Society and Mental Health 2018, Vol. 8(3) 214–230 Should I Stay or Should I Go? Ó American Sociological Association 2017 DOI: 10.1177/2156869317748713 Religious (Dis)Affiliation and journals.sagepub.com/home/smh Depressive Symptomatology

Matthew May1

Abstract Religious affiliation is generally associated with better mental health. The nonreligious, however, currently constitute one of the fastest-growing religious categories in the United States. Since most of the nonre- ligious were raised in religious homes, their growth raises important questions about the mental health of those who consider dropping out of . In this article, I use longitudinal data from the Portraits of American Life Study to examine the impact of religious affiliation on mental health. Specifically, I compare individuals who dropped out of religion (leavers) with individuals who considered dropping out (stayers) and individuals who are more consistent in their religious (stable affiliates) and nonreligious (stable Nones) affiliations. I find that stayers experience more depressive symptoms than any other group and that they experience a greater increase in depressive symptoms over time. My findings are consistent with identity theories in sociology, and they provide evidence that a strong religious or secular identity is an important contributor to mental health.

Keywords religion, depression, Portraits of American Life Study (PALS), identity theory

Nones—atheists, agnostics, and the unaffiliated— (Koenig 2012), but these benefits are not universal occupy one of the fastest growing “religious” (Ellison 1995). A number of circumstances limit, groups in the United States today. According to or even reverse, the positive relationship between recent reports, nearly one in four Americans is religious participation and mental health. On the not affiliated with a religious institution (Pew one hand, disaffiliation is associated with more Forum on Religion and Public Life 2015). Inter- negative health outcomes (Fenelon and Danielsen estingly, though, nearly three quarters of these 2016; Hayward et al. 2016). On the other hand, women and men grew up in religious homes (Kos- security in a particular nomos—whether the can- min and Keysar 2008). This means that the vast opy is sacred or secular—appears to improve men- majority of Nones are religious dropouts who tal and physical health (Mochon, Norton, and actively left religion behind. Are these women Ariely 2011; Schnittker 2001). Thus, the decision and men making the best decision for their mental to drop out of may play an well-being? What about those who consider drop- ping out but decide to stay? The extraordinary rate of growth among the 1Oakland University, Rochester, MI, USA Nones in the United States raises important ques- Corresponding Author: tions about the health and well-being of those who Matthew May, Department of Sociology, Anthropology, choose to disaffiliate from religious institutions. Social Work, and Criminal Justice, Oakland University, Active involvement in a religious community is 371 Varner Drive, Rochester, MI 48309-4485, USA. generally associated with better health outcomes Email: [email protected] May 215 important role in the relationship between religion The growing number of religiously unaffiliated and mental health. Moreover, since more people Americans raises important questions about the will consider dropping out of religion than will process of leaving organized religion behind. actually leave, it is important to consider the dif- Nationally representative data from the Portraits ferences between those who depart and those of American Life Survey (Emerson and Sikkink who stay. 2012) indicate that 13 percent of religiously affil- In this paper, I use longitudinal data from the iated Americans considered dropping out of reli- Portraits of American Life Study (PALS) to exam- gion altogether between 2003 and 2006. Individu- ine the relationship between religious affiliation als who consider dropping out of religion and mental health. Religious affiliation varies altogether are generally motivated by religious from unwavering devotion to a religious group skepticism, political attitudes, and the experience (or no group at all) to the decision to completely of stressful life events (Vargas 2012). abandon organized religion. Therefore, I compare According to Vargas (2012), religious skepti- individuals who considered dropping out of reli- cism or a recent financial crisis increases the gion but remain affiliated with a religious group odds that a person will consider opting out of reli- (stayers) to individuals who never considered gion. Certain political attitudes tend to have oppo- dropping out of religion (stable affiliates), individ- site effects on religious commitment, though. uals who were never affiliated with a religious More specifically, support for Christian political group (stable Nones), and individuals who drop- groups reduces the odds that women and men ped out of religion after being affiliated with a reli- will consider dropping out of religion altogether, gious group (leavers). I find that stayers report while “support for same-sex marriage is positively more depressive symptoms than any other group associated with considering and actually leaving and that they experience a greater increase in religion” (Vargas 2012:214). Men and whites are depressive symptoms over time. My findings are also more likely to consider leaving religion alto- consistent with other research on identity and gether, but according to Vargas (2012) only 38 mental health (e.g., Cast and Burke 2002; Large percent of the people who consider dropping out and Marcussen 2000), and they provide evidence of religion altogether actually leave. This means that a strong religious or secular identity is an that between 2003 and 2006, 8 percent of reli- important contributor to mental health. giously affiliated Americans—or more than 6 per- cent of the U.S. population—considered opting out of religion but remained affiliated with an RELIGIOUS AFFILIATION organized religious group. Though disparate in their religious affiliations, The number of Nones has grown from less than 10 “stayers” constitute a group of people that is larger percent in the early 1990s to nearly a quarter of the than atheists (3.1 percent), agnostics (4.0 percent), U.S. population (Pew Forum on Religion and Pub- and all of the major non-Christian religious tradi- lic Life 2015). Not surprisingly, this trend has tions in the United States (5.9 percent total; sparked many attempts to understand who these includes Jews, , Muslims, and Buddhists) Nones are and what they believe (e.g., Baker and (Pew Forum on Religion and Public Life 2015). Whitehead 2016; Manning 2013; Taylor and With the exception of Vargas (2012), whose com- McPeek 2013). One Pew Forum study, for exam- parison of “stayers,” “leavers,” and “stable affili- ple, suggests that Nones are disproportionately ates” highlights the mechanisms that push leavers concentrated among young adults (Pew Forum out of religion and pull stayers back into the pews, on Religion and Public Life 2010), and another though, stayers are generally ignored amid a grow- indicates that a sizable majority—68 percent— ing body of empirical research focused on the grow- still believe in (Pew Forum on Religion and ing population of Nones. To date, no empirical Public Life 2012). Today, Nones are the second research directly examines the mental health of largest “religious” group in the United States stayers compared to stable affiliates and those who (22.8 percent) behind Evangelical Protestants opt out of religion altogether. Identity theories in soci- (25.4 percent) and just ahead of Catholics (20.8 ology and research on the relationship between orga- percent) (Pew Forum on Religion and Public nized religion and mental health suggest there may be Life 2015). important differences between these groups, though. 216 Society and Mental Health 8(3)

