You are on page 1of 12

DAVID C. ATKINS AND DEBORAH E.

KESSEL Fuller Theological Seminary

Religiousness and Infidelity: Attendance, but not


Faith and Prayer, Predict Marital Fidelity

High religiousness has been consistently linked Edwards & Booth, 1976; Liu, 2000; Treas & Gie-
with a decreased likelihood of past infidelity but sen, 2000). Yet, in all of the previous research on
has been solely defined by religious service infidelity, religiousness has been measured solely
attendance, a limited assessment of a complex by attendance at religious services, a very limited
facet of life. The current study developed nine assessment of a rich and complex facet of life.
religiousness subscales using items from the This raises a host of questions: Are there nuances
1998 General Social Survey to more fully in the association of religiousness and infidelity
explore the association between religiousness that attendance fails to capture? Is the negative
and infidelity. Interestingly, logistic regressions association found in past research specific to
using currently married participants ( N attendance or is attendance merely acting as
1,439) demonstrated that attendance, but not a proxy for other religious variables? If further
faith, nearness to God, prayer, and other reli- dimensions of the religious life were assessed,
gious attributes, was related to infidelity. would attendance show the strongest association
Exploratory analyses also found that individu- with infidelity out of a variety of religious fac-
als with high religious importance but low tors? The present research focused on precisely
attendance were more likely to have had an these questions.
affair and weak evidence that marital happiness Most Americans share the expectation that mar-
moderated the association between religious- ried couples will remain sexually faithful to their
ness and infidelity. spouse (Johnson et al., 2002; Treas & Giesen,
2000), yet there is a disconnect between this com-
mon expectation and the not uncommon occur-
You shall not commit adultery (Exodus 20:14;
rence of infidelity. In a nationally representative
New English Bible). The major monotheistic re-
sample, Wiederman (1997) found that 23% of
ligions (i.e., Christianity, Judaism, Islam) pro-
men and 12% of women reported an affair at some
mote fidelity and explicitly prohibit infidelity in
point in their marriages, and 4.1% of men and
marriage, and in this same vein, the majority of
1.7% of women reported having engaged in infi-
individuals believe infidelity in marriage is
delity within the past 12 months. Likewise, Choi,
wrong (Johnson et al., 2002). Past research has
Catania, and Dolcini (1994) found that 2% of
consistently shown an association such that more
a nationally representative sample of American
religious individuals are less likely to have had af-
couples reported having engaged in infidelity
fairs compared to the less religious (Allen et al., within the past 12 months. Moreover, these figures
2005; Atkins, Baucom, & Jacobson, 2001;
are almost certainly conservative estimates, as
rates of infidelity can be notably higher when
Graduate School of Psychology, Fuller Theological Semi- anonymous survey methods are used versus face-
nary, 180 N. Oakland Ave., Pasadena, CA 91101 (datkins@ to-face interviews (Whisman & Snyder, 2007).
fuller.edu). When infidelity occurs, it is often associated
Key Words: general social surveys, infidelity, religiousness. with a number of deleterious outcomes for
Journal of Marriage and Family 70 (May 2008): 407418 407
408 Journal of Marriage and Family

marriages and individuals. Research shows that approval and possible loss of relationship and
infidelity is positively associated with divorce community. Thus, individual (e.g., theological
(Atkins et al., 2001), including one prospective teachings and cognitive consistency with viewing
study showing that earlier infidelity raises the oneself as a religious person), relational (e.g., mar-
likelihood of later divorce (Previti & Amato, ital sanctification, shared activities and beliefs
2004). Moreover, infidelity is associated with between spouses), and contextual processes (e.g.,
poor marital quality (Glass & Wright, 1977; Pre- shared community values and relationship) may
viti & Amato; Spanier & Margolis, 1983), which affect the likelihood of infidelity and may not be
itself is a predictor of divorce. Research also sup- wholly captured by attendance. Finally, the find-
ports links between infidelity and psychopathol- ings from at least one study suggest that religious-
ogy such as depression and symptoms of ness may interact with other factors in its
posttraumatic stress disorder (Cano & OLeary, association with infidelity. Atkins et al. (2001)
2000; Gordon, Baucom, & Snyder, 2004). Infi- found that the association of religiousness with
delity not only has implications for individual infidelity was moderated by marital happiness.
psychological health and strength of relation- Although religiousness and infidelity may
ships, it also has increasingly been implicated interact from individual beliefs up to social struc-
as a transmission route for sexually transmitted tures, the assessment of religiousness has focused
infections including HIV/AIDS (Finer, Darroch, almost entirely on attendance, typically employ-
& Singh, 1999). Although these negative con- ing single-item measures (Mahoney et al.,
comitants of infidelity point to the importance 2001). The present study extended earlier
of understanding factors associated with research by taking a multifaceted view of the
decreased likelihood of infidelity, some have sug- association of religiousness with infidelity. As
gested positive impacts of affairs. In particular, an initial step, a multidimensional scale of reli-
affairs can provide supportive relationships in, giousness was created, including dimensions of
and a way out of, a bad marriage (Vaughn, 1990). religiousness such as perceived closeness to
As noted above, past research has identified God, religious service attendance, forgiveness,
a consistent negative association between reli- prayer, faith, and negative perceptions of reli-
giousness and infidelity. Allen et al. (2005) pro- gion. These dimensions of religiousness were
posed a framework for studying infidelity, then related to infidelity controlling for demo-
wherein one dimension is the level of analysis, graphic factors based on past research (Atkins
consisting of individual, relational, and contex- et al., 2001; Treas & Giesen, 2000; Wiederman,
tual. Religiousness may interact with infidelity 1997), including age, gender, years of education,
within each of these spheres. At the individual race, respondents income, and previous divorce.
level, theological teachings underscore the impor- Although there was no direct research upon
tance of marriage and fidelity and the prohibition which to base hypotheses, the exploratory
of infidelity. In addition to specific messages about hypothesis was that additional dimensions of reli-
infidelity, there are other indirect ways that reli- giousness beyond religious service attendance
gion may help prevent infidelity. For example, would be associated with infidelity. Moreover,
religious teachings emphasize forgiveness, care it was hypothesized that the association of reli-
toward others, and admonishments about anger, giousness with infidelity would be moderated
which will foster individual attitudes that in turn by marital happiness, replicating the earlier find-
could strengthen marital relationships (Mahoney, ings of Atkins et al. (2001).
Pargament, Tarakeshwar, & Swank, 2001). At
the relational level, Mahoney and Tarakeshwar
METHOD
(2005) have studied how religious couples sanc-
tify their marriages through viewing their relation-
Participants
ships as having spiritual significance and pointing
to God as the source of the relationship. Within The participants for this study were drawn from
couples that share similar religious beliefs, spiri- the 1998 General Social Survey (GSS) conducted
tual practices can be shared practices that sustain by the National Opinion Research Center at the
and improve the marital relationship. Moreover, University of Chicago. The initial GSS took place
religious communities provide contexts that in 1972 and since then more than 42,000 respond-
explicitly or implicitly underscore the costs of infi- ents have participated, answering over 3,300
delity to the marriage but also reinforce social dis- different questions. The surveys are structured,
Religiousness and Infidelity 409

