1

Questions on religion are increasingly rare in modern censuses. Information on religious affiliation has been absent from French national statistics since 1872, (see article on French protestants in Population, English Edition, 1-2009), but is still recorded in all Swiss censuses, and for all events entered in the Swiss civil records (births, marriages, deaths). Mathias Lerch, Michel Oris, Philippe Wanner and Yannic Forney have analysed these data sources to reveal links between religious affiliation reported in the 1990 census and the probability of dying over the period 1991-2004. Mortality is higher among persons reporting no religion, and lower among minority Christian groups (other than Roman Catholic and Protestant) where social cohesion is stronger and religious practice more intense. These disparities can be linked to lifestyle differences, as confirmed by analysis of the risks of dying associated with various causes of death such as accidents, suicide, cardiovascular disease and certain cancers.

2Cultural in essence, religion carries a strong emotional charge and inflames current debate. Viewing it as a private concern, most European nations have excluded this question from the sphere of public intervention. The trauma of the Second World War reinforced this trend in most countries, but Switzerland remains an exception, [1] and the religion of its residents is systematically recorded on census forms and death certificates.

3Seen as a regulator of social relationships, membership of a religious group is a key sociodemographic indicator. Switzerland is a pluralist country, including in terms of religion, and thus provides an excellent testing ground for analysing the demographic traits associated with religious affiliation. While up to the nineteenth century religious identity was defined mainly by territorial factors (linked notably to the geographical spread of Protestantism), growing secularization led to a separation of the religious sphere from the other dimensions of social interaction. In parallel with recent demographic developments – migration, settlement by foreign communities, diversification of lifestyles – the religious landscape of Switzerland has become more heterogeneous. Alongside the decline of the traditional churches, the growing numbers of minority communities have steadily increased their membership.

4Religious belief was traditionally a key component of Swiss demographic behaviour, as evidenced by analyses of fertility decline in the country (Lesthaeghe and Neels, 2000). Much like the effects of religiosity observed in the United States, changes in the Swiss religious landscape are liable to affect the longevity of its population, which stood at 79.7 years for men and 84.4 years for women in 2008. In the Anglo-Saxon world, religiosity is defined as a multi-dimensional religious commitment which involves participation in collective rituals and sharing of beliefs as much as individual religious knowledge and practice (Swatos and Kivisto, 1998). However, differences in life expectancy by religious affiliation have rarely been studied by demographers (Vallin et al., 2002, p. 282), although their importance has been highlighted by epidemiologists (George, 2003; Ryff et al., 2004). Individuals are born into a community or religious tradition that will shape their behaviour throughout their life. While ecological analysis no longer reveals a systematic mortality difference between Roman Catholic and Protestant cantons (Wanner et al., 1997), the relationship between religion and mortality is addressed here at a finer level through information on individual religious affiliation.

5In the first section of this article, we present this shifting religious landscape by examining the age-sex structure of the different religious groups. We will then introduce mortality data and discuss the quality of information on religious affiliation before proposing an adjustment method. The enigmatic phenomenon of religious “mobility” before death is measured and taken into account to estimate the cross-sectional life tables presented in part three. The fourth section provides a multivariate assessment of the relationship between religion and mortality which confirms the existence of survival differentials between religious groups. Last, analysis by cause of death highlights the importance of lifestyle differences in explaining the differentials observed.

I – Census information on religious affiliations in Switzerland

6The results of the most recent census in 2000 show that the two majority religious denominations in Switzerland have dwindled in size since the census of 1970 (Table 1). While Roman Catholics and Reformed Protestants occupied 96% of the religious landscape in 1970, they represented only 42% and 33% of the resident population, respectively, in 2000. The downward trend is more marked for the Protestants (13 percentage points) than for Roman Catholics (8 points) because of strong immigration from Roman Catholic countries (Italy, Spain and Portugal; see Bovay and Broquet, 2004) up to 1990. The decline of both majority groups was particularly intense in the last decade of the twentieth century.

Table 1

Relative weights of the different religious groups in 1970, 1980, 1990 and 2000 (%)

Religion 1970 1980 1990 2000 Reformed Protestant 46.4 43.9 38.5 33.0 Roman Catholic 49.4 47.6 46.2 41.8 Other Christian denominations(a) 2.0 2.3 3.5 4.4 Jewish 0.3 0.3 0.3 0.3 Other non Judeo-Christian religions(b) 0.4 1.1 2.6 5.0 of which Muslim 0.3 0.9 2.2 4.3 No religion 1.1 3.8 7.4 11.1 Non-response 0.4 1.1 1.5 4.3 Total population 6,269,783 6,365,960 6,873,687 7,288,010 (a) Evangelical Methodist Church, New Apostolic Church, Jehovah’s Witnesses, Christian Catholic Church, Orthodox Church and other Christian churches and communities. (b) Muslims, Buddhists, Hindus, etc. Sources: Federal population censuses (Bovay and Broquet, 2004).

Relative weights of the different religious groups in 1970, 1980, 1990 and 2000 (%)

7For all minority groups, their share in the total population increased. Note also the growth in the “no religion” category. This was a marginal choice in the 1970s, but concerned one-tenth of the enumerated population in 2000. Membership of “other” Christian churches also increased, from 2% of the population in 1970 to 4.4% in 2000. This group is comprised mainly of evangelicals (60%), Orthodox Christians and Christian Catholics. Among the non Judeo-Christian religions, the Muslims contribute most to the rapid growth observed, with a share that rises from 1% to 4% of the Swiss population between 1980 and 2000. Buddhists, Hindus and other religions represent a very small share of the total. The proportion of persons who did not answer the question on religious affiliation increased substantially between 1990 and 2000, to above 4%. However, there is no way of knowing whether they have a religious affiliation but refuse to report it, or whether they have no religion. It is nonetheless certain that problems linked to the organization of the 2000 census are partly responsible for this increase (see next section).

