relatively low religiousness and religious services attendance rates, and stress that modernisation was not involved in the secularisation of East Germany. First, East Germany was ‘very’ socialist before the Soviet Union took control over the country which already led to relatively high levels of secularisation decades before the secularisation process in West Germany, and Western Europe, happened. During the communist regime this only enhanced as scientific atheists and atheist institutions were given an advantage over religious institutions. The second reason was the reregulation of religious life which came along with the reunification of the German Democratic Republic and the Federal Republic of Germany. Anyone who wanted to join a church or participate in other religious activities had to pay certain taxes which demotivated the East Germans to revive their religiosity (Ibid.). All of this makes it impossible to study secularisation, through panel data, in the former German Democratic Republic as the secularisation process happened before panel studies existed. Even the oldest respondents would be too young to give any sensible information about the secularisation process that occurred in the former German Democratic Republic.

(17)

17 between 1992 and 2015. The selected sample has an average duration of observation of 4 waves. Within the selected sample, 14 125 respondents who were observed in at least one older version of the SOEP are not observed in 2015, the last survey year in the selected sample. These respondents either died or actively left the panel before they could be observed in 2015.

Table 1: Sample selection

N(amount of respondents cut) Percentage

Starting sample: 84 862 100

Exclusion criteria

Outcome not measured 21 252 25.1

Oversamples 18 206 21.1

Not West German 12 217 14.4

Immigration background 9 332 11.0

Younger than 17 or older than 86 183 0.22

Cohort of 1900-1910 147 0.17

Missing values outcome variable 556 0.66

Selected sample: 22 969 27.1

Note: SOEP, v33, release 2017.

3.2 Operationalisation

The variable ‘Attend Church Or Other Religious Events’ will act as the fundament for measuring the respondents their religious service attendance rate. This variable was measured in 1992, 1994, 1996, 1997, 1999, 2001, 2005, 2007, 2009, 2011 and 2015. In each of those years respondents are asked “In which of the following activities do you spend your free time? Please indicate how often you spend your free time doing this ~ every week, every month, seldom or never?”. This categorical variable is recoded into the outcome variable ‘Attendance of Religious Services (Days per Year)’, were ‘every week’ results into 52 days per year, ‘every month’ into 12 days per year, ‘seldom’ into 4 days per year and ‘never’ into 0 days per year. Recoding the original variable in this particular way allows one to read the results in a relatively easy way, as it can be treated as a continuous variable. ‘Table 2’ shows that the mean Attendance of Religious Services (Days per Year) is roughly 8 days with a standard deviation of 15.

(18)

18 The original variable ‘Attend Church Or Other Religious Events’ is also recoded into another variable that will measure the percentage of individuals who attend religious services on a weekly basis. It consists of the proportion of respondents who attend religious services weekly times one hundred to portray the outcome values in percentages. This variable will be used to test whether the hypothesised movement towards stability, of attendance of religious services through birth cohorts, is due to the fact that the most devoted religious group is remaining fateful to their religion. The mean of the variable ‘Percentage of Weekly Religious Service Attendance’ is 10.42 with a standard deviation of 30.56 (see ‘Table 2’).

The variables ‘Survey Year’ and ‘Birth Year’ will be used to constitute the variable ‘Age’ this variable itself will be used to indicate if aging has an influence on the outcome variable ‘Attendance of Religious Services (Days per Year)’. The mean age of the selected sample is 48.64 with a standard deviation of 17.48 (see ‘Table 2’).

The variable ‘Birth Year’ will also be used to categorise respondents into different birth cohorts. These birth cohorts consist of individuals who were born, roughly, at the same time (Elder & George, 2016, p. 60). Birth cohorts will be used to analyse whether social change has effect on the outcome variable. The birth cohorts which are used in this paper are ‘1910-1919’, ‘1920-1929’, ‘1930-1939’, ‘1940-1949’, ‘1950-1959’, ‘1960-1969’, ‘1970-1979’, ‘1980-1989’ and ‘1990-1999’.

