| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From Teacher’s College, Columbia University, NY (R.B.T.) and the Department of Epidemiology and Public Health (S.V.K., A.S.D.), Yale University School of Medicine, New Haven, CT.
Address correspondence to: Roni Beth Tower, PhD, 186 Indian Trail Road, New Milford, CT 06776. Email: RoniBTower{at}AOL.com
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
METHOD: Closeness is defined as 1) naming one’s spouse as a confidant or source of emotional support (vs. not naming) and 2) being named by spouse on at least one of the two dimensions (vs. not being named). The survival effects of both naming and being named are examined in Cox proportional hazard regressions, controlling for sociodemographic, health status, and behavioral variables.
RESULTS: Husbands who were named by their wives but did not name them were least likely to have died after 6 years. Compared with them, husbands in marriages with the other three styles of closeness were from 3.30 to 4.68 times more likely to be dead. Wives’ results showed the same pattern of effects, with the same marital style being most protective as for husbands, but the effects were weaker. However, wives’ results were strongly moderated by parenting status: those who had ever had children who were in the marital closeness pattern of wife naming husband but not being named by him were highly protected. Compared with these wives, others who had had children were from 8.26 to 10.95 times less likely to be alive after 6 years.
CONCLUSIONS: The same pattern of marital closeness most benefited husbands and those wives who had had children. These findings are not explained adequately by social support or marital role theory although they fit the latter more closely.
Key Words: Older couples • mortality • marital closeness • confidant • emotional support • gender differences
Abbreviations: CES-D = Center for Epidemiological Studies of Depression Scale;; RR = rate ratio;; CI = confidence interval; LR = likelihood ratio.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Research comparing those who are married with those who are not has shown marriage to be associated with longer life (4–9), although rare exceptions exist (10). These findings are stable across developed countries and time periods (6, 8, 10) and are particularly strong for men (5, 6). A few epidemiological investigations have found marriage to be a statistically significant protective factor for men only (11).
The benefits of marriage extend beyond its impact on longevity. Marriage is associated with better health (4, 12), higher well-being (4, 13), and lower depression (14, 15). Being married often has larger positive effects for husbands than for wives (12, 14, 16).
|
Variations within Marriage |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Ample evidence of variations in style in stable marriages exists (20, 21). Family theorists have suggested that the driving force in marital organization is "distance regulation," (22) the management of feelings of conjugal closeness. Clearly, "closeness" can mean different things to different people (23). Nonetheless, one would expect feelings of love to contribute to feelings of closeness and substantial empirical support for definitions of "love" indicates that passion, intimacy, and commitment comprise its major domains (24, 25). Self-disclosure or "feeling free to talk about anything," with its associated trust, are consistently considered a primary component of intimacy across studies in cognitive, social and clinical psychology (23–27), psychiatry (28, 29), family relations (30), epidemiology (31), and sociology (29). Indeed, some studies have used the presence of a confidant as the sole relationship predictor of outcomes such as the nomination of the spouse as a primary caregiver (31), good mental health and morale (29), or marital satisfaction (30). More globally, data documenting the physical and psychological benefits of self-disclosure for both men and women are accruing at a rapid pace (32, 33). Similarly, perceiving another to be a source of emotional support is a component of commitment, a second major aspect of love (23–25). Researchers have shown that emotional support also has powerful consequences. For example, in a prospective study of older adults who suffered myocardial infarction, perceiving no sources of emotional support (regardless of marital status) predicted increased risk of death both in the hospital and during 1-year follow-up (34). In a study of older Australians, perceived emotional support (which included the existence of a confidant) was significantly protective of survival in women (35). Feeling that one is trusted as a friend or relied on to provide caring and support are also components of love (23–25), although we have less empirical data on their consequences.
In our earlier studies of over 300 older spouse-pairs, we used a specific operational approach, defining marital closeness as naming the spouse as a confidant or as a source of emotional support and as being named by one’s spouse on at least one of the two variables (1, 2, 36–38). Although the meanings and distributions of the confidant and emotional support variables differ (2), the combined index provides a good approximation of closeness in older couples that has effectively predicted spouse levels of depressive symptoms (2). We have also found that husbands and wives who name each other are more similar to each other in levels of depressive symptoms (and both positive and negative affect) than are couples who do not (1), that these associations increase over time (36) and that, in these mutually close couples, a husband is much more distressed when his wife is cognitively impaired than in more distant couples (37). Wives in marriages in which neither spouse names the other have significantly more depressive symptoms than wives in other marital configurations (2). Their husbands are significantly less depressed than those in marriages in which spouses name each other (2), We also noted that the factors that contributed significantly to levels of depressive symptoms varied by closeness group.
In the current study, we examine marital closeness influences from two perspectives. The first focuses on an individual’s perception; that is, the potential implications of naming the spouse (or not) and being named by him or her (or not) are of primary interest. The second approach stresses the dyadic relationship by investigating the predictive power of four possible types of marriages: both naming, wife only naming, husband only naming, and neither naming the spouse spontaneously as a source of closeness.
|
Social Support Theory: Naming and Being Named |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
The person who is named by his or her spouse may receive information that he or she is important to the well-being of the person doing the naming. This information can be communicated in many ways, including through behaviors associated with intimacy, verbal confiding, reassurance, or interactions suggesting appreciation, reliance, or even dependency. Cobb (19) suggested that the resulting feelings of being valued contribute to well-being by activating the beneficial effects of self-esteem. Positive feelings about the self might contribute to longevity both directly, through positive affect generated when one feels competent in the spousal role, and indirectly, through a desire to remain alive in order to continue being useful. This latter dimension is reminiscent of the "meaning" construct in Antonovsky’s "sense of coherence," an experience described as salutogenic (43, 44). Findings on the benefits of volunteer activity in older adults indirectly support this notion (45), as does literature on the rewards of caregiving (46–48). Thus, we would expect positive effects from being named, although this argument has weaker empirical support.