IDENTITY THEORY AND MENTAL clear, though, that religion does not improve the mental health of all participants. HEALTH Several conditions limit, or even reverse, the Identity theory provides a useful framework for positive relationship between religion and mental understanding the complex relationship between health. According to Ellison and Lee (2010), there religious affiliation and mental health. According is a “dark side” to religion (see also Krause and to identity theory, the self is the composite of mul- Wulff 2004). More specifically, disturbances tiple identities reflecting the many roles that an such as doubt (Ellison and Lee 2010; Krause individual occupies in the society (Stryker 1980). 2006; Krause and Ellison 2009), disaffiliation Each identity is a control system composed of (Fenelon and Danielsen 2016; Hayward et al. four components: an input, an identity standard, 2016), moderate levels of commitment (Eliassen a comparator, and an output. Together, these com- et al. 2005), and negative interactions in the ponents function to regulate the meanings per- Church (Ellison et al. 2009) tend to diminish the ceived by the identity in the way a thermostat otherwise positive relationship between religion functions to regulate the temperature it perceives and mental health. According to Speed and (Burke and Stets 2009). More specifically, people Fowler’s (2017) recent study of Ontario’s adult alter their behaviors (the outputs) in order to population, the relationship between religion and achieve semantic congruence between their self- mental health is also moderated by religious affil- perceptions of meaningful feedback from others iation. More specifically, there is a positive rela- (the inputs) and the meanings attached to the iden- tionship between church attendance and well- tity (the identity standard). When the comparator being for Christians and a negative relationship detects a disturbance between the input and the between church attendance and well-being for identity standard (i.e., an incongruence), distress the unaffiliated (Speed and Fowler 2017). The is experienced in the form of negative emotions relationship between church attendance and men- such as depression or anxiety (Burke 1991, tal health also varies within (Schwa- 1996). The experience of negative emotions is del and Falci 2012) . Comparing Nebraska’s main- more pronounced when core (i.e., salient) identi- line Protestant, evangelical Protestant, and ties are threatened (Stryker 2002). Catholic populations, Schwadel and Falci (2012) For many, religion is a core identity, “affecting find that only mainline Protestants report fewer how they understand themselves as religious or depressive symptoms when they attend church spiritual beings . . . as well as determining their more often. social identification with a particular religious A couple of recent studies make use of identity group” (Park and Edmondson 2012:150). Not sur- theory to explain the contingent relationship prisingly, the majority of empirical research on between religion and mental health. First, Ellison religion and mental health shows that people et al. (2013) used identity theory to explain why who are more better mental the pernicious association between doubts about health than people who are less religious (Koenig God and psychiatric symptoms is most pro- 2012). Some notable examples reveal a positive nounced among people who expressed strong reli- relationship between religious engagement and gious commitments. Second, Galek and her col- subjective well-being (Ellison 1991; Ellison, leagues used identity theory to explain why the Gay, and Glass 1989; Lim and Putnam 2010) that life lacks meaning and purpose caused and a negative relationship between religious more psychological distress among the highly reli- engagement and various forms of psychological gious than among the less religious or the highly distress (Acevedo, Ellison, and Xu 2014; Ellison religious who did not believe that life lacks mean- et al. 2012; Eliassen, Taylor, and Lloyd 2005). ing or purpose (Galek et al. 2015). Both of these Acevedo et al. (2014), for example, find that reli- studies echo prior research on the conditional rela- gious involvement buffers the deleterious effects tionship between core religious identities and of perceived financial strain and neighborhood mental health (e.g., Krause 1994, 1998; Krause disadvantage on psychological distress. Similarly, and Wulff 2004), and all of these studies are con- Ellison et al. (2012) find that a secure attachment sistent with Stryker’s (2002) argument that nega- to God buffers against the pernicious effects of tive emotions are more pronounced when more stressful life events on distress. It is increasingly salient roles are threatened. May 217

My hypotheses are based on the propositions of affiliated) and the new identity (i.e., unaffiliated). Burke and Stets’s (2009) identity theory and the By establishing an ex-role, disaffiliates can reduce growing body of research on religion and mental the negative emotions caused by doubt and other health described above.1 Generally speaking, dis- disturbances. This process can be understood turbances produce negative emotions like anxiety using Cast and Burke’s (2002) extension of iden- and depression (Burke 1991, 1996). More specifi- tity theory to explain the self-verification process. cally, religious doubts are the single largest pre- Self-verification occurs when there is semantic dictor of who considers dropping out of religion congruence between the self-perceptions of mean- (Vargas 2012), and they tend to increase psycho- ingful feedback from others (the inputs) and the logical distress (Krause 1994, 1998; Krause and meanings attached to the identity (the identity Wulff 2004). Thus, I hypothesize that people standard) (Burke and Stets 2009). Generally, who consider dropping out of religion altogether changes in behavior (outputs) are enough to bring will report more negative health outcomes than these perceptions back into alignment with the those who do not. identity standard (Burke 1991, 1996). In the case of persistent disturbances, however, people may Hypothesis 1: Individuals who consider drop- abandon the identity in order to avoid the negative ping out of religion altogether will experi- emotions caused by the disturbance (Cast and ence more depressive symptoms than indi- Burke 2002). viduals who do not consider dropping out Some identities are easier to abandon than of religion altogether. other identities, though (Thoits 2003). Ebaugh Hypothesis 2: Individuals who consider drop- (1988) points out that some people never move ping out of religion altogether will experi- out of the first stage of becoming an ex. Regarding ence more depressive symptoms than indi- religious affiliation, this includes people who con- viduals who were never affiliated with sider dropping out of religion altogether but do not a religious institution. actually leave (i.e., stayers). People who consider dropping out of religion altogether and actually Considering dropping out of religion altogether leave, on the other hand, are able to establish is only the first step in the disaffiliation process. In a new ex-role (i.e., leaver). According to role exit order to disaffiliate, a person must actually leave. theory then, stayers are more likely to experience Research on religion and mental health suggests prolonged disturbances within a core identity, and that completing the disaffiliation process can be they are also more likely to experience prolonged better for the mental health of people who consider distress (Cast and Burke 2002; Ebaugh 1988). dropping out of religion altogether. Schnittker Therefore, I hypothesize that people who begin the (2001), for example, describes an inverted U- disaffiliation process but remain affiliated will report shaped relationship between religious salience more depressive symptoms than people who actually and depression. In other words, the negative rela- leave, and they will report a greater increase in neg- tionship between religion and depression is stron- ative health outcomes over time than people who gest when religion is very important and when complete the disaffiliation process. religion is not important at all. Similarly, Ross (1990) finds that stronger religious beliefs are Hypothesis 3: Individuals who consider drop- associated with lower levels of distress than ping out of religion altogether but remain weaker religious beliefs, while no religious beliefs affiliated with a religious institution are also associated with low levels of distress (see (stayers) will experience more depressive also Eliassen, Taylor, and Lloyd 2005). symptoms than individuals who drop out The research described above is consistent with of organized religion altogether (leavers) Ebaugh’s (1988) role exit theory and the self- after a period of several years. verification process (Cast and Burke 2002). Role Hypothesis 4: Individuals who consider drop- exit is “the process of disengagement from a role ping out of religion altogether but remain that is central to one’s self-identity and the rees- affiliated with a religious institution tablishment of an identity in a new role that takes (stayers) will experience a larger increase into account one’s ex-role” (Ebaugh 1988:1). in depressive symptoms than individuals Therefore, disaffiliation is the process of creating who drop out of organized religion alto- an ex-role that incorporates the old identity (i.e., gether (leavers). 218 Society and Mental Health 8(3)

