EVERYDAY DISCRIMINATION IN LATER LIFE: A SOCIAL NETWORK APPROACH

Abstract

What factors shape everyday discrimination among older adults? Existing perspectives focus on individual identities and social group membership (e.g., race/ethnicity, age) as key determinants of perceived discrimination. This paper examines the idea that individuals’ broader social contexts – including their personal social networks – also shape perceived discrimination, and in ways that may differ across racial groups. Using data from Round 3 of the National Social Life, Health, and Aging Project (N = 3,312), I consider how properties of personal networks are associated with how frequently older adults report everyday discrimination. Results indicate that more kin-centric personal networks protect against more frequent everyday discrimination, but that this protective effect may be stronger among White older adults. I propose why more kin-centric networks may play a different role in the perceived discrimination of White and Black older adults, and close by suggesting that social network composition may be a source of heterogeneity in the link between everyday discrimination and inequality in later life outcomes such as health.

INTRODUCTION

Personal discrimination in everyday life contributes to the persistence of racial, gender, and other forms of inequality in outcomes such as employment, criminal legal system contact, housing, and health (Kessler, Mickelson, and Williams 1999; Pager and Shepherd 2008; Quillian 2006). Discrimination is a key driver of racial health disparities in the United States (Goosby et al. 2015; Lewis, Cogburn, and Williams 2015; Pascoe and Smart Richman 2009; Williams and Mohammed 2009), and is associated with myriad physical and mental health problems later in life (Chang et al. 2020; Lewis et al. 2009, 2010), including heightened mortality risk (Farmer, Wray, and Thomas 2019).

To date, sociological perspectives on discrimination have focused largely on those markers of social group membership that are often the basis of this experience – namely age, gender, race/ethnicity, religion, and sexual orientation (e.g., Burt, Simons, and Gibbons 2012; Rivera 2017; Tilcsik 2011; Vogt Yuan 2007). A smaller set of studies draws attention to factors beyond an individual’s social group membership, finding that neighborhood socioeconomic status and social composition also shape the frequency of everyday discrimination (Dailey et al. 2010; Hunt et al. 2007; Stokes and Moorman 2016). This research highlights the role of social-environmental processes in determining the potential for discriminatory social interactions to occur (Feagin 1991; Hagestad and Uhlenberg 2005).

Scholars have not paid as much attention to the role of the social networks in which individuals are embedded. As I argue in this paper, older adults’ personal social networks have the potential to shape perceptions of everyday discrimination through their multifaceted intersections with individual identities, belief structures, and everyday social exposures (e.g., Hogg and Rinella 2018; Marsden and Friedkin 1993; McFarland and Pals 2005; Valente 2010). Individuals depend on personal network members for social support following experiences that are perceived as stressful or threatening (e.g., Thoits 2011; Cohen and McKay 1984; Ajrouch et al. 2010), such that network members have opportunities to influence how individuals come to understand and attribute their everyday social interactions. Additionally, sharing time and space with personal network members in the context of routine daily activities could influence an individual’s exposure to contexts where discriminatory social interactions may occur (e.g., medical offices, housing searches), furthering opportunities for network members to influence one another’s perspectives and understandings of such encounters.

In this paper, I use data from Round 3 of the National Social Life, Health, and Aging Project (NSHAP) to shed light on the question: How are features of older adults’ personal social networks associated with the frequency of perceived everyday discrimination? In addition to the relevance of social networks and everyday discrimination to later-life well-being, exploring this research question among older adults means that the sample as a whole is broadly vulnerable to various forms of discrimination, including ageism, even though perceived discrimination as a whole tends to decline with age (Gee, Pavalko, and Long 2007). The analyses suggest that personal network composition – in particular, the extent to which one’s personal network is comprised of kin – plays a role in shaping the frequency of everyday discrimination among older adults, but that the nature of this association differs between Black and White older adults.

THE SOCIAL BASES OF PERCEIVED EVERYDAY DISCRIMINATION

Individual reports of everyday discrimination – that is, the awareness or judgment that one has been mistreated because of their identity or social group membership – represent a consequential measure of discriminatory experience that is generally treated by social scientists at face value (Major, Quinton, and McCoy 2002). Awareness of everyday discrimination is often emphasized as a primary means through which such mistreatment is consequential for psychological and physical well-being (Ajrouch et al. 2010, p. 418). Self-reports include personal assessments of everyday interactions or experiences that individuals identify as discriminatory, which can include more overt, blatant discriminatory interactions – for example, being verbally insulted or told that one is excluded from an opportunity on the basis of their race, gender, or other identity – as well as less overt forms of mistreatment on the basis of an attribute or identity. Audit studies and other experimental designs are a more objective means of assessing the prevalence of discrimination in access to opportunities such as employment and housing (e.g., Pager 2007; Tilcsik 2011). Individuals who are discriminated against in these scenarios are not always aware of this mistreatment, though these experiences are nevertheless profoundly consequential for attainment and well-being (e.g., Quillian et al. 2017). Whether or not someone perceives an given interaction as discriminatory is a key factor in whether the interaction contributes to stress, as well as the range of subsequent adverse mental and physical health consequences that are tied to the stress response (Ajrouch et al. 2010; Birditt et al. 2018; Goosby, Straley, and Cheadle 2017; Williams 2018; Williams and Mohammed 2009). Put simply, perceived everyday discrimination is important to study in and of itself (Small and Pager 2020).

Not surprisingly, most approaches to understanding everyday discrimination focus on age, race/ethnicity, gender, and other indicators of individuals’ social location (e.g., immigration status, sexual orientation). that are most often the attributed bases of discriminatory treatment. Although generally more concentrated among historically marginalized populations, discrimination is reported across social groups in the United States (e.g., Bratter and Gorman 2011; Kessler et al. 1999; Kim, Sellbom, and Ford 2014). Indeed, approximately 49% of individuals report experiencing some form of discrimination, either rarely, occasionally, or regularly – including over 70% of Black individuals and over 30% of White individuals (Lee et al. 2019). As I expand on in subsequent sections, there is good consider that the social contexts of these experiences differ across social groups, particularly among Black and White individuals, and in ways that are likely to intersect with properties of individuals’ personal networks.

Social Context and Discrimination

Sociologists have begun to consider the role of larger social contexts in shaping individuals’ reports of everyday discrimination (Bobo and Fox 2006; Feagin 1991; Welch et al. 2001). This work stems in part from the idea that the social environments where individuals spend significant portions of their time structure opportunities for social interactions that are potentially discriminatory (Cagney 2006; Feagin 1991; Hagestad and Uhlenberg 2005). Ethnic density theory, for instance, posits that the proportion of ethnic minority residents in a particular community or neighborhood shapes opportunities for social contact with individuals within or outside of one’s own ethnic group (Bécares, Nazroo, and Stafford 2009; Halpern and Nazroo 2000). Potentially discriminatory interactions with majority group members are less likely when spending time in places with a higher composition of minority group members (Halpern and Nazroo 2000; Hunt et al. 2007). Research on neighborhood social composition supports the general notion that being surrounded by in-group members protects against perceptions of discrimination against the group. Among Black individuals, for example, perceived racial discrimination is lowest for those residing in neighborhoods with a higher percentage of Black residents (Hunt et al. 2007). Likewise, neighborhoods that have a higher proportion of older adults protect against perceptions of ageism among individuals entering later life (Stokes and Moorman 2016).