8The various religious groups have very distinct age structures (Figure 1). Among the Reformed Protestants and the Roman Catholics – the oldest and still by far the largest religious groups in the country – the numbers of young people are falling while the proportion of over-60s increased between 1990 and 2000. The members of these two religious groups are ageing, therefore, although the old-age dependency ratio is lower among Roman Catholics (24 per 100 working-age members aged 20-64) than among Protestants (36 per 100) thanks to an inflow of young adult immigrants up to 1990. The proportion of women in the two majority groups is also tending to increase because women live longer than men.

Figure 1

Age-sex structure by religious affiliation, Switzerland, 2000

Age-sex structure by religious affiliation, Switzerland, 2000

9Individuals with no religious affiliation are most numerous in the 30-49 age group, where they represent 15% of the total. Men also more often report having no religion than women, whatever their age. Beyond age 50, the proportion of individuals with no religion decreases, doubtless due to a generation effect. The absence of religious affiliation in Switzerland is a recent phenomenon; as we have seen, very few people reported no religion in 1970 (Table 1). While the earliest non-believers were individuals who rejected traditional religious institutions, their strong presence among persons of reproductive age has led to the transmission of such attitudes from one generation to the next (Bovay and Broquet, 2004). Young children are strongly represented, testifying to recent change in the mode of reproduction.

10The other Christian denominations have a pyramid similar to that of the population with no religion. The development of the evangelical churches is the main reason for this strong growth among young adults, and secondarily, the arrival of Orthodox Christian immigrants from former Yugoslavia in the 1990s. This group of “other” Christian denominations – neither Roman Catholic nor Calvinist or Lutheran Protestant – has higher fertility than the national average (Bovay and Broquet, 2004) and a “capacity to keep its children inside the community” (Favre, 2006, p. 280). This explains why the cohorts of children are more strongly represented than in other religious communities. The expansion of this category is recent, so the over-60s are few in number – as is also the case for the “no religion” group.

11The non Judeo-Christian population is very young. This is due largely to the Muslim population, of which 39% was aged below 20 in the most recent census. Beyond age 50, the numbers are much smaller. The bottom-heavy shape of the pyramid reflects the growing diversity of migration flows to Switzerland in recent years. The share of Italians, who represented almost 60% of foreigners in the 1960s, had fallen to one-third of its earlier level by 2000, while the relative weight of Turks and citizens of former Yugoslavia had increased thanks to recent immigration. The proportion of immigrants from countries with other majority religions, formerly rare in Switzerland, has increased as a consequence. Immigrants from North Africa and Asia have also contributed to this trend. The family component of this immigration is recent, however. Up until 1990, most immigrants were working-age men, but the immigrant sex ratio has since fallen from 205 men per 100 women aged 20-39 to just 114 in 2000. This diversification of the religious landscape has also been accentuated by relatively high fertility among these foreigners after their arrival in Switzerland (Wanner and Fei, 2005).

12Apart from its limited overall presence, the Jewish community is characterized by an ageing process similar to that of the traditional churches, although less pronounced.

II – Data and critical assessment of religious affiliation recorded in censuses and death certificates

13Mortality by religious affiliation over the period 1991-2004 in Switzerland can be described with data from the 1990 and 2000 population censuses, which provide the age-sex structure of sub-populations, and from the deaths registered in the civil records, which also give the cause of death. [2]

14As mentioned at the outset, information on religious affiliation is rare in Europe, but the experience of Swiss data highlights the interest and importance of a critical assessment. In particular, the phenomenon of “religious mobility” between the affiliation reported in the census and that recorded at the time of death is liable to bias the ratio between numerator and denominator. Certain points relating to data quality thus deserve a mention.

15In principle, the information recorded in the census is the most reliable because it is self-reported. While the institutional dimension of religion is losing ground, the sense of individual religious faith is at the centre of Swiss religiosity today (Becci and Bovey, 2004). This information can nonetheless be biased if the “reference person” fills in the questionnaire for all household members, without necessarily consulting them, as is typically the case of parents for their children. Moreover, the mode of administration changed between the two censuses. In 1990, census agents checked that questionnaires had been fully completed by all households, while in 2000 the respondents returned them by mail. This increased the potential sources of bias: according to the Federal Statistical Office, 1.3 questionnaires per 1,000 (around 10,000 in all) [3] were not returned, so information on religion is unknown for these individuals. In addition, incorrect responses by Roman Catholics (who ticked the “Christian Catholic” box) had to be rectified by the Federal Statistical Office with a certain degree of approximation. [4]

16Sources of bias also exist in the reporting of religion at the time of death, since the information recorded on the death certificate is given by a third person. Moreover, it is expected, by convention, to correspond to the choice of funeral ceremony requested by the family. For older adults with no family, the declaration is made by hospital staff or social workers, or by a legal guardian, normally on the basis of a statement made before death by the decedent. Hence, for a variable such as religion, the potential for approximation cannot be totally ruled out.