(19)

19

3.3 Models

In the analysis, outcomes while be estimated with generalized least squares random effects models. It is necessary to use such a model due to the nature of panel data. While depending on panel data one must realise that different observations are nested in one individual, which results into high correlation values among the observations that are connected to one individual. A generalized least squares random effects model takes this nested nature of data into account and structures the data to prevent said high correlation values. Using such a model also enables one to estimate change in religious service

Table 2: Descriptive statistics

Variables N(obs.) Range Mean

values

S.D. Measurement

Attendance of Religious Services (Days per Year)

90 827 0-52 8.07 15.41 52 = Weekly attendance 12 = Monthly attendance 4 = Sometimes 0 = Never Percentage of Weekly Religious Service Attendance

90 827 0-100 10.42 30.56 Proportion of respondents who attend religious services once a week times 100

Age 90 827 17-86 48.64 17.48 Survey year minus birth year

Birth cohorts 90 827 Birth years combined into 10-year

intervals 1910-1919 1 328 0-1 0.02 1920-1929 6 271 0-1 0.07 1930-1939 12 590 0-1 0.14 1940-1949 14 472 0-1 0.16 1950-1959 16 536 0-1 0.18 1960-1969 20 804 0-1 0.23 1970-1979 10 998 0-1 0.12 1980-1989 6 122 0-1 0.07 1990-1999 1 704 0-1 0.02

(20)

20 attendance when individuals age, while simultaneously estimating change in religious service attendance between individuals across cohorts.

The models that are used within this paper contain independent variables for age and birth cohort to estimate the dependent, or outcome, variable ‘Attendance of Religious Services (Days per Year)’. It is important that the parameters in the random effects generalised least squares model represent the trend of the descriptive graph. Below, ‘Figure 1’ shows the descriptive trend of the effects that aging, the coloured lines, and social change, the distances between the coloured lines, have on the average amount of days that an individual attends religious services.

‘Figure 1’ reports that the variable ‘Age’, is rather constant, apart from some minor fluctuations. This leads one to believe that the original continuous variable ‘Age’ is a sufficient way of measuring the age effect. However, the effect that the continuous variable ‘Age’ has on the outcome variable might not be linear. It could well be that measuring the age effect as a quadric or cubic parameter results into a ‘better’ estimation. An estimation is ‘better’ when the R2

(model fit) of a model has a higher value. The quadric version of ‘Age’ is no longer significant but a cubic measurement of the variable ‘Age’ has a significant (p < 0.001) effect on the outcome variable, just like the continuous variable ‘Age’, and predicts that the age effect on the outcome variable is not linear for the youngest and oldest respondents. However, the model fit (R2) does not improve. Utilizing a cubic version of the variable ‘Age’ does complicate the interpretation of the results. As using the cubic version of ‘Age’ will complicate the analysis, it will be more fruitful to parameterise age as its original continuous variable.

The descriptive graph, ‘Figure 1’, reports that the birth cohorts do not follow a totally linear trend, the youngest birth cohorts are more similar to each other compared to the older birth cohorts. When the birth cohort variable is parametrised as a continuous variable, the social effect is treated as a linear effect. In this configuration every cohort is estimated to have the same effect on the outcome variable and thus the ‘1910-1919’ cohort appears to be the ‘most’ religious cohort, while the descriptive graph clearly tells otherwise. Utilising the birth cohort variable as dummy variables solves this ‘misinterpretation’ as a separate estimation is made for each birth cohort. For this model the birth cohort of 1910-1919 is used as reference category as it is the oldest cohort in the selected sample. Thus, the model that is best suited for the analysis is a model where ‘Age’ is used as a continuous variable while the variable ‘Birth Cohorts’ is used as a dummy variable with the birth cohort of ‘1910-1919’ as reference category. Note that a variation on this model will be used to test whether the hypothesised

(21)

21 absence of an age effect on the outcome variable is true for every birth cohort. To test this, the model will incorporate an interaction between ‘Age’ and the dummy variables of ‘Birth Cohorts’. ‘Figure 1’ reports that the birth cohort of 1960-1969 is the most ‘stable’ in its age effect. Using this birth cohort as the reference category will facilitate testing whether the age effect of the other birth cohorts differ in a statistical and meaningful way.