At a more micro-level, social psychological literature prioritizes interpersonal processes that shape individuals’ conceptualizations of those identities that may be targeted by discriminatory treatment. Indeed, discrimination is psychologically painful because it directly threatens and devalues a core part of one’s social identity (Branscombe, Schmitt, and Harvey 1999; Harris-Britt et al. 2007; Schmitt and Branscombe 2002). When an aspect of one’s identity is made salient – perhaps through social interaction with others – individuals are more aware of stigmatization and discrimination based on that identity (Major et al. 2002). Thus, more personal aspects of the social context (e.g., close interpersonal ties) may be just as important as its more general socio-demographic composition (e.g., neighborhood or institutional composition) in shaping perceived discrimination.

THE RELEVANCE OF PERSONAL SOCIAL NETWORKS

The primary focus of this is paper is on the relevance of personal social networks – that is, one’s core set of social confidants – as a concrete aspect of the social context that may shape the frequency of everyday discrimination among older adults. These networks are typically comprised of one’s most intimate and strongest social ties, with whom individuals exchange support and discuss issues of personal importance (Fischer 1982, 2011; Marsden 1987; Wellman and Wortley 1990). Core network ties are a main pathway through which “influence processes and normative pressures operate” (Marsden 1987, p. 123), and through which individuals are integrated into society (Fischer 1982).

A vast literature on personal networks establishes that different network properties can facilitate access to social and other resources in distinct ways. Network size is a general measure of social integration and access to alternative sources of support and other resources (Berkman et al. 2000; Marsden 1987). The frequency of contact between network members represents the opportunity structure for support exchange (e.g., Hank 2007), and is a general measure of relationship strength (Swartz 2009; Ward, Deane, and Spitze 2014). Network composition is often examined in circumstances where the type of network tie is relevant, referring to which social relationships are included in the social network, and the associated social roles (Fischer 1982; Marsden 1987).

Personal networks can serve as a prism for understanding social interactions or behaviors. Individuals may turn to personal network members following discriminatory interactions, as close social ties are primary sources of social support, acceptance, and enhanced psychological well-being following distressing, or otherwise negative events (Branscombe et al. 1999; Schmitt and Branscombe 2002; Thoits 2011; Williams 2012). Network members’ evaluations of such interactions as discriminatory or otherwise can influence the meaning that individuals assign to those interactions, including whether they were mistreated on the basis of a personal attribute or group identity. Larger networks may offer a larger pool of collective discrimination experience for an individual to draw on when evaluating their own interactions, and include more alternatives for supporting and corroborating one’s experiences as discrimination. More frequent contact with network members can indicate greater availability of close ties who can be called on to evaluate such experiences as discriminatory or otherwise.

But the fact that discrimination is tied to social identities and social group membership suggests that social network processes may extend to those network properties most pertinent to the particular social identity that is the target of discriminatory treatment. In this regard, network composition, or who comprises one’s personal network, may be especially pertinent. Akin to the application of ethnic density theory in studies of neighborhoods, personal networks could be comprised of members who are similar (“in-group”) or different (“out-group”) from an individual along various identities and social group memberships.

Network homogeneity refers to the degree to which members of a social network are similar along one or more identities (Louch 2000; Marsden 1988) – i.e., in-group members. Network homogeneity along a particular trait – e.g., race, religion, social class – can increase the salience of that identity for network members (McFarland and Pals 2005; Walker and Lynn 2013) and help to identify discriminatory treatment by out-group members on that basis. Personal networks that are more heterogeneous along certain identities may also function to lessen perceptions of discriminatory treatment, dissuading individuals (or otherwise not corroborating) that they were mistreated based on that identity, especially if network members have not personally experienced mistreatment on the basis of that same identity.

Social networks are contexts that influence the salience and meaning of various identities (Stryker and Burke 2000; Walker and Lynn 2013). Network homogeneity may influence the salience of certain identities in different ways with respect identifying social interactions as discriminatory or not. Similarly-identifying network members may be a source of solidarity, affirming certain social interactions as discriminatory, particularly if they have also experienced discrimination on that basis (Sellers and Shelton 2003). Differently-identifying network members may be less likely to corroborate mistreatment on the basis of identities that are not shared and may even dissuade an individual from identifying negative interactions as discriminatory.

Age and race can be used to illustrate these processes. The social meaning of age (i.e., whether one is “old” versus “young”) is relative across the life course and across social and institutional contexts. An individual’s identity is not inherently rooted in the category of “older adult” across the entire life span. A personal network that includes members of diverse ages may offer more social support to persuade an individual that potential age-based mistreatment from others is misattributed (e.g., “You are not old”), reducing reports of discrimination on these bases.

Racialization and the racial hierarchy in the U.S., however, make an individual’s race a life-long, ascribed and socially constructed identity. For racial groups that have been historically marginalized and socially excluded, discrimination may be anticipated and more likely experienced throughout the life course (Gee and Ford 2011; Hicken et al. 2013; Kessler et al. 1999). Kin network members are likely to exhibit racial similarity but, even more so, are likely to represent shared belief systems and collective histories that originate in the family context and that have implications for everyday discrimination in ways that are distinct from other types of network members. Indeed, scholarship at the intersection of race and family discusses racial socialization practices that begin in the family context during early childhood and endure across the life course, preparing children for the types of prejudiced interactions that they may encounter in the future (Hagerman 2014; Hughes et al. 2006; Hughes and Chen 1997). Through this socialization, families (often parents) cultivate individuals’ knowledge about racial hierarchy, conditioning their expectations for the types of treatment that they can expect based on their racial identity (Bonilla-Silva 1997, 2006; Rollins and Hunter 2013).

For Black families, these practices can involve fostering a heightened sense of vigilance against race-based mistreatment, and cultivating a strong sense of racial consciousness, pride, and positive racial identity (Hicken et al. 2013; Hughes et al. 2006). A higher presence of kin in Black individuals’ personal networks may help individuals to identify discriminatory interactions. White families, for whom discrimination is overall less prevalent, are more likely to effectively and implicitly socialize children not to expect mistreatment based on race or other identities (Bonilla-Silva 2006; Hagerman 2014).

Personal networks can also shape everyday discrimination by influencing an individual’s broader range of social interactions. The ways in which time is spent with network members may draw older adults into certain spaces or contexts that present additional opportunities for social interactions or encounters with others that may be discriminatory. For example, network members may assist one another with medical appointments and other routine activities (e.g., grocery shopping) – contexts where there is potential for discrimination from others outside of the network (Stepanikova and Oates 2017). Further, being in the company of social network confidants allows for greater occasions for close others to influence the interpretation and attribution of social experiences. The more frequently that one accesses their social network members, the greater the potential to share time in social contexts (e.g., Hank 2007), while larger networks can also offer more diverse opportunities for exposure by way of spending time in different social spaces with different network members.

PERCEIVED EVERYDAY DISCRIMINATION IN LATER LIFE

Although discrimination occurs across the life course (Gee et al. 2007), everyday discrimination is associated with a number of adverse health outcomes among the older adult population, including elevated blood pressure (Lewis et al. 2009), inflammation (Lewis et al. 2010), and mortality (Barnes et al. 2008; Farmer et al. 2019), and may be especially harmful for the health of Black older adults (Lewis et al. 2009, 2010). Although everyday discrimination tends to decline with age (Gee et al. 2007), ageism may become more prominent in later life. Ageism is associated with worse physical and mental well-being (Chang et al. 2020), and is considered to be a form of elder abuse and mistreatment (Bugental and Hehman 2007). Thus, understanding the social-contextual basis of everyday discrimination is a key direction for better understanding the social determinants of later-life health and health disparities.

