Introduction

Through multiple benefits, participation in daily social activities promotes physical and mental health, and ultimately, survival (House et al. 1982; Hendricks and Hendricks 1998; Welin et al. 1992; Glass et al. 1999; Lennartson and Silverstein 2001). Participation provides social contacts and thereby fulfills a phylogenetically determined need for affiliation (Cantor and Sanderson 1999; Reis et al. 2000). Activity theory (Lemon et al. 1972; Longino and Kart 1982) postulates that social activity is associated with life satisfaction because social activity provides opportunities for role support, which in turn reaffirms the self-concept. Researchers distinguished between informal social activity (with friends, relatives, and neighbors), formal social activity (e.g., participation in voluntary associations), and solitary activity. Activity theory holds that informal social activity has a stronger association with life satisfaction than does formal social or solitary activity, because informal social activity is more intimate and occurs more frequently, and consequently it is more rewarding and provides more specific role support. However, using informal social activity as predictor, researchers were not able to disentangle the two different explanations for the beneficial effect—role support may have been provided through social interactions (activity content) or through the mere presence of other individuals (social contexts).

In this paper, we use the term “social participation” to designate both time spent in social interaction as well as time spent in the presence of others. Both direct social interaction and the mere presence of others affirm an individual’s worth (Buunk and Schaufeli 2000), and increase the subjective meaning of whatever activity is performed (Thoits 1983; Rowe and Kahn 1998). Moreover, recent research has shown that social participation and specifically social interactions can alter basal risk profiles and attenuate acute stress reactivity (Seeman et al. 1994; Seeman and McEwen 1996). The influence on health appears to be exerted by promoting psychobiological recovery processes that play a central role in the onset of age-related illnesses such as cardiovascular diseases, type-II diabetes, and dementia (Sapolsky 1993; McEwen 1998). Because of the heterogeneity of measures and the neglect of intermediate concepts, progress in the accumulation of knowledge regarding the mediating processes has been slow (Herzog et al. 2002).

In order to be able to advance existing knowledge, it is important to map social activities into a theory-guided activity categorization and disaggregate heterogeneous categories (Herzog et al. 2002). Basing our work on Baltes’ two-component model (Baltes et al. 1999), we distinguished two broad categories of daily activities (Klumb and Baltes 1999a; Klumb and Maier 2002; see Fig. 1). On the one hand, there are regenerative activities that have to be carried out by physiological necessity (e.g., personal hygiene, eating, and resting). On the other hand, there are discretionary activities that one can do by choice on the basis of individual abilities and preferences. The third-party criterion (Reid 1934) served to further subdivide discretionary activities into productive and consumptive ones. If an activity is performed predominantly due to its outcomes and can, therefore, be delegated to a third party without losing its benefit (e.g., doing laundry, housecleaning, running errands), then it is productive. In contrast, if an activity is performed primarily for its own sake and cannot, therefore, be delegated to a third party without losing benefit (e.g., meeting friends, reading a novel, watching TV), then it is consumptive. We consider “social activity” to be a part of “consumptive activities”. Because heterogeneous activity categories do not easily lend themselves to testing causal pathways, we followed Herzog’s suggestion (Herzog et al. 2002) and further disaggregated “social activity” into its more basic components “face-to-face talks”, “visiting”, “phone conversations”, and “other social interaction”. Moreover, we distinguished four specific social contexts of performing an activity, namely, (1) being alone, (2) being with one’s spouse, (3) being with family members, and (4) being with friends.

Fig. 1
figure 1

A theory-based categorization of all activities during waking day

Full size image

The aim of the present study was twofold. First, we investigated if time spent on social activities and time spent in social contexts is associated with survival among older persons. Second, in an earlier study, we found that time spent on consumptive/leisure activities was related to survival among older adults (Klumb and Maier 2002). In the present study, we investigated the relative importance of social activities and social context for the effect of consumptive/leisure activities on survival. If the beneficial effect of performing leisure activities on mortality were driven through the effects of social activity, then social activities should be more strongly associated with a lower mortality risk than are non-social leisure activities. If, however, the driving force behind the effect of leisure activities were the mere presence of other people during the performance of any activity, then—independently of specific activity content—the social contexts “with spouse”, “with family”, and “with friends” should be associated with a decreased mortality risk when compared to the social context “alone”.