Social support theory also suggests that there may be negative consequences or costs of social relationships (49–51). Particularly relevant for our purposes, "social support" is not always beneficial and, indeed, can have a particularly negative impact in a context of nonreciprocity. Social exchange theory (52) would claim that reciprocity is required for a positive interpersonal impact.
Therefore, from the perspective of social support theory, we hypothesize that 1) naming a spouse will have beneficial main effects for both husbands and wives; 2) being named by a spouse will have beneficial main effects for both husbands and wives; and 3) the effects will be synergistic, with reciprocal conditions being most beneficial.
|
Dyadic Perspective: Marital Role Theory |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Subjects
The data are from the Yale Health and Aging Project (YHAP) in New Haven, Connecticut, one of the four sites funded by the National Institute on Aging as part of its Established Populations for Epidemiologic Studies of the Elderly (EPESE) program (3). The full YHAP sample is a probability sample of 2811 noninstitutionalized men and women, aged 65 years and older living in the city of New Haven in 1982.
Samples were drawn from three housing strata reflecting the most common types of housing for those 65 and older: 1) public elderly housing, which is age- and income-restricted; 2) private elderly housing, which is age-restricted; and 3) private community houses and apartments. Three goals guided sample selection: 1) to obtain roughly equal proportions of men and women at baseline, 2) to maximize the number of respondents from the public age-segregated housing units, and 3) to oversample those in private age-restricted housing. The first goal recognized that simple random samples of elderly will yield many more women than men. The second reflected awareness that the needs of those in public housing would have a potentially disproportionate future impact on community health and resources, and thus accurate information on aging within that stratum could be particularly valuable. The third acknowledged that, although approximately one sixth of those over 65 in 1982 lived in age-segregated housing, the proportion was expected to increase in New Haven during the respondents’ lifetimes.
In public age-segregated housing, all individuals aged 65 and over were included in the sample, yielding 239 men and 486 women. In private age-segregated housing, where men were expected to comprise 24% of the residents, the sample consisted of all men over the age of 65, consisted of 1 in 2.5 women, and yielded 331 men and 536 women. In the community, the sampling frame was a utilities listing. All men who were enumerated were included in the sample, whereas 1 in 1.5 women were included, yielding 519 male and 619 female respondents. A detailed description of the stratified sample design for the full cohort is given elsewhere (55). The overall response rate was 82% and did not vary significantly by marital status or gender.
The spouse-pair sample of 317 married couples (634 individuals) emerged when both partners participated in YHAP. They had to have been selected through the above process, that is, when both spouses were chosen independently as a result of the sampling frame. Within these 317 couples, there were 305 couples in which both spouses provided data on the existence of a confidant and the availability of emotional support in their initial interviews. These 305 couples comprise the sample for this study of mortality. Approximately 85% of the husbands and wives were in their first marriage; the average length of marriage was 43.4 years.
Our study focuses on the prediction of survival over 6 years for the husbands and wives in our spouse-pairs. The analyses consider the period from the 1982 baseline interview to December 31, 1988, or to death if earlier. Within the 305 couples, 130 husbands (42.6%) and 71 wives (23.3%) died during this period.
Data Collection and Measures
Husbands and wives were interviewed individually in their homes by trained interviewers who used a 75-page structured interview. At the baseline interviews, the enumeration was done on the doorstep and interviewers were instructed to interview as many eligible people in the household as they could at the time of that first contact. If more than one potential respondent was available at the time of the initial contact, researchers who traveled in pairs (which was the case in dangerous neighborhoods or if the study coordinator had previously been informed that a married couple or siblings lived together) were instructed to interview husbands and wives simultaneously in separate rooms; those researchers who were alone were told to interview respondents sequentially. In 213 couples, both spouses were interviewed on the same day. In spite of the repeated emphasis given to interviewers during their training sessions to conduct separate interviews, in a substantial number of cases (63.9% for husbands and 58.4% for wives), the spouse was present in the home for at least part of the time of the approximately ninety minute interview. We have no data from which we can determine whether the spouse was in the same room and within hearing distance. Therefore, in post hoc analyses, we examined the possibility that a spouse’s presence during an interview may have affected our results. Measures of the control and predictor variables used in this study are described below; baseline information for our sample appears in Table 1.
|
Health Status
Chronic illness.
Respondents were asked whether a doctor had ever told them they had each of 10 chronic conditions: heart disease, stroke, cancer, diabetes, liver disease, hip fracture, other broken bones, hypertension, arthritis, and Parkinson’s disease. Responses of "yes" were summed to create the total number of chronic conditions experienced. The means of 1.3 for husbands and 1.7 for wives reflect the relatively good health of this community-dwelling sample.
Functional disability.