The model that predicts the outcome of the ‘Percentage of Weekly Religious Service Attendance’ is constituted along the same rationale. This model is similar to a generalized least squares random effects model, but is in fact a linear probability model because the outcome values of this model are estimated in percentages. ‘Appendix: Figure 4’ shows the descriptive trend of the effects of aging, the coloured lines, and social change, the distance between the coloured lines, on the percentage of individuals who attend religious services on a weekly basis.

‘Figure 4’ shows something that is similar to ‘Figure 1’. ‘Figure 4’ shows that the values of the variable ‘Age’ are rather stable. Again, it seems that the original continuous variable for ‘Age’ is sufficient in this model as its effect on the outcome variable is significant (p < 0.001). However, it might well be that a quadric or cubic measurement for an age effect fits the model ‘better’. As before the quadric version is not significant, while the cubic version is (p < 0.001). In this case the model fit (R2) does improve when ‘Age’ is parametrised as cubic. However, it only improves by a minimal 0.001. Using a cubic version of the variable ‘Age’ will complicate the analysis, thus it will be more useful to parameterise the age effect as its original continuous variable. The social effect in ‘Figure 4’ is not a totally linear effect, younger birth cohorts differ less from each other than older birth cohorts and the social effect is reversed for the oldest birth cohort. For this reason it is necessary to separately estimate the effect for each birth cohort, thus the birth cohorts must be parametrised as a dummy variables within the linear probability model. The birth cohort of ‘1910-1919’ is used as the reference category since it is the oldest birth cohort in the selected sample. Thus, the model that is best suited for the analysis is a model where ‘Age’ is used as a continuous variable while the variable ‘Birth Cohorts’ is used as a dummy variable with the birth cohort of ‘1910-1919’ as reference category.

(22)

22

Figure 1: Descriptive life course and cohort trends of religious service attendance. Note: SOEP version 33

(1992–2015), release 2017.

3.4 Period effects

While conducting research along the lines of an age, cohort study it is recommended to investigate whether or not any events occurred, during the time lapse of the panel study, that could have had an impact on the outcome variable. Such an event is called a period effect and could distort the analysis if it is not incorporated into the model. Period effects might move respondents towards sudden changes in their behaviour, and can thus influence the age effect and the social effect. Period effects differ from an age effect and a social effect in the sense that they affect everyone in the selected sample while an age effect influences the life course of an individual and a social effect is only applicable to certain birth cohorts. It is unlikely that anything of such magnitude, connected to the attendance of religious services, occurred in Germany while the outcome variable was measured between 1992 and 2015. If anything could classify as period effect it might be the election of a German pope, Benedictus XVI, in 2005. But more often than not period effects must only be seen as ground-breaking events that really change the way a whole society perceives itself. The election of a pope is not necessarily something that can force a whole society to attend religious services more often after decades of secularisation. Moreover, there are no sources that hint or even hypothesise towards an increase in religious service attendance since the German pope was elected. Thus the models are not influenced by a period effect.

0 5 10 15 20 25 A v e ra g e D a y s o f R e lig io u s S e rv ic e A tte n d a n c e p e r Y e a r 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 Age 1910-1919 1920-1929 1930-1939 1940-1949 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 Birth Cohorts

(23)

23

4. Results

Below, ‘Figure 2’ and ‘Figure 3’ report the visual representations of the results that were gained through the ‘random effects generalised least squares’ models. In all figures the coloured lines represent different birth cohorts, and each of them represents ten years. Birth cohort ‘1910-1919’, black, is in all figures the oldest birth cohort and birth cohort ‘1990-1999’, sienna, is in all figures the youngest. The length of the lines, which are following the x-axis, indicate the ages that are represented in the different birth cohorts and how they vary in relation to the outcome variable on the y-axis, thus indicating whether there exists an age effect or not. Within ‘Figure 2’ this y-axis represents the, predicted, average days of religious service attendance per year, while the y-axis of ‘Figure 3’ represents the, predicted, percentage of weekly religious service attendance. The distance between the ‘birth cohort lines’ indicate the differences between the various cohorts, meaning that the difference in religious service attendance or percentage of weekly service attendance is more dramatic if the ‘birth cohort lines’ are, relatively to the others, further apart. Thus, if the ‘birth cohort lines’ are closer to each other the difference in religious service attendance or percentage of weekly service attendance is less dramatic or minimal if they share the same values.