DATA AND METHODS experienced the symptom. I combined these meas- ures to produce a count of respondents’ depressive To examine the relationship between religious symptoms during the past 12 months in 2006 affiliation and depressive symptomatology, I use (Chronbach’s a = .824) and 2012 (Chronbach’s longitudinal data from PALS (Emerson and Sik- a = .826). The variable ranges from 0 to 3. It is kink 2012). PALS is a nationally representative negatively skewed, with more than half of the multistage panel study of 2,610 American adults wave 2 sample (57.6 percent) reporting no depres- (wave 1, 2006) with follow-up data on 1,314 of sive symptoms lasting for a period of more than the original respondents (wave 2, 2012). PALS’s two weeks in the 12 months prior to the 2006 multistage design generated a probability sample and 2012 PALS interviews. with a response rate of 58 percent at wave 1; this includes oversamples of African Americans, Hispanics, and Asian Americans. Specifically, Independent Variables PALS is designed to address the impact of religion My key independent variable measures respond- and religious change across a number of individual ents’ religious affiliation at waves 1 and 2 of the and family-level experiences. PALS is particularly PALS data set. PALS respondents were asked to well suited for the present study because it is the provide their religious affiliation at wave 1 and only nationally representative longitudinal study wave 2. In total, 1,280 respondents provided their with questions about the religious disaffiliation religious affiliation at both waves. Of these process and mental health.2 respondents, 1,028 (80.3 percent of the sample) The longitudinal nature of PALS makes it pos- were affiliated with a specific religious institution sible to model the effect of changes in religious at wave 2. A total of 252 respondents (19.7 percent) affiliation on mental health. Specifically, the reported no religious affiliation at wave 2. The per- wave 1 interview guide includes a 38-question centage unaffiliated in the PALS data set is highly health module gauging respondents’ physical, consistent with other summations of the total num- mental, and subjective well-being. During the ber of Nones in the United States in 2012 (e.g., wave 2 interview, respondents were asked the Pew Forum on Religion and Public Life 2012). same set of questions about their physical, men- “Affiliated” and “unaffiliated” do not completely tal, and subjective health. Since PALS includes capture the nuance in these two groups, though. interviews with the same respondents at two The unaffiliated in this sample include 178 points in time, I am able to examine the effect respondents with no religious affiliation at wave of religious affiliation on mental health at two 1 and 74 respondents who cut ties with a specific points in time and the impact of religious affilia- religious institution between waves 1 and 2. The tion on the change in mental health between affiliated, on the other hand, include 138 respond- waves. Thus, the current study provides a better ents who seriously considered dropping out of reli- test of the hypothesis that religious affiliation gion between 2003 and 2012. More specifically, affects people’s mental health than studies utiliz- PALS respondents were asked the following two ing cross-sectional data on religion and mental questions at waves 1 and 2, respectively: “In the health. past three years, have you seriously considered dropping out of religion altogether?” and “In the past six years, have you seriously considered drop- ping out of religion altogether?” Combining both Dependent Variable questions with questions about respondents’ reli- My dependent variable is a count of the number of gious affiliation at waves 1 and 2 makes it possible depressive symptoms experienced by respondents to create a multinomial variable of respondents’ measured at two points in time. PALS includes religious affiliation over a nine-year span.3 Fol- three measures of behaviors commonly associated lowing Vargas (2012), I refer to individuals who with depression. Specifically, PALS respondents considered dropping out of religion altogether were asked if they experienced feelings of depres- but were still affiliated with a specific religious sion, worthlessness, or hopelessness for a period group at wave 2 as stayers, and I refer to individ- lasting more than two weeks during the past 12 uals with a religious affiliation at wave 1 and no months. Each of these variables is a dichotomous religious affiliation at wave 2 as leavers. My anal- measure wherein 1 indicates that the respondent yses also include two unique, but stable, groups. May 219

Table 1. Descriptive Statistics for All Study Variables.

Wave 1 (n =1,183) Wave 2 (n = 926) Mean/ Standard Mean/ Standard Variable Proportion Deviation Proportion Deviation Range

Dependent variable Depressive symptoms scale 0.57 1.02 0.53 0.98 0–3 Lagged (wave 1) 0.57 1.02 0–3 Depression 0.27 0.44 0.25 0.43 0–1 Lagged (wave 1) 0.26 0.44 0–1 Hopelessness 0.15 0.36 0.16 0.36 0–1 Lagged (wave 1) 0.15 0.35 0–1 Worthlessness 0.15 0.35 0.13 0.33 0–1 Lagged (wave 1) 0.16 0.37 0–1 Religious affiliation Stable affiliate 0.77 0.42 0.70 0.46 0–1 Stayer 0.09 0.29 0.11 0.31 0–1 Leaver 0.04 0.20 0.06 0.25 0–1 Stable None 0.10 0.30 0.12 0.33 0–1 Control variable Religious service attendance 3.71 2.17 3.64 2.28 1–7 Age 44.84 15.96 50.18 15.8 18–80a Female 0.51 0.50 0.52 0.50 0–1 White 0.69 0.46 0.70 0.46 0–1 Married 0.58 0.49 0.58 0.49 0–1 Education (1 = some college) 0.38 0.49 0.45 0.50 0–1 Income 3.16 1.67 3.25 1.78 1–6 Region (1 = South) 0.33 0.47 0.34 0.47 0–1 Subjective health 3.48 1.10 3.35 1.08 1–5 a24–80 (wave 2).

Stable affiliates are religious affiliates who did not 2010; Ellison 1991; Fenelon and Danielsen seriously consider dropping out of religion alto- 2016). Some people retain their affiliation with gether between 2003 and 2012. Stable Nones are a religious group without actively participating respondents with no religious affiliation at wave in any religious organization; others continue to 1 or wave 2. Due to the small number of cases, participate in religious activities when they are the wave 2 analyses do not include women and unaffiliated.4 This phenomenon makes it impor- men who were unaffiliated at wave 1 but are affil- tant to consider the divergent experiences of active iated with a religious organization at wave 2. The and inactive affiliates (both stable affiliates and mean for each group in Table 1 is the percentage stayers) and active and inactive disaffiliates of each group in the sample after accounting for (both leavers and stable Nones). Religious service missing data on the other variables in my analyses. attendance is an ordinal measure ranging from 1 (never)to7(more than once a week). In addition to religious service attendance, I Control Variables also control for age, sex (1 = female), race (1 = In each of my models, I control for multiple meas- white), marital status (1 = married), education ures related to religious commitment and mental (1 = some college), household income, region of health. Each of the variables described below is the country (1 = South), and respondents’ subjec- available in both waves of PALS. Religious ser- tive health. PALS respondents were asked to rate vice attendance is a common predictor of mental their overall health from 1 (poor)to5(excellent). health (Barkan and Greenwood 2003; Childs Self-rated health is generally considered a measure 220 Society and Mental Health 8(3) of physical health (Krause and Jay 1994) and is comparing the means across groups does not a good indicator of respondents’ quality of life account for possible covariates that might explain (Jylha¨ 2009). Several of the models described any observed relationship between religious affil- below also include a lagged measure of the depen- iation and mental health. Therefore, I also estimate dent variable (i.e., the count of depressive symp- a series of regression models controlling for age, toms reported at wave 1). Controlling for respond- sex, race, marital status, education, household ents’ mental health at wave 1 allows me to income, region of the country, subjective health, examine the effect of religious affiliation on and religious service attendance, as described changes in mental health over time. This is an above. important strength over cross-sectional studies of In order to account for the clustered sample religion and health. Table 1 shows the range, design of this dataset, I use the “svy” command mean, and standard deviation for all of the varia- in STATA 14.0 for each of the regressions bles described above. described below. Since PALS oversamples Afri- can Americans, Asians, and Hispanics, I also use sample weights that more accurately reflect the Missing Data racial makeup of the United States in each of my analyses. The weight variable also accounts for Wave 2 of PALS includes 1,314 cases. There are issues of nonresponse and the likelihood of being 1,183 cases with complete data at wave 1 and selected based on family size (Emerson, Sikkink, 926 cases with complete data at wave 2. The and James 2010). The first analysis examines the majority of missing cases (7 at wave 1 and 299 relationship between religious affiliation and at wave 2) are missing on the dependent variable. depressive symptomatology at wave 1. The second Another 44 cases (10 at wave 1 and 34 at wave 2) analysis examines the same relationship at wave 2. are missing on the key independent variable (reli- The final analysis examines the change in depres- gious affiliation). I ran two separate analyses: one sive symptomatology between waves across the for the cases without missing data, and one using different types of religious affiliation. multiple imputation in STATA 14.0 to impute My first set of analyses tests the hypothesis that the eligible cases that were missing on one or stayers and leavers will experience more depres- more of the control variables (see Allison 2009 sive symptoms than stable affiliates and stable for a discussion of cases eligible for imputation Nones. Examining the relationship between reli- with multiple imputation). The results of the gious affiliation and depressive symptomatology imputed analyses produced results that are sub- in the cross-section at wave 1 shows the short- stantively and significantly similar to the analysis term consequences of disturbances in religious of complete cases. Therefore, I present only results affiliation on mental health. Identity theories in for the primary analyses below since they are sociology (see Burke and Stets 2009) and a more conservative estimate of the relationship Ebaugh’s (1988) role exit theory, however, are between religious affiliation and mental health. clear that any mental health problems observed in 2006 (wave 1) should be resolved by 2012 ANALYTIC STRATEGY (wave 2) if the disturbance is removed and an ex-role identity is established. Therefore, my sec- I begin by plotting the pattern of change in mental ond set of analyses provides a test of my hypoth- health between waves for stable affiliates, stayers, esis that stayers will experience more depressive leavers, and stable Nones. Comparing the average symptoms after a period of several years by exam- number of depressive symptoms across all four ining the relationship between religious affiliation groups in 2006 and 2012 provides a simple test and depressive symptomatology at wave 2. of all four of my hypotheses: (1) Do stayers and My final set of analyses tests the hypothesis leavers experience more depressive symptoms that stayers will experience a greater increase in than stable affiliates? (2) Do stayers and leavers depressive symptoms than leavers over time. In experience more depressive symptoms than stable order to examine the change in depressive symp- Nones? (3) Do stayers experience more depressive toms between waves, I regress mental health on symptoms than leavers after several years? and (4) religious affiliation and respondents’ baseline Do stayers experience a greater increase in depres- mental health. Including depressive symptomatol- sive symptoms than leavers over time? Of course, ogy at wave 1 shifts the interpretation of the other May 221

