Methodologic issues in research on religion and health.

Abstract: This study examines several methodologic issues in research on religion and health, including the measurement of the concept of religion, research designs, sampling, and statistical controls for assessing the "net" effects of religion on health outcomes. It briefly discusses differences in analytical perspectives that have contributed to the debate about the effects of religion on health. The authors review some of the methodologic problems of past research in this area of study and address what needs to be done to enhance the quality of the research. The authors conclude that the research methodology used in studies of religion and health has improved over time and that it continues to do so.

Key Words: religion, health, methodology, research

**********

Claude Bernard, (1) the father of experimental medicine, wrote in his 1865 book that "In scientific investigation, minutiae of method are of the highest importance ... In a word, the greatest scientific truths are rooted in details of experimental investigation, which form as it were, the soil in which these truths emerge." The importance of sound methodology cannot be overstated in medicine today. In her useful book, How to Read a Paper, Trisha Greenhalgh (2) quotes Dr. Stephen Lock, former editor of the British Medical Journal: "Few things are more dispiriting to a medical editor than having to reject a paper based on a good idea but with irremediable flaws in the methods used" (page 39).

One might think that medical research papers are so carefully and thoroughly scrutinized before their publication in a professional journal that the reader should be able to accept their findings at face value. However, Bland and Peacock (3) argue persuasively that this is not the case, and that despite the rigors of the peer-review process, the reader must be prepared to critically evaluate research articles. This point is driven home in a series of articles on evidence-based medicine, which appeared in The Journal of the American Medical Association between 1993 and 2000. (4-8) Therefore, the present article addresses the methodologic issues that should be considered when reading a scientific paper on the relation between religion and health. Many of the same issues must be considered when reading any scientific study, but we discuss them within the context of religion/spirituality and health.

After a brief discussion of the place of religion and spirituality in medicine and the potential ways in which religion may positively or negatively affect health, several methodologic issues will be considered. These issues cover the areas of measurement, research design, sampling methodology, and statistical control. Finally, we briefly discuss differences in analytical perspectives that have contributed to the debate about the effects of religion on health.

Religion, Spirituality, and Medicine

Many medical researchers naturally may be reluctant to study the influence of religion and spirituality on health. Early scientific progress in medicine was a battle against vitalism, the philosophical belief that animals possess a nonmaterial vital force or energy that gives them life. (1, 9) William Harvey's discovery in the 1600s that the heart acted as a pump to circulate the blood was one of the earliest scientific successes of mechanism over mysticism. Yet, Harvey himself believed in vitalism. (9) When Rene Descartes applied mechanical principles to the action of the muscles in 1650, he still maintained that they were activated by "animal spirits" that flowed through the nerves. (10)

In the 1800s, Claude Bernard (1) was concerned that medicine lagged behind the physical sciences in its use of the experimental method. In consequence, he said "We see [medicine] more or less mingled with religion and with the supernatural. Superstition and the marvelous play a great part in it. Sorcerers, somnambulists, healers by virtue of some gift from Heaven, are held as the equals of physicians. This state of affairs is the clearest of proofs that the experimental method has by no means come to its own in medicine." (1)

Over time, the mechanistic and deterministic underpinnings of medicine have prevailed. The workings of the body, with the exception of human thought, have been shown to follow physical laws and to be deterministic, in that they are attributable to natural causes. Yet, health and disease are the result of a complex interaction of natural processes, not all of which have been specified or, in some cases, even identified.

Having freed medicine from the shackles of the supernatural, physicians rightfully may feel squeamish about reintroducing it into their research and practice. However, the scientific study of religion and spirituality in relation to health is not the return of vitalism, nor is it the enemy of mechanism and determinism. Instead, a growing body of high quality research treats religion and spirituality as constructs that can and should be considered scientifically, as individual-level variables that may have direct and indirect effects on morbidity and mortality.

Positive ways religion can affect health

Religion can positively influence health in a number of ways. One way is simply that some religions endorse positive health-related behaviors and prohibit negative health-related behaviors. (11-13) In addition, many religious groups also tend to encourage moderation and discourage risk-taking behaviors. (13, 14) Adherence to religious norms may reduce exposure to various types of stressful events and conditions including family, interpersonal, legal, and other types of problems. (12, 14) Religion also may mediate physical and mental health by providing social support. Religious institutions often provide formal support through church programs, clergy assistance, and other types of activities. (15, 16) Religious involvement may also enhance access to support by strengthening marital quality, intergenerational ties, and otherwise contributing to family cohesiveness. (14, 17) It is well-established that social support has a strong positive effect on health, and people who regularly attend religious services have larger social networks, which provide companionship and access to the resources of the religious community, including financial and other forms of assistance (eg, providing meals, transportation, etc). (11, 12)

Psychologic resources are another possible mechanism through which religion may mediate health outcomes. (11, 12) These include feelings of self-esteem and self-worth, which have been associated with a range of physical and mental health outcomes. (11) There is substantial evidence that such psychologic resources are associated with religious participation. (18, 19) Religion may also engender positive emotions and provide a sense of meaning that helps individuals to cope with adverse situations, including health problems. (11, 12) Moreover, certain styles of religious coping (eg, perceived collaboration with God) tend to foster feelings of control and enhance one's ability to manage difficulties, which contribute to better health outcomes. At the physiologic level, we might expect the positive effects of religion to be mediated by stress hormones or immunologic functioning. (12, 20-22)

Negative ways religion can affect health

Researchers recently have started to examine possible deleterious effects of religion on mental and physical health. (23) These include various examples of maladaptive or problematic religious coping practices, including (a) passive coping styles, in which individuals surrender control or responsibility over problems to a higher power; (b) feelings of divine abandonment; (c) belief that personal problems reflect divine judgment or punishment; and (d) feelings of anger at God. Empiric studies have linked these negative religious coping practices with outcomes ranging from depression and anxiety (24) to poor rates of recovery from broken bones (25) and to increased risk of death in at least one clinical sample. (26)

Chronic religious doubting may also pose health problems, for at least two reasons: (a) chronic doubters may lack (what for many is) an important source of existential certainty and clarity, which may increase vulnerability to stressful conditions; and (b) they may feel guilt or embarrassment and consequently refrain from sharing their concerns with others. Although much work remains to be done in this area, several recent studies have shown that significant religious doubts are linked with psychologic distress and declines in physical health status. (27-29)

Tests and Measurement

When physicians think about tests and instruments, laboratory tests and surgical instruments are probably more likely to come to mind than questionnaires, which are commonly called tests and instruments by social scientists. Social scientists rely on questionnaires that ask people about their religious activities because it is not feasible to directly observe such activities on a large scale. And, since it is simply impossible to make an independent determination of what people believe or experience, self-reports of individuals' religious experiences and beliefs are an essential element of research on religion and health. Such self-reports can be obtained through self-administered tests, telephone interviews, or in-person interviews.