Figure 2: Predicted life course and cohort trends of religious service attendance. Note: SOEP version 33 (1992–

2015), release 2017. 0 5 10 15 20 25 A v e ra g e D a y s o f R e lig io u s S e rv ic e A tte n d a n c e p e r Y e a r 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 Age 1910-1919 1920-1929 1930-1939 1940-1949 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 Birth Cohorts

(24)

24

4.1 Aging and religious service attendance

‘Table 3, Model 1’ shows the results of the regression analysis on which ‘Figure 2’ is based. ‘Figure 2’ shows that aging has a minimal negative effect on the outcome variable ‘Attendance of Religious Services (Days per Year)’. This negative trend appears to be very small, as the cohort lines are declining in a rather conservative way. ‘Model 1’ in ‘Table 3’ proves that the effect that aging has on the attendance of religious services is indeed small. However, said predicted effect is significant (p < 0.001) which might allow one to think that aging truly has an effect on the attendance of religious services. In reality the religious services attendance rate of the individual will decline by 0.02 for each year that they age, effectively meaning that it takes an individual fifty years to drop their religious service attendance rate by one day per year.

Aging was hypothesised to have no significant effect on the attendance of religious services. Past research reported that any age effect, if it could be measured at all, would never appear to be significant. On the contrary the above proves that there is indeed a significant, negative effect where aging influences the attendance of religious services among the West German population. However, translating this significant, negative effect to the empirical and theoretical world is anything but satisfying. As portrayed the negative effect that aging has on the attendance of religious services is minimal. Every year an individual ‘ages’, results into him or her attending religious services 0.02 days less compared to the year that came before. This negative effect does not appear to be strong enough to be explained by anything other than Bahr his stability model. Both the traditional and family-cycle model predict dramatic changes in religious service attendance during the life course. It is clear that neither of these models can explain the trend as shown in ‘Figure 2’ as there does not seem to be any kind of dramatic change during the life course. One may find arguments to categorise this effect as Bahr his disengagement model, as individuals do seem to attend religious services less often as they get older. The disengagement model states that the elderly in society are cut off from social activities due to their age and physical state. In this context this disengagement results into an almost complete absence of the elderly during religious services. Due to the predicted minimal negative effect, this reasoning cannot explain the trend as shown in ‘Figure 2’. In general, individuals attend religious services at, a somewhat, stable rate during their life course; the religious services attendance seems to be established at a young age. This leads to the conclusion that aging has, however significant its effect is, no substantive effect on religious service attendance rates among the West German population.

(25)

25

Table 3: Random Effects GLS Models, Predicted Days of Attendance of Religious Services per Year

Model 1 Model 2 B SE B SE Age -0.02*** 0.01 0.00 0.01 Birth Cohorts 1910-1919 . . 24.29*** 7.25 1920-1929 2.77*** 0.73 17.23*** 1.89 1930-1939 -0.39 0.69 6.76*** 1.09 1940-1949 -3.92*** 0.69 1.93* 0.92 1950-1959 -6.09*** 0.70 5.05*** 0.78 1960-1969 -7.55*** 0.71 . . 1970-1979 -8.81*** 0.74 -1.24 0.72 1980-1989 -9.76*** 0.77 1.16 0.91 1990-1999 -9.91*** 0.83 4.17 2.48

Age x Birth Cohort

Age x 1910-1919 -0.22* 0.09 Age x 1920-1929 0.10*** 0.03 Age x 1930-1939 0.00 0.02 Age x 1940-1949 0.02 0.02 Age x 1950-1959 -0.08*** 0.02 Age x 1960-1969 . . Age x 1970-1979 0.01 0.02 Age x 1980-1989 -0.13*** 0.03 Age x 1990-1999 -0.30* 0.12 Constant 14.31*** 0.77 5.87*** 0.46 R2 0,045 90 827 22 969 0.046 90 827 22 969 Observations Individuals

