This is the accepted version of the article: Lam, Jack; García Román, Joan. «Contour of the day : social patterning of time in later life and variation in reported well-being in activities». Journal of Aging and Health, (22 April 2021). DOI 10.1177/08982643211011800 This version is available at https://ddd.uab.cat/record/240018 under the terms of the license Author's postprint: LAM, Jack; GARCÍA-ROMÁN, Joan (2021) “Contour of the Day: Social Patterning of Time in Later Life and Variation in Reported Well-Being in Activities”. Journal of Aging and Health (First Published: April 22, 2021): 1-13 (ISSN: 0898-2643). DOI: https://doi.org/10.1177/08982643211011800 Contour of the day: Daily activity patterns and variation in reported wellbeing during activities among older Americans Objective To contextualize experiences of activities during the day and investigate whether the contour of the day is correlated with wellbeing during activities. Methods Drawing on American Time Use Surveys, we employ sequence and cluster analyses to create distinct typologies of daily life patterns, and bivariate analyses to describe whether wellbeing across activities varies by these typologies. Results We identified four typologies, characterized by different primary activity of the day: leisure (22.7%), TV (22.4%), housework (47.5%), and work (7.5%). Individuals in the Work cluster on average reported more positive wellbeing and individuals in the Housework and TV clusters reported more negative wellbeing in experiences of activities during the day. We also found that wellbeing experiences in the same activity also differed across the individuals in the different typologies. Conclusion Understanding the daily life patterns of older adults may be important given its correlation with wellbeing during activities. KEYWORDS: Activity levels; successful aging; health-related quality of life; time use 1 INTRODUCTION Later life is a period during which individuals may engage in a wide range of activities given the diversity of social roles and forms of engagement (Burr et al. 2007; Morrow-Howell et al. 2014). As evidence suggests, much of the structuring of time of working-age adults centers on the social institutions of work and family (Vagni and Cornwell 2018). As such, older adults may develop more heterogeneous time use patterns than younger or middle-aged adults as they exit the labor market and complete the child-rearing stage of the life course. Activity and engagement theories have therefore been predominant in gerontology (Havinghurst 1961; Johnson and Mutchler 2014; Lemon et al. 1972), emphasizing the importance of ongoing engagement. This approach has provided a broad overview of individuals’ activities, empirically demonstrating the diversity and complexity of social engagement among older adults (Chen et al. 2019), and their links with health and wellbeing. Existing knowledge on activity engagement later in life has typically been developed by drawing on broad measures of social roles and activities (Burr et al. 2007; Morrow-Howell et al. 2014). These studies have utilized measures in datasets such as the Americans Changing Lives (ACL) survey (Burr et al. 2007) or the Health and Retirement Study (HRS) (Chen et al. 2019; Morrow-Howell et al. 2014). Due to the nature of the measures, however, the operationalization of the activities has been necessarily de-contextualized. For instance, existing studies have drawn on indicators that captured how many hours individuals may engage in an activity per week or per month (Chen et al. 2019; Morrow-Howell et al. 2014), or in some cases, considering activity engagement through a dichotomous variable indicating whether a participant may engage in an activity at all (Burr et al. 2007). Arguably, these indicators are vague and give little sense of whether the context upon which these activities are performed matters. This is the case even as recent research has shown that the context upon which activities are performed indeed helps provide further explanations for understanding 2 variations in wellbeing, made possible in recent years by the availability of time use datasets (Gershuny 2011, 2019; Musick et al. 2016). For example, investigations of the enjoyment of activities have been made possible through reports of momentary wellbeing in activities (Kahneman et al. 2004), in which respondents indicate how happy, tired, sad, stressed, or in pain they felt while engaged in specific activities during the day in a time diary. Importantly, these findings show how the experience of the same activity may vary across individuals. In the current study, we draw on time diary information from the American Time Use Survey (ATUS), utilizing sequence and cluster analyses to identify and describe patterns of daily activities over the course of a single day for a large sample of older Americans aged 65 or older. This builds on and extends current knowledge on the diversity and multiplicity of social roles and forms of engagement in later life. Our paper makes three contributions to the existing literature. First, to our knowledge, we do not have information about the daily activities of older Americans. The only study that we are aware of that depicts the life of older adults uses time use data from South Africa (Grapsa and Posel 2016). Drawing on a rich, national survey, we therefore provide for first time an overall view of the different types of days older Americans might have. This fills an important knowledge gap, and extends prior literature on the activity profiles of older adults. This also begins to build an evidence base, and provides a benchmark for cross-national or historical comparisons, against older adults in other countries or perhaps future cohorts of older Americans, to observe how daily life might differ or change over time. Second, we contribute by highlighting how the experiences of activities varies across the types of days. This is an advancement of the current knowledge on how active engagement is related to wellbeing, showing how the contour of the day correlates with experiences of activities throughout the day. Furthermore, drawing on measures that captures various dimensions of wellbeing, we provide a more nuanced account of the implications of being active for older adults, to show that being active could also relate to different components of 3 wellbeing during activities (such as in meaningfulness, stress, etc.). Our third contribution, is by shifting the framing underpinned by the unit