17Rather than simply relying on assumptions, reporting of religious affiliation can be tested for consistency by comparing the information given in two different sets of records, thanks to an original database available in Switzerland: as part of the Swiss National Cohort project, all death certificates established in the 1990s were coupled with the 1990 population census. After determinist record linkage based on variables common to both databases, [5] a probabilist procedure using these same variables – in addition to information on each individual’s spouse and family structure – was applied (Bopp et al., 2008 [6]). In all, 94.3% of the 629,141 deaths that occurred between 5 December 1990 and 4 December 2000 were matched with a person enumerated on 5 December 1990. Thanks to these matched deaths, the religious “mobility” rate can be estimated by comparing religious affiliation reported in the census with that given at the time of death so as to correct the numerator of mortality rate by religious group.

18Analysis of the mobility rates of individuals enumerated in 1990 and who died up to December 2000 shows a positive relation with the minority status of a given religious affiliation. For 95% of Protestants and Roman Catholics, the information recorded at the time of death is the same as that given in the census. By contrast, there is substantial mobility of affiliation among other Christian communities (72% instability), individuals with no religion (59%) and, to a lesser extent, among Jews (34%) and other non Judeo-Christian religions (32%).

19Table 2 confirms these differences. It presents a logistic regression for the risk of a difference of affiliation between the two sources by sex, marital status and socioeconomic status (estimated for persons in employment by their occupational status and for inactive persons by their educational level) given in the census. Women have a significantly higher risk of religious mobility. Admittedly, they survive to older ages than men and are more often widows, but this is not a valid explanation since the most affected age group is that of women aged 25-59. Religious mobility is also stronger for individuals with a high socioeconomic status or a high educational level. This tallies with a survey of religious practice conducted in Switzerland in 1988-1989 which found that the sense of attachment to a religious institution is weakest among young people and among highly educated individuals (Campiche et al., 1992).

Table 2

Logistic regression of the risk of religious mobility between the census and death by characteristics reported in the 1990 census, matched deaths 1991-2000

Covariate Odds ratio Sex  Male (Ref.) 1.00   Female 1.11 *** Age group Age 25-59 (Ref.) 1.00 Age 60-69 0.80 *** Age 70-79 0.69 *** Age 80-89 0.64 *** 90+ 0.61 *** Marital status Single 1.06 * Married (Ref.) 1.00 Divorced, widowed 1.04 n.s. Occupational status In employment Higher level occupations(a) 1.16 *** Intermediate occupations(b) (Ref.) 1.00   Lower level occupations(c) 0.96 ** Not in employment(d) Higher education 1.19 *** Upper secondary education 0.97 *** Lower secondary education 0.92 *** Religion Protestant 0.78 *** Roman Catholic (Ref.) 1.00   Other Christian denomination 46.09 *** Jewish 9.19 *** Other religion 7.48 ** No religion 23.66 *** Truncated N 489,194 N mobility 49,221 (a) Professional, intellectual and managerial occupations. (b) Skilled clerical and sales workers and self-employed (farmers, craft and service occupations). (c) Skilled and unskilled manual workers. (d) Unemployed, inactive and retired. Significance level: ***: p < 0.0001; **: p < 0.001; *: p < 0.05; x: p < 0.1; n. s.: non-significant. Sources: Linked data (deaths from 5 December 1990 to 4 December 2000; 1990 population census).

Logistic regression of the risk of religious mobility between the census and death by characteristics reported in the 1990 census, matched deaths 1991-2000

20The two religious groups at highest risk of religious mobility are those with no reported religion and “other Christians”. Longitudinal survey data indicate very low religious mobility in Switzerland (Becci and Bovey, 2004). [7] The stability among the majority religions observed in the linked data provides an explanation for this overall result which suggests that mobility can largely be attributed to errors or non-response and to the data recording problems mentioned earlier. The stability of religious mobility rates by time elapsed between the census and death (9% in the first year and 11% in the tenth) confirms this assessment of data quality. For example, for members of “other” Christian communities, the risk of incorrect recording of religious affiliation is high if their community is confused with the Swiss majority churches, and civil registrars may place them in the wrong category. [8] Inconsistencies relating to persons with no religion may also arise at the time of death if a third party reports the decedent’s religious affiliation as that into which he or she was born. By contrast, the religious mobility observed among Jews and non Judeo-Christian religions is difficult to explain. Given the small size and high geographical mobility of the populations concerned, these inconsistencies may be due to the probabilistic method used to match deaths and census data. We will return later to this potential source of bias.

21Bearing in mind these data problems, we chose to use the affiliation recorded in the census, which is less biased because generally reported by the individual him/herself or by the household reference person, and not by a potentially distant third party. The life tables by five-year age groups are based on the mean numbers of deaths registered between 1991 and 2004. The mean numbers by religious affiliation are redistributed by applying “mobility” coefficients estimated by broad age group on the basis of death certificates matched to the 1990 census. In this way, we aim to correct the numerator of the mortality rates so that it can be expressed in terms of the population exposed to risk, estimated as the mean of the population sizes in the two most recent censuses of 1990 and 2000. An approximation of this kind is necessary not only because there are few deaths in the minority groups, but also because of the substantial (and sometimes unstable) migratory flows which prevent any accurate estimation of person-years during the intercensal period. [9] As indicated by the findings presented below, the group of non Judeo-Christian affiliations is particularly concerned by this problem. Their mortality is perhaps over-estimated. There are similarities, however, between the corrected estimates for these migrant groups and the mortality conditions observed in the sending countries.

