While a focus of increasing interest, the role of religion in mental illness and mental health remains understudied (1–5). In a prior article, we addressed this issue by asking 10 questions about religious beliefs and practices in a sample of more than 1,900 female twins from a population-based registry (6). We extracted two factors reflecting religious devotion and religious conservatism. Both of these dimensions were consistently associated with levels of substance use and misuse, had an inconsistent relationship with depression, and bore no association with anxiety or eating disorder symptoms or diagnoses.
A limitation of this and many other prior studies in this area (7, 8) has been the relatively simplistic approach taken toward the measurement of religiosity. Religion is a complex, multidimensional construct (9). Each of the numerous dimensions of religious beliefs, attitudes, and behavior might relate in a different way to the risk for psychiatric and substance use disorders.
This report has two goals. Using results from a new survey containing 78 items reflecting a diversity of religiosity constructs, we first sought to elucidate the dimensions of religiosity. Second, we attempted to clarify the relationship of these dimensions to the risk for lifetime psychiatric and substance use disorders.
Sample and Diagnostic Methods
The twin sample for this study was derived from two interrelated projects that used the population-based Virginia Twin Registry (10). The female-female twin pairs, with birth years between 1934 and 1974, became eligible if both members previously responded to a mailed questionnaire, to which the response rate was ∼64%. The eligible female twin pairs were approached for four waves of personal interviews from 1988 to 1997. The male-male and male-female twin pairs, with birth years between 1940 and 1974, were ascertained in a separate study and were approached for two waves of interviews from 1993 to 1998.
In late 1999, we mailed out questionnaires to all prior participants in these two studies (N=7,230). Very limited resources were available for follow-up of nonresponders. We received 2,621 questionnaires, a 36.3% response rate.
In prior waves, clinical interviewers assessed psychiatric and substance use disorders by personal interview using an adaptation of the Structured Clinical Interview for DSM-III-R (11) and the DSM-III-R criteria, with six exceptions. First, diagnostic hierarchies were not used. Second, a 1-month rather than a 6-month minimum duration of illness was used for generalized anxiety disorder (12). Third, assessment of drug abuse and dependence, which examined seven substance classes (cannabis, sedatives, stimulants, cocaine, opiates, hallucinogens, and other drugs), utilized the DSM-IV criteria (13). Fourth, nicotine dependence was defined as a score of ≥7 on the Fagerstrom Tolerance Questionnaire (14) during the period of heaviest lifetime use. Fifth, we utilized a broad definition of panic disorder requiring a history of spontaneous panic attacks meeting ≥2 symptomatic criteria reaching maximal intensity within 30 minutes (15). Sixth, phobia was defined as an irrational fear with objective behavioral impact on the respondents’ behavior as judged by the trained interviewer (16). The research reported here was approved by the Institutional Review Board of Virginia Commonwealth University, which required participants to provide signed consent before face-to-face interviews and mailed questionnaires and verbal assent before phone interviews.
Assessment of Dimensions of Religiosity
We sought to assess broadly religiosity, spirituality, and related attitudes, including forgiveness and gratitude. Our selection of scales and measures was based on an attempt to saturate the empirically ill-defined multivariate space that we call religiosity. As Little et al. (17) have argued, when constructs are poorly defined, attempts to maximize heterogeneity among items, all else being equal, will lead to the best representations of the underlying constructs.
After reviewing the literature, we selected 78 items from a number of sources (A1). First, we included all 24 items from the Religious Attitudes and Practices Inventory (18), which had four subscales: theism, with eight items; spirituality, with six items, social support, with seven items, and religious views on drug use, with three items. Second, we used all 10 items from our previous study of religiosity (6) (selected from items used in the National Comorbidity Survey , a Gallup poll , and the religiousness scale of Strayhorn et al. ). Third, we utilized six items from the "God images" scale (22): three each from the "God as love" and "God as authority" subscales. Fourth, from the Multidimensional Measurement of Religiousness/Spirituality (23), we included 11 items from the religious/spiritual coping scale and five items from the daily spiritual experiences scale. In four areas we were unable to find satisfactory items and developed our own: 1) "nature of God" scale, reflecting the level of perceived involvement of God in his creation, which consisted of four items (developed by K.S.K.), 2) forgiveness versus revenge, consisting of six items (developed by M.E.M. partly on the basis of a previous scale ), 3) gratitude versus ingratitude, consisting of six items (developed by M.E.M), and 4) love and caring, consisting of six items (developed by K.S.K.). With few exceptions, the items had four to six response options.
