lable at ScienceDirect

Social Science & Medicine 69 (2009) 519–528

Contents lists avai

Social Science & Medicine

journal homepage: www.elsevier .com/locate/socscimed

Association between attendance at religious services and self-reportedhealth in 22 European countriesq

Amanda Nicholson a,*, Richard Rose b, Martin Bobak a

a International Centre for Health and Society, Department of Epidemiology and Public Health, University College London, United Kingdomb Centre for the Study of Public Policy, University of Aberdeen, United Kingdom

Keywords:EuropeSelf-rated healthAttendanceReligious involvementReligion

q This study was funded by a small research grantSocial Research Council (ESRC): RES-000-22-2429.

* Corresponding author. Division of Primary Care anSussex Medical School, Brighton BN1 9PH, United King

E-mail address: a.c.nicholson@bsms.ac.uk (A. Nich

0277-9536/$ – see front matter � 2009 Elsevier Ltd.doi:10.1016/j.socscimed.2009.06.024

a b s t r a c t

There are consistent reports of protective associations between attendance at religious services andbetter self-rated health but existing data rarely consider the social or individual context of religiousbehaviour. This paper investigates whether attendance at religious services is associated with better self-rated health in diverse countries across Europe. It also explores whether the association varies witheither individual-level (gender, educational, social contact) or country-level characteristics (overall levelof religious practice, corruption, GDP). Cross-sectional data from round 2 of the European Social Surveywere used and 18,328 men and 21,373 women from 22 European countries were included in multilevelanalyses, with country as higher level.

Compared to men who attended religious services at least once a week, men who never attended werealmost twice as likely to describe their health as poor, with an age and education adjusted odds ratio of1.83 [95% CI, 1.49–2.26]. A similar but weaker effect was seen in women, with an age and educationadjusted odds ratio of 1.38 [1.19–1.61]. The associations were reduced only marginally in men bycontrolling for health status, social contact and country-level variables, but weakened in women. Therelationships were stronger in people with longstanding illness, less than university education and inmore affluent countries with lower levels of corruption and higher levels of religious belief.

These analyses confirm that an association between less frequent attendance at religious services andpoor health exists across Europe, but emphasise the importance of taking individual and contextualfactors into account. It remains unclear to what extent the observed associations reflect reverse causalityor are due to differing perceptions of health.

� 2009 Elsevier Ltd. All rights reserved.

Introduction

Decades of research in Britain and elsewhere have demonstratedthat there are social as well as biomedical causes of ill-health(Marmot, 2004) and a social gradient in health is perhaps the mostrobust and consistent finding in epidemiology. However, noteveryone in stressful or deprived circumstances suffers ill-health.Exploring the causes of resilience, that is, characteristics enablingindividuals to withstand adverse social conditions injurious to health,is now a priority in medical sociology (Jammer & Stokols, 2000).

The existing explanatory models for social inequalities in healthfocus on materialist causes, health behaviours and psycho-social

from the UK Economic and

d Public Health, Brighton anddom. Tel.: þ44 1273 641974.olson).

All rights reserved.

factors. The concept of perceived relative deprivation is importantin the psycho-social model in explaining social gradients in healthin affluent societies, where few are in absolute poverty (Marmot &Wilkinson, 2001) (Bartley, 2004). However, these models haveignored the role of personal belief in interpreting the externalworld. An individual’s beliefs and value systems will influence his/her reaction to social situations but the role of religious involve-ment in the context of social deprivation and health has not beensystematically investigated.

In recent years religion has been theorized as a potential influ-ence on health. There are extensive data linking religious involve-ment to better health. Meta-analyses have shown small protectiveassociations with reduced mortality in those with more religiousinvolvement (Chida, Steptoe, & Powell, 2009; McCullough, Hoyt,Larson, Koenig, & Thoresen, 2000). In studies in which differentdimensions of religious involvement have been studied, associa-tions are stronger for religious attendance than for other dimen-sions such as spirituality or prayer (McCullough et al., 2000). The

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528520

mechanisms underlying these associations are unclear. Betterhealth behaviours are potential confounders and there is ongoingdebate about whether the observed associations are independentof these variables (McCullough, Hoyt, & Larson, 2001; Sloan &Bagiella, 2001). The role of social contact and networks in anyimpact on health is also debated and there is clear correspondencewith research on social networks and their impact on health(Berkman & Glass, 2000). The contribution of reverse causality tothe observed associations is another explanation, with ill-healthpotentially preventing religious involvement. Data using self-ratedhealth as an endpoint are often cross-sectional but again there areconsistent reports of protective associations between attendance atreligious services and better self-rated health (Arredondo, Elder,Ayala, & Campbell, 2005; Hyyppa & Maki, 2001; Krause, Ingersoll-Dayton, Liang, & Sugisawa, 1999; Maselko & Kubzansky, 2006;McCullough & Laurenceau, 2005; Veenstra, 2000).

Research into the health impacts of religious involvement and ofsocial influences has been conducted largely separately. There isa literature on race differences in the religion-health association(Schieman, Pudrovska, Pearlin, & Ellison, 2006) and the health effectof religious involvement at a neighbourhood level has been studied(Jaffe, Eisenbach, Neumark, & Manor, 2005) but variations acrosssocial status have not been a focus of research. With some exceptions(la Cour, Avlund, & Schultz-Larsen, 2006), much of the data on theeffect of religious involvement on both mortality and self-ratedhealth comes from populations in North America. The distinctivereligious context of the US, with the confounding of race, socio-economic deprivation and Protestant religiosity, cautions againstgeneralizing such findings to European or OECD countries generally.

Stark and Bainbridge (Stark & Bainbridge, 1996) argue that theextent to which a specific religion is dominant in a society affectsindividuals. Insofar as a country’s official religion influences culturalvalues and beliefs internalized by individuals (Norris & Inglehart,2004), then the influence of religion on health may vary betweencountries. The overall degree of secularisation in a country is anotherpotential influence on the individual-level association betweenattendance and health. We could predict that the impact of religiousinvolvement on health would be less in more secular societies(Bruce, 2003; Davies, 2000; Martin, 2005). Similarly in wealthiercountries, where health is better overall, the support offered byreligion may be less important. In more unequal societies, religiousinvolvement may make people indifferent to relative deprivationand material inequalities, and hence religion may have a moreimportant role in protecting health. Corruption can be associatedwith poor health insofar as it is a sign of bad governance that in turngenerates stressful social conditions. Increased corruption ina society could increase reliance on religious support or generatemistrust of all authority, including religious institutions. Unfortu-nately, the specific theoretical background in this area is sparse, soprediction rapidly becomes conjecture. This paper represents aninitial attempt to study these influences in a systematic manner.

