Sources of Friendship and Structurally Induced Homophily across the Life Course

Abstract

How people meet new friends changes throughout life in ways that change the potential for diverse friendships. This study presents results from the first U.S. survey with data on how respondents met their friends, specifically the two nonfamily friends they most often socialize with. The most common sources of new friendships shift across life from the dominance of schooling during youth, to the centrality of work in midlife, to neighbors and voluntary groups in later life. Educational homophily peaks for friendships made in midlife, and is strongest for friendships made in higher education and at work. Racial homophily generally declines as people age but is lowest for men in midlife, while decreasing later for women. Friendship sources largely account for life course changes in racial homophily, but not educational homophily. The racial homophily induced by friendship sources also changes as people age, but in different ways for women and men.

Keywords

social networks friendship homophily gender

Friendship choice is an exemplar of a decision that is both intensely personal and profoundly shaped by social structure. It is the most voluntary form of strong ties in modern societies, entailing far more choice than family, with easier exit than long-term romances and marriages. Yet this freedom of association mostly results in friendships formed among those closely situated in social structures, creating ties that are homogeneous on most social characteristics. The informal social networks created by friendships are conduits for information, social influence, aid, and introductions to others, all of which become sequestered within group boundaries when networks are segregated (Tilly 1998), heightening inequality between groups. Crucial to the reproduction of this segregation are the sources of new friendships, which are typically the settings and networks that dominate time and identity. As the contexts people are embedded within change through their lives, so too do the characteristics of the people within them, which in turn impact the characteristics of the new friendships they produce. The life course of the sources of friendships and friendship homophily has been largely overlooked, but this study shows that the combination of when in life and through which sources friends are made has important implications for the segregation of interpersonal social networks, and the reinforcement or weakening of group boundaries.

Where and how people find new friends has been a remarkably persistent blind spot in our empirical knowledge of social life, despite attention to the importance of friendship networks in general. In fact, this is the very first nationally representative study of the sources of friendships in the United States,¹ and perhaps only the second of any country (Mollenhorst, Völker, and Flap 2008). This is also the first to show a link between the sources of friendships and racial homophily, a proposition that has long been accepted (McPherson, Smith-Lovin, and Cook 2001) but never really tested at a societal level. This study also breaks new ground by exploring how life stages and friendship sources interact to induce different kinds of homophily, specifically educational and racial/ethnic homophily. While there has been evidence of friendship sources’ effects on some homophilies, such as age assortativity (Feld 1982), and there has been evidence of life stage effects on those homophilies (Kalmijn and Vermunt 2007), the relationships between these two homophily-inducing factors have not been established. This is partly a question of mediation, whether and the extent to which the life stage effects on homophily are due to life stage effects on friendship sources. This is also a question of moderation, whether the sources of friendships change in their diversity across life. Some of the answers to these questions are different for men and women, and as such, this study adds an important new contribution to the body of research on gender differences in social networks (Smith-Lovin and Mcpherson 1993). As life courses are gendered, changes in friendship sources and the homophilies they induce across life are likewise gendered.

I begin by defining the sources of friendship in terms of social foci and network brokerage, and how they structurally induce homophily. I then discuss what our existing knowledge suggests should be the primary sources of friendships in each major life stage, gender differences in these, and the typical racial/ethnic and social-class diversity of potential friends within those sources. These descriptions lead to hypothesized relationships between life stage, friendship sources, gender, and structurally induced homophily.

Social Foci, Catalyst Brokerage, and Structurally Induced Homophily

Potential new friends are a small subset of the overall population, constrained by the opportunities presented through the social structures people participate in (Blau 1977). These opportunity structures are constrained not only by geography but also by the formal and informal groups, settings and activities that constitute social lives, or “social foci” (Feld 1981). All foci entail opportunities for creating new social ties, but these opportunities are not equal. Settings vary in the quality, quantity, and cultural appropriateness of interactions within them, and may hold different friendship-creation potential for different subgroups of people, which can also vary by life stage.

The sources of friendships can also be thought of in terms of network brokerage, specifically catalyst brokerage (Obstfeld 2005; Stovel and Shaw 2012), the third-party introduction and facilitation of new relationships. Introduction to potential new friends through family and existing friends is a network process of triadic closure (Granovetter 1973), but so too is introduction through co-participation in an organization, when viewed as a two-mode network of persons and groups (Breiger 1974). Even informal social foci can broker new ties without interpersonal brokerage, such as a self-introduction between unacquainted surfers at a popular surfing spot. While it may seem strange to attribute the social action of brokerage to nonpersons such as beaches or churches, conscious decision making is not required for interpersonal brokerage either, as people often accidentally act as conduits for new friendships among those they know, sometimes without being aware of it. Just as social settings can be conceived as brokering nodes in networks, social network entities such as families or friendship circles, can also be considered social foci (Feld 1981:1018–19). Friendship origins typically begin within a social focus, and entail catalyst brokerage that is interpersonal and/or organizational.

These settings and network components vary in their composition along many social dimensions, and thus the potential diversity of friendships begun within them likewise varies. Whether we call it homophily, segregation, or assortativity,² disproportionate homogeneity in relationship pairings is a likely outcome from disproportionate homogeneity in the social subunits that relationships begin within (Blau 1977; Feld 1982). This is structurally induced homophily, often contrasted with choice homophily (McPherson and Smith-Lovin 1987), or the tendency toward homogeneous ties that can be attributed to individual preferences. Also referred to as supply-side versus demand-side perspectives (Mollenhorst et al. 2008), this distinction does not include segregating mechanisms that can exist in between opportunities and preferences, such as easier communication between those with more shared cultural knowledge (Carley 1991), or interaction dynamics that temporarily alter perceptions of group boundaries (Lewis 2013). Yet among the wide array of potential homophily mechanisms, the supply constraints on opportunities for interaction set the range of possibilities for diverse relationships before most other mechanisms can even come into play.

The Structure of Opportunities for Friendship and Homophily across the Life Course

The pools of potential new friends that people encounter can vary dramatically across their life courses, and so too the types and extent of structurally induced homophily can vary by life stage. The descriptions below, following earlier work on social foci and homophily (McPherson et al. 2001), are based on generalizations about the recent structure of social life in the United States but are general enough to apply more widely. Life stages are not universal nor universally sequenced (Elder 1975), however, and the foci discussed in the following often overlap within lives.

Childhood Foci: Family and Residential Propinquity

The first social ties in life are typically made through the family, negotiated by parents, creating initial networks that are reflections of caregivers’ networks. Ties based on residential proximity are often the first childhood friendships that are not strictly caregiver-redundant. School settings soon come to dominate children’s time, and schools’ internal networks and hierarchies (McFarland et al. 2014) typically dominate their social lives. Most schools reflect the composition of the neighborhoods the students live within (Rivkin 1994). Many children also become involved in youth voluntary organizations, sports, and clubs, which typically connect them to homogeneous subsets of their schools and neighborhoods (Clotfelter 2002). The general effect of these youthful foci are to create early friendships that reflect neighborhood composition, which demographically resemble families: homogeneous by race and ethnicity, religion, and social class (Charles 2003; Lichter et al. 2007; Massey and Denton 1993). However, note that all of these homogeneities are with respect to the child’s social statuses at the time of friendship formation. Most of these statuses will not change over the life course, but significant social-class mobility from childhood to adulthood can render once homogeneous childhood friendships more class-diverse later in life.

Social Foci in Adulthood

The transition into adulthood typically entails social opportunities less correlated with residential proximity. For many, including most of the newer generation of adults (Ryan and Bauman 2016), the transition from childhood to adult social foci begins through postsecondary education, including colleges and vocational training, even for those who do not eventually earn a degree. The full spectrum of these ranges from total institutions that entirely encompass their students’ lives to more instrumental settings with few out-of-classroom social opportunities, but they typically involve exposure to a more racially and ethnically diverse range of people than childhood schooling entailed (Hinrichs 2015). The role of colleges in social stratification, however, ensures a degree of social-class homophily in the friendships they facilitate, as they produce a more educationally homogeneous group of alumni from the body of students that enter them.

