Primary Caregivers in a Network Context

Primary Caregivers in a Network Context Abstract Objectives Caregiving to individuals affected by Alzheimer’s disease and related dementia (ADRD) is a family-systems process where tasks are distributed between multiple caregivers in a network. We evaluate the extent to which multiple network members nominate one another as filling primary caregiver (PCG) roles and factors associated with nomination. Method Data come from the Caregiving Roles and Expectations Networks project, which aimed to characterize the caregiving networks of families affected by ADRD. All persons affected by ADRD were either full-time residents in residential care facilities or community-dwelling adult day-care participants. Generalized Poisson regression was used to model the count of incoming PCG nominations of each network member. Results On average, there were multiple network members identified as PCGs across different network contexts. Network members who were perceived to perform essential caregiving tasks, such as making decisions on behalf of and spending time with the care recipient, received more primary caregiving nominations from their network peers, adjusting for personal attributes, and the context of care. Discussion Having multiple PCGs in a network may result in lack of consensus in who fills those roles, potentially putting families at risk for interpersonal conflicts. Future work aimed at intervention development should fully assess the social contexts surrounding caregiving processes in order to better understand how network composition might impact outcomes. Alzheimer’s disease, Caregiving, Social networks The need for long-term care and support of older adults continues to increase as the population ages, particularly for those affected by chronic neurodegenerative disease. Currently, there are more than 15 million individuals who provide informal care for a family member or friend who has Alzheimer’s disease or related dementia (ADRD) in the United States. This number is expected to grow rapidly in the coming years due to the aging population (Alzheimer’s Association, 2017) even as the prevalence rate of Alzheimer’s declines (Langa, Larson, & Crimmins, 2016). With the growth in both incidence and demand for caregiving to those affected by ADRD comes a need to understand the social structural processes at play in determining who fills what roles in caregiving networks. Past research has emphasized understanding the characteristics of caregivers, with a particular focus on the health and burden outcomes of single primary caregivers (PCGs) to people with ADRD (i.e., care recipients). Less emphasis, however, has been placed on understanding the social processes within family systems that determine perceptions of who is filling particular roles, because the necessary network data has only recently been collected (Koehly, Ashida, Schafer, & Ludden, 2015). The process through which individuals acquire caregiving roles within a family involves these individuals’ interactions with the care recipient and other family members, and has health implications such as role strain or difficulty performing the caregiving role, sense of uncertainty, and family conflicts (Schumacher, 1995). In particular, the process by which people embedded within such network systems are identified as PCGs becomes increasingly important for research, policy, and practice to optimally support their efforts, as PCGs tend to experience more caregiver burden than other caregivers, especially when they are the sole caregiver within the family. In this article, we take a network perspective on identifying PCGs from the set of network members surrounding care recipients in both residential care home dwelling and community-dwelling populations. Specifically, we address two research questions: (a) to what extent is there a single PCG present in these caregiving networks and (b) how do personal characteristics, role fulfillment, and caregiving context lead to individual caregivers being seen as filling PCG roles by their caregiving network peers? This research contributes to the literature on caregiving by highlighting how PCG roles are identified within broader caregiving network systems. In particular, we shed light on the interpersonal process by which perceptions of who appears to be a PCG may vary depending on the particular perspective of the observer and the context in which care is given. This can inform future research and practice by increasing clarity about the roles and expectations of members involved in care and identifying contexts in which additional support is needed through interventions that aid role transitions. Background The National Alliance for Caregiving defines the role of PCG as that who “is the sole caregiver” or provides the most unpaid (i.e., informal) care among a group of other unpaid caregivers. Historically, PCGs have received greater attention in caregiving research because of the higher levels of caregiving burden and greater access to care recipients that they experience, compared to other caregivers who may provide care to a lesser extent or indirectly by supporting PCGs (National Alliance for Caregiving & AARP, 2009). However, many families and caregiving networks have multiple members who share caregiving tasks and responsibilities (Koehly et al., 2015). In one survey of people identified as PCGs, nearly half of the respondents reported being embedded in a network of additional unpaid caregivers. In total, 70% of the respondents, both primary and non-PCGs, indicated that they shared caregiving responsibilities with other unpaid caregivers (National Alliance for Caregiving & AARP, 2009). Since only a single participant from these caregiving networks responded, it’s unknown whether other caregivers would have identified themselves (or others) as PCGs. Certainly, previous research suggests that there is typically one individual who takes on the role of PCG, providing most of the care (Joiling et al., 2013; Schulz & Martire, 2004). However, caregiving involves the performance of a large variety of tasks and roles, including providing hands-on assistance, helping with finances, coordinating services, and making care decisions for the care recipients (Schulz & Tompkins, 2010). Such a heterogeneous set of provisions often necessitates cooperation among multiple caregivers. Different social support network members may take on one or more of these functions in the provision of care to a person affected by dementia. As such, PCGs may be distinguished from other members filling supporting roles by their differential adoption of selective caregiving tasks (Dilworth-Anderson et al., 1999). Caregiving networks are also composed of different types of people, who have various other relationships with care recipients. Potential network members may consist of family members, health care professionals, friends, and others. Having such a variety of formal and informal caregivers may complicate the identification of a single PCG. Most previous studies on informal caregiving obtained information from only one informant’s perspective: usually, it is a single unpaid caregiver who is more involved in supporting the care recipient than others. However, solely relying on this caregiver’s report without considering the reports from other paid and unpaid caregivers can result in under-estimating the care provided through a broader caregiving network (Neubauer, Holle, Menn, Grossfeld-Schmitz, & Graesel, 2008). This, in turn, may also result in mis-identifying which network members fill what roles, potentially leading to off-target or under-powered interventions by not involving other key network members. For example, some network members may not self-identify as PCGs, but may be fulfilling the role of PCG from the perspectives of others in the caregiving network. This is consequential on the well-being of caregivers because individuals who do not see themselves as PCGs but are seen by others as such may be a source of psychological distress in the network as they may fail to meet the expectations of the group (Ashida et al. 2017). In consideration of these possibilities, we use a broader framework that incorporates various perspectives from multiple network members using a large dataset consisting of 30 caregiving networks. This approach will deepen our understanding of caregiving social contexts and inform past findings related to PCGs in the extant literature. Moreover, individuals may find themselves filling PCG roles unexpectedly; some may not realize that they are playing a caregiver role initially, and others may become PCGs due to a lack of other family members unable or unwilling to provide care (Koehly, 2017; Pearlin & Aneshensel, 1994). By taking a network perspective to study caregiving processes we gain the perspectives of multiple caregiving network members. This sheds light on who is performing the functions of a PCG, even when those individuals do not perceive themselves as such, and whether there is a shared understanding of these caregiving roles. Insight into who are the individuals identified as PCGs within the family, and the social functions they take on, may in turn inform our conceptual models and theories of expectations family members develop for each other in caregiving participation. To provide a frame around this question, we join a social network perspective with role labeling theory. The social network perspective treats social actors as part of a system of interaction that structures relationships, provides access to resources, and gives rise to social standing. It is an increasingly prevalent perspective in the field of gerontology and is predicted to play an important part in shaping the future of gerontological and life course research (Cornwell et al., 2015). The social network—which consists of a set of actors and the relationships between them (Wasserman & Faust, 1994)—is both a constraint on and product of the availability of interaction partners over the life course (Blau, 1977; Marcum, 2013). This dependence strongly shapes individual perspectives on their own social position and the social positions of others (Casciaro, 1998). Correspondingly, role labeling theory posits that a social role exists to the extent that there is necessity in the social system for that role (Becker, 1973; Biddle, 1986). Individuals are labeled as fulfilling a role when they meet the expectations set by the group—either society generally or the network specifically—for a particular behavior or set of behaviors. While past work on role-labeling theory predominantly focused on individuals’ own perspectives of the roles they fill, the theory is ultimately about a social process by which roles emerge through group consensus net of the specific perspectives of the individuals labeled with a role. Therefore, we advance the theoretical aspects of role labeling theory by more appropriately matching the theory with measurement as we take account of multiple perspectives on the set of expectations emerging in a caregiving network. Primary caregiving, for example, involves such a set of expectations for behavior within caregiving networks. Members from a caregiving network perceived to meet those expectations by their peers are seen as occupying such primary caregiving roles and labeled as PCGs, who in turn may adopt or reject that label. This perspective has gained traction in understanding how different family and social environments define expectations for intergenerational caregiving (Mui, 1992; Burnette, 1999). Of course, due to the variability across such caregiving contexts, the set of expectations and mechanisms giving rise to PCG roles differs from network to network; however, some plausible mechanisms have been identified in the literature. One of the mechanisms that may give rise to primary caregiving is role acquiescence; or, the process by which one takes on the responsibilities of care because nobody else has stepped in, regardless of one’s reluctance to do so. Here, individuals may nominate themselves, or be seen by others, as fulfilling PCG roles because of a lack of others available to take on such responsibilities. This may possibly arise out of filial piety or a sense of responsibility (Piercy, 1998) on the part of the caregiver, thus, family member’s kinship to the care recipient may be important in PCG role nomination. Indeed, past work has demonstrated that first degree relatives of people affected by ADRD, especially spouses and adult children, are more likely to adopt and be seen as taking on the role of PCG (Cahill, 1999). Additionally, qualitative research often reveals just a process, with self-identified PCGs reporting a lack of choice, or that they were stepping-up to provide care, when nobody else volunteered to help their loved one (Cahill, 1999; Merrill, 1996; O’Connor, 2006). There is also quantitative data consistent with this evidence: one AARP study found that half of PCGs stated that they “did not have a choice in taking on the caregiving role” compared to 37% of other non-PCGs (National Alliance for Caregiving & AARP, 2009). Such lack of choice may be due to the unavailability of other potential caregivers (Pearlin & Aneshensel, 1994) or because of their competing responsibilities such as demanding careers or parenting (Stoller & Pugliesi, 1989). Therefore, caregiving network members that have children and careers may be less likely to be seen by others as fulfilling PCG roles, factors considered in the current study. Another potential mechanism identified in the sociological literature that may give rise to primary caregiving roles is the division of caregiving roles. Individuals perceived to take on the majority of caregiving tasks may be seen to occupy PCG roles in the network. This mechanism is embodied in how Joiling et al. (2013) define PCGs as those, “who coordinated the caring processes, usually the person who spent most hours on caregiving tasks.” Of course, unequal distribution in the type of care tasks may differ from that of the quantity of time spent providing care. Observers may value particular care tasks, such as decisions about care or direct care tasks, over quantity when ascribing PCG role labels to caregivers; spending a lot of time doing caregiver tasks may not be sufficient to be identified as a PCG if those tasks are not perceived to be of high importance in care. For instance, women were more likely to be identified as PCGs—and may be more motivated become caregivers (Cahill, 1999)—even though men reported spending more time doing caregiving tasks in a nationally representative study of care provided to older people by Dwyer & Seccombe (1991). Additionally, network members that spend the most time in caregiving activities also may make decisions about care provision for others (Chadiha et al., 2011). Thus, both the type of care task and quantity of care task performance may shape who is perceived to fill PCG roles. Additionally, PCG roles may arise from an expectation of reciprocity among caregivers themselves (George, 1986); that is, caregiving networks may demonstrate a tendency for caregivers to nominate each other as fulfilling PCG roles as part of the norms of the group. In the latter case, individuals seen as being more helpful to other caregivers (and not necessarily directly to the care recipient) may be more likely to also be seen as a PCG. Thus members who are identified by family members as support providers to family caregivers may be also be identified as a PCG themselves. Finally, the context of caregiving is another important consideration into how individuals may fill PCG roles. Care recipients are heterogeneous with respect to their personal characteristics and disease progression (George & Gwyther, 1986; Karlawish, Casarett, Klocinski, & Clark, 2001). Such differences may generate variability in the need for specific caregiving functions, caregiving demand, and thus affect the number and composition of caregivers needed in the network. Several important factors shaping the context of caregiving include the care recipient’s age and years since diagnosis Fitzpatrick, Kuller, Lopez, Kawas, & Jagust (2005), gender (Hayes, Zimmerman, & Boylstein, 2010; Ott, Tate, Gordon, & Heindel, 1996), residential status (e.g., community dwelling or not) (Mittelman, Ferris, Shulman, Steinberg, & Levin, 1996; Zimmerman