Assessment of multiple constructs of social integration for older adults living in nursing homes

ABSTRACT A variety of terms and measures have been used in the literature to denote being socially integrated, and many studies of older adults focus on only social networks or social support and often only include those living in the community. The purpose of this study was to assess multiple constructs of social integration (i.e., social networks, social capital, social support, and social engagement) for older adults in nursing homes. Data were collected from 140 older adults at 30 nursing homes in Kansas. We interviewed older adults’ in-person using a survey questionnaire, and used multilevel confirmatory factor analysis to analyze the data. The final model that included the four constructs had acceptable fit (χ2 = 174.71; df = 112; p < .01; CFI = .93; RMSEA = .06; SRMR-W = .06; SRMR-B = .12). The results showed that the proposed model was supported at the individual level. At the between-level, social networks and social support were supported. Study results have methodological and practice/policy implications for the study of older adults in long term care settings. In particular, this study contributes to understanding how to operationally define and differentiate social integration variables in studies of older adults, particularly when study data are hierarchical.

participation, can help to determine an individual's level of social integration .
Various terms and measures have been used in the literature to denote being socially integrated. Many studies of social integration have focused only on older adults living in the community (see Holt-Lunstad, Smith, & Layton, 2010) or only on one or two aspects of social integration of nursing home residents, such as social networks and/or social support (e.g. Abott & Pachucki, 2016;Casey et al., 2016). For example, the Lubben Social Network Scale (LSNS-6) is a widely used, valid, and reliable measure that simultaneously assesses social networks and social support and screens for isolation among older adults; however, this measure does not differentiate social networks from social support and was designed for older adults living in the community (Lubben et al., 2006). A more holistic understanding of social integration of nursing home residents is needed in order for social workers to identify opportunities to increase the social integration of residents.
The purpose of this study was to examine the measurement structure of social networks, social capital, social support, and social engagement of older adults in nursing homes. The intent of this research is to help distinguish these variables from one another, help eliminate confusion in the literature about how to define and measure the variables, and help establish consistency in how various researchers discuss and measures these variables. This work can help guide future research efforts to develop, implement, and replicate interventions to increase the social integration of nursing home residents. This paper is part of a larger study presented in Leedahl, Chapin, and Little (2015) that examined relationships between social integration, cultural change practices, the role of social workers, and physical and mental health for older adults living in nursing homes.

Social integration for older adults living in nursing homes
Nursing home residents face significant barriers to social integration because "historically nursing homes have been cut off from the wider world by both institutional walls and societal segregation" (Anderson & Dabelko-Schoeny, 2010;p. 274;Goffman, 1961). According to Victor, Scambler, and Bond (2009), older adults living in facility settings and those who struggle with mobility often experience a sense of geographical social isolation, and social workers working in long-term care help to ensure quality of life by providing psychosocial care for consumers through physical, psychological, and social interventions as well as family support, with the goal of promoting optimal levels of psychological, physical, and social functioning. Increasing the social integration of older adults who utilize long-term care services, who are some of the most vulnerable in their communities, is of vital importance for social workers in working to fulfill their commitment to enhance human well-being and help meet the basic human needs of all people (NASW, 2008). A clearer understanding of the concepts related to social integration (social networks, social capital, social support, and social engagement) in long-term care facilities can help social workers and administrators to develop more targeted programs and interventions.
In the following segment, we review information related to four key concepts of social integration (i.e., social networks, social capital, social support, and social engagement), how each concept is connected to health outcomes, and why each concept is important when assessing nursing home residents.