Additionally, personal networks are primary social contexts through which older adults experience, interpret, and generally make sense of daily life (e.g., Alwin, Felmlee, and Kreager 2018; Antonucci, Akiyama, and Takahashi 2004) – discriminatory or otherwise. Later life is characterized by numerous transitions and changes, such as retirement, widowhood, and declines in health, many of which involve shifts in daily contacts and routine contexts (Cornwell et al. 2014; Kalmijn 2012). Older adults often demonstrate heightened reliance on their personal networks as they navigate these transitions, making social network ties an especially critical source of support, influence, and integration during this later period of the life course (Seeman 2000; Umberson, Crosnoe, and Reczek 2010). Thus, any role of personal networks in shaping everyday discrimination may be especially pronounced among older adults.

The life course perspective also prompts consideration of how earlier discrimination contextualizes these experiences later in life, and in ways that are likely to differ across racial groups. Following a lifetime of relatively un-stigmatized social status, White older adults (particularly those of higher socioeconomic status) enter later life with fewer strategies for dealing with discrimination compared to Black older adults who are more likely to have endured race-based mistreatment in varying degrees throughout their lifetime (Abramson 2015; Alwin, Thomas, and Sherman-Wilkins 2018; Gee et al. 2019) – not only through interpersonal interactions, but also through institutions and organizations (e.g., education, healthcare, employment) (Burt et al. 2012; Gee et al. 2019; Stepanikova and Oates 2017). In this regard, investigating everyday discrimination among older adults sheds light on the role of the later life social context above and beyond the role of prior experiences earlier in the life course.

THE PRESENT STUDY

While an in-depth consideration of potential mechanisms is beyond the scope of this study, this paper presents as an initial inquiry into whether personal social networks are an aspect of the social context that is relevant for frequency of discrimination, and that has not been as extensively explored as other social contexts such as neighborhoods. This study examines how aspects of older adults’ personal networks shape the frequency of everyday discrimination, above and beyond other personal and social-environmental factors. Throughout the analysis I attend to racial differences. This focus is motivated by the stark differences among Black and White older adults in both expected and experienced discrimination across the life course (Alwin et al. 2018; Gee, Walsemann, and Brondolo 2012; Hicken, Lee, and Hing 2018). Prior research has explored how neighborhood-level and organization-level characteristics shape racial differences in everyday discrimination (Dailey et al. 2010; Lee 2000), but less research has extended this study to the social network context. This paper presents as an initial inquiry into whether personal social networks are an aspect of the social context that is relevant for frequency of discrimination, and that has not been as extensively explored as other social contexts.

Some prior studies find that Black and White older adults differ in their personal networks, with Black older adults tending to have smaller and more frequently contacted personal networks compared to their White counterparts (e.g., Ajrouch, Blandon, and Antonucci 2005). In part, these differences are thought to reflect a heightened reliance on close network members as sources of social and economic resources, or fewer opportunities for developing larger, more expansive personal networks. reflecting a lifetime of differences in institutional events, exclusion, and structural sources of racism and socioeconomic disadvantage that disproportionately affect Black communities (Ajrouch, Antonucci, and Janevic 2001; Alwin et al. 2018). Thus, a key emphasis of this study is on how personal networks are a social-structural context that may differentially shape perceptions of everyday discrimination among Black and White older adults.

Personal network relationships and everyday discrimination are likely to intersect in dynamic ways over time, and in ways that I cannot fully account for in this study. I hypothesize that personal network properties influence perceptions of everyday discrimination, but I cannot analytically distinguish this relationship from the possibility that perceived discrimination also influences social network properties. Indeed, the potential for reverse causality is an important limitation of this study. Exposure to discrimination across the life course can shape the types of social networks that individuals cultivate and maintain later in life. Racial discrimination, for example, can lead Black individuals to form and maintain network ties with same-race peers and kin who can understand their experiences with this form of mistreatment (Tatum 2004; Ueno 2009).

DATA AND METHODS

The National Social Life, Health, and Aging Project

To begin to shed light on the question of how personal social networks shape the frequency of everyday discrimination, I use data from the National Social Life, Health, and Aging Project (NSHAP). The NSHAP is a population-based panel study of community-dwelling older adults in the United States (Suzman 2009). The overall goal of the NSHAP is to better understand how health and the social context intersect to influence older adults’ well-being as they age. At each round, data collection consisted of in-home interviews conducted by the National Opinion Research Center (NORC), which included the collection of personal social network information. Following the in-home interview, respondents were also asked to complete a leave-behind questionnaire (LBQ) to be returned to NORC by mail. This study relies on data collected at Round 3 (2015–2016). The sample includes returning respondents who participated in Rounds 1 and/or 2 and their partners (N = 2,409), if applicable, as well as a new cohort of respondents and their co-resident partners (N = 2,368). Although the NSHAP includes three rounds of data collection, each of which collects information about respondents’ personal social networks, Round 3 is the first survey round to ask about everyday discrimination.

Perceived Everyday Discrimination

The Everyday Discrimination Scale (EDS) is among the most widely used survey-based scales to measure perceptions of everyday discrimination – that is, those perceptions of unfair treatment based on gender, race, ethnicity, age, appearance, or other identities, attributes, and social group memberships, and that take place as part of what are considered to be routine daily encounters (Kim et al. 2014; Lewis et al. 2012; Williams et al. 1997). Whereas instances of objective discrimination may be represented in these self-reports, this scale further reflects individuals’ subjective assessments of how often they have been unfairly treated by others (Bratter and Gorman 2011; Goosby et al. 2018). Beginning in Round 3, the NSHAP included two items from the nine-item EDS as part of the LBQ. Respondents were asked: “In your day-to-day life, how often have you been treated with less courtesy than other people?” and, separately, “In your day-to-day life, how often have people acted as if they’re better than you are?” Responses to each of the two EDS items were collected using the following scale: 0 = never, 1 = less than once a year, 2 = about once or twice a year, 3 = several times a year, 4 = about once a month, 5 = every week, and 6 = several times a week.

The EDS scale was created to assess perceptions of mistreatment experiences across different social groups (e.g., race, gender, age, etc.), without explicit reference to racism, discrimination, or prejudice (Williams et al. 1997). Prior research has identified two subscales of the 9 EDS items among older adults: “unfair treatment” and “personal rejection.” The two items included in the NSHAP are both considered to be part of the “unfair treatment” subscale (factor loadings of .54 or higher) (Barnes et al. 2004). Other work on the nine EDS items indicates measurement equivalence on these two items among Black and White respondents (Kim et al. 2014). The main outcome in this analysis is a scale that is the average of these two items, representing how often respondents report everyday discrimination (alpha = .80).

Social Network Measures

At each round, the NSHAP asked respondents to name up to five individuals with whom they discussed important matters over the last 12 months. This “important matters” name generator is commonly used in surveys to elicit individuals’ core social confidants, including their strongest and most intimate social ties, with whom they are most likely to exchange resources and social support (Marsden 1987). Following the enumeration of their network members (i.e., “alters”), respondents were asked to indicate their relationship to each alter (e.g., spouse, friend, neighbor) and how often they speak with each alter (1 = “less than once a year,” 8 = “every day”).