of analysis, away from types of individuals towards a focus on types of days. Previous research, focused on the individual level, necessarily typecast or reify individuals into types of individuals. Our focus on days however, focuses to the varying types of days individuals might have, and how this correlates with wellbeing. Not only is it a more realistic account of daily life, but also may have more usefulness for practitioners for informing and educating older adults of the associations of the different types of days, contributing to informing and enhancing older adults’ wellbeing. LITERATURE REVIEW Existing studies have shown that the enjoyment of activities varies across activities (Gershuny 2011, 2019). On average, housework and paid work tend to be the least pleasant, while leisure activities tend to be the most pleasurable. Other studies have sought to explain other potential sources of variation, such as the importance of social roles. In their study, Musick and colleagues (2016) observed that parents report better subjective wellbeing during activities with their children than without them, with some variation between mothers and fathers. People also may engage in activities differently, with implications for wellbeing. In their study, Yamashita and colleagues (2019) found that compared to passive leisure, physically active leisure was linked to higher levels of subjective wellbeing among older adults, including higher levels of reported happiness and meaningfulness during the activity and lower levels of sadness. Nevertheless, as described above, the majority of the current research that examines the activity engagement of older adults draws on broad measures of activity engagement (Burr et al. 2007; Morrow-Howell et al. 2014). Furthermore, individuals may also vary in their involvement in tasks and activities across days of the week, given that our daily life is 4 structured by social institutions. For instance, a full-time worker may nevertheless have days in which they are engaged in active leisure, while a retiree may similarly have days where they are predominantly engaged in housework or active versus passive leisure activities. With innovative data from time use diary information, the first contribution of this paper is to identify and describe the experiences of older adults at the daily level. The current study aims to identify and describe clusters of activity patterns among older adults, moving beyond aggregate time statistics and broad measures of social roles and activities. While individuals may follow different patterns of activities, there may also be variation across groups of individuals. Prior research has demonstrated that patterning of time is a reflection on household and social structures in later life (Chen et al. 2019). Existing research has shown variations in the experiences of activities and time engagement by class and gender (Lesnard 2008; Sullivan 1997; Sullivan and Katz-Gerro 2007). We therefore to explore associations between different activity profiles and individual characteristics. Further, we investigate variations in the experiences of activities across activity clusters, drawing on reported measures on momentary wellbeing across activities. We examine whether the structure and sequence of daily life may be associated with reports of wellbeing across the same activities. Given research around activity theory, we expect that daily profiles that depicts the respondents as being more active to also report more positive wellbeing. While current research drawing on activity theory proposes that individuals experience better wellbeing if they remain active (Menec 2003), and while the concept of ‘successful aging’ argues that older adults who remain actively engaged receive benefits through mental stimulation, a sense of routine and greater self-esteem and efficacy (Wahrendorf et al. 2008), our study further investigates whether another potential pathway may exist through nuanced contextual variation in the experiences of activities as related to the broader structure of the day. This extends prior research in the social sciences about the temporal regularity of daily 5 life and social time as a tool for analysis (Lewis and Weigert 1981; Zerubavel 1981) and demonstrates whether and how the structure and temporality of daily life are linked to individual wellbeing (Zisberg, Gur-Yaish and Shochat 2010). While the patterning of time has been of interest to scholars in recent years (Chen et al. 2019), the current study also aims to advance existing knowledge by highlighting the extent to which the degree of enjoyment of activities might vary given the structure of the day. The study thus underscores how daily-level experiences might be associated with emotional valence at the activity level. RESEARCH DESIGN To conduct our study, we draw on data from the ATUS. We use data from the ATUS Data Extract Builder (ATUS-X) from IPUMS-Time Use, which provides harmonized data over the survey period (Hofferth et al 2020). The ATUS is a nationally representative time diary survey in the United States. Households that have previously participated in the Current Population Survey (CPS) are selected to join the ATUS. The CPS is a monthly U.S. household survey conducted jointly by the U.S. Census Bureau and the Bureau of Labor Statistics (BLS) and whose main purpose is to collect information about the labor market and labor force participation. For each selected household, sociodemographic information of the household and its members is gathered, and one member of the household aged 15 or older completes a 24-hour diary of all activities. We select data for 2010, 2012 and 2013, which were the years when a wellbeing module was included in the survey. In addition to diary and sociodemographic information, the wellbeing module collects information on momentary wellbeing in 3 randomly selected activity episodes during the diary day. For each selected episode, individuals are asked to report their perceived pain, happiness, sadness, fatigue, stress and meaningfulness during the activity on a 6 scale from 0 to 6. As this study targets the older adult population, we select people aged 65 and over, which yields a sample size of 7,326 respondents. We apply sequence analysis techniques to the selected dataset in order to establish clusters of individuals with similar behaviors. These methods are appropriate to categorical data that consist of sequentially linked categorical states and our overall goal is to