Social networks
Berkman and colleagues (2000) introduced a model for social network theory, and in doing so discussed that studies often used the terms social networks, social support, social ties, and social integration interchangeably, which can be confusing. To help clarify terms,  conceptually defined social networks as a "web of social relationships that surround an individual and the characteristics of those ties" (p. 847). Further, in a review by Smith and Christakis (2008), social network research is recognized as broad and as studying webs of social relations in contrast to the more common, social support research, which assesses the quality of a person's social relations. Operationally, research studies often use numbers of close friends and relatives, marital status, or affiliation or membership in religious and voluntary associations to measure one's social network (e.g., Berkman, 1995;Cohen, 1988;House, Landis, & Umberson, 1988). Research shows that social networks positively influence cognitive and emotional states, including self-esteem, social competence, self-efficacy, depression, and affect (Smith & Christakis, 2008). For nursing home residents, research suggests that 1) having a large social network is associated with a higher quality of life, particularly for residents with dementia (Abbott & Pachucki, 2016), 2) non-familial friendship networks have been associated with an increased risk of social isolation for nursing home residents (Casey et al., 2016), and 3) the quality of the social network is also important for the wellbeing of residents (Bergland & Kirkevold, 2006;Casey et al., 2016). communication for mutual benefit" (p. 67). More recently, Ferlander (2007) defined social capital using structural and cognitive aspects (i.e., social networks, norms of reciprocity, and trust), stating that social networks are the core element, but that the networks break down without reciprocal norms and trust. The most common measurement indicators for social capital, particularly for health studies, have been membership in voluntary associations and generalized social trust (Ferlander, 2007). Research suggests that social capital is linked to better outcomes related to morbidity/mortality, selfrated health, mental health, and health behaviors, such as reduced smoking, increased physical activity, better diet, less disease, and improved survival when ill (Ferlander, 2007;Forsman, Herberts, Nyqvist, Wahlbeck, & Schierenbeck, 2012;Forsman, Nyqvist, & Wahlbeck, 2011). For older adults, particularly those in long term care settings, social capital can help older adults access emotional and practical support. Similarly, for nursing home residents, Leedahl and colleagues (2015) found that greater social capital was associated with increased mental health and functional health and well-being for nursing home residents.

Social support
One of the most commonly used social integration terms is social support. Social support consists of emotional, social, physical, and financial resources as well as other types of care provided by others . Callaghan and Morrisey (1993) conceptually defined social support as "an exchange of resources between at least two individuals perceived by the provider or recipient to be intended to enhance the well-being of the recipient" (p. 203). As such, social support is considered the primary pathway by which social networks influence physical and mental health status (Berkman, Glass, Brissette, & Seeman, 2000). Social support often encompasses the qualitative aspects of social relations and generally includes emotional, instrumental, appraisal, and informational support . Specific measures of social support can include structural features (size and composition of network), frequency of interactions, content and quality of support, and perceptions of adequacy (Siebert, Mutran, & Reitzes, 1999). Having adequate social support has been shown to increase positive outcomes for older adults specifically in the areas of long-term care, health, community participation, security, and sustained independent living (Antonucci & Akiyama, 1987;Drageset et al., 2015;World Health Organizations, 2002). Research suggests that both giving and receiving social support is important for the well-being of older adults (Thomas, 2009). In a study of nursing home residents with cancer, Drageset et al. (2015) found that social support that reinforces self-esteem and worth can reduce loneliness.

Social engagement
Social engagement refers to the enactment of potential ties in real life activities within the communities in which people live and is defined as "performance of meaningful social roles for either leisure or productive activity" (Glass, Mendes De Leon, Bassuk, & Berkman, 2006, p. 606). Social engagement has been conceptualized to include productive activities (i.e., activities that generate goods or services for an economic value, such as preparing meals, completing volunteer work, or having paid employment) and social or leisure activities (i.e., activities that involve talking with others or taking part in activities with others that are enjoyable, such as going to a movie or playing cards). Measures of social engagement often include a single summary index that encompasses multiple aspects of engagement without differentiating between the different aspects (i.e., productive versus social) (Mendes de Leon, Glass, & Berkman, 2003). Social engagement has been shown to influence health, namely mortality, disability, and health care expenditures (Kiely & Flacker, 2003;Mendes de Leon et al., 2003). Research suggests that residents enjoy both structured activities and unstructured socializing (Forsam, et al., 2012;Nakrem, Vinsnes, Harkless, Paulesen, & Seim, 2013;Roberts & Bower, 2015). For nursing home residents, low social engagement at admission, limitations in physical and cognitive functioning, and vision or communication difficulties can lead to low engagement overtime (Bliss et al., 2017).
In conclusion, the terms social networks, social capital, social support, and social engagement, each have distinct features related to social integration, yet all relate to one another with some overlapping definitions. Additionally, each of these concepts has been shown to predict or influence various physical and mental health outcomes in multiple studies that include comparison groups and longitudinal design, and each are important when assessing social integration for nursing home residents.