From this information, I constructed three measures to capture the structural and compositional aspects of respondents’ personal networks that are likely to influence individuals’ reports of everyday discrimination, based on the range of mechanisms outlined earlier. Network size represents the range of potential sources for discussion and exchange following potentially discriminatory interactions, as well as the range of diverse exposures to people and contexts that are potential sources of mistreatment. Frequency of contact indicates the availability of close ties who could be activated following discrimination, as well as the opportunities to be present with network members during daily activities. Proportion kin captures network composition and the extent to which one’s social network members are drawn from a single social context (i.e., the family). Family ties tend to be long-standing relationships with individuals of similar social background, and who may be particularly influential in shaping how individuals perceive aspects of their identity as part of everyday interactions. While data on network members’ race, age, and other characteristics could be used to construct other measures of network homogeneity, this information is not collected as part of the NSHAP.¹

Social network size is measured as a count of the total number of alters named, ranging from 1 to 5, and serves as a general measure of social integration (Marsden 1987), as well as the number of distinct sources of potential support, advice, and personal experience that one has to call on in their network. Frequency of contact is an average of how often a respondent reports being in touch with their network members, and is a useful proxy for potential support exchange, familiarity, and communication (e.g., Hank 2007). Proportion kin is measured as the proportion of alters who are related to the respondent as either a spouse/partner, parent, parent-in-law, child, stepchild, grandchild, sibling, other in-law, or other relative.

Number of kin serves as an absolute measure of network composition, and proportion of kin serves as a relative measure (Marsden 1987; Moore 1990).² I ultimately use proportion kin in the analyses. A count of kin network members is highly correlated with network size given that the NSHAP limits network enumeration to a maximum of five alters, and that older adults’ personal networks tend to be relatively kin-centric (Fung, Carstensen, and Lang 2001; Marsden 1987). Nevertheless, respondent networks do vary in size. By using network size as the denominator in calculating proportion kin, I can effectively compare respondents who may have similar network composition (i.e., the majority or minority of network members are kin), but who differ in network size. Therefore, the relative measure represents a single estimate of how heavily older adults’ core personal networks draw on kin as a main source of confidants.

Sociodemographic and Social Environment Measures

I account for several sociodemographic and social environmental measures that are associated with everyday discrimination. Individual-level measures include gender and age (in years). I also control for whether the respondent is Hispanic. Race in the NSHAP is self-reported by the respondent as either White, Black/African American, American Indian or Alaskan Native, Asian or Pacific Islander, or other. Residential census tract characteristics are drawn from the 2011–2015 American Community Survey (ACS). I include the proportion of Black, non-Hispanic residents and the proportion of residents over the age of 65 as covariates. These measures capture aspects of neighborhood social composition that are relevant to two of the most prevalent types of everyday discrimination reported in the sample: racial/ethnic discrimination and ageism (see Appendix Figure 1).

Older adults’ subjective assessments of residential neighborhoods are informative of how they perceive their residential social context outside of their household. As the local neighborhood provides a key opportunity structure for older adults to interact with others (e.g., Stokes 2019; Stokes and Moorman 2016), respondents’ perceptions of their neighborhoods may influence their likelihood of spending time and socially engaging in their neighborhood (Riina et al. 2013). I account for respondents’ perceptions of their neighborhood collective efficacy using a scale that averages and standardizes eight items assessing respondents’ perceptions of neighborhood characteristics such as social cohesion, trust, and social connectedness within their neighborhood (α =.79) (Sampson, Morenoff, and Earls 1999). I also account for respondents’ perceptions of neighborhood danger as a scale that averages and standardizes three items assessing perceptions of fear, expectations of “trouble” in certain places, and risk of walking alone after dark (α =.83) (York Cornwell and Cagney 2014). Higher scores on these two scales indicate greater perceived neighborhood collective efficacy and danger, respectively. More broadly, these measures are informative of older adults’ perceptions of a salient social space in their daily lives that carries potential opportunities for everyday discrimination (Hunt et al. 2007; Stokes 2020), in ways that may explain or render less significant the role of personal network characteristics.

Control Variables

I include several life-course and health-related covariates that are likely to be related to both everyday discrimination and social network measures (Gee et al. 2007; Kessler et al. 1999; Williams 2018). These include whether the respondent has attended college/university, is married/partnered, and whether the respondent is retired at Round 3. Health covariates include whether the respondent reports good, very good, or excellent self-rated physical health (=1). I also account for respondents’ functional health, as mobility issues may limit the extent to which respondents can move about independently, thus limiting their exposure to social interactions that may be discriminatory. I measure functional health as the average of standardized responses to 9 items assessing respondents’ difficulty completing basic daily activities independently (e.g., eating, dressing), with higher scores indicating greater functional impairment. Finally, religious communities serve as formal sources of in-group solidarity and support for many older adults (Lim and Putnam 2010). I therefore control for frequency of attending religious services using a 6-point scale (0 = “never,” 5 = “several times a week”).

Analytic Strategy

The analyses proceed in two stages. First, I use descriptive analyses to consider racial differences in the frequency of perceived everyday discrimination, as well as social network and other key covariates. In the second stage, I conduct multivariable analyses to examine whether the frequency of everyday discrimination is a function of social network properties, above and beyond individual-level and other social-structural predictors. I use ordinary least squares regression models to examine how social network properties are associated with the frequency of reporting everyday discrimination, using the scale of the two EDS items as the dependent variable.³ In a first set, I present separate (within-race) models for Black and White respondents given race-associated differences in the lived experiences across racial groups, and in ways that intersect with the frequency of everyday discrimination. Separate models allow for estimates that are fully-interacted with respondent self-reported race, recognizing that the role of individual-level, personal network and neighborhood-level factors may differentially shape frequency of discrimination between these groups. It is important to note that these analyses rely on self-reported race, while individuals’ lived experiences and social interactions may vary based on how others perceive them (White et al. 2020). In a second set of results, I present a single model that includes both Black and White respondents, and interaction terms between each of the three network variables and respondent race. Round 3 of the NSHAP included interviews with 4,115 Black and White respondents ages 50 and older, from which the analytic sample is drawn. Although the NSHAP collects data from respondents who self-identify as American Indian, Alaskan Native, Asian or Pacific Islander, the relatively small number of respondents included in these categories does not yield sufficient statistical power for inclusion in the multivariable analysis.

The most significant source of missing data comes from the fact that everyday discrimination was measured using the leave-behind questionnaire, which some respondents (approximately 13.4%) did not return to NORC. Of the remaining respondents, 86 were excluded because they did not respond to the EDS items, or they reported that either they did not have a social network at Round 3, or they were missing data on social network characteristics. An additional 4.7% of respondents were excluded due to missing data on one or more covariates, with the majority of non-response coming from missing data on religious service attendance and neighborhood collective efficacy. Supplementary analyses that substitute missing values with values from prior rounds, to the extent possible, yield similar findings.

Unweighted t-tests reveal that respondents excluded from the analytic sample due to missing covariates did not differ significantly from those included in the analysis based on how frequently they interact with their network members, or the proportion of network members who are kin. Those included in the analytic sample did have significantly larger personal networks compared to those respondents who were excluded on the basis of missing data (M = 3.93 vs. M = 3.64, p < .001). The final sample size includes 3312 respondents.

I use inverse probability weighting to attenuate bias that may result from exclusion based on not returning the LBQ or missing data on other covariates. I use a logit model to predict respondents’ inclusion in the final analytic sample based on a number of sociodemographic, life course, and health measures. Next, I multiply the inverse of the predicted probabilities derived from the logit model by the person-level weights provided by the NSHAP that adjust for selection and non-response.⁴ I then apply these final weights to all model estimates. This process gives greater weight to those respondents who most resemble those excluded from the models due to missing data, generating estimates that better approximate those that would be generated had all respondents been included in the analysis (Morgan and Todd 2008). This weighting procedure follows the approach used in both cross-sectional and longitudinal studies that use the NSHAP (e.g., Cornwell and Laumann 2015; Hawkley et al. 2014; Schafer and Koltai 2015). I conduct all analyses using Stata 14. Standard errors are adjusted to account for clustering and stratification in the NSHAP sampling design (‘svy’ commands in Stata 14).