Repeated measures (mean=89.5 months apart, SD=5.1) were available for the 10 items on religiosity previously assessed (6) among ∼1,000 female twins. According to standard or weighted kappas (25, 26), item stability ranged from 0.33 to 0.67, with stability for most items between 0.45 and 0.60.
To examine how many separable dimensions existed in our item pool, we submitted a product-moment correlation matrix for all 78 items to a factor analysis and VARIMAX rotation (27). Ten factors had an eigenvalue ≥1.0. However, the scree plot indicated seven important factors (the final three having eigenvalues just above 1 and substantial loadings on only two or three items). In the service of parsimony, we utilized seven factors that explained 57.3% of the total variance. We examined the replicability of the factor structure by repeating our factor analysis in split-halves of the study sample. The observed structure was quite stable, with congruency coefficients (28) of 0.96 for factor 7 and ≥0.98 for factors 1–6.
Factor-derived scales were formed by assigning an item to the scale on which it loaded most heavily, given that the loading exceeded 0.40. Individuals with missing data for ≥50% of scale items were removed. For those completing ≥50% but <100% of the items (0.5% of the sample), scores on the missing items were imputed from the endorsed items. Three items did not load substantially on any factor.
We report three groups of analyses. First, given our modest response rate, we examined the representativeness of our sample by predicting cooperation from demographic, psychopathological, and (for female subjects only) religious variables. Since the analyses utilized a dichotomous dependent variable, they were performed by using logistic regression. To correct for correlated observations within twin pairs, we utilized generalized estimating equations implemented in PROC GENMOD in SAS (29). Second, we predicted our dimensions from demographic variables of sex, age, and educational level by using linear regression operationalized in PROC MIXED in SAS (29), treating individuals within pairs as repeated observations. Third, we examined the relationship between scores on these scales and risk for lifetime psychopathology and substance use disorder using logistic regression, with age, sex, and years of education as control variables. The religiosity measures were standardized, so the odds ratio reflected the change in risk for a disorder associated with a one standard deviation change in the religiosity factor.
Representativeness of Sample
One each of the 10 previously assessed religiosity items (6) and nine psychiatric and substance use disorders significantly predicted cooperation, a pattern not different from chance expectation (30). However, participation was substantially predicted by female sex (odds ratio=2.14, χ2=192.0, df=1, p<0.0001), increasing age (odds ratio=1.33 [per decade], χ2=82.1, df=1, p<0.0001), increasing education (odds ratio=1.12 [per year], χ2=113.4, df=1, p<0.0001), and monozygosity (odds ratio=1.47, χ2=43.3, df=1, p<0.0001).
Dimensions of Religiosity
The first dimension included 30 items reflecting 1) the person’s concern and involvement with spiritual issues, including sensing his/her place within the universe and 2) his/her active involvement with God on a day-to-day basis and at times of crisis (A1). The items in this factor came from two subscales of the Religious Attitudes and Practices Inventory (spirituality and theism), two subscales of the Multidimensional Measurement of Religiousness/Spirituality (daily spiritual experiences and religious coping), and the scale developed for our previous study (6). No single term captures entirely this factor, but general religiosity seemed appropriate.
The second dimension consisted of 12 items reflecting the degree of interaction with other religious individuals, the frequency of church attendance, and attitudes about substance use. Ten of the 12 items came from the Religious Attitudes and Practices Inventory subscales for social support and religious views on drug use. We called this dimension social religiosity. Because an association between this factor and substance use disorders might be "driven" by the three substance use items, we also used another version of this dimension, called social religiosity—M, from which these items were omitted.
Factor 3 included seven items from a range of sources, all of which included the word "God" and reflected a belief in a deity who is actively and positively involved in human affairs. We called this factor involved God.
Factor 4 consisted of seven items reflecting a caring, loving, and forgiving approach to the world. All of these items, which were positively worded, came from the love and caring or forgiveness versus revenge scales. The term God did not appear in these items. We termed this factor forgiveness.
All six items in factor 5 also contained the word "God" but differed from those in factor 3 by emphasizing the judgmental and punitive nature of the divinity. All three items from the "God as authority" subscale of the "God images" scale (22) were in this factor, which we called God as judge.
Factor 6 consisted of eight items reflecting an attitude toward the world emphasizing personal retaliation rather than forgiveness. With one exception, these items came from the forgiveness versus revenge and gratitude versus ingratitude scales and included all of the negatively worded items. For consistency, we scored these items in the positive direction, so that the factor was termed unvengefulness.
