Most Likely to Succeed

Abstract

Sociological explanations for economic success tend toward measures of embeddedness in long-standing social institutions, such as race and gender, or personal skills represented mainly by educational attainment. Instead, we seek a distinctively social foundation for success by investigating the long-term association between high school popularity and income. Using rich longitudinal data, we find a clear and persistent association between the number of friendship nominations received and adult income, even after accounting for the mediating influences of diverse personal, family, and work characteristics. This skill is distinct from conventional personality measures such as the Big Five and persists long into adulthood. We hypothesize that popularity encapsulates a socioemotional skill recognized by peers as the practice of being a good friend rather than an indicator of social status.

Keywords

Do you know how popular I am? . . . I am so popular. Everybody loves me so much at this school.

—Claire Standish, from The Breakfast Club

Introduction

The iconic image of one’s high school past, filtered through old yearbooks with awkwardly out-of-fashion pictures, would lead many of us to hope that the time was truly buried and forgotten, having no relevance to our older, more mature, current selves. Yet high school holds a place in the collective American psyche that seems to overshadow its brief temporal span. The sum of its formative experiences occupies a crucial point in our trajectory through adulthood. But despite its central place in our collective memories, we have remarkably few examinations of how one’s position in that distant high school microcosm relates to the “real world” we inhabit as adults.

We expect that much of the imprint of high school stems from its origin position in the arc of one’s social identity, occupying a key developmental point shaping one’s social interaction style. These four years mark a time of exploration during which individuals achieve a greater subjective sense of adulthood (Arnett 2000). High school is the first stage of the process of becoming one’s self, independent of parents (Schulenberg, Sameroff, and Cicchetti 2004; Turner 1967). Adolescents in high school begin to engage in multiple facets of identity exploration involving romantic relationships and worldviews as they wrestle with a sense of self, which continues and matures into emerging adulthood (Arnett 2000; Erikson 1968), shaped in no small measure by the school peer environment (Cairns and Cairns 1995; Gest, Sesma, Masten, and Tellegen 2006; Giordano 2003; Hansell 1981). One’s position in the high school social network and interactions with peers inform different dimensions of identity, ranging from attitudes toward intimacy to misbehavior (Brown 2004). Despite occupying only an early point in the developmental process of becoming a fully adult self, one’s relationship to peers during high school can have lasting associations with adult outcomes via skill development, ways of acting, or preference formation. Unfortunately, empirical investigation of such effects—particularly how position within a peer context shapes later life outcomes—has been severely hampered due to lack of long-term data with sociometric components.

Here, we explore one dimension of this potential long reach of adolescent social embeddedness by asking whether popularity in school predicts economic success as an adult. We take advantage of the longitudinal nature of the National Longitudinal Study of Adolescent to Adult Health (Add Health), which has the added benefit of sociometric data. Our analyses document a clear and consistent signal linking popular status in youth to economic success as an adult: each additionally received friendship nomination yields an earnings premium, 15 years later, of 1.4 percent after controlling for a host of plausible mediating factors. This net effect is equivalent to a 5.3 percent earnings premium when comparing students who differ in popularity by 1 standard deviation.

We know that high school popularity is correlated with a number of other factors, including some with genetic foundations (Christakis and Fowler 2014) that can independently drive success in the adult workplace (Case and Paxson 2008; Hamermesh and Biddle 1994; Persico, Postlewaite, and Silverman 2004). Yet we identify a long-lasting imprint of school popularity that carries over into adult income even after accounting for the types of personal characteristics that drive school popularity. The findings hold independent of the mediating influences of educational success, family background, and physical appearance. Although the same skills and traits that manifest in popularity may lead to more schooling and higher grades, educational variables explain only one quarter of the popularity effect. Similarly, family characteristics, measures of personality traits, working hours/habits, and beauty only account for a small share of aggregate returns to popularity.