Work dominates most adults’ pre-retirement life, typically drawing together coworkers from multiple and diverse neighborhoods. Workplaces tend to be segregated along most social dimensions but to different extents than the youth settings discussed earlier. Although workplace segregation by race in the United States remains quite significant (Tomaskovic-Devey et al. 2006), workplaces are on average much more racially and ethnically diverse than the other settings in peoples’ lives. In fact, coworker is the role in Americans’ ego-networks most likely to be interracial (Marsden 1990). This greater racial diversity, however, is matched by stronger social-class segregation. While work settings can be educationally diverse, the norms of the workplace, structures of collaboration, and professional homophily all strongly encourage friendships between occupational equals (Brass 1985; Lazega and van Duijn 1997). Coworkers have been found to be more educationally similar than other types of relationships (Marsden 1990; Mollenhorst et al. 2008). The effects of the workplace on friendship homophily may differ by gender, however, as men continue to spend more of their lives formally employed than do women (Toossi and Morisi 2017), and men’s social networks’ size and role composition have been shown to be more affected by their employment status (Moore 1990).

Adults who become parents are drawn back into children-centered foci. Parenting duties continue to disproportionately fall to women (Umberson, Pudrovska, and Reczek 2010), affecting women’s daily lives and networks more strongly. New parents often experience an initial reduction in the size of their ego networks (Bost et al. 2002), but this rebounds as they are soon drawn into local networks through their children’s activities, typically with residentially proximate parents of similarly aged children (Klärner, Keim, and der Lippe 2016). Friendships formed in child-centered settings may be similar in homophily for both the parents and their children, but this can be quite different from the structurally induced homophily the parents experienced in their childhoods. Each recent new cohort of Americans has been more racially and ethnically diverse than the last (Frey 2014), and as a result, the baseline probability for childhood interracial friendships has been increasing. While U.S. parents are part of a more homogeneous generation of Americans than their children, the parents of their children’s friends will be a disproportionately racially diverse subset of the parents’ generation.

Adult social life outside of family, education, and work can be structured by organized groups and activities that vary in their formality. These can include sports, hobby clubs, political groups, religious organizations, fraternal organizations, adult enrichment classes, volunteering, and so on. For some people, such activities are primary sources of friendships, for others not at all. While the voluntary organizations of youth are mostly drawn from within schools and neighborhoods, often with heavy membership overlap, adult organizations are more likely to draw from a wider geographic slice, with less overlap. Considerable variation exists in the diversity within these organizations. Organizations that primarily recruit through occupations or occupational interests tend to have little social-class diversity (McPherson and Smith-Lovin 1987), but their racial and ethnic diversity depends on the diversity of their corresponding occupations. Organizations based on ethnicity, religion, or cultural interests that are correlated with ethnicity can be expected to be more segregated on that dimension, but perhaps less by social class. Churches have been characterized as the most racially segregated subunits of U.S. society (King 1963), and the great majority of religious congregations continue to be homogeneous by race and ethnicity (Edwards, Christerson, and Emerson 2013). Voluntary organizations have also historically differed in size by their gender composition, with men participating in larger groups than women (McPherson and Smith-Lovin 1982). Larger groups tend to produce more expansive networks of weak ties but more homogeneous strong-friendship ties within them (McPherson and Smith-Lovin 1987). Women tend to be more involved in religious activities than men, and also tend to perceive inter-religious differences as more important (Brashears 2008), which may lead their networks to more resemble the composition of religious congregations.

Later life stages are characterized by a shift away from workplace and child-centered settings, and toward voluntary and religious organizations, existing family and friendship networks, and residential and neighborhood social foci (Cornwell, Laumann, and Schumm 2008), though the extent of this varies by life course trajectory (Crosnoe and Elder 2002). Gender inequality in network size reverses later in life (Fischer and Oliker 1983), as men’s networks shrink, while women’s tend to expand and become less kin-based (McLaughlin et al. 2010). This may cause women’s networks to see a late life bloom in race and class diversity, while men’s may see a decline in the racial diversity gained through work.

Introductions to new friends through existing relationships occur throughout life. This is a transitive process, which has been shown to amplify homophily (Goodreau, Kitts, and Morris 2009), creating more homogeneous ties than would be expected from personal preferences alone. One’s friends’ friends tend to be similar to one’s friends, so befriending them typically creates more homophily. However, interpersonal introductions can also act to desegregate when the broker bridges group boundaries, or when a diverse foci creates friendships that in turn broker more intergroup ties (Mark and Harris 2012). Even friends made in the racially homogeneous settings of childhood age into more diverse adult settings, potentially bridging those networks. Men’s and women’s networks differ (Smith-Lovin and Mcpherson 1993) in ways that have important implications for the homophily induced by introductions. Among children, boys’ friendship circles tend to be more expansive than girls’ (Eder and Hallinan 1978). This continues into adulthood, as men’s networks more often bridge into diverse areas of social space (Brashears 2008), while women’s are more often densely interconnected with redundant pathways (Smith-Lovin and Mcpherson 1993). Men’s networks have been shown to be more affected by life transitions such as divorce (Kalmijn 2007; McLaughlin et al. 2010), employment status (Moore 1990), and retirement (Fischer and Oliker 1983; Kalmijn 2012), while women’s networks are more affected by parenthood (Campbell 1988). Even once factors such as employment are accounted for, women’s networks tend to be more kin-based than men’s, and women’s employment status affects the kin-proportion of their networks, while it does not for men (Moore 1990). Friendship introductions may be more likely to produce a diverse connection for those with more expansive networks, but more often foster demographically similar new ties for those with kin-heavy networks. The greater stability of women’s networks across most life course transitions may limit the impact of life stages on their homophily, compared with men.

Summary and Hypotheses

To summarize, racial and ethnic homogeneity within friendship formation settings is expected to be at its peak at the beginning of the life course, when social foci are centered on family, residential proximity, and residentially derived organizations. The stages of life centered on work then represent the peak potential for interracial and interethnic friendships, as workplaces and their related midlife networks and settings create more diverse opportunities. Postwork social foci may entail less racial diversity but are still likely less segregated than youth foci. Social-class homogeneity in new friendships is expected to be greatest in workplace-dominant midlife. Early life settings are segregated on the parents’ social class, and intergenerational social mobility results in some once-homogeneous childhood friendships becoming class-diverse in adulthood. As adults enter later life stages more centered on communities and voluntary organizations, the potential for social-class diversity in new friendships increases.

Hypothesis 1: Friendships formed in midlife are the most likely to be drawn from work, compared with early life friendships drawn more from schooling, family, and neighborhoods, and later-life-formed friendships drawn more from voluntary organizations and neighborhoods.

Hypothesis 2: Friendships formed through family and early schooling are more racially and ethnically homogeneous than those formed through other sources, while those formed through work are the most diverse.

Corollary 2: Assuming Hypotheses 1 and 2, then friendships formed in early life are the most racially and ethnically homogeneous, while those formed in midlife are the least.

Hypothesis 3: Friendships formed in the workplace and through postsecondary education are the most homogeneous by social class.

Corollary 3: Assuming Hypotheses 1 and 3, then friendships formed in midlife are the most homogeneous by social class.

While the aforementioned corollaries concern the life stage effects on homophily that occur through the mechanism of social foci change across the life course, I also hypothesize that there are life stage effects on homophily not mediated by the sources of friendship. This can occur from second order and “ripple effects” of social foci: If a setting is very dominant within a life stage, it can become the pathway into other major friendship sources as well and produce homogeneity across social foci that is similar to the dominant setting. For instance, if a person’s midlife is largely organized around their workplace, that organization can also determine the voluntary organizations and informal activities they engage in, and the composition of much of their acquaintanceship network. All of these would then induce similar homophily. This can likewise occur in education-focused stages of life, or in a life stage based around a residential community. In addition to these structural ripple effects, there also may be changes in choice homophily across the life course, and these may coincide and align with the shifts in the dominate settings at each life stage. This could occur as a cultural ripple effect of a dominant social focus, as it becomes the central social organizing principle of a life stage.