Figure 1. Mean depressive symptoms by wave and change in religious affiliation. covariates in the model to the prediction of change differences between the model coefficients (e.g., in depressive symptomatology. The assumed stayers vs. leavers). causal effect of Y1 on Y2 plus the time-invariant covariates such as race and sex make the lagged RESULTS dependent variable model the preferred model to test the effect of religious affiliation on changes Figure 1 plots the pattern of change in depressive in mental health over time (Finkel 1995; but also symptoms between waves for stable affiliates, see Johnson 2005).5 stayers, leavers, and stables Nones. Not surpris- The primary dependent variable in all of my ingly, stable affiliates report the lowest number analyses, the number of depressive symptoms of depressive symptoms at wave 1; stable affiliates experienced in the past 12 months, is a count and leavers are also the only groups to report that ranges from 0 to 3. Count variables typically a decrease in the number of depressive symptoms require a Poisson or negative binomial model between waves. Stayers and stable Nones, on the (Long 1997). Unlike the Poisson model, the nega- other hand, report a modest increase in depressive tive binomial regression model does not assume symptoms between waves. Notably, stayers report that the conditional mean is equal to the condi- the highest number of depressive symptoms at tional variance, a highly restrictive assumption. wave 1 (0.93 or 31 percent of all symptoms) and Therefore, I use the negative binomial regression wave 2 (0.98 or 33 percent of all symptoms); model in order to account for overdispersion.6 and both stayers and stable Nones report an Following the negative binomial regression, I increase in the number of depressive symptoms examine each component of the count variable experienced between waves. Based on these unad- separately to determine whether the results of the justed means, it appears there is support for my negative binomial regressions are driven by a com- hypotheses that individuals who consider dropping ponent of the count-dependent variable. Each out of religion altogether will experience more component of my primary dependent variable is depressive symptoms than individuals who do a dichotomous measure of respondents’ mental not consider dropping out of religion altogether health. The construction of the count variable and individuals who were never affiliated with gives equal weight to each of these measures. a religious institution. There is also support for Analyzing each of these measures separately is my hypothesis that stayers will experience more a way to test whether the count variable is biased depressive symptoms than leavers when there is by any of the unique measures. I use a series of time for the latter to establish an ex-role identity. logistic regression models to test the effect of reli- More specifically, stayers and leavers reported gious commitment on each of these dichotomous the greatest number of depressive symptoms at outcomes. For each model, the results of postanal- wave 1, and stayers reported the greatest number ysis Wald tests are presented in order to test for of symptoms at wave 2. In addition, the change 222 Society and Mental Health 8(3)

Table 2. Mental Health Regressed on Religious Affiliation, Wave 1 (n = 1,183).

Depressive Symptomsa Depressionb Hopelessnessb Worthlessnessb Model 1 Model 2 Model 3 Model 4

Religious commitment Stayer 0.69*** 0.95** 1.32** 1.17** Leaver 0.33 0.25 0.91 0.73 Stable None 0.03 20.11 0.21 0.53 Religious behavior Religious service attendance 20.02 20.07 20.02 0.04 Sociodemographics Age 20.01** 20.02** 20.02* 20.02 Female 0.34* 0.52* 0.47 0.31 White 0.21 0.34 0.42 0.36 Married 20.24 20.31 20.06 20.50 Education (1 = some college) 20.11 0.08 20.18 20.16 Income 20.10* 20.08*** 20.23** 20.17 Region (1 = South) 0.11 0.16 0.25 0.23 Health Subjective health 20.36*** 20.48** 20.62*** 20.71*** Wald test Stayer vs. leaver ns ns ns ns Stayer vs. stable None ****ns Leaver vs. stable None ns ns ns ns aNegative binomial coefficients. bLogistic regression coefficients. *p \ .05. **p \ .01. ***p \ .001 (two-tailed tests).

in the unadjusted means between waves in Figure leavers after a period of several years. Each model 1 provides support for my hypothesis that stayers in Tables 2 and 3 shows the effect of religious will experience a greater increase in depressive affiliation on a unique dependent variable. The symptoms between waves than leavers will. As first model in both tables displays the results for previously noted, stayers report an increase in the negative binomial model. Each subsequent depressive symptoms between 2006 and 2012, model in Tables 2 and 3 displays the results of but leavers report a decrease in depressive symp- a logistic regression model predicting one of the toms during the same period. Of course, the pat- depressive symptoms that make up the count vari- tern in Figure 1 does not account for the possible able. The coefficients in the negative binomial covariates, nor does it show whether the relation- regression models represent the percentage change ships observed are statically significant. The mod- in depressive symptoms for a unit change in the els in Tables 2 through 4, however, provide sup- independent variable after computing [expb – 1]. port for all of my hypotheses and confirm most The coefficients in the logistic regression models, of the patterns shown in Figure 1. on the other hand, represent the odds of experienc- The models in Table 2 (wave 1) are tests of the ing the dependent variable after computing hypotheses that people who consider dropping out [expb]. Similar to Figure 1, the models in Table of religion altogether will experience more depres- 2 provide partial support for the hypotheses that sive symptoms than individuals who do not con- stayers and leavers will report more depressive sider dropping out of religion altogether and indi- symptoms than stable affiliates and stable Nones, viduals who were never affiliated with a religious and the models in Table 3 provide clear support institution. The models in Table 3 (wave 2), on the for the hypothesis that stayers will experience other hand, are tests of the hypothesis that stayers more mental health problems than leavers over will experience more depressive symptoms than time. May 223

Table 3. Mental Health Regressed on Religious Affiliation, Wave 2 (n = 926).