RESULTS

My main goal is to assess whether the frequency of perceived discrimination among Black and White older adults could be a function of social network properties. I begin by comparing the distributions of the main variables used in the regression models among Black and White respondents. I compare means using survey adjusted Wald tests and proportions using unweighted Chi-squared tests.

Descriptive findings.

As shown in Table 1, Black respondents report significantly more frequent everyday discrimination than do White respondents, although the majority of respondents in both racial groups report some discrimination (.69 of White respondents and .68 of Black respondents, with no statistically significant difference in these proportions). Respondents tend to maintain relatively large and intimate personal social networks, with significant differences across racial groups. On average, respondents name between three and four network members (with a maximum of 5 allowed). Black respondents’ networks are significantly smaller than those of White respondents, but also exhibit significantly more frequent contact with network members (“several times a week,” on average) than do White older adults’ networks (between “once a week” and “several times a week”). Kin members comprise a notable proportion of network members, including 65% of network members, on average, among Black respondents, and 62% of network members among White respondents. Proportion kin does not significantly differ by race.

Table 1.

Descriptive Statistics of Key Variables, by Race.

	Proportion or Mean (SD) a
	Black Respondents (N = 527)	White Respondents (N = 2,785)
Perceived Everyday Discrimination
Everyday discrimination scale (Average of 2 EDS items, Range: 0 to 6)	1.65 (1.69)**	1.40 (1.36)
Proportion of respondents reporting any everyday discrimination (i.e., more than 0 on EDS scale)	.68	.69
Social Network Variables
Network size (Range: 1 to 5)	3.67 (1.47)**	3.96 (1.23)
Average frequency of interaction with network members (Range: 0 “never” to 8 “every day”)	7.00 (.91)***	6.73 (.84)
Proportion kin in the network (Range: 0 to 1)	.65 (.36)	.62 (.31)
Sociodemographic, Life Course, and Health Covariates
Age	62.00 (9.67)**	64.27 (9.61)
Hispanic (1 = yes)	.01***	.06
Female (1 = yes)	.59	.55
Attended college (1 = yes)	.49***	.64
Retired (1 = yes)	.49**	.56
Married/partnered (1 = yes)	.53***	.73
Self-rated physical health (1 = Excellent/very good/good; 0 = Fair/poor)	.71***	.80
Functional limitations (Average of 9 standardized items, Range: −3.46 – 6.14)	.08 (.86)***	−.07 (.62)
Social and Contextual Covariates
Frequency of attending religious services (0 = “never”; 5 = “several times a week)	2.83 (1.77)***	1.99 (1.74)
Perceived neighborhood collective efficacy (Average of 8 standardized items, range: −2.22 – 1.79)	−.16 (.70)***	.04 (.64)
Perceived neighborhood danger (Average of 3 standardized items, range: −1.35 – 2.59)	.36 (.96)***	−.18 (.80)
Proportion Black, non-Hispanic residents in tract	.53 (.36)***	.07 (.12)
Proportion residents age 65 years and older in tract	.13 (.05)***	.16 (.07)

^*p < .05

^**p < .01

^***p < .001 (Two-tailed tests).

Asterisks indicate significantly different from the mean or proportion calculated among White respondents, using survey weighted adjusted Wald tests (to test mean differences) and unweight chi-squared tests (to test differences between proportions).

^aMeans are weighted using NSHAP Round 3 respondent weights (adjusted for selection) and are survey adjusted. Standard deviations appear in parentheses for continuous variables. Proportions are unweighted.

Multivariable findings.

Next, I examine the results of multivariable ordinary least squares regression models to consider how social network properties predict the frequency of everyday discrimination when accounting for individual, neighborhood, and other social factors, among Black and Whites respondents, separately. Among White respondents (Table 2, Model 1), network size and frequency of contact with network members do not appear to shape how frequently respondents report everyday discrimination. A higher proportion of kin in one’s personal network, however, is significantly associated with less frequent everyday discrimination (b = −.43, p < .001), suggesting that a more kin-centric network appears to be protective against more frequent everyday discrimination.

Table 2.

Coefficients from Ordinary Least Squares Regression Models Predicting the Frequency of Perceived Everyday Discrimination. ^a

	White Respondents	Black Respondents	All Respondents
Predictor	Model 1	Model 2	Model 3	Model 4
Network size	0.01	−0.02	0.01	0.01
	(0.02)	(0.07)	(0.02)	(0.02)
Network size × Black	--	--	--	−0.03
				(0.06)
Frequency of contact with network	0.04	−0.12	0.02	0.04
	(0.04)	(0.11)	(0.04)	(0.04)
Frequency of contact × Black	--	--	--	−0.17
				(0.10)
Proportion kin in network	−0.43*** b	0.17	−0.34**	−0.45***
	(0.11)	(0.26)	(0.10)	(0.11)
Proportion kin × Black	--	--	--	0.67*
				(0.30)
Age	−0.04***	−0.03*	−0.04***	−0.04***
	(0.00)	(0.01)	(0.00)	(0.00)
Female	−0.18**	−0.41*	−0.21***	−0.21***
	(0.06)	(0.18)	(0.05)	(0.05)
Black	--	--	0.22	1.07
			(0.12)	(0.69)
Hispanic	−0.16	−0.72	−0.21	−0.20
	(0.12)	(0.54)	(0.12)	(0.11)
Attended college	0.08	0.16	0.09	0.09
	(0.08)	(0.22)	(0.08)	(0.08)
Retired	−0.24**	−0.22	−0.23***	−0.23***
	(0.07)	(0.22)	(0.06)	(0.06)
Married/partnered	−0.08	0.26	−0.03	−0.02
	(0.07)	(0.17)	(0.06)	(0.06)
Self-rated health	−0.21*	−0.13	−0.18*	−0.18*
	(0.09)	(0.15)	(0.08)	(0.08)
Functional limitations	0.10	−0.03	0.07	0.07
	(0.08)	(0.09)	(0.06)	(0.06)
Frequency of attending religious services	0.04*	0.10	0.04*	.05*
	(0.02)	(0.05)	(0.02)	(.02)
Perceived neighborhood collective efficacy	−0.20***	−0.33**	−0.22***	−0.22***
	(0.05)	(0.10)	(0.05)	(0.05)
Perceived neighborhood danger	0.26*** b	0.50***	0.30***	0.30***
	(0.04)	(0.08)	(0.04)	(0.04)
% Black, non-Hispanic in tract	−0.75**	−0.78**	−0.74***	−0.71***
	(0.24)	(0.28)	(0.19)	(0.20)
% Age 65 and older in tract	−0.14	1.10	−0.03	−0.01
	(0.42)	(1.31)	(0.40)	(0.40)
Constant	4.67***	4.42***	4.42***	4.28***
	(0.43)	(0.90)	(0.39)	(0.41)
N	2,785	527	3,312	3,312
F (df)	34.86*** (16, 80)	10.91*** (16, 55)	38.06*** (17, 79)	31.77*** (20,76)
R2	.18	.18	.17	.18

^*p < .05

^**p < .01

^***p < .001 (Two-tailed tests).

Standard errors appear in parentheses.

^aAll estimates are weighted using the NSHAP Round 3 respondent-level weights that adjust for selection, non-response, and inclusion in the analytic sample, and are survey adjusted.

^bCoefficient is statistically significantly different from coefficient for Black respondents in Model 2 (p < .05).

Older age, being female, and being retired are each also significantly inversely associated with more frequent everyday discrimination (b = −.04, p <.001, b = −.18, p < .01, b = −.24, p <.01, respectively) among White respondents. Neighborhood context also emerges as relevant. Higher levels of perceived neighborhood collective efficacy and a higher percentage of Black residents in respondents’ residential tracts are each associated with less frequent everyday discrimination among Whites (b = −.20, p < .001, and b = −.75, p < .01, respectively), while higher levels of perceived neighborhood danger is associated with more frequent everyday discrimination (b = .26, p <.001).