face-to-face interviews and involve a national, Infidelity. Infidelity was measured through a sin-
cross-sectional sample of noninstitutionalized, gle question: Have you ever had sex with some-
English-speaking individuals between the ages one other than your spouse while you were
of 18 and 89 in the continental United States. married? Previous studies with the GSS (e.g.,
At present, the GSS is conducted every other year Atkins et al., 2001) have noted that the wording
with approximately 3,000 participants, and the of the infidelity question is somewhat ambiguous.
1998 survey (N 2,832) included a special bat- In particular, the wording limits infidelity to mar-
tery of questions related to religion and spiritu- riage relationships as opposed to other committed
ality. Of participants surveyed in 1998, 283 relationships (e.g., cohabiting partners or same-
(10%) were widowed, 446 (16%) were sex partners in all likelihood); at the same time,
divorced, 663 (23%) were never married, and 1 it does not identify a specific marital relationship
had missing data for marital status. Of partici- for those participants who have been married
pants who had been married at some point (n more than once. Finally, although most survey re-
2,169, 76.7% of total sample), 82.1% reported spondents would likely interpret sex as indicating
never having had sex with another person while sexual intercourse, this is not a universal interpre-
they were married, whereas 17.9% reported sex- tation (see discussion in Allen et al., 2005).
ual infidelity at some point while married Where possible, we attempt to address these def-
(14.7% women and 22.5% men). initional ambiguities in the analyses and consider
Because one of the hypotheses related to mar- their implications in interpreting the results.
ital happiness, participants who were married or
married but separated at the time of the interview Marital happiness. Marital happiness was mea-
were included in the analysis, yielding a sample sured through a single question on the GSS:
size of 1,439. (Note that participants who have Taking all things together, how would you
been divorced but were remarried at the time of describe your marriage? Would you say that your
the survey were included in all analyses.) Sensi- marriage is very happy, pretty happy, or not too
tivity analyses (described in Data Analyses) happy? Although it would be preferable to have
examined whether religiousness associations a scale of marital happiness, multi-item scales are
generalized to participants who were previously often not feasible in large surveys. At the same
married (i.e., widowed or divorced). time, Goodwin (1992) has shown that a single-
As general indicators of the religiousness of item assessment of marital happiness correlates
the sample, 48.2% of participants indicated that highly with total marital adjustment and differen-
they attend religious services once a month or tiates between adjusted and distressed couples.
more frequently, and 59.5% reported that they
find strength and comfort in religion most days Religiousness. Thirty-seven items from the 1998
or more frequently. Moreover, 82% (n 2,290) GSS religion battery were used to assess a variety
of the sample were affiliated with a branch of of religious domains, including beliefs, attitudes,
the Christian church (i.e., Protestant, Catholic, behaviors, and affective experience. For exam-
Nondenominational), 4% (n 110) were affili- ple, items reflecting personal or corporate reli-
ated with other religious traditions (e.g., Jewish, gious behavior or both included frequency of
Buddhist, Muslim), and 14% (n 396) listed church attendance, religious activities, and
no religious affiliation. In addition, the majority prayer. Negative attitudes toward religion were
of participants reported stable, long-term reli- assessed with items such as I believe religion
gious beliefs: 85.6% endorsing that they are brings conflict and I believe religious people
current believers and always have been; 2.9% are often too intolerant. Items such as I find
endorse no beliefs currently or previously; 6.2% strength and comfort in religion and I feel deep
endorse that they believe now but did not previ- inner peace or harmony assess religious experi-
ously; and 5.3% report no current spiritual be- ence. Specific items are discussed later in the con-
liefs whereas previously they did. text of scale development.

Other variables. Demographic variables that


Measures
have been shown to be important in previous
All of the variables used in this project were taken assessments of infidelity were entered as covari-
from questions that were included in the 1998 ates in the regression model (Atkins et al.,
GSS (Davis & Smith, 2002). 2001; Treas & Giesen, 2000; Wiederman,
410 Journal of Marriage and Family