Depressive Symptomsa Depressionb Hopelessnessb Worthlessnessb Model 1 Model 2 Model 3 Model 4

Religious commitment Stayer 0.97*** 1.16*** 1.52*** 1.41** Leaver 20.19 20.33 20.21 20.02 Stable None 20.15 20.01 20.44 20.25 Religious behavior Religious service attendance 20.12** 20.11* 20.15* 20.13 Sociodemographics Age 20.02** 20.03** 20.02 20.03* Female 0.00 0.02 20.18 20.02 White 0.31 0.43 0.63 0.83* Married 20.37 20.63** 20.25 20.32 Education (1 = some college) 20.12 20.05 20.35 20.33 Income 20.05 20.09 20.05 20.07 Region (1 = South) 0.21 0.20 0.50 0.13 Health Subjective health 20.36*** 20.43** 20.62*** 20.55** Wald test Stayer vs. leaver *** * * * Stayer vs. stable None *** ** *** *** Leaver vs. stable None ns ns ns ns aNegative binomial coefficients. bLogistic regression coefficients. *p \ .05. **p \ .01. ***p \ .001 (two-tailed tests).

The models in Table 2 indicate that people who depressed and hopeless than stable affiliates. seriously considered dropping out of religion Leavers, on the other hand, are not significantly between 2003 and 2006 but were still affiliated different from stable affiliates or stable Nones, with a religious institution in 2006 experienced and they are less likely than stayers to report feel- more depressive symptoms than religious affiliates ings of depression or hopelessness. who did not seriously consider dropping out of Models 1 through 4 in Table 3 show a similar religion. Being a stayer at wave 1 increases the pattern. The difference, however, is that religious expected number of depressive symptoms by a fac- affiliation in measured over a nine-year period tor of 1.98 (expb) compared to stable affiliates (from 2003 to 2012). Model 1 in Table 3 indicates (p \ .001). A postanalysis Wald test also indicates that being a stayer from 2003 to 2012 increases the that the difference in the expected number of expected number of depressive symptoms in 2012 depressive symptoms for stayers and stable Nones by a factor of 2.63 (expb) compared to stable is statistically significant (p \ .05). On the other affiliates (p \ .001). Older persons, people in bet- hand, being a leaver does not significantly ter physical health, and people who attend reli- increase the expected number of depressive symp- gious services more often also report fewer toms compared to stable affiliates or stable Nones. depressive symptoms. The latter variable demon- These same patterns repeat across all of the logis- strates the unique relationship between religious tic regression models in Table 2 with one excep- affiliation and mental health. That is, net of reli- tion: The coefficient for stayers is not significantly gious service attendance, the expected number of different from the coefficient for stable Nones in depressive symptoms is higher for stayers than it model 4. Thus, stayers are more likely to feel is for stable affiliates. Like model 1 in Table 2, depressed, hopeless, and worthless than stable however, the differences between stable affiliates affiliates, and they are more likely to feel and leavers in model 1 of Table 2 is not 224 Society and Mental Health 8(3)

Table 4. Mental Health Regressed on Religious Affiliation and Prior Mental Health, Wave 2 (n = 926).

Depressive Symptomsa Depressionb Hopelessnessb Worthlessnessb Model 1 Model 2 Model 3 Model 4

Religious commitment Stayer 0.70** 0.87** 1.30*** 1.18* Leaver 20.08 20.23 20.11 0.07 Stable None 20.25 20.08 20.51 20.35 Religious behavior Religious service attendance 20.13*** 20.12* 20.17* 20.14 Sociodemographics Age 20.02** 20.02** 20.02 20.03* Female 20.11 20.13 20.35 20.18 White 0.27 0.31 0.52 0.73* Married 20.25 20.51* 20.10 20.18 Education (1 = some college) 20.06 0.12 20.22 20.20 Income 20.03 20.06 20.03 20.04 Region (1 = South) 0.16 0.13 0.43 0.04 Health Subjective health 20.27** 20.29* 20.49** 20.44* Lagged dependent variable 0.39*** 0.64*** 0.52*** 0.47*** Wald test Stayer vs. leaver * ns ns ns Stayer vs. stable None *** * *** ** Leaver vs. stable None ns ns ns ns aNegative binomial coefficients. bLogistic regression coefficients. *p \ .05. **p \ .01. ***p \ .001 (two-tailed tests).

