religiousness and depression
DESCRIPTION
Religie si psihologieTRANSCRIPT
-
Religiousness and Depression: Evidence for a Main Effect and theModerating Influence of Stressful Life Events
Timothy B. SmithBrigham Young University
Michael E. McCulloughUniversity of Miami
Justin PollBrigham Young University
The association between religiousness and depressive symptoms was examined with meta-analyticmethods across 147 independent investigations (N 98,975). Across all studies, the correlation betweenreligiousness and depressive symptoms was .096, indicating that greater religiousness is mildlyassociated with fewer symptoms. The results were not moderated by gender, age, or ethnicity, but thereligiousnessdepression association was stronger in studies involving people who were undergoingstress due to recent life events. The results were also moderated by the type of measure of religiousnessused in the study, with extrinsic religious orientation and negative religious coping (e.g., avoidingdifficulties through religious activities, blaming God for difficulties) associated with higher levels ofdepressive symptoms, the opposite direction of the overall findings.
Depression and depressive symptoms are among the most com-mon of all mental disorders and health complaints. Worldwide, asmany as 330 million people may suffer from depression at anygiven time, with prevalence estimates ranging from 2%3% formen and 5%12% for women (American Psychiatric Association,2000). Approximately 20 million visits to physicians in 19931994 involved reports of depressive symptoms (Pincus et al.,1998). Depression is costly as well as prevalent: The worldwidemarket for antidepressants in 1998 was approximately $7 billion(Spirit of the Age, 1998). In the United States alone, approxi-mately $12 billion in labor is lost to depressive symptoms eachyear. Moreover, depression is also a leading cause of physicaldisability and has even been found to be a risk factor for cardio-vascular mortality (Saz & Dewey, 2001; Wulsin, Vaillant, &Wells, 1999).
Given the prevalence of depression and the burdens it creates,investigators have spent a great deal of effort trying to identify
factors that may be useful for improving its detection and diagno-sis. Well-established risk factors include (a) genetic factors (Na-tional Institute of Mental Health Genetics Workgroup, 1998); (b)gender (Nolen-Hoeksema, Larson, & Grayson, 1999); (c) socialisolation (Barnett & Gotlib, 1988; Joiner & Coyne, 1999; Lara,Leader, & Klein, 1997); (d) personality traits such as dependency,Introversion, and Neuroticism (Barnett & Gotlib, 1988; Jorm et al.,2000); and (e) stressful life events (e.g., Monroe & Simons, 1991;Nolen-Hoeksema & Morrow, 1991).
Another variable that has recently received attention in thedepression literature is religious involvement. Several recent high-profile studies (e.g., Braam et al., 2001; Koenig, George, & Peter-son, 1998; Murphy et al., 2000) indicate that certain aspects ofreligiousness (e.g., public religious involvement, intrinsic religiousmotivation) may be inversely related to depressive symptoms(with greater religious involvement associated with fewer symp-toms of depression). Of note, Braam et al. (2001) reported thatpublic religious involvement (viz., church attendance) was in-versely related to depression among older individuals from Euro-pean countries that were included in the EURODEP collaboration.Also Koenig, George, and Peterson (1998) reported that amongclinically depressed older adults, intrinsic religiousness wasstrongly associated with the speed with which individuals depres-sive episodes abated, even after controlling for a variety of poten-tial confounds. Reviewers of the overall literature on religion anddepression (e.g., Koenig, McCullough, & Larson, 2001; McCul-lough & Larson, 1999) have reached similar conclusions.
Studies on these issues have been accumulating for more than100 years (see McCullough & Larson, 1999). Indeed, studies as farback as the 1880s have pointed to religion as a possible influenceon the occurrence and severity of depression (Koenig et al., 2001).Given the perceived size and scope of this literature, scholars haverecently recommended that investigators quantify and summarize
Timothy B. Smith and Justin Poll, Department of Counseling Psychol-ogy, Brigham Young University; Michael E. McCullough, Department ofPsychology, University of Miami.
Work on this article was supported by grants from the John TempletonFoundation, the Campaign for Forgiveness Research, TP Industrials Inc.,and the Religious Research Association. We thank Sharon Black, RickIngram, and Michael Lambert for their helpful comments on previousversions of this article and Brandon Dickson, Tiffany Martin, and JimScriber for their work as coders.
Correspondence concerning this article should be addressed to TimothyB. Smith, Department of Counseling Psychology, Brigham Young Univer-sity, 340 MCKB, Provo, Utah 84602-5093 or to Michael E. McCullough,Department of Psychology, University of Miami, P.O. Box 248185, CoralGables, Florida 33124-2070. E-mail: [email protected] or [email protected]
Psychological Bulletin Copyright 2003 by the American Psychological Association, Inc.2003, Vol. 129, No. 4, 614636 0033-2909/03/$12.00 DOI: 10.1037/0033-2909.129.4.614
614
-
this literature through meta-analysis so that the potential utility ofreligiousness as a predictor of depressive symptoms could beassessed more objectively (McCullough & Larson, 1998), just asrecent meta-analytic work has contributed to understanding of thenature of the relationship of religiousness and longevity (McCul-lough, Hoyt, Larson, Koenig, & Thoresen, 2000). A similar meta-analysis on the relationship of religiousness and depressive symp-toms also would permit an examination of several subsidiaryquestions.
Possible Mechanisms of Association
If religiousness and depressive symptoms are indeed related,what factors influence that relationship? It is challenging to offera theoretical account that is both elegant and comprehensive be-cause research has yet to confirm which variables moderate therelationship, let alone which variables mediate it. Identification ofmoderating and mediating variables is important but difficult be-cause both religiousness and depressive symptoms are influencedby a host of biological, social, and psychological factors. Tocomplicate matters, the variables with which religion and depres-sion are correlated typically are not exclusively one-way relation-ships: Because of the possibility of reciprocal relationships, manyfactors that are believed to be causes of depression may also beconsequences of depression, and the same is true of religiousness.Regardless, insofar as a relationship between religiousness anddepressive symptoms does exist, any of a variety of mechanismsmight explain the association. Previous research suggests somepossibilities that should inform future efforts, although not all ofthese possibilities can be addressed with meta-analytic methods atpresent.
Potentially Common Influences on Religiousness andDepressive Symptoms
Genetic influences. Religiousness might be associated withdepressive symptoms because of similar genetic influences. Sev-eral studies suggest that 40%50% of the variance in religiousnessmay be attributable to additive genetic factors (DOnofrio, Eaves,Murrelle, Maes, & Spilka, 1999; Waller, Kojetin, Bouchard,Lykken, & Tellegen, 1990). To the extent that certain genes thatconfer resistance to depressive symptoms also contribute to thedevelopment of religious sentiments, one might expect a negativecorrelation between religious involvement and depressive symp-toms. However, the multivariate behaviorgenetic studies thatwould shed light on this issue have not been conducted to date.
Developmental influences. Developmental factors that conferresistance to depressive symptoms may also foster the develop-ment of religious interests in children and adolescents. For exam-ple, warm and caring childparent bonds may be both a protectivefactor against depression (Ingram & Ritter, 2000) and a positivefactor in development of religious interests (at least within reli-gious families, J. Wilson & Sherkat, 1994). Therefore, parentalwarmth, closeness, and caring may explain some of the associationof religiousness and resistance to depressive symptoms. Con-versely, negative life events during childhood that might conferrisk for depressive symptoms might also disincline people frominterests in or concerns with religious matters. Some research hasindicated that poor relationships with parents may predispose an
individual to depression and contribute to a lack of interest inreligious matters (Hunsberger, 1980). It may also be that someindividuals seek out religion to compensate for poor relationshipswith parents (Granqvist, 1998; Kirkpatrick & Shaver, 1990).
Depressive Influences on ReligiousnessIn addition to the possibility that the associations between
religiousness and depression are influenced by common develop-mental factors, there are good reasons to believe that depressivesymptoms might influence religiousness. People who are experi-encing high levels of depressive symptoms may find a lack ofpleasure in former religious involvements, which may over timeerode their public and even private engagements with their reli-gious faith. Moreover, to the extent that a persons depressivesymptoms include a lack of energy or a physical disability, previ-ously religious people may find themselves unable to engage inreligious pursuits, by extension making them seem less religiouson many metrics for assessing religiousness. On the other hand,depressive symptoms apparently prompt some people to seekcomfort in their religion (e.g., by attending religious services orreading holy scripture), which might, de facto, increase theirapparent religiousness (Ferraro & Kelley-Moore, 2000).