Methods

Participants and sampling

We used data from the first measurement occasion of the Berlin Aging Study (Baltes and Mayer 1999), which took place in the time period 1990–1993. The study was designed to be representative of the West-Berlin population aged 70+, while oversampling men and the very old. Samples originated from a random draw of addresses from the general registry (Landeseinwohneramt) of West Berlin. To obtain the final sample of 516 individuals stratified by age and sex, a much larger number of addresses had to be drawn. The study design consists of a hierarchical sequence of four levels of participation, with increasing numbers of variables but decreasing numbers of participants at each consecutive level: (1) the verified parent sample (n=1,908); (2) the short-contact sample (n=1,264); (3) the intake-assessment sample (n=928); and (4) the intensive-protocol sample (n=516) used in the present study, with 14 sessions of multidisciplinary assessment. Extensive selectivity analyses (Lindenberger et al. 1999) showed the intensive-protocol sample to be a somewhat positive selection of the parent sample. The magnitude of the selectivity effects was largest for general intelligence, but it did not exceed half a standard deviation for any of the analyzed domains of functioning. With one exception (dementia prevalence), selectivity effects did not interact with age or gender. Furthermore, comparisons of the sample with the Berlin census data showed no significant differences in indices such as marital status, proportion of institutionalized persons, and educational and income levels.

The intensive-protocol sample includes individuals ranging in age from 70 to 103. The sample was stratified for age and sex, resulting in 43 women and 43 men in each of six age/cohort groups: 70–74 years (born 1915–1922), 75–79 years (born 1910–1917), 80–84 years (born 1905–1913), 85–89 years (born 1900–1908), 90–94 years (born 1896–1902), and 95–103 years (born 1883–1897). Based on judgments made by one of us (P.K.), we excluded 31 participants from our analyses, due to implausible activity data. Most of these persons were diagnosed as suffering from dementia.

Mortality status information and the date of death for the deceased participants were obtained from the State Registry Office. Mortality information on 12 individuals could not be obtained because they had moved out of the Berlin area. These individuals were not considered in our analyses. This means that we utilized a total sample of 473 persons (230 women and 243 men) in our study.

Measures

Three types of measures were relevant: activity measures including measures of social activity, social context measures, and a set of covariates. As covariates, we rigorously chose common determinants of activity involvement and mortality (Rowe and Kahn 1998). The set of covariates included age, sex, years of education, measures of health and cognitive status, and an indicator of whether or not participants lived in an institution.

Activity measures

The “yesterday interview” (YI, Moss and Lawton 1982) was used to reconstruct the participants’ day preceding the interview, from waking up to falling asleep. With the YI we recorded the participants’ activities as well as the amount of time allocated to each activity. The YIs took place in the participants’ homes and lasted an average of about 50 min. In a separate study, we compared self-reports assessed with the YI to time samples of activities in daily life, and found acceptable agreement (Klumb and Baltes 1999b).

Interview data were first categorized into 44 activity codes. We quantified the levels of intercoder agreement with the kappa statistic (Cohen 1960). Kappas for all of the 44 activity codes were above 0.8, suggesting high levels of intercoder agreement. The 44 activity codes were then condensed into 13 activity domains. Activity domains were in turn assigned to three broad activity categories: regenerative, productive, and consumptive activities (Klumb and Baltes 1999a).

Regenerative activities serve to maintain one’s physical existence. This activity category comprised the activity domains “resting” and “self-maintenance”. The category “productive activities” resulted from collapsing the following five activity domains: “gardening”, “helping others” (including volunteer work and provision of care for relatives), “housework” (including maintenance of home and possessions), “paid work”, and “running errands”. The six remaining domains were aggregated into the category “consumptive activities”. Specifically, this category comprised the activity domains “active leisure” (such as attending adult education courses or performing sports), “locomotion” (such as walking, driving with own vehicle, or riding as a passenger), “health-related activities” (including visits to doctors), “reading”, “watching TV/listening to radio/records/tapes”, and “social activity”. “Social activity” in turn comprised the more fine-grained subcategories “face-to-face talks”, “visiting”, “phone conversations”, and “other social interaction” such as interaction with professional helpers (see Fig. 1). “Face-to-face talks” and “visiting” were coded as different categories because opportunities for the two kinds of activities differ. Specifically, “face-to-face talks” can occur in an individual’s own apartment, in his or her building, or outside the building. However, in order to be coded as “visiting”, a person has to leave his or her apartment, and walk or drive to that of somebody else.

We were interested in distinguishing the effects of social activity from those of other consumptive activities. Thus, we also examined the category “consumptive activities without social activities” (see “consumptive w/out social” in tables). This category comprised the activity domains “active leisure”, “locomotion”, “health-related activities”, “reading” and “watching TV/listening to radio/records/tapes”—but not “social activity”. For the purpose of the present analyses, activity measures were coded as either high or low, based on a median split (see Table 1). For all activity measures with a median of zero, this coding is equivalent to the dichotomy “does not do/does the activity”.