et al., 2005) and disease progression (Norton et al., 2009). As a result, caregivers provide various types of care depending on the “caregiving trajectory” that spans from early stage to end-of-life (Schulz & Tompkins, 2010). Those in early stages of dementia may need help with paying bills or other forms of Instrumental Activities of Daily Living (IADLs), personal care, and emotional support (Wolff, Spillman, Freedman, & Kasper, 2016). Later in the trajectory, those in advanced stages often need additional help with performing Activities of Daily Living (ADL), taking medications, making decisions about care (Gitlin & Wolff, 2011), and others may require constant supervision to ensure safety (Black et al., 2013; Schulz & Tompkins, 2010). From the perspective of role labeling theory, if someone is performing the tasks that are considered essential to the daily life of the affected individual such as these, and especially spending more time on the aforementioned tasks than others (Stommel, Given, Given, & Collins, 1995), they may be identified by others as fulfilling the role of the PCG. Variability in these factors between care recipients may account for whom within their caregiving networks are seen to be PCGs. In this study, provision of ADL assistance, discussing about health of the care recipient, and helping with health decision making, as well as spending time with the affected individual (which is essential in providing assistance, companionship, and emotional support to the care recipient) are all considered as potential attributes from prior literature that lead others to see one as fitting PCG roles. Caregiving networks necessarily involve a nested structure. Relationships are embedded between actors, which are themselves embedded in the context of care surrounding a recipient. How individual actors in these networks are seen to be filling primary caregiving roles, then, involve their own perceptions of role fulfillment, the perceptions of the other caregivers, and role demands and constraints arising from the needs of the care recipient. The network may give rise to a single PCG as has been the focus of past studies, or it may give rise to multiple PCGs, as is the focus here. In the balance of this paper, we use data from multiple caregiving networks surrounding residential care home dwelling and community-dwelling people affected by ADRD from the greater Memphis, Tennessee region to shed light on the emergence of multiple primary caregiving roles using a model that preserves the nested structure. Design and Methods Our data come from the Caregiving Roles and Expectations Networks (CaRENet) project, which was conducted to characterize the caregiving networks of families affected by ADRD (Koehly et al., 2015). Participants in this study were recruited in 2012 from four facilities around Memphis, Tennessee that provide services to older adults affected by ADRD. The four sites consisted of an assisted living facility specializing in dementia care with approximately 50 beds, two memory care units that were part of larger continuing care campuses with about 150 beds; and an adult day service that provides daily care to about 100 individuals at two locations. Each participant was asked to refer in other caregiving network members to the study. As previously described by both Koehly et al. (2015) and Ashida, Marcum, & Koehly (2017), the caregiving network was defined to include those who are involved in care, support family members involved in care, and/or could be involved in care due to their relationships with the affected relative. Collectively, the caregiving networks included: immediate family members of the affected relative including spouses and first-degree relatives; other people important to the affected relative or to other participants (including kin and non-kin); and, paid formal caregivers from the facility staff who have been important to the affected relative and/or the participant. A total of 72 individuals (30 index and 42 non-index) from 30 families along with 43 formal care providers from four facilities were interviewed. The first (index) participant from each family was recruited passively through materials posted at these facilities and actively by research staff handing out materials or introducing the study at resident/family meetings. Both biological and non-biological (spouses, step- and adopted children and siblings) family members of individuals receiving services at these facilities were eligible to participate. These index participants enumerated members of the care recipient’s caregiving network by listing other people that met at least one of the following criteria: (a) immediate family members of the affected relative including spouse and first-degree relatives, (b) persons, either family or non-family, who are important to the affected relative, (c) persons, either family or non-family, who are important to the participant, and (d) facility staff members who have been important to the affected relative and/or the participant. Participants were contacted approximately one week after their interview to obtain contact information of eligible network members (biological and non-biological family members of the affected individual and facility staff members) for whom they gave us permission to recruit. Referred individuals were contacted, consented, interviewed, and asked to refer other eligible network members into the study. Each participant enumerated their own network members based on the same elicitation questions. This respondent-driven network sampling (Goodman, 1961; Heckathorn & Cameron, 2017) approach continued until recruit saturation was met: all referred members participated, refused, or could not be reached. All participants were fluent in English and did not have physical, cognitive, or mental difficulties influencing their ability to answer interview questions. Figure 1 details the recruitment, eligibility, and response flowchart for the index and non-index participants. This study was approved by the Institutional Review Boards at the National Human Genome Research Institute and University of Memphis. Figure 1. View largeDownload slide Recruitment flowchart for participating families and caregiving network members. Figure 1. View largeDownload slide Recruitment flowchart for participating families and caregiving network members. Network informants were asked to nominate who among the members they enumerated were considered PCGs, including, potentially, themselves. This initial “primary caregiving nomination network” is a directed network with loops (self-nominations). The network data used in the present study differ from that reported in our previously published research [i.e., Koehly et al. (2015)] in two important ways: first, we include reports from both formal and informal caregivers whereas the previous study only included reports from informal caregivers; and second, our relational ties are PCG nominations between network members whereas the previous study simply used the total network enumeration roster (e.g., to establish who is in the caregiving network without respect to specific roles or network ties). Thus, the former difference results in a larger sample of actors and the latter difference results in a more specific network. Because not everyone from each network answered the questions, we only have a sample of the members informing on each network. This approach yields a distribution of different perspectives on who may, or may not, fill PCG roles. The total number of individuals within all caregiving networks we collected information on was 679 (including 115 network informants); missing covariate data were imputed by predictive mean matching on 148 of these individuals while three individuals were dropped due to too much missing data (more than 50% of their observations). The total amount of imputed data represented about 2% of all observations, which had no substantive effect on the analysis or results. Dependent Variable Our dependent variable is the “in-degree” of each network member. That is, we take the number of primary caregiving nominations a network member receives from all informants (including, perhaps, oneself) in the caregiving network. Formally, we sum each network member’s (ai) in-degree (DI) on caregiving network G, where Gij = 1 is informant j’s declaration that network member i fills a PCG role: DI(ai)=∑jnGij,∀i. Predictor Variables Our primary independent variables are of three types: socio-demographic attributes of the network members, functions that the network members are perceived to perform in the caregiving network, and contextual attributes of the caregiving network. As we have multiple accounts of both network and caregiver attributes, we take either the mean (for numeric) or mode (for categorical) of each covariate across the set of reports. Our network member background controls include age and age2 (as middle-age caregivers are expected to be more likely than younger and older caregivers to fulfill PCG roles), gender, relationship to the person affected—adult child (ref), spouse, other relative, friend, formal caregiver, or other important person—a five level categorical variable for the number of children top-coded at 3+ and including unreported number of children (no children = ref), four category employment status variable including unreported employment status (employed = ref), and a five category marital status variable (married = ref) including unreported marital status. Each of these covariates is thought to tap into an aspect of role labeling theory discussed above. For instance, we hypothesize that middle-age adults, women, and relatives of the affected individual are more likely to fulfil primary caregiving roles as the normative expectations about attributes of caregivers map to those variables via filial piety. We also expect that individuals perceived to be exposed to competing roles associated with employment and family life (being married and having children) will have fewer PCG nominations as they will be perceived by the network to be unable to fulfill the functions of PCGs. We include five measures of caregiver roles, all derived from informant reports of each member fulfilling a particular function. Specifically, we count the in-degree for the number of times an individual is said (by their caregiver peers) to help with caregiving tasks, to discuss the health of, to spend enough time with, to make decisions on behalf of, and to assist the care recipient with activities of daily living, respectively. Collectively, these measures tap into the expectations of functions and tasks done by PCGs. From role-labeling theory, we expect that as more network members perceive individuals as fulfilling these tasks and roles, the higher their PCG nomination in-degree will be. To capture the context of caregiving (level two in the hierarchical regression model described below), we include five variables that measure characteristics of the care recipient. First, we measure the severity of the care recipient’s condition as the average sum of scores from informants’ reports that the care recipient needs assistance with 18 different activities of daily living rated on a four-item scale. Each specific item asked whether the care recipient needs assistance “no assistance” (coded 0), “a little assistance” (coded 1), “a lot” (coded 2), and “total assistance” (coded 3) with the activity. The internal reliability of this measure is high, with a Cronbach’s α = 0.89. We also include the gender of the care recipient (1 = male), the age of the care recipient and number of years since their diagnosis (these two variables are moderately positively correlated r = 0.27), and whether the care recipient resides in an assisted-living recruitment site or resides at home but is enrolled in a community-center adult day-care program (1 = resident). Methods We have two-levels of analysis—the network level and the individual (network member) level. As our dependent variable, in-degree, is a count of the number of PCG nominations each network member receives from the informants, we use a hierarchical Poisson regression to model in-degree as a function of caregiver characteristics jointly with the within-network variation accounted for at level two (Christiansen & Morris, 1997) by the network-level covariates. Poisson regression with a log-link is an appropriate choice of model for such count data (McCullagh & Nelder, 1989). The model is sufficiently powered to detect an effect size of |0.15| at the α = 0.05 with 80% precision (Browne, Lahi, & Parker, 2009). To ensure that the estimated rate model is homogeneous across families, we offset each network member’s in-degree by the number of informants in their respective caregiving network. The offset also effectively controls for network size in the model as the two variables are highly correlated (r = 0.65). Thus, the model coefficients represent the effect difference in the log rate of receiving primary caregiving nominations given the number of nominations that were possible in a particular network. Additionally, the level 2 variance components adjust for sources of variation arising from the clustering of individual responses within each network. We report the coefficients and significance levels associated with the level 1 effects and the point-estimates of the variance components associated with the level 2 effects. Results Univariate descriptive statistics are reported in Table 1. The table is divided into three sections corresponding to the three types of covariates we include in our multivariate model: network member background characteristics, perceptions of caregiver functions from network members, and care recipient background characteristics (these are our level 2 covariates). Means and standard deviations are reported. As expected, our sample is older than the population on average at a mean age of about 53 years old, and predominantly female (two-thirds of caregivers). The plurality of caregivers are relatives of the care recipient (about 39%). About half of the caregivers have at least one child under the age of 18 living at home, and most are employed at least part of the time. While most caregivers are married, we were unable to ascertain marital statuses of about a third of the caregivers either due to omission or informant uncertainty (primarily, this information was missing for formal caregivers). While not shown here, just under half of our informants (35 or 49%) self-identified as a PCG. To examine whether self-nomination was related to a caregiver’s indegree or other factors, we conducted exploratory logistic regression models predicting self-nomination as a function of each covariate in the final model (adjusting only for the number of informants per family as we lack degrees of freedom to do more). The results revealed that the only significant predictor of self-nomination was, in fact, indegree (our primary outcome). Specifically, a single caregiving nomination from another network member multiplies the odds of self-nominating by 2.57 (β = 0.945, p < .001) on average. Table 1. Descriptive Statistics of the Sample Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 a Adjusted for the number of informants. View Large Table 1. Descriptive Statistics of the Sample Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 a Adjusted for the number of informants. View Large Adjusting for the number of network informants, the average number of PCG nominations is 4 and the standard deviation is high (about 3.7) owing to the fact that, on average, many network members receive no nominations and a few receive many—this phenomena is consistent with a Matthew Effect commonly found in social networks [Barabási, 2009; Strogatz, 2001)]—and all networks had at least one nominated PCG. Concordant with the low connectivity of these networks, only the number of health discussion partners exceeds this number on average (about 6), whereas network members who are perceived to provide support to other caregivers, those seen as spending enough time with the care recipient, those making decisions on their behalf, and those providing the care recipient with ADL assistance are all relatively sparse. While our outcome is a count variable, it is still useful to consider how much consensus in primary caregiving nominations exists, on average, in these networks. The