Discovering this association opens the door to thinking about the social mechanisms linking popularity and success. We imagine three interwoven social processes. To begin with, there may be a “collective intelligence” in the aggregate friendship votes that defines popularity, picking up an unknown and heretofore unmeasured trait recognized by adolescents that ends up being a key indicator of success. Any such trait would need to signal sociality, openness, or some similarly attractive feature that operates independently of beauty, socioeconomic status (SES), or broad personality characteristics already measured. Such a trait would imply that the observed popularity effect is spurious on the unidentified selection feature. If the effect is causal, we expect it operates distally. That is, we highly doubt that this effect reflects traditional “social capital” mechanisms, such as having contacts pass along employment opportunities (Lin 2002). Rather, we suspect two mechanisms that can connect teenage popularity to adult success. First, the experience of being popular—of navigating the multiple constituencies that confer status—might hone a still-unmeasured skill that translates into later success. This might include the ability to ease tensions between conflicted peers, switch language or behavior quickly between microsocial crowds, or easily pick up on social cues that allow one to navigate later employment settings with greater ease. Second, the experience of popularity might help define a self-identity, carrying over into a type of confidence that translates into success, a popularity variant of the general social comparison hypothesis (Coleman and Fults 1982; Davis 1966; Festinger 1954; Marsh and Parker 1984). Unfortunately, we cannot test these mechanisms directly here; rather, we are able to identify a remarkable social signature, where the collective friendship activity of teens translates into a powerful predictor of future success.

Background and Prior Work

The simplest explanation for a long-term association between popularity and adult income is that popular adolescents have a socioemotional skill advantage recognized by their peers. When explaining income differentials, economists have traditionally focused on returns to cognitive skills measured through total education. Recently, scholars have moved away from a simple conception of human capital to consider how both cognitive and socioemotional skills work together to explain economic, educational, and health behaviors (Bowles and Gintis 1976; Bowles, Gintis, and Osborne 2001; Cunha et al. 2006; Heckman and Rubinstein 2001; Heckman, Stixrud, and Urzua 2006). This reflects the simple fact that employers value socioemotional traits like collaboration and creativity beyond standard measures of cognitive ability. For example, a 2013 employer survey by the staffing firm Adecco found that 44 percent of senior executives cite a lack of social skills as the most pressing gap in the U.S. workforce. Technical and computer skills trailed behind at 22 and 12 percent, respectively (Adecco 2013).

Socioemotional skills are difficult to measure, however, and researchers usually resort to one of two strategies. They either use self-reports on attitudinal and personality scales (Fortin 2008; Lundberg 2014; Mueller and Plug 2006) or track behaviors related to workplace discipline, such as school attendance and grade point average (GPA; Aucejo 2015; Glaeser, Laibson, and Sacerdote 2002; Kuhn and Weinberger 2005). Popularity represents an alternative approach relying on the idea that kids who have good social skills are well-liked in school and continue to carry those social skills into adulthood. This line of argument builds on a nascent literature that relies on network measures to explain long-term economic success (Conti et al. 2013; Fletcher 2014).

Here, we engage a growing literature that crosses economics, sociology, and psychology, underscoring the importance of socioemotional skills for future success. Although advocates of the intelligence-as-ability approach emphasize the role of cognitive skills for explaining variation in outcomes (Herrnstein and Murray 1996; Jensen 1998), empirical evidence consistently shows that the vast majority of earnings variation remains unexplained after controlling for conventional measures of cognitive ability (Bowles et al. 2001). Individuals with observationally similar cognitive attributes often diverge markedly in educational and employment outcomes. As a result, researchers are increasingly relying on personality and behavioral attributes such as persistence, organization, and dependability to explain the residual variance (Almlund et al. 2011; Duckworth and Seligman 2005; Jencks 1979; Wolfe and Johnson 1995).

Popularity may also have a direct influence on individuals’ social skills above and beyond that captured by some existing skill signal. That is, the developmental significance of friendships seems to promote one’s sense of well-being and provide a range of competencies that facilitate future success (Gest et al. 2006; Hartup 1996; Sullivan 1953). There are two subtly different mechanisms proposed here. On one hand, there is an internalization of popularity that leads to a stronger sense of self, thus acting through an identity formation mechanism. For example, longitudinal empirical evidence shows that early friendship predicts adult adjustment, with socially accepted individuals exhibiting higher levels of self-worth and perceived competence, whereas peer rejection undermines social competencies (Bagwell, Newcomb, and Bukowski 1998; Ladd 2006). On the other hand, popular youth likely hone a practical skill by engaging with a more diverse group of peers that allows them to more easily traverse diverse workplaces.