create a finite set of clusters made of homogenous sequences (Fasang and Liao, 2014; Vagni, 2020). Although there are other techniques such as Latent Class Analysis that can be applied for the same purpose, there is not a clear superiority of one method and sequence analysis has been widely used in time use studies (Barban and Billari, 2012; Lesnard, 2010; Minnen et al. 2016; Vagni and Cornwell, 2018; Vagni, 2020) We use TraMineR, which is an R package for mining and visualizing sequences of categorical data (Gabadinho et al 2011). To create the clusters, we identify and select the reported activity at minutes 0, 10, 20, 30, 40 and 50 of each hour 1. We only consider main activities reported by the respondents in their diary of activities. Though the dataset contains predefined activities, for simplicity and in order to estimate our models, we recoded the activities disseminated by ATUS-X in 11 groups of activities that are detailed in Appendix Table 1. For example, ‘Grooming’, ‘Health related activities,’ and ‘Personal care’ were predefined as distinct activities in the data, but we combined these together to create the category of ‘Personal Care’. BLS applied some data imputation methods to nonresponses and missing values in order to create a dataset with “complete” cases (BLS, 2020). The data we used do not contain any missing values so we did not apply any method to deal with missingness. In total, each individual contributes a sequence of 144 observations that are the inputs for the analysis. From these inputs, the package computes the optimal matching distances using substitution costs based on the transition rates observed in the data and an indel 1 We examine only activity reported at every ten-minute interval, as it is too computationally intensive for programming if we consider every minute of activity. 7 cost of 1 2. The output is a distance matrix between the sequences of each individual that is used to create the cluster in the following step. Using the cluster library in R, we build a Ward hierarchical cluster of the sequences from the optimal matching distances and retrieve the cluster membership from each individual’s sequence. Using the dendrogram of the cluster procedure (Appendix Figure 1), we select the most meaningful number of typologies of the trajectories. Three statistics suggested by Han et al. (2017:321) were used to determine the quality of clusters in terms of their size (see appendix 2). Each individual is then assigned to one of the clusters. According to each individual’s cluster assignment, we first explore time use by groups by considering the average time spent in each activity for each group and visually displaying the daily life patterns over a 24-hour period. The average time is computed as the mean of the total time spent for each individual in the cluster in the selected activities. Patterns are displayed by tempo-graphs that represent the proportion of individuals performing a selected activity for each moment of the day. Second, we report the sociodemographic characteristics of individuals in each cluster and test whether there are significant differences across the typologies. The analysis consists of the distribution of the individuals in each cluster according to their gender, living arrangement, race, age, level of education, employment status and self-reported health. Finally, we examine momentary wellbeing during the activities of individuals across each cluster to investigate potential differences, drawing on information collected from the wellbeing module. This enables us to consider how the structure of the day is related to the enjoyment of similar activities. In this case, for every activity, we compute averages of the different measures of wellbeing reported by each individual during 3 random episodes. 2 Optimal matching analysis transforms sequences into distances between individuals (Abbot, 1995; Lesnard, 2004). The algorithm evaluates the dissimilarity between two sequences. Dissimilarity is the cost required to make two sequences identical through the insertion, deletion and substitution of the elements of every sequence. The method requires a cost for each operation (indel parameter) that traditionally is one unit (Lesnard, 2004) 8 RESULTS Clustering daily time use patterns of the elderly population After applying the TraMineR package to create the distance matrix and the cluster algorithm, we choose to classify individuals into 4 different clusters according to the dendrogram (Appendix Figure 1). These 4 groups are clearly identified in the dendrogram, and the distances decrease considerably in the following step. In Table 1, we show the number of cases and percentages of each cluster. As shown, the largest cluster comprises slightly less than half of the respondents (47.5%), while the smallest cluster contains approximately 7.5% of the sample. TABLE 1 ABOUT HERE Figures 1 and 2 plot the daily life pattern of each cluster, and Table 2 shows the mean time spent in selected activities for individuals in each cluster. Figure 1 plots the percentage of respondents in each cluster performing a certain activity for every moment of the day. Figure 2 shows the sequence of activities for a random selection of 250 cases for every cluster. Figures illustrates that the daily patterns of each cluster are very different, not only in the activities performed but also in the rhythm of how they are performed, as shown through the timing of activities throughout the day. Taking into account the most characteristic activity in each cluster we label the clusters as Leisure, TV, Housework, and Work. The other activities are less relevant, probably because individuals generally spend less time in those activities. We note that care for others is, however, likely an underreported main activity, which is one of the limitations of the time use surveys (Duran and Rogero, 2010), because it can be reported as a secondary activity and our analyses is based on the primary activity. We mention this as a limitation in the Discussion. 