Research questions
This study works to fill a gap in the literature by helping to conceptualize and operationalize these multiple aspects of social integration (i.e., social capital, social networks, social support, and social engagement), using well-conceptualized definitions and operational measures, for studying the lives of older adults. In this study, we build on the measurement literature related to assessing various aspects of the social worlds of older adults by asking two specific questions: (1) Do the data support the proposed multilevel model differentiating the multiple constructs and indicators of social integration for older adults in nursing homes?
(2) Which indicators (i.e., measures) explained the most meaning (i.e., had the highest factor loadings and R 2 values) for each of the latent variables?
This could help differentiate these variables and determine what particular aspects of social integration could be targeted in interventions for improving physical and mental health of older adults in nursing homes.

Data & sample
This study utilized a two-stage, multilevel random sampling technique to obtain a sample of 140 older adult residents from 30 nursing homes in Kansas during 2011-2012. Using random sampling, nursing homes were recruited through letters and follow-up phone calls. If administrators agreed to participate, we then contacted the social service director who assisted us with identifying nursing home residents who could participate in the study. Residents were also identified using random sampling. To do this, the social service director was provided a list of the four inclusion criteria (i.e., 65 years of age or older, lived in the nursing home for at least four months, did not have a legal guardian, did not have moderate to severe cognitive impairment), identified residents who were eligible, gave them identification (ID) numbers (#s), and then worked with us to randomly select participants. Informed consent was obtained from every resident included in the study, and extra precautions were taken to ensure residents consented to participation prior to interviewers entering the nursing home. Survey data were collected through in-person interviews, and three trained interviewers conducted the interviews; each resident was provided $20 cash or a gift card. We used a planned missing data design as described by Graham, Taylor, Olchowski, & Cumsille (2006) to ensure we could use the full set of questionnaire items while also reducing respondent burden and keeping the interviews to no longer than one hour. The approach involved collecting cross-sectional data and analyzing it at the individual level (i.e., within level) and group level (i.e., between level). See Leedahl (2017) for the full details of the sampling, data collection, and missing data procedures used to gather data in this IRB-approved study.

Measures
The survey forms were developed using multiple standardized measures for this study. We first pilot tested the questions to ensure respondents could answer the questions without difficulty and to verify that interviews would last no more than one hour. This also provided us with an opportunity to assess our measures for construct validity, as we intentionally asked nursing home staff and residents about the questions as they related to the study concepts. Based on these conversations, some slight wording changes and examples were added to the measures to ensure construct validity. Interviewers asked participants questions from the survey, and recorded responses. The latent variables for this study included social networks, social capital, social support, and social engagement. See Table 1 for a summary of the latent variables used in this study along with their conceptual definitions and indicators.

Social networks
Social network characteristics (i.e., size, frequency of contact, proximity) were measured using the concentric circle (i.e., egocentric network) approach (Antonucci & Akiyama, 1987;Ashida & Heaney, 2008). Following the interview, interviewers calculated totals for: size = number of total people; proximity = number of people within the nursing home & within 1-hour drive; and frequency = number of people with at least once a week contact. As Table 1. Study latent variables, conceptual definitions, and indicators

Latent Variables Conceptual Definition Indicators Social Networks
Web of social relationships that surround an individual and the characteristics of those ties (1) Size-# total people (e.g., family, friends, staff, residents) (2) Frequency-# with frequent contact (at least once/week) (3) Proximity-# live close by or live/work in the nursing home Social Capital Features of social organization that facilitate coordination and communication for mutual benefit (4) Level of reciprocity (i.e., belief that people help each other out) (5) Level of trust in others (e.g., people in the community, residents, staff in the facility, people in your family) (6) # of religious/professional/community/social groups (Social groups) Social Support Exchange of resources between at least two individuals perceived by the provider or the recipient to be intended to enhance the well-being of the recipient (7) Informational (e.g., getting advice) (8) Tangible (e.g., getting gifts or other goods) (9) Emotional (e.g., getting hug, talking about concerns) (10) Provided (e.g., giving advice to others) Social Engagement Taking part in real life activities with others within the communities people live Frequency of Participation: (11) Formal Activities (e.g., day trips, playing games, music, church) (12) Socializing (e.g., talking on phone, visiting) (13) Civic or Productive Activities (e.g., helping in dining area, being with other residents in need) previous research has shown the importance of within facility relationships for nursing home residents (Chou, Boldy, & Lee, 2002;McGilton & Boscart, 2007), the count of within facility network members was used to enhance the proximity measure for this population.