1997). These included age, gender, years of edu- from the 1998 GSS. Factors that had at least two
cation, race (i.e., White, Black, Other), respond- loadings of .4 and above were retained (Gorsuch,
ents income, and previous divorce. The GSS 1997). Subscales suggested by the factor analysis
assessed income in 23 separate income blocks were created by adding the relevant items. The
from less than $1,000 to $110,000 and higher. resulting subscale was standardized, and internal
Income was transformed to approximately consistency was examined using Cronbachs
$10,000 increments, except that the three highest alpha.
increments were $15,000, $20,000, and open-
ended based on the way that the GSS originally Logistic regression. A series of logistic regres-
coded the variable. There was considerable miss- sion models were fit with infidelity as the out-
ing data for income (n 446, 31.0%), reflecting come (0 no, 1 yes) and covariates and
in part that some individuals were not working religiousness subscales as predictors. An addi-
(i.e., at home, retired, laid off), though 21.3% of tional model examined whether the association
these individuals still reported income. Other of religiousness was moderated by marital hap-
participants simply refused to report income piness. Because of the multiple imputation,
although they specified that they were working regression models were fit to each replicated
full- or part-time. Two models examined (1) data set. Regression coefficients are simply
imputing zero for those with missing data and averaged across replications, but standard errors
who were not working, and (2) using multiple take into account the variability within and
imputation (as described below) for all missing between replicated data sets (Schafer & Olson,
income data. No substantive differences were 1998).
found treating the missing income data in these
two ways. Analyses below report multiply Sensitivity analyses. Several additional analyses
imputed data. were conducted to ascertain whether the findings
from the logistic regressions were sensitive to the
particular subset of the GSS used or the wording
Data Analysis
of the infidelity question. Because of the open-
Missing data. The GSS uses a planned missing ended phrasing of the infidelity question, partici-
data design with three separate ballots. Certain pants who have been divorced may have had an
core items (e.g., age, gender, race) are asked on affair in their previous marriage. Thus, one sensi-
all ballots and have little to no missing data (less tivity analysis examined the moderating effects
than 0.1% for age, gender, race). Rotating ques- of previous divorce. To examine whether the reli-
tions are included on two out of three ballots giousness associations generalized beyond cur-
and asked to two thirds of the sample by design, rently married participants, a second sensitivity
which is becoming commonplace in large sur- analysis included participants who had ever been
veys (Schafer & Graham, 2002). Missing data married (N 2,168), excluding the marital hap-
were imputed using multiple imputation, in piness question. Finally, because the infidelity
which a number of data sets are generated from question does not specify a timeframe, it is pos-
the initial data containing missing data (see sible that the affair may have occurred many
Acock, 2005, for an overview). In each replica- years previously even if in the current marriage,
tion, all of the observed data are repeated, and whereas religiousness items are all assessed at
the missing data are imputed based on a regres- the time of the interview. Thus, the final sensi-
sion model predicting each variable from every tivity analysis examined a model restricted to
other variable plus a random component (i.e., a younger cohort of individuals.
there is variability across the replicated data sets
in the imputed values). Multiple imputation has General issues. Like most national surveys, the
been shown to be one of the most accurate meth- GSS does not use a simple random sample of in-
ods for handling missing data (Acock, 2005), par- dividuals in the United States. Instead, an initial
ticularly in cases of planned missing data like sample of primary sampling units (PSUs) is
much of the missing data in the GSS. selected; in rural areas, these are counties,
whereas in densely populated metropolitan areas,
Factor analysis. An exploratory factor analysis, they are subsets of cities. A second stage of selec-
using Principal Axis Factoring and Promax rota- tion occurs within each PSU. The two-stage
tion, was conducted using 37 religiousness items design is more efficient (i.e., a simple random
Religiousness and Infidelity 411

sample with in-person interviews would be pro- Table 1. Factor Loadings Based on Principal Axis Factor-
hibitively expensive) but affects the statistical ing Using Promax Rotation With Coefficient Alphas
analyses. To account for the clustered, nonran-
Factor 1: Nearness to God (a .93)
dom nature of the data, standard errors in the
1. I find strength and comfort in religion .94
regression analyses are based on robust variance
2. I feel Gods presence .93
estimates that take into account the clustered
3. I feel Gods love directly or through others .91
nature of the data (Harrell, 2001).
4. I desire to be closer to God .81
Exploratory graphs examined the association
5. I feel deep inner peace or harmony .79
of each continuous predictor with the probability
6. I work together with God as partners .77
of infidelity, using a loess smoother to detect non-
7. I try to carry my beliefs into other dealings .61
linear associations, and interactions with key cat-
8. I look to God for strength and support .58
egorical variables (e.g., gender and previous
9. I think about life as part of a larger force .53
divorce) were explored. All analyses were done
10. I believe God is concerned .48
in R version 2.5.0 (R Development Core Team,
with human beings personally
2007) and made extensive use of the Hmisc
11. I consider myself a religious person .48
(Harrell, 2007), Design (Harrell, 2005), and mi-
Factor 2: Attendance (a .82)
tools (Lumley, 2004) packages for multiple impu-
1. How often do you attend religious services? .78
tation and cluster-adjusted logistic regression.
2. How often do you take part in .53
religious activities?
RESULTS Factor 3: Prayer (a .89)
1. How often do you pray privately? .92
Factor Analysis: Religiousness Subscales 2. How often do you pray? .89
Factor 4: Religion as Problem (a .71)
Results of the exploratory factor analysis showed 1. I believe that religion brings more conflict .69
that 29 out of the 37 religiousness items from the 2. I believe religious people are often too intolerant .66
GSS loaded onto nine factors and accounted for Factor 5: Transformation (a .74)
66% of the total item variance. Table 1 lists each 1. Have you ever had a born again experience? .77
of the nine factors with factor loadings and inter- 2. Have you ever had a turning point .76
nal consistency estimates for each resulting sub- in your life for religion?
scale. The subscales cover many dimensions of Factor 6: Doubts (a .75)
religious and spiritual life, including importance 1. Have evil in the world ever caused you doubt? .80
and experience of religion (e.g., Nearness to 2. Has personal suffering caused you doubt? .73
God, Transformation), specific religious behav- Factor 7: Forgiveness (a .61)
iors and attitudes (e.g., Attendance, Prayer, For- 1. Have you ever forgiven .67
giveness), and negative aspects of religion (e.g., yourself for mistakes?
Religion as Problem, Punitive God). The sub- 2. Do you forgive those who hurt you? .56
scales have good internal consistency with two Factor 8: Faith (a .77)
exceptions; Punitive God and Forgiveness sub- 1. Has a death in the family .96
scales both have alphas less than .70, in part strengthened your faith?
because of their skewed distributions. 2. Has the birth of a child .60
strengthened your faith?
Descriptive Correlations: Religiousness Factor 9: Punitive God (a .53)
and Infidelity 1. I often wonder whether God has abandoned me .79
2. I feel God is punishing me for sin .47
Spearman correlations between religiousness sub-
scales and infidelity are shown in Table 2. Punitive
God and Doubts are relatively independent of tion at .12, whereas the next strongest associa-
other factors, whereas moderate to strong correla- tion with infidelity is Faith (r .07) at almost
tions are seen among the other religiousness sub- half the strength of Attendance.
scales. Nearness to God and Attendance are
Logistic Regression: Religiousness
notable in their correlation with one another and
and Infidelity
also the other religiousness subscales. At the same
time, the correlations with infidelity are generally Based on previous research suggesting that
quite low. Attendance shows the strongest correla- age has a notably nonlinear relationship with
412 Journal of Marriage and Family