Theories of resilience take into account how individuals maintaincompetent functioning in the face of stressors (van Breda, 2001).They suggest that individuals who are vulnerable due to poor socio-economic resources would benefit more from religious involve-ment. Under this hypothesis, individuals with low human capital(such as the elderly or less educated), low economic capital or poorsocial networks (those who are otherwise socially isolated) mayshow stronger associations between health and religious variables.Religious belief as a source of mental resilience been examined inrelation to specific stressors e.g. discrimination (Bierman, 2006), ill-health (Wink, Dillon, & Larsen, 2005), substance abuse (Pardini,Plante, Sherman, & Stump, 2000), stroke (Giaquinto, Spiridigliozzi, &Caracciolo, 2007) but not in relation to social deprivation. Straw-bridge et al. showed that religious involvement buffered the effect of

some stressors on depression but exacerbated the effect of others(Strawbridge, Shema, Cohen, Roberts, & Kaplan, 1998).

The purpose of this article is to test systematically under whatconditions and to what extent religious involvement, heremeasured as church attendance, can influence self-assessed healthby analyzing individual-level data from the 2004–2005 round ofthe European Social Survey. The availability of data from respon-dents in a wide variety of religious contexts, from secular Scandi-navia to countries with strong religious traditions, such as Irelandand Poland, makes it possible to test for differences in varyingcontexts. In addition, individual-level data makes it possible tocontrol for differences in socio-economic conditions of people whohave the same religious practices. The research questionsaddressed were: 1. Is increased attendance at religious servicesassociated with better self-rated health across Europe? 2. Does theassociation vary with individual characteristics such as educationand gender? 3. Does the association vary across countries, and, if so,what influences the association between countries?

Methods

Study populations: European social survey

This paper uses data from round 2 of the European Social survey.The Norwegian Social Science Data Services (NSD) holds the dataarchive and are distributors of the ESS data. Full details of the ESSmethodology are available elsewhere (Jowell, 2005) but in briefdata were collected during 2004–2005 in face-to-face interviewsfrom a probability sample of all adults aged 15 years and over in 26countries. The questionnaire covered domains including subjectivewell-being, health and socio-economic profile. The response rate inindividual countries varied from 44 to 70%.

Variables used

Self-rated health was the dependent variable for these analyses.This was asked as ‘‘How is your health in general? Would you say itis very good, good, fair, bad or very bad’’ and was recoded toa dichotomous variable with bad and very bad health classed aspoor health. The SRH scale was used dichotomously rather than asa continuous scale because much of the published validation workuses the scale either as discrete values or in dichotomised form(Idler & Benyamini, 1997; Singh-Manoux et al., 2008; Wanname-thee & Shaper, 1991). As a continuous scale it is less clear whata unit increase represents in objective health terms.

The frequency of attendance at religious services was asked inthe question ‘‘Apart from special occasions such as weddings andfunerals, about how often do you attend religious services nowa-days?’’ The response levels were recoded to: regular (every day(1%), more than once a week (3%), at least once a week (13%));infrequent (at least once a month (10%)); sporadic (only on holydays (20%), less often (20%)); and never (32%). Education wasassessed using the recoded ESS variable, primary, secondary,university. Because of differences in currencies, conversion ratesand prices across Europe, education was used as the primarymeasure of socio-economic status as well as a measure of humancapital. Respondents aged less than 18 years were not included inanalyses and those aged under 30 years who reported being in full-time education were recoded as having university level education(2011 individuals were recoded this way in the full ESS dataset withparticipants aged less than 18 years 45,763). Other individual-levelvariables included in analyses related to social contact: maritalstatus (not married/cohabiting), frequency of social contact (takingpart in social activities less than once a month) and not havingsomeone to discuss personal matters with. A measure of social

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528 521

isolation was created including those who were not married orcohabiting and had no-one to discuss personal matters with.Frequency of social contact was not included in the summarymeasure due to potential overlap with attendance at services.Additional health variables included limitation from longstandingillness, assessed by the question ‘‘Are you hampered in your dailyactivities in any way by any longstanding illness, or disability,infirmity or mental health problem?’’ with response levels: No, notat all; Yes, to some extent; and Yes, a lot.

Country-level variables were added into the data from othersources (full details are given in the web references list after mainreferences). GDP per capita in PPS in 2003 (from the EUROSTAT NewCronos dataset, adjusted so that 100 is the European average) wasused along with the Gini index of inequality in 2003 (from the WorldInequality Database, range 0–100, with higher score indicating moreinequality). A corruption score was created based on using thecorruption perception index for 2003 from Transparency Interna-tional. The index defines corruption as the abuse of public office forprivate gain and measures the degree to which corruption isperceived to exist among a country’s public officials and politicians.It is a composite index, drawing on 14 polls and surveys from 12independent institutions, which gathered the opinions of busi-nesspeople and country analysts. The index was reversed in theseanalyses so that highest score (10) indicated worst corruption(corruption¼ 10-transparency index). A measure of overall religiousbehaviour in each country was created using the religious variablesin the ESS dataset. In addition to frequency of attendance at religiousservices (original 7 point scale), questions were included onfrequency of private prayer (7 point scale) and self-assessed reli-giousness (0–10). These were all scored so that increasing scoreindicated less religious behaviour. Country-level means for each ofthese three questions were calculated within sexes and summed tocreate a secularity index score for men and women in each country,with a potential range 0–24. All country- level variables were usedboth as continuous variables and also quartiled in analyses. Thesecularity index was quartiled within sexes.

Of the 26 countries included in ESS round 2, four countries werenot included in these analyses. In the United Kingdom the ques-tions on education level differed from those in other countries andwere not considered as compatible. The Italian sampling frame wasnot approved by ESS. No GINI data were available for Iceland andUkraine was not included in Eurostat GDP data.

Statistical analysis

All analyses were performed separately for men and womenusing in Stata version 9 (Statacorp LP, Texas). Men and women wereanalysed separately because, due to sex differences in morbidityreporting and in the extent of religious involvement, associationsmay differ between sexes. Analyses were performed within countryinitially to describe the health and attendance variables and then toassess the relationship between attendance and health in eachcountry. When countries were analysed together, multilevel logisticregression models (xtlogit command) were used with a randomintercept (country as higher level variable) and self-rated health asthe dependent, dichotomous variable. Successive models werecreated with the first model including attendance, age and educa-tion as independent covariates, then models adjusting for health,social and context variables in turn, and finally a combined fully-adjusted model. Individual-level variations in the associations weretested by stratification by the third variable (such as educationallevel) and interactions tested by adding cross-product variable andcomparing significance of models using the likelihood ratio tests.Mean proportions of poor self-rated health by attendance withinstrata were estimated using the adjust command following the

regression with attendance as a categorical variable. Country-level-variations were similarly tested by stratification by country-levelvariables and interaction testing in the multilevel logistic regres-sion models but also by using the collapse and statsby commands inStata to produce country-level mean variables (including effect sizefor attendance).

