|Return to: Spirituality Articles
Empirical Support for a Model of Well-Being, Meaning in Life,
Importance of Religion, and Transcendent Experiences
James E. Kennedy and H. Kanthamani
Published on the internet in pdf at http://jeksite.org/research/path.pdf.
Copyright 1995 James E. Kennedy
Abstract: A model developed in an investigation of the effects of transcendent experiences on subjective well-being may provide insight into the weak, positive correlation between religious commitment and well-being. This model suggests that religious commitment influences a person's sense of meaning in life, which, in turn, influences well-being. The model also suggests that transcendent experiences can affect religious commitment, which then influences meaning in life and well-being. The data from a convenience sample of 182 people are very consistent with this causal chain model. More importantly, numerous other studies of the relationships between specific components of the model are consistent with the model. However, the available data and structural equation methods are ambiguous about the direction of causation along this chain path, and reciprocal or bi-directional causation is likely. Although the direction of causation may vary, the intervening or mediating roles appear to apply with either causal direction.
The weak relationship between religious involvement and psychological well-being is somewhat surprising given the prevalence of religious beliefs in the United States. Gallup surveys report that 94% of Americans say they believe in God, 54% say that religion is very important in their lives, and 38% say their religious involvement has been a "very positive" experience (Gallup and Castelli, 1989:35,45). Beliefs that have such widespread affirmation might be expected to have a strong positive impact on peoples' lives. In addition, as Koenig (1990) noted, a strong association between religious commitment and well-being would be expected based on individual testimonials. However, reviews of research on the relationship between religious commitment and psychological well-being find only a weak positive association overall (Bateson, Schoenrade and Ventis, 1993:287; Bergin, 1983; Koenig, 1990; Witter, Stock, Okun, and Haring, 1985).
In a recent project that investigated the effects of transcendent experiences on people's lives, our initial data supported a model that may offer some insight into the relationship between religious commitment and well-being. Although our data were from a nonrandom sample from a selected population, the resulting model appears remarkably consistent with and integrates a wide body of data from other sources. The data and model support the hypothesis proposed by Zika and Chamberlain (1992b) that a sense of meaning in life mediates or intervenes between religious beliefs and well-being. A relatively low correlation between religious commitment and well-being follows naturally from this model. Our work also extends the analysis to include the role of transcendent experiences.
The purposes of this paper are (a) to describe a model of the relationships among well-being, religious commitment, meaning in life, and transcendent experiences, (b) to summarize data that pertain to this model, and (c) to discuss the implications of the model. We start with brief summaries of the concepts and limitations for each component of the model and for the statistical methods used to evaluate the model.
COMPONENTS OF THE MODEL
Subjective well-being is a global assessment of all aspects of a person's life and includes a cognitive-judgement component, life satisfaction, and two emotional components, positive affect and (absence of) negative affect (Diener, 1984). Well-being consists of stable dispositions or personality traits combined with short-term states resulting from transient events or environmental conditions (Chamberlain and Zika, 1992a; Diener, 1984; Pavot and Diener, 1993). Available data generally support the picture that good or bad events cause corresponding fluctuations in well-being that subsequently tend to return to a relatively stable baseline level, but major losses can cause long-term decreases in well-being (Braumeister, 1991:226-229; Chamberlain and Zika, 1992a; Diener, 1984; Lehman, et al., 1993).
The inadequate understanding of the relative roles of the trait and state aspects of well-being and the associated inability to identify factors that can cause long-term positive shifts in well-being are major gaps in well-being research. This situation results from the primarily correlational nature of the existing studies, which cannot disentangle factors that influence well-being from factors that are influenced by well-being. The stable trait aspects of well-being presumably are more likely to influence other factors, whereas the fluctuating state aspects of well-being are more likely to reflect influences by other factors.
Meaning and Purpose in Life
Certain psychologists believe that meaning in life is essential for psychological health in general (e.g., Maddi, 1967; Yalom, 1980) and various others propose that meaning in life protects against adverse health effects from stressful events (e.g., Antonovsky, 1987; Kobasa, 1979; Wortman, Silver, and Kessler, 1993). The term meaning in life indicates that a person is committed to a concept, framework, or set of values that (a) makes life understandable, (b) offers goals to attain, and (c) provides fulfillment (Battista & Almond, 1973). The most widely used meaning in life scale is the Purpose in Life test (Crumbaugh & Maholick, 1964).