Social capital
Social capital was measured using three indicators: social groups, norms of reciprocity, and trust, based on a study by Narayan and Cassidy (2001). For social groups, residents were asked: how many groups or organizations do you belong to? We encouraged them to name religious/professional/community groups, social clubs, resident groups, or just groups of people who they got together or corresponded with regularly. For norms of reciprocity, residents were asked one question about whether they think people mostly look out for themselves or try to be helpful, and one question about whether they think people try to take advantage of others or be fair. Trust was assessed using ten Likert scale questions about how much trust they had for different groups of people (e.g., people in your community, residents in the facility, staff in the facility, people in your family, local government, judges/police). Previous research has shown these measures to be largely stable and consistent across data sets and demonstrably reliable and valid, particularly for community-dwelling older adults (Norstrand & Xu, 2011). In this study, internal consistency for social capital was quite high (α = .79).
Social support Social support was measured using a modified version of the Krause and Markides (1990) version of the widely used Inventory of Socially Supportive Behaviors (ISSB) (Barrera, Sandler, & Ramsay, 1981). The Krause and Markides (1990) version uses Likert scale questions to generate subscale scores for informational, tangible, emotional, and provided support. In previous research, reported internal consistency estimates ranged from 0.67 to 0.83 (Krause & Shaw, 2002). In this study, questions tailored to those only living in the community were eliminated. For the tangible support questions, we eliminated five of the nine questions because they did not pertain to those living in a nursing home. The five questions asked about people providing them with a place to stay, watching over possessions while they were away, providing a place to stay overnight, and looking after a family member while being away. For the provided support questions, we eliminated a question about other people depending on the person for transportation. In this study, internal consistency for social support was high (α = .92).

Social engagement
Social engagement was measured using Likert scale questions about participation in various social activities within and outside the nursing home. The questions tapped into whether or not residents participated as well as the frequency of participation. The questions were derived from previous work (e.g., Glass et al., 2006;Jang, Mortimer, Haley, & Borenstein Graves, 2004;Mitchell & Kemp, 2000) and tailored to include activities pertinent to nursing homes. Based on the literature and conversations with nursing home social services staff, we identified three indicators: productive/civic, formal activities, and socializing. These indicators assessed the prevalence and importance of productive/civic engagement for older adults in nursing homes and differentiated between informal socializing (i.e., having conversations) versus participating in more formal activities facilitated by nursing home staff that involved mental stimulation and active involvement (e.g., playing a game, singing, doing art work). We screened the item responses to ensure they conformed to the expected pattern. Internal consistency for social engagement in this study was acceptable (α = .65).

Covariates
The covariates for the study included: activities of daily living (ADLs), cognitive status, socioeconomic status (SES), and self-reported health status. The Katz Index of Independence in Activities of Daily Living was used to assess ADLs (Wallace & Shelkey, 2008). Inclusion criteria eliminated those with moderate to severe cognitive impairments. However, to control for mild cognitive impairment, a simplified and non-invasive test called the 6 Item Cognitive Impairment Test (6CIT) was used (Brook & Bullock, 1999). Years of education was collected as a proxy for SES (Lee, Paultre, & Mosca, 2005). Finally, self-reported health status was measured by asking the question, "In general, would you say your health is excellent, very good, good, fair, or poor."