III – Mortality by religious affiliation

22While illustrating the life expectancy differentials for the period 1991-2004, Figure 2 shows the sensitivity of this indicator to the corrections for religious mobility between the census and the time of death. [10] The estimates for the two main religious groups, representing the overwhelming majority of the Swiss population, are barely affected by the correction coefficient. However, the reporting errors described above produce a strongly over-estimated mean length of life, not only for “other” Christians, but also for individuals with no religious affiliation. The situation is reversed, however, for the non Judeo-Christian religions: the redistribution of certain deaths from this latter group produces an upward adjustment of their largely under-estimated life expectancy.

Figure 2

Life expectancy at birth by religious affiliation, before and after correction for religious mobility, Switzerland, 1991-2004

Life expectancy at birth by religious affiliation, before and after correction for religious mobility, Switzerland, 1991-2004

23The “other” Christian communities nonetheless have longer life expectancies, notably for women (83.6 years after correction), but also for men (76.8 years). They are followed by Roman Catholics (82.5 and 76.5 years, respectively) and Protestants (81.7 and 75.6 years). While the differential between the two majority groups has remained consistently low over the last thirty years, it has reversed since the 1990s (Haug and Wanner, 1998). As we will see below, the increase in premature mortality has probably played a role. At the other extreme, persons with non Judeo-Christian religious affiliations have a shorter life expectancy: 77.4 years for women and 74.3 years for men. This finding should be viewed with great caution, however. Migrant communities are strongly represented in this group, giving rise to numerator-denominator bias. In particular, regular waves of Muslim immigration were recorded in the 1990s with the arrival of refugees from the Balkans, a large share of whom returned to their home country before the 2000 census. Consequently, the mean number of Muslims and Orthodox Christians in the two most recent censuses is not an accurate estimate of the mean population exposed to the risk of dying between 1991 and 2004. The group is also very heterogeneous in terms of origin. Nonetheless, our estimates are slightly higher than the life expectancies observed in the Balkans, which ranged from 67 to 72 years for men and from 76 to 81 years for women in 2000 (Mrdjen and Penev, 2003).

24The patterns of age-specific mortality vary substantially by religious affiliation. Table 3 gives various other indicators taken from the estimated life tables for the period 1991-2004 (after adjustment for religious mobility). Persons in the “other” Christian churches group have the lowest risk of dying at all ages, except in childhood. The difference with respect to the other groups is largest between ages 30 and 64 for men and ages 15 and 79 for women. Persons in this group also have the strongest rectangularization of their survival curve, which indicates – assuming a fixed life span – that while individuals live to older ages, they also die out more rapidly. This concentration of deaths over a narrow age range is the consequence of declining morbidity at younger ages and indicates that an advanced stage in the mortality transition has been reached (Fries, 1980): the interquartile range of ages at death is 15 years for males and 12 years for females in the “other” Christian churches.

Table 3

Level and structure of mortality by religious affiliation, after correction of information given at time of death, Switzerland, 1991-2004

Males Roman Catholic Protestant Other Christian church Other religion(a) No religion Life expectancy at birth (e0) 76.5 75.6 76.8 74.3 75.7 Relative probabilities by age (Q) 1Q0 100 0.0046 97 112 149 122 4Q1 100 0.0017 104 109 151 121 10Q5 100 0.0016 110 106 141 152 15Q15 100 0.0123 119 99 78 135 35Q30 100 0.1378 113 89 108 108 15Q65 100 0.4079 104 99 130 98 10Q80 100 0.7403 103 99 120 107 Rectangularization Q1(b) 70.7 69.8 71.2 73.0 70.3 IQR(c) 15.9 16.3 15.1 10.6 16.0 N mean population 1,564,334 1,233,302 137,011 161,936 373,414 N mean deaths (1991-2004) 12,049 15,567 859 358 2,361 Females Roman Catholic Protestant Other Christian church Other religion(a) No religion Life expectancy at birth (e0) 82.5 81.7 83.6 77.4 79.4 Relative probabilities by age (Q) 1Q0 100 0.0035 112 128 181 126 4Q1 100 0.0012 125 128 183 139 10Q5 100 0.0011 105 98 139 162 15Q15 100 0.0043 127 88 98 164 35Q30 100 0.0719 115 84 160 131 15Q65 100 0.2219 107 92 178 130 10Q80 100 0.5709 105 94 148 136 Rectangularization Q1(b) 78.2 77.2 79.7 72.6 75.5 IQR(c) 13.1 13.4 12.3 13.7 12.4 N mean population 1,638,413 1,367,680 153,922 123,148 309,201 N mean deaths (1991-2004) 12,229 17,038 879 189 1,764 Note: Individuals of unknown religious affiliation are distributed proportionally to the size of the different religious groups. (a) Given the small numbers concerned, Jews are not included in this aggregate overview. (b) Q1 = first quartile of age at death. (c) IQR = interquartile range (Q3-Q1) of age at death. Sources: Population censuses of 1990 and 2000, vital records, 1991-2004.

Level and structure of mortality by religious affiliation, after correction of information given at time of death, Switzerland, 1991-2004

25There is little difference between the mortality patterns of Reformed Protestants and Roman Catholics. However, more deaths occur at working ages among Protestants, women in particular. The sex differential is also higher for the two majority religious groups.

26Individuals reporting no religious affiliation have high childhood mortality and, in contrast to the other religious groups, mortality intensity is highest at young adult ages with, yet again, a more pronounced difference for women. Female excess mortality continues to advanced ages and is only exceeded by persons in the non Judeo-Christian religious group. Consequently, only minimal sex differentials in risk of dying are observed in this non-religious group.