overall agreement rate, that is the fraction of informants agreeing on whether or not others in the caregiving network fill PCG roles is 89%. The conditional agreement rate, that is net of those not seen as filling PCG roles, is 39%. Thus, on average, there is more consensus in the networks surrounding perceptions of who does not fill PCG roles than there is around who does fill PCG roles. Our other network-level variables are summarized under “Network Perceptions of Caregiver Functions” in Table 1. First, the average caregiving network included 24 (±7.9) members. As expected, most of the affected people were female (71%), were relatively advanced in their condition with average severity being 20 out of a possible 54, had been living with a diagnosis an average of 6 years since our study began, were in their early 80s, and most were residents of assisted-living sites rather than day-care visitors (61% versus 39%, respectively). As the descriptive statistics suggest, individual caregiving networks varied greatly in terms of size, number of caregiving nominations, and composition. Figure 2 graphically illustrates 3 (out of 30) selected PCG nomination networks from the data. Network members that were interviewed (the informants) are represented by dark gray circles while non-interviewed network members are colored in light gray. The size of the circles is scaled in proportion to the network member’s in-degree (our dependent variable). Isolates (network members receiving no PCG nominations) are suppressed in the illustration. Also not shown is the person affected by ADRD as they did not participate directly in this study. Directed arrows indicate that the informant nominates the other member as a PCG. Loops indicate that a person declared his- or herself to be a PCG. The figure illustrates the variability across the networks, including: evidence of cases where a clear consensus for the PCG role is present (Network A); cases where non-informants are at least as preferred as informants (Network B); and pluralistic cases where no specific caregiver is preferred over any other in the network (Network C). The frequency distribution of networks by the number of uniquely nominated PCGs is depicted in Figure 3, which demonstrates such variation across networks. The histogram highlights how a single PCG is present in only a minority of the caregiving networks. Figure 2. View largeDownload slide Primary caregiver (PCG) nominations in three selected caregiving networks. Circles represent members of a caregiving network. Circle color indicates whether the network member was an informant (dark gray) or an enumerated non-informant (light gray). Arrows represent PCG nominations emanating from an informant and directed toward either themselves (i.e., the loops or self-ties) or another network member. The network layout was accomplished via the Fruchterman–Reingold algorithm and then tuned slightly for aesthetics. Figure 2. View largeDownload slide Primary caregiver (PCG) nominations in three selected caregiving networks. Circles represent members of a caregiving network. Circle color indicates whether the network member was an informant (dark gray) or an enumerated non-informant (light gray). Arrows represent PCG nominations emanating from an informant and directed toward either themselves (i.e., the loops or self-ties) or another network member. The network layout was accomplished via the Fruchterman–Reingold algorithm and then tuned slightly for aesthetics. Figure 3. View largeDownload slide Frequency distribution of networks by the number of uniquely nominated primary caregivers. Figure 3. View largeDownload slide Frequency distribution of networks by the number of uniquely nominated primary caregivers. The results from the hierarchical regression model of caregiver in-degree are reported in Table 2. Three models are presented representing effects separately for network member background covariates in Model 1, caregiver functions in Model 2, and both in Model 3. All three models adjust for within-network variation arising from the differences in network context by including level two variance components in this hierarchical model. Table 2. Results from the Hierarchical Poisson Regression Model of the Number of Primary Caregiver Nominations on Network Member-Level Covariates and Network-Level Variance Components Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Note. ADL = Activities of Daily Living; ALSO = Akaike Information Criteria; AP = Affected Person. *** p < .001 ** p < .01 * p < .05. View Large Table 2. Results from the Hierarchical Poisson Regression Model of the Number of Primary Caregiver Nominations on Network Member-Level Covariates and Network-Level Variance Components Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Note. ADL = Activities of Daily Living; ALSO = Akaike Information Criteria; AP = Affected Person. *** p < .001 ** p < .01 * p < .05. View Large When only caregiver covariates are considered (Model 1), we find no effect from either age or gender, net of other factors and controlling for the context of caregiving variance components in level two. Adult children of the care recipient receive more PCG nominations than spouses, other family members, friends, medical providers, and other people important to the care recipient as indicated by the respective significant negative coefficients. Childless caregivers tend to receive more PCG nominations than those with 1, 2, 3, or an unknown number of children (which do not differ). Unemployed, compared with employed, caregivers receive fewer PCG nominations (no other employment status significantly differed from the employed). There were no significant differences in marital status with respect to the number of PCG nominations. Level 1 effects in Model 2 represent the network informant’s perceptions of themselves and other network member’s roles. We find that the number of times a network member is said to support another caregiver significantly decreases the number of PCG nominations he or she receives. By contrast, the number of times one is nominated as discussing the health of, being seen as spending enough time with, making decisions on behalf of, and assisting with tasks for the care recipient all significantly increase the number of PCG nominations one receives. However, taking both sets of covariates together in Model 3, only relationship of the network member to, being seen as spending enough time with, and being seen as one who makes decisions on behalf of, the care recipient remain significant predictors of PCG nominations net of the other factors. This final model, which is preferred by AIC, strongly implicates the importance of a network member’s relationship to their relative affected with ADRD jointly with their role fulfillment in being perceived by others to be PCGs. Finally, the variance components at Level 2 meant to capture sources of variability surrounding the context of care in different networks from this model demonstrate that, all other things constant, there is relatively large variability in the average number of PCG nominations across caregiving networks (i.e., intercept variance term that is >9). Interestingly, while we find no significant gender difference in the number of PCG nominations at Level 1 among caregiving network members (β = 0.1751, p < .1714), there is more variability in the average number of PCG nominations among families whose affected relative is male at Level 2 (Var(Y | Gender=MALE) = 1.5) in this model. All other variance component estimates were close to zero, indicating that those factors do not contribute to the between-family variability of the outcome. Discussion The multiple caregiving network comparisons are a unique feature of our data; this is one of the only studies to date to have collected multiple networks of this nature. By tapping into the caregiving networks surrounding people affected by ADRD, we have revealed that there is variation in the number of network members thought to fill PCG roles across families. While several networks (n = 6 or 20%) had only a single PCG uniquely nominated, the plurality of networks contained multiple PCGs (n = 24 or 80%). We also find that certain support members have a greater likelihood of being perceived by their network peers as PCGs than others. Controlling for personal factors and within-network covariates, we find that caregivers who are children of, are seen as making decisions on behalf of, and thought to spend enough time with, the affected person receive more PCG nominations from their peers. These findings support two of the mechanisms through which family members may acquire the PCG role discussed earlier; through filial piety (Piercy, 1998), quantity of time spent (Dwyer & Seccombe, 1991) and engagement in specific caregiving tasks (Joiling et al., 2013) with the affected individual. From a role-label theoretic, these findings suggest evidence of a shared mental model of who fulfills primary caregiving roles based in part on perceptions of specific task fulfillment by network members and in part based on knowledge of prior relationships to the care recipient. Importantly, by taking into account multiple-perspectives on whom fulfills primary caregiving roles in these networks we’ve more closely mapped the measurement and theoretical aspects of role labeling theory, which emphasizes how labels arise through the social process of consensus from the group (here, the caregiving network) that individuals fill a particular role. That there are, on average, more than one PCG in these networks implies that these tasks may be distributed among many caregivers rather than concentrated on a single individual. It is plausible that such a label “sticks” only as consensus builds within the network, and more individuals are perceived to be involved in these responsibilities by a greater number of observers. Future work should evaluate the extent to which individual actors adopt primary caregiving roles as a function of their peers’ building such a consensus. Interpreting these results in light of the finding that there are, on average, more than one uniquely identified PCG in caregiving networks, we conclude that the sole PCG model is not sufficient to capture the heterogeneity within networks and between individual caregivers. Rather, researchers should take care to cast a wide net—focusing on decision makers and being attuned to the personal characteristics of the affected person—when designing studies to investigate key caregiver roles in the context of ADRD. Moreover, interventions for alleviating caregiver burden may be better designed to take a network perspective, relying on multiple informants rather than solely on self-reports, to identify optimal targets. At the network level (level two in our model), few variance components were substantial enough to warrant discussion. Certainly, there was evidence of between network heterogeneity, and meaningful variance arising from differences in the gender and residency status of the person affected by ADRD. It is possible that the small level two variance components are the result of the fact that we have only 30 networks for comparison. This highlights one of the trade-offs and limitations affecting our study. On the one hand, there are few past studies with multiple whole-networks as their data, which we address here by collecting 30 multiple-informant networks at great cost (Marcum et al., 2017). However, on the other hand, this is still a small sample when the network is itself a source of variation in outcome, which constrains making inferences using multivariate analyses. As a result, we limit our discussion to the network-member level analysis. Another limitation of our study is that we focus on a population of people with ADRD that have access to formal care facilities. Certainly, many (if not most) of the ADRD population do not have this luxury and the caregiving networks—including the distribution of PCGs in them—may differ between individuals with and without access to formal care. However, our study highlights an under-examined aspect of the context of caregiving: different perspectives on the caregiving network arising from a mixture of informal and formal caregivers being present. Moreover, according to the Family Caregiver Alliance (2015), roughly a quarter of U.S. adults ages 65 and older received a combination of informal and formal care. On a typical day in 2012, there were 273,200 and 22,200 older adults being served at adult day centers and residential care home or similar residential care communities, respectively (Harris-Kojetin, Sengupta, Park-Lee, & Valverde, 2013). This is a sizeable population worthy of study. We note, however, that our results may reflect only the population of such networks residing in the greater Memphis area. More research is needed to evaluate the extent of generalizability and regional variation in these types of caregiving networks. We also recognize that such family caregiving plans may involve the preferences of and proximity to the care recipient. Pillemer & Suitor (2006) showed that mothers considered physical proximity and employment status of their children as they select who they wanted to become their PCG, suggesting that care recipients may prefer those who can spend enough time with them as their caregivers potentially illuminating the importance of high quality social interactions. As a limitation, we are unable to directly account for this effect in our study. We argue that this availability preference is manifest in our result that network members who were thought to spend enough time with the care recipient were more likely to be perceived as a PCG among informants. This likely reflects the perceived availability and ability of these network members to provide needed assistance and support. In addition to being seen as spending enough time with the care recipient, our results also point to the importance of decision-makers in the caregiving network. Controlling for the relationship between network members and the person affected by ADRD, network members who are seen to participate in care decision making are more often identified as PCGs by other network members. An increasing number of adults have comorbid illness and experience more complex medical conditions (Wolff & Kasper, 2006), therefore making decisions about medical treatment and care coordination increasingly important. It may be that those who are able and willing to make such decisions are seen as playing key roles within these familial caregiving networks, even if they do not involve themselves in day-to-day caregiving tasks per se. The potential importance of this role, making decisions related to coordinating care, in addition to the provision of direct care should be considered in future research and practice as social and economic contexts surrounding caregiving continue to shift in our society with shifting medical landscapes. Networks wherein multiple people are identified as serving PCG roles may also pose a challenge to legal and institutional precedents that establish compensatory and record-keeping mechanisms around a single individual who makes these important decisions. For example, a single individual needs to be identified for the purposes of insurance billing or granting power-of-attorney (Arling & McAuley, 1983; Grabowski, Stevenson, Huskamp, & Keating, 2006; Kapp, 2003), suggesting some reform may be needed to ease institutional dissonance between the multiple caregiver context of caregiving and the broader social system. Overall, our results suggest that there is heterogeneity in caregiving processes across families as well as within each caregiving network. This highlights the importance of fully assessing the social contexts surrounding caregiving to understand those social processes important in defining caregiving roles and expectations and to develop interventions that facilitate effective performance of the wide variety of tasks needed to support the care recipient. For example, although the vast majorities of the participating families had multiple PCG nominations, a minority of 20% had only one PCG nominated. Such families likely will need different types of interventions (e.g., facilitating member participation in caregiving or distribution of CG roles) compared to those with multiple PCGs (e.g., discussion of who fulfills what