Hypothesis 4: Friendships formed in midlife are the least racially homogeneous, regardless of the source of friendship.

Hypothesis 5: Friendships formed in midlife are the most homogeneous by social class, regardless of the source of friendship.

Gender differences in foci and networks across the life course lead me to hypothesize that life stage effects on racial and social-class homophily should be stronger for men than women. The mid-life dominance of work is typically stronger for men, and more strongly affects their social networks (Moore 1990), and so the related midlife reduction in racial homophily and peak in social-class homophily should be strongest for men. Women, on the other hand, typically see a late life expansion of their personal networks, as men’s shrink, and this expansion is related to an increase in nonkin ties (Fischer and Oliker 1983). This should result in a reduction of racial homophily later in life for women, whereas men’s racial homophily can be expected to rebound as the effects of workplace networks fade and their personal networks shrink. The overall effect should be flatter relationships between life stage of friendship onset and both homophilies for women compared with men.

Hypothesis 6: The life stage of friendship formation has a stronger effect on racial homophily for men than for women, with less of a midlife homophily trough for women.

Hypothesis 7: The life stage of friendship formation has a stronger effect on social-class homophily for men than for women, with less of a midlife homophily peak for women.

Method

In the analyses that follow, I use a unique U.S.-representative dataset, the Geographic Mobility and Homophily Survey (GMHS), which includes open-ended responses on how people met the two friends they most often socialize with. I designed the survey, and contracted Knowledge Networks (KN; since acquired by the German company GfK) to administer it to a sample of 1,077 respondents in January of 2007. KN selected these respondents at random from their panel of more than 40,000, which were in turn drawn from a probability sampling of the U.S. population, selected through a stratified national telephone random digit dialing method. Persistent recruitment by registered mail and telephone follow-ups resulted in an initial panel recruitment rate of 32.6 percent, reduced to 18.5 percent when factoring in the completion rate for the basic demographic survey (Callegaro and DiSogra 2008). KN’s surveys were conducted through a web interface, and they supplied respondents with Internet access through a television set-top box, if needed. Respondents received three to four surveys a month, which required an average of 15 minutes to complete. Out of 1,666 panel members who received my survey, 1,077 respondents (or 64.6 percent) completed it.³ KN samples have been show to provide more accurate population estimates than traditional random digit dial telephone surveys (Chang and Krosnick 2009; Fricker et al. 2005). See Appendix Table A1 for unweighted descriptives of all measures included in the analyses that follow.

Sampling Friendships

The ego network generator (or name generator) used in this survey focuses on identifying nonkin and nonsignificant-other relationships that constitute the core of the respondents’ social lives. The survey began with the instruction, “Think about the two friends you most often socialize with face-to-face. Do not include family members or boyfriends/girlfriends.” I chose the phrase “most often socialize with” to elicit the friendships most involved in the respondents’ nonwork, recreational time. This did not ask the respondents’ to choose based on strength of bond or affection but purely on frequency of shared social activities, in the present tense. Claude S. Fischer’s (1982) comparison of ego network name generators found that joint social activities was the best predictor that the alter would be labeled a “friend,” better than discussing personal matters, or any of the other several name generators he used. The specification “face-to-face” discouraged respondents from listing long-distance confidants or other people who were outside of their regular social activities. Although nonlocal ties can be very strong, and important in many ways, they cannot provide many types of support that local ties can (e.g., child care), cannot participate in most physical social events (dinners, parties, etc.), and are thus much less likely to have ties to other members of the respondents’ local social networks (Martin and Yeung 2006), or introduce new alters. The focus on local friendship ties in this study is an effort to measure the characteristics of the cores of the nonkin informal social networks that respondents are embedded within. Twenty-seven of 1,077 respondents declared that they had no such friends (2.51 percent), and 23 responded that they only had one such friend (2.14 percent), while more than 95 percent of the respondents gave information about two of their friends.

Social Foci and Brokerage: How They Met

The survey asked respondents how they met each friend, and provided a text box to type in their answers in their own words. The majority of answers were very simple and to the point, such as “at church” or “work.” Some answers were too ambiguous or minimal to provide any clear information, such as “since childhood” or “at home,” which I coded as missing. I defined voluntary organizations loosely to include hobbies, sports, and similar activities as well as more formal organizations. I use the label “college” to include all postsecondary education, including when respondents used the word “school” to describe an educational setting in adulthood. I coded as “neighbor” friends who were introduced by a neighbor, but the great majority of this category are friends who are/were the respondents’ neighbors. A small number of answers involved chains of introductions, such as “he’s the brother of my wife’s friend.” In these instances, I coded the response by the connection most immediate to the respondent (e.g., spouse, in the previous sentence), as the most likely gateway to the opportunity for meeting this person. I condensed these codes into two categorization schemes: a detailed 11-category coding, and one of five broader categories. The later simpler scheme is useful in easing model convergence when friendship source is the outcome variable, as well as maximizing statistical power in all models. It also simplifies interpretation of model results, which is particularly helpful when interacting those categories with other independent variables. However, there are some interesting differences in the structurally induced homophily among the more detailed categories, so I include results using both the five- and 11-category schemes of friendship sources in the models predicting homophily.

Alters’ Characteristics

The survey asked for each friend’s race/ethnicity and highest educational degree. The measure of the respondents’ alters’ race and ethnicity mirrored the KN panel’s question, which included Hispanic as one of the mutually exclusive options, along with White and Black, but combined Asian and Native American with the remaining Other Race category. Nonetheless, this simple measure does map onto the major social divides in U.S. society (Massey and Denton 1993). Education is only a partial measure of social class (or a proxy), but it is strongly predictive of the social and economic outcomes associated with class in the United States (Hout 2012). The question about alters’ highest educational degree also mirrors KN’s, distinguishing between less than high school, a high school degree or equivalent, some college, a college degree, or a graduate degree. To measure the educational heterogeneity of friendships, I use the absolute difference between the respondent’s and the friend’s value on this 5-point scale. Models that instead predict heterogeneity by a binary indicator of college degree status, or predict a less-versus-same-versus-more educated friend with a multinomial model, produced consistent results (see Appendix Figure A2 and Appendix Table B4).

Age of Friendship Formation

Ideally, the life stage of friendship formation could be most accurately determined by using detailed data on the respondents’ timings of major life events and transitions in conjunction with their timing of friendship formation. This data set lacks that level of detail, so I use the age the respondent met each friend as a proxy for the life stage the friendship was formed within. This may be a coarser measure, as there is variance in the timing of life transitions (Shanahan 2000). However, for the broadly defined life stages crucial to the hypotheses, keyed on the transition from schooling to work and parenting, and away from both later in life, the institutionalization of timing is strong (Meyer 1986): The variance between people is measured in years, not decades (Ravanera, Rajulton, and Burch 2004), and as such, age is a good proxy here.

Other Control Variables

In the models that follow, I control for a number of potentially confounding factors, including respondent’s gender, age, marital status, children in the household, employment, and income. I also control for the frequency of the respondents’ contact with relatives: Variance in embeddedness within extended family networks could be related both to how/when friends are met and to the homogeneity of those friendships. The survey asked, “How often do you see or visit relatives? (including your parents, grandparents, aunts/uncles, etc.),” with responses on a 6-point scale, including “almost every day,” “almost once a month,” to “less often than once a year or never.” Contextual controls include U.S. region and the population size of the place or city the respondent lives in, as well as measures of local diversity, including the percentage of college graduates in their current zip code, and the percentage of people in the same racial/ethnic category as the respondent, both of their zip code and their metropolitan statistical area or county.