27The differences are even smaller in the non Judeo-Christian group, but mortality levels are higher. As we have seen, this group is characterized by a lower life expectancy and much higher mortality at all ages, with an especially large mortality differential before age 5 and above age 65. The specific fragility of children and older adults is striking, although the numerator-denominator bias is probably highest at these ages. [11] For the older age groups, we fear that the strong rectangularization of the survival curve indicates an under-estimation of the population exposed to risk. Young persons aged 15-24 are an exception, however, with lower mortality than all other age groups. This may be due to a selection effect of migrants in good health.

IV – Impact of religious affiliation on mortality

28The relationship between religion and mortality may also be attributable to a socioeconomic composition effect of the different religious groups. To control for these confounders, we use the linked data described in the second section to adopt a longitudinal individual perspective. Because of the lower record linkage rates at young ages, the multivariate analyses focus exclusively on individuals aged 25 and above. [12] These 4.8 million inhabitants of Switzerland in 1990 can thus be followed up to the year of their death, their emigration (if they are foreigners), or the end of the observation period, i.e. the year 2000. Note that the religious affiliation considered here is that reported by individuals in the census.

29To identify the possible influences of lifestyle on the relationship between religion and mortality, we estimate not only overall mortality risk but also mortality risk for selected causes of death. Figure 3 gives the probabilities for different age groups of dying from various causes.

Figure 3

Probabilities of dying from a specific cause for different age groups (taken from a multiple decrement table), population living in Switzerland, 1991-2004

Probabilities of dying from a specific cause for different age groups (taken from a multiple decrement table), population living in Switzerland, 1991-2004

30Death from external causes is the leading cause of death among young adults (accounting for 49% of male mortality and 31% of female mortality), notably suicide (around 15% for both sexes). Cardiovascular diseases and cancers predominate at older ages. While the probability that death is due to cancer – especially lung cancer among men – is highest at ages 45-64 (39% for men and 54% for women), heart disease takes a steadily heavier toll as age increases (10% of mortality at ages 25-44 and 43% beyond age 64). Digestive diseases (including alcohol-induced cirrhosis) account for around 5% of mortality risk and are highest at intermediate ages.

31Table 4 shows the results of discrete-time survival analyses of the risk of dying of any cause during the ten years following the 1990 census. The results by cause of death are given in the Appendix Tables. The logistic regressions estimate the specific impact of religious affiliation on mortality after controlling for the effects of other individual characteristics (Allison, 2001).

Table 4

Logistic regressions for the risk of dying by age reached, Switzerland, 1991-2000

Covariate Odds ratio Age 25-44 Age 45-64 65+ Age group       25-29 (Ref.) 1.00       30-34 1.14 ***     35-39 1.31 ns     40-44 2.01 ***     45-49 (Ref.) 1.00     50-54 1.57 ***   55-59   2.41 ***   60-64   3.58 ***   65-79 (Ref.)   1.00   80+   4.39 *** Sex  Male 2.46 *** 2.79 *** 1.92 *** Female (Ref.) 1.00   1.00   1.00   Marital status       Single 1.66 ** 1.87 *** 1.38 *** Married (Ref.) 1.00   1.00   1.00   Divorced/widowed 2.39 *** 1.70 *** 1.39 *** Occupational status       In employment Higher level occupations(a) (Ref.) 1.00   1.00     Intermediate occupations(b) 1.28 *** 1.32 ***   Lower level occupations(c) 1.55 * 1.53 ***   Not in employment(d) Higher education 1.53 ns 2.02 *** 0.70 *** Upper secondary education 1.96 *** 2.37 *** 0.83 *** Lower secondary education 2.97 *** 2.68 *** 1.00   No qualifications     0.94 *** Nationality       Swiss (Ref.) 1.00   1.00   1.00   European 0.68 *** 0.83 ns 0.88 ns Former Yugoslavian, Turkish and other 0.62 *** 0.71 *** 0.78 *** Religion       Protestant 1.09 *** 1.06 *** 1.07 *** Roman Catholic (Ref.) 1.00   1.00   1.00   Other Christian denomination 0.79 *** 0.78 * 0.87 x Jewish 1.16 ns 0.60 ** 0.82 * Other religion 0.63 *** 0.66 *** 0.65 *** No religion 1.21 *** 1.08 *** 0.98 *** N deaths 18,473 75,716 455,516 N truncated person-years 16,673,423 16,607,255 10,134,775 BIC – Differential with respect to model without religious affiliation – 138   – 261   – 1,002   (a) Professional, intellectual and managerial occupations. (b) Skilled clerical and sales workers and self-employed (farmers, craft and service occupations). (c) Skilled and unskilled manual workers. (d) Unemployed, inactive and retired. Significance level: ***: p < 0.0001; **: p < 0.001; *: p < 0.05 ; x: p < 0.1; n.s.: non-significant. Sources: Linked data (deaths from 5 December 1990 to 4 December 2000; 1990 population census).

Logistic regressions for the risk of dying by age reached, Switzerland, 1991-2000

32The risk differentials by sex or by increasing age are clearly visible and become stronger after controlling for marital status and occupational status. Divorced, widowed and single men have a higher risk of dying than married men, especially among the population aged below 65. Occupational status is negatively associated with risk of dying at all ages. Mortality among professional, intellectual and managerial occupations is lower than that of both skilled and unskilled workers. Although inactive persons follow this same mortality gradient by educational level, their risks are markedly higher than for persons with equivalent characteristics who are in employment.