Differences in cognitive and socioemotional skills, whether reflected by popularity or shaped directly from peer relationships, can manifest via several labor market mechanisms that might reasonably explain the relationship between popularity and earnings. For example, we consider the mediating roles of hours worked, employment disruptions, and sorting by job content. Popular students work more hours and are less likely to be laid off or fired from a job. Although occupational sorting plays a trivial role in explaining the popularity premium, evidence is nevertheless consistent with individuals sorting by job attributes. Individuals with more social ties tend to work in professions that place greater value on analytic ability, attention to detail, and leadership. Together, these labor market characteristics reduce the popularity premium to 1.2 percent for each friendship nomination, but do not erase it.

Data

Our data come from the National Longitudinal Study of Adolescent Health (Add Health). The first wave of data collection began in 1994–1995, when students were in Grades 7 to 12. Initial sampling selected 80 high schools from a total of 26,666 schools sorted on size, type, grade span, racial composition, urbanicity, region of origin, and sector. Participating high schools then identified feeder institutions. This produced a nationally representative sample of 144 middle, junior high, and high schools comprising 90,118 students. Respondents in eligible schools selected up to five male and five female peers from a roster of students in their own school and the corresponding sister school.¹ Thus, each individual could nominate up to 10 friends. Information on friendship networks allows researchers to construct the structure of respondents’ school environments and measure their relative placement in these social contexts.

In addition to school-level and network information, the longitudinal component of Add Health collected data from more than 20,000 respondents in the first survey wave. The latest round, Wave 4, occurred between April 2007 and February 2009 when respondents were between 24 and 34 years old. All together, 15,701 individuals remain in Wave 4 from the original longitudinal sample. Among these, 10,926 have comprehensive network data. The longitudinal nature of Add Health enables us to measure individuals’ network status almost 15 years prior to their realized economic outcomes. In this way, we are able to avoid endogeneity concerns stemming from contemporaneous observations of socio-emotional skills and earnings.

Network and Non-network Measures

Network Measures

We use standard network measures to capture a person’s position in the social network. A basic and intuitive measure of a person’s popularity is in-degree centrality, defined as the sum of friendship nominations received (Wasserman and Faust 1994). We differentiate between in-degree and out-degree in a directed network. Out-degree is the number of individuals that the respondent lists as a friend and serves as an indicator of gregariousness or sociality.

Table 1 shows the summary statistics on in- and out-degrees. Respondents have a mean of 4.62 nominations and a standard deviation of 3.8. The distribution skews to the left with a mode of 2 nominations and more than 8 percent of the sample receiving zero nominations (Figure 1). Some respondents are strikingly popular with their peers, receiving nominations well into the double digits. We top code the sample at 15 or more friends to minimize the role of outliers. Approximately 80 percent of the sample falls within 1 standard deviation of the mean.

Table 1.

Summary Statistics: Network Measures.

Measure	Description	Mean	Standard deviation	Maximum	Count
In-degree	Number of alters that nominate ego as friend	4.62	3.8	32	9,519
Out-degree	Number of alters nominated by ego	4.66	3.0	10	9,519
Same-sex in-degree	Number of same-sex alters that nominate ego	2.74	2.1	14	9,519
Male	Number of same-sex alters that nominate male ego	2.51	2.1	13	4,557
Female	Number of same-sex alters that nominate female ego	2.98	2.1	14	4,962
Cross-sex in-degree	Number of cross-sex alters that nominate ego	1.87	2.4	25	9,519
Male	Number of cross-sex alters that nominate male ego	1.87	2.5	19	4,557
Female	Number of cross-sex alters that nominate female ego	1.86	2.4	25	4,962
Bonacich centrality on in-degree	Ego’s centrality weighted by centrality of nominating alters	0.67	0.8	10.1	9,519
Same-sex Bonacich centrality	Ego’s centrality weighted by centrality of same-sex alters	0.59	0.7	6.0	9,468
In-degree transitive triples	Number of nominating alters belonging to transitive triples	2.64	3.4	30	9,519
In-degree nontransitive triples	Number of nominating alters that do not belong to transitive triples	1.99	1.6	12	9,519
Reciprocity	Number of alters’ nominations that ego returns with nomination	1.87	1.8	10	9,519
In-degree—not reciprocated	Number of alters’ nominations that ego does not return	2.75	2.8	26	9,519
Out-degree—not reciprocated	Number of ego’s nominations that alter does not return	2.79	2.3	10	9,519

Note. Ego is the respondent, whereas alter refers to students in the same school eligible to be nominated as friends.

Figure 1.

Popularity density and cumulative distribution function (CDF).