9 We conduct ANOVA tests to determine whether there are significant differences in the average time spent in the activities between clusters, though the results show that the differences are significant for all activities. Thus, we apply Scheffe’s method (Salkind, 2010) as a post hoc test to determine which pairs of means are significantly different. The lines denote for which clusters differences are significant at p-value=0.05. FIGURE 1 ABOUT HERE FIGURE 2 ABOUT HERE TABLE 2 ABOUT HERE We find that the main defining time use characteristic of individuals in the first cluster is time engaged in non-TV leisure activities (light red). Respondents in this cluster spend an average of six hours and 36 minutes in this type of activity. Such activities might include socializing, relaxing, engaging in hobbies, practicing or attending sports, participating in religious activities, volunteering and studying. These activities are more common during the middle of the day, with shorts breaks for eating. According to Figure 1, more than half of the individuals in this group perform leisure activities between 8 am and 8 pm. They also spend a considerable amount of time in housework, with slightly more time spent in the morning, and watching TV between 8 pm and 11 pm. Respondents in the second cluster are characterized by the large amount of time they spend watching TV (red), i.e., an average of approximately 8 hours and 40 minutes per day. Watching TV is especially common in the evening, as almost 75% of individuals in this group watch TV at approximately 8 pm. In the other clusters, the proportion of individuals watching TV at 8 pm is slightly less than 50%. Other activities that are clearly identified in cluster 2 are 10 non-TV leisure and housework, to which they dedicate a total average of almost 3 and a half hours. Regarding cluster 3, it is the largest group in our sample (47.5%). We find that the defining characteristic of this group is time spent in housework (yellow). Individuals in this group spend an average of 4 hours and 35 minutes doing housework. Housework is more prevalent between 9 am and 4 pm, when almost 40% of the individuals engage in this activity, with a peak of approximately 50% at around 10 am. In the evening, this group changes their activities to watching TV, which is the most common activity at 9 pm. This group spends more time sleeping (540 minutes), in general personal care (58 minutes) and eating (87 minutes) than the other groups. They also spend more time in childcare than the other groups, although the total time in this activity is low (13 minutes). Finally, regarding the fourth and last cluster, the main defining characteristic is the time the respondents spend in paid work (violet), i.e., an average of 7 hours and 32 minutes over the observed day. As a point of comparison, the other groups spend practically 0 minutes per day in paid work. As shown in Figure 2, around 9 am, approximately 75% of individuals in this cluster are performing some paid work activity. This proportion decreases after 4 pm, when a larger proportion of individuals in this group are engaged in housework or watching television, which becomes the most popular activity by 9 pm. This group spends the least amount of time sleeping and engaging in leisure activities, but these respondents also travel more than those in the other clusters. The differences in these activities between this cluster and the other clusters are all significant. Through the sequence analysis and cluster algorithm, we have created clusters to define typologies that describe different patterns of daily life. In the next section, we examine and report the sociodemographic characteristics of each cluster. In contrast to the activities, these 11 characteristics are not used to create the cluster; rather, the bivariate analysis will allow us to explore and describe differences in activities across sociodemographic characteristics. Sociodemographic characteristics of the clusters Table 3 presents the main sociodemographic characteristics of the individuals in each cluster. For all characteristics, we additionally perform chi2 tests to examine whether their distributions are independent of the cluster, and in all cases, they are not. This means that there is a significant difference between the expected frequencies when the cases are independently distributed among the clusters and the observed frequencies. Characteristics are not equal in all clusters, and some characteristics of individuals are overrepresented in certain clusters. TABLE 3 ABOUT HERE Gender is one of the most differential characteristics between clusters. Women are more prevalent in the Housework cluster, in which they represent approximately two out of three individuals, or 63.9%. As a point of comparison, in the Work cluster (characterized by a large amount of time in paid work), women constitute 40.6%. In general, the Leisure and TV clusters show similar characteristics, and regarding the gender distribution, women represent 52.1% and 48.3%, respectively. For the rest of the characteristics analyzed, Housework cluster has a distribution very close to the overall population. Turning to the Work cluster, individuals in this have characteristics that are most different from those of the sample averages. The respondents in this cluster are more likely to be men (15.5 percentage points higher than the sample average), live only with a partner (10.7 points percentage higher), have ‘Some College’ (13.1 points percentage higher) and have better health (14.6 points percentage higher). They are also younger, with 81.7% of the members in 12 the group being aged 65-74, as compared with 56.2% of the full sample belonging to this age group. Individuals in the TV cluster are less likely to have ‘Some College’ (10.4 percentage points lower than the sample average), be white (4.8 percentage points lower) and have good health (9.9 percentage points lower) than the sample average. Regarding living conditions, this cluster has a lower proportion of those living only with a partner. In general, individuals from the Leisure cluster (those who spend more time in non-TV leisure activities) are very similar to the overall sample. Wellbeing during activities across clusters In this section, we describe the averaged reported wellbeing of individuals in each cluster across their activities. Table 4 presents the means of each measure for which the respondents indicate how they feel during three randomly selected activities. As in table 2, the lines shows for which clusters the differences are significant at p-value=0.05. TABLE 4 ABOUT HERE In general, we observed significant differences between clusters in all measures of reported wellbeing during activities. Leisure was the cluster in which individuals reported more positive measures (mainly higher levels of happiness and meaningfulness in activities) while the TV cluster was the one with more negative measures of wellbeing in