Religious Influences on Depressive SymptomsAnother possibility, that factors associated with religiousness
influence symptoms of depression, has received the most attentionin the recent empirical literature. Many researchers have suggestedthat religiousness may reduce vulnerability to depressive symp-toms by way of a variety of substantive psychosocial mechanisms.Although an exhaustive list of these potential mechanisms is notfeasible here, we illustrate with four possible mechanisms. (For amore extensive review, see Koenig et al., 2001.)
Lower substance use. National surveys point to high rates ofcomorbidity between depressive symptoms and both drug abuseand drug dependence (Grant, 1995), which suggests that drugabuse and dependence may be risk factors for the development ofdepressive symptoms. Evidence also suggests that religious in-volvement is related to lower rates of substance use (e.g., tobacco,alcohol, and drugs) among both adolescents (DOnofrio, Murrelle,et al., 1999) and adults (Kendler, Gardner, & Prescott, 1997). Theeffects of religion in deterring drug use may be largely throughendorsing moral proscriptions and discouraging interaction withdrug-using peers (DOnofrio, Murrelle, et al., 1999). Therefore, bydeterring drug use among adolescents and adults, religious in-volvement might indirectly influence peoples susceptibility todepressive symptoms.
Social support. Religious involvement may afford people op-portunities for social support, which has been found to protectagainst depressive symptoms (Koenig et al., 2001; George, Larson,Koenig, & McCullough, 2000). People who are involved in reli-gion have substantially more informal social contacts and are moreactive in civic engagements than people who are not (Putnam,2000). Insofar as religious involvement puts people in touch withsuch sources of social support, this participation may be a mech-anism that accounts for some of the inverse association of reli-giousness and depressive symptoms.
615RELIGIOUSNESS AND DEPRESSIVE SYMPTOMS
-
Appraisal of life events. Religiously involved people mayhave resources for appraising negative life events that reduce theperceived stressfulness of those events (Pargament, 1997). To theextent that religious people believe that their lives are controlledby a higher power or that negative life events happen for a reasonor that life events are opportunities for spiritual growth, they mayexperience life events as less threatening and less stressful (Georgeet al., 2000). As a result, these positive appraisals may help toprotect some religious individuals from depressive symptoms byhelping them to perceive negative events as less stressful.
Coping with stress. Pargament (1997) also described how avariety of religious cognitions and behaviors that help religiouslyinvolved people to cope with stress may deter negative mentalhealth outcomes. Relatedly, Koenig et al. (1992) reported thatmedically ill men who reported using religion to cope with theirphysical health problems also reported less severe depressivesymptoms, even with a variety of other demographic, physical, andpsychosocial predictors of depression controlled. Such findingssuggest that religiousness may protect people against depressivesymptoms by helping them to avert the negative psychologicalsequelae that are frequently associated with stressful life events.
Main Effect or Stress-Buffering Effect?Insofar as religious involvement is inversely related to depres-
sive symptoms, it is conceivable that this association appliesequally to persons irrespective of the amount of life stress that theyexperience. This hypothesis has been called the main effect hy-pothesis, signifying that the association of religiousness and de-pressive symptoms is significant and negative when averagedacross all levels of life stress. On the other hand, it is conceivablethat, like many psychosocial factors that are related to depressivesymptoms, the negative association of religious involvement anddepression is even stronger for people who have recently under-gone high amounts of life stress. This view has been called thebuffering hypothesis (S. Cohen & Wills, 1985).
The main effect hypothesis and buffering hypothesis are notmutually exclusive, of course: It is possible for religious involve-ment to be related to significantly lower degrees of depressivesymptoms for all persons and for the association to become evenstronger for people who are undergoing particularly high levels oflife stress. Indeed, some investigators have found evidence forboth hypotheses (e.g., Kendler, Gardner, & Prescott, 1999). Re-latedly, it is conceivable that the negative association betweenreligious involvement and depressive symptoms becomes particu-larly strong among people who are undergoing moderate or severedepressive difficulties. Depressive symptoms themselves can be asubstantial source of life stress that may require secondary copingefforts for the individual to avoid being caught in a downwardspiral. Theory development in the study of religion and mentalhealth would benefit greatly from knowing whether the religiousinvolvementdepressive symptoms relationship is consistent withthe main effect hypothesis, the buffering hypothesis, or both(Schnittker, 2001).
Other Potential Moderators of the AssociationGiven reports of differences in the rates of religiousness and
depression across sociodemographic groups (e.g., American Psy-
chiatric Association, 2000; Levin, Taylor, & Chatters, 1994;Nolen-Hoeksema et al., 1999; Taylor, Chatters, Jayakody, &Levin, 1996), it is conceivable that the religiousnessdepressionassociation is not uniform across ethnicity, age, and gender. If so,the apparent disparities among some of the results that investiga-tors have obtained may be caused by differences in the nature ofthe samples studied. For example, the religiousnessdepressionrelationship may be stronger for women than for men, for olderadults than for younger adults, or for African Americans than forEuropean Americans (Blaine & Crocker, 1995; Musick, Koenig,Hays, & Cohen, 1998).
Relatedly, it is likely that the methods used to measure reli-giousness and/or depression within these studies have a reliableimpact on the strength of the associations (Schnittker, 2001). In arecent narrative review, McCullough and Larson (1999) noted thatthe negative association of religiousness and depressive symptomsappeared to be particularly strong when religiousness was mea-sured in terms of public religious involvement or intrinsic religiousmotivation (Allport & Ross, 1967) but less so when religiousnesswas measured in terms of private religiousness, such as thestrength with which people hold particular religious beliefs. More-over, McCullough and Larson (1999) also noted that manifesta-tions of religiousness that are extrinsically motivated, perhaps as ameans to nonreligious ends (Allport & Ross, 1967), may actuallybe associated with a higher risk for depressive symptoms. Thepostulate that extrinsic religiousness is a maladaptive form ofreligiousness, predictive of negative outcomes such as mentalillness as opposed to mental health, has been supported by others(e.g., Donahue, 1985; Richards & Bergin, 1997). For example,Burris (1994) found that extrinsic religiousness is positively pre-dictive of depression unless intrinsic religiousness is either veryhigh or very low.
These findings are conceptually similar to claims concerningnegative forms of religious coping, such as blaming God for lifesproblems or avoiding problems through religious activities, thatmay also predict mental illness as opposed to mental health (e.g.,Pargament, 1997). Nevertheless, the purported association be-tween religiousness and depression as well as the effects of dif-ferent types of religiousness on that association observed in nar-rative reviews have yet to be confirmed using empirical methodsfor synthesizing scientific literatures.
We investigated issues raised above that were possible to ad-dress using extant data from the literature on religiousness anddepressive symptoms. In this meta-analysis, our major goals wereto (a) estimate the average magnitude of the association betweenmeasures of religiousness and depressive symptoms; (b) assesswhether the association of religiousness and depressive symptomsis consistent with a main effect model, a stress-buffering model, orboth; and (c) identify characteristics of study samples and methodsthat might explain variation in the religiousnessdepression asso-ciation. In addition, we assessed the vulnerability of our conclu-sions to publication bias (Rosenthal, 1979; Sutton, Song, Gilbody,& Abrams, 2000).
Method
Literature SearchTo identify published and unpublished studies examining the association
of religious involvement with symptoms of depression, we used three
616 SMITH, MCCULLOUGH, AND POLL
-
techniques. First, we conducted searches with several electronic databases:Cumulative Index to Nursing and Allied Health Literature (CINAHL),Dissertation Abstracts International, Education Resources InformationCenter database, HealthSTAR, MEDLINE, Mental Health Abstracts, Pro-gramme Applique a` la Selection et la Compilation Automatique de Lit-terature, PsycINFO, Religion Index, Social Sciences Abstracts, SocialSciSearch, Sociological Abstracts via SocioFile, Social Work Abstracts,and TGG Health & Wellness database. To capture the broadest possiblesample of relevant articles, we used multiple search terms: all wordsbeginning with the root depress, as well as terms such as affective disorder,mood, and affect. These search terms were crossed with a series of searchwords related to religion and religious spirituality (all words beginningrelig, spirit, church, mosque, synagogue, temple, worship, and pray). Toreduce inadvertent omissions, databases yielding the most citations (CI-NAHL, Dissertation Abstracts International, MEDLINE, PsycINFO, andSocial Science Abstracts) were searched one to three additional times bydifferent members of the research team. Second, we manually examinedthe reference sections of past reviews and of studies meeting the inclusioncriteria for articles not identified in the database searches. Finally, we sentsolicitation letters to authors who had published three or more articles onthe topic to obtain several additional unpublished studies. Unpublishedstudies and dissertations obtained by contacting the authors or universitylibraries eventually accounted for 24% of included reports.