role/tasks). To inform such interventions, future research is needed to identify whether and how this variability in PCG nominations is related family adaptation to caregiver stress and caregiver well-being. For instance, it’s possible that a large number of caregivers being perceived to fill PCG roles in a single caregiving network may be the result of a lack of consensus among network members. This, in turn, may be an indicator of underlying vulnerabilities of negative outcomes in the network (such as family conflict or lack of cohesion). Finally, leveraging multiple informants from caregiving networks allows for the identification of shared perceptions about caregiving roles that particular network members may play (Koehly et al., 2015). In turn, this may help identify not only the core members of caregiving networks, but also the key characteristics of network members who would be optimal to mobilize in efforts to reinforce support systems to families touched by ADRD. For instance, intervention studies targeting PCGs should consider the possibility that more than one person in the caregiving network may need to be reached to achieve optimal outcomes. Funding This work was supported by the Intramural Research Program of the National Institutes of Health (grant number ZIA HG200395 to L.M.K.). Conflict of Interest We have no COIs to report. Acknowledgements We would like to thank the anonymous reviewers and associate editor Dr. Suitor as well as Drs. Deborah Carr, Judith Treas, and Jielu Lin for their thoughtful comments on previous drafts of this manuscript. C.S.M. conceived of the current study, performed the analyses, and wrote the article. S.A. and L.M.K. designed the original study, collected the data, and contributed to writing and revising the article. References Alzheimer’s Association . ( 2017 ). Alzheimer’s Disease facts and figures (Report) . Chicago, IL : Alzheimer’s Association . PubMed PubMed Arling , G. , & McAuley , W. J . ( 1983 ). The feasibility of public payments for family caregiving . The Gerontologist , 23 , 300 – 306 . doi:10.1093/geront/23.3.300 Google Scholar CrossRef Search ADS PubMed Ashida , S. , Marcum , C. S. , & Koehly , L. M . ( 2017 ). Unmet expectations in Alzheimer’s family caregiving: interactional characteristics associated with perceived under-contribution . The Gerontologist . doi: 10.1093/geront/gnx141 Barabási , A. L . ( 2009 ). Scale-free networks: a decade and beyond . Science (New York, N.Y.) , 325 , 412 – 413 . doi: 10.1126/science.1173299 Google Scholar CrossRef Search ADS PubMed Becker , H. S . ( 1973 ). Labelling theory reconsidered . In Becker H. S. (ed.), Outsiders: Studies in the Sociology of Deviance . New York: Free Press. Biddle , B. J . ( 1986 ). Recent development in role theory . Annual Review of Sociology , 12, 67 – 92 . doi:10.1146/annurev.so.12.080186.000435 Black , B. S. , Johnston , D. , Rabins , P. V. , Morrison , A. , Lyketsos , C. , & Samus , Q. M . ( 2013 ). Unmet needs of community-residing persons with dementia and their informal caregivers: findings from the maximizing independence at home study . Journal of the American Geriatrics Society , 61 , 2087 – 2095 . doi:10.1111/jgs.12549 Google Scholar CrossRef Search ADS PubMed Blau , P. M . ( 1977 ). A macrosociological theory of social structure . The American Journal of Sociology , 83 ( 1 ), 26 – 54 . doi:10.1086/226505 Google Scholar CrossRef Search ADS Browne , W. J. , Lahi , M. G. , & Parker , R. M . ( 2009 ). A guide to sample size calculations for random effect models via simulation and the MLPowSim software Package (Tech. Rep. No. 0-903024-96-9) . Bristol, UK : University of Bristol . Burnette , D . ( 1999 ). Social relationships of Latino grandparent caregivers: a role theory perspective . The Gerontologist , 39 , 49 – 58 . doi:10.1093/geront/39.1.49 Google Scholar CrossRef Search ADS PubMed Cahill , S. M . ( 1999 ). Caring in families: what motivates wives, daughters, and daughters-in-law to provide dementia care ? Journal of Family Studies , 5 , 235 – 247 . doi: 10.5172/jfs.5.2.235 Google Scholar CrossRef Search ADS Casciaro , T . ( 1998 ). Seeing things clearly: Social structure, personality, and accuracy in social network perception . Social Networks , 20 ( 4 ), 331 – 351 . doi:10.1016/S0378-8733(98)00008-2 Google Scholar CrossRef Search ADS Chadiha , L. A. , Feld , S. , & Rafferty , J . ( 2011 ). Likelihood of African American primary caregivers and care recipients receiving assistance from secondary caregivers: A rural-urban comparison . Journal of Applied Gerontology , 30 , 422 – 442 . doi: 10.1177/0733464810371099 Google Scholar CrossRef Search ADS PubMed Christiansen , C. L. , & Morris , C. N . ( 1997 ). Hierarchical Poisson regression modeling . Journal of the American Statistical Association , 92 ( 438 ), 618 – 632 . doi:10.1080/01621459.1997.10474013 Google Scholar CrossRef Search ADS Cornwell , B. , Marcum , C. , & Silverstein , M . ( 2015 ). The social network approach in gerontological research . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 70 , 87 – 90 . doi: 10.1093/geronb/gbu157 Google Scholar CrossRef Search ADS Dilworth-Anderson , P. , Williams , S. W. , & Cooper , T . ( 1999 ). Family caregiving to elderly African Americans: Caregiver types and structures . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 54 , S237 – S241 . doi:10.1093/geronb/54B.4.S237 Google Scholar CrossRef Search ADS Dwyer , J. W. , & Seccombe , K . ( 1991 ). Elder care as family labor: The influence of gender and family position . Journal of Family Studies , 12 ( 2 ), 229 – 247 . doi:10.1177/019251391012002006 Family Caregiver Alliance . ( 2015 ). Selected long-term care statistics (Report) . San Francisco, CA : National Center on Caregiving . Retrieved from: https://www.caregiver.org/selected-long-term-care-statistics Fitzpatrick , A. L. , Kuller , L. H. , Lopez , O. L. , Kawas , C. H. , & Jagust , W . ( 2005 ). Survival following dementia onset: Alzheimer’s disease and vascular dementia . Journal of the Neurological Sciences , 229-230 , 43 – 49 . doi: 10.1016/j.jns.2004.11.022 Google Scholar CrossRef Search ADS PubMed George , L. K . ( 1986 ). Caregiver burden: Conflict between norms of reciprocity and solidarity . In Pillemer K. A. & Wolf R. S. (Eds.), Elder abuse: Conflict in the family (pp. 67 – 92 ). New York : Auburn House . George , L. K. , & Gwyther , L. P . ( 1986 ). Caregiver well-being: a multidimensional examination of family caregivers of demented adults . The Gerontologist , 26 , 253 – 259 . doi:10.1093/geront/26.3.253 Google Scholar CrossRef Search ADS PubMed Gitlin , L. N. , & Wolff , J . ( 2011 ). Family involvement in care transitions of older adults: What do we know and where do we go from here ? Annual Review of Gerontology and Geriatrics , 31 ( 1 ), 31 – 64 . doi:10.1891/0198-8794.31.31 Google Scholar CrossRef Search ADS Goodman , L. A . ( 1961 ). Snowball sampling . The Annals of Mathematical Statistics , 32 ( 1 ), 148 – 170 . doi:10.1214/aoms/1177705148 Google Scholar CrossRef Search ADS Grabowski , D. C. , Stevenson , D. G. , Huskamp , H. A. , & Keating , N. L . ( 2006 ). The influence of Medicare home health payment incentives: Does payer source matter ? INQUIRY , 43 , 135 – 149 . doi: 10.5034/inquiryjrnl_43.2.135 Google Scholar PubMed Harris-Kojetin , L. , Sengupta , M. , Park-Lee , E. , & Valverde , R . ( 2013 ). Long-term care services in the United States: 2013 (Vital Health Statistics No. 3(37)) . San Francisco, CA : National Center for Health Statistics . Retrieved from: https://www.caregiver.org/selected-long-term-care-statistics Hayes , J. , Zimmerman , M. K. , & Boylstein , C . ( 2010 ). Responding to symptoms of Alzheimer’s disease: husbands, wives, and the gendered dynamics of recognition and disclosure . Qualitative Health Research , 20 , 1101 – 1115 . doi: 10.1177/1049732310369559 Google Scholar CrossRef Search ADS PubMed Heckathorn , D. D. , & Cameron , C. J . ( 2017 ). Network sampling: From snowball and multiplicity to respondent-driven sampling . Annual Review of Sociology , 43 , 101 – 119 . doi:10.1146/annurev-soc-060116-053556 Google Scholar CrossRef Search ADS Joiling , K. J. , Bosmans , J. E. , van Marwijk , H. W. , van der Horst , H. E. , Scheltens , P. , Vroomen , J. L. M. , & van Hout , H. P . ( 2013 ). The cost-effectiveness of a family meetings intervention to prevent depression and anxiety in family caregivers of patients with dementia: a randomized trial . Trials , 14 ( 1 ), 305 . doi: 10.1186/1745-6215-14-305 Google Scholar CrossRef Search ADS PubMed Kapp , M. B . ( 2003 ). Family caregivers’ legal concerns . Generations , 27 ( 4 ), 49 – 55 . Karlawish , J. H. , Casarett , D. , Klocinski , J. , & Clark , C. M . ( 2001 ). The relationship between caregivers’ global ratings of Alzheimer’s disease patients’ quality of life, disease severity, and the caregiving experience . Journal of the American Geriatrics Society , 49 , 1066 – 1070 . doi:10.1046/j.1532-5415.2001.49210.x Google Scholar CrossRef Search ADS PubMed Koehly , L. M . ( 2017 ). It’s interpersonal: Family relationships, genetic risk, and caregiving . The Gerontologist , 57 , 32 – 39 . doi: 10.1093/geront/gnw103 Google Scholar CrossRef Search ADS PubMed Koehly , L. M. , Ashida , S. , Schafer , E. J. , & Ludden , A . ( 2015 ). Caregiving networks-using a network approach to identify missed opportunities . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 70 , 143 – 154 . doi: 10.1093/geronb/gbu111 Google Scholar CrossRef Search ADS Langa , K. M. , Larson , E. B. , Crimmins , E. M. , Faul , J. D. , Levine , D. A. , Kabeto , M. U. & Weir , D. R . ( 2017 ). A comparison of the prevalence of dementia in the United States in 2000 and 2012 . JAMA: Internal Medicine , 177 ( 1 ), 51 – 58 . doi: 10.1001/jamainternmed.2016.6807 Google Scholar CrossRef Search ADS PubMed Marcum , C. S . ( 2013 ). Age differences in daily social activities . Research on Aging , 35 , 612 – 640 . doi: 10.1177/0164027512453468 Google Scholar CrossRef Search ADS PubMed Marcum , C. S. , Lienert , J. , Goldring , M. , Lin , J. , Miggins , A. , Moss , M. , … Koehly , L . ( 2017 ). Ego-centered cognitive social structures of close personal networks in the United States . Unpublished preprint retrieved from: osf.io/4ugk6 McCullagh , P. , & Nelder , J. A . ( 1989 ). Generalized linear models, no. 37 in Monograph on statistics and applied probability . ( 2nd Ed .) Chapman & Hall : London . doi:10.1007/978-1- 4899-3242-6 Merrill , D. M . ( 1996 ). Conflict and cooperation among adult siblings during the transition to the role of filial caregiver . Journal of Social and Personal Relationships , 13 ( 3 ), 399 – 413 . doi: 10.1177/0265407596133006 Google Scholar CrossRef Search ADS Mittelman , M. S. , Ferris , S. H. , Shulman , E. , Steinberg , G. , & Levin , B . ( 1996 ). A family intervention to delay nursing home placement of patients with Alzheimer disease. A randomized controlled trial . JAMA , 276 , 1725 – 1731 . doi:10.1001/jama.1996.03540210033030 Google Scholar CrossRef Search ADS PubMed Mui , A. C . ( 1992 ). Caregiver strain among black and white daughter caregivers: a role theory perspective . The Gerontologist , 32 , 203 – 212 . doi:10.1093/geront/32.2.203 Google Scholar CrossRef Search ADS PubMed National Alliance for Caregiving, & AARP . ( 2009 ). Caregiving in the U.S. 2009 (Report) . Washington, DC : National Alliance for Caregiving . Retrieved from http://www.caregiving.org/data/Caregiving_in_the_US_2009_full_report.pdf Neubauer , S. , Holle , R. , Menn , P. , Grossfeld-Schmitz , M. , & Graesel , E . ( 2008 ). Measurement of informal care time in a study of patients with dementia . International Psychogeriatrics , 20 , 1160 – 1176 . doi: 10.1017/S1041610208007564 Google Scholar CrossRef Search ADS PubMed Norton , M. C. , Piercy , K. W. , Rabins , P. V. , Green , R. C. , Breitner , J. C. , Ostbye , T. ,… Tschanz , J. T . ( 2009 ). Caregiver-recipient closeness and symptom progression in Alzheimer disease. The Cache County Dementia Progression Study . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 64 , 560 – 568 . doi: 10.1093/geronb/gbp052 Google Scholar CrossRef Search ADS O’Connor , D. L . ( 2006 ). Self-identifying as a caregiver: exploring the positioning process . Journal of Aging Studies , 21 ( 2 ), 165 – 174 . doi: 10.1016/j.jaging.2006.06.002 Google Scholar CrossRef Search ADS Ott , B. R. , Tate , C. A. , Gordon , N. M. , & Heindel , W. C . ( 1996 ). Gender differences in the behavioral manifestations of Alzheimer’s disease . Journal of the American Geriatrics Society , 44 , 583 – 587 . doi:10.1111/j.1532-5415.1996.tb01447.x Google Scholar CrossRef Search ADS PubMed Pearlin , L. I. , & Aneshensel , C. S . ( 1994 ). Caregiving: The unexpected career . Social Justice Research , 7 ( 4 ), 373 – 390 . doi:10.1007/BF02334863 Google Scholar CrossRef Search ADS Piercy , K. W . ( 1998 ). Theorizing about family caregiving: The role of responsibility . Journal of Marriage and the Family , 60 ( 1 ), 109 – 118 . doi:10.2307/353445 Google Scholar CrossRef Search ADS Pillemer , K. , & Suitor , J. J . ( 2006 ). Making choices: a within-family study of caregiver selection . The Gerontologist , 46 , 439 – 448 . doi:10.1093/geront/46.4.439 Google Scholar CrossRef Search ADS PubMed Schulz , R. , & Martire , L. M . ( 2004 ). Family caregiving of persons with dementia: prevalence, health effects, and support strategies . The American Journal of Geriatric Psychiatry , 12 , 240 – 249 . doi:10.1097/00019442-200405000-00002 Google Scholar CrossRef Search ADS PubMed Schulz , R. , & Tompkins , C. A . ( 2010 ). Informal caregivers in the United States: Prevalence, caregiver characteristics, and ability to provide care . In Olson S. (Ed.), The role of human factors in home health care (pp. 117 – 144 ). Washington, DC : National Academies Press . Schumacher , K. L . ( 1995 ). Family caregiver role acquisition: role-making through situated interaction . Scholarly Inquiry for Nursing Practice , 9 , 211 – 26; discussion 227 . Google Scholar PubMed Stoller , E. P. , & Pugliesi , K. L . ( 1989 ). Other roles of caregivers: Competing responsibilities or supportive resources . Journal of Gerontology , 44 , S231 – S238 . doi:10.1093/geronj/44.6.S231 Google Scholar CrossRef Search ADS PubMed Stommel , M. , Given , B. A. , Given , C. W. , & Collins , C . ( 1995 ). The impact of the frequency of care activities on the division of labor between primary caregivers and other care providers . Research on Aging , 17 ( 4 ), 412 – 433 . doi: 10.1177/0164027595174003 Google Scholar CrossRef Search ADS Strogatz , S. H . ( 2001 ). Exploring complex networks . Nature , 410 , 268 – 276 . doi: 10.1038/35065725 Google Scholar CrossRef Search ADS PubMed Wasserman , S. , & Faust , K . ( 1994 ). Social network analysis: Methods and applications . Cambridge : Cambridge University Press . Google Scholar CrossRef Search ADS Wolff , J. L. , & Kasper , J. D . ( 2006 ). Caregivers of frail elders: updating a national profile . The Gerontologist , 46 , 344 – 356 . doi:10.1093/geront/46.3.344 Google Scholar CrossRef Search ADS PubMed Wolff , J. L. , Spillman , B. C. , Freedman , V. A. , & Kasper , J. D . ( 2016 ). A national profile of family and unpaid caregivers who assist older adults with health care activities . JAMA Internal Medicine , 176 , 372 – 379 . doi: 10.1001/jamainternmed.2015.7664 Google Scholar CrossRef Search ADS PubMed Zimmerman , S. , Sloane , P. D. , Williams , C. S. , Reed , P. S. , Preisser , J. S. , Eckert , J. K. ,… Dobbs , D . ( 2005 ). Dementia care and quality of life in assisted living and nursing homes . The Gerontologist , 45 , 133 – 146 . doi:10.1093/geront/45.suppl_1.133 Google Scholar CrossRef Search ADS PubMed Published by Oxford University Press on behalf of The Gerontological Society of America 2017. This work is written by (a) US Government employee(s) and is in the public domain in the US. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Journals of Gerontology Series B: Psychological Sciences and Social Sciences Oxford University Press

Loading next page...
 
/lp/ou_press/primary-caregivers-in-a-network-context-9KpgRzD2jP
Publisher
Oxford University Press
Copyright
Published by Oxford University Press on behalf of The Gerontological Society of America 2017. This work is written by (a) US Government employee(s) and is in the public domain in the US.
ISSN
1079-5014
eISSN
1758-5368
D.O.I.