Probably the best-known health questionnaire, or scale, is the Short Form-36 (SF-36), which has been used in thousands of studies. The SF-36 was developed in the early 1990s as part of the Medical Outcomes Study, which was conducted in Boston, Chicago, and Los Angeles to look at health outcomes in patients with chronic medical and psychiatric problems. (30-34) It consists of a number of subscales that measure social functioning, physical functioning, level of energy/fatigue, bodily pain, general physical health, and mental health.

In medicine, the usefulness of a diagnostic test depends, in part, on two criteria: its sensitivity and specificity. These are, respectively, the degree to which it is able to detect a disease when it is present and its ability to exclude a disease when it is absent. (25-36) In the words of Jaeschke et al (6): "The accuracy of a diagnostic test is best determined by comparing it with the 'truth'" (page 390). Hence, self-report scales such as the SF-36 are tested in different clinical groups that are known to differ in the severity of their medical and psychiatric conditions to assess their accuracy in terms of sensitivity and specificity. (31, 32)

Validity and Reliability of Scales

The concepts of sensitivity and specificity are akin to the concept of validity in the psychologic field of tests and measurement, or psychometrics. (37, 38) The validity of a scale refers to the extent to which it measures what it is supposed to measure. Five types of validity are recognized, but the two most rigorous types are criterion validity and concurrent validity. Criterion validity is the most stringent type of validity and is similar to comparing a patient's score on a self-report scale with their actual clinical diagnosis to see if the results of the scale are consistent with the diagnosis, for example, the criterion. Concurrent validity is less stringent in that it uses scores from another, established scale that measures the same (or a similar) concept as the standard for judging the validity of the scale in question. This is done by calculating the correlation between the scores on the two scales.

The second important standard for evaluating a scale is its reliability. (37, 38) Although there are several types of reliability, the primary types are test-retest reliability and internal reliability. Test-retest reliability measures the degree to which people's scores on a scale are the same each time they take it. The correlation between the two scores provides the measure of reliability. Test-retest reliability may or may not be reported in a paper, so do not be surprised if it is not. However, one should always expect to be told about a scale's internal reliability.

Internal reliability measures the internal consistency of a scale. A scale is internally consistent to the degree that all the items in the scale are measuring aspects of a single concept. One should not include questions about geography in a chemistry test or questions about economics on a religiosity scale. Presumably, gross errors like this would not arise when developing a scale, but more subtle ones do, so the only way to be sure that a scale is internally consistent is to have people take it and compare how they respond to each of the questions or items on the scale. Internal reliability is presented in a form similar to a correlation coefficient with values ranging from zero to positive one. The most commonly used measure of internal reliability or consistency is the Cronbach [alpha]. (39) Authors sometimes present the results of a statistical procedure called factor analysis, which is another way of testing the internal consistency of a scale. (39, 40)

Multidimensionality of Religion

Religion has long been recognized by social scientists as a multidimensional concept, and a number of valid and reliable scales have been developed to measure the various aspects of religion. (41) However, since few of them were specifically designed for use in health research, researchers have argued for the development of such scales. (11, 42) One such scale is the "Brief Multidimensional Measure of Religiousness/Spirituality: 1999."

In 1995, the National Institute on Aging and the Fetzer Institute began a collaborative effort to identify and develop scales to measure the dimensions of religion that were considered to be the most likely to influence health. (43) Specialists from psychology, sociology, public health, and other health fields agreed on 10 religious dimensions they thought were important to study in relation to health. (44, 45) These are (1) religious affiliation or membership, (2) religious participation (attending formal services and activities), (3) religious beliefs and values, (4) private religious practices, such as prayer, meditation, and reading sacred literature, (5) religious commitment (the importance of religion relative to other aspects of life), (6) religious motivation for regulating and reconciling relationships, (7) religious support (tangible social support from members of one's religious group), (8) religious coping to deal with stressful events, (9) religious history in terms of upbringing and "turning points" in religious participation and beliefs, and (10) religious experiences with divine or sacred, as reflected in emotions and sensations. Research has mainly focused on 5 of the 10 dimensions: religious affiliation, religious participation, private religious practices, and religious coping. (12) These four dimensions tend to be moderately and positively associated with one another, and all four have been found to be associated with positive health outcomes. (12, 13, 46, 47)

The core work group involved in the project developed separate scales for each dimension, using a comprehensive nine-step strategy to develop the individual items that began with focus groups and in-depth interviews. (48) The culmination of their 3-year effort is the 38-item Brief Multidimensional Measure of Religiousness/Spirituality, which includes all 10 dimensions and global self-ratings of religiousness/religiousness and spirituality. The reliability and validity of the scale are presented in Idler et al. (44) It is available on-line in the appendix to the 2003 Fetzer report. (43)

Commonly Used Measures of Religion in Health Research

Religious affiliation is one of the most widely used measures of religion in medical and nursing research. (49-53) Religious affiliation provides two general classes of information: information about an individual's personal beliefs and information about their social environment. From the social epidemiologic perspective, knowing a person's religious affiliation can help us to understand the influence of social institutions on health and health behaviors. (11, 12, 54) However, church membership and religious affiliation are rather crude indexes of social involvement in religious institutions that do not measure the extent of social interaction, the amount and type of assistance provided by the social group, or other characteristics of social support. (12, 55-58) Religious affiliation is even less informative in terms of an intrapersonal or psychological perspective. Although knowing that a person belongs to a specific religious denomination implies something about the nature of one's religious beliefs, it does not provide information about the strength of those beliefs or one's adherence to the religious practices of that religious faith. Despite these limitations, numerous studies have found religious affiliation to be a predictor of health outcomes. (59-64) Some differences in health outcomes among religious denominations clearly are related to lifestyle differences. (12, 59, 65)