statistically significant. This pattern persists across who drop out of organized religion altogether all of the models in Table 3 and is consistent with (leavers) over time. To test if stayers also experi- the assumption that stayers will fare the worst over enced a greater increase in depressive symptoms a prolonged period of time (hypothesis 3). than leavers did between 2006 and 2012, though, Compared to stable affiliates, stayers are more the models in Table 4 include a lagged measure likely to report having felt depressed in the past 12 of the dependent variable in models 1 through 4 months, they are more likely to report having felt of Table 3. worthless in the past 12 months, and they are more The inclusion of the lagged dependent variable likely to report having felt hopeless in the past 12 in the models in Table 4 shifts the interpretation of months. Postanalysis Wald tests for all of the mod- the other covariates in the negative binomial els in Table 3 also indicate that people who con- model and the logistic regression models to the sidered dropping out of religion between 2003 prediction of change in depressive symptoms and 2012 experienced poorer mental health in between waves. These models provide a test of the 12 months prior to the 2012 PALS interview my hypothesis that stayers will experience a larger when they remained affiliated with a religious increase in depressive symptoms than leavers will. group (stayers) than when they abandoned orga- Being a stayer increases the expected change in nized religion between 2006 and 2012 (leavers). the number of depressive symptoms by a factor Together, these models provide clear support for of 2.01 (expb) compared to stable affiliates. The the hypothesis that individuals who consider drop- postanalysis Wald tests for model 1 in Table 4 ping out of religion altogether but remain affili- also indicate that stayers experience a greater ated with a religious institution (stayers) will expe- increase in the expected number of depressive rience more depressive symptoms than individuals symptoms than leavers (p \ .05) even though May 225 the difference between stayers and leavers is not dropping out of religion actually leave. Vargas statistically significant in models 2 through 4.7 based his assessment on data from wave 1 of Thus, model 1 in Table 4 provides support for PALS, but this still appears to be the case six years my hypothesis that stayers will experience a larger later. More than 23 percent of religiously affiliated increase in depressive symptoms than leavers will, Americans seriously considered dropping out of and it confirms the pattern shown in Figure 1. religion between 2003 and 2012. At wave 2 of PALS, 53.5 percent of these people still identified with a specific religious institution. Although this DISCUSSION is a sizeable change from the 62 percent reported by Vargas, it means that 11.4 percent of the U.S. The percentage of Americans with no religious population is still positioned to experience “the affiliation has nearly tripled since the early dark side of religion” based on the current results. 1990s (Pew Forum on Religion and Public Life To some extent, this is consistent with previous 2012), yet most of the unaffiliated were raised in research on religion and mental health, but the cur- religious homes (Kosmin and Keysar 2008). Since rent study provides greater nuance to our understand- active involvement in a religious community is ing of the role that religious affiliation plays in the generally associated with better mental health out- relationship between religion and mental health. comes (Koenig 2012), it is important to consider Few studies on religion and mental health actu- the role that religious affiliation plays in the health ally consider the role of religious affiliation (see and well-being of this growing population. My Fenelon and Danielsen 2016 for an exception). investigation was motivated by two suspicions. Instead, behaviors like religious service atten- First I suspected that people who considered drop- dance or / are generally the key ping out of religion (stayers) and people who left focal points (Koenig 2012). Like most of these (leavers) would experience more mental health studies, there is a negative relationship between problems than people who did not consider drop- religious service attendance and my measure of ping out of religion (stable affiliates and stable depressive symptomatology. However, my results Nones). Second, I suspected that people who also show there is a contingent relationship remained affiliated (stayers) would experience between religious affiliation and mental health greater distress than people who did not (leavers) net of religious behaviors like attending religious after several years. Indeed, I did find that religious services. It is important for future research to con- affiliation affects depressive symptomatology, but sider the distinctive ways that religious affiliation not exactly in the expected way. and other religious behaviors affect mental health. In sum, the results described above suggest that In addition to highlighting the contingent role stayers experience more mental health problems that religious affiliation plays in shaping the men- than stable affiliates, leavers, and stables Nones tal health of the affiliated and the unaffiliated, this and that they experience a greater increase in men- research calls into question the importance of reli- tal health problems over time. This is consistent gious affiliation and disaffiliation for other life with Burke and Stets’s (2009) identity theory outcomes. Since there is clear variability in the and with Ebaugh’s (1988) role exit theory. More mental health of stayers and leavers, stable affili- specifically, disturbances like considering drop- ates, and stable Nones based on my results, it is ping out of religion cause negative emotions like also likely that there is variability between these depression (Burke 1991, 1996). However, aban- groups in other areas. Recent scholarship demon- doning an identity can be a useful way to eliminate strates religion’s important role in marriage mar- disturbances (Cast and Burke 2002). Therefore, it kets (McClendon 2016), welfare attitudes (Van- makes sense that stayers experience more depres- Heuvelen 2014), cross-racial interactions (Park sive symptoms and a greater increase in depressive and Bowman 2015), and the work-family interface symptoms over time than the other groups, while (Reynolds and May 2014), to note just a few leavers are not significantly different from stable examples. Future research should consider the affiliates or stable Nones. There are several impor- possibility that religious affiliation plays an tant implications, limitations, and directions for important role in shaping people’s experiences in future research to consider in light of these results. these and other domains. According to Vargas (2012), less than half of This study is also an improvement over cross- the women and men who seriously consider sectional studies of religion and mental health. A 226 Society and Mental Health 8(3) key issue with cross-sectional studies of religion consider the role of religious tradition. Belonging and mental health is causality. Do religious atti- to a religious institution is more important for the tudes/behaviors cause changes in mental health, members of particular religious traditions (Bender or do changes in mental health shape religious atti- et al. 2013). Unfortunately, less than 10 percent of tudes/behaviors? Cross-sectional studies of reli- PALS respondents belong to non-Christian . gion and health generally assume the former (see Small sample sizes across different religious Ellison and Taylor 1996 for an exception), but groups also make it difficult to raise meaningful a number of longitudinal studies show that some comparisons between groups. It is possible that people turn to (or away from) religion during times stayers experience better mental health outcomes of trial in their lives (Ferraro and Kelley-Moore when they belong to religious institutions that 2001; Oates 2013; Pargament et al. 2004). One place a strong emphasis on community and formal strength of the current study is my test of the rela- . It is also possible that leavers experience tionship between religious affiliation and changes better health outcomes when they leave religious in depressive symptomatology over time. traditions that place less emphasis on the congre- Last, my findings raise important questions gational aspects of religious life. Future research about religious identities and role exit. Identities should examine the roles that context and religious are an important source of meaning and purpose tradition play in the relationship between religious (Thoits 1983, 1986), but some identities are affiliation and mental health. more salient than others (Stryker 2002). Thus, it Second, my measure of depressive symptom- is easier to exit some identities (e.g., voluntary atology is based on just three items: depression, identities such as friend, religious/spiritual, volun- worthlessness, and hopelessness. More traditional teer/social) than others (e.g., obligatory identities measures of depressive symptomatology like the such as worker, partner/spouse, or parent) (Thoits Center for Epidemiological Studies Depression 2003). Although Thoits (2003) considers religious Scale (Radloff 1977) or the Hospital Anxiety identities easier to exit, Park and Edmonson and Depression Scale (Zigmond and Snaith (2012) point out that religion is a core identity 1983) also include behavioral (e.g., “I talked less for most people. If stayers experience more than usual.”), physiological (e.g., “My sleep was depressive symptoms than leavers do because restless.”), and attitudinal (e.g., “I have lost inter- they are unable to give up the identity, religious est in my appearance.”) measures of depression. identities are likely more than voluntary identities These items are not available in PALS, and it is for these people. Understanding this difference is possible that the experiences of stayers, leavers, an important avenue for future research on identi- stable affiliates, and stable Nones described here ties and role exit. Vargas (2012) provides some are limited to the emotional components of insight into the process of staying versus leaving, depression that are captured by my dependent var- but PALS does not specifically ask respondents iable. Nevertheless, my findings provide an impor- to explain why they considered dropping out of tant starting point for thinking about the contin- religion. Likewise, PALS does not ask respond- gent relationship between religious affiliation ents who considered dropping out of religion and mental health. Future research should consider why they decided to stay or why they decided to the relationship between religious affiliation and leave. Qualitative research can provide important the other dimensions of depressive symptomatol- insights into the reasons behind the commitment ogy and other measures of mental health. of religious participants and ex-participants (e.g., Last, there is no way to know if repeated Packard and Hope 2015). It is possible that poor changes in religious affiliation affect the observed mental health motivates people to stick with their changes in mental health in different ways. religious institution. This, in turn, might drive the PALS’s multistage panel design makes it possible observed increase in mental health problems for to model changes in mental health during a six- those who remain affiliated as they expected their year period, but we are constantly adjusting our feelings of depression, hopelessness, or worthless- behaviors in order to achieve semantic congruence ness to wane. Future research should consider this between our perceptions of meaningful feedback complex relationship between religion, identity, from others and the meanings we attach to an iden- and mental health. tity (Burke and Stets 2009). Thus, leavers might A couple of additional limitations are also rejoin organized religion, and stayers might decide worth noting here. First, my analyses do not to leave. It is possible, for example, that an May 227 individual considered dropping out of religion in test the predictions of affect control theory (see Rob- 2006, dropped out of religion in 2007, and inson 2007). returned to religion in 2010. This individual is 2. The Americans’ Changing Lives study is another considered a stayer in the current analysis if he nationally representative longitudinal study with questions about religious behavior and health. Unlike or she is still affiliated in 2012, but his or her expe- the Portraits of American Life Study (PALS), how- rience is likely different from that of someone who ever, Americans’ Changing Lives does not ask persistently thinks about dropping out of religion respondents whether they have seriously considered but maintains his or her affiliation with the reli- dropping out of religion. gious group. Reducing the number of years 3. Religious affiliation at wave 1 is based on respond- between the measures of religious affiliation and ents’ identification (or lack of) with a specific reli- mental health can provide a better test of the con- gious group in 2006 and their response to the follow- trol system described above. ing question: “In the past three years, have you seriously considered dropping out of religion altogether?” This measure assumes that leavers CONCLUSION were affiliated with an organized religious group In this article, I examined the role that religious between 2003 and 2006; it also assumes that stable Nones were not. Although not ideal, it is the same affiliation plays in the relationship between reli- measure used by Vargas (2012) before PALS wave gion and depressive symptomatology. My findings 2 was released. indicate that individuals who seriously consider 4. In the current sample, only 62.5 percent of leavers dropping out of religion altogether but maintain and 62.2 percent of stable Nones never attend reli- their affiliation with a religious institution experi- gious services, whereas 7.1 percent of leavers and ence more depressive symptoms than people who 13.4 percent of stable Nones attend religious services do not seriously consider dropping out of religion several times a year or more. Stayers are also quite altogether, people who were never affiliated, and active in their religious groups; 25.0 percent report people who actually leave. Likewise, they experi- attending religious services two or three times per month or more, and only 28.0 percent report never ence a greater increase in depressive symptoms attending religious services. than the members of these other groups do over 5. Ordinary least squares regression models predicting time. Moving forward, it is important to consider the change in depressive symptomatology between how the different causes of variability in religious waves showed little variation from the lagged depen- affiliation are associated with health and well- dent variable model (see Johnson 2005). being and how the relationship between religious 6. Despite the large number of PALS respondents who affiliation and mental health may vary across dif- experienced no depressive symptoms (i.e., 0 on the ferent religious traditions, different settings, and dependent variable), there is little theoretical justifi- when there are different motivations to stay or cation for a zero-inflated negative binomial model leave. Research on these topics will further our with these data. More specifically, zero-inflated neg- ative binomial models require theoretical explana- understanding of one of the fastest-growing reli- tions for cases that are always zero on the dependent gious groups in the United States during the past variable. See Allison (2012) for more on the limita- 25 years and the mental health of these Nones tions of zero-inflated models for count-dependent and almost Nones. variables. 7. In separate analyses (not shown), I included interac- ACKNOWLEDGMENTS tion terms for religious service attendance and reli- gious affiliation. Interestingly, stayers experience The author thanks Ashley Barr, Dennis Condron, Katie a greater increase in depressive symptoms when James, and participants in the Georgia Workshop on Cul- they attend religious services more often, but I am ture, Power, and History for their helpful comments on hesitant to draw conclusions about this finding due earlier drafts of this manuscript. to the small number of cases in each cell of the inter- action term (i.e., stayers who attend more than once NOTES a week, stayers who attend once week, etc.). 1. These hypotheses largely reflect the predictions of other control theories in sociology (e.g., Heise REFERENCES 2007), but the current data do not include other rele- vant measures such as evaluation, potency, and activ- Acevedo, Gabriel A., Christopher G. Ellison, and Xiaohe ity ratings for religious identities in order to formally H. Xu. 2014. “Is It Really Religion? Comparing the 228 Society and Mental Health 8(3)