10.1093/geronb/gbx165
Publisher site
See Article on Publisher Site

Abstract

Abstract Objectives Caregiving to individuals affected by Alzheimer’s disease and related dementia (ADRD) is a family-systems process where tasks are distributed between multiple caregivers in a network. We evaluate the extent to which multiple network members nominate one another as filling primary caregiver (PCG) roles and factors associated with nomination. Method Data come from the Caregiving Roles and Expectations Networks project, which aimed to characterize the caregiving networks of families affected by ADRD. All persons affected by ADRD were either full-time residents in residential care facilities or community-dwelling adult day-care participants. Generalized Poisson regression was used to model the count of incoming PCG nominations of each network member. Results On average, there were multiple network members identified as PCGs across different network contexts. Network members who were perceived to perform essential caregiving tasks, such as making decisions on behalf of and spending time with the care recipient, received more primary caregiving nominations from their network peers, adjusting for personal attributes, and the context of care. Discussion Having multiple PCGs in a network may result in lack of consensus in who fills those roles, potentially putting families at risk for interpersonal conflicts. Future work aimed at intervention development should fully assess the social contexts surrounding caregiving processes in order to better understand how network composition might impact outcomes. Alzheimer’s disease, Caregiving, Social networks The need for long-term care and support of older adults continues to increase as the population ages, particularly for those affected by chronic neurodegenerative disease. Currently, there are more than 15 million individuals who provide informal care for a family member or friend who has Alzheimer’s disease or related dementia (ADRD) in the United States. This number is expected to grow rapidly in the coming years due to the aging population (Alzheimer’s Association, 2017) even as the prevalence rate of Alzheimer’s declines (Langa, Larson, & Crimmins, 2016). With the growth in both incidence and demand for caregiving to those affected by ADRD comes a need to understand the social structural processes at play in determining who fills what roles in caregiving networks. Past research has emphasized understanding the characteristics of caregivers, with a particular focus on the health and burden outcomes of single primary caregivers (PCGs) to people with ADRD (i.e., care recipients). Less emphasis, however, has been placed on understanding the social processes within family systems that determine perceptions of who is filling particular roles, because the necessary network data has only recently been collected (Koehly, Ashida, Schafer, & Ludden, 2015). The process through which individuals acquire caregiving roles within a family involves these individuals’ interactions with the care recipient and other family members, and has health implications such as role strain or difficulty performing the caregiving role, sense of uncertainty, and family conflicts (Schumacher, 1995). In particular, the process by which people embedded within such network systems are identified as PCGs becomes increasingly important for research, policy, and practice to optimally support their efforts, as PCGs tend to experience more caregiver burden than other caregivers, especially when they are the sole caregiver within the family. In this article, we take a network perspective on identifying PCGs from the set of network members surrounding care recipients in both residential care home dwelling and community-dwelling populations. Specifically, we address two research questions: (a) to what extent is there a single PCG present in these caregiving networks and (b) how do personal characteristics, role fulfillment, and caregiving context lead to individual caregivers being seen as filling PCG roles by their caregiving network peers? This research contributes to the literature on caregiving by highlighting how PCG roles are identified within broader caregiving network systems. In particular, we shed light on the interpersonal process by which perceptions of who appears to be a PCG may vary depending on the particular perspective of the observer and the context in which care is given. This can inform future research and practice by increasing clarity about the roles and expectations of members involved in care and identifying contexts in which additional support is needed through interventions that aid role transitions. Background The National Alliance for Caregiving defines the role of PCG as that who “is the sole caregiver” or provides the most unpaid (i.e., informal) care among a group of other unpaid caregivers. Historically, PCGs have received greater attention in caregiving research because of the higher levels of caregiving burden and greater access to care recipients that they experience, compared to other caregivers who may provide care to a lesser extent or indirectly by supporting PCGs (National Alliance for Caregiving & AARP, 2009). However, many families and caregiving networks have multiple members who share caregiving tasks and responsibilities (Koehly et al., 2015). In one survey of people identified as PCGs, nearly half of the respondents reported being embedded in a network of additional unpaid caregivers. In total, 70% of the respondents, both primary and non-PCGs, indicated that they shared caregiving responsibilities with other unpaid caregivers (National Alliance for Caregiving & AARP, 2009). Since only a single participant from these caregiving networks responded, it’s unknown whether other caregivers would have identified themselves (or others) as PCGs. Certainly, previous research suggests that there is typically one individual who takes on the role of PCG, providing most of the care (Joiling et al., 2013; Schulz & Martire, 2004). However, caregiving involves the performance of a large variety of tasks and roles, including providing hands-on assistance, helping with finances, coordinating services, and making care decisions for the care recipients (Schulz & Tompkins, 2010). Such a heterogeneous set of provisions often necessitates cooperation among multiple caregivers. Different social support network members may take on one or more of these functions in the provision of care to a person affected by dementia. As such, PCGs may be distinguished from other members filling supporting roles by their differential adoption of selective caregiving tasks (Dilworth-Anderson et al., 1999). Caregiving networks are also composed of different types of people, who have various other relationships with care recipients. Potential network members may consist of family members, health care professionals, friends, and others. Having such a variety of formal and informal caregivers may complicate the identification of a single PCG. Most previous studies on informal caregiving obtained information from only one informant’s perspective: usually, it is a single unpaid caregiver who is more involved in supporting the care recipient than others. However, solely relying on this caregiver’s report without considering the reports from other paid and unpaid caregivers can result in under-estimating the care provided through a broader caregiving network (Neubauer, Holle, Menn, Grossfeld-Schmitz, & Graesel, 2008). This, in turn, may also result in mis-identifying which network members fill what roles, potentially leading to off-target or under-powered interventions by not involving other key network members. For example, some network members may not self-identify as PCGs, but may be fulfilling the role of PCG from the perspectives of others in the caregiving network. This is consequential on the well-being of caregivers because individuals who do not see themselves as PCGs but are seen by others as such may be a source of psychological distress in the network as they may fail to meet the expectations of the group (Ashida et al. 2017). In consideration of these possibilities, we use a broader framework that incorporates various perspectives from multiple network members using a large dataset consisting of 30 caregiving networks. This approach will deepen our understanding of caregiving social contexts and inform past findings related to PCGs in the extant literature. Moreover, individuals may find themselves filling PCG roles unexpectedly; some may not realize that they are playing a caregiver role initially, and others may become PCGs due to a lack of other family members unable or unwilling to provide care (Koehly, 2017; Pearlin & Aneshensel, 1994). By taking a network perspective to study caregiving processes we gain the perspectives of multiple caregiving network members. This sheds light on who is performing the functions of a PCG, even when those individuals do not perceive themselves as such, and whether there is a shared understanding of these caregiving roles. Insight into who are the individuals identified as PCGs within the family, and the social functions they take on, may in turn inform our conceptual models and theories of expectations family members develop for each other in caregiving participation. To provide a frame around this question, we join a social network perspective with role labeling theory. The social network perspective treats social actors as part of a system of interaction that structures relationships, provides access to resources, and gives rise to social standing. It is an increasingly prevalent perspective in the field of gerontology and is predicted to play an important part in shaping the future of gerontological and life course research (Cornwell et al., 2015). The social network—which consists of a set of actors and the relationships between them (Wasserman & Faust, 1994)—is both a constraint on and product of the availability of interaction partners over the life course (Blau, 1977; Marcum, 2013). This dependence strongly shapes individual perspectives on their own social position and the social positions of others (Casciaro, 1998). Correspondingly, role labeling theory posits that a social role exists to the extent that there is necessity in the social system for that role (Becker, 1973; Biddle, 1986). Individuals are labeled as fulfilling a role when they meet the expectations set by the group—either society generally or the network specifically—for a particular behavior or set of behaviors. While past work on role-labeling theory predominantly focused on individuals’ own perspectives of the roles they fill, the theory is ultimately about a social process by which roles emerge through group consensus net of the specific perspectives of the individuals labeled with a role. Therefore, we advance the theoretical aspects of role labeling theory by more appropriately matching the theory with measurement as we take account of multiple perspectives on the set of expectations emerging in a caregiving network. Primary caregiving, for example, involves such a set of expectations for behavior within caregiving networks. Members from a caregiving network perceived to meet those expectations by their peers are seen as occupying such primary caregiving roles and labeled as PCGs, who in turn may adopt or reject that label. This perspective has gained traction in understanding how different family and social environments define expectations for intergenerational caregiving (Mui, 1992; Burnette, 1999). Of course, due to the variability across such caregiving contexts, the set of expectations and mechanisms giving rise to PCG roles differs from network to network; however, some plausible mechanisms have been identified in the literature. One of the mechanisms that may give rise to primary caregiving is role acquiescence; or, the process by which one takes on the responsibilities of care because nobody else has stepped in, regardless of one’s reluctance to do so. Here, individuals may nominate themselves, or be seen by others, as fulfilling PCG roles because of a lack of others available to take on such responsibilities. This may possibly arise out of filial piety or a sense of responsibility (Piercy, 1998) on the part of the caregiver, thus, family member’s kinship to the care recipient may be important in PCG role nomination. Indeed, past work has demonstrated that first degree relatives of people affected by ADRD, especially spouses and adult children, are more likely to adopt and be seen as taking on the role of PCG (Cahill, 1999). Additionally, qualitative research often reveals just a process, with self-identified PCGs reporting a lack of choice, or that they were stepping-up to provide care, when nobody else volunteered to help their loved one (Cahill, 1999; Merrill, 1996; O’Connor, 2006). There is also quantitative data consistent with this evidence: one AARP study found that half of PCGs stated that they “did not have a choice in taking on the caregiving role” compared to 37% of other non-PCGs (National Alliance for Caregiving & AARP, 2009). Such lack of choice may be due to the unavailability of other potential caregivers (Pearlin & Aneshensel, 1994) or because of their competing responsibilities such as demanding careers or parenting (Stoller & Pugliesi, 1989). Therefore, caregiving network members that have children and careers may be less likely to be seen by others as fulfilling PCG roles, factors considered in the current study. Another potential mechanism identified in the sociological literature that may give rise to primary caregiving roles is the division of caregiving roles. Individuals perceived to take on the majority of caregiving tasks may be seen to occupy PCG roles in the network. This mechanism is embodied in how Joiling et al. (2013) define PCGs as those, “who coordinated the caring processes, usually the person who spent most hours on caregiving tasks.” Of course, unequal distribution in the type of care tasks may differ from that of the quantity of time spent providing care. Observers may value particular care tasks, such as decisions about care or direct care tasks, over quantity when ascribing PCG role labels to caregivers; spending a lot of time doing caregiver tasks may not be sufficient to be identified as a PCG if those tasks are not perceived to be of high importance in care. For instance, women were more likely to be identified as PCGs—and may be more motivated become caregivers (Cahill, 1999)—even though men reported spending more time doing caregiving tasks in a nationally representative study of care provided to older people by Dwyer & Seccombe (1991). Additionally, network members that spend the most time in caregiving activities also may make decisions about care provision for others (Chadiha et al., 2011). Thus, both the type of care task and quantity of care task performance may shape who is perceived to fill PCG roles. Additionally, PCG roles may arise from an expectation of reciprocity among caregivers themselves (George, 1986); that is, caregiving networks may demonstrate a tendency for caregivers to nominate each other as fulfilling PCG roles as part of the norms of the group. In the latter case, individuals seen as being more helpful to other caregivers (and not necessarily directly to the care recipient) may be more likely to also be seen as a PCG. Thus members who are identified by family members as support providers to family caregivers may be also be identified as a PCG themselves. Finally, the context of caregiving is another important consideration into how individuals may fill PCG roles. Care recipients are heterogeneous with respect to their personal characteristics and disease progression (George & Gwyther, 1986; Karlawish, Casarett, Klocinski, & Clark, 2001). Such differences may generate variability in the need for specific caregiving functions, caregiving demand, and thus affect the number and composition of caregivers needed in the network. Several important factors shaping the context of caregiving include the care recipient’s age and years since diagnosis Fitzpatrick, Kuller, Lopez, Kawas, & Jagust (2005), gender (Hayes, Zimmerman, & Boylstein, 2010; Ott, Tate, Gordon, & Heindel, 1996), residential status (e.g., community dwelling or not) (Mittelman, Ferris, Shulman, Steinberg, & Levin, 1996; Zimmerman et al., 2005) and disease progression (Norton et al., 2009). As a result, caregivers provide various types of care depending on the “caregiving trajectory” that spans from early stage to end-of-life (Schulz & Tompkins, 2010). Those in early stages of dementia may need help with paying bills or other forms of Instrumental Activities of Daily Living (IADLs), personal care, and emotional support (Wolff, Spillman, Freedman, & Kasper, 2016). Later in the trajectory, those in advanced stages often need additional help with performing Activities of Daily Living (ADL), taking medications, making decisions about care (Gitlin & Wolff, 2011), and others may require constant supervision to ensure safety (Black et al., 2013; Schulz & Tompkins, 2010). From the perspective of role labeling theory, if someone is performing the tasks that are considered essential to the daily life of the affected individual such as these, and especially spending more time on the aforementioned tasks than others (Stommel, Given, Given, & Collins, 1995), they may be identified by others as fulfilling the role of the PCG. Variability in these factors between care recipients may account for whom within their caregiving networks are seen to be PCGs. In this study, provision of ADL assistance, discussing about health of the care recipient, and helping with health decision making, as well as spending time with the affected individual (which is essential in providing assistance, companionship, and emotional support to the care recipient) are all considered as potential attributes from prior literature that lead others to see one as fitting PCG roles. Caregiving networks necessarily involve a nested structure. Relationships are embedded between actors, which are themselves embedded in the context of care surrounding a recipient. How individual actors in these networks are seen to be filling primary caregiving roles, then, involve their own perceptions of role fulfillment, the perceptions of the other caregivers, and role demands and constraints arising from the needs of the care recipient. The network may give rise to a single PCG as has been the focus of past studies, or it may give rise to multiple PCGs, as is the focus here. In the balance of this paper, we use data from multiple caregiving networks surrounding residential care home dwelling and community-dwelling people affected by ADRD from the greater Memphis, Tennessee region to shed light on the emergence of multiple primary caregiving roles using a model that preserves the nested structure. Design and Methods Our data come from the Caregiving Roles and Expectations Networks (CaRENet) project, which was conducted to characterize the caregiving networks of families affected by ADRD (Koehly et al., 2015). Participants in this study were recruited in 2012 from four facilities around Memphis, Tennessee that provide services to older adults affected by ADRD. The four sites consisted of an assisted living facility specializing in dementia care with approximately 50 beds, two memory care units that were part of larger continuing care campuses with about 150 beds; and an adult day service that provides daily care to about 100 individuals at two locations. Each participant was asked to refer in other caregiving network members to the study. As previously described by both Koehly et al. (2015) and Ashida, Marcum, & Koehly (2017), the caregiving network was defined to include those who are involved in care, support family members involved in care, and/or could be involved in care due to their relationships with the affected relative. Collectively, the caregiving networks included: immediate family members of the affected relative including spouses and first-degree relatives; other people important to the affected relative or to other participants (including kin and non-kin); and, paid formal caregivers from the facility staff who have been important to the affected relative and/or the participant. A total of 72 individuals (30 index and 42 non-index) from 30 families along with 43 formal care providers from four facilities were interviewed. The first (index) participant from each family was recruited passively through materials posted at these facilities and actively by research staff handing out materials or introducing the study at resident/family meetings. Both biological and non-biological (spouses, step- and adopted children and siblings) family members of individuals receiving services at these facilities were eligible to participate. These index participants enumerated members of the care recipient’s caregiving network by listing other people that met at least one of the following criteria: (a) immediate family members of the affected relative including spouse and first-degree relatives, (b) persons, either family or non-family, who are important to the affected relative, (c) persons, either family or non-family, who are important to the participant, and (d) facility staff members who have been important to the affected relative and/or the participant. Participants were contacted approximately one week after their interview to obtain contact information of eligible network members (biological and non-biological family members of the affected individual and facility staff members) for whom they gave us permission to recruit. Referred individuals were contacted, consented, interviewed, and asked to refer other eligible network members into the study. Each participant enumerated their own network members based on the same elicitation questions. This respondent-driven network sampling (Goodman, 1961; Heckathorn & Cameron, 2017) approach continued until recruit saturation was met: all referred members participated, refused, or could not be reached. All participants were fluent in English and did not have physical, cognitive, or mental difficulties influencing their ability to answer interview questions. Figure 1 details the recruitment, eligibility, and response flowchart for the index and non-index participants. This study was approved by the Institutional Review Boards at the National Human Genome Research Institute and University of Memphis. Figure 1. View largeDownload slide Recruitment flowchart for participating families and caregiving network members. Figure 1. View largeDownload slide Recruitment flowchart for participating families and caregiving network members. Network informants were asked to nominate who among the members they enumerated were considered PCGs, including, potentially, themselves. This initial “primary caregiving nomination network” is a directed network with loops (self-nominations). The network data used in the present study differ from that reported in our previously published research [i.e., Koehly et al. (2015)] in two important ways: first, we include reports from both formal and informal caregivers whereas the previous study only included reports from informal caregivers; and second, our relational ties are PCG nominations between network members whereas the previous study simply used the total network enumeration roster (e.g., to establish who is in the caregiving network without respect to specific roles or network ties). Thus, the former difference results in a larger sample of actors and the latter difference results in a more specific network. Because not everyone from each network answered the questions, we only have a sample of the members informing on each network. This approach yields a distribution of different perspectives on who may, or may not, fill PCG roles. The total number of individuals within all caregiving networks we collected information on was 679 (including 115 network informants); missing covariate data were imputed by predictive mean matching on 148 of these individuals while three individuals were dropped due to too much missing data (more than 50% of their observations). The total amount of imputed data represented about 2% of all observations, which had no substantive effect on the analysis or results. Dependent Variable Our dependent variable is the “in-degree” of each network member. That is, we take the number of primary caregiving nominations a network member receives from all informants (including, perhaps, oneself) in the caregiving network. Formally, we sum each network member’s (ai) in-degree (DI) on caregiving network G, where Gij = 1 is informant j’s declaration that network member i fills a PCG role: DI(ai)=∑jnGij,∀i. Predictor Variables Our primary independent variables are of three types: socio-demographic attributes of the network members, functions that the network members are perceived to perform in the caregiving network, and contextual attributes of the caregiving network. As we have multiple accounts of both network and caregiver attributes, we take either the mean (for numeric) or mode (for categorical) of each covariate across the set of reports. Our network member background controls include age and age2 (as middle-age caregivers are expected to be more likely than younger and older caregivers to fulfill PCG roles), gender, relationship to the person affected—adult child (ref), spouse, other relative, friend, formal caregiver, or other important person—a five level categorical variable for the number of children top-coded at 3+ and including unreported number of children (no children = ref), four category employment status variable including unreported employment status (employed = ref), and a five category marital status variable (married = ref) including unreported marital status. Each of these covariates is thought to tap into an aspect of role labeling theory discussed above. For instance, we hypothesize that middle-age adults, women, and relatives of the affected individual are more likely to fulfil primary caregiving roles as the normative expectations about attributes of caregivers map to those variables via filial piety. We also expect that individuals perceived to be exposed to competing roles associated with employment and family life (being married and having children) will have fewer PCG nominations as they will be perceived by the network to be unable to fulfill the functions of PCGs. We include five measures of caregiver roles, all derived from informant reports of each member fulfilling a particular function. Specifically, we count the in-degree for the number of times an individual is said (by their caregiver peers) to help with caregiving tasks, to discuss the health of, to spend enough time with, to make decisions on behalf of, and to assist the care recipient with activities of daily living, respectively. Collectively, these measures tap into the expectations of functions and tasks done by PCGs. From role-labeling theory, we expect that as more network members perceive individuals as fulfilling these tasks and roles, the higher their PCG nomination in-degree will be. To capture the context of caregiving (level two in the hierarchical regression model described below), we include five variables that measure characteristics of the care recipient. First, we measure the severity of the care recipient’s condition as the average sum of scores from informants’ reports that the care recipient needs assistance with 18 different activities of daily living rated on a four-item scale. Each specific item asked whether the care recipient needs assistance “no assistance” (coded 0), “a little assistance” (coded 1), “a lot” (coded 2), and “total assistance” (coded 3) with the activity. The internal reliability of this measure is high, with a Cronbach’s α = 0.89. We also include the gender of the care recipient (1 = male), the age of the care recipient and number of years since their diagnosis (these two variables are moderately positively correlated r = 0.27), and whether the care recipient resides in an assisted-living recruitment site or resides at home but is enrolled in a community-center adult day-care program (1 = resident). Methods We have two-levels of analysis—the network level and the individual (network member) level. As our dependent variable, in-degree, is a count of the number of PCG nominations each network member receives from the informants, we use a hierarchical Poisson regression to model in-degree as a function of caregiver characteristics jointly with the within-network variation accounted for at level two (Christiansen & Morris, 1997) by the network-level covariates. Poisson regression with a log-link is an appropriate choice of model for such count data (McCullagh & Nelder, 1989). The model is sufficiently powered to detect an effect size of |0.15| at the α = 0.05 with 80% precision (Browne, Lahi, & Parker, 2009). To ensure that the estimated rate model is homogeneous across families, we offset each network member’s in-degree by the number of informants in their respective caregiving network. The offset also effectively controls for network size in the model as the two variables are highly correlated (r = 0.65). Thus, the model coefficients represent the effect difference in the log rate of receiving primary caregiving nominations given the number of nominations that were possible in a particular network. Additionally, the level 2 variance components adjust for sources of variation arising from the clustering of individual responses within each network. We report the coefficients and significance levels associated with the level 1 effects and the point-estimates of the variance components associated with the level 2 effects. Results Univariate descriptive statistics are reported in Table 1. The table is divided into three sections corresponding to the three types of covariates we include in our multivariate model: network member background characteristics, perceptions of caregiver functions from network members, and care recipient background characteristics (these are our level 2 covariates). Means and standard deviations are reported. As expected, our sample is older than the population on average at a mean age of about 53 years old, and predominantly female (two-thirds of caregivers). The plurality of caregivers are relatives of the care recipient (about 39%). About half of the caregivers have at least one child under the age of 18 living at home, and most are employed at least part of the time. While most caregivers are married, we were unable to ascertain marital statuses of about a third of the caregivers either due to omission or informant uncertainty (primarily, this information was missing for formal caregivers). While not shown here, just under half of our informants (35 or 49%) self-identified as a PCG. To examine whether self-nomination was related to a caregiver’s indegree or other factors, we conducted exploratory logistic regression models predicting self-nomination as a function of each covariate in the final model (adjusting only for the number of informants per family as we lack degrees of freedom to do more). The results revealed that the only significant predictor of self-nomination was, in fact, indegree (our primary outcome). Specifically, a single caregiving nomination from another network member multiplies the odds of self-nominating by 2.57 (β = 0.945, p < .001) on average. Table 1. Descriptive Statistics of the Sample Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 a Adjusted for the number of informants. View Large Table 1. Descriptive Statistics of the Sample Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 Network Member Background Mean Std. Dev. Network Perceptions of Caregiver Functionsa Mean Std. Dev. Age 52.782 19.832 Is primary caregiver 4.389 3.655 Female 0.615 0.487 Helps other caregiver 2.757 1.616 Male 0.385 0.487 Discusses health of recipient 6.362 3.213 Spouse of recipient 0.086 0.280 Spends enough time with recipient 4.102 4.375 Adult child of recipient 0.181 0.386 Makes decisions for recipient 5.584 6.425 Other family member of recipient 0.122 0.327 Assists recipient with ADLs 3.476 2.700 Friend of recipient 0.221 0.416 Care Recipient Background Formal caregiver of recipient 0.073 0.260  Age 81.728 6.587 Another important person 0.316 0.465  Female 0.717 0.451 Has 1 child 0.091 0.288  Male 0.283 0.451 Has 2 children 0.191 0 .393 Symptom severity score 19.778 7.475 Has 3 children 0.109 0.312 Years since diagnosed 5.560 2.329 Has 4+ children 0.056 0 .231 Is day-care participant 0.385 0.487 Does not have children 0.494 0.500 Is resident 0.615 0.487 Unknown # of children 0.060 0.238 Network size 23.957 7.859 Employed 0.546 0.498 Retired 0.160 0.367 Unemployed 0.107 0.309 Employment status unknown 0.187 0.390 Married 0.405 0.491 Never married 0.181 0.386 Divorced/separated 0.065 0.247 Widowed 0.065 0.247 Marital status unknown 0.283 0.451 a Adjusted for the number of informants. View Large Adjusting for the number of network informants, the average number of PCG nominations is 4 and the standard deviation is high (about 3.7) owing to the fact that, on average, many network members receive no nominations and a few receive many—this phenomena is consistent with a Matthew Effect commonly found in social networks [Barabási, 2009; Strogatz, 2001)]—and all networks had at least one nominated PCG. Concordant with the low connectivity of these networks, only the number of health discussion partners exceeds this number on average (about 6), whereas network members who are perceived to provide support to other caregivers, those seen as spending enough time with the care recipient, those making decisions on their behalf, and those providing the care recipient with ADL assistance are all relatively sparse. While our outcome is a count variable, it is still useful to consider how much consensus in primary caregiving nominations exists, on average, in these networks. The overall agreement rate, that is the fraction of informants agreeing on whether or not others in the caregiving network fill PCG roles is 