Table 1 Average time allocated to activity categories and time spent in social contexts (in minutes)
Full size table

Social context measures

In the “yesterday interview”, participants also reported the social context in which each activity took place. On the basis of the social partner’s name, and his or her relationship to the participant, we coded four social contexts of each activity: alone, with spouse, with family, and with friends. Specifically, we recorded the amount of time spent in each of these four contexts. For the purpose of the present analyses, social context measures were coded as either high or low, based on a median split. For all social context measures with a median of zero, this coding is again equivalent to the dichotomy “does not spend time/does spend time in this context”. Table 1 displays the average time allocated to activity categories and that spent in social contexts.

Education

We used the number of years spent in formal educational settings as an indicator of socioeconomic status. In addition to the number of years spent in elementary school and the different types of high school in Germany (graduation after 10 to 13 years of schooling), this variable also includes formal professional (e.g., apprenticeships) and academic (e.g., university) training. On average, participants in this sample had 10.8 years of education (SD=2.3).

Number of diagnoses

We selected the number of diagnosed moderate or severe illnesses as an externally assessed indicator of participants’ general health status. Diagnoses were determined in the course of consensus conferences of the research physician and psychiatrist, based on a standardized summary of clinical findings from all diagnostic procedures. Diagnosed moderate and severe illnesses were summed up to form the variable “number of diagnoses”. On average, participants in this sample had eight diagnoses of moderate or severe illnesses (SD=4).

Balance/gait

We selected a measure of sensorimotor functioning as an indicator of participants’ functional health. Sensorimotor functioning was represented by a unit-weighted composite of clinical assessments of balance and gait, the Romberg stance, and the “turn 360” tasks (Tinetti 1986). In the Romberg stance task, participants stood upright for about 1 min, with legs as close together as possible, arms extended in front of the body, palms turned up, and eyes closed. Performance was scored by a physician on a six-point scale ranging from “no swaying” to “not able to stand upright at all”. In the “turn 360” task, subjects were asked to perform a full turn around their body axis as fast as they could without risking a fall. The score corresponded to the number of steps needed to finish the circle. For the purpose of the present analyses, the balance/gait composite was represented as a z-score (mean=0, SD=1), with higher scores indicating higher levels of functional health.

Digit-letter test

We employed the digit-letter test, a measure of perceptual speed, as an indicator of cognitive functioning. On a large table visible throughout the whole procedure, each of nine different letters was assigned to a digit. Participants were presented with tables containing six digits, and their task was to name the corresponding letters. The score consisted of the number of correct answers given within 3 min. Stimulus presentation and data collection were supported by a Macintosh SE30 personal computer equipped with a Micro Touch Systems touch-sensitive screen. For the purpose of the present analyses, the digit-letter test was represented as a z-score (mean=0, SD=1), with higher scores reflecting higher levels of cognitive functioning.

Living in institution

We included an indicator reflecting whether a participant was living in the community or in an institution. This information was based on self-reporting by the participants, and it was verified by interviewers’ observation. In all, 409 persons (86%) lived in the community, and 64 (14%) in institutions.

Vital status

The vital status of participants in the Berlin Aging Study is monitored at regular intervals. By August 2003 (representing a 10–13 year period after baseline assessment), 368 individuals, or 78% of this sample, were registered in the state records as deceased, and 95 persons, or 20%, were registered as living. Ten persons, or 2% of this sample, were registered in the state records as alive in February 2000, but were subsequently lost due to follow-up. We included the exposure times of these ten individuals, and treated them as right-censored in the analyses. As is to be expected for a sample of this advanced age, a larger proportion of the oldest old had died (older than 85 years: n=223 decedents vs. n=7 survivors) than in the younger age group (70–84 years: n=145 decedents vs. n=88 survivors). As is also to be expected, a larger proportion of men had died (n=200 decedents vs. n=37 survivors) than was the case for women (n=168 decedents vs. n=58 survivors).

Statistical analyses

Cox proportional hazards regression models (Cox 1972) were evaluated for the effects of risk factors. We used the PHREG procedure (Allison 1995) from the SAS software package to estimate Cox regression models. We report relative risks and their 95% confidence intervals.