Models with random slope as well as random intercept wereattempted both within Stata 10 (xtmelogit command) and usingMLwiN (Institute of Education, London) but it was not possible to fitthese models for all analyses in men, probably due to limitednumber of countries (22) and small numbers within countries.

Results

Descriptive data

Descriptive data on the main variables by sex and country aresummarised in Table 1. The prevalence of poor self-rated health washigher in women than men and showed substantial variationbetween countries, ranging from 2% to 3% for men and womenrespectively in Ireland to 16% and 20% in Hungary. The prevalenceof poor self-rated health increased with age in both sexes and washigher in respondents with lower educational qualifications. Thepresence of limitation from longstanding illness showed similar,expected trends with age and education (results not shown).

The frequency of attendance at religious services also showedmarked variations across countries with the percentage reportingregular attendance ranging from 3% to 63%. Attendance increasedwith age in all countries (results not shown). The associationbetween attendance and education varied across countries. In menthere was a strong interaction between education and secularityquartiles. Less educated men were more likely to attend services inless secular countries but in the most secular countries the trend wasabolished or even reversed (p for interaction term between educa-tion and secularity< 0.001). In women the pattern was less clear inthe least secular countries (due largely due to low attendance in lesseducated in women in Turkey) but the other 3 tertiles of secularityindex showed a pattern of decreasing influence of education onattendance with increasing secularity similar to that seen in men (pfor interaction term¼ 0.02). The association between worse self-rated health and low education also weakened in more secularcountries (p for interaction term between education and secularityindex quartile was< 0.001 in both sexes). The relationship ofincreasing poor health and increasing attendance at religiousservices with both increasing age and low education suggest thatthese variables act as important negative confounders in the rela-tionship between not attending religious services and poor health.

Association between poor self-rated health and attendance

Analyses in individual countries showed associations betweenless attendance at religious services and poor SRH in nearly allcountries (Table 1). The odds ratio for attendance as a linear vari-able (controlled for age and education) in men varied from 2.39 inGreece to 0.96 in Hungary and was significant at the p� 0.05 levelin 4 countries. In women, effect sizes ranged from 0.78 in Estonia(significant at 0.05 level) to 1.58 in Sweden with 5 countriesshowing associations between less attendance and poor healthsignificant at the p <¼ 0.05 level.

Table 2 shows the results of multilevel logistic regressionmodels, with a random intercept and country as higher level vari-able. Less attendance was associated with poor SRH in men andwomen across Europe, although the effect was stronger in men(interaction term for sex and attendance, p< 0.001). Men whonever attended religious services had an 80% increase in reporting

Table 1Descriptive data of sample in 22 European countries.

Code Country Responserate (%)

N (mean age in yrs) Poor SRH (%) Attend religious services Quartile of Odds ratio for poor SRH (per unit decrease in attendance)

M F M F regularly (%) never (%) Secularity corruption GDP M F

M F M F M F

A Austria 62 927 (43.9) 1100 (45.9) 3.7 5.6 15.2 15.8 30.6 25.1 2 2 2 3 1.36 * 1.27*B Belgium 61 829 (45.7) 887(47.4) 2.9 6.1 8.9 10.5 57.1 50.6 3 3 2 3 1.10 1.13Cz Czech Republic 55 1175 (48.7) 1391 (49.6) 11.5 12.5 6.6 9.3 63.5 53.3 4 4 4 1 1.20 1.02Dk Denmark 65 678 (48.0) 718 (48.0) 5.3 4.5 2.5 3.6 40.3 25.8 3 4 1 3 1.18 1.57Est Estonia 79 747 (46.8) 1099 (50.4) 13.4 16.7 1.7 5.4 41.6 24.8 4 4 3 1 1.18 0.76**Fin Finland 71 907 (47.7) 1007 (49.8) 4.5 5.9 2.9 6.1 28.0 18.6 2 2 1 2 1.90** 1.71**Fr France 44 814 (49.5) 942 (49.6) 7.1 8.8 4.9 8.7 57.9 45.0 4 4 3 2 1.14 0.98D Germany 53 1276 (47.6) 1360 (48.9) 9.6 11.0 6.4 10.6 48.7 38.5 4 3 2 3 1.15 1.09Gr Greece 79 1013 (50.7) 1312 (50.6) 3.6 8.2 15.9 31.6 5.8 1.8 1 1 4 2 2.63*** 1.50***H Hungary 67 606 (46.9) 810 (48.9) 15.7 20.1 6.9 13.6 43.6 34.1 3 3 3 1 0.95 1.04Irl Ireland 63 894 (49.2) 1198 (48.3) 2.1 3.0 50.5 63.4 9.8 6.0 1 1 2 4 2.68*** 1.45**L Luxembourg 50 781 (45.2) 707 (45.0) 7.2 8.5 9.6 17.5 34.2 29.0 3 3 2 4 1.19 1.00Nl Netherlands 64 764 (48.9) 1060 (50.9) 4.1 6.0 11.1 14.4 55.6 47.7 2 3 1 4 1.57* 1.22N Norway 66 877 (46.2) 810 (47.0) 6.0 7.4 5.1 5.1 40.0 35.2 3 4 1 4 1.14 1.07Pl Poland 74 749 (43.0) 803 (44.9) 12.8 13.7 49.1 61.0 7.1 3.4 1 1 4 1 1.16 1.52***P Portugal 71 751 (48.1) 1135 (51.2) 10.1 19.1 19.7 35.9 31.6 16.7 1 1 3 2 1.17 1.02Sk Slovakia 63 612 (43.2) 623 (44.0) 9.5 12.7 28.3 35.8 27.3 20.2 1 1 4 1 1.27* 0.90Slo Slovenia 70 582 (46.0) 698 (48.2) 10.0 15.2 11.2 20.3 27.3 22.1 2 2 3 2 1.09 1.14E Spain 55 779 (45.5) 748 (46.3) 7.6 12.3 14.9 23.8 44.9 34.4 2 2 3 2 1.10 1.32**S Sweden 66 924 (47.6) 922 (49.1) 3.0 5.2 4.1 3.8 45.9 36.3 4 4 1 3 1.04 1.58*Ch Switzerland 47 912 (47.5) 1164(49.9) 2.4 3.6 11.0 14.4 27.9 22.4 1 2 1 4 1.34 1.28Tr Turkey 51 737 (41.5) 915 (39.5) 5.0 13.8 62.3 7.8 11.7 48.9 1 1 4 1 1.41** 1.18

Total 18,323 21,373

SRH – self-rated health.Q: quartile. 1¼ bottom 4¼ top.Secularity index created from country means of three indices of religious involvement – higher quartile¼more secular Corruption index: higher quartile¼more corrupt.GDP¼ gross domestic product per capita in PPS. Higher quartile¼more affluent.*significance of attendance in logistic regression models *0.1> p> 0.05, **0.05> p> 0.01, ***p< 0.01 Odds ratios adjusted for age and education.