Main and Stress-buffering Effects of Religious and Ellison, Christopher G., Matt Bradshaw, Nilay Kuyel, Secular Civic Engagement on Psychological Dis- and Jack P. Marcum. 2012. “Attachment to God, tress.” Society and Mental Health 4(2):111–28. Stressful Life Events, and Changes in Psychological Allison, Paul D. 2009. “Missing Data.” Pp. 72–89 in The Distress.” Review of Religious Research 53(4): Sage Handbook of Quantitative Methods in Psychol- 493–511. ogy, edited by R. E. Missap and A. Maydeu- Ellison, Christopher G., Qijuan Fang, Kevin J. Flannelly, Olivares. Thousand Oaks, CA: Sage. and Rebecca A. Steckler. 2013. “Spiritual Struggles Allison, Paul D. 2012. “Do We Really Need Zero-in- and Mental Health: Exploring the Moderating flated Models?” Statistical Horizons. Retrieved Effects of Religious Identity.” International Journal March 20, 2017 (http://statisticalhorizons.com/ for the 23(3):214–29. zero-inflated-models). Ellison, Christopher G., David A. Gay, and Thomas A. Baker, Joseph O., and Andrew L. Whitehead. 2016. Glass. 1989. “Does Religious Commitment Contrib- “Gendering (Non)Religion: Politics, Education, and ute to Individual Life Satisfaction?” Social Forces Gender Gaps in Secularity in the United States.” 68(1):100–23. Social Forces 94(4):1623–45. Ellison, Christopher G., and Jinwoo Lee. 2010. Barkan, Steven E., and Susan F. Greenwood. 2003. “Spiritual Struggles and Psychological Distress: Is “Religious Attendance and Subjective Well-being There a Dark Side of Religion?” Social Indicators among Older Americans: Evidence from the General Research 98(3):501–17. Social Survey.” Review of Religious Research 45(2): Ellison, Christopher G., and Robert Joseph Taylor. 116–29. 1996. “Turning to Prayer: Social and Situational Bender, Courtney, Wendy Cadge, Peggy Levitt, and Antecedents of Religious Coping among African David Smilde. 2013. “Religion on the Edge: Americans.” Review of Religious Research 38(2): De-centering and Re-centering.” Pp. 1–20 in Reli- 111–31. gion on the Edge: De-centering and Re-centering Ellison, Christopher G., Wei Zhang, Neal Krause, and the , edited by C. Bender, W. John P. Marcum. 2009. “Does Negative Interaction Cadge, P. Levitt, and D. Smilde. New York: Oxford in the Church Increase Psychological Distress? Lon- University Press. gitudinal Findings from the Presbyterian Panel Sur- Burke, Peter J. 1991. “Identity Processes and Social vey.” Sociology of Religion 70(4):409–31. Stress.” American Sociological Review 56(6): Emerson, Michael O., and David Sikkink. 2012. 836–49. “Portraits of American Life Survey, 2nd Wave.” Burke, Peter J. 1996. “Social Identities and Psychosocial Emerson, Michael O., David Sikkink, and Adele D. Stress.” Pp. 141–74 in Psychosocial Stress: Perspec- James. 2010. “The Panel Study on American Reli- tives on Structure, Theory, Life Course, and Meth- gion and Ethnicity: Background, Methods, and ods, edited by H. B. Kaplan. San Diego, CA: Aca- Selected Results.” Journal for the Scientific Study demic Press. of Religion 49(1):162–71. Burke, Peter J., and Jan E. Stets. 2009. Identity Theory. Fenelon, Andrew, and Sabrina Danielsen. 2016. New York: Oxford University Press. “Leaving My Religion: Understanding the Relation- Cast, Alicia D., and Peter J. Burke. 2002. “A Theory of ship between Religious Disaffiliation, Health, and Self-esteem.” Social Forces 80(3):1041–68. Well-being.” Social Science Research 57:49–62. Childs, Ellen. 2010. “Religious Attendance and Happi- Ferraro, Kenneth F., and Jessica A. Kelley-Moore. 2001. ness: Examining Gaps in the Current Literature: A “Religious Seeking among Affiliates and Non- Research Note.” Journal for the Scientific Study of affiliates: Do Mental and Physical Health Problems Religion 49(3):550–60. Spur Religious Coping?” Review of Religious Ebaugh, Helen Rose Fuchs. 1988. Becoming an Ex: The Research 42(3):229–51. doi: 10.2307/3512568. Process of Role Exit. Chicago: University of Chicago Finkel, Steven E. 1995. Causal Analysis with Panel Press. Data. Thousand Oaks, CA: Sage. Eliassen, A. Henry, John Taylor, and Donald A. Lloyd. Galek, Kathleen, Kevin J. Flannelly, Christopher G. Elli- 2005. “Subjective Religiosity and Depression in the son, Nava R. Silton, and Katherine R. B. Jankowski. Transition to Adulthood.” Journal for the Scientific 2015. “Religion, Meaning and Purpose, and Mental Study of Religion 44(2):187–99. Health.” Psychology of Religion and Ellison, Christopher G. 1991. “Religious Involvement 7(1):1–12. and Subjective Well-being.” Journal of Health and Hayward, R. David, Neal Krause, Gail Ironson, Peter C. Social Behavior 32(1):80–99. Hill, and Robert Emmons. 2016. “Health and Well- Ellison, Christopher G. 1995. “Race, Religious Involve- being among the Non-religious: Atheists, Agnostics, ment and Depressive Symptomatology in a South- and No Preference Compared with Religious Group eastern U.S. Community.” Social Science & Medi- Members.” Journal of Religion & Health 55(3): cine 40(11):1561–72. 1024–37. May 229