89%. The conditional agreement rate, that is net of those not seen as filling PCG roles, is 39%. Thus, on average, there is more consensus in the networks surrounding perceptions of who does not fill PCG roles than there is around who does fill PCG roles. Our other network-level variables are summarized under “Network Perceptions of Caregiver Functions” in Table 1. First, the average caregiving network included 24 (±7.9) members. As expected, most of the affected people were female (71%), were relatively advanced in their condition with average severity being 20 out of a possible 54, had been living with a diagnosis an average of 6 years since our study began, were in their early 80s, and most were residents of assisted-living sites rather than day-care visitors (61% versus 39%, respectively). As the descriptive statistics suggest, individual caregiving networks varied greatly in terms of size, number of caregiving nominations, and composition. Figure 2 graphically illustrates 3 (out of 30) selected PCG nomination networks from the data. Network members that were interviewed (the informants) are represented by dark gray circles while non-interviewed network members are colored in light gray. The size of the circles is scaled in proportion to the network member’s in-degree (our dependent variable). Isolates (network members receiving no PCG nominations) are suppressed in the illustration. Also not shown is the person affected by ADRD as they did not participate directly in this study. Directed arrows indicate that the informant nominates the other member as a PCG. Loops indicate that a person declared his- or herself to be a PCG. The figure illustrates the variability across the networks, including: evidence of cases where a clear consensus for the PCG role is present (Network A); cases where non-informants are at least as preferred as informants (Network B); and pluralistic cases where no specific caregiver is preferred over any other in the network (Network C). The frequency distribution of networks by the number of uniquely nominated PCGs is depicted in Figure 3, which demonstrates such variation across networks. The histogram highlights how a single PCG is present in only a minority of the caregiving networks. Figure 2. View largeDownload slide Primary caregiver (PCG) nominations in three selected caregiving networks. Circles represent members of a caregiving network. Circle color indicates whether the network member was an informant (dark gray) or an enumerated non-informant (light gray). Arrows represent PCG nominations emanating from an informant and directed toward either themselves (i.e., the loops or self-ties) or another network member. The network layout was accomplished via the Fruchterman–Reingold algorithm and then tuned slightly for aesthetics. Figure 2. View largeDownload slide Primary caregiver (PCG) nominations in three selected caregiving networks. Circles represent members of a caregiving network. Circle color indicates whether the network member was an informant (dark gray) or an enumerated non-informant (light gray). Arrows represent PCG nominations emanating from an informant and directed toward either themselves (i.e., the loops or self-ties) or another network member. The network layout was accomplished via the Fruchterman–Reingold algorithm and then tuned slightly for aesthetics. Figure 3. View largeDownload slide Frequency distribution of networks by the number of uniquely nominated primary caregivers. Figure 3. View largeDownload slide Frequency distribution of networks by the number of uniquely nominated primary caregivers. The results from the hierarchical regression model of caregiver in-degree are reported in Table 2. Three models are presented representing effects separately for network member background covariates in Model 1, caregiver functions in Model 2, and both in Model 3. All three models adjust for within-network variation arising from the differences in network context by including level two variance components in this hierarchical model. Table 2. Results from the Hierarchical Poisson Regression Model of the Number of Primary Caregiver Nominations on Network Member-Level Covariates and Network-Level Variance Components Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Note. ADL = Activities of Daily Living; ALSO = Akaike Information Criteria; AP = Affected Person. *** p < .001 ** p < .01 * p < .05. View Large Table 2. Results from the Hierarchical Poisson Regression Model of the Number of Primary Caregiver Nominations on Network Member-Level Covariates and Network-Level Variance Components Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Model 1 Model 2 Model 3 (Intercept) −3.3346 (0.8237)*** −3.7606 (0.1863)*** −3.8810 (0.8640)*** Age 0.0384 (0.0238) 0.0053 (0.0245) Age2 −0.0002 (0.0002) 0.0000 (0.0002) Female 0.1816 (0.1522) 0.1751 (0.1714) Relationship (ref = adult child)  Other family member −2.2523 (0.4755)*** −1.2887 (0.5616)*  Friend −1.8342 (0.2422)*** −0.8301 (0.3051)**  Formal caregiver −3.2300 (0.6229)*** −1.7909 (0.7435)*  Other important person −2.5733 (0.2435)*** −1.4776 (0.3376)***  Spouse −1.4157 (0.3373)*** −0.7025 (0.3667) Number of children (ref = 1)  No. children = 2 −0.1354 (0.5170) −0.1533 (0.5747)  No. children = 3 −0.3017 (0.5830) −0.2047 (0.6221)  No. children > 3 −0.9111 (0.8532) −0.8804 (0.8917)  No. children 0.9928 (0.4759)* 0.5600 (0.5361)  No. children unknown 1.3832 (0.7725) 0.7965 (0.8172) Employment status (ref = employed)  Unknown 0.5270 (0.3344) 0.5677 (0.4651)  Retired 0.3296 (0.2679) 0.1888 (0.2941)  Unemployed −1.7440 (0.7514)* −1.4498 (0.7932) Marital status (ref = married)  Unknown 0.7260 (0.3474)* 0.8611 (0.4909)  Never married 0.1409 (0.2317) 0.3523 (0.2710)  Divorced/separated −0.5662 (0.5340) −0.1613 (0.5689)  Widowed −0.6149 (0.4980) −0.1112 (0.5528) Supports other caregivers −0.5195 (0.1487)*** −0.1752 (0.1536) Discusses health of AP 0.0510 (0.0173)* 0.0099 (0.0221) Spends enough time with AP 0.7696 (0.1142)*** 0.5766 (0.1156)*** Makes decisions for AP 0.8223 (0.1228)*** 0.7107 (0.1533)*** Assists AP with ADLs 0.3367 (0.1190)* 0.1809 (0.1355) AIC 832.2308 764.8345 719.3758 Log likelihood −374.1154 −355.4172 −312.6879 Variance: (Intercept) 4.2284 1.6439 9.3144 Variance: AP gender = Male 0.2719 0.8559 1.4487 Variance: AP severity score 0.0001 0.0012 0.0005 Variance: AP years since Diag. 0.0027 0.0366 0.001 Variance: AP age 0.0002 0.0004 0.0013 Variance: AP is resident 0.1444 1.3425 0.8383 Note. ADL = Activities of Daily Living; ALSO = Akaike Information Criteria; AP = Affected Person. *** p < .001 ** p < .01 * p < .05. View Large When only caregiver covariates are considered (Model 1), we find no effect from either age or gender, net of other factors and controlling for the context of caregiving variance components in level two. Adult children of the care recipient receive more PCG nominations than spouses, other family members, friends, medical providers, and other people important to the care recipient as indicated by the respective significant negative coefficients. Childless caregivers tend to receive more PCG nominations than those with 1, 2, 3, or an unknown number of children (which do not differ). Unemployed, compared with employed, caregivers receive fewer PCG nominations (no other employment status significantly differed from the employed). There were no significant differences in marital status with respect to the number of PCG nominations. Level 1 effects in Model 2 represent the network informant’s perceptions of themselves and other network member’s roles. We find that the number of times a network member is said to support another caregiver significantly decreases the number of PCG nominations he or she receives. By contrast, the number of times one is nominated as discussing the health of, being seen as spending enough time with, making decisions on behalf of, and assisting with tasks for the care recipient all significantly increase the number of PCG nominations one receives. However, taking both sets of covariates together in Model 3, only relationship of the network member to, being seen as spending enough time with, and being seen as one who makes decisions on behalf of, the care recipient remain significant predictors of PCG nominations net of the other factors. This final model, which is preferred by AIC, strongly implicates the importance of a network member’s relationship to their relative affected with ADRD jointly with their role fulfillment in being perceived by others to be PCGs. Finally, the variance components at Level 2 meant to capture sources of variability surrounding the context of care in different networks from this model demonstrate that, all other things constant, there is relatively large variability in the average number of PCG nominations across caregiving networks (i.e., intercept variance term that is >9). Interestingly, while we find no significant gender difference in the number of PCG nominations at Level 1 among caregiving network members (β = 0.1751, p < .1714), there is more variability in the average number of PCG nominations among families whose affected relative is male at Level 2 (Var(Y | Gender=MALE) = 1.5) in this model. All other variance component estimates were close to zero, indicating that those factors do not contribute to the between-family variability of the outcome. Discussion The multiple caregiving network comparisons are a unique feature of our data; this is one of the only studies to date to have collected multiple networks of this nature. By tapping into the caregiving networks surrounding people affected by ADRD, we have revealed that there is variation in the number of network members thought to fill PCG roles across families. While several networks (n = 6 or 20%) had only a single PCG uniquely nominated, the plurality of networks contained multiple PCGs (n = 24 or 80%). We also find that certain support members have a greater likelihood of being perceived by their network peers as PCGs than others. Controlling for personal factors and within-network covariates, we find that caregivers who are children of, are seen as making decisions on behalf of, and thought to spend enough time with, the affected person receive more PCG nominations from their peers. These findings support two of the mechanisms through which family members may acquire the PCG role discussed earlier; through filial piety (Piercy, 1998), quantity of time spent (Dwyer & Seccombe, 1991) and engagement in specific caregiving tasks (Joiling et al., 2013) with the affected individual. From a role-label theoretic, these findings suggest evidence of a shared mental model of who fulfills primary caregiving roles based in part on perceptions of specific task fulfillment by network members and in part based on knowledge of prior relationships to the care recipient. Importantly, by taking into account multiple-perspectives on whom fulfills primary caregiving roles in these networks we’ve more closely mapped the measurement and theoretical aspects of role labeling theory, which emphasizes how labels arise through the social process of consensus from the group (here, the caregiving network) that individuals fill a particular role. That there are, on average, more than one PCG in these networks implies that these tasks may be distributed among many caregivers rather than concentrated on a single individual. It is plausible that such a label “sticks” only as consensus builds within the network, and more individuals are perceived to be involved in these responsibilities by a greater number of observers. Future work should evaluate the extent to which individual actors adopt primary caregiving roles as a function of their peers’ building such a consensus. Interpreting these results in light of the finding that there are, on average, more than one uniquely identified PCG in caregiving networks, we conclude that the sole PCG model is not sufficient to capture the heterogeneity within networks and between individual caregivers. Rather, researchers should take care to cast a wide net—focusing on decision makers and being attuned to the personal characteristics of the affected person—when designing studies to investigate key caregiver roles in the context of ADRD. Moreover, interventions for alleviating caregiver burden may be better designed to take a network perspective, relying on multiple informants rather than solely on self-reports, to identify optimal targets. At the network level (level two in our model), few variance components were substantial enough to warrant discussion. Certainly, there was evidence of between network heterogeneity, and meaningful variance arising from differences in the gender and residency status of the person affected by ADRD. It is possible that the small level two variance components are the result of the fact that we have only 30 networks for comparison. This highlights one of the trade-offs and limitations affecting our study. On the one hand, there are few past studies with multiple whole-networks as their data, which we address here by collecting 30 multiple-informant networks at great cost (Marcum et al., 2017). However, on the other hand, this is still a small sample when the network is itself a source of variation in outcome, which constrains making inferences using multivariate analyses. As a result, we limit our discussion to the network-member level analysis. Another limitation of our study is that we focus on a population of people with ADRD that have access to formal care facilities. Certainly, many (if not most) of the ADRD population do not have this luxury and the caregiving networks—including the distribution of PCGs in them—may differ between individuals with and without access to formal care. However, our study highlights an under-examined aspect of the context of caregiving: different perspectives on the caregiving network arising from a mixture of informal and formal caregivers being present. Moreover, according to the Family Caregiver Alliance (2015), roughly a quarter of U.S. adults ages 65 and older received a combination of informal and formal care. On a typical day in 2012, there were 273,200 and 22,200 older adults being served at adult day centers and residential care home or similar residential care communities, respectively (Harris-Kojetin, Sengupta, Park-Lee, & Valverde, 2013). This is a sizeable population worthy of study. We note, however, that our results may reflect only the population of such networks residing in the greater Memphis area. More research is needed to evaluate the extent of generalizability and regional variation in these types of caregiving networks. We also recognize that such family caregiving plans may involve the preferences of and proximity to the care recipient. Pillemer & Suitor (2006) showed that mothers considered physical proximity and employment status of their children as they select who they wanted to become their PCG, suggesting that care recipients may prefer those who can spend enough time with them as their caregivers potentially illuminating the importance of high quality social interactions. As a limitation, we are unable to directly account for this effect in our study. We argue that this availability preference is manifest in our result that network members who were thought to spend enough time with the care recipient were more likely to be perceived as a PCG among informants. This likely reflects the perceived availability and ability of these network members to provide needed assistance and support. In addition to being seen as spending enough time with the care recipient, our results also point to the importance of decision-makers in the caregiving network. Controlling for the relationship between network members and the person affected by ADRD, network members who are seen to participate in care decision making are more often identified as PCGs by other network members. An increasing number of adults have comorbid illness and experience more complex medical conditions (Wolff & Kasper, 2006), therefore making decisions about medical treatment and care coordination increasingly important. It may be that those who are able and willing to make such decisions are seen as playing key roles within these familial caregiving networks, even if they do not involve themselves in day-to-day caregiving tasks per se. The potential importance of this role, making decisions related to coordinating care, in addition to the provision of direct care should be considered in future research and practice as social and economic contexts surrounding caregiving continue to shift in our society with shifting medical landscapes. Networks wherein multiple people are identified as serving PCG roles may also pose a challenge to legal and institutional precedents that establish compensatory and record-keeping mechanisms around a single individual who makes these important decisions. For example, a single individual needs to be identified for the purposes of insurance billing or granting power-of-attorney (Arling & McAuley, 1983; Grabowski, Stevenson, Huskamp, & Keating, 2006; Kapp, 