Self-reported religious attendance is another widely used measure of religion, and it has been found to be strongly related to physical and mental health. (12, 66-68) Although it can be difficult to establish the validity of self-report data, several studies have examined whether people attend religious services as often as they say the do when they answer a survey. One interesting study compared church attendance reported in a telephone poll in a county in Ohio to church attendance records provided by churches in the county. (69) When attendance records were not available, the authors counted the number of people who attended Sunday services or estimated the number of people who were attending services based on the number of cars in the church parking lot. The consensus from this and other studies is that people tend to overestimate the frequency with which they attend religious services. (69-72) The size of the overestimate can be affected by the specific wording of the question and how it is interpreted by the survey participant. (69, 71-73)

There is some evidence to suggest that the overestimate arises because many people see a question about religious attendance as if it were asking about their level of religious involvement, so they respond in such a way as to "truthfully" convey that they are "church-going" people. (69) Thus, they might say they went to church last Sunday, even if they didn't, simply because they usually do, or because they see themselves as religious or faithful individuals. (69) Individuals who think of themselves as religious (eg, because they pray, read the Bible, and so forth) could also be tempted to provide somewhat exaggerated accounts of their attendance because they think the question is intended to measure their overall religiousness rather than a literal accounting of their attendance. This could, in fact, help to explain why the relation between religious attendance and health is so robust. (74-76) In particular, the beneficial effects of church attendance on allcause mortality rates is the most solidly established positive effect of religion on health. (75)

Various reviews and critiques of the literature have noted that many studies that have examined the relation between religion and health used other, simple measures of religion instead of using a validated scale. (49-53) This is not surprising, given that religion often is not the primary focus of studies that have examined religion and health. (42, 77, 78) Many of the studies that have found a relation between religion and health have been large epidemiologic surveys in which religion was just one of a large set of variables. (42, 78) What is striking and impressive to these authors is that the health and mortality effects of religion are robust despite the severe limitations of the religious measures that are often used. (42, 78)

We used the Handbook of Religion and Health (79) to take a closer look at the sophistication of the religious measures used in four of the major areas covered in the book: anxiety, depression, well-being, and coping with physical disorders. Using the table in Chapter 24, we identified 283 studies, published between 1953 and 2000, in which all the data fields in the table were completely coded for the studies' findings and research methodology, including the number and kinds of religious measures used. We divided the studies into two categories with respect to whether or not they focused on religion. A study was classified as being focused on religion if its title contained the words religion, spirituality, church, the name of a religious denomination, or some other religious term.

We found that 215 of the 283 studies (76.0%) in the sample focused on religion. Thirty-four (15.8%) of these measured religion with a single question and 14 (6.5%) only measured religious affiliation. By comparison, 40 (58.5%) of the 68 studies in which religion was not the major focus only measured religion with a single question, and 11 (16.2%) only measured religious affiliation. On average, studies that focused on religion used significantly more questions (P < 0.001) to measure religion (mean = 9.5) than those in which religion was not the primary focus (mean = 2.3). Studies in which religion was the primary focus also measured significantly more (P < 0.001) religious dimensions (mean = 2.3) compared with studies in which religion was incidental to the study's purpose (mean = 1.3). The number of questions used to measure religion increased significantly over time (r = 0.17, P < 0.01).

These results indicate a greater degree of research sophistication with respect to measurement than we had expected to find based on past reviews of the literature, although Ellison and Levin (11) and Miller and Thoresen (78) note that the research methodology in this area has improved in recent years. An important point to be drawn from these results is that research critiques that lump studies in which religion is a minor or incidental variable together with studies that are specifically designed to investigate the relationship between religion and health give an inaccurate picture of the nature and adequacy of the religious measures being used in this field.

Designs, Samples, and Controls

Although Claude Bernard was a staunch advocate of the superiority of the experimental method, he recognized a continuum between the methods of the experimental and the observational sciences, such as astronomy, and he recognized that natural experiments might provide much the same kind of information as true experiments. Moreover, he believed that experimental reasoning, or the logic of experimentation, could help compensate for the lack of experimental control in observational studies. (1) Epidemiologic research is, by its nature, observational research that uses experimental reasoning in combination with appropriate methodological and statistical controls. (36, 80) The adequacy of these controls is one of the major aspects of a study that the reader must evaluate to see if its results are valid and meaningful and that the interpretation of the results is reasonable.

Research Designs

A variety of research designs are used in health research and the study of religion and health. Different designs are used for different purposes, and they also differ in the extent to which they allow strong causal inferences. Greenhalgh (2) presents a hierarchy of study designs that is consistent with the notion of strength of inference (4, 7) (the degree of certainty one should place on the results) and the assessment of the quality of different designs by other authors. (35, 36, 81, 82) From highest to lowest in terms of research design, they are (1) randomized controlled trials, (2) cohort studies, (3) case-control studies, (4) cross-sectional surveys, and (5) case reports. Randomized controlled trials are ranked first because they contain the three hallmarks of the true experiment: (1) random assignment of participants to treatment and control conditions, (2) manipulation of the treatment (the independent variable), and (3) control of potentially confounding variables. (83, 84) Since randomized clinical trials are true experiments, they provide the best evidence by which to make clinical decisions about the effects of different treatments. (4-7) Despite their advantages, randomized clinical trials are not always practical or applicable to the research question of interest. For example, experiments cannot answer questions about the incidence or prevalence of disease. This type of information is most often obtained from cross-sectional research. Thus, although randomized clinical trials are suited for studying clinical populations with respect to treatment, recovery, or disease management, they are not suited for studying the general population with respect to disease rates or disease prevention. The distinction between research on the general population and research on clinical populations is an important one in that there is no reason to assume that religious variables will exhibit the same relationships with a given outcome (eg, mortality) within clinical populations as they do among the largely healthy, general population.