Among Black respondents (Table 2, Model 2), none of the three personal network characteristics emerges as significantly associated with frequency of everyday discrimination. Whereas older ages and being female are each inversely associated with frequency of everyday discrimination (b = −.03 p <.05 and b = −.41, p <.05, respectively), other statistically significant covariates suggest that neighborhood characteristics are particularly important factors. Indeed, a higher proportion of Black residents in one’s census tract and higher levels of perceived collective efficacy are each associated with significantly less frequent everyday discrimination (b = − 0.78, p < .01 and b = − 0.33, p < .01, respectively). Higher levels of perceived neighborhood danger are associated with significantly greater frequency of perceived discrimination (b = .50, p <.001) among Black older adults. Adjusted Wald tests indicate that the coefficient for proportion network kin among White respondents in Model 1 (b = −.43) is significantly different than the proportion network kin coefficient among Black respondents in Model 2 (b = .17) (F = 4.93, p < .05). The coefficient for neighborhood danger among White respondents in Model 1 (b = .26) is also significantly smaller than the coefficient for neighborhood danger among Black respondents in Model 2 (b = .50) (F = 8.44, p < .01). Other tests of equality across coefficients in Models 1 and 2 do not reach statistical significance.

Estimates in Model 3 represent those when both Black and White respondents are included in a single model, while Model 4 of Table 2 includes interaction terms between respondent race and each of the three social network measures of interest: network size, frequency of contact, and proportion kin. Consistent with the findings from the within-race models, Model 4 reveals little evidence that the association between network size or frequency of contact and everyday discrimination differs by race, however a higher proportion of network kin appears to be more protective against more frequent everyday discrimination for White older adults than for Black older adults.

Predicted probabilities from this interaction term indicate that White respondents with no kin in their networks report discrimination nearly “about once or twice a year” (1.68), while White respondents who have entirely kin-based personal networks report discrimination closer to “less than once a year” (1.23). Among Black older adults, social network characteristics do not emerge as statistically significant. However, neighborhood characteristics may play a more significant role. In a supplemental model, an interaction term between race and perceived neighborhood danger reaches marginal statistical significance (b = 0.19; p < .10). The difference in predicted frequency of everyday discrimination between older adults who perceive their neighborhood to be relatively safe versus unsafe is nearly 1.7 times as large among Black older adults than the same difference among White older adults.

Supplemental Analyses.

A comprehensive examination into underlying mechanisms requires more detailed data on the types of interactions and contexts that inform respondents’ reports of mistreatment, including differences in the roles of network kin in these social interactions. In the absence of such data, I conducted a supplemental analysis to assess whether perceptions of family social support are associated with frequency of everyday discrimination, and whether its inclusion explains the race by proportion kin interaction. Perceived family support is intended to shed light on the proposed mechanism that close kin ties may assist in interpreting events or interactions as discriminatory or not, potentially diminishing the salience of those attributes and identities (e.g., age, appearance) that White older adults in the sample report more frequently as the basis for others’ mistreatment, thereby leading to less frequent discrimination. To measure family social support, I create a scale by averaging and standardizing two items that ask respondents how often they feel they can rely on family and how often they feel they can open up to family, where 1 = never, 2 = sometimes, and 3 = often (alpha = .74). The family support measure is not necessarily specific to the individuals named in the network roster, but nevertheless represents a key dimension of the quality of respondents’ family ties (Wong et al. 2020).

Consistent with the hypothesized mechanisms, the results in Table 3 suggest that more family social support is significantly associated with less frequent everyday discrimination (b = −.08, p <.05). Proportion kin remains significantly associated with lower levels of everyday discrimination in this model, and more so among White respondents (b = .65, p <.05). Supportive family relationships are important in protecting against perceived discrimination in the sample as a whole, but the kin-centric nature of one’s network appears to buffer against discrimination among Whites in ways that are not fully captured by this measure.

Table 3.

Coefficients from Ordinary Least Squares Regression Models Predicting the Frequency of Perceived Everyday Discrimination, Accounting for Family Support, Among All Respondents.^a

Predictors	Model 1
Network size	0.02
	(0.03)
Network size × Black	−0.04
	(0.06)
Frequency of contact with network	0.05
	(0.04)
Frequency of contact × Black	−0.17
	(0.09)
Proportion kin in network	−0.40***
	(0.11)
Proportion kin × Black	0.65*
	(0.29)
Family social support	−0.08*
	(0.04)
Age	−0.04***
	(0.004)
Female	−0.19***
	(0.06)
Black	1.08
	(0.67)
Religious service attendance	0.05**
	(0.02)
Perceived neighborhood collective efficacy	−0.20***
	(0.05)
Perceived neighborhood danger	0.29***
	(0.04)
% Black, non-Hispanic in tract	−0.70**
	(0.20)
% Age 65 and older in tract	.02
	(.41)
Constant	4.15***
	(0.42)
N	3,308
F(df)	31.77*** (21, 75)
R2	0.18

^*p < .05

^**p < .01

^***p < .001 (Two-tailed tests).

Standard errors appear in parentheses.

^aAll estimates are weighted using the NSHAP Round 3 respondent-level weights that adjust for selection, non-response, and inclusion in the analytic sample, and are survey adjusted. All models include controls for ethnicity, education, retirement, marital status, self-rated health, and functional limitations (not shown due to space constraints).

DISCUSSION

Understanding perceptions of everyday discrimination is key to understanding an important source of variation in older adults’ health and well-being (e.g., Farmer et al. 2019; Lewis et al. 2009, 2010). The majority of sociological perspectives consider everyday discrimination to be a function of those identities or social groups that are historically socially excluded and marginalized, as well as characteristics of the broader social environment that collectively shape risk of exposure to everyday discrimination (Gee et al. 2007; Hunt et al. 2007; Monk 2015; Stokes and Moorman 2016). This study extends prior structural perspectives by examining the possibility that the frequency of everyday discrimination may be partly shaped by properties of older adults’ core network relationships.

The main analyses suggest that a more kin-centric personal network is associated with less frequent everyday discrimination, and that this association is stronger among White older adults compared to Black older adults. The findings further indicate that the association between proportion kin and frequency of perceived discrimination is not explained by general measures of family support. These findings prompt a more pointed consideration of ways in which a more kin-centric network – and not other personal network characteristics (e.g., size, frequency of interaction) – are relevant for frequency of everyday discrimination, and in ways that differ by race.

One possibility is that there are overall fewer collective experiences of perceived everyday discrimination among kin in White older adults’ personal networks compared to Black older adults’ personal networks. Whereas Black individuals are more often exposed to racism and racial discrimination throughout the life course (Gee et al. 2019, 2012), those identities or attributes that White individuals are more likely to cite as reasons for discrimination (e.g., age, weight, “other reasons”), may be represented more heterogeneously among network kin and may peak at different periods over the life span (Gee et al. 2007), leading to a less consistent pool of discrimination experience to share and draw from within Whites’ personal networks. Thus, among White older adults, more kin-based personal networks may be less likely to recognize or corroborate negative social encounters as discrimination, or may be more likely to dissuade older adults from determining certain social interactions as discrimination, and this may be especially so in the case of identities or attributes that such as age, weight, or disability status that may not be shared among kin.