33An initially surprising finding, confirmed by other analyses (Wanner et al., 2000), is a lower risk of death for foreigners, especially non-Europeans. Immigration from outside Europe probably selects individuals in better health, and these migrants increasingly tend to have high levels of education. Moreover, some have returned to their home country, as was the case for many Portuguese and Spanish in the 1990s. While the linkage procedure seeks to limit this bias, it cannot be eliminated entirely. In these cases, the actual survival status remains unknown and the risk of dying may be under-estimated. Note, however, the stability of these confounding effects on cause-specific mortality given in the Appendix Tables.

34After controlling for these socioeconomic profiles, the impact of religious affiliation remains significant. We observe that the specific effect of religious affiliation, versus no affiliation – like the differences between religious groups – are very similar for both sexes (analyses not presented here). The differentials by affiliation are stronger among young adults and decrease in the older age groups. The small differential between Roman Catholics and Protestants and the importance of minority religions, notably their impact on mortality or longevity, are the key findings of this analysis.

35Regarding the first point, the model does indicate a certain excess mortality among Protestants compared to Roman Catholics at all ages, but it is very small and varies with age. Young people have a higher risk of death from external causes (OR = 1.09; p < 0.05), especially suicide (OR = 1.14; p < 0.1) (Appendix Table A.1.), while persons aged over 65 are even more strongly exposed to suicide risk than the reference group (OR = 1.64; p < 0.05) (Appendix Table A.3). Older persons are also at higher risk of dying from cardiovascular diseases (OR = 1.08; p < 0.0001) and digestive diseases (OR = 1.06; p < 0.05).

36The minority groups, for their part, which are small but have been growing in recent decades, have various specific traits and we will discuss our estimates in the light of these interesting features.

37First, individuals aged 25-44 who reported no religious affiliation at the time of the 1990 census have a higher risk of dying (OR = 1.21; p < 0.0001) than the reference group of Roman Catholics. While the phenomenon disappears with age, the decrease in excess mortality does not concern all causes of death. Individuals aged 65 and over have a higher risk of dying from lung cancer (OR = 1.25; p < 0.0001) or alcohol-induced cirrhosis (OR = 1.20; p < 0.05). There are also large differentials for deaths from external causes. In fact, introducing religious affiliation into the model for young adults improves the prediction [13] solely for mortality from external causes. Excess suicide mortality of individuals with no religion is observed at all ages but is highest for young people (OR = 1.33; p < 0.001) and for persons over age 64 (OR = 3.05; p < 0.0001), for whom the most frequent method is assisted poisoning (Spoerri et al., forthcoming). Assisted suicide, managed by specialist organizations since the 1980s in Switzerland, is tolerated under two conditions: the candidate for suicide must be mentally competent and the person who assists him or her must be free of personal interest (Fisher et al., 2008). Note, however, that while these differentials with regard to suicide appear to be real, they are probably over-estimated due to the stigmatization of suicide by certain religious groups. In such cases, family doctors may tend to record a different cause of death.

38A second sub-population stands out, positively this time, with a lower risk of dying at all ages for members of “other” Christian communities. Let us examine the reversed profile of this lower mortality with respect to Protestants and persons with no religion. Young people from “other” Christian churches have much lower mortality from external causes (OR = 0.64; p < 0.0001) and especially from suicide (OR = 0.56; p < 0.0001), although this advantage decreases slightly with age. At ages 45-64, we also observe a lower risk of dying from cardiovascular disease (OR = 0.78; p < 0.1) or cancer (OR = 0.77; p < 0.1) – particularly lung cancer (OR = 0.58; p < 0.1; and OR = 0.68; p < 0.05 beyond age 64) – highlighting the beneficial effects of a healthy lifestyle (Appendix Table A.2). Jews have a very similar profile, except with regard to excess suicide mortality among young people.

39Last, the mortality risk of the third expanding group, that of non Judeo-Christian religions, appears to be even lower than that of the other groups. Note, however, the bias introduced by migration, which strongly influences the results for young adults. [14] Despite these reservations, we nonetheless see a low mortality profile similar to that of the “other” Christian churches. Exposure to death from external causes is low for young people (OR = 0.64; p < 0.05), while individuals aged 45-64 enjoy low mortality from cardiovascular diseases (OR = 0.71; p < 0.05) and cancer (OR = 0.64; p < 0.001), the two leading causes of death in Switzerland. As is the case for the minority Christian churches, this might point to a more healthy lifestyle which has positive consequences in old age. We will return to this interpretation in the final discussion.

40No final conclusion can be drawn regarding this latter group, however, since two different sets of bias affect the analysis of their mortality. We have already mentioned the difficulties involved in estimating the number of people exposed to risk who are affiliated to the other religions, notably because of recent migration and residential instability within these populations. Their life expectancy is therefore under-estimated. Regarding individual data, the linkage rate – i.e. the share of observed deaths matched to enumerated individuals – is inversely proportional to the relative size of the sub-population concerned. The problem is thus the reverse of that encountered with life tables, i.e. an under-estimation of deaths occurring in the population enumerated in the 1990 census and exposed to risk during the ten observation years. There has been little discussion of the effect on statistical results of biases linked to unmatched deaths (O’Reilly et al., 2008) but the question deserves further investigation which is beyond the scope of this study. Simulating probabilistic linkage of data previously matched by deterministic linkage would, for example, enable systematic comparison of results.