Disability was assessed by combining responses to questions about 1) impairment in activities of daily living: walking, bathing, grooming, dressing, eating, transferring, and using the toilet (57); 2) gross mobility: doing housework, climbing stairs, walking distances (58); and 3) physical activities: moving heavy objects, bending and kneeling, carrying heavy objects, reaching, fine motor control (59). A Guttman-like scale was developed from these three scales (55); scores ranged from 1 (no disability) to 5 (severe disability).
Cognitive impairment.
Interviews included the Short Portable Mental Status Questionnaire (60), with the question "What is the name of this place?" changed to "What is your address?" because respondents were interviewed in their homes rather than in institutions. Responses to the 10 items were grouped into a 3-point scale of no impairment (0 or 1 error), mild impairment (2 or 3 errors), and severe impairment (4 or more errors) (61).
Depressive symptoms
The Center for Epidemiologic Studies of Depression Scale (CES-D) was included in the interviews as a measure of depressive symptoms suitable for older community-dwelling adults (55, 62). Prorated scores were computed for respondents who completed at least 17 items but not all 20. Symptoms were scored for presence in the previous week from rarely or none of the time (0) to most or all of the time (3). The range of scores obtained by our sample was 0 to 41 of a possible 60; 23 (7.8%) of the husbands and 63 (21.0%) of the wives scored 16 or higher.
Indicators of Health Behaviors
Body mass index.
The variable that we used as a global proxy for nutritional status is the body mass index, computed by dividing weight in kilograms by height in meters squared. Height and weight were self-reported. Low body mass index has been associated with increased risk for mortality in older adults (63); the association between body mass index and mortality in our data are also linear.
Alcohol use.
Respondents were questioned concerning ingestion of alcoholic beverages in the previous year. They were then classified as those who reported any alcohol use and those who reported no alcohol use.
Smoking.
Respondents were asked if they currently smoked cigarettes and, if not, if they had ever smoked cigarettes. Responses were coded as current smoker, ex-smoker, and never smoked.
Self-Rated Health
Interviewers asked the global question "How would you rate your health at the present time?" Possible answers ranged from excellent (1) to bad (5). Ample literature documents that, in the presence of statistical controls for objective health status such as the variables described above, poor self-rated health is associated with increased mortality (64).
Predictor Variables: Marital Closeness
Marital closeness was measured by two variables: the husband spontaneously naming the wife on at least one of two questions and the wife spontaneously naming the husband on at least one of the same two questions. The questions were "Is there any one special person you know that you feel very close and intimate with [italics ours]—someone you share confidences and feelings with, someone you feel you can depend on? What is this person’s relationship to you?" and "Can you count on anyone to provide you with emotional support? (Talking over problems or helping you make a difficult decision)? In the last year, who has been most helpful in providing you with emotional support?" Respondents were identified as either naming the spouse at least once or not naming him or her at all. Among husbands, 156 (51.1%) named the wife at least once; 141 wives (46.2%) named the husband at least once. The slight excess of husbands naming the wife results from a slightly larger number of husbands naming their wives on the confidant variable than being named by them on that variable (2).
The above variable construction permits the category "did not name spouse" to be heterogeneous with respondents naming someone else or "no one." In actuality, few respondents (<10%) said "no one" to both questions. Consequently, contrasts in the analyses are overwhelmingly "named spouse" vs. "named someone else," rather than "named spouse" vs. "named no one;" that is, the contrast is primarily closeness in marriage vs. closeness outside, rather than closeness in marriage vs. social isolation. We return to this issue and find additional support for our measure in our post hoc analyses.
When the spouse-pairs are considered as dyadic units, 47 (15.4%) had the wife naming the husband but not being named; 62 (20.3%) had the reverse; in 102 couples (33.4%) neither spouse named the other, and in 94 couples (30.8%) both spouses named each other. Given the moderate association of naming across spouses, the reciprocal dyad types are roughly twice as common as the nonreciprocal ones.
Correlates of Closeness
To provide the reader with a better understanding of our closeness variables (and of the four possible types of marriage), we examined several classes of variables, available in our extensive dataset, as potential correlates of closeness. These variables can be grouped as follows: 1) sociodemographics and background, 2) reports of allocation of responsibilities and support with tasks within the marriage, 3) social relations with family, friends, and neighbors, 4) husband–wife similarity, and 5) (the dynamics of) symptoms of depression. Although this is a purely descriptive analysis, we are aware of the caution required when interpreting statistical significance in examination of such a large number of variables. However, the most noteworthy finding was that there are relatively few correlates of closeness, suggesting that our closeness indicators are not readily reducible to other dimensions.
Background variables.
We were surprised to find no significant differences among the four marital groups on many background variables: income level, education, work or retirement status; husband–wife similarity of religion, ethnic identity, birthplace; number of siblings or birth order of husband or wife; length of time at the current address or in New Haven; or length of marriage, childlessness, or number of children. Husbands who named their wives tended to be slightly (p < .07) more educated than those who did not.
Responsibilities within the marriage.
Husbands who named their wives were also more likely to name them as a source of task help. Their naming them was not associated with allocation of tasks within the marriage, but that of their wives was; wives who named their husbands were also highly likely to name them as a primary source of task support. They saw their husbands as more involved in handling household finances, tracking medical appointments, managing health matters, and cleaning the house than did wives who did not name their husbands. Their husbands, who were the "named," also saw themselves as more involved in all these tasks, except cleaning house, than did husbands who were not named. Correlations of husband–wife perceptions were comparable across marital types.