The frequency of discrimination among Black individuals may be more a function of macro-level, social-environmental factors that reflect and structure systemic and institutionalized sources of racism, such as educational, neighborhood, and workplace composition, policies, and norms (Abramson, Hashemi, and Sánchez-Jankowski 2015; Gee et al. 2019; Hunt et al. 2007; Seaton and Yip 2009). These broader factors may render the more proximate personal network less relevant. Indeed, models stratified by race indicate that neighborhood racial composition and collective efficacy are significant predicators of frequency of everyday discrimination among Black older adults, with perceived neighborhood danger being more significant in shaping frequency of discrimination among Black older adults compared to White older adults. These findings suggest that more meso-level (i.e., neighborhood) social factors may be relevant in shaping everyday discrimination among Black older adults, especially subjective assessments of neighborhood safety.

Heterogeneity among network kin along those identities or attributes that White older adults are more likely to report as a basis for discrimination could also be relevant in buffering against discrimination when spending time together. In the case of age-based mistreatment, for example, more kin-centric networks may reflect a stronger presence of younger ties, as network kin represent older adults’ primary source of cross-age relationships and integration (Hagestad and Uhlenberg 2005). Being present with network kin who are younger, or otherwise distinct from the older adult along traits that may be targeted (e.g., able-bodied, different appearance, different health status)⁵ could buffer treatment from strangers or acquaintances (e.g., cashiers, medical personnel, and other professions encountered during routine daily activities) in everyday interactions that might otherwise be reported as discriminatory. For example, an older adult at a medical visit may be less likely to report mistreatment from others on the basis of age if they are with a younger adult child as opposed to when they are alone or with a same-aged companion.

Limitations and Future Directions

Although the NSHAP includes three rounds of data, questions about discrimination were only administered at Round 3. The cross-sectional analysis means that the findings could reflect endogeneity due to reverse causality. Frequency of discrimination could play a role in shaping social network characteristics, and in ways that differ between Black and White older adults based on the type of – or attributed reason for - everyday discrimination. For example, as Gee and colleagues (2012) emphasize, Black individuals’ experiences of racial discrimination may be at their height between early adulthood and midlife through interactions with employers, the education system, health care professionals, and other institutions. More frequent racial discrimination earlier in the life course from non-kin network members may lead Black older adults to maintain a closer-knit network of trusted kin. Future analyses should draw on longitudinal data on the frequency of everyday discrimination and overtime changes in personal network composition to more rigorously test the directionality hypothesized in this study. In the absence of this data, I examined supplemental models that use lagged (Round 2) social network measures to predict the frequency of everyday discrimination at Round 3. The estimates from these models lend support to the main findings. A higher proportion of kin in one’s personal network at Round 2 is associated with significantly less frequent everyday discrimination at Round 3 (b = −.22, p <.05), and this association differs between Black and White respondents consistent with the main models (interaction term: b = .57, p < .05).

Future extensions should also collect more detailed information about the context of older adults’ discriminatory experiences, such as where and who they were with, the nature of the experience, and the source of the discrimination in relation to the social network members. This type of data would allow for a richer understanding of how network characteristics shape older adults’ perceptions of everyday discrimination. Additional information on network members such as their race, age, and other social identities could allow for a more nuanced analysis into how network homogeneity along multiple identities shapes the frequency of discrimination in ways that intersects with the type(s) of discrimination experienced.

Other limitations include the possibility that discrimination is under or overreported, and in ways I am unable to account for in this study. Members of socially stigmatized groups may not report discrimination given perceived social costs, to protect one’s self-esteem or sense of justice (Kaiser and Major 2006; Mayrl and Saperstein 2013), or to avoid the distress that accompanies being the victim of discrimination (Williams et al. 2012). In other cases, discrimination may be overreported if, for example, an individual experiences an especially salient or unexpected discriminatory interaction, or experiences negative outcomes that they attribute to a discriminatory experience (Mayrl and Saperstein 2013; Quillian 2006). Thus, there may be greater variation in reporting among Black and White older adults in this sample depending on the contexts of their experiences with discrimination. Additionally, the NSHAP only collects reports of respondents’ everyday discrimination. Other studies demonstrate the significance of major discrimination such as being fired from a job (e.g., Kessler et al. 1999), which could also intersect with personal network properties. Additional work should also explore the role of social networks among older adults of other racial and ethnic groups, and among other populations for whom everyday discrimination is prevalent and consequential. Whereas this analysis relies on respondents’ self-reported race, a more multidimensional measure of race would allow for examination of how self-assigned and socially-assigned race intersect with social network characteristics, as these distinctions are especially relevant to everyday discrimination (White et al. 2020).

CONCLUSION

Recent research emphasizes the need to more fully explore a range of individual and social factors that predict perceptions and reports of everyday discrimination (Lewis et al., 2015; Monk 2015; Williams 2018). This work emphasizes that individuals’ identities and social group memberships are situated in social contexts such as personal networks and neighborhoods that intersect with broader stratification systems and that are uniquely positioned to shape the meaning of social identities and opportunities for social interactions in everyday life. These findings suggest that aspects of older adults’ personal networks are a relatively overlooked source of within- and between-group differences in the link between everyday discrimination experiences and health, as well as other outcomes for which racial disparities persist. Future research should more fully consider the social network context as a potential source of heterogeneity in these associations.

Appendix Figure 1. Proportion of White and Black respondents who report each “main reason” for discriminatory treatment.

Note: Proportions for each racial group sum to more than 1 because respondents were able to select multiple “main reasons.” Proportions are unweighted and are based on respondents in the main analytic sample who report at least some experience of discrimination, and who had non-missing data on the “main reason” for discrimination experience.

Appendix Table A1.

Odds Ratios from Generalized Ordered Logit Model Predicting the Frequency of Others Acting “As if They’re Better than You Are” (N = 3,301).^a

Variables	Odds of reporting experiencing discrimination more than:
	Never	Less than once a year	About once or twice a year	Several times a year
Network size	1.04	1.00	1.05	1.06
	(0.97 – 1.12)	(0.93 – 1.07)	(0.97 – 1.14)	(0.94 – 1.19)
Frequency of contact	0.98	1.02	1.15	1.14
	(0.85 – 1.12)	(0.88 – 1.18)	(0.98 – 1.35)	(0.91 – 1.43)
Proportion network kin	0.65**	0.63**	0.56**	0.58*
	(0.47 – 0.90)	(0.46 – 0.86)	(0.38 – 0.83)	(0.35 – 0.95)
Age	0.94***	0.94***	0.94***	0.94***
	(0.93 – 0.95)	(0.93 – 0.95)	(0.92 – 0.95)	(0.91 – 0.96)
Female	0.79**	0.76**	0.68***	0.67*
	(0.67 – 0.94)	(0.63 – 0.93)	(0.54 – 0.85)	(0.48 – 0.92)
Black	.99	1.07	1.54*	2.01**
	(0.68 – 1.42)	(0.74 – 1.55)	(1.06 – 2.23)	(1.20 – 3.38)
Attended college	1.48**	1.20	0.88	0.82
	(1.17 – 1.87)	(0.94 – 1.53)	(0.66 – 1.17)	(0.57 – 1.18)
Married/partnered	1.13	0.97	1.12	0.95
	(0.91 – 1.39)	(0.79 – 1.20)	(0.88 – 1.43)	(0.66 – 1.38)
Retired	0.79*	0.66***	0.61***	0.67
	(0.64 – 0.98)	(0.54 – 0.81)	(0.47 – 0.79)	(0.42 – 1.09)
Self-rated health	0.96	0.83	0.76*	0.77
	(0.76 – 1.21)	(0.66 – 1.04)	(0.57 – 1.00)	(0.51 – 1.16)
Functional health	1.02	1.05	1.14	1.07
	(0.85 – 1.13)	(0.88 – 1.19)	(0.96 – 1.34)	(0.80 – 1.45)
Perceived neighborhood collective efficacy	0.85	0.82*	0.70***	0.64***
	(0.72 – 1.01)	(0.71 – 0.95)	(0.60 – 0.82)	(0.51 – 0.80)
Perceived neighborhood danger	1.23***	1.35***	1.51***	1.67***
	(1.09 – 1.37)	(1.19 – 1.52)	(1.32 – 1.73)	(1.39 – 2.01)
Proportion Black residents in census tract	0.49**	0.47**	0.28***	0.23**
	(0.31 – 0.81)	(0.28 – 0.81)	(0.15 – 0.51)	(0.08 – 0.66)
Proportion residents ages 65 and older	0.77	1.76	2.03	1.17
	(0.24 – 2.48)	(0.54 – 5.77)	(0.45 – 9.21)	(0.14 – 9.72)
Hispanic	0.48**	0.71	0.82	0.49*
	(0.31 – 0.76)	(0.45 – 1.14)	(0.54 – 1.24)	(0.29 – 0.86)
Religious service attendance	1.04	1.05	1.12**	1.06
	(0.98 – 1.10)	(0.99 – 1.12)	(1.05 – 1.20)	(0.98 – 1.15)
Constant	148.99***	54.83***	12.96**	5.32
	(35.71 – 621.64)	(14.97 – 200.76)	(2.91 – 57.79)	(0.70 – 40.33)