activities (mainly higher reported levels of pain and sadness in activities). Looking at the differences in greater details, respondents from the Leisure cluster were significantly happier than those in the other 13 groups and reported lower levels of pain, sadness and stress than the TV and Housework clusters during their activities. Individuals in the TV cluster had higher scores for sadness and pain and the lowest scores for meaning and happiness during their activities. The most significant differences are with clusters leisure and work that are significant in almost all measures while differences with cluster housework are only significant in terms of sadness, fatigue and meaning of the activities. Those in the Housework cluster had higher levels of fatigue than those in the Leisure and TV clusters, but not compared with individuals in the Work cluster. The Housework cluster was in a middling position in terms of sadness, with significantly lower levels than the TV cluster and significantly higher levels than the Leisure and Work clusters. They are also more stressed than individuals in the Leisure cluster, but less stressed than those in the Work cluster. In terms of pain, they reported higher levels than both the Leisure and Work clusters. Wellbeing measures for individuals in the Work cluster were led by the fact that they were more likely to be employed. They reported higher levels of fatigue and stress than individuals in the other clusters. On the other side, they reported lower levels of sadness than those in the clusters with more negative indicators (TV and Housework). They also report significantly higher scores in terms of meaning of their activities than the three other clusters. In Table 5, we also present the measures of wellbeing for activities with at least 30 observations in each cluster. This provides a more nuanced understanding of whether reported wellbeing in the same activity also varies across clusters. TABLE 5 ABOUT HERE We observed that TV, travel, housework and eating were the activities with the most significant differences across the clusters, and the significant differences were often observed 14 for the measures of negative feelings during the activities (pain, sadness, fatigue, and stress) while less significant differences are observed for happiness and meaning. The results were in the same direction of what we observed for the overall activities. Individuals in the TV cluster had the highest scores for negative feelings compared to the individuals in the other clusters. For instance, when engaging in the activity ‘eating,’ respondents in this cluster reported higher levels of sadness and lower levels of meaningfulness, as compared with individuals in at least one other cluster. They also reported lower levels of meaningfulness in their leisure activities, and higher levels of pain when engaged in housework. Perhaps due to the nature of the day and the activities they engaged in during the other parts of the day, individuals in this cluster reported lower levels of fatigue when engaged in active leisure activities as compared with individuals in the other clusters; they also reported lower levels of fatigue when engaged in television watching, travelling and housework. On the other side, individuals from the Leisure cluster reported lower levels in these negative indicators. They reported lower levels of sadness and fatigue when engaged in housework, as compared with individuals in the other clusters who were engaged in the same activity. Regarding travelling they reported higher levels of happiness and lower levels of stress when they travel, which can be related to the fact that they commute to participate in more appreciated activities. The differences during leisure activities show are only significant in terms of fatigue being in a middle point between the individuals in the cluster work (who reported higher scores) and individuals in cluster TV (who reported the lower scores). If we looked at the cluster Work, we observed significant differences when they are travelling in terms of fatigue and stress, that might be explained because they are in the labor market and have to commute more. 15 Overall, these findings suggest that individuals do not experience the same activity in the same way. Rather, the structure of the day contextualizes their experience, enjoyment, and reported wellbeing during the activity. DISCUSSION With time use diaries, researchers have collected detailed information on the daily lives of individuals, including data on all activities engaged over a 24-hour period. However, existing studies have typically focused their efforts on detailing the lives of working-age adults (Vagni and Cornwell 2018). In this study, we extend this research to explore and describe meaningful clusters of daily activities among older adults aged 65 and older. We aimed to extend the research on the activity engagement of older adults (Chen et al. 2019). Although a stream of research has produced knowledge on the activity profiles of older adults (Burr et al. 2007), this research has been somewhat limited, as it has drawn on crude measures of activity engagement. Utilizing sequence and cluster analyses of daily activities drawing on time use diaries, we identified four distinct clusters of activities among a sample of older adults from a national survey that characterize four different types of days. To distinguish the identified clusters, we also visually show both the sequence and duration of different activities throughout the day, suggesting important differences in the time use of the older adult population. We go beyond summary statistics of the time engaged in select activities to highlight the rhythm and temporality of daily activities across clusters. In addition, we show that these groups are heterogeneous, in terms of not only their time use but also their sociodemographic characteristics. Finally, regarding the extent to which individuals across clusters vary in their reported wellbeing during their activities, we find that individuals in certain clusters report lower wellbeing based on their reported experiences of activities 16 during the day. Furthermore, we show that even in experiences of the same activity, some variations in experiences emerge and could be linked to the broader contour of the day. Although our study has advanced the research in this area by identifying and describing salient activity clusters in later life, it is not without limitations. First, our