A.N

icholsonet

al./Social

Science&

Medicine

69(2009)

519–528522

Table 2Odds ratio for poor self-rated health from multilevel logistic regression models.

Model 1: adj for age and education Model 2 :þ health Model 3:þ social contact Model 4:þ country-level Final model: Fully adjusted

Men N [ 18,328OR for poor SRH with attendance as a categorical variableRegular 1 1 1 1 1Infrequent 1.19 [0.91–1.55] 1.19 [0.87–1.62] 1.16 [0.89–1.52] 1.19 [0.91–1.55] 1.16 [0.85–1.58]Sporadic 1.40 [1.14–1.72] 1.40 [1.10–1.77] 1.37 [1.12–1.68] 1.40 [1.14–1.71] 1.37 [1.09–1.75]Never 1.83 [1.49–2.26] 1.75 [1.37–2.23] 1.74 [1.41–2.15] 1.83 [1.48–2.25] 1.69 [1.32–2.16]

OR for poor SRH with attendance as a linear variableAttendance 1.23 [1.15–1.23] 1.21 [1.12–1.31] 1.21 [1.13–1.29] 1.23 [1.15–1.31] 1.20 [1.10–1.30]

Other co-variatesAge 1.05 [1.04–1.05] 1.02 [1.02–1.03] 1.05 [1.04–1.05] 1.05 [1.04–1.05] 1.02 [1.01–1.02]

Educationprimary 1 1 1 1 1secondary 0.49 [0.41–0.58] 0.56 [0.46–0.68] 0.51 [0.43–0.60] 0.49 [0.42–0.58] 0.59 [0.48–0.72]university 0.33 [0.27–0.42] 0.49 [0.38–0.63] 0.37 [0.29–0.46] 0.34 [0.27–0.42] 0.53 [0.41–0.69]

No Longstanding illness 1 1Yes – a little 18.91 [15.5–23.1] 18.48 [15.1–22.6]Yes – a lot 124.3 [98.9–156] 118.7 [94–149]

Not married/cohabiting 1.13 [1.07–1.19] 1.06 [0.99–1.13]No close discussion 1.41 [1.20–1.66] 1.34 [1.11–1.63]Little social contact 1.91 [1.67–2.19] 1.60 [1.36–1.88]

Secularity indexa 1.14 [0.89–1.47]GDPa 0.68 [0.55–0.83]GINIa 0.94 [0.73–1.21]Corruption scalea 1.39 [1.10–1.74] 1.48 [1.16–1.88]

Random effectsSD of Intercept (SE) 0.692 (0.111) 0.761 (0.123) 0.656 (0.106) 0.496 (0.083) 0.610 (0.102)

Women N [ 21,373OR for poor SRH with attendance as a categorical variableRegular 1 1 1 1 1Infrequent 0.95 [0.79–1.13] 0.94 [0.76–1.17] 0.93 [0.78–1.11] 0.95 [0.79–1.13] 0.93 [0.75–1.16]Sporadic 1.12 [0.97–1.29] 1.06 [0.89–1.26] 1.12 [0.97–1.29] 1.12 [0.97–1.29] 1.07 [0.90–1.27]Never 1.38 [1.19–1.61] 1.18 [0.99–1.42] 1.34 [1.15–1.57] 1.38 [1.19–1.61] 1.18 [0.98–1.42]

OR for poor SRH with attendance as a linear variableattendance 1.12 [1.06–1.18] 1.06 [1.00–1.13] 1.11 [1.05–1.17] 1.12 [1.06–1.18] 1.06 [1.00–1.13]

Other co-variatesAge 1.05 [1.04–1.05] 1.02 [1.01–1.02] 1.04 [1.04–1.05] 1.05 [1.04–1.05] 1.02 [1.01–1.02]

Educationprimary 1 1 1 1 1secondary 0.50 [0.44–0.58] 0.61 [0.52–0.71] 0.52 [0.46–0.60] 0.51 [0.45–0.58] 0.62 [0.53–0.73]university 0.26 [0.21–0.32] 0.40 [0.32–0.51] 0.27 [0.22–0.33] 0.26 [0.21–0.32] 0.43 [0.34–0.54]

No Longstanding illness 1 1Yes – a little 16.9 [14.4–19.8] 16.6 [14.1–19.4]Yes – a lot 128.4 [106–156] 125.1 [103–152]

Not married/cohabiting 1.10 [1.05–1.15] 1.04 [0.99–1.10]No close discussion 1.72 [1.51–1.95] 1.56 [1.33–1.83]Little social contact 1.50 [1.34–1.68] 1.34 [1.16–1.54]

Secularity indexa 0.97 [0.78–1.22]GDPa 0.66 [0.57–0.77]GINIa 1.04 [0.84–1.30]Corruption scalea 1.46 [1.23–1.74] 1.57 [1.27–1.96]

Random effectsSD of Intercept (SE) 0.606 (0.096 0.774 (0.122) 0.566 (0.090) 0.373 (0.062) 0.554 (0.089)

Models with attendance entered as a categorical variable were adjusted for same covariates as the models with attendance entered as a linear variable. Values for the othercovariates and for the variance of the intercept are from the models with attendance as a linear variable.SD – standard deviation.

a All country-level variables divided into quartiles with odds ratio given for unit increase. Due to collinearity, odds ratios given for country-level variables are from individualmodels with each country-level variables added separately. Odd ratios for other co-variates are from a model with all country-level variable together.

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528 523

poor SRH compared to men who attended regularly whereaswomen who never attended had a 40% increase in poor SRH.

Longstanding illness was, as expected, strongly associated withpoor SRH but adjusting for longstanding illness weakened theassociation between attendance and SRH only marginally in men. In

women adjusting for longstanding illness reduced the associationbetween attendance and poor SRH to borderline significant level. Inboth sexes, adjusting for longstanding illness reduced the associa-tion between education and self-rated health and the variance of theintercept was increased in both men and women after adjustment.

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528524

This suggests that the differences in the mean level of poor SRHbetween countries were not explained by differences in reportedlongstanding illness, indeed that the country differences in SRHwere increased by adjusting for longstanding illness. Reduced socialcontact was also associated with poor SRH but again adjustment didnot affect the association between attendance and health or explainthe variance in the intercept.