Models

The models presented here predict sources of friendships and friendship heterogeneity, using friendship dyads as the unit of analysis, with one or two friendships nested within each respondent. I use multinomial logistic regressions to predict friendship source, binomial logistic regressions to predict whether the friendships cross racial/ethnic categories, and ordinary least squares (OLS) regressions to predict the absolute difference between ego and alter on the educational degree scale. The national sampling strategy makes it highly unlikely that any of the respondents are also alters of another respondent, so network autocorrelation is not a concern. I report single-level models with clustered standard errors here, but results were substantively consistent when using multilevel models. All models are adjusted with sampling weights provided by KN. Below, I interpret the logit coefficients as odds ratios, but I also report average marginal effects, as both predicted probabilities and changes in predicted probabilities, reporting p values from z tests where appropriate. Some of the hypotheses posit nonlinear relationships between age at friendship formation and outcomes, such as a midlife minimum in the educational heterogeneity of new friends, which I test with a squared term for age-met. See Appendix Tables A2 to 3 for robustness tests comparing the effects of age at friendship formation using categorical versus linear and polynomial coding. I exclude 12.9 percent of the cases from the analyses that follow due to missing data on some of the variables, and also exclude one of the two possible friends for 8.8 percent additional cases. Listwise deletion of cases with missing data is the most conservative solution for assessing statistical significance without underestimating uncertainty (Allison 2009), and the results presented here are consistent with the results from models using multiple imputation of the missing data.

Results

Sources of Friendship across the Life Course

The foci and brokers that introduce new friendships vary significantly across the life course. The top panels of Figure 1 display locally weighted regression lines for the stacked proportions of friendships made through each source, for women and men separately, by the respondent’s age when the friend was first met.⁴ The general pattern is as predicted by Hypothesis 1. The most common sources of new friendships are very distinct by age: Educational settings are the biggest source in early life, workplace-formed friendships predominate in the middle stages, and brokerage through neighboring (as neighbors or introduced by neighbors) is most common in the later stages of life. The formally organized settings of education and work are the most exceptionally life stage dependent, brokering almost no friendships outside of their dominant stages. Voluntary organizations become a more common friendship source later in life, while religious organizations are more steadily a minority source of friendships across life. Introductions through friends and family, viewed together, are also a mostly steady source of friendships, but the specific type of kin brokerage changes with age, with parents and other relatives the most common introducer of children’s new friends, spouse and children-brokered ties more common in adulthood. Neighbor ties, as expected, are a significant source of lasting friendships from early childhood but decrease in importance in midlife, and then become the most common source of new friends late in life. There are some notable gender differences in the sources of friends: Women’s friendships are less often made at work than men’s, by a difference of 6 percent (p < .05) and more often made through their children, by 3.8 percent (p < .01). Lumping together introductions through children, spouses, and other family, women’s friends are 3.7 percent more often made through familial brokerage than men’s (p < .05). The broad relationships visible in Figure 1 between the age a friend was met and the source of the friendship hold when controlling for potentially confounding factors. See Appendix Section A for discussion of results from models predicting how the friend was met from age-met (Appendix Table A2).

Figure 1.

Stacked proportions of how friends were met, by gender, age-met, and age at survey.

Figure 1 also shows the local regression lines for stacked proportions of friendship sources by the respondent’s age at the time of the survey (bottom row), again by gender. This illustrates that the importance of friendship sources often lasts well beyond the life stage they occur within. Even though friendships formed in educational settings begin almost entirely during youth, they represent a substantial portion of friendships throughout life. For U.S. adults in their 40s, an estimated 21.6 percent of their core friendships began in school or college, with this estimate still over 11 percent for those over the age of 65. Likewise, workplace-formed friendships remain common after the typical retirement age. The persistence of friendships has a fairly smooth and heteroskedastic linear pattern, with age of friendship formation both increasing in mean and variance as age increases. For instance, Americans in their 40s met their friends on average at age 29.3, with an observed interquartile range from 20 to 38, while those older than 65 met their friends at an estimated mean age of 44.4, with an observed interquartile range from 25 to 60. Friendships from the earliest life stages are common in every age group, but so too are new friendships.

Structurally Induced Homophily across the Life Course

Racial and ethnic homophily

Sources of friendship are predictive of racial and ethnic homophily in ways consistent with Hypothesis 2. Table 1 displays logistic regression coefficients and odds ratios predicting a racially/ethnically diverse friendship, using both the broader five-category scheme of friendship sources as well as the more detailed 11-category scheme. Compared with friendships formed in workplaces, friendships made in educational settings are much less likely to cross racial and ethnic boundaries, with less than one quarter the odds (Table 1, Model 2). The educational effect is even stronger when focusing on friends made in pre-adulthood schooling (Model 4), which have less than one-fifth the odds of being interracial compared with work friends. Friends made through informal social networks of kin or other friends have less than half the odds of crossing racial and ethnic boundaries as work friends. Of the more specific types of introductions, however, only spousal introductions have a statistically significant (and very strong) negative effect on racial diversity compared with work. Nonreligious voluntary organizations also produce more racially homogeneous friendships than do workplaces, but neighbor friendships and introductions through children are predicted to produce roughly the same racial friendship diversity as workplaces.

Table 1.

Logit Coefficients and Odds Ratios Predicting a Racially/Ethnically Diverse Friendship (N = 1758).

	(1)	(2)		(3)		(4)		(5)
	b	b	OR	b	OR	b	OR	b	OR
Age met friend	0.027**			0.013				0.010
Friendship source (vs. work)
Educational organization		−1.531***	0.216	−1.364***	0.256
School						−1.686***	0.185	−1.537***	0.215
College						−1.230^†	0.292	−1.149^†	0.317
Family/Friends		−0.738*	0.478	−0.702*	0.496
Family						−1.538^†	0.215	−1.398	0.247
Spouse						−1.850*	0.157	−1.867^†	0.155
Children (as parent)						−0.040	0.961	−0.070	0.932
Other friend						−0.549	0.577	−0.523	0.593
Voluntary organizations		−0.757^†	0.469	−0.750^†	0.472
Nonreligious						−0.976*	0.377	−0.993*	0.370
Religious						−0.561	0.571	−0.543	0.581
Neighbor/Public		−0.003	0.997	−0.022	0.978
Neighbor						0.096	1.101	0.077	1.080
Public place						−0.269	0.764	−0.264	0.768
McFadden’s pseudo R²	.441	.455		.456		.460		.461
BIC	1,116.053	1,115.914		1,121.020		1,151.715		1,157.858

Note. All models control for age, gender, race/ethnicity, household income, education level, U.S. nativity, marital status, frequency of contact with relatives, region of the United States, population size of metro area, % of metro area R’s race/ethnicity, % of zip code R’s race/ethnicity, % of zip code college graduates. For full models, see Appendix Table B2. OR = odds ratio; BIC = Bayesian information criterion.

†

p < .1. *p < .05. **p < .01. ***p < .001.

The panels on the left in Figure 2 display predicted probabilities of interracial friendship for each setting, using average adjusted predictions from Models 2 and 4, with the top left panel using the broader categories of friendship sources, the bottom left using the more detailed categories. Among the broad categories, education-formed friends stand out as the least likely to be racially diverse, at a predicted rate of 12.6 percent, while workplace- and neighbor/public-formed friendships stand out as the most likely to be interracial, both at 23.3 percent. Introductions through family/friends and voluntary/religious organizations also have very similar predicted rates of diversity (17.4 and 17.3 percent, respectively), in between the extremes of schooling and work. The differences in the predicted rates between family/friends and workplaces (p < .05) and neighbors/public (p < .05) are both statistically significant, but the difference between family/friends and educational settings are not. Voluntary/religious organizations’ predicted rate of interracial friendships is only significantly different from that of workplaces (p < .05). For the most part, the more specific sources of friendship have similar effects on racial friendship diversity within their broader categories, with the exception of family/friend introductions. Friendships made through one’s children are predicted to be surprisingly racially diverse, with a similar rate (22.9 percent) to those made at work (23.3 percent) and through/as neighbors (24.1 percent). Friendships made through spouses (11.0 percent) and other family (12.6 percent), however, are among the least likely to be racially diverse. Note that while most social foci effects on racial homophily do not differ significantly between racial/ethnic groups, I do find that Black Americans are more likely to find a different-race friend through schooling rather than through friends and family, whereas the opposite is true for White Americans (interaction effects model not shown, p < .05). Social foci effects on racial homophily are not significantly different between educational or income groups.