Table 2. Correlations of Religiousness Subscales and Infidelity

Variables 1 2 3 4 5 6 7 8 9 10

1. Infidelity
2. Nearness to God 0.06
3. Attendance 0.12 0.61
4. Prayer 0.03 0.27 0.19
5. Religion as Problem 0.05 0.38 0.36 0.06
6. Transformation 0.04 0.54 0.40 0.16 0.29
7. Doubts 0.04 0.10 0.10 0.12 0.08 0.00
8. Forgiveness 0.02 0.40 0.28 0.15 0.19 0.28 0.09
9. Faith 0.07 0.42 0.34 0.13 0.18 0.35 0.14 .18
10. Punitive God 0.03 0.01 0.05 0.02 0.04 0.10 0.20 0.12 0.11
Note: Correlations based on original, nonimputed data. Sample sizes for correlations range from 584 to 1,218. Because cor-
relations are used for descriptive purposes, inferential statistics (i.e., p values) are not reported.

infidelity (Atkins et al., 2001; Wiederman, 1997), enhance their interpretability. Specifically, two
age was modeled using a restricted cubic spline new variables were created: one in which Near-
(Harrell, 2001). A spline divides up a predictor ness to God and Attendance were summed (Near
into two or more segments and fits a separate 1 Attend), and a second in which Attendance
regression slope within each segment, and cubic was subtracted from Nearness to God (Near 
splines fit quadratic regressions within segments. Attend). These two new variables assess the con-
Splines can be effective tools for modeling gruence of belief and behavior (Near 1 Attend)
nonlinear relationships that are not well fit by but also the divergence (Near  Attend). Those
polynomial contrasts. In the current instance, participants with high scores on Near  Attend
likelihood-ratio tests and information criteria indicate that they feel very close to God and that
(i.e., AIC and BIC) all preferred the model using religion is important but rarely attend religious
the restricted cubic spline over polynomials for services. These two derivative variables are
age. Exploratory analyses revealed a nonlinear almost totally unrelated (r .05), and thus well
association for years of education and for income, suited for assessing congruence and divergence
which were both modeled using linear and qua- of religious belief and behavior. These simple
dratic terms. Exploratory graphs also supported transformations result in two predictors that are
an interaction between education and gender. directly interpretable, as opposed to the com-
The GSS included only three categories of race mon strategy of entering variable interactions.
in 1998 (i.e., White, Black, Other). Because the Twenty multiply imputed data sets were cre-
Other category was quite small (only 5% of the ated from the complete data with religiousness
sample) and appeared to have a similar relation- subscales. Three logistic regressions were fit to
ship with infidelity as White, it was combined the multiply imputed data: (a) covariates only,
with White. Thus, race was a single dummy- (b) covariates plus nine religiousness subscales,
coded predictor comparing non-Black and Black, and (c) an exploratory model with covariates plus
where non-Black includes participants who indi- two transformed religiousness variables (i.e.,
cated White or other racial categories. Categori- Near 1 Attend and Near  Attend). Table 3 con-
cal predictors were dummy-coded with the tains odds ratios and 95% confidence intervals for
following reference categories: females, Black odds ratios for the three models just described. In
participants, very happy marriages, and no previ- examining the covariates, age has a nonlinear
ous divorce. association that is similar to what has been seen
Religiousness subscales were initially in previous research. Nonlinear associations
included in a model with covariates. A further, (i.e., age, income, education) were interpreted
exploratory model was tested following methods through plotting the predicted regression lines
outlined by Gorsuch and Walker (2006). These (not shown). The likelihood of having had an
authors have recently examined a simple trans- affair reaches a peak in approximately the late
formation of religious attendance and importance 40s, early 50s, and drops off on either side of this.
(Nearness to God in the present analyses) to At the same time, there are a few differences from
Religiousness and Infidelity 413

Table 3. Logistic Regression Results of Infidelity on Religiousness Controlling for Demographic Covariates (N 1,439)