Of the 4 country-level variables examined, only GDP and corrup-tion were associated with health, with corruption the strongest andhence included in the final model. Results for country-level variablesin model 4 Table 2 are presented for quartiled scores for ease ofinterpretation, but the results were similar when used continuously.Due to collinearity between GDP and corruption, odds ratios are givenfor models with each country-level variable added separately to theindividual-level characteristics. As inclusion of the country-levelvariables, together or individually, made no material difference to theeffect sizes for attendance or the other covariates, the effect sizes for

Table 3Variations in the association between attendance at religious services & poor SRH. (Odds r

Men

N (n poor SRH) OR for poor SRH in s

Stratified by individual-level variables

Educational levelPrimary x 3267 (413) 1.30 [1.16–1.46]Secondary 10,511 (698) 1.25 [1.14–1.37]University 4550 (164) 1.06 [0.88–1.26]

Limited by longstanding illness?No/a little x 17,342 (704) 1.15 [1.05–1.25]A lot 986 (571) 1.27 [1.10–1.47]

**

Socially isolateda

No x 17,277 (1116) 1.18 [1.10–1.27]Yes 1051 (159) 1.42 [1.16–1.74]

Stratified by country-level variables

Secularity indexLow secular x 5662 (344) 1.38 (1.23–1.55)2 3959 (223) 1.23 (1.06–1.43)3 3771 (264) 1.07 (0.92–1.25)High 4936 (444) 1.16 (1.02–1.32)

Corruption indexLow corruption x 5056 (211) 1.32 [1.10–1.60]2 4707 (256) 1.32 [1.13–1.53]3 4279 (446) 1.09 [0.97–1.22]High 4286 (362) 1.32 [1.17–1.48]

*

GDPLowest quartile x 4626 (521) 1.19 [1.07–1.32]2 4846 (328) 1.25 [1.10–1.42]3 4634 (245) 1.16 [0.99–1.36]Highest quartile 4222 (181) 1.46 [1.22–1.76]

*

Combined corruption/secularityb

Low sec, low corr x 4398 (147) 1.73 [1.44–2.08]low sec, high corr 5223 (420) 1.24 [1.12–1.38]high sec, low corr 5365 (320) 1.13 [0.98–1.30]high sec, high corr 3342 (388) 1.11 [0.97–1.28]

***

From multilevel logistic regression models with country as higher level variable x refere* significance LR test for model with interaction term * 0.1> p> 0.05, ** 0.05> p> 0.01,Low secularity: low corruption: Austria, Finland, Ireland, NL(m). Switzerland.Low secularity: high corruption Greece, Poland, Slovakia, Slovenia, Spain, Turkey, PortuHigh secularity: low corruption: Belgium, Denmark, Germany, Luxembourg, Sweden, NHigh secularity: high corruption: Czech Republic, Estonia, France Hungary.

a socially isolated ¼ not married/cohabiting and no-one to discuss personal matters wb low¼ bottom 2 quartiles, high ¼ top 2 quartiles.

other covariates in model 4 are estimated in models with all 4country-level variables. The variance in the intercept was, however,reduced by inclusion of the country-level variables.

Individual-level variation in relationship betweenattendance and health

Analyses of individual-level variation in the effect on attendanceon poor SRH considered educational level, presence of longstandingillness and social isolation (Table 3). There was a trend seen in bothmen and women for the effect of attendance on poor SRH to be less inthose with university education, although the interaction term didnot achieve significance in men and was borderline significant inwomen. When stratified by the presence of longstanding illness, theeffect of attendance was stronger in both men and women reportinga lot of limitation from longstanding illness, with the interactioneffect significant in both sexes. The relationship between social

atio for poor SRH for attendance as linear variable shown, age & education adjusted).

Women

tratum N (n poor SRH) OR for poor SRH in stratum

4868 (999) 1.20 [1.11–1.29]11,617 (961) 1.11 [1.03–1.19]4888 (146) 0.95 [0.80–1.14]

19,987 (1189) 1.06 [0.99–1.13]1386 (917) 1.10 [0.98–1.24]

***

20,177 (1763) 1.13 [1.07–1.20]1196 (343) 1.01 [0.88–1.15]

**

5986 (676) 1.15 (1.05–1.26)4717 (361) 1.27 (1.13–1.42)4794 (488) 1.08 (0.98–1.20)5876 (581) 1.01 (0.91–1.12)

*

5675 (305) 1.34 [1.15–1.55]5222 (359) 1.17 [1.04–1.31]5432 (845) 1.04 [0.96–1.13]5044 (597) 1.14 [1.03–1.25]

***

5641 (836) 1.02 [0.94–1.10]5842 (665) 1.15 [1.05–1.26]4951 (343) 1.19 [1.05–1.35]4939 (262) 1.19 [1.04–1.36]

***

4469 (199) 1.46 [1.26–1.68]6234 (838) 1.16 [1.07–1.25]6428 (465) 1.13 [1.02–1.26]4242 (604) 0.97 [0.88–1.08]

***

nce group for interaction terms.*** p< 0.01.

gal.orway, NL (f).

ith.

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528 525

isolation and the impact of attendance of religious services on healthshowed opposite trends in men and women. There was a strongrelationship between attendance and health in socially isolated menbut weaker in men who had social ties. This high odds ratio (1.42)was based on small numbers of men reporting poor health so theinteraction was not significant. Further analysis showed this effect isdriven by the high prevalence of poor SRH in men who are sociallyisolated and never attend religious services (results not shown intable). By contrast, socially isolated women had a weaker relation-ship between health and attendance but reported high levels of poorSRH regardless of how frequently they attend services.

Country-level variation in relationship betweenattendance and health

Table 3 shows that effect of attendance at religious services wasweaker in more secular countries in both men and women but thiswas not a statistically significant interaction. The effect of attendanceon health was stronger in countries with less corruption, with thiseffect particularly strong in women. One group of countries in thethird quartile of corruption (Estonia, France, Hungary, Slovenia, Spainand Portugal) showed weak associations between attendance andhealth. When countries were classified according to secularity and

mean corruption score in country

A

BCzDk Est

Fin

FD

Gr

H

Irl

L

Nl

PlSk

SloES

Tr

P

Ch

N

A

B

Cz

Dk

Est

Fin

F

D

Gr

H

Irl

L

Nl

Pl

Sk

Slo

E

S

Tr

P

Ch

N

-.5

0

.5

1

0 2 4 6 8 0 2 4 6 8

Male Female

lo

g o

dd

s ratio

fo

r atten

dan

ce

country corruption index

secularity index in country

A

BCzDk Est

Fin

FD

Gr

H

Irl

L

Nl

PlSk

SloE S

Tr

P

Ch

N

A

B

Cz

Dk

Est

Fin

F

D

Gr

H

Irl

L

Nl

Pl

Sk

Slo

E

S

Tr

P

Ch

N

10 15 20 10 15 20

Male Female

country secularity index

Graphs by Gender

Graphs by Gender

-.5

0

.5

1

lo

g o

dd

s ratio

fo

r atten

dan

ce

a

b

Fig. 1. Country –level differences in effect size for attendance (adjusted for age and education)mean level of poor SRH in country.