Factor 7 contained four items from the gratitude versus ingratitude and religious coping scales that reflected feelings of thankfulness versus anger toward life and God. We called this scale thankfulness.
Religiosity and Demographic Variables
Only two of the seven factors (God as judge and unvengefulness) were unrelated to age (t1). Five factors were positively associated with age, with the strongest relationship seen for social religiosity.
Significant sex differences were seen for all seven factors. For six factors, higher levels were seen for female than for male subjects, with the effect being particularly large for general religiosity and also substantial for social religiosity, involved God, and forgiveness. For one factor—God as judge—men had significantly higher levels than did women.
For four of the seven factors, a significant association was seen with years of education. For three of these factors (social religiosity, involved God, and God as judge), the relationship was negative. One factor, unvengefulness, was positively and significantly associated with years of education. The magnitude of the association with education was strongest for God as judge.
Religiosity and Lifetime Psychiatric and Substance Use Disorders
t2 shows the association, with age, sex, and years of education controlled, between the seven religiosity factors examined one at a time and the lifetime risk for nine psychiatric or substance abuse syndromes. We divided these nine syndromes into two groups: 1) five internalizing disorders—major depression, generalized anxiety disorder, phobia, panic disorder, and bulimia nervosa; and 2) four externalizing disorders—nicotine dependence, alcohol dependence, drug abuse or dependence, and adult antisocial behavior.
The level of general religiosity was significantly and positively related to risk for one of five internalizing disorders (panic disorder) and inversely and significantly linked with all four externalizing disorders. Levels of social religiosity were significantly and inversely related to risk for three of the five internalizing and all four externalizing disorders. The odds ratios were substantially lower for externalizing than internalizing disorders. Repeating these analyses with the modified social religiosity factor produced little change. Scores on the involved God factor were not significantly associated with risk for any internalizing disorder, but were inversely associated with risk for all four externalizing disorders.
Levels of forgiveness were related significantly to two disorders, with high levels associated with low risk for nicotine dependence and drug abuse or dependence. High levels of the God as judge factor were significantly associated only with a decreased risk for alcohol dependence and drug abuse or dependence. High levels of unvengefulness were associated with a significantly decreased risk for four of five internalizing disorders but no externalizing disorders. Increased levels of thankfulness were associated with a decreased risk for all disorders except panic disorder and adult antisocial behavior.
This report sought to 1) clarify the number of meaningful religiosity dimensions and 2) determine the association between these dimensions and risk for common psychiatric and substance use disorders.
Our results confirmed the complexity of the construct of "religiosity" (23). While we make no claims for coverage of all relevant dimensions (a recent book reviewed 126 published scales for religiosity !), our findings do strongly support the multidimensionality of religiosity. We emphasize six results. First, we identified a "social religious" factor that is similar to what others have termed religious "social support" (18, 23). Second, although others have claimed separable dimensions of religiosity and spirituality (9, 18), we could not distinguish these two domains. For example, most items from the Religious Attitudes and Practices Inventory spirituality subscale (18) and the daily spiritual experiences scale (23) loaded most heavily on the first general religiosity factor. A factor analysis using an oblique rotation (PROMAX) produced similar results. Third, we found no evidence for a separate "religious coping" factor (23) distinguishable from more general religious beliefs. Fourth, we identified a factor—God as judge—that resembles what we previously termed "religious conservatism" (6) and reflects beliefs about the nature of God most prominently seen in fundamentalist American Protestants. This factor had two distinct sociodemographic correlates, being more strongly endorsed by men and most strongly correlated with education. Fifth, attitudes often but not always associated with religiosity (e.g., forgiveness, gratitude, and love) could be separated from more formal religious and spiritual beliefs (9, 23). However, we did not discriminate these dimensions from one another. Rather, they split on positive or negative wording rather than on their item content.
Dimensions of Religiosity Associated With Psychiatric and Substance Use Disorders
Our seven factors were divisible into three groups with differing patterns of illness association. Two factors (social religiosity and thankfulness) were related to lifetime risk for both internalizing and externalizing disorders. Four factors (general religiosity, involved God, forgiveness, and God as judge) appeared to be more specific, with high scores predicting reduced risk only for externalizing disorders. One factor (unvengefulness) had the opposite pattern, as it was associated with internalizing but not externalizing disorders.