Next, we distinguish between nominations by friends’ sex (see McPherson, Smith-Lovin, and Cook 2001; Tuma and Hallinan 1979). To compare homophily in male and female friendship networks, we separate in-degree by same-sex versus cross-sex nominations. On average, respondents have 2.7 friends of the same sex, compared with 1.9 of the opposite sex. The distribution of same-sex in-degree is similar to that of the overall in-degree measure (Figure 2). Nearly 40 percent of students receive one or two nominations. The number of respondents decreases quickly in same-sex in-degree after the first three nominations, with less than 1 percent having 10 or more same-sex friends. The distribution of cross-sex nominations peaks at zero nominations with a comparatively lower mean than same-sex nominations. Nearly 40 percent of all respondents have no cross-sex friends (Figure 3). Both men and women are more likely to nominate members of the same sex, reflecting well-known gender homophily in adolescent networks (Table 1).

Figure 2.

Same-sex in-degree distribution.

Figure 3.

Cross-sex in-degree distribution.

Although in-degree centrality captures local popularity, it is agnostic on the person’s status relative to the position in the larger social hierarchy (McFarland et al. 2014). Bonacich centrality builds the structure of the wider school network into the popularity measure, as an individual’s centrality is a positive function of the centrality of their friends. Holding the number of connections equal, those connected to more influential individuals have higher Bonacich centrality scores.² The mean Bonacich centrality score is 0.7, with a standard deviation of 0.8. Although correlated with degree centrality (.83), the two are not identical (Figure 4). As with degree, we separately examine same-sex and cross-sex Bonacich centrality.³

Figure 4.

Bonacich centrality and in-degree.

Non-network Measures

Table 2 presents summary statistics for non-network variables in the final sample. The dependent variable comes from a survey question in Wave 4 that asks, “Now think about your personal earnings. How much income did you receive from personal earnings before taxes—that is, wages or salaries, including tips, bonuses, and overtime pay, and income from self-employment?” We log transform all positive earnings.⁴ The final sample, excluding individuals from schools without an assigned weight, yields 9,519 respondents across 113 schools.

Table 2.

Summary Statistics: Non-network Measures.