^*p < .05

^**p < .01

^***p < .001 (Two-tailed tests). 95% confidence intervals in parentheses.

^aAll estimates are weighted using the NSHAP Round 3 respondent-level weights that adjust for selection, non-response, and inclusion in the analytic sample, and are survey adjusted. Model is conducted using the “glogit2” package in Stata 14 (Williams 2016).

Appendix Table A2.

Odds Ratios from Generalized Ordered Logit Model Predicting the Frequency of Being “Treated with Less Courtesy than Other People” (N = 3,304).^a

Variables	Odds of reporting experiencing discrimination more than:
	Never	Less than once a year	About once or twice a year	Several times a year
Network size	1.05	1.05	1.02	0.98
	(0.97 – 1.15)	(0.98 – 1.13)	(0.93 – 1.13)	(0.87 – 1.10)
Frequency of contact	0.95	1.05	1.12	1.12
	(0.84 – 1.07)	(0.94 – 1.17)	(0.97 – 1.29)	(0.92 – 1.36)
Proportion network kin	0.70*	0.55***	0.47***	0.53**
	(0.51 – 0.97)	(0.42 – 0.73)	(0.30 – 0.71)	(0.34 – 0.83)
Age	0.96***	0.95***	0.96***	0.95***
	(0.94 – 0.97)	(0.94 – 0.97)	(0.94 – 0.97)	(0.93 – 0.97)
Female	0.85*	0.72**	0.70**	0.69*
	(0.71 – 1.01)	(0.59 – 0.88)	(0.55 – 0.89)	(0.50 – 0.94)
Black	1.29	1.30	1.48	1.80*
	(0.90 – 1.85)	(0.86 – 1.96)	(0.92 – 2.36)	(1.06– 3.05)
Attended college	1.47***	1.26*	1.11	0.81
	(1.18 – 1.83)	(1.01 – 1.56)	(0.84 – 1.45)	(0.57 – 1.08)
Married/partnered	1.07	1.03	0.94	0.79
	(0.87 – 1.31)	(0.84 – 1.28)	(0.73 – 1.22)	(0.57 – 1.08)
Retired	0.81*	0.77*	0.61**	0.61
	(0.66 – 1.00)	(0.62– 0.96)	(0.45 – 0.82)	(0.37 – 1.00)
Self-rated health	1.04	0.78	0.68**	0.64**
	(0.83 – 1.07)	(0.60 – 1.01)	(0.51 – 0.90)	(0.45 – 0.92)
Functional health	1.01	1.01	1.11	1.05
	(0.85 – 1.21)	(0.83 – 1.23)	(0.89 – 1.40)	(0.83 – 1.33)
Perceived neighborhood collective efficacy	0.84	0.69***	0.75***	0.74**
	(0.71 – 1.00)	(0.69 – 0.80)	(0.65 – 0.89)	(0.60 – 0.96)
Perceived neighborhood danger	1.30***	1.36***	1.60***	1.68***
	(1.16 – 1.45)	(1.21 – 1.52)	(1.38 – 1.85)	(1.37 – 2.07)
Proportion Black residents in census tract	0.55*	0.55*	0.43*	0.22**
	(0.35 – 0.89)	(0.33 – 0.92)	(0.22 – 0.83)	(0.08 – 0.57)
Proportion residents ages 65 and older	0.74	1.11	0.73	0.61
	(0.21 – 2.57)	(0.27 – 4.58)	(0.10 – 5.09)	(0.09 – 4.22)
Hispanic	0.71	0.76	1.04	0.75
	(0.47 – 1.06)	(0.50 – 1.16)	(0.64 – 1.70)	(0.38 – 1.48)
Religious service attendance	1.07**	1.04	1.06	1.04
	(1.02 – 1.12)	(0.98 – 1.10)	(0.98 – 1.14)	(0.95 – 1.13)
Constant	38.19***	12.31***	3.95*	4.24
	(9.59 – 152.04)	(3.87 – 39.20)	(1.03 – 15.15)	(0.66 – 27.21)

^*p < .05

^**p < .01

^***p < .001 (Two-tailed tests). 95% confidence intervals in parentheses.

^aAll estimates are weighted using the NSHAP Round 3 respondent-level weights that adjust for selection, non-response, and inclusion in the analytic sample, and are survey adjusted. Model is conducted using the “glogit2” package in Stata 14 (Williams 2016).

Appendix Table A3.

Average Marginal Effects (AMEs) from Logit Models Predicting Whether Respondents Experience Everyday Discrimination “Once or Twice a Year” or More Frequently.^a

	“Better than You Are”		“Less Courtesy than Others”
	AME	Standard error	AME	Standard error
Probability of experiencing discrimination more often than “once or twice a year”
White respondents, No kin in network	.29***	.03	.26***	.03
Black respondents, No kin in network	.24***	.06	.26***	.06
White respondents, All kin in network	.18***	.02	.13***	.01
Black respondents, All kin in network	.32***	.04	.26***	.04
Difference between “No kin” and “All kin”, by race (first differences)
White respondents	.12**	.04	.13***	.03
Black respondents	−.09	.08	.004	.09
Tests of second differences	.20*	.09	.13	.09

^*p < .05

^**p < .01

^***p < .001 (Two-sided tests).

^aEstimates are derived based on two logit models that predict whether respondents report experiencing each EDS item more often than “once or twice a year,” separately. The models include controls for all covariates included in the main models presented in Table 2. Tests of second difference are used to test for the statistical significance of proportion kin by race interaction terms, following the guidance for interpreting interaction terms in non-linear models outlined in Mize (2019).

Findings suggest that across both outcomes, the difference in the predicted probability of experiencing these forms of discrimination more often than “once or twice a year” is significantly higher among White respondents with no kin in their network (proportion kin = 0) compared to White respondents with entirely kin-based networks (proportion kin = 1) (p < .01 and p < .001, respectively). Among Black respondents, however, there is no statistically significant difference in the predicted probability among those with no kin in their network and those who have entirely kin-based networks. Based on the tests of second differences, the interaction between race and proportion network kin is statistically significant (p < .05) for whether respondents report that others act as if they are better than they are more than once or twice a year. The interaction is not statistically significant in predicting whether respondents report that they are treated with less courtesy than others.