corruption, the strongest effect seen in countries with low secularityand low corruption – Austria, Finland, Netherlands, Ireland andSwitzerland. When countries were stratified by quartile of GDP, therewas a stronger association between attendance and poor health in themost affluent countries (Ireland, Netherlands, Luxembourg, Norway,Switzerland) in men and in all countries above the bottom quartile inwomen. The scatter graphs in Fig. 1 confirm the trend in men andwomen for weaker effectof attendance in countries with higher levelsof secularity and corruption. Fig. 1d shows the scatter graph ofcountry-specific mean effect size for attendance and poor SRH againstthe country mean prevalence of poor SRH. This shows an inverselinear trend of higher effect sizes in countries with lower meanprevalence of poor SRH. Thus, in countries which had better healthoverall, the effect of attendance on health was stronger. This negativecovariance between level of SRH in a country and the strength ofassociation between attendance and health was confirmed in themultilevel models with random slopes which were fitted.

Discussion

Summary of results

These analyses extend the existing literature by examining theassociation between attendance at religious services and health in

GDP of country

A

BCz DkEst

Fin

FD

Gr

H

Irl

L

Nl

PlSk

SloES

Tr

P

Ch

N

A

B

Cz

Dk

Est

Fin

F

D

Gr

H

Irl

L

Nl

Pl

Sk

Slo

E

S

Tr

P

Ch

N

0 100 200 300 0 100 200 300

Male Female

country GDPGraphs by Gender

mean level of poor SRH in country

A

BCzDk Est

Fin

F D

Gr

H

Irl

L

Nl

PlSk

SloES

Tr

P

Ch

N

A

B

Cz

Dk

Est

Fin

F

D

Gr

H

Irl

L

Nl

Pl

Sk

Slo

E

S

Tr

P

Ch

N

0 .05 .1 .15 .2 0 .05 .1 .15 .2

Male Female

mean prevalence of poor SRH

-.5

0

.5

1

lo

g o

dd

s ratio

fo

r atten

dan

ce

-.5

0

.5

1

lo

g o

dd

s ratio

fo

r atten

dan

ce

Graphs by Gender

c

by, a. mean corruption score in country, b. secularity index in country, c. GDP of country, d.

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528526

diverse European populations. We confirm that an associationbetween less frequent attendance at religious services and poorself-rated health exists across Europe, but it is weaker in womenthan in men and the effect varies according to both individual andsocial context. The association between attendance at religiousservices and self-rated health is stronger in people who reportlimitation due to longstanding illness and it tends to be strongerin those with less education. The association is stronger for thoseliving in more affluent countries with higher level of religiouspractice and low levels of corruption.

Many other studies of SRH and attendance give results ascorrelation or regression coefficients rather than relative risks orodds ratios so it is difficult to compare our effect sizes with thosereported previously. Hyppa reports odds ratios of having goodhealth for every unit increase in religious participation of 2.36 inmen and 1.47 in women although the number of units in the scale isnot reported (Hyyppa & Maki, 2001). Studies of attendance andmortality have found relative risks of around 1.3 for less attendance(McCullough et al., 2000). Thus odds ratios found in this study ofbetween 1 and 2 are consistent with the literature and indicatea small but potentially important effect.

Potential explanations for the association betweenSRH and attendance

Following Idler’s typology (Idler, 1995), in these cross-sectionaldata there are three possible categories of explanation for theobserved association between less frequent attendance at religiousservices and worse self-rated health: social contact improveshealth, either directly or via health behaviours; poor healthprevents social contact (an example of reverse causality); or reli-gious attendance affects perception of health rather than healthitself.

Reverse causalityBecause these data are cross-sectional we cannot exclude the

possibility that those in poor health were prevented fromattending services and that this accounts for the observed associ-ations. The effect persists, but is weaker in men and insignificant inwomen, in participants who report no or slight limitation fromlongstanding illness, suggesting this endogeneity accounts forsome but not all of the observed effect. Further analyses (notshown) showed that the linear effect of attendance on SRH in thosewith longstanding illness is largely driven by the higher prevalenceof poor SRH in those never attending. This is consistent with theeffect seen being due to ill-health preventing attendance. We haveincomplete control for health status because limitation fromlongstanding illness is also subject to reporting bias. Additionalmeasures, such as number of times attended doctor, were presentin the ESS dataset but did not affect associations and were notincluded in final models. More objective measures of health status,such as the PF10 subscale from the SF36 or biological measures,would be helpful and should be considered in future work on thistopic.

Social contactThe role of social contact in mediating the observed associations

between attendance and better health are difficult to assess inthese data. Adjusting for other social contact variables made littledifference to associations between attendance and health in men orwomen. Such adjustment would only explain the observed asso-ciations if the other social contact measures included were thecause of better health rather than religious-based contact and thusthis adjustment does not address the mediating hypothesis. Strat-ifying by social isolation showed that for men with little socialcontact the effect of attendance was stronger than in men who had

other social resources. However it was weaker in the equivalentgroup of women. This may reflect difference in the structure ofsocial networks and capital between the sexes. Attending servicesmay not act as a source of support to women who are otherwiseisolated but was associated with better self-reported healthamongst isolated men.

In adjusted models, the association between attending servicesand poor SRH is only borderline significant in women and isstronger in men. A similar stronger effect in men with SRH wasreported in US by Maselko (Maselko & Kubzansky, 2006) andScandinavia by Hyppa (Hyyppa & Maki, 2001) but this contrastswith studies of mortality where the effect of attendance onlongevity is stronger in women (Chida et al., 2009; la Cour et al.,2006; McCullough et al., 2000) Our findings of a weaker associ-ation in women suggest that meaningful social contact for womenmay come from more informal contacts whereas men can derivebenefit from more structured contact. This however does notexplain the contradiction with mortality studies and the differ-ence may reflect the more subjective endpoint used in theseanalyses.

Perceptions of healthThere is some uncertainty about what is being measured when

using self-rated health as an endpoint. Self-rated health is ‘‘clearlysomething more and something less’’ (Maddox & Douglass, 1973)than more objective measures of health and this ‘‘something more’’may provide some clues to the mechanisms involved. Despitestrong and consistent associations with mortality (Idler & Benya-mini, 1997), SRH is measuring an individual’s perception of healthrather than health itself. Idler described a non-physical sense of selfwhich might be included in an individual’s assessment of theirhealth (Idler, 1995). Hence, if religious involvement leads tostronger positive non-physical sense of self, this might lead toratings of SRH being higher. Sen has also argued that SRH may besensitive to differences in expectation, perception and experience(Sen, 2002) and religious involvement may affect such perceptionsor expectations. In those with longstanding illness, the strongereffect of attendance on health might be due to this psychologicalimpact rather than reflect reverse causality. Such improvedperception may in turn actually affect health but longitudinal datawith more objective health endpoints would be needed todemonstrate such an effect.