Measure	Description	Mean	Standard deviation	Minimum	Maximum
Log earnings	W4: log of income from personal earnings before taxes	10.18	1.0	0.0	13.8
Age	W4: age at time of interview	28.71	1.8	24.5	34.1
Male	W4: sex	0.52	0.5	0.0	1.0
White	W1: race	0.68	0.5	0.0	1.0
Black	W1: race	0.16	0.4	0.0	1.0
Hispanic	W1: race	0.09	0.3	0.0	1.0
Maternal education	W1: response to “how far in school did mom go?”	13.29	2.3	8.0	18.0
Parents married	W1: parental response to question on marital status	0.66	0.5	0.0	1.0
PVT score	W1: PVT	102.91	13.3	14.0	137.0
Education	W4: highest level of education achieved to date	14.31	2.2	8.0	21.0
GPA	W1: average of math and English GPA	2.81	0.8	1.0	4.0
W1 grooming	W1: interviewer remarks on respondent’s grooming	3.58	0.8	1.0	5.0
W2 grooming	W2: interviewer remarks on respondent’s grooming	3.51	0.7	1.0	5.0
W3 grooming	W3: interviewer remarks on respondent’s grooming	3.52	0.7	1.0	5.0
W1 personality attractiveness	W1: interviewer remarks on respondent’s personality attractiveness	3.62	0.8	1.0	5.0
W2 personality attractiveness	W2: interviewer remarks on respondent’s personality attractiveness	3.61	0.7	1.0	5.0
W3 personality attractiveness	W3: interviewer remarks on respondent’s personality attractiveness	3.68	0.8	1.0	5.0
W1 physical attractiveness	W1: interviewer remarks on respondent’s physical attractiveness	3.58	0.9	1.0	5.0
W2 physical attractiveness	W2: interviewer remarks on respondent’s physical attractiveness	3.57	0.7	1.0	5.0
W3 physical attractiveness	W3: interviewer remarks on respondent’s physical attractiveness	3.51	0.7	1.0	5.0
No active sport past week	W1: number of times respondent played active sport	0.25	0.4	0.0	1.0
Active sport one to two times past week	W1: number of times respondent played active sport	0.28	0.4	0.0	1.0
Active sport three to four times past week	W1: number of times respondent played active sport	0.21	0.4	0.0	1.0
Active sport five or more times past week	W1: number of times respondent played active sport	0.26	0.4	0.0	1.0
Friends zero times past week	W1: number of times respondent hung out with friends	0.08	0.3	0.0	1.0
Friends one to two times past week	W1: number of times respondent hung out with friends	0.23	0.4	0.0	1.0
Friends three to four times past week	W1: number of times respondent hung out with friends	0.28	0.4	0.0	1.0
Friends five or more times past week	W1: number of times respondent hung out with friends	0.40	0.5	0.0	1.0
Sport clubs (W1)	W1: current/planned participation in sports this year	0.57	0.5	0.0	1.0
Self-esteem	W1: standardized index of the sum of six personality questions	0.00	1.0	−5.4	1.5
Lot of energy—strongly agree	W1: response to statement “I have a lot of energy”	0.29	0.5	0.0	1.0
Lot of energy—agree	W1: response to statement “I have a lot of energy”	0.41	0.5	0.0	1.0
I try very hard to do my best	W1: respondent tries very hard to his or her best	0.37	0.5	0.0	1.0
I try hard enough	W1: respondent tries hard enough but not as hard as he or she can	0.49	0.5	0.0	1.0
Never skipped school	W1: zero unexcused full-day absences from school	0.75	0.4	0.0	1.0
Skipped school without excuse once	W1: one unexcused full-day absence from school	0.07	0.3	0.0	1.0
Ever suspended	W1: ever received an out-of-school suspension	0.24	0.4	0.0	1.0
Extraversion	W4: standardized score of preconstructed variable	−0.00	1.0	−4.3	2.2
Neuroticism	W4: standardized score of preconstructed variable	0.00	1.0	−3.8	3.6
Agreeableness	W4: standardized score of preconstructed variable	0.00	1.0	−6.3	2.0
Conscientiousness	W4: standardized score of preconstructed variable	−0.00	1.0	−5.5	2.0
Openness	W4: standardized score of preconstructed variable	−0.00	1.0	−5.4	2.1
Anxious personality scale	W4: standardized score of preconstructed variable	−0.00	1.0	−2.8	2.6
Optimistic personality scale	W4: standardized score of preconstructed variable	−0.00	1.0	−4.5	2.1
Anger/hostility personality scale	W4: standardized score of preconstructed variable	0.00	1.0	−2.1	3.4
Internal locus of control	W4: standardized score of preconstructed variable	−0.00	1.0	−5.3	2.0
Weekly hours	W4: average hours/week at most recent or current job	41.65	11.7	0.0	120.0
Fired or laid off since 2001	W4: ever fired, let go, or laid off from job since 2001	0.30	0.5	0.0	1.0
Observations = 9,519

Note. Self-esteem variable includes respondent answers to (1) you have a lot of good qualities, (2) you have a lot to be proud of, (3) you like yourself just the way you are, (4) you feel like you are doing everything just about right, (5) you feel socially accepted, and (6) you feel loved and wanted. PVT = Picture Vocabulary Test; W1 = Wave 1; W2 = Wave 2; W3 = Wave 3; W4 = Wave 4; GPA = grade point average.

The nationally representative sample is 68 percent white, 16 percent black, and 9 percent Hispanic. Respondents are between 24 and 34 years of age at the time earnings data are provided. We group regressors by potential mechanisms. First, we include controls for maternal education and marital status to account for differential family background that can influence the habits and skills of students in the sample. These differences can inform the individuals’ position in the school network and the associated number of friendship ties. Mothers attain 13.3 years of education on average, and more than 66 percent of the sample comes from households with married parents.

Second, previous research shows that wage premiums from social skills operate via educational attainment and better academic performance (Cawley, Heckman, and Vytlacil 2001). We are interested in parsing out popularity’s effects, net of its influence via educational achievement and cognitive performance. Variables representing these mediating influences come from the highest level of education attained by Wave 4, average math and English grades in Wave 1, and the Picture Vocabulary Test (PVT). There exists substantial variation in educational attainment among a sample with more than 95 percent of its members at or exceeding 25 years at the time of surveying. Average schooling level is 14.3 years, and the standard deviation is 2.2 years. Some remain in school during Wave 4 (though overall results are not sensitive to their exclusion). Regarding the grade variables, it is well documented that teachers reward personality traits such as diligence and lack of disruptiveness (Bowles and Gintis 1976; Farkas 2003). Controlling for the average of math and English grades as well as effort invested in school enables us to focus on the components of popularity that operate independently of work ethic to influence future earnings.