Data analysis
This study utilized multilevel confirmatory factor analysis (MCFA) to analyze the data. Prior to conducting MCFA analysis, a number of steps were performed, including data assessment and cleaning (before and after data imputation) and missing data analysis using multiple imputation. We used a combination of SPSS Statistics version 20 and Mplus version 7 for quantitative analyses. CFA is a statistical technique used to assess measures and test relationships between measures and latent variables, and researchers must have a priori knowledge (e.g., past research, theory) about the indicators and latent variables and their patterns of relationships (Brown, 2006). The value of the standardized factor loadings of each indicator on the latent variables is used to interpret latent variables in CFA. Indicators with the highest standardized factor loadings, or correlation, with the factor explain the meaning of the latent variables.
Because the data for this study have a hierarchical structure, meaning the residents are nested within nursing homes, we analyzed the data using multilevel analysis in order to account for the between group influence. As discussed by Heck (2001), fitting multilevel data structures requires multiple steps to work including: a) fitting the single-level confirmatory factor analysis (CFA) model; b) examining the intraclass correlation coefficients to determine if data should be examined at multiple levels; c) fitting the MCFA model. This study followed these steps to answer the research questions.
The intraclass correlation coefficient (ICC) is a statistical measure of the proportion of variance among groups, and higher values indicate that the independence of errors is violated and that grouping matters and should not be ignored in analyses (Tabachnick & Fidell, 2007). When ICCs are less than 0.05, there is little need to perform multilevel analyses. When the ICCs are considered high (range of .10 to .25), this suggests considerable similarities across groups (Heck, 2001). We examined ICCs for each indicator to identify which latent variables should be included in the between-level model.
As stated by Brown (2006), "often a CFA model will need to be revised. . . to improve the fit of the model" (p. 157). In this study, we were able to determine acceptable model fit for the measurement model using established statistical techniques discussed in the literature. To fit the final MCFA model, we included three correlated residuals to help improve model fit based on our assessment of modifications indices provided by Mplus. Because the various questions for the sub-scales and scales used in the study had different metrics (e.g., 0-5 Likert scales vs. 0-3 Likert scales), after analyzing descriptive statistics, we standardized all variables by scaling them between 0 and 1 to ensure they were on a similar metric. Finally, because the histograms indicated some concerns regarding normality for the social networks variable, we used robust maximum likelihood estimation rather than transforming the indicators, as this would have made the data harder to interpret (Yuan, Chan, & Bentler, 2000).

Fit statistics
In this study, model fit was evaluated by examining the model chi-square (χ 2 ), the normed chi-square (χ 2 /df), the root mean square of approximation (RMSEA), comparative fix index (CFI), and the standardized root mean square residual (SRMR) (Kline, 2005). The literature suggests looking at multiple fit indices to determine overall fit of the model, rather than relying on one statistic in making decisions because each of these fit indices have their inherent limitations. χ 2 by itself is often not helpful, particularly in complicated models with small sample and cluster sizes, because it is somewhat unrealistic to expect a model to be "perfect population fit." When sample sizes are not large, it can be helpful to examine the normed chi-square, calculated by dividing the χ 2 value by the degrees of freedom (df). The "rule of thumb" is that a ratio of less than 2.0 indicates a good-fitting model, though values of 3.0 or 5.0 can indicate reasonable fit (Kline, 2005). RMSEA values less than or equal to .05 indicate close approximate fit, values between .05 and .08 indicate acceptable or reasonable error of approximation, and values over .10 suggest poor fit (Browne & Cudeck, 1993;Kline, 2005). SRMR values less than .10 are considered acceptable model fit (Kline, 2005). CFI values greater than roughly .90 indicate acceptable model fit (Kline, 2005).

Resident demographics
The mean age for the entire sample was 83.07 (SD = 9.02). The sample was 74.3% female and 25.7% male, and most participants identified as White (92.7%). See Table 2 for a summary of basic descriptive information for the nursing home resident participants.

Nursing home demographics
For the 30 nursing homes that participated in the study, the mean number of licensed beds (bed size) was 106.63, and ranged rather considerably from 46 to 269 beds. The average bed size in Kansas at the time was 82. The mean number of deficiencies for participating nursing facilities was 12.5. For Medicare Star Ratings, which are based on health inspection, staffing, and quality measures ratings, at the time of the study, 13.3% were rated much above average, 30% above average, 13.3% average, 30% below average, and 13.3% far below average. Most nursing homes (83%) were located in semiurban or urban counties, though 15% of nursing homes were located in rural counties. Most nursing homes (93.3%) accepted Medicaid residents, and most (66.7%) were for-profit nursing homes, with 33.3% being either nonprofit or government-owned. Over half (56.7%) were affiliated with large chains, and most (76.7%) were not Continuing Care Retirement Communities (CCRCs). Table 3 includes descriptive statistics for participating nursing homes.