Inclusion and Exclusion Criteria for Relevant StudiesStudies included in the present investigation met the following criteria:
First, they were written in English and were identified by searches con-ducted through February 2000. Second, they provided an estimate of thebivariate association of religiousness and depressive symptoms at the levelof an individual person (studies reporting group-level data on incidencerates were not included). Thus, each study involved at least one measure ofan individuals depressive symptoms and at least one measure of that sameindividuals religiousness.
Religiousness was defined broadly as any attitude, belief, motivation,pursuit, or behavior involving spiritual or religious content or processes.The terms religiousness and spirituality can both refer to an individualssearch for that which is sacred, with the former implying group or socialpractices and doctrines and the latter tending to refer to personal experi-ences and beliefs. This meta-analysis uses the word religiousness becausethe majority of research studies reviewed used similar terminology andbecause religion encompasses pursuits and behaviors that are not neces-sarily spiritual (e.g., seeking social support from a congregation) (Hill etal., 2000). Because existing measures of religiousness and spiritualitygreatly overlap in content (Tsang & McCullough, 2003), studies that usedeither term were included in the analyses. Studies that reported onlyreligious affiliation or spiritual group membership were excluded becausemembership status alone cannot be reliably inferred to represent an indi-viduals beliefs, motivations, and behaviors (e.g., Richards & Bergin,1997).
We defined symptoms of depression more narrowly, only includingstudies with measures that explicitly used the term depression and thatassessed a cluster of symptoms consonant with and specific to depressivedisorder (American Psychiatric Association, 2000). Consistent with theview that depressive symptoms and depressive disorder reside on a singlecontinuum (see also Hankin & Abramson, 2001), studies were includedthat involved measures of depressive symptoms as well as those thatdiagnosed participants into depressed and not depressed categories on thebasis of psychiatric diagnoses or the use of cut-off scores. Studies that usedonly measures of global mental health or affect or measures specific to lifesatisfaction, loneliness, or suicidal ideation were excluded.
Data CodingCoders received extensive training to increase the reliability of their
efforts. To minimize the likelihood of rater bias, Method sections of studies
were coded separately from the results (Owin, 1994; Sacks, Berrier, Reit-man, Ancova-Berk, & Chalmers, 1987). To decrease the likelihood ofhuman error in coding data, teams of two raters reviewed each article.Team members helped one another to verify the accuracy of coding anddata entry. Subsequently, each article was coded by a different team of tworaters (57%) or by Timothy B. Smith (43%). The coders were advancedpsychology students who coded the Method sections of the studies sepa-rately from the Results sections.
Coders extracted several objectively verifiable characteristics of thestudies: (a) the field of inquiry (psychology, medicine, etc.) and outlet(journal article, dissertation, etc.); (b) the number of participants and theircomposition by gender, ethnicity, age, and religious affiliation if reported;(c) the location of data collection if reported (nursing home, universityclass, etc.); (d) the design type (cross-sectional, longitudinal, experimental,quasi-experimental, and archival, with archival studies using data collectedfrom extant records [archives] and not from recruited participants); and (e)the name of the measure or measures of religious involvement. Becausemost of the information listed above was extracted verbatim from thereports, interrater agreement was quite high for categorical variables (Co-hens .85.97 across variables, mean .90) and for continuousvariables (intraclass rs .65.99, M .89; Shrout & Fleiss, 1979). Aftercomputation of interrater reliability coefficients, interrater discrepancieswere resolved by discussion and consensus.
Coders also made inferences, on the basis of information in the reports,regarding three variables: participants average levels of stress, partici-pants average levels of depression, and the aspect of religiousness mea-sured in the study. The inferred average level of depression of the partic-ipants was coded as minimal, mild, or moderate. For each effect sizeinvolving a standard measure of depression, we determined the severity ofdepressive symptoms by comparing the mean depression score for thesample with published norms. When mean scores were not reported, codersexamined the descriptions of the study samples. If no pertinent informationwas provided, the sample was coded as having minimal levels of depres-sion. When descriptions of the sample yielded information suggesting ahigher rate of depressive symptoms than in normal populations, the samplewas coded as having a mild level of depression. Finally, if the descriptionsindicated that participants had been diagnosed with major depression, thestudy was coded as having a moderate level of depression.
Similarly, the research participants average level of stress associatedwith life events was coded on the basis of descriptions of sample charac-teristics. If the authors of the studies reported no unusual circumstances orsample characteristics, the study was coded as having minimal life eventstress. Wherever any descriptions yielded pertinent information about thesample, these characteristics were compared with the ratings of perceivedlife event severity provided by Hobson et al. (1998), based on the originalwork by Holmes and Rahe (1967). The continuous ratings provided byHobson and colleagues were trichotomized in such a way that peopleundergoing life events that were of similar severity to being the victim ofa crime or going through a recent divorce were rated as undergoing severelife stress. People undergoing events similar to having major disagreementswith an employer or being involved in an auto accident were characteristicof the bottom two thirds of Hobson et al.s scale. People undergoing suchstressors were rated as experiencing mild to moderate life stress.
If the dimension of religiousness measured in a study was not explicitlystated, it was inferred by the coders. Each measure was assigned to one ofthe following seven categories, which were similar to those used by Hilland Hood (1999) in their comprehensive review: (a) religious behaviors(measures of the frequency of saying prayers, participating in worshipservices, reading sacred texts, and/or performing religious service), (b)religious attitudes and beliefs (measures of the perceived importance ofreligion, belief in deity, religious values), (c) religious orientation (mea-sures of the motivation for religious behaviors, intrinsic and extrinsic; e.g.,the Religious Orientation Scale; Allport & Ross, 1967), (d) religiouscoping (measures of the use of positive and negative strategies for dealing
617RELIGIOUSNESS AND DEPRESSIVE SYMPTOMS
-
with difficulties, such as seeking support from God vs. blaming God for theproblem; e.g., the Religious Coping Activities Scale; Pargament et al.,1990), (e) religious well-being (measures of life satisfaction in relation toreligious beliefs, such as personal feelings of a connection with God; e.g.,the Religious Well-Being subscale of the Spiritual Well-Being Scale;C. W. Ellison, 1983), (f) God concept (measures that assessed perceptionsof the extent to which one holds a positive or negative image of God), and(g) multidimensional aspects of religiousness (measures that included morethan one of the above content areas).
In coding the three inferred variables, the two coding teams demon-strated adequate reliability (Cohens .85, .87, and .86 for inferreddepression, life stress, and type of religiousness, respectively ). As withother inferences and classifications, discrepancies were resolved by dis-cussion and consensus.
Computation of Effect Size EstimatesThe majority (76%) of identified studies reported the association of
religious involvement and depressive symptoms with the correlation coef-ficient. The data from other studies were transformed to correlation coef-ficients using the Meta-Analysis Calculator software (Lyons, 1996). Whenno statistic was provided but an analysis was reported as significant, wedetermined the correlation coefficient corresponding to the reported alphalevel (assuming two-tailed .05, unless reported otherwise). When ananalysis was reported as nonsignificant but no additional information wasavailable, we set the correlation coefficient to .00. These proceduresyielded conservative effect size estimates. For the purposes of aggregation,correlation coefficients reported in individual studies were transformedinto Fishers stabilizing z scores (Shadish & Haddock, 1994) to avoid thesmall amount of bias in Pearsons r estimates. The resulting average acrossall studies was then converted back to the metric of a correlation. Thedirection of all effect sizes was coded uniformly, such that positive valuesindicated that religiousness was associated with greater symptoms ofdepression and negative values indicated that religiousness was associatedwith fewer symptoms of depression.
Several studies reported data on either multiple measures of religiousinvolvement or multiple measures of depression. For example, some stud-ies assessed attendance at religious services and frequency of personalprayer as well as aspects of religious belief or commitment. Use of each ofthese effect sizes in the omnibus analysis would have violated the assump-tion of independent samples (Cooper, 1998; Hedges & Olkin, 1985). Wetherefore used the shifting units of analysis approach to analysis, whichminimizes the threat of nonindependence in the data while allowing moredetailed follow-up analyses to be conducted (Cooper, 1998). Each singleeffect size estimate within each study was coded as if it were an indepen-dent event. However, in determining the overall association of religiousinvolvement and depression, the effect sizes within each study were aver-aged so that each study contributed only one data point to the omnibusanalysis (Mullen, 1989). This average effect size was used in subsequentanalyses except where more detailed information was required. In suchcases, the specific effect size representing the information needed was usedin the analysis rather than the average of all effect sizes extracted from thestudy.