Our measure of cognitive skills is an abridged version of the PVT administered for Add Health. The PVT assesses verbal comprehension and general cognitive aptitude, and is shown to maintain high construct validity with a number of intelligence tests (Baker et al. 1993). We expect substantial outcome heterogeneity to remain after controlling for educational attainment and cognitive ability, as achievement reflects only a subset of the abilities valued by employers.

Another channel that increases one’s propensity to win friends and become gainfully employed is beauty (Hamermesh and Biddle 1994). As beauty is positively correlated with wages, omitting physical appearance from the model will upwardly bias the popularity coefficient. We therefore include measures of physical attractiveness along with indicators for grooming and personality attractiveness. All data are taken from interviewers’ assessments at the end of a home visit during Waves 1 to 3, in which they rate the respondent’s appearance or personality on a scale of 1 to 5, ranging from very unattractive to very attractive.⁵

We also include daily activities measures on sports participation and hanging out with friends. A willingness to participate frequently may signal existing preferences for competition and ambition. In addition to signaling, the experience can foster the development of certain attributes that are advantageous in the labor market, including teamwork, discipline, and leadership. The cultivation of friendships through frequent interactions suggests that the respondent possesses certain interpersonal skills that translate into higher future earnings.

A lingering question that remains after the consideration of these covariates is the content of popularity. If popularity is a broad proxy for socioemotional skills, then we need to account for potential overlap with common measures of personality traits. To address this shortcoming, we use a number of survey questions from Waves 1 and 4 to construct relevant personality attributes. We use measures for vigor, school effort, and self-esteem from Wave 1 in-school and in-home questionnaires.⁶ Finally, we take advantage of information on unexcused absences and suspensions during Wave 1 to measure student delinquency. We add dummy variables for never having an unexcused absence during the school year, having one unexcused absence, and ever being suspended from school.

Personality questions from Wave 4 are inputs for the remaining five scaled measures: (1) internal locus of control/mastery, (2) Big Five, (3) anxiety, (4) optimism, and (5) anger/hostility. The locus of control dimension of personality describes the extent to which one regards life outcomes as being under one’s own control (Rotter 1966). The Big Five components of extraversion, neuroticism, agreeableness, conscientiousness, and openness account for a substantial portion of variance in personality-relevant adjectives.⁷ The remaining variables of anxiety, optimism, and anger/hostility capture other psychosocial facets that influence behavior. Each scale comprises four questions, including “I worry about things,” “I’m always optimistic about my future,” and “I get angry easily,” respectively. To ease interpretation, we normalize personality scales for self-esteem, locus of control, anxiety, optimism, anger/hostility, and Big Five by grade enrolled to a mean of 0 and a standard deviation of 1.⁸

Finally, job attributes such as occupation, task content, hours worked, and employment disruptions can shape income trajectories, though their status as exogenous to any popularity effect is questionable. To be conservative, we use Wave 4 data on weekly hours worked during the respondent’s most recent or current job. Individuals who stay with jobs for sustained periods of time may accrue benefits from increased promotional opportunities, whereas those who switch jobs frequently lose income during unemployment spells. Employment disruption data characterize the number of times respondents have been fired, let go, or laid off from a job since 2001.

We construct detailed measures of job attributes using the Occupational Information Network (O*NET). O*NET classifies each occupation by required knowledge, skills, abilities, and common activities and tasks. The full taxonomy comprises almost 1,000 distinct occupations using data collected from job incumbents or occupation experts. We rely on one facet of the database called “Work Styles,” an input into worker characteristics that describe how well different personalities can perform the job. Each style of personality is ranked on an importance scale of 1 to 5, with 1 representing minimum importance and 5 representing maximum importance.⁹

Empirical Strategy

We regress log-transformed earnings on popularity, out-degree, and a vector of covariates ranging from family background to physical appearance. We include grade and school fixed effects to permit a clearer identification of popularity’s influence while minimizing the confounding effects of school tenure and institutional context. Students in higher grades have interacted with more peers by virtue of spending more time at the school. The formation of social networks furthermore depends on grade and school characteristics such as size, presence of academic tracking, extracurricular activities, and demographic composition. Grade fixed effects remove the effect of mean differences between grades, whereas school fixed effects control for the mediating effect of school-level factors such as average educational attainment, teacher quality, and other resource inputs.¹⁰