Multilevel confirmatory factor analysis model (research question 1)
In this study, the ICCs ranged from 0.04-0.24, with most indicators having small effect sizes. This suggests a need to examine the data using multilevel analyses for some variables but that group level variance may be difficult to detect for certain variables. Because the indicators for social groups, norms of reciprocity, trust, productive/civic, socializing, and activity participation had ICCs lower than 0.10, we determined that these variables would not have enough group-level variance for testing them on the between-level model. Therefore, only social networks and social support were included as latent variables for the between-level model because the indicators for size, frequency, proximity, informational, tangible, emotional, and provided had ICCs above 0.10. For the within-level model, the only measurement modification made involved the social capital construct. Specifically, the parameter estimate for the social groups indicator of social capital was not significant and did not highly correlate with the other indicators (trust and norms of reciprocity). Therefore, we included social groups as an observed variable in the model rather than as an indicator of social capital (Brown, 2006). See Figure 1 for pictorial representation of the final MCFA model. The final MCFA model had overall acceptable fit based on χ 2 /df, CFI, RMSEA, and SRMR (χ2 = 174.71, df = 112, p < .01; χ2/df = 1.56; CFI = .93; RMSEA = .06; SRMR-W = .06; SRMR-B = .12). Considering the complexity of the model and its χ 2 /df, CFI, RMSEA, and SRMR scores, the final model appears to offer a reasonably close fit to the data. Therefore, the constructs and indicators are supported as proposed, with one modification made to the social capital construct. See Table 4 for the relationships between indicators and their respective constructs. The amount of variance in each indicator that was accounted for by its latent construct (i.e., standardized loadings) ranged from 0.50 to 0.96. Some of these values are lower than the ideal standardized factor loadings of .70 or higher, but all values are higher than the cut-off value of .30. Floyd and Widaman (1995) recommend removing indicators if the standardized factor loadings fall below .30. The influence of the covariates on each of the social integration constructs was non-significant, 1 thus activities of daily living, cognitive status, and SES did not explain a significant amount of the variance in these constructs.

Best indicators for each latent variable (research question 2)
We examined the factor loadings to determine which indicator had the greatest amount of variance accounted for by each of the latent variables, basically working to identify "how good" each indicator is at measuring the latent variable. For the social network latent variable, social networks explained 85.8% of the variance in proximity, 79.1% of the variance in frequency, and 67.8% of the variance in size. Therefore, proximity was the best indicator of social networks for older adults living in nursing homes in this study. For the latent variable social capital, social capital explained 78.3% of the variance in social trust and explained 25.3% of the variance in norms of reciprocity; therefore, social trust was the much better indicator of social capital in this sample. For social support, social support explained 91.4% of the variance in emotional support, 40.5% of the variance in provided support, 36.2% of the variance in tangible support, and 33.8% of the variance in informational support. This shows that emotional support was the best indicator of social support. In examining the social engagement latent variable, social engagement explained the greatest amount of variance in participating in activities (58.6%) compared to 32.3% of the variance in socializing, and 29.2% of the variance in productive/civic activities. Activity participation was the best indicator for social engagement in this sample. Comrey and Lee (1992) suggest interpreting indicators such that more than 50% overlapping variance is considered excellent, 40% very good, 30% good, 20% fair, and 10% poor. Therefore, in this study, proximity, frequency of contact, size of network, social trust, emotional support and participating in nursing home activities all fell within the excellent category of having more than 50% overlapping variance.