Analyses
To aggregate effect sizes and to estimate the reliability of these aggre-gates, we calculated random effects models with Comprehensive Meta-Analysis software (Borenstein, 2000). A random effects approach yieldsresults that generalize beyond the sample of studies actually reviewed(Hedges & Vevea, 1998). The assumptions made in this meta-analysisclearly warrant this method: The belief that certain variables serve asmoderators of the observed association between symptoms of depressionand religious involvement implies that the studies reviewed would estimate
different population effect sizes. Random effects models take suchbetween-studies variation into account, whereas fixed effects models donot (Mosteller & Colditz, 1996).
Following the computation of the overall association of symptoms ofdepression to religious involvement, random effects weighted regressionmodels and mixed effects analyses of variance (ANOVAs) were conductedusing SPSS macros developed by Lipsey and Wilson (2001) to examine theinfluence of potential moderating variables. Such analyses are useful indetermining circumstances under which the association between symptomsof depression and religious involvement varies in strength, such that a moreaccurate depiction of the association is provided. To correct for the numberof analyses conducted in this study, we set the level of statistical signifi-cance at .0035.
Results
Descriptive Characteristics
Effect sizes based on nonoverlapping samples of participantswere extracted from 147 studies. Table 1 provides an overview ofdescriptive information about these studies. Table 2 provides adescription and the effect size estimate for each individual study.Across all studies, data were reported from a total of 98,975participants. Participant gender was reported in 137 studies (93%),
(text continues on page 623)
Table 1Characteristics of Studies Included in the Meta-Analysis
CharacteristicNo. of studies
(k)
Year of reportBefore 1990 2719901994 5419952000 66
Field of inquiryPsychology 56Medicine 42Nursing 15Sociology 15Religious studies 13Other (e.g., education) 6
Population sampledAdults from the general population 44College students 30Adults receiving medical care 23Adults with psychological concerns
(e.g., outpatients, bereaved) 19Adults from religious settings 15Other (e.g., caregivers, adolescents) 16
Sampling procedure usedConvenience 87Representative 28Unable to determine from report 32
Sample size 50 1350100 36101250 392511,000 37 1,000 22
DesignCross-sectional 125Longitudinal 15Other (e.g., archival, experimental) 7
618 SMITH, MCCULLOUGH, AND POLL
-
Tabl
e2
Des
crip
tions
oft
he14
7In
divi
dual
Stud
ies
Incl
uded
inth
eM
eta-
Anal
ysis
Stud
yN
Mea
nag
eSa
mpl
ety
peSt
ress
full
ifeev
ents
Dep
ress
ion
leve
lTy
peofr
elig
ious
mea
sure
Effe
ctsiz
e(r)
95%
CI
Aus
tin&
Lenn
ings
(1993
)57
33G
rievi
ng/b
erea
ved
adul
tsM
oder
ate
Mod
erat
eM
ultid
imen
siona
l
.03
.29
,.24
Bai
rd(19
90)
164
16H
igh
scho
olst
uden
tsM
inim
alM
ildM
ultid
imen
siona
l
.17
.32
,.02
Bek
arog
luet
al.(
1991
)29
272
Nur
sing
hom
ere
siden
ts&
oth
erold
erad
ults
Mild
Mild
Rel
igio
usbe
havi
ors
.31
.41
,.21
Bel
avic
h(19
95)
222
19U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usco
ping
.09
.22
,.04
Ber
gin
etal
.(19
87)
32(Y
oung
adul
ts)R
elig
ious
univ
ersit
yst
uden
tsM
inim
alM
inim
alR
elig
ious
orie
ntat
ion
.02
.34
,.36
Bic
kele
tal
.(19
98)
245
53Ch
urch
mem
bers
Min
imal
Min
imal
Rel
igio
usco
ping
.00
.13
,.13
Bie
nenf
eld
etal
.(19
97)
89(S
enior
adul
ts)R
elig
ious
nurs
ing
hom
ere
siden
tsM
ildM
inim
alR
elig
ious
orie
ntat
ion
.40
.58
,.22
Big
gare
tal
.(19
99)
205
35H
IV-p
ositi
vead
ults
&nonin
fect
edad
ults
Mod
erat
eM
inim
alM
ultid
imen
siona
l.00
.14
,.14
Bish
opet
al.(
1987
)17
3(A
dults
)Ps
ychi
atric
inpa
tient
s&
contr
olad
ults
Seve
reM
oder
ate
Rel
igio
usat
titud
es
.16
.31
,.02
Bla
ine
&Cr
ocke
r(19
95)
146
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.06
.22
,.11
Blu
mel
(1993
)12
,007
(Adu
lts)
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.06
.08
,.04
Boh
rnste
dtet
al.(
1968
)3,
666
(You
ngad
ults)
Uni
vers
ityst
uden
tsM
inim
alM
inim
alR
elig
ious
attit
udes
.18
.22
,.15
Boz
arth
(1997
)43
(Adu
lts)
HIV
-pos
itive
adul
tsSe
vere
Mod
erat
eR
elig
ious
wel
l-bei
ng
.10
.40
,.20
Bra
am,B
eekm
an,D
eeg,
etal
.(19
97)
177
(Sen
iorad
ults)
Seni
orad
ults
Mild
Mild
Rel
igio
usat
titud
es
.30
.43
,.16
Bra
am,B
eekm
an,v
anTi
lbur
g,et
al.(
1997
)2,
817
70A
dults
Min
imal
Min
imal
Mul
tidim
ensio
nal
.09
.13
,.05
Bro
wn
&G
ary
(1985
)95
(Adu
lts)
Une
mpl
oyed
adul
tsM
ildM
inim
alR
elig
ious
beha
vior
s
.03
.23
,.18
Bro
wn
etal
.(19
90)
438
42A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.17
.26
,.08
Bro
wn
etal
.(19
92)
927
43A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.23
.29
,.17
Bro
wn
&G
ary
(1994
)53
743
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.15
.23
,.06
Buc
hana
n(19
93)
160
74Ps
ychi
atric
inpa
tient
s,co
ntr
olad
ults,
&m
enta
lhea
lthoutp
atie
nts
Mod
erat
eM
oder
ate
Rel
igio
usw
ell-b
eing
.54
.65
,.43
Bur
ris(19
94)
100
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.11
.30
,.09
Carro
ll(19
92)
450
56Ch
urch
mem
bers
Min
imal
Min
imal
God
conce
pt
.06
.14
,.03
Chan
cey
(1997
)34
940
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l.00
.11
,.11
Chan
get
al.(
1998
)12
762
Care
give
rsofa
dult
patie
nts
Mod
erat
eM
inim
alR
elig
ious
copi
ng
.09
.26
,.08
Clin
e(19
92)
8073
Nur
sing
hom
ere
siden
tsM
ildM
inim
alM
ultid
imen
siona
l
.21
.41
,.02
Coat
es(19
96)
100
38A
buse
vic
tims
Mod
erat
eM
inim
alR
elig
ious
orie
ntat
ion
.02
.22
,.18
Cole
man
(1996
)11
638
HIV
-pos
itive
adul
tsSe
vere
Mod
erat
eR
elig
ious
wel
l-bei
ng
.21
.38
,.03
Com
mer
ford
(1996
)81
81N
ursin
gho
me
resid
ents
Mild
Mild
Rel
igio
usorie
ntat
ion
.05
.27
,.17
Corz
o(19
81)
123
36Ps
ychi
atric
inpa
tient
s&
contr
olad
ults
Mild
Mod
erat
eG
odco
nce
pt.00
.18
,.18
Crow
e(19
90)
389
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.07
.03
,.17
Dom
anic
o&
Craw
ford
(2000
)96
37H
IV-p
ositi
vead
ults
Seve
reM
inim
alR
elig
ious
wel
l-bei
ng
.45
.61
,.29
Dox
eyet
al.(
1997
)4,
920
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usat
titud
es
.10
.13
,.07
Duc
harm
e(19
88)
5739
Psyc
hiat
ricin
patie
nts
&co
ntr
olad
ults
Seve
reM
oder
ate
God
conce
pt
.29
.53
,.05
C.G
.Elli
son
(1991
)2,
956
43A
dults
Min
imal
Min
imal
Mul
tidim
ensio
nal
.02
.02
,.05
Faul
kner
(1994
)2,
834
59A
dults
Min
imal
Min
imal
Mul
tidim
ensio
nal
.02
.01
,.06
Fehr
ing
etal
.(19
87)
9519
Uni
vers
ityst
uden
tsM
inim
alM
inim
alR
elig
ious
wel
l-bei
ng
.04
.24
,.16
619RELIGIOUSNESS AND DEPRESSIVE SYMPTOMS
(tabl
eco
ntin
ues)
-
Stud
yN
Mea
nag
eSa
mpl
ety
peSt
ress
full
ifeev
ents
Dep
ress
ion
leve
lTy
peofr
elig
ious
mea
sure
Effe
ctsiz
e(r)
95%
CI
Fehr
ing
etal
.(19
87)
7520
Uni
vers
ityst
uden
tsM
inim
alM
inim
alM
ultid
imen
siona
l
.25
.46
,.04
Fehr
ing
etal
.(19
97)
100
73Te
rmin
ally