Model 1 Model 2 Model 3


B B B
Variables e 95% CI e 95% CI e 95% CI

Intercept 0.18 0.01 4.29 0.14 0.01 3.08 0.16 0.01 4.06
y y
Age 7.59* 7.17 7.19
Gender (female) 0.02* 0.00 0.92 0.02* 0.00 1.08 0.08* 0.00 1.09
y y y
Education 0.69 0.44 1.08 0.72 0.45 1.14 0.70 0.45 1.11
2
Education 1.02* 0.99 1.03 1.02* 0.99 1.03 1.02* 0.99 1.03
y y
Race (Black) 0.75 0.45 1.22 0.71 0.43 1.18 0.71 0.43 1.17
Income 0.70* 0.55 0.90 0.69* 0.53 0.90 0.69* 0.54 0.89
2
Income 1.03** 1.01 1.06 1.04** 1.01 1.06 1.04** 1.01 1.06
Divorce (no divorce) 2.39** 1.62 3.52 2.29** 1.53 3.43 2.28** 1.54 3.39
Marital happiness (very happy)
Pretty happy 2.18** 1.48 3.20 2.18** 1.47 3.23 2.15** 1.46 3.17
Not too happy 2.97** 1.21 7.26 3.00** 1.19 7.54 2.97** 1.21 7.34
Sex 3 Education 2.20** 1.22 3.96 2.22** 1.19 4.13 2.17** 1.19 3.96
2
Sex 3 Education 0.97** 0.95 0.99 0.97** 0.95 0.99 0.97** 0.95 0.99
Nearness to God 1.19 0.90 1.57
Attendance 0.71** 0.53 0.94
Prayer 0.98 0.79 1.22
Religion as problem 1.04 0.83 1.32
Transformation 1.02 0.80 1.29
Faith 0.91 0.72 1.17
Doubt 1.02 0.83 1.26
Forgiveness 1.08 0.86 1.35
Punitive God 1.04 0.81 1.34
Near 1 Attend 0.84* 0.70 0.99
Near  Attend 1.25* 1.01 1.55
2
v 148.5 190.9 163.1
Df 15 24 17
AIC 1158.4 1155.6 1149.3
Note: 95% CI 95% Confidence Interval for odds ratio. AIC Akaike Information Criterion. Reference categories for cat-
egorical predictors are noted in parentheses. There is no single regression coefficient for age because of the spline fit. The like-
B
lihood ratio chi-square is reported under e .
y
p , .10. *p , .05. **p , .01.

past research (e.g., Atkins et al., 2001; Wieder- with years of schooling until college education
man, 1997): This peak is at a somewhat earlier and notably at graduate levels of education
age, the dropoff in older cohorts is not as pro- (i.e., 171 years of schooling), at which point
nounced, and the model suggests that rates of infi- their likelihood of reporting an affair is greater
delity increase in the youngest cohort (i.e., ages than men. There is a weak suggestion of differ-
18 30). The relationship of years of education ences in rates of infidelity between non-Black
on infidelity is strongly dependent on gender. and Black participants, with fewer non-Black
At lower levels of education (e.g., less than 10 participants reporting infidelity. Income has
years), mens and womens rates of infidelity a strongly nonlinear relationship with infidelity,
are similar; as years of education increase, mens with an overall U shape. Participants at the low-
likelihood of having had an affair also increases, est income levels (i.e., under $10,000) and high-
to a peak level at roughly a college education (i.e., est income levels (i.e., $110,000 and higher)
15 16 years of schooling). At yet higher levels share similar rates of infidelity that drop off in
of schooling, mens rate of infidelity decreases. between the two income extremes. Participants
Womens reports of infidelity do not increase earning between $40,000 and $60,000 per year
414 Journal of Marriage and Family

are the least likely to have reported having an God and behavior show decreased odds of infi-
affair. delity whereas those with divergence show
Finally, relationship variables were strong pre- increased odds of infidelity.
dictors of infidelity. Previous divorce is associ-
ated with a 2.5 times greater likelihood of
Moderating Effects of Marital Happiness
infidelity. In addition, marital happiness has
a strong relationship with infidelity. Those partic- We also examined whether marital happiness
ipants who report that their marriages are pretty would moderate the associations of either At-
happy are 2.2 times more likely to have had an tendance or the two transformed religiousness
affair than those with very happy marriages; par- variables on infidelity by including their interac-
ticipants with not too happy marriages are 3 times tions in the model with covariates. Neither the
2
more likely to report an affair relative to those model with Attendance, v (2, N 1,423)
with very happy marriages. 1.22, p .54, nor2 the model with the trans-
The second model that includes the religious- formed variables, v (2, N 1,422) 0.98, p
2
ness subscales in addition to covariates is inter- .61 and v (2, N 1,422) 1.42, p .49,
esting both for what it shows and what it fails to respectively, revealed significant interactions
show. Attendance proved to be the only signifi- between the religious variables and marital hap-
cant predictor of infidelity. Although Nearness piness. Exploratory analyses had supported that
to God is approaching significance, the odds ratio there may be a moderating effect of marital hap-
is over one, implying a positive relationship piness, specifically for the not too happy group
between participants sense of being near to of participants; one challenge to testing this
God and the likelihood of infidelity, although hypothesis is that the GSS assesses marital hap-
early correlation analyses had suggested exactly piness on a 3-point scale, and few individuals
the opposite. Follow-up analyses without Atten- indicated that their marriages were not too
dance demonstrated that the positive association happy (n 37, 2.8% in original, nonimputed
between Nearness to God and infidelity is an arti- data).
fact of the correlation between the two subscales. The Attendance subscale is almost entirely
When Attendance is removed, Nearness to God determined by a single item on religious service
has a negative but nonsignificant (p .42) rela- attendance (i.e., a 9-point Likert scale), which
tionship with infidelity. Additional facets of was asked of participants in each year the GSS
religiousness do not add significant prediction was administered. Thus, as a final test of the mar-
over and above Attendance. ital happiness moderating hypothesis, we exam-
Because each subscale has been standardized, ined the interaction of marital happiness and
the odds ratio for Attendance (i.e., 0.71) is inter- religious service attendance using data from the
preted as a 29% decrease in the likelihood of hav- 1991 2004 GSS, those surveys that included
ing had an affair for each standard deviation the infidelity question. The exact same proce-
increase in Attendance. Furthermore, those partici- dures that were described previously for the
pants who rarely if ever attend services (i.e., 2 1998 GSS were used in this analysis. Using those
SD) are approximately 4.0 times more likely to participants who were married or married but sep-
have had an affair compared to those participants arated yielded 10,671 total participants, and more
who attend services with great frequency (i.e., importantly, 249 participants who had indicated
12 SD). not too happy marriages. An identical model to
The third, exploratory regression includes co- that shown in Table 3 including covariates was
variates and the two transformed religiousness run, using the single item assessment of religious
variables described earlier (i.e., Near 1 Attend, service attendance as the sole religiousness pre-
Near  Attend). The two transformed predictors dictor and including year of the survey as a set
had opposing and significant associations. The of dummy-coded contrasts, though no significant
odds ratio for Near 1 Attend is 0.84, indicating differences across years were observed. In this
a 16% reduction in the likelihood of a past affair analysis with a greatly expanded sample size,
for each 1 SD increase. Conversely, the odds the interaction of marital happiness and religious
ratio for Near  Attend is 1.25, indicating service
2
attendance just reached significance,
a 25% increase in the likelihood of a past affair v (2, N 10,645) 6.0, p .05.
for each 1 SD increase. Thus, those individuals The interaction is driven by the slope of atten-
with congruence between sense of nearness to dance for the not too happy group, which partly
Religiousness and Infidelity 415