Prospective cohort studies are ranked second in Greenhalgh's (2) hierarchy of research designs because they do not manipulate the independent variable, nor do they assign participants to conditions. (35) Since the independent variable is not manipulated by the researchers, it is uncertain that it is the actual cause of the observed outcome (the dependent variable). And, since the participants are not assigned to conditions, the samples within each condition could be biased in some way. Cohort studies are able to exert some control over extraneous variables by matching the demographic characteristics of people in the conditions, and they are able to infer causality from the temporal relation between exposure and subsequent outcomes. Still, exposure to the variable of interest may be confounded with other variables that co-vary with exposure to the independent variable. (2, 35)

Determining the cause of a condition is even more difficult in case-control studies because the studies are retrospective, with the comparison groups established after the fact. The temporal sequence of cause and effect that is possible to see in experimental and cohort studies cannot come into play. Instead, we have an effect and are looking back in time to try to determine the possible cause.

Cross-sectional studies are ranked just above case reports because they are mainly descriptive. Cross-sectional surveys should be given greater weight, in our opinion, to the degree that they statistically control for confounding variables, but it is difficult to draw causal inferences from them because independent and dependent (or out-come) variables are measured concurrently. (35, 81) In the sample of 283 studies we looked at in the Handbook of Religion and Health, (79) 81.3% were cross-sectional studies. Longitudinal studies have begun to emerge in the literature on religion and health, however, which support the findings from cross-sectional studies on the salutary effects of religion on depression, well-being, and overall health. (11, 85-87) The consistent findings of the salutatory effects of religion on mortality were obtained from longitudinal studies. (75)

Based on Greenhalgh's (2) hierarchy of study designs, we assigned a quality rating to each of the 283 studies in our sample from the Handbook of Religion and Health (79): 5 randomized controlled trials, 4 cohort studies, 3 case-control studies, 2 cross-sectional surveys, and 1 case report. A correlation analysis performed on the ratings revealed a small (r = 0.12) but statistically significant increase in the quality of study designs over time (P < 0.05).

Sampling

Sampling refers to the process of selecting a portion of a population to represent a population. (88,89) The more closely a sample reflects the characteristics of the population, the more representative it is of the population. In turn, the more representative a sample is, the more confident one can be in generalizing the findings from the sample to the population from which it is drawn (eg, patients with lung cancer or by-pass surgery or women over age 60 years). Therefore, the major criterion for evaluating the adequacy of a sample should be the degree to which it accurately represents the population. Unfortunately, it is often impossible to know this. Therefore, we typically judge the quality of a sample by the methods used to select it.

The sampling method, like the research design, should be selected in accordance with the purpose and goals of the study. Sampling methods fall into two broad categories: random (probability) sampling, and nonrandom (nonprobability) samples. (88,89) The hallmark of random sampling is that each person in the population of interest has a specific probability of being included in the sample. Findings based on data from a random sample can be generalized only to the population from which the given sample was drawn. Hence, George et al (12) caution that some of the findings about the relation between religion and health may not be generalizable to the population at large because much of the research in this area has been conducted on older people. Positive effects of religion on mental health have been found in the few studies that have used random samples of the general population of the United States, in which the representativeness of the samples could be verified by comparing them to the US Census. (90-92) A number of studies of religion and mortality have also used probability samples of the general population of the United States (93,94) or specific geographical areas in the United States. (95-98)

Random sampling is especially important when one wishes to generalize to the population at large, such as research on the salutary effects of religion, which typically use samples of community dwelling (mainly healthy) individuals. The need for random sampling is not as great in studies of individuals with specific health conditions, which have key characteristics in common, namely the features of the disease, and it is fairly common for the latter kinds of research to use nonprobability samples.

Although there are a number of procedures for selecting nonprobability samples, the most commonly used are purposive and convenience sampling. (88, 89) Purposive sampling entails choosing participants for a study because they have certain specific characteristics. This procedure is widely used in case-control studies. Convenience sampling refers to the selection of a sample that is conveniently available to the researchers. Psychologic experiments, for example, commonly use convenience samples of college students enrolled in psychology courses, but they assign participants randomly to the experimental conditions (eg, treatment versus control groups). It is important to remember that random assignment to conditions is different from random sampling and that random assignment and random sampling serve somewhat different functions. Random assignment is performed to enhance the ability to make causal inferences from a study by ensuring that the observed effects of treatment are not due to initial differences between groups. Random sampling is conducted to enhance the ability to generalize the findings of a study to the population of interest. Nonetheless, they both serve a similar function by attempting to reduce the likelihood that the results are biased by the particular, or peculiar, nature of the individuals in the study. (84)

Statistical controls, which are discussed in the next section, are a way of mimicking random assignment. Obviously, people cannot be randomly assigned to different sexes, races/ethnicities, social classes, ages, religious backgrounds, and so forth. Statistical controls adjust for these differences allowing researchers to estimate the "net" effect of religiosity (or any independent variable) on a given outcome, even when random assignment is not possible.

Statistical Controls

A number of factors or causes often act in concert to produce a disease outcome, including intrinsic personal factors, environmental factors, and disease agents. Although some variables may contribute to the outcome, others might ameliorate it or be irrelevant. Technically, statistical control means the use of statistical methods to "identify, isolate, or nullify" the variations in an outcome variable that are caused by other variables that are extraneous to the particular relation or relations under study. (99) Kleinbaum et al (100) provide an example: "In describing the relationship between blood pressure and physical activity, we would probably consider 'age' and 'sex' as control variables because they are related to blood pressure and physical activity, and unless taken into account, could confound any conclusions regarding the primary relationship of interest."

To statistically control for a variable, it must be measured and its effect on the variable of interest must be analyzed. In their Handbook of Religion and Health, Koenig et al (79) classified the studies they reviewed into 3 categories in terms of the quality of their statistical control for confounding variables: (1) no statistical controls, (2) some important variables not controlled for, and (3) statistical control of important variables using regression analysis. Although this classification system is somewhat rudimentary, it provides another measure by which to assess changes in the sophistication of research methodology across time. And, once again, bivariate correlation showed that the research methodology of the 283 studies that we analyzed substantially improved over the years (r = 0.30, P < 0.001).