analysis focused only on individuals’ main activities. In other words, individuals might engage in multiple activities simultaneously. For example, caring activities may be reported in the ATUS as a secondary activity, and some older adults may care for family members as a secondary activity. Nevertheless, due to our analytic approach, we were restricted to only examining the primary activity. Further, due to our interests in examining correlates of wellbeing as linked to the activity clusters, we limited our analysis to three years of available data, restricting the size of our analytic sample. Future research using a larger sample and more years of available data may be able to explicate nuances by also considering respondents’ secondary activities. A second limitation of our study is our use of cross-sectional data, which prevented us from disentangling how health may similarly be an antecedent of activity engagement. Future research using longitudinal data may be better positioned to adjust for selection into activity engagement while considering its wellbeing effects. Despite these limitations, our findings advance the existing knowledge by providing detailed illustrations of the rhythm and patterns of older adults’ daily lives. The study moves beyond single, broad measures of activities to delineate activity engagement over a 24-hour period. By reporting sociodemographic characteristics that correlate with different activity clusters, the study also alerts practitioners and family members to potentially vulnerable groups of older adults and the importance of detecting older adults’ potentially more negative days based on the activities they are performing, which may shape their experiences of other activities. 17 CONCLUSION Activity and engagement theories have been advanced in the gerontology literature to underscore the importance of remaining active and engaged in later life (Havighurst 1961; Johnson and Mutchler 2014). In line with this, a number of empirical studies have begun to articulate the diversity of activity engagement (Burr et al. 2007; Morrow-Howell et al. 2014). These studies, drawing upon available data, have shown the different activity profiles of older adults, and importantly, established links between activity profiles and wellbeing outcomes (Chen et al. 2019). Nevertheless, due to the nature of the data drawn upon, existing quantitative studies around the activity engagement of older adults have been necessarily de-contextualized. Drawing on a rich time-use survey, with information on activities over a 24-hour period, we further describe and elaborate on the daily lives of older Americans. Here, we move towards contextualizing the experiences of activities, within the context of a single day, and also empirically test whether this correlates with their wellbeing during activities. Specifically, we showed that in the aggregate, experienced wellbeing varies across different activity clusters, replicating prior research linking activity profiles and wellbeing outcomes (Chen et al. 2019, Freedman et al. 2019). In addition, we also showed that the experience of wellbeing in the same activity also differs across individuals in different activity clusters. This is an advancement of current knowledge, as it suggests that the context of the day also matters for how older adults may experience their activities. With some caveats, we showed that respondents in the TV cluster report more negative wellbeing across activities, though they also report less fatigue. In contrast, respondents in the Work cluster report more fatigue across some activities, though they also found activities more meaningful. Our findings contribute and extend research on activity theory, moving beyond binary distinction between uniformly positive or negative wellbeing associations, highlighting nuances in the engagement of activities for older adults, such as in energy depletion. 18 While our analyses described bivariate associations, they also point to important future directions for research. They highlight that while activities matter, the context in which they are performed may also be salient and worthy of future investigations. Understanding the context in which activities are performed advances activity and engagement theories. In this study, we empirically show that the context of the day is related to the experiences of activities, as well as pointing to the fact that activities are multi-faceted. This highlights that future research may wish to investigate other characteristics of activities that may similarly shape the experiences of activities, such as with whom activities are engaged (Lam and Garcia 2020), where the activities are performed, and how the activity may correspond to the individual’s social role. Future research that aggregates across more waves of data might also be able to tease out whether there might be important differences by types of passive and leisure activities, as well as volunteering or religious activities. Understanding whether and how various components of activities might matter for older adults’ wellbeing could open up opportunities to consider how to move different levers to improve the experience of activities. This would advance current knowledge, which heavily focuses on the intensity of activity engagement and whether an activity is engaged in or not, to include other factors that would also be of salience. Our research findings also have implications for research, as well as programs and practice. While prior research may reify individuals as belonging to certain activity clusters, our study moves the focus from individuals to types of days. Potentially, this could alert researchers and practitioners to variations in the type of day older adults may have, as well as how this could shape their wellbeing. For example, while an employed older adult may spend much of their time engaged in activities related to work, there may also be days when they are predominantly engaged in active leisure or television watching. Shifting the focus of analysis from individuals to days could alert researchers and practitioners to how structures of the day is related to experiences of activities, 19 with implications for wellbeing. Given advances in available data and statistical modeling, new conceptualizations of activities and their implications could move beyond the ways it has traditionally been operationalized and towards providing richer contexts. 