Social relations.
Naming and being named were not significantly associated for either husbands or wives with closeness to relatives, the number of close friends they had, giving or receiving help from neighbors, or participation in a religious congregation. Both spouses in the "Neither names" group reported the fewest neighbors known well and both spouses in the "He names only" marriages reported the most. "Both names" couples reported providing significantly more childcare to their children, and husbands in this type of marriage reported the highest level of gift-giving to their children. No group had an advantage in receiving help from children.
Similarity.
Husband–wife correlations within each marital type suggested some variations in spouse similarity. "Both names" couples had notably higher husband–wife correlations on a few attitudes (such as depth of religiosity, comfort and strength derived from religion; and desired amount of contact with their children) and social activities (such as going to movies, restaurants and events, and taking day trips and overnights) than the other three marital types, although the amounts of any of these variables did not vary among the groups. Husbands and wives in the "She names only" marriages had the least correlated perceptions of neighborhood and building safety, of their own health, and of the extent of their shared friends.
Dynamics of symptoms of depression.
As previously noted, levels of depressive symptoms were lowest in wives who were named by their husbands, but the husbands who were named were not particularly high in depressive symptoms. The factors that contributed significantly to the variance in depressive symptoms differed across the four marital groups. In "Both names" couples, one spouse’s level of depressive symptoms was highly predictive of that of the other. In "She names only" couples, levels of depressive symptoms in husbands were primarily influenced by their own sociodemographic characteristics, with a small contribution from their wife’s affective state. Wives were depressed by financial strain, her husband’s health, and his poor hearing. "She names only" couples had husbands who were depressed by their level of disability and wives who were minimally impacted by the state of their own health. "Neither names" spouses’ depression scores were related to their perceived financial strain (although the level was no greater than that of other couples); husbands were also depressed by their own poor health and cognitive impairment.
Mortality
Mortality was monitored through surveillance of obituaries in local papers, hospital records, and annual phone calls to respondents, or, if they had died, appropriate proxies. In all cases, copies of death certificates were obtained and dates of death were confirmed through them. Survival time was estimated as days from baseline interview to death; respondents surviving beyond December 31, 1988 were right-censored. The death of the spouse is accounted for by a time-dependent covariate (widowhood) in the survival analysis.
Analytic Strategy
Investigating marital-closeness influences on survival requires us to control for potential confounders, i.e., those variables that may influence mortality and may be also associated with closeness. These include sociodemographic variables of age, race (65), and financial strain, often seen as affecting survival through influences on access to health care and exposure to stressful living conditions (66). Health status variables were also controlled in our analyses: these include chronic illnesses (67), functional disability (63, 68–70), and cognitive impairment (68, 71, 72). These variables are not particularly correlated with closeness in this study; nonetheless, controlling for them minimizes potential confounding.
In addition, we control for potential confounders that have been identified in earlier research or seem particularly vulnerable to marital influences. Education has been associated with somewhat higher levels of closeness in husbands (2) and is also protective in mortality studies, especially in the United States (73). Health behaviors, including nutrition, alcohol consumption, and smoking, are associated with mortality (63) and may be subject to social influences that could vary with marital closeness (74).
Our models control also for depressive symptoms and self-rated health. The latter is a robust predictor of mortality (64), whereas the former is sometimes a predictor (75) and other times not (76). Because these two variables are associated with our marital closeness variable (1, 2, 36, 38), they could be pathways through which closeness affects mortality. Adjusting for them means we wish to document an effect of closeness that is independent of these possible mediators.
Finally, an analysis of the influence of marital closeness on survival must take into account the impact of a spouse’s death. We know that spousal bereavement can negatively influence the longevity of the surviving spouse (77), especially if the survivor is the husband (78, 79), although this gender difference is not found consistently (8, 80). Thus, we control for widowhood by using spouse death as a time-dependent covariate.
In looking at survival, first we computed the unadjusted percent of deaths by closeness group for husbands and wives. Next, we examined the effect of marital closeness on mortality using Cox proportional hazards regression models (81) with days since the baseline interview as the time scale. When analyzing time to event data, some researchers argue that age is a more appropriate time scale than time on study (82). Final models, reanalyzed with age as the time scale, yielded nearly identical rate ratios for the closeness variables. Thus the choice of time scale does not affect our marital closeness results, and these additional analyses will not be presented. All data are analyzed separately for husbands and wives.
To examine the assumptions underlying the Cox regression models, we visually examined plots of the log of the negative log of the estimated survival function against log time for parallelism across closeness strata over the period from the baseline interview to December 31, 1988. In addition, we included time dependent covariates (modeled by closeness variables * log time) in multivariate models to assess the significance of a general form of nonproportionality, and we were reassured that the rate was stable across time in a model including all our control variables. These procedures revealed that the underlying Cox assumption of constant rate ratios does not appear to be violated for marital closeness in the 6-year period.