Figure 2.

Predicted probability of racial/ethnic heterogeneity and educational degree difference in friendships, by friendship source.

The age at which a friendship began is linearly predictive of racial/ethnic heterogeneity, partially supporting Corollary 2, once potentially confounding factors are controlled for (Table 1, Model 1). Each year later in life that a friendship began increases the odds that it is interracial by 2.7 percent, which nearly doubles the odds with a twenty-five-year difference in age-met. The predicted curvilinear relationship between age-met and racial heterogeneity is not generally supported (but see the discussion of interactions with gender that follows): Polynomial age-met terms are not significant (not shown), and the age-met categories’ effects (see Appendix Table A3) do not clearly exhibit a nonlinear pattern. The top left panel of Figure 3 illustrates the linear relationship from Model 1 with average adjusted predictions. The range of these predicted probabilities is a little over .15. Given that 17 percent of friendships in these data were observed to be interracial, this represents a substantial change across life. Contrary to the prediction of Hypothesis 4, the effect of the age of friendship formation on racial homophily is largely mediated by the setting of friendship formation (Models 3 and 5): The inclusion of the 11 categories of friendship sources reduces the age-met coefficient by more than 63 percent and renders it statistically insignificant. Both workplace (p < .05) and educational settings (p < .01) each separately have statistically significant mediating effects on the age of friendship formation’s coefficient predicting racial homophily (see Appendix Table B5 for the full mediation test models, following Sobel [1986]). On the other hand, the differences in homophily between social foci are not changed much by controlling for the age of friendship formation.

Figure 3.

Predicted friendship diversity by respondent’s age when friend was met.

Educational homophily

Friends who met through work are more educationally homogeneous than those who met through family/friend introductions and neighbors/public settings (Table 2, Model 3), as predicted by Hypothesis 3. Looking at more specific social foci, work friends are more homogeneous than friends made through other friends, spouses, public places, and neighbors (Table 2, Model 6). The sizes of these significant effects are all within a range of one sixth to more than one third of step increase in the five-category highest degree scale. Counter to expectations (Hypothesis 3), however, workplace friends are not significantly more educationally homogeneous than those made through schooling or voluntary organizations, including religious settings. The two rightmost panels of Figure 2 display predicted education scale differences for each friendship source, from Models 3 and 6. Here, education and work settings roughly line up as the most homogeneous friendship sources (top right panel), while family/friend brokerage and neighbors/public settings are predicted to be the most educationally diverse. Looking at the more specific categories (bottom right panel), college friendships are predicted to be exceptionally educationally homogeneous, significantly more so than those from every other source except for work. Workplace friendships are predicted to be less educationally diverse than those made as neighbors (p < .01), through one’s spouse (p < .05), in public settings without an introduction (p < .05) and through other friends (p < .05). Religious organizations are also predicted to be particularly educationally homogeneous, more so than neighbors (p < .05) and spouse-introduced friends (p < .05). These effects do not differ significantly by gender, race, education, or income.

Table 2.

OLS Regression Coefficients Predicting Absolute Educational Scale Difference Between Ego and Alter (N = 1,758).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Age met friend	0.001	−0.015**		−0.001	−0.018**		0.000	−0.016*
(Age met friend)²		0.000**			0.000**			0.000*
Friendship source (vs. work)
Educational organization			0.036	0.028	−0.060
School						0.093	0.096	−0.004
College						−0.152^†	−0.151^†	−0.175^†
Family/Friends			0.205**	0.204**	0.175*
Family						0.223^†	0.225^†	0.129
Spouse						0.353*	0.353*	0.325*
Children (as parent)						0.094	0.094	0.093
Other friend						0.209*	0.209*	0.191†
Voluntary organization			0.108	0.109	0.073
Nonreligious						0.198^†	0.197^†	0.159
Religious						0.036	0.036	0.011
Neighbors/Public			0.231**	0.232**	0.187*
Neighbor						0.228**	0.227**	0.182*
Public place						0.245*	0.245*	0.222^†
Adjusted R²	.064	.071	.075	.074	.082	.079	.078	.083
BIC	4,173.047	4,165.553	4,172.348	4,179.691	4,171.792	4,203.274	4,210.739	4,207.692

†

p < .1. *p < .05. **p < .01. ***p < .001.

The expected nonlinear relationship between the age a friend was met and educational homophily, peaking in midlife (Corollary 3), is supported by both the age-met category effects (see Appendix Table A3) and the polynomial regression models (Table 2, Models 2, 5, and 8). Figure 3’s top right panel illustrates the curvilinear age-met-squared effect from Model 2: The transition from childhood to early adulthood entails less educationally diverse new-friendships, but the transition to later-middle adult life stages increases their diversity once again, soon producing more educationally diverse friendships than ever. While this curve seems to neatly mirror the rise and fall of the importance of work in the life course, this relationship remains unchanged when controlling for the sources of friendships (Table 2, Models 5 and 8), supporting Hypothesis 5. Unlike racial homophily, the age-met effect on educational homophily is not significantly mediated by friendship sources (not shown, following Sobel [1986]), meaning that there is no statistical evidence here that the age-met effect on educational homophily is even partly due to the age-met effect on friendship sources.

Gendered Differences in Structurally Induced Racial Homophily across the Life Course

There are no detectable gender differences in the changes in educational homophily across the life course (not shown), contrary to Hypothesis 7, but there are significant differences between men and women in how their racial homophily changes as they age, supporting Hypothesis 6. The linear age-met effect on racial homophily is much stronger for men than for women (p < .05; Figure 3, middle left panel), but the gender difference in the squared effect of age-met (p < .01) is more telling (Figure 3, middle right panel; see Appendix Table B6 for full interaction model coefficients). The hypothesized nonlinear relationship between age of friendship formation and racial homophily (Corollary 2) is observed for men, with their interracial friendship potential peaking in later-midlife (a 27.5 percent probability at age 51), and then dropping as they age. Women’s potential for making new friends that are interracial does not peak, but instead rises later in life, and is largely flat before late adulthood, with a slight trough in early adulthood (a 17.8 percent probability at age 23). This may correspond to the previously noted later life expansion of women’s networks, and the dominance of kinship ties prior to that (Fischer and Oliker 1983; McLaughlin et al. 2010). Note that the models control for current employment status, but they do not control for employment at the time of friendship formation, which may be responsible for some of this gender difference. However, men’s networks have been shown to be more dependent on their employment status than are women’s (Moore 1990), and as such, men’s racial friendship homophily may simply be more affected by transitions into and out of the working stages of life.