Other potential explanations for the association betweenSRH and attendance

Health behaviours are undoubtedly important in the in theassociations between attendance at religious services and health,since those not attending are more likely to have adverse healthbehaviours such as smoking, alcohol use and poor diet. Forexample, Arredondo showed that in Hispanic women the associa-tion between church attendance was positively and independentlyassociated with healthier diet and increased physical activity butfound no independent association with self-rated health (Arre-dondo et al., 2005). Because ESS did not collect data on healthbehaviours, we are unable to assess the contribution that behaviourmakes to the associations or to the interactions.

Confounding by personality is another explanation for theobserved associations. It is plausible that positive affect may beassociated both with more social encounters, hence attendance atreligious services, and with better reporting of health. Conversely,negative affectivity may reduce social contact and lead to worsereporting of health status. In a rare longitudinal study of attendanceand SRH, McCullough found that attendance at religious serviceswas predictive of better SRH in women and that the adjustment forthe Big Five factors of personality made very little difference to the

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528 527

associations (McCullough & Laurenceau, 2005). ESS data includethe Human Values Scale and in future work it might be interestingto consider how dimensions on the HVS contribute to the observedassociations.

Individual-level variations in associations between religiousinvolvement and SRH

The relationship between educational level and the effect ofreligious involvement on health is of interest because of thepotential role of relative deprivation as a psychosocial explanationfor social inequality in health. Under this hypothesis, the awarenessof your position in society and perceived relative deprivationcontributes to the adverse effect of low social status on health(Bartley, 2004). We could predict that such perceived relativedeprivation might have less salience for those with religiousinvolvement, and thus the social gradient in health would beweaker for those who attend services regularly. In this way,attendance might be a source of resilience for those with lesseducation. The education gradient was weaker in men who atten-ded regularly (OR for university education and poor SRH comparedto primary education is 0.47 [0.26–0.84] for men who attendregularly and 0.28 [0.19–0.39] for those who never attend) but theinteraction was not significant. The trends are however consistentwith the psychosocial hypothesis.

Several of the observed trends or interactions indicate thatattendance at religious services might offer solace or a buffer tostress for men who are at risk due to, for example, social isolation orill-health. This is consistent with attendance at services acting asa source of resilience, although it is not clear whether this is actingvia perception or via an actual effect on health. Other authors havefound that men in adverse circumstances benefit from socialsupport or religious involvement (Antonucci, Ajrouch, & Janevic,2003; Vaillant, Templeton, Ardelt, & Meyer, 2008).

Country-level-variations in associations between religiousinvolvement and SRH

These analyses offer some evidence that context is important indetermining the influence that religious involvement has on health.There was a trend, as might be expected, that attendance hasa stronger effect on health in countries where religious behaviour ismore common (Fig. 1b). This could be due to the adverse psycho-social effect of breaking social expectations or may reflect reversecausality. The greater the social pressure to attend, the stronger theeffect of ill-health preventing attendance may be. Higher levels ofcorruption within a country were associated with reduced associ-ations between attendance and health. This effect was strongest inwomen and this suggests that general distrust of authority includedreligious institutions. When secularity and corruption werecombined the strongest effect was, as expected from the aboveresults, seen in more religious countries with low corruption andthese include stable and affluent countries such as Netherlands,Switzerland and Finland.

Countries with a strong effect of attendance on SRH have a lowprevalence of poor SRH overall (Fig. 1d). This may reflect the role ofaffluence, such that when conditions are good and overall health isbetter, there is more scope for other social factors to affect health assuggested by Maslow’s hierarchy of needs (Maslow, 1943). Idler’sconcept of the non-physical sense of self is a related concept, andagain it is plausible that in countries where life is easier, there ismore scope for other dimensions to be included in the con-ceptualisation of health (Idler, 1995). If people with religiousinvolvement have an increased sense of this non-physical sense

this will contribute to the observed associations. This explanationwould also account for the finding that the association betweenheath and attendance in men was stronger in more affluentcountries. Although at an individual-level there was some evidencethat religious involvement offered resilience to those in poor socialcircumstances, this was not seen at a country-level.

Limitations to analysis

Caution is needed in the interpretation of differences betweencountries because we have used only fairly crude descriptivefeatures of countries and so these analyses cannot determinewhether the stronger effect of attendance on health in somecountries is related to wealth, lack of corruption or some otheraspect of the societies which has not been measured. Equally, thestrong observed effects in some countries, such as in men in Greeceand Ireland, which are influencing the analyses of country groups,are based on small numbers of reports of poor health in non-attenders.

Our data is based on surveys which seek to achieve represen-tativeness. The response rate was between 50% and 70% in mostcountries and so the issue arises of how representative the data areof the social processes in these countries. This variation in responserates across different countries may have introduced differentialself-selection response bias and this may have potentially affectedthe findings at a societal level. However within a country we haveno reason to believe that selection bias should distort the associa-tion with health. It is impossible to predict in which direction anysuch biases would occur.

In these analyses we have only considered the associationbetween attendance and health, not other dimensions of religiousinvolvement. This is because attendance is the most widely usedmeasure, especially for comparative analyses. It is possible howeverthat the protective effect of attendance occurs because attendanceis strongly associated with other features of religious behaviour(e.g. prayer) which are the ‘‘true’’ cause of better health. Maselkoconsidered other dimensions as predictors of SRH and found thatattendance was the most strongly associated with health (Maselko& Kubzansky, 2006). This is consistent with research looking atmortality as an endpoint which has also found that attendance isthe strongest predictor (Chida et al., 2009; McCullough et al., 2000).Further work will consider the importance of other dimensionssuch as prayer and self-assessed religiousness.

Conclusions

This study has shown that religious attendance was associatedwith better reported health across Europe. The effect varies withboth individual and country-level variables and is weaker inwomen. We have found some evidence that attendance at religiousservices is associated most strongly with better health in men whoare vulnerable due to the low education, presence of longstandingillness or social isolation. The effect is strongest in countries whichare more affluent, less secular and less corrupt.

These findings highlight the importance of taking othercontextual factors into account when considering the role of reli-gion on health and suggest that religious involvement may bea source of social capital and resilience. It remains unclear to whatextent the observed associations reflect reverse causality or are dueto differing perceptions of health. Longitudinal studies are requiredto clarify the direction of effect and to explore whether improvedperception of health is translated into more concrete healthbenefits.

A. Nicholson et al. / Social Science & Medicine 69 (2009) 519–528528

References

Antonucci, T. C., Ajrouch, K. J., & Janevic, M. R. (2003). The effect of social relationswith children on the education-health link in men and women aged 40 andover. Social Science & Medicine, 56(5), 949–960.