First, we modeled the impact of marital closeness, measured by the two closeness variables, on 6-year mortality without adjusting for any covariates. Second, we assessed the independent impact of marital closeness by adding the previously mentioned "control" variables to the models. Cox regressions yielded adjusted rate ratios for the effects of her naming him and him naming her. Results appear in Table 2 for husbands and Table 4 for wives. All of the control variables were retained in the multivariate models, including those with a nonsignificant independent effect and/or those that did not confound the effect of marital closeness. To gain a better understanding of which covariates, if any, may have contributed to confounding the effect of closeness, we analyzed a sequential series of nested Cox regressions (data not shown in tables but results reported in appropriate places of the text). Third, to examine the extent to which the effect of him naming her may depend on whether she names him, and vice versa, we added the interaction of the two closeness variables to the previous multivariate model. Cox regressions yielded parameter estimates of the two closeness variables and their interaction. Our second set of analyses was designed to test the marital role theory. We reparameterized the last set of analyses, using the most protected marriage (she names him; he does not name her) as the reference group, against which the other configurations are compared. Results appear in Table 3 for husbands and Table 5 for wives. In addition to the above two sets of analyses, we performed post hoc exploratory analyses to examine specific questions raised by our data. The results of these additional analyses are reported in appropriate places of the text and discussed more fully at the end of the results section.
|
|
|
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
For wives, the crude mortality rates showed only small differences for the main effects: 24.8% of those who named their husbands were dead 6 years later compared with 22.0% of those who did not; 25.0% of those whose husbands named them were dead after 6 years compared with 21.5% of those whose husbands did not. Examining the interaction of the husband and wife responses, again a wife naming her husband but not being named was associated with the fewest deaths over 6 years (17.0%), compared with a husband naming her but her not naming him (19.4%), neither naming the other (23.5%), or both naming each other (28.7%). We note that the numbers of wives dying over the 6-year period was small (N = 71) compared with the number of husbands (N = 130). Thus, the survival analyses for wives have lower statistical power.
Survival Analyses
Husbands.
In a first step (see Table 2) using Cox proportional hazard regressions to predict a husband’s survival, unadjusted for covariates, both a husband naming his wife and being named by her were significant, with a wife naming her husband being protective (RR = .58; 95% CI = 0.40–0.83; p = .003) and with him naming her significantly increasing his risk of death (RR = 1.56; 95% CI = 1.09–2.24; p = .015).
In the second step, the control variables were added to the model. This model, which retained the two closeness variables, was significantly more explanatory than a model that included the covariates but not the closeness variables (LR test = 15.24; df = 2; p =.0005). Again, a wife naming her husband was significantly protective with a relative risk almost identical to the unadjusted rate ratio. For a husband naming his wife, the risk became stronger compared with the unadjusted relative risk (RR = 2.10 compared with RR = 1.56). Older age, being white, and more chronic conditions significantly increased mortality risk. Less education, more functional disability, and low body mass increased risk with marginal significance. Positive self-rated health was highly protective. Widowhood was not significant (p = .339). Nested sequential models, not shown, revealed that education had suppressed the magnitude of the negative effect of his naming her: husbands who were more educated were more likely to name their wives and also to live longer.
To allow for the possibility that poor wife health might be an additional confounder, as caregiver burden, operating independently of respondent health status variables, the model was then rerun (results not shown) with the inclusion of the wife’s four health status variables as additional controls. The results remained unchanged.
In a third step, the interaction term was added to the closeness variables. The interaction was significant (LR chi-square test of the significance of the interaction = 4.49; df = 1; p = .034). The presence of the interaction revealed that the main effect of being named by the wife is particularly protective if the husband does not name her (RR = .30), and less protective if he names her (RR = .82). At the same time, the main effect of the husband naming his wife particularly increased the risk if she also named him (RR = 3.82) and attenuated the risk if she did not name him (RR = 1.42). The rate ratios for the covariates remained highly similar between the models that did and did not include the closeness interaction term.
To examine the results from the perspective of marital role theory, we reparameterized the model, creating the four possible dyadic configurations. Adjusted rate ratio results are shown in Table 3. The marriage in which a wife named her husband but was not named in return was the reference group. Husbands in each of the other marital configurations were at substantially increased mortality risk compared with those in this most protected marriage. We also compared husbands in the combination of the other three marital closeness groups with those in the reference group marriages. The RR = 3.80 (95% CI = 1.86, 7.73; p < .001). The effects in the other three marital groups were not significantly different from each other (p = .491).
Wives.
In the first step, using Cox proportional hazard regressions to predict a wife’s survival, unadjusted for covariates, neither of the closeness variables was significant (see Table 4).
In the second step, shown in Table 4, the control variables were added to the model. The full model, although itself highly significant statistically, did not significantly improve the fit from a model that included the covariates but not the closeness variables. The effect of a wife naming her husband remained nonsignificant. The effect of a husband naming his wife increased somewhat, reflecting heightened risk, and attained marginal significance. Comparison to the husband’s model reveals that the specific effect of him naming her increased both his risk (RR = 2.10) and hers (RR = 1.75). Nested sequential models (not shown) revealed that baseline sociodemographic variables and self-rated health suppressed the effect of the husband naming his wife. Among the covariates significantly increasing the mortality risk were older age, more chronic conditions, more functional disability, and being a nondrinker and a current smoker. As expected, more education and a higher body mass index tended to be protective. The effect of self-rated health was marginally significant and was surprisingly in the opposite direction from the results on the husbands and from general expectations. In additional analyses (not shown), we added the husband’s four health status variables to the adjusted model in Table 4. The results remained unchanged, as had been the case for the husband mortality analysis when the wife’s health status was added.