illad
ults
Seve
reM
ildR
elig
ious
orie
ntat
ion
.44
.59
,.28
Fern
ando
(1978
)87
(Adu
lts)
Med
ical
inpa
tient
sSe
vere
Mod
erat
eR
elig
ious
orie
ntat
ion
.25
.44
,.05
Fior
i(19
91)
722
16H
igh
scho
olst
uden
tsM
ildM
ildG
odco
nce
pt
.31
.38
,.24
Fitc
hett
etal
.(19
99)
9565
Med
ical
inpa
tient
sSe
vere
Min
imal
Mul
tidim
ensio
nal
.01
.21
,.19
Gen
ia&
Shaw
(1991
)30
929
Chur
chm
embe
rsM
inim
alM
inim
alR
elig
ious
orie
ntat
ion
.02
.09
,.13
Gra
y(19
87)
5016
Grie
ving
/ber
eave
dad
ults
Seve
reM
ildR
elig
ious
attit
udes
.30
.55
,.04
Grif
fin(19
88)
171
(Ado
lesce
nts)
Hig
hsc
hool
stud
ents
Min
imal
Mild
Rel
igio
usat
titud
es.00
.15
,.15
Gro
sse-
Hot
forth
etal
.(19
96)
9769
Med
ical
inpa
tient
sM
oder
ate
Mild
Mul
tidim
ensio
nal
.25
.43
,.06
Gup
ta(19
83)
313
17H
igh
scho
olst
uden
tsM
inim
alM
inim
alR
elig
ious
attit
udes
.24
.13
,.34
Hal
lstro
m&
Pers
son
(1984
)45
347
Men
talh
ealth
outp
atie
nts
&co
ntr
olad
ults
Seve
reM
oder
ate
Mul
tidim
ensio
nal
.07
.16
,.02
Her
tsga
ard
&Li
ght
(1984
)76
044
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.13
.20
,.06
Het
tler&
Cohe
n(19
98)
124
48Ch
urch
mem
bers
Min
imal
Min
imal
Mul
tidim
ensio
nal
.01
.18
,.17
Hin
tikka
etal
.(19
98)
1,17
941
Men
talh
ealth
outp
atie
nts
Mild
Mild
Rel
igio
usbe
havi
ors
.11
.17
,.06
Hoh
man
net
al.(
1990
)65
3(A
dults
)A
dults
Min
imal
Min
imal
Rel
igio
usw
ell-b
eing
.11
.03
,.19
Hon
g&
Gia
nnak
opou
los
(1994
)1,
722
25A
dults
Min
imal
Min
imal
Rel
igio
usat
titud
es
.06
.10
,.01
Hon
g&
May
o(19
88)
208
(You
ngad
ults)
Uni
vers
ityst
uden
tsM
inim
alM
inim
alR
elig
ious
attit
udes
.04
.09
,.18
Hus
aini
etal
.(19
99)
995
72A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.15
.21
,.08
Hyl
and
(1996
)60
81N
ursin
gho
me
resid
ents
Mild
Min
imal
Mul
tidim
ensio
nal
.19
.42
,.07
Idle
r(19
87)
2,75
674
Old
erad
ults
Min
imal
Mod
erat
eM
ultid
imen
siona
l
.13
.17
,.09
Idle
r&K
asl(
1992
)1,
447
(Olde
radu
lts)
Old
erad
ults
Min
imal
Min
imal
Mul
tidim
ensio
nal
.04
.09
,.01
Jens
enet
al.(
1993
)3,
835
21U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usat
titud
es
.06
.09
,.03
Ken
dler
etal
.(19
97)
1,86
030
Twin
sM
inim
alM
inim
alM
ultid
imen
siona
l
.04
.08
,.01
Ken
nedy
etal
.(19
96)
1,85
575
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.07
.12
,.03
Kiv
ela
etal
.(19
96)
679
74A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.35
.42
,.29
Kla
as(19
98)
7782
Adu
ltsM
inim
alM
inim
alR
elig
ious
attit
udes
.11
.33
,.11
Koe
nig
(1995
)96
57Pr
isone
rsSe
vere
Min
imal
Mul
tidim
ensio
nal
.17
.37
,.02
Koe
nig
etal
.(19
88)
9374
Med
ical
outp
atie
nts
Min
imal
Mild
Mul
tidim
ensio
nal
.20
.40
,.01
Koe
nig
etal
.(19
92)
841
70M
edic
alin
patie
nts
Seve
reM
ildR
elig
ious
copi
ng
.16
.23
,.09
Koe
nig
etal
.(19
92)
202
(Sen
iorad
ults)
Med
ical
inpa
tient
sSe
vere
Mild
Rel
igio
usco
ping
.24
.37
,.11
Koe
nig
etal
.(19
97)
4,00
073
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.07
.10
,.04
Koe
nig,
Geo
rge,
&Pe
ters
on(19
98)
8767
Med
ical
inpa
tient
sM
oder
ate
Mod
erat
eM
ultid
imen
siona
l
.11
.32
,.10
Koe
nig,
Parg
amen
t,&
Nie
lsen
(1998
)57
768
Med
ical
inpa
tient
sSe
vere
Mod
erat
eM
ultid
imen
siona
l.06
.02
,.14
Koh
bod
(1996
)78
41A
dults
Min
imal
Min
imal
Rel
igio
usw
ell-b
eing
.35
.55
,.16
Kot
eske
yet
al.(
1991
)10
919
Uni
vers
ityst
uden
tsM
inim
alM
inim
alR
elig
ious
orie
ntat
ion
.29
.46
,.12
Levi
net
al.(
1996
)61
755
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.02
.10
,.05
Lind
gren
&Co
urse
y(19
95)
3041
Men
talh
ealth
outp
atie
nts
Seve
reM
ildR
elig
ious
beha
vior
s
.42
.72
,.12
Littr
ell&
Bec
k(19
99)
9138
Hom
eles
sad
ults
Seve
reM
inim
alR
elig
ious
copi
ng.00
.21
,.21
Mae
set
al.(
1998
)2,
844
48A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.06
.10
,.02
Mal
tby
&D
ay(20
00)
360
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.11
.01
,.21
Mal
tby
etal
.(19
99)
474
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.00
.09
,.09
Mar
tin(19
94)
100
76M
edic
aloutp
atie
nts
Mod
erat
eM
inim
alM
ultid
imen
siona
l
.14
.33
,.05
Mat
on(19
89)
8146
Grie
ving
/ber
eave
dad
ults
Mod
erat
eM
inim
alR
elig
ious
copi
ng
.23
.44
,.02
620 SMITH, MCCULLOUGH, AND POLLTa
ble
2(co
ntin
ued)
-
Stud
yN
Mea
nag
eSa
mpl
ety
peSt
ress
full
ifeev
ents
Dep
ress
ion
leve
lTy
peofr
elig
ious
mea
sure
Effe
ctsiz
e(r)
95%
CI
May
oet
al.(
1969
)16
5(Y
oung
adul
ts)U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usat
titud
es
.19
.33
,.04
McD
ouga
ll(19
98)
169
68A
dults
Min
imal
Min
imal
Rel
igio
usco
ping
.08
.23
,.07
McI
ntos
h&
Dan
igel
is(19
95)
1,64
4(S
enior
adul
ts)A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.30
.34
,.26
Mic
kley
etal
.(19
98)
9257
Care
give
rsofa
dult
patie
nts
Mod
erat
eM
inim
alM
ultid
imen
siona
l
.16
.35
,.05
Mill
eret
al.(
1997
)60
55A
dults
Min
imal
Mod
erat
eM
ultid
imen
siona
l
.33
.56
,.11
Miro
la(19
90)
700
(Adu
lts)
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l
.13
.20
,.06
Mol
enka
mp
(1993
)39
445
Chur
chm
embe
rsM
inim
alM
inim
alM
ultid
imen
siona
l
.12
.22
,.02
Mor
se&
Wiso
cki(
1987
)14
872
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l
.23
.38
,.08
Mur
phy
etal
.(20
00)
271
(Adu
lts)
Psyc
hiat
ricin
patie
nts
&m
enta
lhea
lthoutp
atie
nts
Seve
reM
oder
ate
Rel
igio
usw
ell-b
eing
.32
.42
,.21
Mus
ick
etal
.(20
00)
1,89
773
Chur
chm
embe
rsM
inim
alM
inim
alR
elig
ious
beha
vior
s.01
.04
,.05
Mus
ick,
Koe
nig,
etal
.(19
98)
235
73A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.03
.16
,.10
Mus
ick
&St
rulo
witz
(2000
)8,
866
42A
dults
Min
imal
Min
imal
Rel
igio
usbe
havi
ors
.04
.06
,.02
Mus
ick,
Will
iam
s,&
Jack
son
(1998
)58
443
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.04
.12
,.04
Nee
lman
&Le
wis
(1994
)51
38Ps
ychi
atric
inpa
tient
s,co
ntr
olad
ults,
&m
enta
lhea
lthoutp
atie
nts
Seve
reM
oder
ate
Mul
tidim
ensio
nal
.16
.12
,.43
Nef
f&H
oppe
(1993
)1,
784
38A
dults
Min
imal
Min
imal
Mul
tidim
ensio
nal
.13
.18
,.08
Nel
son
(1989
a)68
72A
dults
Min
imal
Min
imal
Mul
tidim
ensio
nal
.16
.39
,.08
Nel
son
(1989
b)26
81N
ursin
gho
me
resid
ents
Mild
Min
imal
Rel
igio
usbe
havi
ors
.39
.72
,.06
OCo
nnor
&V
alle
rand
(1990
)14
682
Nur
sing
hom
ere
siden
tsM
ildM
inim
alR
elig
ious
orie
ntat
ion
.08
.09
,.24
OLa
oire
(1997
)88
(Adu
lts)
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.36
.54
,.18
Ole
r(19
97)
273
42Ch
urch
mem
bers
Min
imal
Min
imal
Mul
tidim
ensio
nal
.04
.08
,.16
Olsz
ewsk
i(19
94)
9514
Chur
chm
embe
rsM
inim
alM
inim
alR
elig
ious
copi
ng.01
.20
,.21
Palin
kas
etal
.(19
90)
1,61
575
Adu
ltsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.07
.12
,.02
C.Pa
rket
al.(
1990
)St
udy
183
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.01
.22
,.21
C.Pa
rket
al.(
1990
)St
udy
283
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.05
.26
,.17
H.S
.Par
ket
al.(
1998
)95
35Ch
urch
mem
bers
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.11
.30
,.10
Pato
ck-P
eckh
amet
al.