Results

The analyses begin with a relatively sparse model containing individual demographic and family background variables before folding in additional behavioral traits and personal attributes. The simplest specification in Table 3 finds that an additional nomination raises the individual’s annual earnings by 2.7 percent. Holding in-degree centrality constant, a respondent who claims an extensive network of friends garners no earnings premium relative to a respondent who records few ties. That is, popularity is a significant predictor of labor market outcomes, whereas gregariousness provides little explanatory value (in fact turning negative once other controls are added).

Table 3.

Popularity, Bonacich Centrality, and Log Earnings.

Measure	(1)	(2)	(3)	(4)	(5)	(6)
In-degree	.027***	.020***	.014***
In-degree	(.004)	(.004)	(.004)
Same-sex in-degree				.020**
Same-sex in-degree				(.008)
Cross-sex in-degree				.009*
Cross-sex in-degree				(.005)
Bonacich centrality on in-degree					.070***
Bonacich centrality on in-degree					(.020)
Same-sex Bonacich centrality						.052*
Same-sex Bonacich centrality						(.028)
Out-degree	.000	−.005	−.009*	−.009*	−.009**	−.008*
Out-degree	(.005)	(.005)	(.004)	(.004)	(.004)	(.005)
Male	.410***	.471***	.458***	.462***	.457***	.461***
Male	(.032)	(.031)	(.034)	(.034)	(.034)	(.034)
Black	−.294***	−.277***	−.267***	−.267***	−.263***	−.265***
Black	(.062)	(.065)	(.062)	(.062)	(.063)	(.063)
Hispanic	.013	.024	.031	.032	.034	.036
Hispanic	(.058)	(.059)	(.059)	(.059)	(.058)	(.059)
Asian	.088	−.014	−.020	−.021	−.019	−.021
Asian	(.093)	(.088)	(.085)	(.085)	(.084)	(.084)
Other race	−.174**	−.149*	−.136*	−.134*	−.132*	−.131*
Other race	(.081)	(.078)	(.079)	(.079)	(.079)	(.078)
Maternal education	.024***	−.001	−.004	−.004	−.004	−.004
Maternal education	(.007)	(.007)	(.006)	(.006)	(.006)	(.006)
Parents married	.038	.008	.002	.001	.000	.000
Parents married	(.037)	(.034)	(.034)	(.033)	(.033)	(.033)
PVT score		−.001	−.001	−.001	−.001	−.001
PVT score		(.001)	(.001)	(.001)	(.001)	(.001)
Education		.096***	.090***	.090***	.089***	.091***
Education		(.008)	(.008)	(.008)	(.008)	(.008)
GPA		.102***	.090***	.089***	.089***	.088***
GPA		(.020)	(.020)	(.020)	(.020)	(.020)
Observations	9,519	9,519	9,519	9,519	9,519	9,468
R ²	.148	.193	.206	.206	.206	.206

Note. Attractiveness covariates come from interviewer responses in Waves 1 to 3. Daily activities measures come from Wave 1. All weighted least squares specifications include grade and school fixed effects. Clustered standard errors are at the school level. PVT = Picture Vocabulary Test; GPA = grade point average.

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

The model in Column 2 includes measures of educational attainment, cognitive ability, and school performance. To the extent that popular individuals possess work habits and a school-oriented attitude that induces them to obtain more schooling and higher GPA and test scores, omitting attainment regressors from the model results in upward bias in the popularity coefficient. According to Model 2, an additional friendship nomination is associated with a 2.0 percent earnings premium after controlling for schooling, cognitive ability, and GPA. Three quarters of the original variation remains, even after controlling for social and cognitive skills channeled through achievement.

Next, we include beauty, personality attractiveness, and daily activities measures from Waves 1 to 3.¹¹ Their inclusion lowers the coefficient on in-degree centrality to 1.4 percent. In particular, well-groomed individuals in Wave 3 are significantly more likely to earn higher wages. The positive correlation between physical appearance and earnings is consistent with other studies on the economic returns to attractiveness (Averett and Korenman 1998; Biddle and Hamermesh 1998; Hamermesh and Biddle 1994). Individuals who actively engage in sports and hang out with their friends earn significantly more than their counterparts who engage less frequently. All together, this implies that a 1 standard deviation increase in popularity is associated with a 5.3 percent wage premium for adolescents less than 15 years later.