Arredondo, E. M., Elder, J. P., Ayala, G. X., & Campbell, N. R. (2005). Is churchattendance associated with Latinas’ health practices and self-reported health?American Journal of Health Behavior, 29(6), 502–511.

Bartley, M. (2004). Health inequality. Cambridge: Polity Press.Berkman, L. F., & Glass, T. (2000). Social integration, social networks, social support

and health. In L. F. Berkman, & I. Kawachi (Eds.), Social epidemiology. New York:Oxford University Press.

Bierman, A. (2006). Does religion buffer the effects of discrimination on mentalhealth? Differing effects by race. Journal for the Scientific Study of Religion, 45(4),551–565.

Bruce, S. (2003). God is dead: Secularization in the West. Oxford: Blackwells.Chida, Y., Steptoe, A., & Powell, L. H. (2009). Religiosity/spirituality and mortality:

a systematic quantitative review. Psychotherapy and Psychosomatics, 78(2), 81–90.Davies, G. (2000). Religion in modern Europe: A minority mandate. Oxford: Oxford

University Press.Giaquinto, S., Spiridigliozzi, C., & Caracciolo, B. (2007). Can faith protect from

emotional distress after stroke? Stroke, 38, 993–997.Hyyppa, M. T., & Maki, J. (2001). Individual-level relationships between social

capital and self-rated health in a bilingual community. Preventive Medicine,32(2), 148–155.

Idler, E. L. (1995). Religion, health, and nonphysical senses of self. Social Forces,74(2), 683–704.

Idler, E. L., & Benyamini, Y. (1997). Self-rated health and mortality: a review oftwenty-seven community studies. Journal of Health and Social Behavior, 38(1),21–37.

Jaffe, D. H., Eisenbach, Z., Neumark, Y. D., & Manor, O. (2005). Does living in a reli-giously affiliated neighborhood lower mortality? Annals of Epidemiology, 15(10),804–810.

Jammer, M. S., & Stokols, D. (2000). Promoting human wellness: New frontiers forresearch, practice and policy. Berkeley CA: University of California Press.

Jowell, R. (2005). European social survey 2004/2005. Technical report. London:Centre for Comparative Social Surveys, City University.

Krause, N., Ingersoll-Dayton, B., Liang, J., & Sugisawa, H. (1999). Religion, socialsupport, and health among the Japanese elderly. Journal of Health and SocialBehavior, 40(4), 405–421.

la Cour, P., Avlund, K., & Schultz-Larsen, K. (2006). Religion and survival in a secularregion. A twenty year follow-up of 734 Danish adults born in 1914. SocialScience & Medicine, 62(1), 157–164.

McCullough, M. E., Hoyt, W. T., & Larson, D. B. (2001). Small, robust, and important:reply to Sloan and Bagiella (2001). Health Psychology, 20(3), 228–229.

McCullough, M. E., Hoyt, W. T., Larson, D. B., Koenig, H. G., & Thoresen, C. (2000).Religious involvement and mortality: a meta-analytic review. Health Psychology,19(3), 211–222.

McCullough, M. E., & Laurenceau, J. P. (2005). Religiousness and the trajectory ofself-rated health across adulthood. Personality and Social Psychology Bulletin,31(4), 560–573.

Maddox, G. L., & Douglass, E. B. (1973). Self-assessment of health: a longitudinalstudy of elderly subjects. Journal of Health and Social Behavior, 14, 87–93.

Marmot, M. (2004). Status syndrome: How your social standing directly affects yourhealth and life expectancy. London: Bloomsbury.

Marmot, M., & Wilkinson, R. G. (2001). Psychosocial and material pathways in therelation between income and health: a response to Lynch et al. BMJ, 322(7296),1233–1236.

Martin, D. (2005). On secularization: Towards a revised general theory. Aldershot:Ashgate.

Maselko, J., & Kubzansky, L. D. (2006). Gender differences in religious practices,spiritual experiences and health: results from the US general social survey.Social Science & Medicine, 62(11), 2848–2860.

Maslow, A. H. (1943). A theory of human motivation. Psychol Review, 50, 370–396.Norris, P., & Inglehart, R. (2004). Sacred and secular: Religion and politics worldwide.

New York: Cambridge University Press.Pardini, D. A., Plante, T. G., Sherman, A., & Stump, J. E. (2000). Religious faith and

spirituality in substance abuse recovery: determining the mental healthbenefits. Journal of Substance Abuse Treatment, 19(4), 347–354.

Schieman, S., Pudrovska, T., Pearlin, L. I., & Ellison, C. G. (2006). The sense of divinecontrol and psychological distress: variations across race and socio-economicstatus. Journal for the Scientific Study of Religion, 45(4), 529–549.

Sen, A. (2002). Health: perception versus observation. BMJ, 324(7342), 860–861.Singh-Manoux, A., Gueguen, A., Ferrie, J. E., Shipley, M., Martikainen, P.,

Bonenfant, S., et al. (2008). Gender differences in the association betweenmorbidity/mortality among middle-aged men and women. American Journal ofPublic Health, 98, 2251–2257.

Sloan, R. P., & Bagiella, E. (2001). Religion and health. Health Psychol, 20(3), 228–229.Stark, R., & Bainbridge, W. S. (1996). Religion, deviance and social control. New York:

Routledge.Strawbridge, W. J., Shema, S. J., Cohen, R. D., Roberts, R. E., & Kaplan, G. A. (1998).

Religiosity buffers effects of some stressors on depression but exacerbatesothers. The Journals of Gerontology. Series B, Psychological Sciences and SocialSciences, 53(3), S118–S126.

Vaillant, G., Templeton, J., Ardelt, M., & Meyer, S. E. (2008). The natural history ofmale mental health: health and religious involvement. Social Science & Medi-cine, 66(2), 221–231.

van Breda, A. D. (2001). Resilience theory: A literature review. Pretoria, South Africa:South African Military Health Service.

Veenstra, G. (2000). Social capital, SES and health: an individual-level analysis.Social Science & Medicine, 50(5), 619–629.

Wannamethee, S. G., & Shaper, A. G. (1991). Self-assessment of health status andmortality in middle-aged British men. International Journal of Epidemiology,20(1), 239–245.

Wink, P., Dillon, M., & Larsen, B. (2005). Religion as moderator of the depression-health connection: findings from a longitudinal study. Research on Aging, 27(2),197–220.

Web references

EUROSTAT New Cronos website http://www.esds.ac.uk/international/support/user_guides/eurostat/cronos.asp.

Transparency International http://www.transparency.org/policy_research/surveys_indices/cpi/2003.

World Inequality Database http://62.237.131.23/wiid/wiid.htm.