We proceeded in four stages to test the effects of broad activity types, social activities and social context on mortality risk. We first determined the zero-order relationships, and in a second step we evaluated mortality risks adjusted for the set of covariates (see Tables 2 and 3). A third set of analyses was designed to investigate whether the effects of activity categories diminished or increased with time. We calculated a time-dependent covariate for each of the activity categories, as the product of the activity category and time (see Allison 1995). We then calculated a Cox regression model that included the respective activity category, the associated time-dependent covariate, and the set of covariates. A fourth and final set of analyses was aimed at disentangling the effects of consumptive activities, social activity, and social context measures (Table 4).

Table 2 Mortality risk associated with daily activities (n=473)a
Full size table
Table 3 Mortality risk associated with social context (n=473)a
Full size table
Table 4 Association of consumptive activities, social activity and social context with mortality (n=473)a
Full size table

Results

Broad activity types and mortality risk

All three broad activity types were significantly associated with risk of death in the unadjusted analyses (Table 2). Higher levels of regenerative activities, and lower levels of productive and consumptive activities were associated with an increased mortality risk. The magnitude of the associations was considerably reduced when we controlled for potential confounds. Only consumptive activities continued to be significantly (p<0.05) associated with mortality risk after controlling for the activity×time interaction (see last column of Table 2). The risk of death was then reduced by 45% for individuals whose time spent on consumptive activities was above the median. The significant effect for the time-dependent covariate “consumptive×time” indicates that the effect of consumptive activities decreased with time since baseline assessment.

Social activity and mortality risk

The effects of social activity and its subcategories on mortality risk are shown in Table 2. In the unadjusted analyses, those with a higher level of social activity had a 20% lower risk of death. In the adjusted analyses, those with a high level of social activity still were estimated to have a 16% lower risk of death, although the effect did not reach statistical significance. The subcategories “face-to-face talks” and “phone conversations” were significantly associated with a lower risk of death in the unadjusted analyses, but “visiting” and “other social interaction” were not. None of these effects reached statistical significance in the adjusted analyses. Note, however, that the average amount of time allocated to these subcategories was relatively small (Table 1).

Social context and mortality risk

We distinguished the amount of time spent in four different contexts: alone, with spouse, with family, and with friends. The mortality risk associated with these contexts is shown in Table 3. From the unadjusted analyses it can be seen that a higher amount of time spent in social contexts (with spouse, with family, with friends) was related to a lower risk of death. From the adjusted analyses it appears that, with regard to survival, time spent with friends is more important than are the other social contexts. Specifically, those who spent time with friends had a mortality risk that was reduced by 28%. None of the time-dependent covariates (social context×time) reached statistical significance (data not shown). This suggests that the effects of social context on survival remained fairly stable since baseline assessment.

Disentangling consumptive activity, social activity, and social context

A final set of analyses was designed to disentangle the effects of consumptive activities, social activities, and social context measures. In a first step, we estimated the mortality risk associated with levels of consumptive activities without social activities (“consumptive w/out social” in Table 4). The mortality risk associated with this category (RR=0.58, cf. model 1 in Table 4) was very similar to the mortality risk associated with consumptive activity including social activity (RR=0.55, cf. model 3 in Table 2), indicating that the beneficial effect of consumptive activities is not mediated through social activity.

In a second step, we added “social activity” to the regression model (model 2 in Table 4). This did not alter the association between consumptive activities and survival, again suggesting that social activities contribute little to the beneficial effect of leisure activity on health and survival. In a third step, we added “social context: with friends” to the model (model 3 in Table 4), because this social context measure was found to be associated with survival in the previous analyses (Table 3). Interestingly, “consumptive activities without social activities” as well as “social context: with friends” were both significantly associated with survival. This suggests that time spent with friends affords a survival advantage above and beyond the beneficial effects of consumptive/leisure activities.

Discussion

In this study we investigated the relative importance of activity content and social context for the association between social participation and survival. It appears that social participation is related to survival. Individuals with higher levels of social activity and with more time spent in the presence of others had a lower mortality risk in the unadjusted analyses. In these analyses, measures of social participation carried variance associated with common predictors of differential social involvement and mortality risk. After controlling for covariates, several of the effects of social participation did not reach statistical significance, suggesting that the effects were not very strong. Nevertheless, based on an inspection of the relative risks, we argue that the association appears to be present even after controlling for confounding risk factors. Above and beyond the confounding influences, only the social context “with friends” was significantly associated with a reduced risk of death. The effect was fairly robust over time, as indicated by the absence of a statistically significant interaction with time. Because only little time was spent in each of the social activity categories, and our sample was relatively small, we could not draw firm conclusions with regard to the relative importance of the two mediating processes (i.e., social interaction versus mere presence of other people).