accounts for the borderline significance. The sistent for a broader sample including those par-
attendance slopes of the very happy and pretty ticipants who were married at some point but
happy participants are negative and virtually par- were not at the time of the survey.
allel, with the pretty happy participants 1.9 times Although it is impossible to learn when the
more likely to have reported an affair than the affair occurred, analyzing younger cohorts limits
very happy participants; the slope of the not too the range of time between the GSS assessment
happy participants is steeper and more elevated and the affair to some extent. Thus, a final sensi-
than the pretty happy participants. At frequent tivity analysis reanalyzed the 1998 GSS data set
levels of service attendance (i.e., once a week or including all participants who were currently or
more), the model predicts pretty happy partici- previously married but limited to those partici-
pants to be 6.3% and not to happy to be 11.9% pants who were 40 years of age or younger (n
more likely to have had affairs relative to very 691). Even in the younger cohorts, there was
happy participants (who have a 5.2% overall pre- evidence for both Near 1 Attend (OR 0.74,
dicted likelihood). For those participants who 95% CI 0.57, 0.95, p , .01) and Near 
never went to religious services, the model pre- Attend (OR 1.29, 95% CI 1.00, 1.69, p
dicts pretty happy participants to be 5.4% and .05). Thus, results are similar with a shorter
not too happy to be 22.7% more likely to have timeframe, limiting the temporal distance
had affairs relative to very happy participants between when the affair may have occurred and
(who have an 11.6% overall predicted likeli- the assessment of religiousness.
hood). Thus, the difference in percentages of par-
ticipants reporting affairs is almost constant
across levels of service attendance for very happy DISCUSSION
and pretty happy couples but increases notably The present research underscores the important
for not too happy couples. association between personal religiousness and
infidelity. At the same time, the results are sur-
Sensitivity Analyses prising for what they fail to show. In the first mul-
tifaceted examination of religiousness with
The first sensitivity analysis examined whether infidelity, only attendance at religious services
associations were different for currently married proved to be significantly related to infidelity.
participants who have been divorced, as their af- Thus, among religiousness factors attendance ap-
fairs may have been in a previous marriage (and pears to have a unique and possibly protective
may have contributed to the divorce). Model 3 relationship with infidelity. Moreover, endorsing
in Table 3 was rerun including interactions religion as being very important without the
between divorce and all other predictors. None behavioral component of attending services
of the divorce interaction terms were significant, was positively associated with infidelity. In ad-
nor
2
was the omnibus, likelihood ratio chi-square, dition, there appears to be weak support for
v (16, N 1,405) 19.5, p .24. Thus, it ap- a moderating effect of marital happiness on the
pears that the observed results are similar for relationship between religious service attendance
participants regardless of previous divorce. and infidelity.
A second sensitivity analysis included all par-
ticipants in the 1998 GSS who had ever been mar-
Whats So Special About Attendance?
ried (N 2,168), adding participants who were
divorced or widowed at the time of the inter- Discussions of religion (and particularly Chris-
view. The marital happiness predictor was tianity) often focus on prayer, faith, forgiveness,
excluded for this analysis, and a marital status and the transformative power of religion. Yet
variable (i.e., married, separated, divorced, wid- none of these components were predictive of infi-
owed) was included with married as the refer- delity, whereas religious service attendance was.
ence category. In this analysis, the nonlinear In examining the subscales of the present investi-
relationship of age was no longer significant, gation, the attendance subscale is unique on at
and Near  Attend also failed to reach signifi- least two dimensions. First, the vast majority of
cance (OR 1.11, 95% CI 0.93, 1.32, p items used to assess religiousness on the GSS
.12). Near 1 Attend had a similar association as focus on beliefs/attitudes (e.g., I believe that
seen earlier (OR 0.83, 95% CI 0.71, 0.96, religion brings more conflict; I believe God
p , .01). Thus, findings seem to be largely con- is concerned with human beings personally) or
416 Journal of Marriage and Family