Note: SOEP, v33, release 2017. Ref. Birth Cohort; Model 1 = 1910-1919, Model 2 = 1960-1969

(26)

26 Wingrove & Alston (1974) speculated that different birth cohorts might experience different age effects. Unconvinced by their own cross-sectional research design, Wingrove & Alston could not make any lasting statements about their speculation. It is possible to hint towards any birth cohort differences in the effect that aging has on the attendance of religious services by adding an interaction effect, between age and birth cohort, to the earlier explained model. Model 2 as presented in ‘Table 3’ shows the regression analysis that is illustrated in ‘Appendix: Figure 5’. In this model the birth cohort of ‘1960-1969’ is the reference category of the other birth cohort dummy variables due to the relative stability of its age effect.

At first glance it becomes clear that the age, cohort interactions of the birth cohorts of ‘1930-1939’, ‘1940-1949’ and ‘1970-1979’ do not have a significant effect on the outcome variable ‘Attendance of Religious Services (Days per Year)’ when compared to the reference birth cohort of ‘1960-1969’. Thus, there is no evidence for the existence of an age effect on the predicted attendance of religious services for the birth cohorts of 1930-1939’, ‘1940-1949’ and ‘1970-1979’. However, the other five age, cohort interactions do have a significant effect on the predicted outcome variable ‘Attendance of Religious Services (Days per Year)’ when compared to the reference birth cohort of ‘1960-1969’.

First, the age, cohort interaction of the birth cohort of ‘1910-1919’ has a negative significant (p < 0.05) effect of 0.22 days of religious service attendance per year on the outcome variable. This indicates that the religious service attendance rates of individuals who were born between 1910 and 1919, decreases by one day per year for every 4.55 years that they age. Individuals who belong to the birth cohort of ‘1920-1929’ experience a significant (p < 0.001) decrease of 0.10 days of religious service attendance per year for each year that they age, effectively meaning that it takes them ten years to decrease their religious service attendance rate by one day per year. The age, cohort interaction of the birth cohort of ‘1950-1959’ has a negative significant (p < 0.001) effect of 0.08 days of religious service attendance per year; individuals who are born in this birth cohort decrease their religious service attendance rate by one day every 12.5 years. The birth cohort of ‘1980-1989’ shows a significant (p < 0.001) decline of 0.13 days of religious service attendance per year for each year that an individual ages, resulting in a decline of one day per year over the course of 7.7 years. Finally, individuals who were born between 1990 and 1999 experience a significant (p < 0.05) decrease of 0.30 days of religious service attendance per year for each year that they age, effectively meaning that their religious service attendance rate decreases by one day per year for each 3.33 years that they age.

(27)

27 The results seem to imply that the earlier found significant but minimal negative age effect does statistically change across some of the birth cohorts. First, the age effect does not exists within the birth cohorts of ‘1930-1939’, ‘1940-1949’, ‘1960-1969’ and ‘1970-1979’ as their trends are quite stable. In essence this makes said birth cohorts similar to the minimal general negative age effect. Another birth cohort which is comparable to the general age effect is the birth cohort of ‘1950-1959’ since its age effect is only marginally stronger than the age effect of the combined population, but it is still rather stable and thus not substantively different. The most statistically substantive differences are shown in both the oldest and the youngest birth cohorts. Their age effects are from five times, ‘1920-1929’, up to fifteen, times, ‘1990-1999’, stronger than the age effect of the general population. However, it is hard to fully theorise the more excessive age effects of these birth cohorts as the model is limited by their relatively poor representation of the average life course.