Discussion
The final model had acceptable fit based on the fit statistics, the significance of the factor loadings, and assessment of measurement error. The results showed that the proposed model that included all constructs was supported at the individual-level, with a modification needed for the social capital construct. At the between-level, social networks and social support were supported, showing that social networks and social support varied based on nursing home in this study whereas social engagement and social capital only had individual-level variation. Similar to previous studies of social networks showing that social relationships with others are an important component of quality of life for those living in nursing homes, but that new friendships can be difficult to develop (Casey et al., 2016), results from our study suggest that some nursing homes are better than others in helping residents to grow their social networks. While other studies of social networks of nursing home residents have explored the size, frequency of contact, reciprocation, and density (e.g., Abbott & Pachucki, 2016), our study focused on proximity, frequency, and size. Our results indicate that proximity is particularly important for nursing home residents, meaning it is important for residents to have family and friends close by who can visit on a regular basis and to gain new relationships with staff and residents when living in a nursing home. For those who do not have family or friends close by, building new relationships with residents and staff (or even family members of other residents) within the nursing home is particularly important. Consistent with previous work on social support (e.g. Casey et al., 2016), our results suggest that it is not only the size of the networks that matter, but also the quality of the support received. In this study, emotional support was particularly important for measuring social support among nursing home residents, more so than informational, tangible, or provided support. Interestingly, Malench (2004) conducted a study about activities performed by family members of those in nursing homes. The most common activity reported was bringing gifts, and importantly, facilities with social workers were more likely to offer family supports and to distribute newsletters indicating facility events in order to encourage family participation. Based on the results of this study as related to social work practice, social workers could aid families in providing emotional support and taking part in meaningful activities with their family members.
For example, social workers could help educate family members about strategies to help cope with challenges that often arise from moving a loved one to a nursing home. As family involvement tends to decrease the longer a person stays in the nursing home (Yamamotot-Mitani, Aneshensel, & Lvey-Storms, 2002), social workers can collaborate with family members to identify barriers to continued visits and possible solutions. In addition, social workers could work with family members to design programs and activities that could be taught or led by the family members. For example, a family member who enjoyed photography could work with their loved one and other residents to take pictures and create an art exhibit.
In addition, similar to work by Forsman et al. (2011), our results indicated that trust is an important aspect of a resident's social capital. Qualitative studies (e.g. Nakrem et al., 2013;Roberts & Bowers, 2015) suggest that residents enjoy participating in both structured and unstructured social activities. Our results support these findings and indicate that participating in structured activities, such as group-led art activities, group-led exercise groups, and planned intergenerational activities (e.g., children's musical offerings, children or young adults coming to the nursing home as part of daycare/school activities or volunteer work) may be particularly important for the well-being of nursing home residents compared to unstructured activities, such as spontaneous conversations.
Regarding social engagement, this study found that socializing, nursing home activity participation, and participation in productive/civic activities all contributed to the overall measurement of social engagement in this study, but that level of nursing home activity participation was the best indicator of social engagement. According to Casey et al. (2016), most residents attend activities with the assistance of staff members. Therefore, ensuring that nursing home staff have time to help residents attend and appreciate activities is important to ensuring social engagement for residents. Recognizing the ongoing struggles within nursing homes to maintain an adequate, competent direct care workforce (Lehning & Austin, 2010), further developing volunteer models, such as the "super" volunteer model championed by Falkowski (2016), may be a viable solution for supporting social engagement among residents. Trained volunteers could help to provide support to direct care staff and provide assistance with non-technical tasks, such as identifying meaningful social engagement activities for each resident. The importance of this work is also underscored by findings of a recent large-scale study commissioned by the Centers for Medicare and Medicaid Services in which 160 residents in 40 nursing homes were interviewed about quality of life issues (Allen, 2011). Residents listed "choice of activities" and "activities that amount to something," such as activities that produce or teach something; activities using skills from residents' former work, religious activities, and activities that contribute to the nursing home, as priorities for ensuring dignity. Interestingly, residents rarely mentioned participating in activities as a way to just "keep busy" or just to socialize, stating instead that they "wanted a variety of activities, including those that are not childish, require thinking (such as word games), are gender-specific, produce something useful, relate to previous work of residents, allow for socializing with visitors and participating in community events, and are physically active" (p. 54). Lastly, researchers noted that these ideas were relevant to both interviewable and non-interviewable residents, stating that non-interviewable residents appeared "happier" and "less agitated" in homes with many individualized planned activities for them.
This study also makes an important contribution to measurement literature on the various social variables. By testing multiple constructs of social integration (i.e., social networks, social capital, social support, and social engagement) with multiple indicators for each construct in one study, rather than focusing on one construct or even one indicator of a construct, this work helps to distinguish these variables from one another. This helps to eliminate confusion in the literature about how to operationally define the variables and to establish consistency in how various researchers define and measure these variables. Further work is also needed to identify better measures for social engagement and social capital and in general to identify the best measures for assessing overall social integration.
In order to account for individual variation and experience, it is important to assess multiple aspects of social integration for all older adults, particularly those in nursing homes. For example, some older adults have multiple family members who are available on a daily basis and visit regularly, others may have one friend from church who stops by once a week to play a game of cards, and some may prefer being alone but find they enjoy the daily encounters with staff in whom they trust. In all of these examples, the older adults may be socially integrated, but without testing multiple aspects of social integration, one might have greater tendency to state that older adults are isolated when they are not. Future research could analyze data using data techniques, such as latent class analysis, to establish potential groups of residents to target for interventions based on assessment of multiple aspects of social integration. Additionally, studies that assess social integration among older adults should account for the hierarchical nature/ between-level variance of data because, as this study shows, social networks and social support varied significantly between nursing homes.