In a third step, the interaction of husband and wife closeness responses was added to the model; the addition of the closeness interaction to the previous model did not change it significantly. Comparison of the wives’ closeness rate ratio to the husbands’ is, however, suggestive. The slightly protective main effect on the wife of her naming her husband became somewhat more protective when he did not name her (RR = .52) while the risk was slightly over 1 (RR = 1.26) when he did name her. This pattern echoes the findings for the husbands. Similarly, the higher risk associated with the main effect of a husband naming his wife increased under the condition when she also named him (RR = 3.01) and diminished when she did not (RR = 1.24). Thus even though the interaction term was not significant in the third model for wives, its inclusion reveals an important similarity to the closeness patterns seen for husbands.
In Table 5, the data for wives are reparameterized to reflect the four possible dyadic groups. The only group that shows heightened risk approaching statistical significance compared with wives in the "she names him–he does not name her" marriages is the mutually close group.
Post Hoc Analyses
To better understand our results and to examine their robustness, we conducted six sets of post hoc analyses. The first two have already been reported briefly.
Spouse health.
Controlling for spouse health variables in addition to respondent variables did not change the models in any substantial way for either husbands or wives. Thus, a potential burden presumed to accompany living with an ailing spouse did not modify the impact of marital closeness, presumably because such an effect would be mediated by respondents’ own health status and the models already adjusted for that.
Widowhood.
In addition to controlling for widowhood as a time-dependent covariate in our analyses, we reanalyzed the data both by 1) censoring a case entirely when a respondent became widowed and 2) censoring the closeness variables at the time of widowhood. Neither analysis changed results substantially for husbands or wives. The first strategy reduced the power of the models to detect effects: for husbands, there were 15 fewer events and 15 more cases censored because 15 of the 45 widowers among the 266 husbands died after being widowed. Fifteen of the 100 widows among the 269 wives died after being widowed, leaving only 39 events available for analysis. In the second strategy, testing the models with the effects of the closeness variables censored at the time of death of the spouse, the parameter estimates for the closeness variables did not change drastically for husbands or wives. However, for both husbands and wives, the effect of husband naming wife increased risk slightly less while the effect of wife naming husband decreased risk slightly more. We did not have sufficient statistical power to test for widowhood-by-closeness interactions.
Missing data.
To examine the influence of cases eliminated because of missing data, we restricted the unadjusted models to only those respondents with no missing data (266 husbands and 269 wives). For husbands, the protective influence of a wife naming him decreased less than 5%, and the risk of him naming her increased slightly. For wives, both risk ratios remained nonsignificant. We conclude that the subgroup that had missing data were not substantially different in the influences of the focal variable from that which did not.
Privacy during the interview.
Despite rigorous training of the interviewers, some respondents (195 husbands and 178 wives) were interviewed while their spouse was present in the home during some portion of the interview. We have no way of knowing if the spouse was actually in the same room and within hearing distance, but to reassure ourselves that the presence of a spouse did not affect our results, we correlated the presence or absence of spouse with whether or not the respondent named him or her. The associations were weak and nonsignificant. Next we reran our models, adding the presence or absence of spouse during the interview as an additional control variable. The changes in results were quite minimal.
Number and length of marriage.
We realized that there was a possibility that the protective factor of being married could depend on the number of years of marriage. The addition of years of marriage to the predictive models changed them only minimally and the variable was itself not at all a significant risk factor (RR = .99 for husbands and RR = 1.00 for wives). Interestingly, the risk ratios for age in the expanded models remained virtually unchanged despite the high levels of collinearity between years of marriage and age for the those 258 respondents (84.6%) in first marriages (for husbands the correlation between age and length of marriage was r = .60; for wives, r = .53). We also examined whether being in a first marriage influenced the results. Adding whether the respondent was in a first marriage to our models changed them negligibly; the variable did not itself present significant risk or protection (RR = 1.10, p = .75 for husbands; RR = 1.00; p = .99 for wives). We conclude that the closeness effects are independent of length of marriage and of having remarried.
Other social relationships.
Because of the impressive literature documenting positive influences of social networks on survival, we wanted to be certain that 1) our focal marital variables were not proxies for a more general social involvement and that 2) the presence of other social support did not substantially alter the effects of the marital relationship. We examined two variables that are most relevant to our concerns: number of close friends and the presence of children.
Close friends.
Respondents had estimated the number of close friends they had. We reanalyzed our data by adding this variable: for both husbands and wives, the risk ratios for the marital closeness variables changed less than 3%, reassuring us that the impact of the marital relationship exists independently of other social involvements. The number of close friends did not have an independent impact of its own on mortality.
Children.
We reasoned that the presence or absence of children might moderate the influences of marital closeness. Because the groups of husbands and wives who had never had any children were so small (42 husbands and 46 wives), our analysis concentrated on rerunning our models from Tables 3 and 5 on those who had ever had children (223 husbands and wives). The subgroups of those who had at least one child living at baseline (216 husbands and 218 wives) overlapped so much with the "ever had children" group that no meaningful separate analyses could be carried out. The results are shown in Table 6. For ease of comparison, the RR values from Tables 3 and 5 are repeated here.
|
For wives, however, separating those who had children from those who did not changed the results dramatically. Wives who had ever had children and who were in the reference group of marriages of wife naming husband and not being named by him were highly protected relative to wives in the other three marriages: the husband named his wife but was not named by her (RR = 10.95); both spouses named each other (RR = 8.99); and neither spouse named the other (RR = 8.26).