The birth cohorts of ‘1910-1919’ and ‘1920-1929’ only represent the very end of the life course while the birth cohorts of ‘1980-1989’ and ‘1990-1999’ only roughly represent ‘early’ adulthood. Moreover, theoretical evidence for differences in age effects across cohorts is non-existent. Due to this it is perhaps more likely that the age effects within these cohorts are different because it only manifests itself at a certain age and not in a certain birth cohort. Considering that the birth cohorts without a statistical age effect contain more respondents, it might well be that they compensate the stronger age effect in the birth cohorts of ‘1910-1919’, ‘1920-1929’, ‘1980-1989’ and ‘1990-1999’ when the combined age effect is measured as in Model 1; resulting in a significant negative but relatively stable age effect.

Thus, all in all it seems unlikely that there are meaningful differences in the effect that aging has on the attendance of religious services across different birth cohorts. Furthermore, this new insight of the age effect does add something to the earlier found significant but minimal combined age effect. It seems that the religious service attendance rate is indeed established relative early in the life course, during young adulthood, and that a decline of the religious service attendance rate is more common when one reaches old age. However, this does not change the interpretation of the initial age effect, as it still does not verify any of Bahr his theoretical models. While the young adults seem to leave church at a high rate, they do not seem to return. The elderly, in spite of a declining predicted religious service attendance rate, are still attending religious services in large numbers which falsifies the idea that disengagement is common when one reaches old age.

(28)

28

4.2 Attendance of religious services across birth cohorts

As before, ‘Table 3, Model 1’ shows the results of the regression analysis on which ‘Figure 2’ is based. Both the regression analysis and the graph report that there exists an overall decline in religious service attendance when comparing the youngest birth cohort to the oldest birth cohort. However, the oldest birth cohort of ‘1910-1919’ has a significantly (p < 0.001) lower predicted religious service attendance rate of 2.77 days per year than the birth cohort of ‘1920-1929’. All birth cohorts are significantly (p < 0.001) different from the reference birth cohort of ‘1910-1919’, except for the birth cohort of ‘1930-1939’. This is due to the similarity of the predicted religious service attendance rates of those two cohorts. Every birth cohort that is ‘younger’ than the cohort of ‘1920-1929’ declines in religious service attendance when compared to is predecessor. Younger birth cohorts differ less from each other than older birth cohorts. The predicted religious service attendance rate of the youngest birth cohort of ‘1990-1999’ is ‘only’ 0.15 days per year lower than the birth cohort of ‘1980-1989’, while the predicted religious service attendance of the birth cohort of ‘1940-1949’ is 3.53 days per year lower than the birth cohort of ‘1930-1939’.

It is clear that the predicted religious service attendance rate decreased across birth cohorts over the course of the 20th century in West Germany. The hypothesis stated that it would indeed be likely that younger birth cohorts attend religious services less often than older birth cohorts and that the differences between birth cohorts would be less dramatic among the youngest birth cohorts. As hypothesised the birth cohort which came of age during the cultural changes of the 1960s, ‘1940-1949’, shows the biggest difference in their predicted religious service attendance rate compared to its predecessor the birth cohort of ‘1930-1939’.

Religious service attendance is the social dimension of religiosity; this might indicate that people are starting to prefer to practise religion on their own terms. But most sources report that every facet of religious life is following a trend of decline. Because of this it is almost impossible to specify exactly what causes this decline in religious service attendance. One explanation might be in line with the classical secularisation theories which state that modernisation and industrialisation eliminates the need and symbol of religion and its services. While another classical interpretation might state that religion stopped ‘making sense’ to the rational and educated human being who consults modern science instead of god. Another explanation might argue that individuals gradually stopped attending religious services over time because religion is becoming something personal. The only certainty in all of this is that religious service attendance rates are substantively lower in younger birth

(29)

29 cohorts than in older birth cohorts, but it also seems appear that attendance rates can increase again under specific circumstances.