Limitations
There are a number of recognized limitations in this study. First, this was crosssectional data, so therefore we were not able to assess differences over time.
Regarding validity and reliability of measures of social support and social engagement, we note there is further research needed. For the social support measure, we modified the Krause and Markides (1990) version of the ISSB to ensure it was appropriate for older adult living in nursing homes using strategies recommended by Stewart, Thrasher, Goldberg, and Shea (2012). However, we did not follow a more extensive scale development protocol due to time and resource limitations. For the social engagement, there needs to be additional research to further establish a valid and reliable measure for assessing social engagement for older adults, including those living in nursing homes. A strength of the study was that there were multiple interviewers who conducted the resident interviews, and all interviewers received training, practiced the interviews prior to conducting the study, and discussed questions and strategies for potential follow-up questions. However, the study did not assess inter-rater reliability. Regarding sampling, the study sample was limited to Kansas, so this does limit the generalizability for the resident and the nursing home samples. However, every attempt was made to ensure random sampling at both the nursing home-level and the resident-level, so certainly some strong comparisons can be made to other areas that have similar geographic or population characteristics.
Related to the older adult sample, it is recognized that this sample included older adults with relatively high cognitive functioning, which is potentially only 25% of older adults in nursing homes (Kaye, Harrington, & LaPlante, 2010). This level of cognitive functioning was needed in order to ensure older adults could answer the questions, but this does provide limitations. Nursing home residents with more severe cognitive impairments tend to report having a lower quality of life and fewer opportunities to participate in meaningful activities (Abrahamson, Clark, Perkisn, & Arling, 2012;Naylor et al., 2016). Further research is needed to both identify the concepts of social integration that are most relevant to nursing home residents with severe cognitive impairment and to develop interventions specifically for this population.
Finally, we were unable to control for SES on all constructs due to potential linear dependency issues. Therefore, this finding suggests that future work is needed to better understand the influence of SES on various aspects of social integration for older adults living in nursing homes, as this may be a targeting factor for intervention work.

Conclusion
This study's findings provide important building blocks to support our work to measure social integration, and ultimately to improve social integration for nursing facility residents. Clear definitions and tested models facilitate the involvement of both researchers and social work practitioners in this vital effort. This work also points to the need for more attention to the impact of varying nursing home environments on social engagement and social integration, to understanding the influence of SES, and to developing nursing home staff who recognize the influence social integration has on residents' health, are trained in strategies to promote social integration, and have sufficient time to implement those strategies.
Notes 1 SES had a potential linear dependency with the social support and social engagement latent variables; therefore, SES was not controlled for on these selected constructs. 2 The model included the covariates (ADLs, cognitive status, self-reported health status) for each construct; SES was included for social networks and social capital. 3 Numbers in boxes correspond to the numbers in Table 1.