The Purpose in Life test, like other meaning in life scales, has items on life satisfaction and depression and therefore can be expected to exaggerate relationships with well-being (Dufton and Perlman, 1986; Dyck, 1987; Yalom, 1980:456). This overlap with two components of well-being is a serious problem.
The inadequate understanding of the specific factors that provide meaning in life is a major gap in this area of research. Research to-date has focused on the degree or intensity of an overall sense of meaning in life, with little consideration that different sources of meaning may have different effects. The majority of people report that more than one factor gives their life meaning (Battista and Almond, 1973; Baumeister, 1991:5). Baumeister (1991) provides an interesting review and discussion of the potential sources of meaning in life. The relative roles of meanings or goals that are of a transcendent or mythical nature versus goals that are more tangible and short-term may be particularly important.
Self-Rated Importance of Religion
A self-rated importance of religion question measures the same construct as intrinsic religiosity scales and is a key dimension of religious involvement. These measures assess the degree that religion is the most important motivation or guiding principle in a person's life. The two items on the standard intrinsic religiosity scale that usually have the highest correlations with the full scale are: "My religious beliefs are what really lie behind my whole approach to life," and the negatively loaded "Although I believe in my religion, I feel there are many more important things in life" (Genia, 1993; Gorsuch and McPherson, 1989; Hoge, 1972). The obvious correspondence between intrinsic religiosity and self-rated importance of religion was verified in a factor analysis by Gorsuch and McFarland (1972) that identified the two measures as indicators of one factor.
Intrinsic religiosity is one of the most widely used measures in research on religion (Donahue, 1985). The concept and term originated with Allport (1960:264, 1966), and have evolved into a measure of religious commitment that is relatively independent of specific religious beliefs (Gorsuch, 1991). Standard scales have been developed (Hoge, 1972; Gorsuch and Venable, 1983). Self-rated importance of religion is typically measured with one item asking how important religion is to the person. Although religious involvement is widely presumed to have several dimensions, other proposed dimensions have not caught on the way that intrinsic religiosity has.
The relationships between intrinsic religiosity and the religion questions often asked in large social surveys are unclear. One common survey question is some form of "how religious do you consider yourself to be?" Although this question presumably correlates with questions like "how important is religion to you," the two questions have a different emphasis and we found no data comparing them. Questions about frequency of attending religious services are known to correlate with intrinsic religiosity, but also probably reflect other dimensions of religious involvement, as well as non-religious factors such as physical health (Levin and Markides, 1986) and social support (Ellison and George, 1988; Taylor and Chatters, 1988).
Mystical experiences are perceived as contact or union with a transcendent or ultimate divine reality and have several key characteristics (Spilka, Hood, and Gorsuch, 1985:176). These characteristics include: (a) a profound sense of unity, (b) a sense that the experience is noetic or a source of direct knowledge, (c) a sense that the experience is holy or spiritual, (d) ineffability or impossibility of describing the experience in words, and (e) presence of positive affect.
Mystical experiences can be viewed as part of a larger domain of transcendent experiences, which include similar experiences without the holy or religious connotation (Bourque, 1969; Spilka, Hood, and Gorsuch, 1985:Chapter 8). Transcendent experiences are one extreme on a continuum that includes experiences with varying numbers and intensities of the key characteristics (Spilka, Hood and Gorsuch, 1985:182; Thomas and Cooper, 1980). Also, mystical experiences are a category of the broader domain of religious experiences (Hardy, 1979; Margolis and Elifson, 1979; Spilka, Hood and Gorsuch, 1985).
Mystical experiences can influence a person's religious beliefs and a person's religious beliefs apparently can affect the occurrence of mystical experiences and/or the perception that transcendent experiences are religious experiences (Hay and Morisy, 1985; Spilka, Hood, and Gorsuch, 1985:Chapter 8). This is another situation when it is difficult to sort out the direction of causation.
National surveys consistently find that 30% to 40% of the American people report they have had one or more mystical or religious experiences. These percentages occur consistently in surveys using different questions (Back and Bourque, 1970; Davis and Smith, 1994:124; Gallup and Castelli, 1989:68-69; Greeley, 1975; Spilka, Hood, and Gorsuch, 1985:182-184).