20 REFERENCES Abbott, A. (1995). Sequence analysis: new methods for old ideas. Annual Review of Sociology, 21, 93-113. https://doi.org/10.1146/annurev.so.21.080195.000521 Barban, N., & Billari, F. C. (2012). Classifying life course trajectories: a comparison of latent class and sequence analysis. Journal of the Royal Statistical Society: Series C (Applied Statistics), 61(5), 765-784. Bureau of Labour Statistics (2020). American Time Use Survey User’s Guide. https://www.bls.gov/tus/atususersguide.pdf Burr, J. A., Mutchler, J. E., & Caro, F. (2007). Productive activity clusters among middle-aged and older adults: intersecting forms and time commitments. Journal of Gerontology: Social Sciences, 62, S267-S275. doi:10.1093/geronb/62.4.s267 Chen, Y.-C., Putnam, M., Lee, Y. S., Morrow-Howell, N. (2019). Activity patterns and health outcomes in later life: The role of nature of engagement. The Gerontologist, 59(4), 698- 708. DOI: 10.1093/geront/gny023 Cornwell, Benjamin, Jonathan Gershuny, and Oriel Sullivan. (2019). The Social Structure of Time: Emerging Trends and New Directions. Annual Review of Sociology. https://doi.org/10.1146/annurev-soc-073018-022416 Cornwell, Benjamin. (2011). Age Trends in Daily Social Contact Patterns. Research on Aging, 33, 598-631. https://doi.org/10.1177/0164027511409442 Durán, M. Á.& Rogero-García, J.. (2009). La investigación sobre el uso del tiempo. Cuadernos Metodológicos, Vol. 44. Madrid: CIS, 214. Fasang, A. E., & Liao, T. F. (2014). Visualizing sequences in the social sciences: Relative frequency sequence plots. Sociological Methods & Research, 43(4), 643-676. 21 Flood, S., Hill, R., & Genadek, K. (2018). Daily temporal pathways: A latent class approach to time diary data. Social Indicators Research, 135, 117-142. DOI: 10.1007/s11205- 016-1469-0 Freedman, V., Cornman, J., Carr, D., & Lucas, R. (2019). Time use and experienced wellbeing of older caregivers: A sequence analysis. The Gerontologist. https://doi.org/10.1093/geront/gny175 Gabadinho, A., Ritschard, G., Mueller, N. S., & Studer, M. (2011). Analyzing and visualizing state sequences in R with TraMineR. Journal of Statistical Software, 40(4), 1–37. Grapsa, E., & Posel, D. (2016). Sequencing the real time of the elderly: Evidence from South Africa. Demographic Research, 35(25), 711-744. DOI: 10.4054/DemRes.2016.35.25 Han, Y., Liefbroer, A. C., & Elzinga, C. H. (2017). Comparing methods of classifying life courses: sequence analysis and latent class analysis. Longitudinal and Life Course Studies, 8(4), 319-341. Havighurst, R. J. (1961). Successful aging. The Gerontologist, 1, 8–13. doi:10.1093/geront/1.1.8 Hofferth, Sandra L.; Sarah M. Flood, Matthew Sobek and Daniel Backman (2020). American Time Use Survey Data Extract Builder: Version 2.8 [dataset]. College Park, MD: University of Maryland and Minneapolis, MN: IPUMS, 2020. https://doi.org/10.18128/D060.V2.8 Johnson, K., & Mutchler, J. (2014). The emergence of a positive gerontology: From disengagement to social involvement. The Gerontologist, 54(1), 93-100. Lam, J., & Garcia-Roman, J. (2020). Solitary day, solitary activities, and associations with well-being amongst older adults. Journal of Gerontology: Social Sciences, 75(7), 1585-1596. 22 Lemon, B. W., Bengtson, V. L., & Peterson, J. A. (1972). An exploration of the activity theory of aging: Activity types and life satisfaction among in-movers to a retirement community. Journal of Gerontology, 27, 511–523. doi:10.1093/geronj/27.4.511 Lesnard, L. (2004) Schedules as sequences: a new method to analyze the use of time based on collective rhythm with an application to the work arrangements of French duel-earner couples. Electronic International Journal of Time Use Research, 1, 63-88. DOI: 10.13085/eIJTUR.1.1.60-84 Lesnard, L. (2008). Off-scheduling within dual-earner couples: An unequal and negative externality for family time. American Journal of Sociology, 114(2), 447–490. Lesnard, L. (2010). Setting cost in optimal matching to uncover contemporaneous socio- temporal patterns. Sociological Methods & Research, 38(3), 389–419. Lewis, J. D., & Weigert, A. J. (1981). The structures and meanings of social time. Social Forces, 60(2), 432-462. DOI: 10.2307/2578444 Menec, V. (2003). The relation between everyday activities and successful aging: A 6-year longitudinal study. Journal of Gerontology: Social Sciences, 58(2), S74-S82. https://doi.org/10.1093/geronb/58.2.S74 Minnen, J., Glorieux, I., & van Tienoven, T. P. (2016). Who works when? Towards a typology of weekly work patterns in Belgium. Time & Society, 25(3), 652-675. https://doi.org/10.1177/0961463X15590918 Morrow-Howell, N., Putnam, M., Lee, Y. S., Greenfield, J. C., Inoue, M., & Chen, H. (2014). An investigation of activity profiles of older adults. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 69(5), 809-821. 23 Musick, K., Meier, A., & Flood, S. (2016). How parents fare: Mothers’ and fathers’ subjective well-being in time with children. American Sociological Review, 81(5), 1069-1095. https://doi.org/10.1177/0003122416663917 Ostir, G. V., Markides, K. S., Black, S. A., & Goodwin, J. S. (2000). Emotional well-being predicts subsequent functional independence and survival. Journal of American Geriatrics Society, 48, 473–478. doi:10.1111/j.1532–5415.2000.tb04991.x Ostir, G. V., Markides, K., Peek, M. K., & Goodwin, J. S. (2001). The association between emotional well-being and the incidence of stroke in older adults. Psychosomatic Medicine, 63(2), 210– 215. doi:10.1097/00006842-200103000-00003 Rowe, J. W., & Kahn, R. L. (1997). Successful aging. The Gerontologist, 37, 433-440. https://doi.org/10.1093/geront/37.4.433 Salkind, Neil. 2010. Encyclopedia of Research Design. SAGE. doi: http://dx.doi.org/10.4135/9781412961288.n404 Sullivan, O. (1997). Time waits for no (wo)man: An investigation of the gendered experience of domestic time. Sociology, 31, 331-39. https://doi.org/10.1177/0038038597031002003 Sullivan, O., & Katz-Gerro, T. (2007). The omnivorourness thesis revisited: Voracious cultural consumers. European Sociological Review, 23, 123-37. https://doi.org/10.1093/esr/jcl024 Vagni, G., & Cornwell, B. (2018). “Patterns of Everyday Activities across Social Contexts.” Proceedings of the National Academy of Sciences, 115(24),6183-6188. https://doi.org/10.1073/pnas.1718020115. Vagni, G. (2020). The social stratification of time use patterns. The British journal of sociology, 71(4), 658-679. 