The statistical procedures that are most often used for this purpose are called multiple regression techniques, which simultaneously analyze the effects of many different variables on a dependent (eg, outcome) variable. (99,100) These procedures include ordinary least-squares, maximum likelihood, and logistic regression, among others. Logistic regression is widely used in epidemiologic research for three reasons. First, it is designed for use on dichotomous dependent variables: a case versus a noncase. Second, it can provide estimates of the odds ratio for cases versus noncases. The odds ratio is, of course, the relative probability of having a disease (eg, lung cancer) for specified groups (eg, smokers versus nonsmokers) expressed in the form of odds (eg, 3 to 1). (100) Third, logistic regression provides an index of the combined contribution of several risk factors. So, in addition to its ability to nullify the effects of confounding variables it can mathematically predict cumulative effects of multiple risk factors.

With respect to the study of the relation between religion and health, regression can help identify if a health outcome that is associated with some measure of religion is due to other variables that may themselves be related to religion in some way, such as socioeconomic status or race. The effects of religion/spirituality can only be isolated with certainty if a number of variables that may be related to it are measured and included in the regression model. (99,100) To the degree that women and poorer people are more religious, an observed relation between religion and health might really be due to the sex or the socioeconomic status of the sample. If sex and socioeconomic status are not included in the regression analysis to statistically control for them, we might erroneously conclude that the observed health effect is due to religion.

Analytical Perspectives

It is vital, therefore, that studies on religion and health include statistical controls for a number of potentially confounding variables. These include age, sex, ethnicity, education, and socioeconomic status. (12,76) There is disagreement, however, over which controls beyond these sociodemographic factors should be included in religion-health research, and how the results should be interpreted. In particular, conceptual differences between two analytical perspectives have helped to fuel the debate about the beneficial effects of religion on health. One has been called the "value added" approach, and the other the "web of causality" approach. Briefly, the former refers to the common epidemiologic interest in identifying variables that add unique "variance explained" to models that include well-established predictors of morbidity or mortality. From this perspective, if religion influences mortality partly through reductions in smoking and heavy drinking, partly through social support, partly through coping, and so forth, these indirect effects would be of little interest, because previous research already shows that these factors predict mortality. Researchers from the "value added" school of thought would not care that these pathways may illustrate indirect effects of religion. Instead, the only religious influences of interest would be those that cannot be attributed (or "reduced") to other, known mechanisms. In contrast, proponents of the "web of causality" approach (which is more common in sociology, psychology, and public health) think the indirect effects of religion are important. They help us to understand what factors influence health behaviors, social support, and other key factors that affect health and that this knowledge can have valuable intellectual and practical implications, for example, perhaps influencing public health interventions.

The fundamental differences between "value added" and "web of causality" thinking may have contributed in important ways to the misunderstanding that currently exists between social scientists and some critics within the medical profession. (101) Thus, distinguishing between these perspectives is important when making one's own decisions about research findings. At times it may be important to precisely identify the part of the total religious effect that is distinctively religious. For example, studies suggest that social support from coreligionists has a positive effect on well-being above and beyond that of secular forms of social support. (76) At other times it may be important to look at both the direct and indirect effects of religion, for example, the effect of religion on mortality, which we just mentioned. However, the "value added" approach ignores the influence of religion on lifestyle and behaviors that reduce exposure to disease risk factors. (74)

Summary and Assessment

There has been a tendency in healthcare research to use simple measures of religion, despite the availability of a number of suitable scales. It would be worthwhile, therefore, for researchers to acquaint themselves with such measures. Despite this tendency, our analysis of a sample of 283 studies contained in the Handbook of Religion and Health (79) revealed that studies that were specifically designed to examine the relationship between religion and health used more sophisticated scales that tapped multiple dimensions of religion.

The recent development of "Brief Multidimensional Measure of Religion/Spirituality: 1999" (43) may prove to be a boon for researchers. But whatever scale is used, the reader should be provided adequate information about its validity and reliability.

Unfortunately, most of the studies in this area have relied on cross-sectional designs, which make it difficult to draw causal inferences from their results. However, the number of longitudinal studies is increasing, and their findings provide stronger support for the positive relation between religion and health that is indicated by the large body of cross-sectional research. Our analyses also show that the methodologic sophistication of the research is increasing in terms of quality of measures used, study design, and statistical controls.

Important public health issues, such as the negative effects of cigarette smoking and air pollution, were unresolved for years because the data came from nonexperimental research. (102-108) In this context, Hill (102) set out 8 criteria for inferring causality from nonexperimental studies, which he applied to illustrate that a causal inference can be drawn from the association between smoking and lung cancer: (1) strength of association, (2) consistency of association, (3) specificity of the association, (4) temporality of the relation, (5) that the association exhibits a biological gradient, or dose-response curve, (6) the plausibility of the causal connection, for example, that it is biologically plausible, (7) the coherence of the proposed cause-effect relation with the natural history of a disease, and (8) analogy, that is, proposed effect is analogous to other known effects. But, as Lave and Seskin (105) commented about the association between smoking and cancer over a decade later: "Some scientists will not accept epidemiologic evidence as proof of causation no matter how elaborate it is." This holds true for some of the critics of research on religion and health.

A decade ago, Levin (19) applied Hill's criteria to evaluate the relation between religion and health. While recognizing the methodologic issues involved in such research, Levin (19) concluded that a causal connection between religion and health is generally supported by Hill's criteria. Support is stronger for some religious effects than others, (109) such as the salutary effect of religious involvement on hypertension (22) and the beneficial effects of church attendance on all-cause mortality. (75) A recent review by Powell et al (74) also finds the evidence supporting the benefits of church attendance on mortality rates to be persuasive, but it finds less support for the relation between religion and other health outcomes. However, this is probably due to the fact that it excluded cross-sectional studies, which, as we have found, comprise more than 80% of the research.