However, 3% or less may be a more accurate estimate of the percentage of the population with full or "classic" mystical experiences. These survey questions appear to capture a much broader range of experiences than traditional mystical experiences. The question developed by Hardy (1979:18-19,125) was specifically intended to address a broad range of religious experiences . The question developed by Greeley (1975:43-57) was specifically intended to capture mystical experiences; however, Greeley (1975:77,79) found that although 35% of a national sample indicated one or more mystical experiences, only 3% of the sample described "authentic" mystical experiences that included at least three characteristics of classic mystical experiences. In very similar results with the same question, Thomas and Cooper (1978, 1980) found in two studies that 34% of the subjects in each study reported an experience, but only 2% and 1% reported "classical" mystical experiences. In a recent factor analysis of a multi-item spiritual experience scale, VandeCreek, Ayres, and Bassham (1995) found that this same question correlated only .16 with the spiritual experience factor.
Improved measurement methods and conceptual distinctions are needed for progress in understanding transcendent and religious experiences. Methods that measure various types and degrees of experiences and corresponding distinctions in terminology may be particularly valuable. Hood (1975) and Kass, et al. (1991) have developed mystical or spiritual experience scales that may provide a basis for further research.
Path analysis compares the relative strengths of the correlations among variables and can evaluate which variables appear to have some type of intervening or intermediate role between other variables.
The foundation for path analysis is the realization that if variable B has a causal or mediating role between variables A and C, then the correlation between A and C is equal to the correlation between A and B multiplied by the correlation between B and C (see Asher, 1983, for a readable introduction to path analysis). Note that the correlation between A and C will normally be substantially less than either the correlation between A and B or the correlation between B and C. This precise specification of the relationship among correlations will probably not hold if variable A affects variable C partially or entirely by mechanisms other than variable B.
The usual method to statistically evaluate the mediating role of variable B is a regression with variable C as the dependent variable and variables A and B as predictor variables. If mediation by B is the only mechanism for A to influence C, this regression should find that (a) the relationship between variables A and C becomes zero or nonsignificant when adjusted for variable B, and (b) the relationship between variables B and C is basically unaffected by adjustment for variable A. If variable A affects variable C by other mechanisms in addition to variable B, then the relationship between A and C will be reduced but not be zero when adjusted for B. The path coefficients between variables are the standardized regression coefficients adjusted for the other predictor or causal variables. Of course, this regression approach to path analysis requires that the data meet all the assumptions of ordinary regression analysis.
Path analyses, like other uses of ordinary regression, often fail to meet the assumption that the predictor variables are measured without error and also are susceptible to confounding by unmeasured causal variables (James, Mulaik and Brett, 1982). Both of these problems can introduce significant bias in the results.
Structural Equation Models
Structural equation models are an extension of path analysis that can handle measurement error and a wider range of relationships among variables. In essence, structural equation methods are a merging of regression analysis, path analysis, psychometrics, and factor analysis. All the various equations for path analysis and measurement error are solved simultaneously. The overall fit of a structural equation model is based on the degree that the correlations among the measured variables match the correlations predicted by the model. Various measures of goodness of fit have been developed. In addition to the overall fit, individual path coefficients can be tested to see if they are zero or some other specific value.
The results of structural equation models must be taken with caution at present because fundamental methodological issues have not yet been resolved. The requirement that data be normally distributed is much more important in structural equation methods than in ordinary regression. Hu, Bentler and Kano (1992:351) noted that the assumption of normality is usually violated in practice and found in simulation studies that "normal-theory tests worked well under some conditions but completely broke down under other conditions." Robust methods that do not assume normality have been proposed, but initial investigations suggest that the most widely recommended method (weighted least squares) may sometimes require sample sizes of 5,000 to be valid (Hu, Bentler and Kano, 1992). Other robust methods may perform better, but the current state of structural equation methodology does not provide practical, usable guidelines for the conditions under which the different methods can be used with confidence.
At present, structural equation models may be most useful for investigating the dominant features of the relationships in a model. Subtle distinctions and precise parameter estimates are usually tenuous at best.
The Problem of Causal Direction
Structural equation models and path analysis normally provide evidence consistent with a group of models and rarely provide evidence uniquely supporting just one causal model. Numerous models usually can be developed that fit the data equally well. In a review of 99 published applications of structural equations, MacCallum et al. (1993:190) found that the median number of alternative models that fit the data equally well was 12 using a methodology that counted only a portion of the alternative models.
In particular, structural equation models normally cannot determine or verify the direction of causation. In the earlier example that variable A affects variable B which affects C, identical statistical results will occur if the causal directions are reversed so that variable C influences B which then influences A, or if variable B is the cause of both variables A and C (Asher, 1983:21). In addition, the variables could be correlated without being causally related.