(1998
)26
320
Uni
vers
ityst
uden
tsM
inim
alM
inim
alR
elig
ious
orie
ntat
ion
.03
.09
,.15
Plan
te&
Boc
cacc
ini
(1997
)10
219
Uni
vers
ityst
uden
tsM
inim
alM
inim
alM
ultid
imen
siona
l
.16
.35
,.03
Pres
sman
etal
.(19
90)
30(S
enior
adul
ts)M
edic
alin
patie
nts
Mod
erat
eM
inim
alM
ultid
imen
siona
l
.30
.63
,.03
Quinn
etal
.(19
96)
8172
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l
.21
.42
,.00
Rat
anas
iripo
ng(19
96)
244
41Ch
urch
mem
bers
Min
imal
Min
imal
Rel
igio
usat
titud
es.12
.00
,.24
Ren
froe
(1990
)25
31M
enta
lhea
lthoutp
atie
nts
Mild
Mild
Rel
igio
usw
ell-b
eing
.21
.59
,.17
Ric
hard
s(19
91)
268
20U
nive
rsity
stud
ents
Min
imal
Mild
Rel
igio
usorie
ntat
ion
.07
.05
,.19
Rob
inso
n&
Kay
e(19
94)
1765
Care
give
rsofa
dult
patie
nts
Mod
erat
eM
ildM
ultid
imen
siona
l
.21
.68
,.26
Rob
inso
n&
Kay
e(19
94)
2362
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l.05
.37
,.47
Ros
ik(19
89)
159
64G
rievi
ng/b
erea
ved
adul
tsM
oder
ate
Mild
Rel
igio
usorie
ntat
ion
.16
.01
,.31
Ros
s(19
90)
401
(Adu
lts)
Adu
ltsM
inim
alM
inim
alR
elig
ious
attit
udes
.12
.22
,.02
621RELIGIOUSNESS AND DEPRESSIVE SYMPTOMSTa
ble
2(co
ntin
ued)
(tabl
eco
ntin
ues)
-
Stud
yN
Mea
nag
eSa
mpl
ety
peSt
ress
full
ifeev
ents
Dep
ress
ion
leve
lTy
peofr
elig
ious
mea
sure
Effe
ctsiz
e(r)
95%
CI
Row
ley
(1998
)23
038
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l.00
.13
,.13
Rya
net
al.(
1993
)15
123
Uni
vers
ityst
uden
tsM
inim
alM
inim
alR
elig
ious
orie
ntat
ion
.12
.17
,.05
Rya
net
al.(
1993
)42
35Ch
urch
mem
bers
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.06
.36
,.25
Scho
enba
chet
al.(
1982
)13
6(A
doles
cents
)H
igh
scho
olst
uden
tsM
inim
alM
inim
alR
elig
ious
attit
udes
.18
.01
,.34
Sher
kat&
Ree
d(19
92)
156
(Adu
lts)
Grie
ving
/ber
eave
dad
ults
Seve
reM
ildR
elig
ious
beha
vior
s
.06
.21
,.10
Shul
eret
al.(
1994
)50
30H
omel
ess
adul
tsSe
vere
Mod
erat
eR
elig
ious
copi
ng
.29
.54
,.03
Sim
on(19
92)
5084
Nur
sing
hom
ere
siden
tsM
ildM
ildR
elig
ious
beha
vior
s
.14
.41
,.14
Snyd
er(19
96)
8542
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l.15
.06
,.36
Sore
nson
etal
.(19
95)
261
(Ado
lesce
nts)
Teen
age
moth
ers
Seve
reM
ildM
ultid
imen
siona
l.09
.04
,.21
Stav
ros
(1998
)88
(Adu
lts)
Chur
chm
embe
rsM
inim
alM
inim
alR
elig
ious
beha
vior
s
.29
.48
,.09
Stee
r&Sc
hut(
1979
)20
0(A
dults
)Su
bsta
nce
abus
ers
Seve
reM
ildR
elig
ious
beha
vior
s
.09
.23
,.05
Stra
wbr
idge
etal
.(19
98)
2,53
765
Adu
ltsM
inim
alM
inim
alM
ultid
imen
siona
l
.04
.08
,.00
Stra
yhor
net
al.(
1990
)27
2(A
dults
)A
dults
Mild
Min
imal
Mul
tidim
ensio
nal
.12
.24
,.01
Teel
(1993
)11
347
Grie
ving
/ber
eave
dad
ults
&ca
regi
vers
ofa
dult
patie
nts
Mild
Mild
Mul
tidim
ensio
nal
.00
.19
,.19
Thea
rleet
al.(
1995
)63
1(A
dults
)G
rievi
ng/b
erea
ved
adul
ts&
oth
erad
ults
Mild
Mod
erat
eR
elig
ious
beha
vior
s
.09
.16
,.01
Tren
tet
al.(
1983
)50
21U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usat
titud
es
.42
.65
,.19
Vai
llant
etal
.(19
98)
223
50N
orm
alad
ults
Min
imal
Min
imal
Rel
igio
usat
titud
es.11
.02
,.24
Van
deCr
eek
etal
.(19
95)
150
43Ca
regi
vers
ofa
dult
patie
nts
Seve
reM
inim
alR
elig
ious
copi
ng
.23
.38
,.08
Virg
inia
(1998
)14
2(A
dults
)R
elig
ious
lead
ers
(clerg
y)M
inim
alM
inim
alR
elig
ious
beha
vior
s.06
.11
,.22
Wat
son
etal
.(19
89a)
221
19U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.07
.20
,.06
Wat
son
etal
.(19
89b)
1,39
720
Uni
vers
ityst
uden
tsM
inim
alM
inim
alM
ultid
imen
siona
l
.08
.14
,.03
Wat
son
etal
.(19
90)
203
21U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usco
ping
.27
.40
,.14
Wat
son
etal
.(19
92)
307
21U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.08
.19
,.04
Wat
son
etal
.(19
93)
331
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.02
.09
,.12
Wat
son
etal
.(19
94)
237
21U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.05
.18
,.08
Wat
son
etal
.(19
94)
351
20U
nive
rsity
stud
ents
Min
imal
Min
imal
Rel
igio
usorie
ntat
ion
.16
.27
,.06
H.P
.Wils
on(19
94)
118
40A
buse
vic
tims
Mod
erat
eM
oder
ate
Rel
igio
usw
ell-b
eing
.20
.37
,.03
Woo
dset
al.(
1999
)10
635
HIV
-pos
itive
adul
tsSe
vere
Mild
Mul
tidim
ensio
nal
.24
.42
,.05
Wrig
htet
al.(
1993
)45
1(A
doles
cents
)H
igh
scho
olst
uden
tsM
inim
alM
inim
alR
elig
ious
wel
l-bei
ng
.15
.24
,.06
Wya
ttet
al.(
1999
)69
972
Term
inal
lyill
adul
tsSe
vere
Min
imal
Rel
igio
usorie
ntat
ion
.15
.22
,.08
Zhan
g&
Jin(19
96)
772
21U
nive
rsity
stud
ents
Min
imal
Min
imal
Mul
tidim
ensio
nal
.01
.08
,.06
Zunz
uneg
uiet
al.(
1999
)19
466
Care
give
rsofa
dult
patie
nts
Mod
erat
eM
ildR
elig
ious
attit
udes
.06
.09
,.20
Note
.CI
confid
ence
inte
rval
.