2003), suggesting some reform may be needed to ease institutional dissonance between the multiple caregiver context of caregiving and the broader social system. Overall, our results suggest that there is heterogeneity in caregiving processes across families as well as within each caregiving network. This highlights the importance of fully assessing the social contexts surrounding caregiving to understand those social processes important in defining caregiving roles and expectations and to develop interventions that facilitate effective performance of the wide variety of tasks needed to support the care recipient. For example, although the vast majorities of the participating families had multiple PCG nominations, a minority of 20% had only one PCG nominated. Such families likely will need different types of interventions (e.g., facilitating member participation in caregiving or distribution of CG roles) compared to those with multiple PCGs (e.g., discussion of who fulfills what role/tasks). To inform such interventions, future research is needed to identify whether and how this variability in PCG nominations is related family adaptation to caregiver stress and caregiver well-being. For instance, it’s possible that a large number of caregivers being perceived to fill PCG roles in a single caregiving network may be the result of a lack of consensus among network members. This, in turn, may be an indicator of underlying vulnerabilities of negative outcomes in the network (such as family conflict or lack of cohesion). Finally, leveraging multiple informants from caregiving networks allows for the identification of shared perceptions about caregiving roles that particular network members may play (Koehly et al., 2015). In turn, this may help identify not only the core members of caregiving networks, but also the key characteristics of network members who would be optimal to mobilize in efforts to reinforce support systems to families touched by ADRD. For instance, intervention studies targeting PCGs should consider the possibility that more than one person in the caregiving network may need to be reached to achieve optimal outcomes. Funding This work was supported by the Intramural Research Program of the National Institutes of Health (grant number ZIA HG200395 to L.M.K.). Conflict of Interest We have no COIs to report. Acknowledgements We would like to thank the anonymous reviewers and associate editor Dr. Suitor as well as Drs. Deborah Carr, Judith Treas, and Jielu Lin for their thoughtful comments on previous drafts of this manuscript. C.S.M. conceived of the current study, performed the analyses, and wrote the article. S.A. and L.M.K. designed the original study, collected the data, and contributed to writing and revising the article. References Alzheimer’s Association . ( 2017 ). Alzheimer’s Disease facts and figures (Report) . Chicago, IL : Alzheimer’s Association . PubMed PubMed Arling , G. , & McAuley , W. J . ( 1983 ). The feasibility of public payments for family caregiving . The Gerontologist , 23 , 300 – 306 . doi:10.1093/geront/23.3.300 Google Scholar CrossRef Search ADS PubMed Ashida , S. , Marcum , C. S. , & Koehly , L. M . ( 2017 ). Unmet expectations in Alzheimer’s family caregiving: interactional characteristics associated with perceived under-contribution . The Gerontologist . doi: 10.1093/geront/gnx141 Barabási , A. L . ( 2009 ). Scale-free networks: a decade and beyond . Science (New York, N.Y.) , 325 , 412 – 413 . doi: 10.1126/science.1173299 Google Scholar CrossRef Search ADS PubMed Becker , H. S . ( 1973 ). Labelling theory reconsidered . In Becker H. S. (ed.), Outsiders: Studies in the Sociology of Deviance . New York: Free Press. Biddle , B. J . ( 1986 ). Recent development in role theory . Annual Review of Sociology , 12, 67 – 92 . doi:10.1146/annurev.so.12.080186.000435 Black , B. S. , Johnston , D. , Rabins , P. V. , Morrison , A. , Lyketsos , C. , & Samus , Q. M . ( 2013 ). Unmet needs of community-residing persons with dementia and their informal caregivers: findings from the maximizing independence at home study . Journal of the American Geriatrics Society , 61 , 2087 – 2095 . doi:10.1111/jgs.12549 Google Scholar CrossRef Search ADS PubMed Blau , P. M . ( 1977 ). A macrosociological theory of social structure . The American Journal of Sociology , 83 ( 1 ), 26 – 54 . doi:10.1086/226505 Google Scholar CrossRef Search ADS Browne , W. J. , Lahi , M. G. , & Parker , R. M . ( 2009 ). A guide to sample size calculations for random effect models via simulation and the MLPowSim software Package (Tech. Rep. No. 0-903024-96-9) . Bristol, UK : University of Bristol . Burnette , D . ( 1999 ). Social relationships of Latino grandparent caregivers: a role theory perspective . The Gerontologist , 39 , 49 – 58 . doi:10.1093/geront/39.1.49 Google Scholar CrossRef Search ADS PubMed Cahill , S. M . ( 1999 ). Caring in families: what motivates wives, daughters, and daughters-in-law to provide dementia care ? Journal of Family Studies , 5 , 235 – 247 . doi: 10.5172/jfs.5.2.235 Google Scholar CrossRef Search ADS Casciaro , T . ( 1998 ). Seeing things clearly: Social structure, personality, and accuracy in social network perception . Social Networks , 20 ( 4 ), 331 – 351 . doi:10.1016/S0378-8733(98)00008-2 Google Scholar CrossRef Search ADS Chadiha , L. A. , Feld , S. , & Rafferty , J . ( 2011 ). Likelihood of African American primary caregivers and care recipients receiving assistance from secondary caregivers: A rural-urban comparison . Journal of Applied Gerontology , 30 , 422 – 442 . doi: 10.1177/0733464810371099 Google Scholar CrossRef Search ADS PubMed Christiansen , C. L. , & Morris , C. N . ( 1997 ). Hierarchical Poisson regression modeling . Journal of the American Statistical Association , 92 ( 438 ), 618 – 632 . doi:10.1080/01621459.1997.10474013 Google Scholar CrossRef Search ADS Cornwell , B. , Marcum , C. , & Silverstein , M . ( 2015 ). The social network approach in gerontological research . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 70 , 87 – 90 . doi: 10.1093/geronb/gbu157 Google Scholar CrossRef Search ADS Dilworth-Anderson , P. , Williams , S. W. , & Cooper , T . ( 1999 ). Family caregiving to elderly African Americans: Caregiver types and structures . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 54 , S237 – S241 . doi:10.1093/geronb/54B.4.S237 Google Scholar CrossRef Search ADS Dwyer , J. W. , & Seccombe , K . ( 1991 ). Elder care as family labor: The influence of gender and family position . Journal of Family Studies , 12 ( 2 ), 229 – 247 . doi:10.1177/019251391012002006 Family Caregiver Alliance . ( 2015 ). Selected long-term care statistics (Report) . San Francisco, CA : National Center on Caregiving . Retrieved from: https://www.caregiver.org/selected-long-term-care-statistics Fitzpatrick , A. L. , Kuller , L. H. , Lopez , O. L. , Kawas , C. H. , & Jagust , W . ( 2005 ). Survival following dementia onset: Alzheimer’s disease and vascular dementia . Journal of the Neurological Sciences , 229-230 , 43 – 49 . doi: 10.1016/j.jns.2004.11.022 Google Scholar CrossRef Search ADS PubMed George , L. K . ( 1986 ). Caregiver burden: Conflict between norms of reciprocity and solidarity . In Pillemer K. A. & Wolf R. S. (Eds.), Elder abuse: Conflict in the family (pp. 67 – 92 ). New York : Auburn House . George , L. K. , & Gwyther , L. P . ( 1986 ). Caregiver well-being: a multidimensional examination of family caregivers of demented adults . The Gerontologist , 26 , 253 – 259 . doi:10.1093/geront/26.3.253 Google Scholar CrossRef Search ADS PubMed Gitlin , L. N. , & Wolff , J . ( 2011 ). Family involvement in care transitions of older adults: What do we know and where do we go from here ? Annual Review of Gerontology and Geriatrics , 31 ( 1 ), 31 – 64 . doi:10.1891/0198-8794.31.31 Google Scholar CrossRef Search ADS Goodman , L. A . ( 1961 ). Snowball sampling . The Annals of Mathematical Statistics , 32 ( 1 ), 148 – 170 . doi:10.1214/aoms/1177705148 Google Scholar CrossRef Search ADS Grabowski , D. C. , Stevenson , D. G. , Huskamp , H. A. , & Keating , N. L . ( 2006 ). The influence of Medicare home health payment incentives: Does payer source matter ? INQUIRY , 43 , 135 – 149 . doi: 10.5034/inquiryjrnl_43.2.135 Google Scholar PubMed Harris-Kojetin , L. , Sengupta , M. , Park-Lee , E. , & Valverde , R . ( 2013 ). Long-term care services in the United States: 2013 (Vital Health Statistics No. 3(37)) . San Francisco, CA : National Center for Health Statistics . Retrieved from: https://www.caregiver.org/selected-long-term-care-statistics Hayes , J. , Zimmerman , M. K. , & Boylstein , C . ( 2010 ). Responding to symptoms of Alzheimer’s disease: husbands, wives, and the gendered dynamics of recognition and disclosure . Qualitative Health Research , 20 , 1101 – 1115 . doi: 10.1177/1049732310369559 Google Scholar CrossRef Search ADS PubMed Heckathorn , D. D. , & Cameron , C. J . ( 2017 ). Network sampling: From snowball and multiplicity to respondent-driven sampling . Annual Review of Sociology , 43 , 101 – 119 . doi:10.1146/annurev-soc-060116-053556 Google Scholar CrossRef Search ADS Joiling , K. J. , Bosmans , J. E. , van Marwijk , H. W. , van der Horst , H. E. , Scheltens , P. , Vroomen , J. L. M. , & van Hout , H. P . ( 2013 ). The cost-effectiveness of a family meetings intervention to prevent depression and anxiety in family caregivers of patients with dementia: a randomized trial . Trials , 14 ( 1 ), 305 . doi: 10.1186/1745-6215-14-305 Google Scholar CrossRef Search ADS PubMed Kapp , M. B . ( 2003 ). Family caregivers’ legal concerns . Generations , 27 ( 4 ), 49 – 55 . Karlawish , J. H. , Casarett , D. , Klocinski , J. , & Clark , C. M . ( 2001 ). The relationship between caregivers’ global ratings of Alzheimer’s disease patients’ quality of life, disease severity, and the caregiving experience . Journal of the American Geriatrics Society , 49 , 1066 – 1070 . doi:10.1046/j.1532-5415.2001.49210.x Google Scholar CrossRef Search ADS PubMed Koehly , L. M . ( 2017 ). It’s interpersonal: Family relationships, genetic risk, and caregiving . The Gerontologist , 57 , 32 – 39 . doi: 10.1093/geront/gnw103 Google Scholar CrossRef Search ADS PubMed Koehly , L. M. , Ashida , S. , Schafer , E. J. , & Ludden , A . ( 2015 ). Caregiving networks-using a network approach to identify missed opportunities . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 70 , 143 – 154 . doi: 10.1093/geronb/gbu111 Google Scholar CrossRef Search ADS Langa , K. M. , Larson , E. B. , Crimmins , E. M. , Faul , J. D. , Levine , D. A. , Kabeto , M. U. & Weir , D. R . ( 2017 ). A comparison of the prevalence of dementia in the United States in 2000 and 2012 . JAMA: Internal Medicine , 177 ( 1 ), 51 – 58 . doi: 10.1001/jamainternmed.2016.6807 Google Scholar CrossRef Search ADS PubMed Marcum , C. S . ( 2013 ). Age differences in daily social activities . Research on Aging , 35 , 612 – 640 . doi: 10.1177/0164027512453468 Google Scholar CrossRef Search ADS PubMed Marcum , C. S. , Lienert , J. , Goldring , M. , Lin , J. , Miggins , A. , Moss , M. , … Koehly , L . ( 2017 ). Ego-centered cognitive social structures of close personal networks in the United States . Unpublished preprint retrieved from: osf.io/4ugk6 McCullagh , P. , & Nelder , J. A . ( 1989 ). Generalized linear models, no. 37 in Monograph on statistics and applied probability . ( 2nd Ed .) Chapman & Hall : London . doi:10.1007/978-1- 4899-3242-6 Merrill , D. M . ( 1996 ). Conflict and cooperation among adult siblings during the transition to the role of filial caregiver . Journal of Social and Personal Relationships , 13 ( 3 ), 399 – 413 . doi: 10.1177/0265407596133006 Google Scholar CrossRef Search ADS Mittelman , M. S. , Ferris , S. H. , Shulman , E. , Steinberg , G. , & Levin , B . ( 1996 ). A family intervention to delay nursing home placement of patients with Alzheimer disease. A randomized controlled trial . JAMA , 276 , 1725 – 1731 . doi:10.1001/jama.1996.03540210033030 Google Scholar CrossRef Search ADS PubMed Mui , A. C . ( 1992 ). Caregiver strain among black and white daughter caregivers: a role theory perspective . The Gerontologist , 32 , 203 – 212 . doi:10.1093/geront/32.2.203 Google Scholar CrossRef Search ADS PubMed National Alliance for Caregiving, & AARP . ( 2009 ). Caregiving in the U.S. 2009 (Report) . Washington, DC : National Alliance for Caregiving . Retrieved from http://www.caregiving.org/data/Caregiving_in_the_US_2009_full_report.pdf Neubauer , S. , Holle , R. , Menn , P. , Grossfeld-Schmitz , M. , & Graesel , E . ( 2008 ). Measurement of informal care time in a study of patients with dementia . International Psychogeriatrics , 20 , 1160 – 1176 . doi: 10.1017/S1041610208007564 Google Scholar CrossRef Search ADS PubMed Norton , M. C. , Piercy , K. W. , Rabins , P. V. , Green , R. C. , Breitner , J. C. , Ostbye , T. ,… Tschanz , J. T . ( 2009 ). Caregiver-recipient closeness and symptom progression in Alzheimer disease. The Cache County Dementia Progression Study . The Journals of Gerontology Series B: Psychological Sciences and Social Sciences , 64 , 560 – 568 . doi: 10.1093/geronb/gbp052 Google Scholar CrossRef Search ADS O’Connor , D. L . ( 2006 ). Self-identifying as a caregiver: exploring the positioning process . Journal of Aging Studies , 21 ( 2 ), 165 – 174 . doi: 10.1016/j.jaging.2006.06.002 Google Scholar CrossRef Search ADS Ott , B. R. , Tate , C. A. , Gordon , N. M. , & Heindel , W. C . ( 1996 ). Gender differences in the behavioral manifestations of Alzheimer’s disease . Journal of the American Geriatrics Society , 44 , 583 – 587 . doi:10.1111/j.1532-5415.1996.tb01447.x Google Scholar CrossRef Search ADS PubMed Pearlin , L. I. , & Aneshensel , C. S . ( 1994 ). Caregiving: The unexpected career . Social Justice Research , 7 ( 4 ), 373 – 390 . doi:10.1007/BF02334863 Google Scholar CrossRef Search ADS Piercy , K. W . ( 1998 ). Theorizing about family caregiving: The role of responsibility . Journal of Marriage and the Family , 60 ( 1 ), 109 – 118 . doi:10.2307/353445 Google Scholar CrossRef Search ADS Pillemer , K. , & Suitor , J. J . ( 2006 ). Making choices: a within-family study of caregiver selection . The Gerontologist , 46 , 439 – 448 . doi:10.1093/geront/46.4.439 Google Scholar CrossRef Search ADS PubMed Schulz , R. , & Martire , L. M . ( 2004 ). Family caregiving of persons with dementia: prevalence, health effects, and support strategies . The American Journal of Geriatric Psychiatry , 12 , 240 – 249 . doi:10.1097/00019442-200405000-00002 Google Scholar CrossRef Search ADS PubMed Schulz , R. , & Tompkins , C. A . ( 2010 ). Informal caregivers in the United States: Prevalence, caregiver characteristics, and ability to provide care . In Olson S. (Ed.), The role of human factors in home health care (pp. 117 – 144 ). Washington, DC : National Academies Press . Schumacher , K. L . ( 1995 ). Family caregiver role acquisition: role-making through situated interaction . Scholarly Inquiry for Nursing Practice , 9 , 211 – 26; discussion 227 . Google Scholar PubMed Stoller , E. P. , & Pugliesi , K. L . ( 1989 ). Other roles of caregivers: Competing responsibilities or supportive resources . Journal of Gerontology , 44 , S231 – S238 . doi:10.1093/geronj/44.6.S231 Google Scholar CrossRef Search ADS PubMed Stommel , M. , Given , B. A. , Given , C. W. , & Collins , C . ( 1995 ). The impact of the frequency of care activities on the division of labor between primary caregivers and other care providers . Research on Aging , 17 ( 4 ), 412 – 433 . doi: 10.1177/0164027595174003 Google Scholar CrossRef Search ADS Strogatz , S. H . ( 2001 ). Exploring complex networks . Nature , 410 , 268 – 276 . doi: 10.1038/35065725 Google Scholar CrossRef Search ADS PubMed Wasserman , S. , & Faust , K . ( 1994 ). Social network analysis: Methods and applications . Cambridge : Cambridge University Press . Google Scholar CrossRef Search ADS Wolff , J. L. , & Kasper , J. D . ( 2006 ). Caregivers of frail elders: updating a national profile . The Gerontologist , 46 , 344 – 356 . doi:10.1093/geront/46.3.344 Google Scholar CrossRef Search ADS PubMed Wolff , J. L. , Spillman , B. C. , Freedman , V. A. , & Kasper , J. D . ( 2016 ). A national profile of family and unpaid caregivers who assist older adults with health care activities . JAMA Internal Medicine , 176 , 372 – 379 . doi: 10.1001/jamainternmed.2015.7664 Google Scholar CrossRef Search ADS PubMed Zimmerman , S. , Sloane , P. D. , Williams , C. S. , Reed , P. S. , Preisser , J. S. , Eckert , J. K. ,… Dobbs , D . ( 2005 ). Dementia care and quality of life in assisted living and nursing homes . The Gerontologist , 45 , 133 – 146 . doi:10.1093/geront/45.suppl_1.133 Google Scholar CrossRef Search ADS PubMed Published by Oxford University Press on behalf of The Gerontological Society of America 2017. This work is written by (a) US Government employee(s) and is in the public domain in the US.

Journal

The Journals of Gerontology Series B: Psychological Sciences and Social SciencesOxford University Press

Published: Jan 3, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off