We are unaware of a specific precedent for this pattern of results or of a conceptual framework within which to view them. However, a number of specific findings are consistent with the literature. We comment on six. First, our results accord well with prior evidence that high levels of religious involvement predict a reduced risk for substance misuse (8, 31–35). Second, the observed inverse relationship between social religiosity and risk of illness is consistent with the hypothesis that religious activity reflects, in a "community of faith," a potent form of social integration (36). Third, a prior review of the relationship between religiosity and depression suggested that "intrinsic" religious motivation may be the aspect of religiosity most protective for depressive disorders (37). Is it possible that our measures of unvengefulness and thankfulness best tap those intrinsic attitudes that reduce risk for major depression? Fourth, the pattern for generalized anxiety disorder, phobia, and bulimia nervosa closely resembled that for major depression, with risk inversely related to levels of social religiosity, unvengefulness, and thankfulness. The pattern for panic disorder was different in that, consistent with prior literature (8, 38), high levels of general religiosity were associated with increased risk. Fifth, while several studies have examined the association between religiosity and smoking (6, 39, 40), far fewer have investigated nicotine dependence (6). The relationships between the dimensions of religiosity and nicotine dependence were very similar to those seen for drug use disorders. Finally, our results are consistent with a range of studies reporting an inverse relationship between religiosity and antisocial behaviors (3, 41, 42).
These results should be viewed in the context of six methodologic limitations. First, we did not address the causal nature of the reported associations. Religiosity may alter risk of illness, the experience of illness may have an effect on religiosity, or some third factor may influence both. Further research, particularly using a longitudinal design (43), will clarify this critical question.
Second, although our sample was large, we examined rare disorders, especially panic disorder, bulimia nervosa, and adult antisocial behavior. In several analyses, odds ratios for these disorders were similar to those seen for other conditions, but the results were not significant—a pattern likely due to low power.
Third, given our modest response rate, the representativeness of our sample is questionable. While neither a history of psychiatric and substance use disorders nor levels of religiosity predicted participation, strong effects were seen for age, sex, years of education, and zygosity. We examined the relationship between our religiosity dimensions and the risk for psychiatric or substance use disorders as a function of these four variables. For each variable, 63 analyses were performed (nine disorders times seven factors), and the number of significant interactions detected (6 of 63 for age, 7 of 63 for sex, 5 of 63 for years of education, and 6 of 63 for zygosity) did not significantly exceed chance expectation (30). We also weighted our data to the eligible sample as a function of these four variables and repeated a subset of the analyses seen in t2. No substantial differences emerged. Regarding the relationship between religiosity and psychopathology, our sample is probably representative of the twins who participated in the earlier interview waves.
Fourth, we took an approach to dissecting the dimensions of religiosity that was strictly empirical. This approach has strengths, especially when an extensive pool of items is applied to a large, representative population. However, it also has limitations in ignoring issues of face validity or prior work on scale dimensions. We argue that the distinctive relationships seen between our identified dimensions and both demographic factors and psychopathology speak to their validity.
Fifth, because of the rarity of certain disorders, we did not assess schizophrenia, bipolar illness, or anorexia nervosa and cannot address the relationship of religiosity to risk for these important disorders. Sixth and finally, our sample consisted of white men and women born in Virginia. Our findings may not generalize to other cultural or ethnic groups.
Received Nov. 16, 2001; revisions received April 30 and July 16, 2002; accepted July 23, 2002. From the Virginia Institute for Psychiatry and Behavioral Genetics and the Departments of Psychiatry and Human Genetics, Medical College of Virginia, Virginia Commonwealth University, Richmond; the Department of Psychology, University of Miami, Coral Gables, Fl.; and the International Center for Health and Spirituality, Rockville, Md. Address reprint requests to Dr. Kendler, Department of Psychiatry, Medical College of Virginia, P.O. Box 980126, Richmond, VA 23298-0126; email@example.com (e-mail). Supported in part by grant 519 from the John Templeton Foundation and NIH grants MH-40828, MH/AA-49492, AA-09095, and DA-11287. The authors thank the Virginia Twin Registry, now part of the Mid-Atlantic Twin Registry (MATR), for contributions to ascertainment of subjects for this study. The MATR, directed by Drs. L. Corey and L. Eaves, has received support from NIH, the Carman Trust, the William Keck Foundation, the John Templeton Foundation, and the Robert Wood Johnson Foundation. Dr. Larson died in March 2002.