There are no significant interaction effects⁵ between gender and friendship sources predicting either type of homophily, meaning that, in general, I do not find that social foci induce racial or educational homophily differently for women than for men. I also do not find any significant change in the homophily effects of social foci across the life course, in general, for either homophily. But I do find that the racial homophily associated with social foci changes across the life course in significantly different ways for women and for men. The bottom panels of Figure 3 display average predicted probabilities from a three-way interaction between gender, age at friendship formation, and friendship source (see Appendix Table B6 for the full model’s coefficients). To aid in both the interpretation and statistical power for this interaction, I only use the five-category coding of friendship sources. I also omit the school/college category from these figures (but not from the models), as they are so confined to early life stages that most of the area of these graphs are outside of the range of observations for those foci, and their extrapolated effects into later life would not be reliable. Looking at gender differences in the structurally induced homophily of social foci across the life course, we see that workplaces and introductions through friends and family become increasingly likely to introduce men to new friends of a different race/ethnicity as they age. For instance, men’s friends formed at work at age 22 are predicted to be 18.9 percent interracial, while those made at work at the typical retirement age of 65 have a 41.8 percent probability to cross racial/ethnic lines. The opposite is true for women: As they age, workplaces and family/friends introduce them to more racially homogeneous new friendships than earlier in life. Women’s work friendships made at age 22 are predicted to be 27.5 percent interracial, but by age 65, their new work-formed friends have only a 15.5 percent probability of being interracial. Voluntary organizations show the opposite pattern, becoming more racially segregated for men’s friendships as they age, but slowly increasing the likelihood of interracial friendships for women as they get older. Neighbor friendships show a similar life course trend for both genders, becoming more interracial as people age. As the transition into later life increases neighbor and voluntary organization-formed friendships at the expense of work and other sources, for both men and women, these gender differences in induced racial homophily help create the downward turn in interracial friendships observed for men later in life, and the increase for women.

These interaction effects also predict that, at some life stages, the sources of friendship do not differ much in the racial homophily they induce, but this occurs at different life stages for men and women. Among men, friendships made in their 20s have similar rates of racial homogeneity regardless of where they were formed (including through schooling and college), while the interracial rates of women’s friendships made in their 20s are significantly affected by their source. At age 20, women’s new friends made through work have the highest probability of being interracial, on average an 8.8 percent higher predicted probability compared with neighbor/public-setting-formed friendships (p < .05), a 12.7 percent higher probability than voluntary and religious organization friends (p < .05), and a 17.1 percent higher probability compared with friends met through education (p < .001). Of friendships made between 40 and 60, women’s friendships do not differ significantly in racial homophily by their source, but men’s do. At age 50, men’s new friendships made through voluntary and religious organizations are the least racially diverse, with a 16.9 percent lower predicted probability of being interracial compared with work-formed friends (p < .01), and 14.9 percent lower compared with neighbor/public settings friends (p < .05). Men’s early adulthood social foci are aligned in racial diversity, and then diverge as they age, but for women, this alignment occurs in later-midlife, with divergent foci-induced racial homophily earlier and later.

Discussion

There is clear evidence here that sources of friendship change across life, and are related to differences in the racial/ethnic and educational homophily of the friendships they produce. The workplace and educational settings stand out as the most exceptionally life-stage-dependent sources of friendships, dominant within their stages, and strongly related to both the racial/ethnic and educational composition of friends met within them. Neighbor and voluntary organization friendships are also life-stage dependent, most common later in life, and neighbor friendships are more likely to be racially/ethnically diverse than previous literature has suggested (Marsden 1990). There is also clear evidence that the life stage of friendship formation affects homophily, but with important gender differences. Age at friendship formation is linearly related to racial diversity, though for men this diversity peaks in midlife, while for both genders there is a midlife trough in educational diversity of new friendships. The life course effect on racial homophily is largely statistically accounted for by the effect of where and how friends were met, but this is not the case for educational homophily. The midlife peak in educational homophily, independent of brokerage or foci, suggests either that social-class segregation in midlife extends beyond the workplace to other midlife social settings and networks, or that there is a midlife peak in the preference for educationally similar friends. The racial homophily effects of some friendship sources change across the life course, but differently by gender. Women’s family/friend and work introductions lead to less racially diverse new friends as they age, while both become more diverse later in life for men. On the other hand, women’s voluntary and religious group friendships become more racially diverse later in life, but men’s become less diverse. Correspondingly, there is a midlife, pre-retirement peak in racially diverse new friends for men but a late-life increase in the potential for interracial friendships for women.

These findings do come with some caveats. The friendships sampled here are the two nonkin, nonsignificant-others the respondents’ spend the most time with socially, in person. Friendships sampled by other definitions (e.g., trust, help, discussion of important matters), or in ways that include long-distant ties, may exhibit different patterns of sources and homophily across the life course. These findings are also about relatively strong ties, and may not be generalizable to weaker relationships. Another important caveat is that the respondents’ age when they formed their friendships is used as a proxy for their life stage: This may be appropriate for the broadly defined life stages tested in the hypotheses here, but a more detailed and precise understanding of the effects of life stages on homophily will require detailed data on life event timing. Such information could partly explain the observed gender differences: For instance, the difference between men and women in which time of life all friendship sources converge or diverge in racial homophily may in part be due to gender differences in the timing of child-rearing. It is also important to remember that this is a sample of English speakers within the 50 U.S. states, and Washington D.C. Ethnic and racial minorities were not oversampled, and so their subsamples may have been insufficient to detect racial/ethnic differences in the effects of social foci and life stages on homophily, particularly gendered racial differences. As the findings here are most generalizable to white non-Hispanic English-speaking Americans, one should not assume that these dynamics operate the same for other groups, in the United States or elsewhere.

A surprising finding of this study is that meeting friends through one’s children predicts about as much likelihood of interracial contact as do workplaces, other things being equal. As discussed earlier, this may be due to the greater racial diversity of the networks and settings that younger generations live within (Frey 2014), compared with their parents, which allows children to act as bridges across the stronger racial boundaries of their parents’ generations. Children-brokered ties are also similar to workplace-formed friendships in their educational homophily, suggesting a general alignment of parenting foci with occupational foci. Another counter-intuitive result here is that neighbor-formed friendships are among the most likely to cross racial and ethnic boundaries, other things being equal. This is not only seemingly in contradiction to what we know about residential segregation in the United States but also differs from previous findings about neighbor alters (Marsden 1990; but see Mollenhorst et al. 2008 on the interreligiousness of neighbor alters in the Netherlands). However, neighbor and neighbor-brokered friendships in these data have a nearly identical rate of being interracial as non-neighbor ties, bivariately speaking. For a large subset of respondents who are otherwise particularly unlikely to have interracial friendships, mainly older respondents and white non-Hispanics, neighbors are where those boundary-crossing friendships are especially likely to occur. This does not mean that neighborhoods themselves are actually racially diverse, just that friendships formed through neighboring are less homogeneous than other common sources of friendship.

Though only homophily with respect to race/ethnicity and education are examined here, extensions to other homophilies may be straightforward, depending on how exposure to diversity changes across the life course. For instance, age segregation is strongest in early life, so friendships formed later should be more age diverse, perhaps with a modest return to age segregated new friendships late in life. Gender homophily may be less straightforward, as youthful settings are gender diverse and yet tend to produce gender homophilous friendships by other mechanisms (Smith-Lovin and Mcpherson 1993). Adult settings may conversely be more gender segregated and yet produce more cross-gender ties, as well as friendships made through cross-gender romantic partners for many adults. Extending the aforementioned framework to homophilies by tastes, interests, and values is complicated by the transmission of culture between friends, particularly during the more impressionable early life stages. Cultural similarity among children may be due more to peer influence, while adult cultural homophily may result more from choosing friends and groups based on shared interests and attitudes.

Other types of relationships may have different patterns of life course foci segregation. Romantic couples, for instance, are much more likely than friends to meet without interpersonal or organizational brokerage (Mollenhorst et al. 2008), particularly since the advent of online dating (Rosenfeld and Thomas 2012). This is life-stage dependent, with educational settings more common for early life romantic partners, a time of maximal educational endogamy (Mare 1991), and online romantic sources peaking in early midlife (Rosenfeld and Thomas 2012), when potential partners become scarcer. Weaker ties, such as acquaintances, are not necessarily formed in the same ways as stronger friendships (Mollenhorst et al. 2008). Even when the sources are the same, the effects of foci composition on weak tie networks might be more lasting. Unlike the strong friendships sampled here, acquaintances do not require much time or effort to maintain, and can last a lifetime, reflecting more fully the whole array of settings and networks a person has participated in since childhood, and the cumulative diversity those entailed.