experiences (e.g., I feel Gods love directly or Thus, attendance can be seen as a costly commit-
through others; Have you ever had a born- ment both for practical reasons (i.e., time and
again experience?). Attendance is a purely energy spent) but also for how the social commu-
behavioral assessment, and the only other sub- nity will interpret this public behavior. In a similar
scale with behavioral items is the Prayer subscale. vein, social psychological theories such as self-
The second dimension that differentiates atten- perception theory (Bem, 1972) suggest that our
dance is the corporate or community aspect of attitudes are based in part upon viewing and inter-
attendance. By definition, attending religious preting our actions. Thus, the behavioral activity
services implies that the individual is participat- of attendance can serve to reinforce attitudinal
ing with other people, which is likely to include variables. In that respect, it could be interesting
the spouse. None of the other subscales imply in- to compare weekly attendance at religious serv-
teractions with other people in the context of reli- ices with other regular activities that do not have
giousness, including Prayer. moral components regarding infidelity (e.g.,
Previous research has highlighted religious social clubs or volunteering), as the social and
homogamy between spouses (i.e., similar reli- self attributions would be different. Moreover,
gious values within a couple) as both common future research will need to clarify whether the
and associated with positive, relationship out- present speculations have some veracity.
comes (e.g., greater satisfaction and reduced like-
lihood of divorce; reviewed in Mahoney et al.,
Importance Without Attendance
2001). Thus, attendance in the present study
likely implies attendance with ones spouse for The majority of individuals in the GSS who at-
many participants. This communal aspect of tended religious services regularly also endorsed
attendance could help to prevent infidelity a strong sense of feeling near to God (r .66 for
through several mechanisms. First, it is a shared Nearness to God and Attendance). Yet there is
activity between spouses, and past research has an important subgroup who endorsed a strong
shown that disengagement in couples is associ- sense of feeling close to God and religions
ated with affairs (Atkins, Yi, Baucom, & Chris- importance in their life but did not attend serv-
tensen, 2005). Moreover, spouses that share ices, which was yoked to markedly higher re-
similar religious convictions are likely to share ports of infidelity. Why might this be?
values specifically about the relationship, includ- One possibility is that these are individuals
ing commitment and fidelity but also broader whose affairs were made known to their reli-
convictions of forgiveness and care that may gious communities and were subsequently
serve to strengthen the marital relationship. In ostracized or chose to leave. Thus, they may
addition, an individual who is regularly attending have been attending religious services at some
services will have a network of relationships point, but as a result of the affair, they no longer
within the church, synagogue, or mosque. These do. If this were the mechanism behind this find-
relationships may provide social support to the ing, we would expect there to be a significant
spouses as well as social comparison. Having interaction with divorce, which was not the case.
an affair risks that the infidelity could be revealed Another possibility builds upon the idea of reli-
within the religious community, which could gious homogamy. It is possible that these indi-
have consequences from shame and embarrass- viduals have religious convictions that are not
ment to being ostracized or removed from the shared by their spouse. When religious values
community. Finally, attending religious services are not shared between partners, it may practi-
almost certainly means that an individual is hear- cally interfere with attending services (e.g., the
ing religious teaching on marital fidelity and the spouse may clearly prefer to engage in other
general importance of marriage. activities) but also could undermine the positive
The current findings also may be interpreted in influences of shared religious belief noted ear-
light of costly signaling theory (as applied to reli- lier. If a partners differing views on religion
gious groups, see Sosis, 2000; Sosis & Bressler, does interfere with attendance, then the mecha-
2003). Commitment to social groups varies along nisms mentioned earlier for the protective role
a spectrum, with certain commitments being of attendance (i.e., costly signaling theory and
largely private and entailing little social risk to self-perception theory) would also be undermined.
the individual (e.g., private prayer) whereas other These speculations point to the critical need for
commitments are more public (e.g., attendance). further research focused on religious variables
Religiousness and Infidelity 417

shared between partners and how these relate to a spotlight on the critical need for future research
infidelity. on shared behaviors and beliefs between partners
to examine how the relational nature of religious-
ness may affect infidelity.
Marital Happiness
The moderating effects of marital happiness on
religiousness and infidelity remains largely an NOTE
open question. There was not support for moder- The article is based in part on Deborah Kessels masters
ation in the 1998 sample and only weak support in research. Some of the results were presented at the Association
for Behavioral and Cognitive Therapies, 2005 conference in
the combined 1991 2004 samples. The GSS Washington, DC, and the International Association for Rela-
only used a 3-point measure of marital happiness, tionship Research, 2006 conference in Rethymnon, Crete. We
and the moderating effect that was observed was thank Jim Furrow, Jamie Bedics, Rebeca Marn Cordero, and
restricted to the not too happy group. A more Rich Gorsuch for their insightful comments and feedback that
greatly improved the manuscript.
refined assessment of marital happiness is needed
to provide a precise test of this hypothesis.
The moderating effect that was observed in the REFERENCES
combined sample portrays an intriguing intersec-
tion of marital and religious life. A strong mar- Acock, A. C. (2005). Working with missing values.
riage should provide a robust deterrent to Journal of Marriage and Family, 67, 1012 1028.
having an affair (i.e., Why should I be interested Allen, E. B., Atkins, D. C., Baucom, D. H., Snyder,
in someone else and risk the good relationship D. K., Gordon, K. C., & Glass, S. (2005). Intraper-
that I have?). A frankly unsatisfying marriage sonal, interpersonal, and contextual factors in
may turn the odds in the other direction and engaging in and responding to infidelity. Clinical
may promote an affair given the opportunity. In Psychology: Science and Practice, 12, 101 130.
the face of a bad marriage, the clarity and strength Atkins, D. C., Baucom, D. H., & Jacobson, N. S.
of religious convictions and community aspects (2001). Understanding infidelity: Correlates in
of religious life may be all the more important. a national random sample. Journal of Family Psy-
Again, future research with a refined measure of chology, 15, 735 749.
marital happiness will need to clarify this poten- Atkins, D. C., Yi, J., Baucom, D. H., & Christensen,
tially important relationship. A. (2005). Infidelity in couples seeking marital
therapy. Journal of Family Psychology, 19, 470
473.
Limitations and Conclusions
Bem, D. J. (1972). Self-perception theory. In L.
Although there are many strengths to the present Berkowitz (Ed.), Advances in experimental social
research (e.g., large nationally representative psychology (Vol. 6, pp. 1 62). New York: Aca-
sample, multidimensional assessment of reli- demic Press.
giousness), there are also some limitations. In Cano, A., & OLeary, K. D. (2000). Infidelity and
addition to the limitations mentioned earlier and separations precipitate major depressive episodes
addressed by the sensitivity analyses, the GSS and symptoms of nonspecific depression and anxi-
religiousness questions tend to be individually ety. Journal of Consulting and Clinical Psychol-
focused and broadly applicable, which leads ogy, 68, 774 781.
them to be somewhat generic. Thus, there are Choi, K. H., Catania, J. A., & Dolcini, M. M. (1994).
no questions about how individuals view their re- Extramarital sex and HIV risk behavior among US
lationships in light of their religious beliefs (e.g., adults: Results from the National AIDS Behavioral
sanctification; Mahoney & Tarakeshwar, 2005; Survey. American Journal of Public Health, 84,
Mahoney et al., 2001). Moreover, there are no 2003 2007.
questions about their religious communities that Davis, J. A., & Smith, T. W. (2002). General social
might further reveal contextual influences. surveys, 19912002: Cumulative codebook. Chica-
The results of this study indicate a powerful go: National Opinion Research Center.
and complex relationship between religious Edwards, J. N., & Booth, A. (1976). Sexual behavior
behavior with infidelity. It also calls into question in and out of marriage: An assessment of correlates.
the idea that beliefs without behavior may serve Journal of Marriage and the Family, 38, 73 81.
a protective factor with infidelity. Perhaps most Finer, L., Darroch, J., & Singh, S. (1999). Sexual
important of all, the present research focuses partnership patterns as a behavioral risk factor for
418 Journal of Marriage and Family