Given that the results for women were changed so strongly with the deletion of the group of childless women, we examined this group despite its small number. Wives who had never had children were significantly protected by husbands naming them (RR = .09; p = .025) and wives naming their husbands resulted in increased risk (RR =3 8.25; p = .017). The findings for wives with no children are particularly striking because the effects are significant in spite of the small number of cases; they suggest a powerful moderating effect of having had vs. not having had children on the effects of the closeness variables for the wives. We formally tested the moderating effect of parenthood in additional analyses compared with a multivariate model that included closeness variables and a parenthood variable (but no interaction); a model with the addition of the interactions of the two closeness variables with parenthood was statistically significant (difference in 
2 = 10.142; df = 2; p = .006). Because the results for husbands were not changed much, we did not expect the small group of childless husbands to be that different from the total. In fact, for these husbands the main effects were in the same direction as those shown in Table 2 but even stronger, although formal testing for the interaction of closeness and parenthood was not significant (p = .409). Thus, having vs. not having had children is a moderator for wives only.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Husbands who were named by their wife as confidant or source of emotional support were more likely to be alive 6 years later than those who were not named. Husbands who named their wife along the same dimensions were less likely to be alive than those who did not name her. Combinations of both naming and being named reveal interesting additional information: being named by her decreased his risk particularly if he did not name her, and naming her increased his risk particularly if she named him.
For wives, the adjusted main effects model showed no effect of her naming him and an increased risk (marginally significant) of him naming her. The addition of the interaction term showed a similar pattern as for the husbands (with him not naming her and her naming him most protective), but the interaction term was not significant. Suggestive findings were obtained for the subgroup of both naming each other, which was associated with heightened risk compared with the most protected marital group. We note that the wives’ analyses had less statistical power because of the smaller number of deaths during the 6-year period. However, stratifying the sample of wives into those who had had children and those who had not moderated these results even though the resulting smaller number of fatalities in each group reduced statistical power still further. Those wives who had ever had children were maximally protected by naming their husband and not being named by them. Those who had never had children had the opposite pattern, being protected if they did not name their husband but were named by him.
When the data are viewed dyadically, as summarized in Table 6, considerable similarity exists in which type of marriage is most protective for husbands and for those wives who had ever had children. This similarity, as well as the gender similarity when we discuss the main effects of "husband names wife" and "wife names husband," is striking. Gender differences emerge only when we think more generically of 1) the consequences of naming or not naming, and 2) the consequences of being or not being named. Note that we assume that the verbal behavior of naming or not naming a spouse is not itself the actual influence on survival but rather is a marker variable that captures sets of attitudes, perceptions, feelings, and behaviors that characterize different relationship dynamics.
Because of the strong moderating effect of ever vs. never having had children on the wives, the above summary of gender differences does not apply. Rather, wives who had not had children were protected if they were named and were at increased risk if they did the naming.
Theoretical Predictions
Social support theory predicted beneficial main effects for naming and for being named and an interaction effect in which reciprocity resulted in increased protection for both husbands and wives. A beneficial effect for naming was supported only for the wives who had had children and a beneficial effect for being named only for the husbands and for wives who had not had children. Other results were generally in the direction opposite that which was predicted: husbands who named their wives were at significantly increased risk; wives who had had children who were named by their husbands were also at increased risk; reciprocal naming was detrimental to both husbands and wives. The failures of social support theory to account for the results in part stemmed from the fact that the theory deals with generic naming and being named rather than with the gender-specific consequences of who does the naming and who is being named.
Marital role theory gathered greater empirical support. The normative or "ideal" marital configuration that had been prescribed for this cohort when they were young (53, 54) was beneficial to both husbands and wives, especially those who had had children, as consistent with expectation. The observed effects were quite strong. Marital role theory also predicted that the role-reversed relationships would be more detrimental to husbands and wives than other nontraditional configurations. This was not supported in that there were no significant differences among the three nontraditional configurations, but the risk ratios did tend to be the highest. Finally, women who had not had children were protected in the role-reversed relationships, where the husband named her and she did not name him, against prediction. This suggests that marital role theory may need to take into consideration the presence vs. absence of children.
Interpretive Possibilities
Because neither theory fully explains our data, we now briefly consider possible additional explanations for our findings.
Naming.
For a husband, naming his wife was associated negatively with his survival. First, this might represent a dependence on her that runs counter to his socialization into "male autonomy," common for this cohort (53, 54, 83) with the resulting detrimental conflict (18) or negative affect (84). However, husbands who named were not more depressed than those who did not. Second, a husband’s naming his wife could lead to a wife’s heightened sense of burden or responsibility and she, in turn, could initiate negative interactions with their unhealthy consequences (85–88). Third, a husband who names his wife may see her as a central resource. Thus the possibility of her eventual death, which may become more real as they age, is a source of stress with its possible negative effects (1, 89). We note that husbands who named their wives were notably low in "faith that [things] would turn out all right," offering some support for this speculation. Fourth, a husband who names his wife may have diminished alternative sources of psychological nurturance such as friends, family, or interests. However, this was not true in our data: these husbands did not report having a smaller number of close friends or close family members.