V – Discussion and conclusions

41We have used datasets unique to Switzerland to analyse mortality by religious affiliation. Over the last forty years, the religious landscape in Switzerland has diversified. The majority religions – i.e. the Roman Catholic and Reformed Protestant Churches – have lost their quasi-monopolistic position in the face of two mounting trends. First, the traditional religious institutions are losing ground as Swiss society becomes increasingly secularized: a growing share of the population report not having, or no longer having a religious affiliation. Second, certain former minority religions (other Christian churches and non Judeo-Christian religions) are expanding. The quest for a more intense religious experience, along with immigration from the Balkans and Turkey, are contributing to this diversification. In this context, religion remains an important cultural variable, contributing to the primary socialization of individuals. Explicit or implicit religious references are liable to determine not only education but also lifestyles in adulthood. They thus have a potential impact on health, especially at the current stage of the epidemiological transition, in which the positive or negative accumulation of lifestyle factors over the life course are key determinants of longevity.

42The analysis of mortality in the 1990s shows a certain diversity among the various religious groups, even though affiliation is difficult to measure. Not only is it liable to change over time, but it also varies by the data collection method used, i.e. self-reporting or declaration by a third party at the time of death. By linking the deaths registered in the 1990s with the individuals enumerated at the start of the period, the enigmatic phenomenon of religious “mobility” between the census and time of death was brought to light. This enabled us to harmonize the “religious affiliation” criterion between the numerator and denominator of mortality rates so that this mobility could be taken into account when estimating aggregate cross-sectional life tables. After these adjustments, the differences in life expectancy do not exceed two years for the groups whose estimates are robust, i.e. those with no religion or with a Christian affiliation. Compared with the other factors of mortality in Switzerland (sex, level of education and marital status), the survival differentials by religion are smaller overall (see Spoerri et al., 2006; Schumacher and Vilpert, 2010). They nonetheless persist after controlling for individual profiles and do not differ by sex.

43Individuals with no religious affiliation have the highest mortality, but this disadvantage decreases with age. Relative to members of religious groups, they have higher risks of dying of one of the main behavioural causes of death: lung cancer, alcohol-induced cirrhosis or death from external causes, especially suicide. Moreover, excess mortality by suicide increases with age. By contrast, there is no excess cardiovascular mortality in this group, perhaps because of its high socioeconomic profile. The members of the two majority religious groups – Roman Catholics and Reformed Protestants – have an intermediate level of mortality. While the differences are very small, they highlight a trend reversal with respect to the past, with Reformed Protestants having higher mortality than Roman Catholics, especially at working ages. Their excess mortality profile is similar to that of persons with no religion, with a higher risk of dying from cardiovascular disease. Last, the “other” Christian churches enjoy the highest life expectancy thanks to lower risks of dying from the main causes of death in Switzerland, namely cardiovascular disease and cancer. Accidents and suicides are infrequent in this group, so premature mortality is low. Given the small numbers concerned and the contradictory estimates for the other minority groups, particular caution is required for the interpretation of results: individuals with Jewish affiliation appear to be protected in the same way, but to a lesser extent, while members of non Judeo-Christian religions also have lower risks of dying from the main behavioural causes of death.

44It is clear that while our study is based on an exceptional and exhaustive set of data, it nonetheless has its limits. We have no direct measure of religious practice. Yet research conducted in the United States shows that religiosity has a positive effect on life expectancy. These studies use a wide range of concepts, distinguishing between public religiosity (attendance at religious services) and private religion, while controlling for socioeconomic profiles as well as behaviours and health status (objective or subjective). The literature on these questions considers the only robust relation to be the beneficial influence of public religious observance on survival, which remains unexplained by the previously mentioned confounding variables (Powell et al., 2003; McCullough et al., 2000). A recent Danish study confirms this observation, although the protective effect is less pronounced in the secularized European context due to lower attendance of religious services (la Cour et al., 2006). Indeed, social and institutional integration in a religious community provides the most vulnerable individuals with a valuable system of support. Powell et al. (2003) even propose an active reading of this phenomenon, seeing public religious practice as a form of self-fulfilment associated with a social role and acts of mutual support rather than as a source of passive well-being linked to a reassuring environment. More generally, there may be stronger social control over lifestyles among members of a spiritually homogeneous community. An earlier study by Idler and Kasl (1992) confirms and qualifies this thesis. Physical activity associated with religious ceremonies may have beneficial effects in old age by increasing physical health, enhancing the capacity to look after oneself without assistance or to carry out strenuous physical tasks. Private religious practice, by contrast, is associated with a higher risk of disability linked to a withdrawal from society. In other words, weekly, seasonal or annual religious ceremonies and celebrations are a potential factor in prolonged survival.

45Musick (1996) focuses on the spiritual dimension and its impact on the acceptance of suffering and the subjective state of health. Private religiosity is a stronger determinant than public religiosity for this aspect, except among individuals with major physical health problems. Yet one of the “ironies of the human condition” (Ryff et al., 2004) is that the individuals in the poorest health tend also to be the most religious (Musick, 1996). So the effect of religion is hidden, or a negative effect emerges if we do not control for this interaction, while in reality, religious practice enhances physical health and perhaps prolongs life. More generally, spirituality moderates emotional stress and its physiological impacts on health. The influence of religion often begins in childhood, serving to stabilize the family environment and limit risk behaviours while favouring the adoption of a healthy lifestyle that may be maintained over the entire life course. These indirect relations between religion and mortality are complex as they operate simultaneously on the psychological, social and behavioural levels (Koenig, 2001). Although our study covers the entire population living in Switzerland, we cannot specify these indirect influences and can only assume their existence.