Interestingly, time spent on consumptive activities other than social activity, such as active leisure, locomotion and watching TV, was found to be associated with lower mortality. We speculate that at least two different mechanisms are involved. One the one hand, it appears that the cognitive stimulation induced by cognitively challenging activities has beneficial effects for intellectual functioning (Schooler and Mulatu 2001), and reduces the risk of dementia (Wilson et al. 2002). On the other hand, successful performance of chosen activities leads to the experience of competence, and increases personal control. Both factors contribute to psychological well-being, and alter the ways in which a person affectively and physiologically reacts to challenge (Mirowsky and Ross 1998).

Social activity, social context, and survival

Previous studies have reported beneficial effects of social activities. For instance, Steinbach (1992) and Menec (2003) found social activities such as visiting or talking to friends or relatives to be related to longevity. Nakanishi and colleagues (Nakanishi and Tatara 2000; Nakanishi et al. 2003) reported an increased mortality risk for individuals who did not participate in any social activities. Walter-Ginzburg et al. (2002) reported a lower risk only for measures of social engagement that explicitly involve others. All of these studies had larger sample sizes than ours, resulting in smaller confidence intervals for similar point estimates.

The pattern of results from our study is surprising because it lends support to the idea that the beneficial effects of social participation do not depend on social activities in the narrow sense, but can be achieved through the mere presence of other people. Interestingly, not all social contexts were equally conducive to acquiring this benefit. The finding that a positive effect was associated only with the context “with friends” is consistent with existing evidence. In contrast to family members, friends can be selected more freely by an individual. Spending time with them is rewarding in itself, and affirms the worth of the persons involved (Johnson and Barer 1997). Contacts within the family, in contrast, tend to be ambivalent in nature. Especially support among family members is liable to have a “dark side”, such as the obligation to reciprocate, devaluation through unwanted support, or loss of autonomy (Kruse and Wahl 1999; Pinquart and Sörensen 2000).

Our results pose new questions for future research. First, do some leisure activities have a higher likelihood of being carried out in the context of other people than others? Second, are there specific combinations of activity content and social context that are more beneficial than others, for instance, dancing, playing games vs. watching TV (Menec 2003)? This question could be investigated by systematically combining activity contents with contexts, and an examination of the effects of all the possible combinations. However, our sample was too small to do this. In addition, not all combinations are logically possible because, for instance, face-to-face talk cannot occur in the social context “alone”.

Strengths and limitations

A strong point of this study is that we employed a theory-guided activity categorization. Furthermore, we used a well-defined sample that was stratified by age and sex, and included a considerable number of very old persons. Assessment of activity involvement and social context based on the “yesterday interview” yielded reliable and valid information. As covariates, we rigorously chose common determinants of activity involvement and mortality from the data protocol of the Berlin Aging Study, in order to reduce the confounding effects of third variables. The chosen covariates were based not only on participants’ self-reporting but also on performance tests and physician-observed diagnoses of illnesses. This selection of covariates minimized confounding through common method variance.

In addition to the small sample size, at least two limitations should be kept in mind. First, we employed only data from a single day, and this day was not necessarily a typical one for all of the participants. It is thus likely that we underestimated the true size of the effects because measurement error in the activity categories may have attenuated these effects. Second, it is obvious that the reported effects are not necessarily causal, even though social participation preceded survival outcomes, and remained associated with mortality risk after controlling for potential confounds. A risk factor can be called causal only if its manipulation changes the outcome (Kraemer et al. 1997), but we did not manipulate social participation in this study. However, the effects of altered engagement in social activity can in principle be investigated, because time spent on social activities is amenable to intervention (Seeman 2000). We suggest that social contexts may contribute considerably to the maintenance of health and to longevity, because they exert their effects on a daily basis and these effects accumulate over the life course (Seeman et al. 2002).

Conclusions

Using time-budget data, we found that time spent in the social context “with friends” and, to a lesser degree, time spent on social activities was related to survival in persons aged 70 and older. This result supports psychological and sociological theorizing on the idea that activity participation and survival are linked through a psychosocial pathway, perhaps involving role support (Lemon et al. 1972; Longino and Kart 1982). The most adequate conception of the association between social activity and health may be a reciprocal one. On the one hand, social activity appears to be beneficial for health outcomes. On the other hand, it is obvious that good health in turn facilitates participation in social activity. Future research on social participation and survival may benefit from the examination of the interaction between specific types of activity and social contexts.