622 SMITH, MCCULLOUGH, AND POLLTa
ble
2(co
ntin
ued)
-
with 61% of the total participants being female. Participant eth-nicity was reported in 95 studies (65%), with a breakdown of 53%European Americans, 24% African Americans, 12% NorthernEuropeans, 5% Hispanic/Latino Americans, 1% Asian Americans,1% other Americans (e.g., Native Americans), and 4% participantsfrom other nations (e.g., Australia, China, India). Religious affil-iation was reported in 45 studies (31%), with 49% Protestants,30% Catholics, 17% unspecified or unaffiliated, 3% Jewish, andless than 1% others (e.g., Muslim).
Descriptions of participant characteristics resulted in 90 (61%)studies being coded as having participants with minimal life eventstress, 32 (22%) having participants with mild to moderate stress,and 25 (17%) having participants with severe stress. Mean scoreson measures of depressive symptoms as well as descriptions ofparticipant characteristics resulted in 104 (71%) studies beingcoded as having participants with minimal symptoms of depres-sion, 25 (17%) studies having participants with mild symptoms,and 18 (12%) studies having participants with moderate symp-toms. Descriptions of instrumentation indicated that 35% of thestudies used multidimensional measures of religiousness, 20%used measures of religious behaviors, 12% used measures ofreligious attitudes and beliefs, 15% used measures of religiousorientation, 8% used measures of religious coping, 7% used mea-sures of religious well-being, and 3% used measures of Godconcept.
Omnibus Analysis
As indicated previously, each study contributed only one datapoint in the omnibus analysis. Across all 147 studies, the randomeffects weighted average effect size was .096 (SE .009, p .000001, 95% confidence interval .11, .08). Of 140 nonzeroeffect sizes, 113 (81%) were negative and 27 (19%) were positive,ranging from .54 to .24. The variability of effect sizes was quitehigh, and the index of heterogeneity was statistically significant,Q(146) 814.00, p .001, suggesting that systematic effect sizevariability was unaccounted for. We therefore conducted addi-tional analyses to determine the extent to which the variability inthe magnitude of the effect size was moderated by other variables.
A visual display of the effect sizes (x axis) by the number ofparticipants per study (logarithmic y axis) is presented in Figure 1.The scatterplot demonstrates a funnel-shaped pattern around themean effect size, with the exception that extreme positive andextreme negative effect size estimates may have been underrepre-sented in studies with small sample sizes. Because studies with lowsample sizes tend to remain unpublished, examination of potentialpublication bias was necessary (Rosenthal, 1979).
Assessment of Publication BiasPublication bias, sometimes called the file drawer effect
(Rosenthal, 1979), is a tendency for studies on a given topic withlarge (and/or statistically significant) effect size estimates to havea higher likelihood of being published than do studies that yieldsmaller (and/or statistically nonsignificant) effect size estimates.Because published studieswhich tend to possess strongest effectsize estimatesare easier for meta-analysts to obtain than areunpublished studies, the effect size estimates resulting from ameta-analysis can be biased against the null hypothesis.
To examine the extent to which our results might be biased bythe file drawer effect, we first compared the mean weighted effectsize from unpublished reports to the mean weighted effect sizefrom published reports. As seen in Table 3, studies that werepublished in journals (weighted mean r .096) yielded slightlylarger effect sizes than did studies that were unpublished (viz.,conference presentations, theses, and dissertations; weighted meanr .087). A mixed effects ANOVA conducted between pub-lished and unpublished studies revealed no statistically significantdifference (QB 0.15, p .70). Thus, although we found thatunpublished studies may yield slightly smaller estimates of thereligiousnessdepressive symptoms association than do publishedones, the amount of bias appears to be trivial (i.e., less than 1correlation point) and is statistically nonsignificant. Calculation ofa fail-safe N (Begg, 1994) showed that 4,128 studies averaging aneffect size of .00 would be required to render the present omnibuseffect size estimate statistically nonsignificant.
Because of limitations of the fail-safe N and similar approaches,nonparametric methods based on the funnel plot distribution (Fig-ure 1) have recently been developed to test and adjust for possiblepublication bias (Sutton et al., 2000). Duval and Tweedie (2000a,2000b) described the trim and fill method of estimating thenumber of missing studies due to publication bias and recalculat-ing the weighted mean effect size accordingly. In an iterativeprocess, outlying studies that have no corresponding values on theopposite side of the funnel plot distribution are temporarily re-moved (trimmed), and the mean effect size is recalculated,repeating the procedure until the distribution is symmetrical withrespect to the mean. In our analyses, we followed the recommen-dations of Duval and Tweedie (2000b) in using L0 to estimate thenumber of missing studies using formulae provided by Jennionsand Moller (2002). The final step in the procedure is to replace thetrimmed studies along with filled estimated values of the missingstudies on the other side of the funnel plot distribution. The valuesfor the filled studies are exactly opposite those trimmed. Theresulting data set inclusive of filled missing studies is then used to
Figure 1. Plot of effect sizes (Pearsons r) as a function of sample size(log).
623RELIGIOUSNESS AND DEPRESSIVE SYMPTOMS
-
calculate a new omnibus effect size and its confidence intervals,with statistically nonsignificant values indicating potential publi-cation bias. In the present study, the recalculated random effectsweighted mean effect size was .088 ( p .00001, 95% confi-dence interval .10, .07). Based on these analyses, publica-tion bias seems an unlikely threat to the results presented here.
Moderation by Sociodemographic VariablesTo examine whether the association of religiousness and depres-
sive symptoms varied as a function of various sociodemographicvariables, we conducted a series of categorical (and where appro-priate, continuous) moderator analyses, using the data from all 147studies (see Table 3). The amount of between-studies variance thatis accounted for by an individual moderator variable is indexed bythe QB statistic. Statistically significant QB values for an individualmoderator variable indicate that the mean effect sizes differ acrossvarious levels of the moderator variable. The amount of remaining
variance within studies is indexed by the QW statistic, for whichsignificant values indicate that the remaining variance is greaterthan zero.
To determine whether differences in the gender composition ofthe samples account for significant between-studies variance, wefirst correlated the percentage of females from the 137 studies thatreported participant gender with the corresponding average effectsize. This random effects weighted correlation was .001 ( p .98).We then analyzed effect sizes calculated with nonoverlappingfemale only or male only samples. As can be seen in Table 3, thisbreakdown by gender did not account for substantial between-studies variation (QB 0.01, p .98). To hold other variations inmethod constant, we then compared effect sizes from 18 studiesthat reported data separately for both men and women. However,the differences observed between the estimates using female par-ticipants (weighted mean r .134) and those using male partic-ipants (weighted mean r .138) were not statistically significant
Table 3Weighted Mean Correlations (r) Across Levels of Several Moderator Variables, 95% Confidence Intervals, and Amount of Between-Groups Variance Accounted For
Variable QB p N k r 95% CI QW p
Data source 0.15 .70Published 66,299 111 .096 .12, .08 156.0 .003Unpublished 32,676 36 .087 .12, .05 58.8 .007
Gender 0.01 .98Female 19,848 37 .123 .16, .08 45.9 .13Male 10,127 28 .122 .17, .08 40.3 .05
Ethnicity 0.51 .78African American 5,632 16 .108 .18, .04 13.8 .54European American 7,017 17 .073 .15, .01 15.6 .48European (Northern) 7,051 10 .105 .19, .02 10.0 .35
Age group 18.90 .009Adolescence (1318) 6,029 10 .067 .13, .01 29.3 .001College age (1924) 15,949 29 .065 .10, .03 27.4 .50Early adulthood (2535) 5,662 14 .092 .15, .03 9.7 .72Middle age (3645) 35,751 37 .098 .13, .06 53.7 .03Late middle age (4655) 4,299 9 .050 .12, .02 7.9 .44Late adulthood (5665) 6,621 11 .023 .09, .04 6.6 .76Retirement age (6675) 24,124 30 .155 .19, .12 54.6 .003Elderly adults (76) 540 7 .093 .20, .01 4.6 .60
Inferred life event stress 15.90 .0004Minimal 88,139 90 .071 .09, .05 116.1 .03Mild to moderate 5,785 32 .141 .18, .10 50.7 .01Severe 5,051 25 .152 .20, .11 38.5 .03
Inferred depression 10.00 .007Minimal 87,357 104 .078 .10, .06 134.3 .02Mild 5,748 25 .135 .18, .09 43.3 .01Moderate 5,870 18 .151 .21, .10 31.1 .02
Measure of religiousness 170.50 .0001Religious behaviors 43,572 30 .124 .16, .09 34.2 .23Religious attitudes and beliefs 16,725 18 .053 .10, .01 35.4 .006Religious orientation
Intrinsic 4,445 22 .175 .22, .13 53.8 .0001Extrinsic 6,361 23 .155 .11, .20 30.2 .11
Religious copingPositive 2,274 11 .167 .24, .10 7.8 .65Negative 1,999 8 .136 .06, .21 8.4 .30
Religious well-being 2,106 11 .197 .27, .12 24.5 .006God concept 1,352 4 .199 .31, .09 4.4 .22Multidimensional 28,345 51 .095 .13, .06 37.4 .91
Note. QB variance between groups; k number of studies; CI confidence interval; QW variance within groups.