Conclusion

In summary, the most common methodologic problems cited in a review of the field are the use of simplistic measures of religion, unrepresentative samples, cross-sectional study designs, simple statistics (such a bivariate or zero-order correlations) that do not control for confounding variables, and post hoc findings from studies that were not specifically designed to study the effects of religion on health. Both the reader and the researcher should be wary of these problems. As Miller and Thoresen (78) note, however, it is "quite easy to find inherent flaws in single studies." What is more important is the replication of findings across a number of studies. In our view, the body of research to date provides consistent and increasingly convincing evidence that various aspects of religion exert positive influences on health. And as we present herein, the research methodology in the field of religion and health is continuing to improve.

He who hesitates is a damned fool.
--Mae West

Accepted August 30, 2004.

References

1. Bernard C. Experimental Medicine. London, Transaction Publishers, 1999 [originally published in 1865].

2. Greenhalgh T. How to Read a Paper: The Basics of Evidence Based Medicine. 2nd edition. London, BMJ Books, 2001.

3. Bland M, Peacock J. Statistical Questions in Evidence-Based Medicine. Oxford, Oxford University Press, 2000.

4. Oxman AD, Sackett MD, Guyatt GH. Users' guides to the medical literature, I: how to get started. JAMA 1993;270:2093-2095.

5. Guyatt GH, Sackett DL, Cook DJ. Users' guides to the medical literature, II: how to use an article about therapy or prevention: are the results of the study valid. JAMA 1993;270:2598-2601.

6. Jaeschke R, Guyatt GH, Sackett DL, Users' guides to the medical literature, III: how to use an article about a diagnostic test: are the results of the study valid. JAMA 1994;271:389-391.

7. Guyatt GH, Sackett DL, Sinclair JC, et al. Users' guides to the medical literature, IX: a method for grading health care recommendations. JAMA 1995;274:1800-1804.

8. Giacomini MK, Cook DJ. Users' guides to the medical literature, XXIII: qualitative research in health care: are the results of the study valid? Evidence-based medicine working group. JAMA 2000;284:357-362.

9. Beck WS. Modern Science and the Nature of Life. Garden City, NY, Doubleday, 1961.

10. Descartes R. The passions of the soul, in Wayne D (ed): Readings in the History of Psychology. New York, NY, Appleton-Century-Crofts, 1927.

11. Ellison CG, Levin JS. The religion-health connection: evidence, theory, and future directions. Health Educ Behav 1998;25:700-720.

12. George LK, Ellison CG, Larson DB. Explaining the relationships between religious involvement and health. Inquiry 2002;13:190-200.

13. Levin JS. Religion and health: is there an association, is it valid, and is it causal? Soc Sci Med 1994;38:1475-1482.

14. Sherkat DE, Ellison CG. Recent developments and current controversies in the sociology of religion. Annu Rev Sociol 1999;25:363-394.

15. Taylor RJ, Ellison, CG, Chatters LM, et al. Mental health services in faith communities: the role of clergy in black churches. Soc Work 2000;45:73-87.

16. Krause N, Ellison CG, Shaw BA, et al. Church-based social support and religious coping. J Sci Study Relig 2001;40:637-656.

17. Ellison CG, George LK. Religious involvement, social ties, and social support in a southeastern community. J Sci Study Relig 1994;33:46-61.

18. Ellison CG. Religious involvement and self-perception among black Americans. Soc Forces 1993;71:1027-1055.

19. Krause N. Religiosity and self-esteem among older adults. J Gerontol Psychol Sci 1995;50:P236-P246.

20. Koenig HG, Cohen HJ, George LK, et al. Attendance at religious services, Interleukin-6, and other biological indicators of immune function in older adults. Int J Psychiatry Med 1997;27:233-250.

21. Levin JS. How religion influences morbidity and health: reflections on natural history, salutogenesis, and host resistance. Soc Sci Med 1996;43:849-864.

22. Levin JS, Vanderpool HY. Is religion therapeutically significant for hypertension? Soc Sci Med 1989;29:69-78.

23. Pargament KI. The Psychology of Religion and Coping. New York, NY, Guilford Press, 1997.

24. Exline JJ, Yali AM, Lobel M. When God disappoints: difficulty for-giving God and its role in negative emotion. J Health Psychol 1999;4:265-279.

25. Fitchett G, Rybarczyk BD. DeMarco GA, Nicholas JJ. The role of religion in medical rehabilitation outcomes: a longitudinal study. Rehabil Psychol 1999;44:333-353.

26. Pargament KI. The bitter and the sweet: an evaluation of the costs and benefits of religiousness. Inquiry 2002;13:168-181.

27. Krause N, Ingersoll-Dayton B, Ellison CG, Wulff KM. Aging, religious doubt, and psychological well-being. Gerontologist 1999;39:525-533.

28. Krause N, Wulff KM. Religious doubt and health: exploring the potential dark side of religion. Sociol Relig 2003;65:35-56.

29. Pargament KI, Koenig HG, Tarakeshwar N, Hahn J. Religious struggle as a predictor of mortality in medically ill elderly patients: a two-year longitudinal study. Arch Intern Med 2001;161:1881-1885.

30. Ware JE, Sherbourne CD. The MOS 36-item short-form health survey (SF-36), I: conceptual framework and item selection. Med Care 1992;30:473-483.

31. McHorney CA, Ware JE, Rogers W, et al. The validity and relative precision of MOS short- and long-form health status scales and the Dartmouth COOP charts: results from the Medical Outcomes Study. Med Care 1992;30:MS253-MS265.

32. McHorney CA, Ware JE, Raczik AE. The MOS 36-item short-form health survey (SF-36), II: psychometric and clinical test of validity in measuring physical and mental health constructs. Med Care 1993;31:247-263.

33. McHorney CA, Ware JE, Lu JFR, Sherbourne CD. The MOS 36-item short-form health survey (SF-36), II: test of data quality, scaling assumptions, and reliability across diverse patient groups. Med Care 1994;32:40-66.

34. Ware JE, Kosinski M, Bayliss MS, et al. Comparison of the methods for the scoring and statistical analysis of the SF-36 health profile and summary measures: summary of results from the Medical Outcomes Study. Med Care 1995;33:AS264-AS279.