These uncertainties are compounded by the fact that reciprocal or bi-directional causation is rampant with human beings. We affect other people and our environment, and we are affected by other people and the environment. As noted in earlier sections, the factors being discussed in this paper are likely candidates for reciprocal causation. Although structural equation methods can be used to investigate reciprocal causation, these analyses require additional and usually questionable assumptions, and are relatively undeveloped (Kenney, 1979:109). In particular, the time periods between sequential causal events and the time lags for causal influences to propagate should be considered in the design and interpretation of structural equation models in general and particularly for the feedback loops with reciprocal causation models.
Structural equation models and path analysis can provide evidence that variable B appears to have some type of intermediate or intervening role between variables A and C, but the exact causal nature of that role usually must be determined by other evidence. This more modest conclusion is often a step forward in our knowledge. In general, randomized experiments provide the most compelling evidence of causation. However, a correlational model that makes useful predictions may be valuable even if the details of causation are uncertain.
As part of a study to investigate how transcendent and paranormal experiences affect people, we collected data on several psychological and health-related measures. The present report focuses on the findings for transcendent experiences and the relationships among variables that have precedents from previous research.
Well-being was measured with six items derived from the Medical Outcomes Study (Stewart, Ware, Sherbourne and Wells, 1992; Veit and Ware, 1983). The respondents indicated how much of the time during the past month they had each of three positive feelings and three negative feelings. The six response options ranged from "all of the time" to "none of the time."
Meaning in life was measured with one item asking "Have you found meaning and purpose for your life?" The four response options ranged from "very much" to "no."
Importance of religion was measured with one item imbedded in a list of ten items with the heading "To what extent do the following values and motivations give your life meaning and purpose?" The religion item was "observe spiritual or religious beliefs." The five response options ranged from "Not at all a purpose of life" to "Extremely important purpose of life." Although this question differs from the usual importance of religion question, it does indicate the importance of religion to the person.
Transcendent experiences were measured with one item that asked "Have you ever had a transcendent or spiritual experience (overwhelming feeling of peace and unity with the entire creation, or profound inner sense of Divine presence)?" The two response options were "yes" and "no."
Questionnaires without missing data on the key variables were obtained from a convenience sample of 182 people. The sample consisted of people who were interested in paranormal phenomena, including people who attended talks on parapsychology, contacted a parapsychology research center, or had ordered books or materials related to paranormal phenomena. The mean age of the respondents was 38 and ranged from 16 to 89. About 35% were under age 25 and 15% over age 60. Women were 70% of the sample.
MODEL DEVELOPMENT AND EVALUATION STRATEGY
The starting point for model development was Chamberlain and Zika's (1992b) hypothesis that meaning in life mediates the effects of importance of religion on well-being. We extended the model to evaluate whether importance of religion mediates the effects of transcendent experiences on meaning in life and on well-being. This sequence from transcendent experiences to importance of religion to meaning in life to well-being gives a causal chain model as shown in Figure 1.
Figure 1. Causal Chain Path Model.
This causal chain model can be evaluated by determining whether: (a) the structural equation goodness of fit measures indicate a good fit overall, (b) the magnitudes of the path coefficients between sequential variables in the chain are significantly different from zero, (c) the magnitudes of the path coefficients between variables that are not sequential in the chain are near zero, and (d) the magnitude of the correlations between variables that are not sequential in the chain match the correlations expected by multiplying the path coefficients for intervening variables. Although these analyses are largely redundant, they bring into focus different aspects of the model and facilitate comparison with other studies.
The chain model was evaluated with and without adjustment for measurement error. The path analysis that assumes no measurement error allows comparison with previous studies that have been analyzed with this assumption. The analysis with measurement error may be more consistent with the true path coefficients and with previous studies that used more reliable multi-item scales. For the measurement error analysis, the observed reliability of .85 was used for the multi-item well-being measure. Because reliability cannot be directly estimated for single-item measures, a range of reliabilities was examined.
We also evaluated the model using methods that do and do not assume the data are normally distributed. All variables were significantly non-normal. Based on the findings of Chou and Benter (1995), the maximum likelihood method was used as the method that assumes normality. The distribution-free method was weighted least squares. The analyses were done using PROC CALIS in SAS for Windows, Release 6.08.2
Initial analyses indicated that age, gender, and education did not affect the path coefficients of interest here. For simplicity and to avoid sample size reduction from additional missing data, these variables were not included in the analysis.