24 Van Tienoven, T., Glorieux, I., & Minnen, J. (2017). Exploring the stable practices of everyday life: A multi-day time-diary approach. The Sociological Review, 65(4), 745-762. https://doi.org/10.1177/0038026116674886 Yamashita, T., Bardo, A., & Liu, D. (2019). Experienced subjective well-being during physically active and passive leisure time activities among adults aged 65 years and older. The Gerontologist, 59(4), 718-726. doi: 10.1093/geront/gny106. Zerubavel, E. (1981). Hidden rhythms: Schedules and calendars in social life. Chicago: University of Chicago Press. Zisberg, A., Gur-Yaish, N., & Shochat, T. (2010). Contribution of routine to sleep quality in community elderly. Sleep, 33(4), 509-514. DOI: 10.1093/sleep/33.4.509 25 Table 1. Distribution of cases by clusters Clu_1 Clu_2 Clu3 Clu_4 Total N 1660 1642 3477 547 7326 % 22.7 22.4 47.5 7.5 Source. Own calculations from Hofferth et al 2020 Table 2. Mean time spent in selected activities by cluster. Minutes per day Activity Leisure TV Housework Work Sleep 536 542 540 472 Personal care 43 37 58 47 Eat 79 70 87 73 Childcare 5 4 13 2 Care for adults 5 4 9 3 Leisure 456 112 170 96 TV 150 520 190 121 Travel 58 31 59 73 Work 3 1 4 452 Housework 85 95 275 84 Other 20 24 33 17 Notes: Reported values are computed using analytic weights. Lines denote significant differences at p_value=0.05 using scheffé test. Source. Own calculations from Hofferth et al 2020 26 Table 3. Sociodemographic characteristics of the individuals in each cluster Leisure TV Housework Work All N 1660 1642 3477 547 7326 % 22.7 22.4 47.5 7.5 Male 47.9 51.7 36.1 59.4 43.9 Female 52.1 48.3 63.9 40.6 56.1 Living alone 31.9 32.3 28.1 22.8 29.2 Only with partner 49.6 42.8 50.9 60.6 49.9 Partner and others 8.0 8.8 9.2 11.9 9.2 Others, without partner 10.5 16.0 11.8 4.6 11.7 White 86.8 82.3 89.0 88.6 87.1 Non-white 13.2 17.7 11.0 11.4 12.9 65-74 52.1 49.4 55.8 81.7 56.2 75-84 36.9 37.8 35.7 16.8 34.6 85+ 11.0 12.8 8.4 1.5 9.2 Less than college 71.7 85.3 74.3 61.8 74.9 Some college 28.3 14.7 25.7 38.2 25.1 Not employed 88.9 93.2 88.1 3.7 81.4 Employed 11.1 6.8 11.9 96.3 18.6 Bad health 22.3 34.0 23.7 9.5 24.1 Good health 77.7 66.0 76.3 90.5 75.9 Notes: Reported values are computed using analytic weights. Source. Own calculations from Hofferth et al 2020 27 Table 4. Average wellbeing measures by cluster Leisure TV Housework Work All Pain scale 1.02 1.32 1.27 0.82 1.18 N 4459 4345 9490 1486 19780 Happiness scale 4.73 4.40 4.46 4.68 4.53 N 4400 4294 9383 1475 19552 Sadness scale 0.56 0.78 0.69 0.47 0.66 N 4437 4340 9469 1485 19731 Fatigue scale 1.62 1.67 1.94 1.92 1.82 N 4445 4335 9476 1485 19741 Stress scale 0.80 0.93 1.01 1.26 0.98 N 4449 4348 9481 1489 19767 Meaningfulness scale 4.56 4.10 4.54 4.85 4.49 N 4358 4227 9305 1471 19361 Notes: Reported values are computed using analytic weights. Lines denote significant differences at p_value=0.05 using scheffé test. Source. Own calculations from Hofferth et al 2020 28 Table 5. Average wellbeing measures by cluster and activity Wellbeing Activity measure Leisure TV Housework Work All sig Eat Pain 0.96 1.25 1.11 0.71 1.08 a e f Happiness 4.95 4.51 4.79 4.69 4.76 a d Sadness 0.41 0.72 0.50 0.46 0.52 a d e Fatigue 1.51 1.67 1.60 1.46 1.58 Stress 0.56 0.93 0.76 0.89 0.77 abc Meaningfulness 4.74 4.54 4.71 4.75 4.69 N 794 943 1611 241 3589 Leisure Pain 1.02 1.16 1.10 0.86 1.06 Happiness 4.83 4.65 4.67 4.75 4.74 Sadness 0.57 0.62 0.65 0.39 0.60 Fatigue 1.50 1.24 1.74 2.01 1.60 abcde Stress 0.81 0.73 0.76 1.03 0.79 Meaningfulness 4.77 4.48 4.78 4.81 4.74 a d N 1353 573 1478 176 3580 TV Pain 1.00 1.39 1.20 0.39 1.21 a c e f Happiness 4.30 4.29 4.18 4.91 4.29 c e f Sadness 0.63 0.86 0.74 0.33 0.76 e Fatigue 2.04 1.76 2.22 2.05 1.99 d Stress 0.86 0.95 0.91 0.38 0.89 e f Meaningfulness 3.74 3.81 3.92 4.58 3.89 c e f N 359 994 888 104 2345 Travel Pain 0.83 0.75 1.04 0.73 0.92 b d f Happiness 4.85 4.47 4.50 4.67 4.59 ab Sadness 0.45 0.66 0.59 0.58 0.57 a Fatigue 1.49 1.10 1.57 2.06 1.55 a cde f Stress 0.85 0.89 1.07 1.39 1.04 bc e f Meaningfulness 4.63 4.17 4.45 4.54 4.46 a d e N 912 635 1642 381 3570 Housework Pain 1.21 1.37 1.46 1.15 1.41 b Happiness 4.52 4.47 4.41 4.33 4.43 Sadness 0.52 0.65 0.67 0.61 0.65 Fatigue 1.66 1.89 2.03 2.41 1.99 bc e f Stress 0.82 1.07 1.19 1.36 1.15 abc Meaningfulness 4.52 4.46 4.66 4.21 4.61 d f N 792 971 3134 250 5147 Notes: Reported values are computed using analytic weights. Letters denote significant differences using scheffé test. a: significant difference between clusters Leisure and TV at the 0.05 level b: significant difference between clusters Leisure and Housework at the 0.05 level c: significant difference between clusters Leisure and Work at the 0.05 level d: significant difference between clusters TV and Housework at the 0.05 level e: significant difference between clusters TV and Work at the 0.05 level f: significant difference between clusters Housework and Work at the 0.05 level Source. Own calculations from Hofferth et al 2020 29 Figure 1. Tempo-graph showing the proportion of members in each cluster performing certain activities at each time point throughout the day (from 4 am to 4 pm) Figure 2. Sequence index plots showing time-specific activities for 250 random cases in each cluster 30 Appendix Table 1. Classification of activities Activities Codes Activities Sleep 10100-10199 Sleeping Personal care 12000-19999 Groomings, health related activities, personal care Eat 110000-119999 Eating and drinking Childcare 30000-30399 & 40000-40399 Caring for and helping household and non-household children Care for adults 30400-39999 & 40400-49999 Caring for and helping household and non-household adults Leisure 120000-159999 (except 120303) & 60000-69999 Education, socializing, sports, religious and volunteering Watch TV 120303 Watching TV Travel 180000-189999 Traveling Paid Work 50000-59999 Work and work-related activities Housework 20000-29999 & 70000-109999 Housework tasks, household management Other 160000-169999 & 500000-509999 Telephone calls, data codes Appendix 2. Measures of statistical cluster quality We use the three statistics suggested by Han et al 2017:321 to determine the optimal number of clusters. According to this paper, ASW and PBC are preferred to be as high as possible while low value of HC indicates good clustering. The estimates for the clusters 2 to 8 are as follow. Indicators N Clusters PBC ASW HC cluster2 0.258 0.122 0.341 cluster3 0.336 0.131 0.270 cluster4 0.417 0.152 0.209 cluster5 0.376 0.116 0.215 cluster6 0.356 0.104 0.214 cluster7 0.339 0.094 0.212 cluster8 0.321 0.086 0.213 31 Appendix Figure 1. Dendogram derived from the cluster procedure 32