Heise, David R. 2007. Expressive Order: Confirming Consolation among Black and White Americans.” Sentiments in Social Actions. New York: Springer. Review of Religious Research 55(4):597–613. Johnson, David. 2005. “Two-wave Panel Analysis: Com- Packard, Josh, and Ashleigh Hope. 2015. Church Refu- paring Statistical Methods for Studying the Effects of gees: Sociologists Reveal Why People Are DONE Transitions.” Journal of Marriage and Family 67(4): with Church but Not Their . Loveland, CO: 1061–75. Group. Jylha¨, Marja. 2009. “What Is Self-rated Health and Why Pargament, Kenneth I., Harold G. Koenig, Nalini Tara- Does It Predict Mortality? Towards a Unified Con- keshwar, and June Hahn. 2004. “Religious Coping ceptual Model.” Social Science & Medicine 69(3): Methods as Predictors of Psychological, Physical 307–16. and Spiritual Outcomes among Medically Ill Elderly Koenig, Harold G. 2012. “Religion, Spirituality, and Patients: A Two-year Longitudinal Study.” Journal Health: The Research and Clinical Implications.” of Health Psychology 9(6):713–30. doi: 10.1177/ ISRN Psychiatry 2012:1–33. 1359105304045366. Kosmin, Barry, and Ariela Keysar. 2008. American Park, Crystal L., and Donald Edmondson. 2012. Nones: The Profile of the No Religion Population. “Religion as a Source of Meaning.” Pp. 145–62 in Hartford, CT: Institute for the Study of Secularity Meaning, Mortality, and Choice: The Social Psy- in Society. chology of Existential Concerns, edited by P. R. Krause, Neal. 1994. “Stressors in Salient Social Roles Shaver and M. Mikulincer. Washington, DC: Amer- and Well-being in Later Life.” Journal of Gerontol- ican Psychological Association. ogy 49(3):P137–48. Park, Julie J., and Nicholas A. Bowman. 2015. “Religion Krause, Neal. 1998. “Stressors in Highly Valued Roles, as Bridging or Bonding Social Capital: Race, Reli- Religious Coping, and Mortality.” Psychology and gion, and Cross-racial Interaction for College Stu- Aging 13(2):242–55. dents.” Sociology of Education 88(1):20–37. Krause, Neal. 2006. “Religious Doubt and Psychological Pew Forum on Religion and Public Life. 2010. “Religion Well-being: A Longitudinal Investigation.” Review among the Millennials.” Retrieved December 19, of Religious Research 47(3):287–302. 2016 (http://www.pewforum.org/files/2010/02/mill Krause, Neal, and Christopher G. Ellison. 2009. “The ennials-report.pdf). Doubting Process: A Longitudinal Study of the Pre- Pew Forum on Religion and Public Life. 2012. “‘Nones’ cipitants and Consequences of Religious Doubt in on the Rise: One-in-five Adults Have No Religious Older Adults.” Journal for the Scientific Study of Affiliation.” Retrieved December 19, 2016 (http:// Religion 48(2):293–312. www.pewforum.org/uploadedFiles/Topics/Religious Krause, Neal, and Keith M. Wulff. 2004. “Religious _Affiliation/Unaffiliated/NonesOnTheRise-full.pdf). Doubt and Health: Exploring the Potential Dark Side Pew Forum on Religion and Public Life. 2015. of Religion.” Sociology of Religion 65(1):35–56. “America’s Changing Religious Landscape.” Krause, Neal M., and Gina M. Jay. 1994. “What Do Retrieved December 19, 2016 (http://www.pewfo Global Self-rated Health Items Measure?” Medical rum.org/2015/05/12/americas-changing-religious-lan Care 32(9):930–42. dscape/). Large, Michael D., and Kristen Marcussen. 2000. Radloff, Lenore Sawyer. 1977. “The CES-D Scale: A “Extending Identity Theory to Predict Differential Self-report Depression Scale for Research in the Forms and Degrees of Psychological Distress.” General Population.” Applied Psychological Mea- Social Psychology Quarterly 63(1):49–59. surement 1(3):385–401. Lim, Chaeyoon, and Robert D. Putnam. 2010. “Religion, Reynolds, Jeremy, and Matthew May. 2014. “Religion, Social Networks, and Life Satisfaction.” American Motherhood, and the Spirit of Capitalism.” Social Sociological Review 75(6):914–33. Currents 1(2):173–88. Long, J. Scott. 1997. Regression Models for Categorical Robinson, Dawn T. 2007. “Control Theories in Sociol- and Limited Dependent Variables. Thousand Oaks, ogy.” Annual Review of Sociology 33(1):157–74. CA: Sage. Ross, Catherine E. 1990. “Religion and Psychological Manning, Christel J. 2013. “Unaffiliated Parents and the Distress.” Journal for the Scientific Study of Religion Religious Training of Their Children.” Sociology of 29(2):236–45. Religion 74(2):149–75. Schnittker, Jason. 2001. “When Is Faith Enough? The McClendon, David. 2016. “Religion, Marriage Markets, Effects of Religious Involvement on Depression.” and Assortative Mating in the United States.” Jour- Journal for the Scientific Study of Religion 40(3): nal of Marriage and Family 78(5):1399–421. 393–411. Mochon, Daniel, Michael I. Norton, and Dan Ariely. Schwadel, Philip, and Christina D. Falci. 2012. 2011. “Who Benefits from Religion?” Social Indica- “Interactive Effects of Church Attendance and tors Research 101(1):1–15. Religious Tradition on Depressive Symptoms and Oates, Gary L. 2013. “Enhanced Religiosity Following Positive Affect.” Society and Mental Health 2(1): Illness? Assessing Evidence of Religious 21–34. 230 Society and Mental Health 8(3)

Speed, David, and Ken Fowler. 2017. “Good for All? Thoits, Peggy A. 1986. “Multiple Identities: Examining Hardly! Attending Church Does Not Benefit Reli- Gender and Marital Status Differences in Distress.” giously Unaffiliated.” Journal of Religion and American Sociological Review 51:259–72. Health 56(3):986–1002. Thoits, Peggy A. 2003. “Personal Agency in the Accu- Stryker, Sheldon. 1980. Symbolic Interactionism: A mulation of Multiple Role-identities.” Pp. 179–94 Social Structural Version. Menlo Park, CA: in Advances in Identity Theory and Research, edited Benjamin-Cummings. by P. J. Burke, T. J. Owens, R. Serpe, and P. A. Stryker, Sheldon. 2002. “Traditional Symbolic Interac- Thoits. New York: Kluwer-Plenum. tionism, Role Theory, and Structural Symbolic Inter- VanHeuvelen, Tom. 2014. “The Religious Context of actionism: The Road to Identity Theory.” Pp. 202–21 Welfare Attitudes.” Journal for the Scientific Study in Handbook of Sociological Theory, edited by J. H. of Religion 53(2):268–95. Turner. New York: Springer. Vargas, Nicholas. 2012. “Retrospective Accounts of Taylor, Peter D., and Mark A. McPeek. 2013. “The Rise Religious Disaffiliation in the United States: Stres- of the ‘Nones’: Does Education Explain the Decline sors, Skepticism, and Political Factors.” Sociology in Religious Affiliation?” American Journal of Edu- of Religion 73(2):200–23. cation 119(3):335–40. Zigmond, Anthony S., and R. Philip Snaith. 1983. “The Thoits, Peggy A. 1983. “Multiple Identities and Psycho- Hospital Anxiety and Depression Scale.” Acta Psy- logical Well-being: A Reformulation and Test of the chiatrica Scandinavica 67(6):361–70. Social Isolation Hypothesis.” American Sociological Review 48:174–87.