35. Katz DL. Clinical Epidemiology and Evidence-Based Medicine: Fundamental Principles of Clinical Reasoning and Research. Thousand Oaks, CA, Sage Publications, 2001.

36. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiological Research: Principles and Quantitative Methods. New York, NY, Van Nostrand Reinhold, 1982.

37. Anastasi A. Psychological Testing, 3rd edition. New York, NY, Macmillan Publishing, 1968.

38. Nunnally JC, Bernstern IH. Psychometric Theory. 3rd edition. New York, NY, McGraw-Hill, 1994.

39. Cronbach LJ. Coefficient alpha and the internal structure of tests. Psychometrika 1951;16:297-334.

40. Gorsuch RL. Factor Analysis. 2nd edition. Mahwah, NJ, Lawrence Erlbaum, 1983.

41. Hill PC, Hood RW (eds): Measures of Religiosity. Birmingham, Religious Education Press, 1999.

42. Hill PC, Pargament KI. Advances in the conceptualization and measurement of religion and spirituality: implications for physical and mental health research. Am Psychol 2003;58:64-74.

43. Fetzer Institute, National Institute on Aging Working Group. Multidimensional Measurement of Religiousness/Spirituality for Use in Health Research: A Report of the Fetzer Institute, National Institute on Aging Working Group. 2003. Available at http://www.fetzer.org/Resources/resources_multidimens.htm. Accessed May 13, 2003.

44. Idler EL, Musick MA, Ellison CG, et al. Measuring multiple dimensions of religion and spirituality for health research: conceptual background and findings from the 1998 general social survey. Res Aging 2003;25:327-365.

45. George LK, Larson DB, Koenig HG, McCullough ME. Spirituality and health: what we know, what we need to know. J Soc Clin Psychol 2000;19:102-116.

46. Levin JS, Larson DB, Puchalski CM. Religion and spirituality in medicine: research and education. JAMA 1997;278:792-793.

47. Levin JS, Chatters LM. Religion, health, and psychological well-being in older adults. J Aging Health 1998;10:504-531.

48. Krause N. A comprehensive strategy for developing closed-ended survey items for use in studies of older adults. J Gerontol Soc Sci 2002;57:S263-S274.

49. Larson DB, Pattison EM, Blazer DG, et al. Systematic analysis of research variables in four major psychiatric journals. Am J Psychiatry 1986;143:329-334.

50. Sherrill KA, Larson DB, Greenwold M. Is religion taboo in gerontology? Am J Geriatr Psychiatry 1993;1:109-117.

51. Orr RD, Issac G. Religious variables are infrequently reported in clinical research. Fam Med 1992;24:602-606.

52. Flannelly LT, Flannelly KJ, Weaver AJ. Religious and spiritual variables in three major oncology nursing journals: 1990-1999. Oncol Nurs Forum 2002;29:679-685.

53. Weaver AJ, Flannelly LT, Flannelly KJ. A review of research on religious and spiritual variables in two primary gerontological nursing journals: 1991-1997. J Gerontol Nurs 2001;27:335-340.

54. Kawachi I, Berkman LF. Social cohesion, social capital and health, in Berkman LF, Kawachi I (eds): Social Epidemiology. Oxford, Oxford University Press, 2000, pp 191-212.

55. Ellison CE, George LK. Religious involvement, social ties, and social support in a southeastern community. J Sci Study Relig 1994;33:46-61.

56. Krause N, Ellison CG, Shaw BA, et al. Church-based social support and religious coping. J Sci Study Relig 2001;40637-656.

57. Krause N. Social support, in Binstock RH, George LK (eds). Handbook of Aging and the Social Sciences. 5th edition. New York, Academic Press, 2001, pp 272-294.

58. Berkman LF, Glass T. Social integration, social networks, social support, and health, in Berkman LF, Kawachi I (eds): Social Epidemiology. Oxford, Oxford University Press, 2000, pp 174-190.

59. Jarvis GK, Northcott HC. Religion and differences in morbidity and mortality. Soc Sci Med 1987;25:813-824.

60. Koenig HG, Larson DB. Use of hospital services, religious attendance, and religious affiliation. South Med J 1998;91:925-932.

61. Ellison CG, Gay DA, Glass TA. Does religious commitment contribute to individual life satisfaction? Soc Forces 1989;68:100-123.

62. King DE, Hueston W. Religious affiliation and obstetric outcome. South Med J 1994;87:1125-1128.

63. Ferraro KF, Koch JE. Religion and health among black and white adults: examining social support and consolation. J Sci Study Relig 1994;33:362-375.

64. Ferraro KF, Albrecth-Jensen CM. Does religion influence adult health? J Sci Study Relig 1991;30:193-202.

65. Fraser GE. Diet as primordial prevention in Seventh-Day Adventists. Prev Med 1999;29:S18-S23.

66. Ellison CG. Race religious involvement and depressive symptomatology in a southeastern US community. Soc Sci Med 1995;40:1561-1572.

67. Koenig HG, George LK, Hayes JC, et al. The relationship between religious activities and blood pressure in older adults. Int J Psychiatry Med 1998;28:189-213.

68. Koenig HG, Hayes JC, Larson DB, et al. Does religious attendance prolong survival? A six-year follow-up study of 3,968 older adults. J Gerontol A Biol Sci Med Sci 1999;54:M370-M376.

69. Hadaway CK, Marler PL, Chaves M. What the polls don't show: a closer look at US church attendance. Am Sociol Rev 1993;58:741-752.

70. Hout M, Greeley A. What church officials' reports don't show: another look at church attendance data. Am Sociol Rev 1998;63:113-119.

71. Presser S, Stinson L. Data collection mode and social desirability bias in self-reported religious attendance. Am Sociol Rev 1998;63:137-145.

72. Smith TW. A review of church attendance measures. Am Sociol Rev 1998;63:131-136.

73. Woodberry RD. When surveys lie and people tell the truth: how surveys over sample church attenders. Am Sociol Rev 1998;63:119-122.