The most surprising results of the analysis might well be the difference between the predicted religious service attendance rate of the birth cohort of ‘1910-1919’ and the birth cohort of ‘1920-1929’, considering that the predicted attendance rate of the former is significantly lower than the latter. On the one hand, it might be a statistical misinterpretation as the age bracket of the oldest cohort is relatively ‘old’ which might cause this deviation from the general trend. On the other hand, the chance that a substantive age effect does exist has proved to be rather slim within this statistical model and other studies. One may argue that people who attend religious services have a higher life expectancy than those who do not because they might be more likely to make ‘good’ lifestyle choices due to their religiosity. But in this case this reasoning makes no sense if one considers that it is the oldest age bracket that has a lower religious service attendance rate than their minors. As said before, Kaufmann et al. (2012), while they are not using panel data, stumble upon the same deviation from the trend when using the data of ten European countries but fail to explain its existence. Other age, period cohort studies conducted with data from the United States and the United Kingdom do not find any evidence for this deviation at all (Schwadel, 2010; Schwadel, 2011; Burkimsher, 2014). Historians who concern themselves with religion and the fluctuations in its popularity are, however, clear in their accounts on religion its state during the 20th century. Like social scientists, historians tend to have fierce discussions about the reasons why religiousness and religious service attendance declined, but they do agree upon the idea that religious service attendance rates started to decline during the early 1900s in West Germany (Brown, 2010). For a brief moment in time this process was reversed after World War II. This last notion might be part of the explanation why the birth cohort of ‘1920-1929’ has a higher predicted religious service attendance than the birth cohort of ‘1910-1919’. The cohort of ‘1920-1929’ came of age during the years that came directly after World War II. As aging has no substantive effect on the attendance of religious services it is likely that an individual their religious service attendance rate is established before or when he or she comes of age. Thus, it could well be that the birth cohort of ‘1920-1929’ has a higher predicted religious service attendance rate than the birth cohort of ‘1910-1919’ due to various social developments that occurred after World War II.

(30)

30

4.3 A movement towards stability?

The general trend of decline in religious service attendance, as explained in the section above, gradually slows down towards what seems to be stability. The literature states that this movement towards stability is either one of two things: a situation that one perceives as stability or true stability where a, devoted, percentage of society is still attending religious services on the same frequent basis as they did throughout all generations. We know that the latter situation is true for the United States, while it is thought that European countries experience a decline in religious service attendance across all the magnitudes of religious following (Schnabel & Bock, 2017).

‘Figure 3’ shows the trend of predicted percentage of weekly religious service attendance in West Germany. This graph is based on the linear probability model of ‘Appendix: Table 4’. ‘Figure 3’ and ‘Table 4’ report that there exists an overall significant (p < 0.001) decline in the predicted percentage of weekly religious service attendance when comparing the youngest birth cohort to the oldest birth cohort. Again the general trend of decline is interrupted by the birth cohort of ‘1920-1929’, which has a considerably higher predicted percentage of weekly religious service attendance, of 5.11 percentage point, than the birth cohort of ‘1910-1919’. From the birth cohort of ‘1920-1929’ up through the birth cohort of ‘1990-1999’, one can see a decline in predicted percentage of weekly religious service attendance. Apart from the birth cohort of ‘1930-1939’, every birth cohort has a significantly (p < 0.001) different predicted percentage of weekly religious service attendance compared to the reference birth cohort of ‘1910-1919’. The trend as pictured in ‘Figure 3’ is similar to the trend that describes the average amount of days that respondents attend religious services in a year in ‘Figure 2’. Both trends show a mostly linear decline in religious service attendance that slows down for the youngest birth cohorts. The difference between the cohorts of ‘1930-1939’ and ‘1940-1949’ is 6.28 percentage point compared the birth cohort of ‘1910-1919’ whereas the difference between the cohorts of ‘1980-1989’ and ‘1990-1999’ is ‘only’ 0.34 percentage point compared the birth cohort of ‘1910-1919’. What this essentially means is that the most devoted religious group in West Germany, the group that attends religious services once a week, has gotten considerably smaller over the years.

The analysis of the most devoted religious group illustrates a comparable trend to that of the general population. Because the most devoted group of religious service attendees in West Germany is shrinking, it is reasonable to assume that the perceived movement towards stability is more likely to be a floor-effect.The graphical representation of the model is unable to illustrate a further negative trend since the amount of respondents that do attend religious