Also unexamined here is how different types of ego networks can shape the effects of foci and brokerage on homophily across the life course. Some people lead lives embedded in dense networks of overlapping relationships, while others maintain more separate ego network components (Bott 1957). Personal networks can also vary greatly by their size, intensity of social activity, and embeddedness in familial and organizational contexts (Burt 1990; Giannella and Fischer 2016). Some people’s rosters of strong friendships remain very stable across their lives, while others change dramatically between life stages and major life transitions (Wrzus et al. 2013). These may all affect the extent to which friendships endure beyond particular life stages, as well as which social foci within stages are more likely to foster new relationships.

Conclusion

I have shown above that both how and when in life social ties are formed affects their racial and educational composition. This study is the first to provide population-generalizable evidence that the sources of friendships affect their racial diversity, and also shows that this largely accounts for changes in racial friendship diversity across the life course. This paper also unearths previously unknown gender differences in personal networks, showing that the life course trend in racial homophily differs for women and men, and also that sources of friendship change in their interracial potential in different and important ways for women and men across their lives. These findings suggest that attempts to understand homophily should consider not just the opportunity constraints for diverse relationships but also their gendered change over life courses.

Where and how friendships and similar relationships are formed has been an unfortunately neglected topic, but need not be. These nonkin relationships are core to the informal social structure of societies, with implications for most social phenomena. Too often, their origins are taken for granted, or treated as a matter of personal choice in a social landscape of minimal structural constraints, or simply considered too difficult to measure. A social demography of interpersonal relationships is incomplete without an account of the origins of those relationships, as these origins shape the composition of the ties that result, as well as reflect the cultural and structural pathways into forming bonds within a society. These are the reasons we care about the origins of romantic partnerships, and the nonromantic and nonfamilial bonds of social life deserve similar attention. The data presented here are only about the two friendships in respondents’ lives that occupy most of their social time, but there is much more to the social networks surrounding people than these ties. There is still very little population-generalizable information about the sources of other kinds of friendships, confidants, and acquaintances. The opportunities are there to greatly expand our knowledge of the recurring origins of informal social structure.

Supplemental Material

LifeCourseFriendshipHomophily-Appendix – Supplemental material for Sources of Friendship and Structurally Induced Homophily across the Life Course

Supplemental material, LifeCourseFriendshipHomophily-Appendix for Sources of Friendship and Structurally Induced Homophily across the Life Course by Reuben J. Thomas in Sociological Perspectives

Footnotes

Acknowledgements

I thank Noah Mark, Michael Rosenfeld, Paula England, Dan McFarland, Mark Granovetter, and panel and audience members at ASA and Sunbelt for their helpful comments and discussion.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for research, authorship, and/or publication of this article: This work was supported by the School of Humanities and Sciences and the Department of Sociology at Stanford University.

Supplemental Material

Supplemental material [the appendix] for this article is available online.

Notes

Author Biography

Reuben J. Thomas is an Assistant Professor of Sociology at the University of New Mexico. He studies social structure at the level of interpersonal relationships, including how people meet their friends and romantic partners, and implications for group boundaries and inequality. He also studies how youth networks and affiliations affect transitions into adulthood, and how social network processes may have contributed to the prehistoric rise of complex social structure.

References

Allison

Paul D.

2009. “Missing Data.” Pp. 72–89 in The SAGE Handbook of Quantitative Methods in Psychology, edited by Millsap

Roger E.

Maydeu-Olivares

Alberto

. Thousand Thousand Oaks, CA: Sage.

Blau

Peter M.

1977. Inequality and Heterogeneity. New York: The Free Press.

Bost

Kelly K.

Cox

Martha J.

Burchinal

Margaret R.

Payne

Chris

. 2002. “Structural and Supportive Changes in Couples’ Family and Friendship Networks across the Transition to Parenthood.” Journal of Marriage and Family 64:517–31.

Bott

Elizabeth

. 1957. Family and Social Network: Roles, Norms and External Relationships in Ordinary Urban Families. London, England: Tavistock.

Brashears

Matthew E.

2008. “Gender and Homophily: Differences in Male and Female Association in Blau Space.” Social Science Research 37:400–15.

Brass

Daniel J.

1985. “Men’s and Women’s Networks: A Study of Interaction Patterns and Influence in an Organization.” Academy of Management Journal 28(2):327–43.

Breiger

Ronald L.

1974. “The Duality of Persons and Groups.” Social Forces 53(2):181–90.

Burt

Ronald S.

1990. “Kinds of Relations in American Discussion Networks.” Pp. 397–410 in Structures of Power and Constraint: Papers in Honor of Peter M. Blau, edited by Calhoun

Meyer

Scott

R. S.

New York: Cambridge University Press.

Callegaro

Mario

DiSogra

Charles

. 2008. “Computing Response Metrics for Online Panels.” Public Opinion Quarterly 72:1008–32.

10.

Campbell

Karen E.

1988. “Gender Differences in Job-related Networks.” Work and Occupations 15:179–200.

11.

Carley

Kathleen M.

1991. “A Theory of Group Stability.” American Sociological Review 56:331–54.

12.

Chang

Linchiat

Krosnick

Jon A.

2009. “National Surveys via RDD Telephone Interviewing versus the Internet: Comparing Sample Representativeness and Response Quality.” Public Opinion Quarterly 73:661–74.

13.

Charles

Camille Zubrinsky

. 2003. “The Dynamics of Racial Residential Segregation.” Annual Review of Sociology 29:167–207.

14.

Clotfelter

Charles T.

2002. “Interracial Contact in High School Extracurricular Activities.” The Urban Review 34:25–46.

15.

Cornwell

Benjamin

Laumann

Edward O.

Schumm

L. Philip

. 2008. “The Social Connectedness of Older Adults: A National Profile.” American Sociological Review 73(2):185–203.

16.

Crosnoe

Robert

Elder

Glenn H.

Jr.

2002. “Successful Adaptation in the Later Years: A Life Course Approach to Aging.” Social Psychology Quarterly 65(4):309–28.

17.

Eder

Donna

Hallinan

Maureen T.

1978. “Sex Differences in Children’s Friendships.” American Sociological Review 43:237–50.

18.

Edwards

Korie L.

Christerson

Brad

Emerson

Michael O.

2013. “Race, Religious Organizations, and Integration.” Annual Review of Sociology 39:211–28.

19.

Elder

Glenn H.

Jr.

1975. “Age Differentiation and the Life Course.” Annual Review of Sociology 1:165–90.

20.

Feld

Scott L.

1981. “The Focused Organization of Social Ties.” American Journal of Sociology 86:1015–35.

21.

Feld

Scott L.

1982. “Social Structural Determinants of Similarity Among Associates.” American Sociological Review 47(6):797–801.

22.

Fischer

Claude S.

1982. “What Do We Mean by ‘Friend’? An Inductive Study.” Social Networks 3:287–306.

23.

Fischer

Claude S.

Oliker

Stacey J.

1983. “A Research Note on Friendship, Gender, and the Life Cycle.” Social Forces 62(1):124–33.

24.

Frey

William H.

2014. Diversity Explosion: How New Racial Demographics Are Remaking America. Washington, DC: Brookings Institution Press.

25.

Fricker

Scott

Galesic

Mirta

Tourangeau

Roger

Yan

Ting

. 2005. “An Experimental Comparison of Web and Telephone Surveys.” Public Opinion Quarterly 69:370–92.

26.

Giannella

Eric

Fischer

Claude S.

2016. “An Inductive Typology of Egocentric Networks.” Social Networks 47:15–23.

27.

Goodreau

Steven M.

Kitts

James A.

Morris

Martina

. 2009. “Birds of a Feather, or Friend of a Friend? Using Exponential Random Graph Models to Investigate Adolescent Social Networks.” Demography 46(1):103–25.