For wives, naming the husband decreased her risk if she had ever had a child and increased it if she had never had children, presumably for the same reasons that naming increased risk for men. (This moderation by parenthood is the greatest gender difference in our data and will be discussed briefly below.) For any wife, naming the husband was associated with lighter practical burdens, ie, she reported that the husband participated in more financial management, house work, tracking medical appointments, managing health care, and that she could turn to him for task support. Possibly, she experienced less stress in daily life than a wife who did not have this help. In addition, wives who named their husbands reported more happiness and less loneliness, perhaps enjoying better health as a result. Third, they were less likely to say that their life was "a failure" than were wives who did not name their husbands, suggesting higher levels of self-efficacy with its beneficial effects (38).
Being named.
Husbands benefited from being named by their wives. Presumably, they knew how important they were to her. The awareness of making a difference in the life of another could contribute to longevity through 1) a heightened "will to live" born from a sense of responsibility for another’s well-being, 2) increased positive feelings that result from feeling valued (19), and 3) a sense of mastery in a traditional role of care provider. A fourth possibility would be that wives who name their husbands may well treat them more kindly in other ways, especially through increased expressions of affection and appreciation, and the husbands may benefit from these more positive interactions.
For wives who had ever had children, risk of mortality increased when husbands named them. Wives who were named by their husbands reported greater hope and happiness and had fewer depressive symptoms than did those who were not, making it unlikely that negative affect or expectations are part of the explanation. They did report more involvement with their children’s home and with childcare, so perhaps the sheer burdens associated with presumably positive social ties took their toll through the stress associated with being relied on by loved ones (49–51). For those wives who had not had children, being named was associated with longer survival. The explanations may be the same as those for men who were named.
Marital dynamics.
One of our strongest findings is that the same marital dynamic is protective for husbands and for those wives who have ever had children. We note that the men in the protected marriage are probably not emotionally aloof. Indeed, the fact that they can be named suggests that their wives perceive them to be sensitive, caring, and emotionally available. Also, the simple consistency with cultural definitions of an ideal marriage for this cohort, as stated in marital role theory, may contribute to protection through lessened conflict (18). On the other hand, it is possible that rather than seeking to understand the protective dynamics in this group, we should examine the heightened risks to both men and women in the other marital styles: perhaps heightened affective contagion in "both names" couples, the tension from the role reversal in "he names her only" couples, and a relative lack of social involvement in "neither names" couples.
Moderation by parenthood.
Perhaps our greatest gender difference was that parenthood moderated findings for women only: not having had children amplifies only slightly the effects of naming and being named on husbands, while it reverses them for women. The results for women are particularly striking because the interaction was significant in spite of the small number in the never-parent group. We do not know how to interpret them because they may reflect the current situation of an older couple with (or without) children, or they may also reflect influences of earlier experiences of having raised a child (or having been childless) and their long-term consequences. We do know that wives in our sample who had and had not ever had a child did not vary significantly on self-perceptions of being "as good as others," in beliefs that their lives had been a failure, in reports of happiness or loneliness, or in beliefs about control over their lives, suggesting that differences in self-esteem and positive affect, which could potentially have health consequences, do not vary with parenthood.
Strengths and Limitations
Our data have several strengths. Mortality was carefully monitored with no cases lost because of missing mortality information. In addition, our heterogeneous community-based sample assures that results can be generalized to other cohorts of a similar sociodemographic composition. Important health status and health behavior variables were available for control of known contributors to mortality that might have confounded our results.
The data also have limitations. Our cohort consisted of married couples in which both spouses were 65 years or older in 1982. They were, therefore, born no later than 1917 and thus belong to generations who came of age with a defined set of beliefs about marriage, appropriate roles within it, and potential benefits to be derived from it. Conditions have changed during the twentieth century and we are not claiming that the impact of specific configurations of marital closeness will be the same for later generations.
In addition, because this report is based on secondary data analyses, one major limitation of our data is its inadequacy in explaining the mechanisms through which the qualities of a marital relationship contribute to survival. For example, we do not have additional data on the nature of marital interaction, such as conflicts and disagreements, and we do not know how naming a spouse during the interview relates to specific behaviors in the marriage. As is typical, secondary data analysis does a better job ruling out explanations when one controls for numerous variables. Thus we can rule out a respondent’s depression and self-rated health, or spousal health status (and, by implication, caregiver burden) as important mediating processes connecting closeness and mortality.
Another limitation of secondary data is that we are unable to link our available measures very closely to relevant theoretical constructs and thereby tie them tightly to appropriate theoretical formulations. We were able to sketch out possible theories that might apply to our findings, but a rigorous testing of these theories awaits future studies. Therefore, at this point we are left to ponder the mystery of the missing links. Like religiousness (90), self-rated health (64), and a sense of coherence (44), marital closeness of a particular type was linked to longer survival for both husbands and wives.
The first task of science is description. Our research suggests that marital dynamics are associated with longevity. Replication with additional samples; the expansion of the concept of closeness from verbal naming behavior to inclusion of physical, emotional, and behavioral correlates; and the investigation into specific mechanisms as they vary with different marital closeness configurations are obvious next steps. Ideally, we could then examine the feasibility, or even the possibility, of helping couples modify marital styles and the efficacy of interventions designed to lower the risk of those who are in high-risk marriages.
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Received for publication January 22, 2001.
Revision received July 23, 2001.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONVariations within MarriageSocial Support Theory: Naming...Dyadic Perspective: Marital Role...METHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