624 SMITH, MCCULLOUGH, AND POLL
-
(QB 0.01, p .93). Therefore, the religiousnessdepressivesymptoms association does not appear to be moderated by gender.
To determine whether differences in the ethnic composition ofthe sample accounts for significant between-studies variance, wecorrelated the percentage of European Americans in the 95 studiesthat reported ethnicity with the corresponding average effect size.This random effects weighted correlation was .05 ( p .60). Wethen analyzed effect sizes calculated exclusively with samples of aspecific ethnic group. These categorical analyses were only pos-sible for contrasting the mean effect sizes for samples of AfricanAmericans, European Americans, and Northern Europeans be-cause fewer than three articles reported data specific to any otherethnic group. As may be seen in Table 3, this ethnic breakdown didnot account for significant between-studies variation in effect sizes(QB 0.51, p .78). To hold other variations in method constant,we then compared effect sizes from six studies that reported dataseparately for African American and European American partici-pants (no other ethnic groups were represented by separate datawithin studies). The differences observed between the estimatesusing African American participants (weighted mean r .132)and those using European American participants (weighted meanr .114) were not statistically significant (QB 0.02, p .88).Thus the relationship between religiousness and depressive symp-toms does not appear to be moderated by ethnicity.
To investigate the potential effect of participant age upon therelationship between religiousness and depressive symptoms, weconducted a random effects weighted correlation with the meanage of the sample and the effect size of the study, with the resultingcoefficient being .19 ( p .007). To further explore this relation-ship, we broke down participants average age into eight catego-ries. As can be seen in Table 3, the magnitude of the association ofreligiousness and depressive symptoms was highest at retirementage. However, the mixed effects ANOVA conducted between theage groups failed to reach the p .0035 level of statisticalsignificance (QB 18.90, p .009). Fewer than three studiesreported data broken down across equivalent age groups, makingit impossible to conduct more detailed analyses within studies.Nevertheless, the association of religiousness and depressivesymptoms does not appear to be moderated by age.
Main Effect Versus Stress-Buffering EffectsIf religiousness buffers people against the effects that stressful
life events can have on depression, then the association of reli-giousness and depression should be stronger at high levels of stressthan at low levels of stress (S. Cohen & Wills, 1985). If religious-ness exerts a main effect on depression, then the association ofreligiousness and depressive symptoms should be significant at alllevels of life stress. Of course, if both the buffering model and themain effect model are accurate, then both sets of conditions mayhold. We evaluated both hypotheses by inspecting the differencesin mean effect sizes obtained from samples with differing degreesof life events stress and with differing degrees of depressivesymptoms (which themselves are a source of significant lifestress).
The inferred amount of life event stress attributed to eachsample accounted for substantial between-studies variation(QB 15.90, p .0004). The religiousnessdepressive symptomsassociation for people with minimal life stress (weighted mean r
.071) was smaller in magnitude than was the association forpeople with mild to moderate (weighted mean r .141) or severelife stress (weighted mean r .152). The random effectsweighted correlation between rated life stress and mean effect sizewas .26, which was statistically significant ( p .0001). Thus,higher rated life stress was associated with stronger negativecorrelation of religiousness and depressive symptoms. These re-sults suggest that the negative association of religiousness anddepressive symptoms is at its strongest when people are undergo-ing stressful life events, which is consistent with the bufferinghypothesis. However, at all levels of inferred life stress the asso-ciation of religiousness and depressive symptoms was significantand negative, which is also consistent with the main effect hypoth-esis. Thus, our results provide support for both models of thereligiousnessdepressive symptoms association.
The degree of depressive symptoms attributed to each samplewas also investigated. The association among studies in which themean level of depression was minimal (weighted mean r .078)was smaller in magnitude than for samples that were judged ashaving a mild (weighted mean r .135) or moderate (weightedmean r .151) degree of depressive symptoms, but these dif-ferences were not statistically significant (QB 10.00, p .007)by the stringent criterion that we adopted for statistical signifi-cance ( p .0035). Thus, the rated level of depression did notmoderate the association between religiousness and symptoms ofdepression.
Insufficient numbers of studies reported data for nonoverlappingsamples with differing levels of stress and/or with differing levelsof depression. We were therefore unable to perform additionalanalyses to confirm these findings within studies. However, as anadditional step complementary to the above analyses, we providehere a narrative summary of the findings from the studies thatexplicitly assessed the influence of level of depression and distresson the religiousnessdepression association.
Only one study was identified that specifically assessed thereligiousnessdepression association across different levels of de-pression. In a survey of older adults, Klaas (1998) found that thecorrelation between religiousness and symptoms of depression was.29 among participants who were depressed compared with .07among participants who were not depressed.
Eight studies were identified that assessed the religiousnessdepression association as a function of life stress. In a survey ofbereaved parents, Maton (1989) reported that the correlation be-tween religiousness and depression was .33 among participantswho had experienced the recent death of a child (a group identifiedas experiencing high distress) compared with .16 among parentswho had lost a child more than 2 years previously (a groupidentified as experiencing lower levels of distress). Similarly,Robinson and Kaye (1994) found a correlation of .21 betweenreligiousness and depressive symptoms in a sample of caregiversof dementia patients compared with a correlation of .05 among asample of partners of healthy adults. Bickel et al. (1998) dividedparticipants into high- and low-stress groups based on scores froma measure of perceived stress and found that correlations amonghigh-stress participants averaged .27 whereas correlations amonglow-stress participants averaged zero. In two separate studies, C.Park, Cohen, and Herb (1990) found interactions between reli-giousness and life stress in the prediction of depression, withreligious coping buffering Catholics depression at high levels of
625RELIGIOUSNESS AND DEPRESSIVE SYMPTOMS
-
controllable life stress and with intrinsic religious orientation buff-ering Protestants depression at high levels of uncontrollable lifestress. Kendler et al. (1997) also found that religious devotionbuffered the relationship between recent stressful life events anddepression. Similarly, Hettler and Cohen (1998) reported a signif-icant interaction effect between intrinsic religiousness and thenumber of negative life events in predicting symptoms of depres-sion 8 months later, with the positive relationship between nega-tive life events and depression being weaker at higher levels ofintrinsic religiousness compared with moderate and low levels ofintrinsic religiousness. However, although Strawbridge, Shema,Cohen, Roberts, and Kaplan (1998) reported that both organiza-tional and nonorganizational religiousness buffered nonfamilystress (medical problems, financial problems, etc.), a finding rep-licated by those results reported above, they also found that in thepresence of family distress organizational religiousness (but notnonorganizational religiousness) was positively correlated withdepressionthe opposite direction of their other results and theresults of the studies cited previously. Similarly, although Belavich(1995) reported that religious coping activities predicted adjust-ment to daily hassles and stressful life events when controlling fornonreligious coping, there were no interaction effects between thenumber of daily hassles and religious coping when predictingdepression. Finally, when dividing participants into three catego-ries of church attendance (weekly, less often, and never), Thearle,Vance, Najman, Embelton, and Foster (1995) found little differ-ence in the magnitude of the religiousnessdepression associationamong bereaved parents (r .10) compared with a control grou