28.

Granovetter

Mark S.

1973. “The Strength of Weak Ties.” American Journal of Sociology 78(6):1360–80.

29.

Hinrichs

Peter

. 2015. “An Empirical Analysis of Racial Segregation in Higher Education.” NBER Working Paper No. 21831. National Bureau of Economic Research, December. Retrieved January 21, 2019 (http://www.nber.org/papers/w21831).

30.

Hout

Michael

. 2012. “Social and Economic Returns to College Education in the United States.” Annual Review of Sociology 38:379–400.

31.

Kalmijn

Matthijs

. 2007. “Gender Differences in the Effects of Divorce, Widowhood and Remarriage on Intergenerational Support: Does Marriage Protect Fathers?” Social Forces 85(3):1079–104.

32.

Kalmijn

Matthijs

. 2012. “Longitudinal Analyses of the Effects of Age, Marriage, and Parenthood on Social Contacts and Support.” Advances in Life Course Research 17:177–90.

33.

Kalmijn

Matthijs

Vermunt

Jeroen K.

2007. “Homogeneity of Social Networks by Age and Marital Status: A Multilevel Analysis of Ego-centered Networks.” Social Networks 29:25–43.

34.

King

Martin Luther

Jr.

1963. “Speech at Western Michigan University.” Q&A. Retrieved December 18, 2018 (http://wmich.edu/sites/default/files/attachments/MLK.pdf).

35.

Klärner

Andreas

Keim

Sylvia

Lippe

Holger von der

. 2016. “Social Network Dynamics in the Course of Family Formation: Results from a Mixed-methods Longitudinal Study.” International Review of Social Research 6(4):245–55.

36.

Lazega

Emmanuel

van Duijn

Marijtje

. 1997. “Position in Formal Structure, Personal Characteristics and Choices of Advisors in a Law Firm: A Logistic Regression Model for Dyadic Network Data.” Social Networks 19:375–97.

37.

Lewis

Kevin

. 2013. “The Limits of Racial Prejudice.” Proceedings of the National Academy of Sciences 110(47):18814–19.

38.

Lichter

Daniel T.

Parisi

Domenico

Grice

Steven Michael

Taquin

Michael C.

2007. “National Estimates of Racial Segregation in Rural and Small-town America.” Demography 44(3):563–81.

39.

Mare

Robert D.

1991. “Five Decades of Educational Assortative Mating.” American Sociological Review 56:15–32.

40.

Mark

Noah P.

Harris

Daniel R.

2012. “Roommate’s Race and the Racial Composition of White College Students’ Ego Networks.” Social Science Research 41:331–42.

41.

Marsden

Peter V.

1990. “Network Diversity, Substructures and Opportunities for Contact.” Pp. 397–410 in Structures of Power and Constraint: Papers in Honor of Peter M. Blau, edited by Calhoun

Meyer

Scott

R. S.

New York: Cambridge University Press.

42.

Martin

John Levi

Yeung

King-To

. 2006. “Persistence of Close Personal Ties Over a 12-Year Period.” Social Networks 28:331–62.

43.

Massey

Douglas

Denton

Nancy

. 1993. American Apartheid: Segregation and the Making of the Underclass. Cambridge, MA: Harvard University Press.

44.

McFarland

Daniel A.

Moody

James

Diehl

David

Smith

Jeffery A.

Thomas

Reuben J.

2014. “Network Ecology and Adolescent Social Structure.” American Sociological Review 79(6):1088–1121.

45.

McLaughlin

Deirdre

Vagenas

Dimitrios

Pachana

Nancy A.

Begum

Nelufa

Dobson

Annette

. 2010. “Gender Differences in Social Network Size and Satisfaction in Adults in Their 70s.” Journal of Health Psychology 15(5):671–79.

46.

McPherson

Miller

Smith-Lovin

Lynn

Cook

James

. 2001. “Birds of a Feather: Homophily in Social Networks.” Annual Review of Sociology 27:415–44.

47.

McPherson

Miller J.

Smith-Lovin

Lynn

. 1982. “Women and Weak Ties: Differences by Sex in the Size of Voluntary Organizations.” American Journal of Sociology 87(4):883–904.

48.

McPherson

Miller J.

Smith-Lovin

Lynn

. 1987. “Homophily in Voluntary Organizations: Status Distance and the Composition of Face-to-face Groups.” American Sociological Review 52:370–79.

49.

Meyer

John W.

1986. “The Self and the Life Course: Institutionalization and its Effects.” Pp. 199–206 in Human Development and the Life Course: Multidisciplinary Perspectives, edited by Sorenson

Weinert

Sherrod

Hillsdale, NJ: Lawrence Erlbaum.

50.

Mollenhorst

Gerald

Völker

Beate

Flap

Henk

. 2008. “Social Contexts and Personal Relationships: The Effect of Meeting Opportunities on Similarity for Relationships of Different Strength.” Social Networks 30:60–68.

51.

Moore

Gwen

. 1990. “Structural Determinants of Men’s and Women’s Personal Networks.” American Sociological Review 55(5):726–35.

52.

Obstfeld

David

. 2005. “Social Networks, the Tertius Iungens Orientation, and Involvement in Innovation.” Administrative Science Quarterly 50(1):100–30.

53.

Ravanera

Zenaida R.

Rajulton

Fernando

Burch

Thomas K.

2004. “Patterns of Age Variability in Life Course Transitions.” Canadian Journal of Sociology 29(4):527–42.

54.

Rivkin

Steven G.

1994. “Residential Segregation and School Integration.” Sociology of Education 67:279–92.

55.

Rosenfeld

Michael J.

Thomas

Reuben J.

2012. “Searching for a Mate: The Rise of the Internet as a Social Intermediary.” American Sociological Review 77(4):523–47.

56.

Ryan

Camille L.

Bauman

Kurt

2016. “Educational Attainment in the United States: 2015.” Current Population Reports, P20-578. Washington, DC: U.S. Department of Commerce, Bureau of the Census.

57.

Shanahan

Michael J.

2000. “Pathways to Adulthood in Changing Societies: Variability and Mechanisms in Life Course Perspective.” Annual Review of Sociology 26:667–92.

58.

Smith-Lovin

Lynn

Mcpherson

J. Miller

. 1993. “You Are Who You Know: A Network Perspective on Gender.” Pp. 223–51 in Theory on Gender/Feminism on Theory, edited by England

Paula

. New York: Aldine.

59.

Sobel

Michael E.

1986. “Some New Results on Indirect Effects and their Standard Errors in Covariance Structure Models.” Sociological Methodology 16:159–86.

60.

Stovel

Katherine

Shaw

Lynette

. 2012. “Brokerage.” Annual Review of Sociology 38:139–58.

61.

Tilly

Charles

. 1998. Durable Inequality. Berkeley, CA: University of California Press.

62.

Tomaskovic-Devey

Donald

Stainback

Kevin

Taylor

Tiffany

Zimmer

Catherine

Robinson

Corre

McTague

Tricia

. 2006. “Documenting Desegregation: Segregation in American Workplaces by Race, Ethnicity, and Sex, 1966-2003.” American Sociological Review 71(4):565–88.

63.

Toossi

Mitra

Morisi

Teresa L.

2017. “Women in the Workforce before, during, and after the Great Recession.” Spotlight on Statistics. Washington, DC: U.S. Department of Labor, Bureau of Labor Statistics.

64.

Umberson

Debra

Pudrovska

Tetyana

Reczek

Corinne

. 2010. “Parenthood, Childlessness, and Well-Being: A Life Course Perspective.” Journal of Marriage and Family 72(3):612–29.

65.

Wrzus

Cornelia

Hӓnel

Martha

Wagner

Jenny

Neyer

Franz J.

2013. “Social Network Changes and Life Events across the Life Span: A Meta-Analysis.” Psychological Bulletin 139:53–80.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.63 MB