sexually transmitted diseases. Family Planning conceptual analysis of links between religion,
Perspectives, 31, 228 236. marriage, and parenting. Journal of Family
Glass, S. P., & Wright, T. L. (1977). The relationship Psychology, 15, 559 596.
of extramarital sex, length of marriage, and sex dif- Mahoney, A., & Tarakeshwar, N. (2005). Religions
ferences on marital satisfaction and romanticism: role in marriage and parenting in daily life during
Athanasious data reanalyzed. Journal of Marriage family crises. In R. F. Paloutzian & C. L. Park
and the Family, 39, 691 703. (Eds.), Handbook of psychology of religion and
Goodwin, R. (1992). Overall, just how happy are spirituality (pp. 177 195). New York: Guilford.
you? The magical Question 31 of the Spanier Previti, D., & Amato, P. R. (2004). Is infidelity
Dyadic Adjustment Scale. Family Therapy, 19, a cause or consequence of poor marital quality?
273 275. Journal of Social and Personal Relationships, 21,
Gordon, K. C., Baucom, D. H., & Snyder, D. K. 217 230.
(2004). An integrative intervention for promoting R Development Core Team (2007). R: A language
recovery from extramarital affairs. Journal of Mar- and environment for statistical computing. Vienna,
ital and Family Therapy, 30, 1 12. Austria: R Foundation for Statistical Computing.
Gorsuch, R. L. (1997). Exploratory factor analysis: http://www.R-project.org.
Its role in item analysis. Journal of Personality Schafer, J. L., & Graham, J. W. (2002). Missing data:
Assessment, 68, 532 560. Our view of the state of the art. Psychological
Gorsuch, R. L., & Walker, D. (2006). Measurement Methods, 7, 147 177.
and research design in studying spiritual develop- Schafer, J. L., & Olsen, M. K. (1998). Multiple impu-
ment. In E. C. Roehlkepartain, P. E. King, L. tation for multivariate missing-data problems: A
Wagener, & P. L. Benson (Eds.), The handbook of data analysts perspective. Multivariate Behavioral
spiritual development in childhood and adoles- Research, 33, 545 571.
cence (pp. 92 104). Thousand Oaks, CA: Sage. Sosis, R. (2000). Religion and intragroup coopera-
Harrell, F. E. (2001). Regression modeling strategies: tion: Preliminary results of a comparative analysis
With applications to linear models, logistic regres- of utopian communities. Cross-Cultural Research,
sion, and survival analysis. New York: Springer. 34, 70 87.
Harrell, F. E. (2005). Design: Design package. R Sosis, R., & Bressler, E. R. (2003). Cooperation and
package version 2.0-12. http://biostat.mc.vanderbilt. commune longevity: A test of the costly signaling
edu/s/Design. theory of religion. Cross-Cultural Research, 37,
Harrell, F. E. (2007). Hmisc: Harrell Miscellaneous. 211 239.
R package version 3.3-2. http://biostat.mc.vanderbilt. Spanier, G. B., & Margolis, R. L. (1983). Marital
edu/s/Hmisc. separation and extramarital sexual behavior. The
Johnson, C. M., Stanley, S. M., Glenn, N. D., Amato, Journal of Sex Research, 19, 23 48.
P. R., Nock, S. L., Markman, H. J., et al. (2002). Treas, J., & Giesen, D. (2000). Sexual infidelity
Marriage in Oklahoma: A 2001 baseline statewide among married and cohabitating Americans. Jour-
survey on marriage and divorce. Oklahoma City, nal of Marriage and the Family, 62, 48 60.
OK: Oklahoma Department of Human Services. Vaughn, D. (1990). Uncoupling: Turning points in
Liu, C. (2000). A theory of marital sexual life. Jour- intimate relationships. New York: Vintage.
nal of Marriage and the Family, 62, 363 374. Whisman, M. A., & Snyder, D. K. (2007). Sexual
Lumley, T. (2004). mitools: Tools for multiple impu- infidelity in a national survey of American women:
tation of missing data. R package version 1.0. Differences in prevalence and correlates as a func-
Vienna, Austria: R Foundation for Statistical Com- tion of method of assessment. Journal of Family
puting. Psychology, 20, 369 377.
Mahoney, A., Pargament, K. L., Tarakeshwar, N., & Wiederman, M. W. (1997). Extramarital sex: Preva-
Swank, A. B. (2001). Religion in the home in the lence and correlates in a national survey. The Jour-
1980s and 1990s: A meta-analytic review and nal of Sex Research, 51, 167 174.

You might also like