74. Powell LH, Shahabi L, Thoresen CE. Religion and spirituality: linkages to physical health. Am Psychol 2003;58:36-52.

75. McCullough ME, Hoyt WT, Larson DB, et al. Religious involvement and mortality: a meta-analytic review. Health Psychol 2000;19:211-222.

76. Ellison CG, Boardman JD, Williams DR, Jackson JS. Religious involvement, stress, and mental health: findings from the 1995 Detroit Area Study. Soc Forces 2001;80:215-249.

77. Flannelly KJ, Weaver AJ, Costa KG. A systematic review of religion and spirituality in three palliative care journals, 1990-1999. J Palliat Care 2004;20:50-56.

78. Miller WR, Thorensen CE. Spirituality, religion, and health: an emerging research field. Annu Rev Sociol 1999;25:363-394.

79. Koenig HG, McCullough ME, Larson DB. Handbook of Religion and Health. Oxford, Oxford University Press, 2001.

80. McNeil D. Epidemiological Research Methods. New York, Wiley, 1998.

81. Silman AJ, Macfarlane GJ. Epidemiological Studies. Cambridge, Cambridge University Press, 2002.

82. Mausner JS, Kramer S. Epidemiology: An Introductory Text. 2nd edition. Philadelphia, WB Saunders, 1985.

83. Edwards AL. Experimental Design in Psychological Research. 5th edition. New York, Harper & Row, 1985.

84. Kerlinger FN. Foundations of Behavioral Research. 3rd edition. New York, Holt, Rhinehart and Winston, 1986.

85. Idler EL, Kasl SV. Religion among disabled and nondisabled persons, II: attendance at religious services as a predictor of the course of disability. J Gerontol B Psychol Sci Soc Sci 1997;52B:S306-S316.

86. Krause N. Neighborhood deterioration, religious coping, and changes in health during late life. Gerontologist 1998;38:653-664.

87. Levin JS, Markides KS, Ray LA. religious attendance and psychological well-being in Mexican Americans: a panel analysis of three-generations data. Gerontologist 1996;36:454-463.

88. Bailey KD. Methods of Social Research. 3rd edition. New York, The Free Press, 1987.

89. Babble ER. The Practice of Social Research. Belmont, CA, Wadsworth, 1975.

90. Ellison CG, Gay DA, Glass TA. Does religious commitment contribute to individual life satisfaction? Social Forces 1989;68:100-123.

91. Levin JS, Chatters LM, Taylor RJ. Religious effects on health status and life satisfaction among black Americans. J Gerontol B Pychol Sci Soc Sci 1995;50:S154-S163.

92. Nooney J. Woodrum E. Religious coping and church-based social support as predictors of mental health outcomes: testing a conceptual model. J Sci Study Religion 2002;41:359-368.

93. Hummer RA, Rogers RG, Nam CB, et al. Religious involvement and U.S. adult mortality. Demography 1999;36:273-285.

94. Musick MA, House JS, Williams DR. Attendance at religious services and mortality in a national sample. J Health Soc Behav 2004;45:198-213.

95. Strawbridge WJ, Cohen RD, Shema SJ, et al. Frequent attendance at religious services and mortality over 28 years. Am J Public Health 1997;87:957-961.

96. Oman D, Kurata JH, Strawbridge WJ, et al. Religious attendance and cause of death over 31 years. Int J Psychiatry Med 2002;32:69-89.

97. Oman D, Reed D. Religion and mortality among the community-dwelling elderly, Am J Public Health 1998;88:1469-1475.

98. Koenig HG, Hays JC, Larson DB, et al. Does religious attendance prolong survival? A six-year follow-up study of 3,968 older adults. J Geronotol 1999;54A:M370-M377.

99. Pedhazur EJ. Multiple Regression in Behavioral Research: Explanation and Prediction. 2nd edition. New York, Holt, Rhinehart and Winston Press, 1982.

100. Kleinbaum DG, Kupper LL, Muller KE, Nizam A. Applied Regression Analysis and Other Multivariable Methods. 3rd edition. Pacific Grove, CA, Duxbury Press, 1998.

101. Sloan RP, Bagiella E, Powell T. Religion, spirituality, and medicine. Lancet 1999;53:664-667.

102. Hill AB. The environment and disease. Proc R Soc 1965;58:295-300.

103. Wynder EL. Goodman MT. Smoking and lung cancer: some unresolved issues. Epidemiol Rev 1983;5:177-207.

104. Loeb LA, Ernster VL, Warner KE, et al. Smoking and lung cancer: an overview. Cancer Res 1984;44:5940-5958.

105. Lave LB. Seskin EP. Epidemiology, causality, and public health. Am Scient 1979;67:178-186.

106. Spiezer FE. Assessment of the epidemiological data relating lung cancer to air pollution. Environ Health Perspect 1983;47:33-42.

107. Ostro B. The association of air pollution and mortality: examining the case for inference. Arch Environ Health 1993;48:336-342.

108. Hemminki K, Pershagen G. Cancer risk of air pollution: epidemiological evidence. Environ Health Perspect 1994;102(Suppl 4):187-192.

109. Levin JS, Vanderpool HY. Is frequent religious attendance really conducive to better health? Toward an epidemiology of religion. Soc Sci Med 1987;24:589-600.

RELATED ARTICLE: Key Points

* Although there are methodologic deficiencies in many of the studies that have reported a positive relation between religion and health, the methodologic sophistication of research in this area has steadily improved, especially in the last several years.

* The application of epidemiologic criteria for inferring causality supports the contention that religion has a positive influence on various health outcomes.

* The increasing use of longitudinal designs and statistical controls will permit researchers to obtain a better understanding of the causal relation between religion and health.

* Some of the salutary effects of religion are dismissed by critics because they result from the indirect mediating influence of religion on risk factors that affect health rather than affecting health directly.

Kevin J. Flannelly, PHD Christopher G. Ellison, PHD, and Adrienne L. Strock, MLIS 

From the Population Research Center and Department of Sociology, University of Texas at Austin, Austin, TX, and The HealthCare Chaplaincy, New York, NY.

Reprint requests to Dr. Kevin Flannelly, Pastoral Care, Education, and Research Center, 307 East 60th Street, New York, NY 10022-1505. E-mail: [e-mail]