46The fact that highest mortality levels are observed for persons with no religion in Switzerland tends to confirm the existence of a mortality gradient by religiosity. Moreover, the causes of their excess mortality at working ages lend support to the behavioural explanation. While individuals with no religious affiliation have a higher mean level of education and occupational status than individuals who belong to a religious group (Bovay and Broquet, 2004), this advantage is not fully played out in terms of life expectancy. Their lifestyles appear to increase their risk of premature death, or in any case their exposure to cumulative disadvantages that make them vulnerable in old age. The absence of religious belief also seems to be associated with the social acceptance – and hence the individual acceptance – of suicide, especially at advanced ages. Spoerri et al. (forthcoming) have observed among the Swiss elderly population a strong gradient of assisted suicide, with Roman Catholics having the lowest risk exposure and non-religious persons the highest.

47In the absence of direct measures of individual religiosity, the sociological surveys available in Switzerland show differentials by religious affiliation. This interpretation of our results enables us to generalize the hypothesis of a gradient of mortality by degree of religiosity. Persons affiliated to the majority churches have the lowest level of public religiosity. Slightly more Protestants than Roman Catholics are thinking about leaving their church (25% and 18%, respectively), feel no need to belong to a religious community (around 50% and 40%, respectively), or say that religion is not important for them (close to 60% and 45%) (Campiche et al., 1992). While this individualism can be explained by the doctrines of Protestantism, both communities place higher value upon the socionormative – or even traditional or customary – ties with their church. Only one-third consider themselves to be practicing Christians (Campiche et al., 1992) and members only occasionally attend church services (Becci et Bovey, 2004). Their excess mortality compared to the “other” Christian churches could thus be explained by this disinvestment in religion, public religion especially.

48Among members of “other” Christian churches, the majority are affiliated to the evangelical churches, and their religiosity is doubtless very strong. These communities have split off from the mainstream Protestants and Roman Catholics in Switzerland (with the exception of the most recent Pentacostal group) and practice a militant brand of Protestantism aiming to attract new teenage and adult converts into the fold (Favre, 2006). The survey of these communities by Olivier Favre showed “abnormally high” levels of both private and public practice by more than 90% of believers (Favre, 2006, p. 230), who felt a strong sense of satisfaction with life. Alongside their intense religiosity, members also benefit from a dense social network. Social cohesion is strengthened by the exclusive nature of the community, the active participation of members and a high level of voluntary activity, which was confirmed by the 2000 census (Bovay and Broquet, 2004). Members are particularly assiduous churgoers, generating strong social capital and active social control (Favre, 2006). Members of this religious group thus appear to benefit cumulatively from the positive influences of religion on survival. It is doubtless by avoiding risk behaviours that their premature mortality is reduced. Members of “other” Christian churches are also less exposed to degenerative diseases since, at the current stage of the epidemiological transition, the benefits of healthy behaviours build up over the life course, with a “payback” at advanced ages in terms of longevity (Alter, 2004). The results obtained appear to be robust, and at the very least consistent.

49Little is known, however, regarding the other religious groups in Switzerland, notably the Muslim community. One may assume that religiosity is stronger than among the more secularized Protestants and Roman Catholics. A majority of migrants from Muslim countries reported being religious or very religious in 2004 (at least 60%; Rommel et al., 2006). However, the different religious traditions in the home country do not appear to determine the religiosity of migrants to Switzerland: religiosity is strongest among the Tamul minority, followed by migrants from the Balkans and, last, from Turkey. Among Muslims, a highly heterogeneous group, practice inevitably varies according to the degree of adherence to the five pillars of Islam and the various rites of passage. [15] Here too, qualitative interviews reveal the existence of non-negligible social capital and social control, notably with “somewhat normative attitudes towards coreligionists” (Schneuwly-Purdie, 2006, p. 327). Dietary laws forbidding the consumption of alcohol and pork appear to be followed by a small majority of migrants, while the prevalence of smoking is comparable to that observed in the native Swiss population (Rommel et al., 2006). As is also the case for Buddhists, Hindus and other non Judeo-Christian religions, these moral codes and lifestyles are explicitly intended to keep the faithful in good health. As a reverse image of what was observed for individuals with no religious affiliation, the socioeconomic disadvantages influencing the survival of individuals with non Judeo-Christian affiliations may be attenuated by the beneficial effects of their religiosity.

50Our findings show similarities with the situation in Northern Ireland, although its religious communities are less diverse (O’Reilly and Rosato, 2008). The non-religious Irish population was not analysed, but here too, members of evangelical churches are also least exposed to the risk of dying, while members of the more traditional churches – Anglican and Roman Catholic – have higher mortality levels. Analysis by cause of death shows that lifestyle has a similar impact to that observed in Switzerland – even though religious affiliation is associated with socieconomic category – a fact which explains some of the differentials among women and all of them among men.

51In Switzerland, while socioeconomic status and marital status are still the main factors explaining differences in life expectancy, the analysis of mortality differences by religious affiliation is also useful for a prospective approach to the Swiss multicultural context. Indeed, there is every reason to believe that the religious landscape will continue to diversify in response to demographic dynamics that vary from one religious community to another. While emerging groups are characterized by a relatively young age structure and fertility above the national average, the congregations of the majority churches are ageing. Differences in longevity by religious affiliation will thus have a progressive impact on life expectancy trends in Switzerland.