In Figure 5, we relax the linearity assumption on the in-degree variable and enter separate intercepts for each popularity level. Relative to socially isolated students without any nominations, those with two friends already hold a significant earnings premium of more than 10 percent. This suggests sizable returns to a minimum threshold of social integration. Isolated students, on the contrary, fare poorly compared with their peers along dimensions such as academic achievement and social cohesion.

Figure 5.

Earnings premium by popularity level.

Next, we redefine the popularity variable according to alters’ attributes. This acknowledges the possibility that friendships differ in intention and substance, such that some types of friendships are more predictive of future labor market success than others. Column 4 shows the relative effects of popularity among same- versus cross-sex peers. We observe two notable trends. First, the popularity earnings premium is concentrated among same-sex nominations. An additional same-sex friend raises the annual earnings by 2.0 percent, relative to 0.9 percent for cross-sex friendships, after controlling for sociodemographic, educational, attractiveness, and daily activities variables. This result is stable across different model specifications.

Second, changes across specifications in the magnitude and significance of the cross-sex variable suggest that students pick friends of the opposite sex for different reasons than same-sex friends. In results not shown, an additional same-sex nomination yields a 2.4 percent increase in annual earnings after controlling for education and cognitive ability, compared with a 1.6 percent increase for cross-sex nominations. The impact of cross-sex popularity decreases to 1.2 percent after accounting for interviewer-rated measures of attractiveness, whereas the influence of same-sex friends decreases only slightly to 2.3 percent. We infer that looks and grooming are key dimensions underlying cross-sex friendship nominations. Same-sex popularity, on the contrary, largely comprises characteristics not related to appearance but that are relevant for future income.¹²

In Model 5, we replace degree centrality with two alternative popularity measures, Bonacich centrality and same-sex Bonacich centrality. Doing so extends the individual’s radius of influence beyond his or her adjacent peers into the wider school-based network. We find that a one-unit increase in Bonacich score yields a 7.0 percent earnings premium. Another way of quantifying the effect is to assess changes in earnings when shifting up a standard deviation. One additional standard deviation in Bonacich centrality yields a 5.6 percent earnings premium. Further evidence of the prominence of same-sex nominations is shown in Column 6. The majority of the Bonacich centrality effect derives from nominations from same-sex peers.

Which Dimensions of Socioemotional Skills Does Popularity Cover?

Despite establishing the popularity premium as a robust finding across different models, we have yet to elaborate on its informational content. What types of personality or behavioral traits does popularity capture? In this section, we build toward an answer by investigating what popularity is not. We present supplementary analyses that characterize the relationship between friendship nominations and common proxies for socioemotional traits including the Big Five, self-esteem, optimism, student effort, and measures of delinquent behavior. The evidence suggests that popularity encompasses traits that are distinct from measures of socioemotional skills referenced in other studies.

Table 4 begins with the baseline result from the full model specification and is subsequently augmented by additional sets of personality traits. Column 2 controls for Wave 1 measures of self-esteem, vigor, school effort, unexcused absences, and suspensions. Their addition reduces the coefficient on popularity to 1.3 percent and only slightly increases the explained residual variation in the earnings function. Vigor is positively associated with earnings, whereas school suspensions lead to a reduction. Somewhat surprisingly, maximum self-reported effort correlates with lower earnings.¹³ Other dimensions of self-evaluation do not appear to explain the popularity premium. Self-esteem is not predictive of earnings, and none of the remaining personality scales demonstrate a relationship with popularity despite their apparent effect on earnings. The inclusion of all personality variables decreases the popularity premium to 1.3 percent. This relatively small change provides further evidence that friendship nominations capture some facets of personality and social skills that do not overlap with common measures of personality and behavioral traits.

Table 4.

Alternative Personality Attributes.

Measure	(1)	(2)	(3)	(4)	(5)	(6)
In-degree	.014***	.013***	.014***	.014***	.014***	.013***
In-degree	(.004)	(.004)	(.004)	(.004)	(.004)	(.004)
Out-degree	−.009*	−.009*	−.011**	−.009**	−.009**	−.011**
Out-degree	(.004)	(.005)	(.004)	(.004)	(.004)	(.005)
Self-esteem		.009				−.007
Self-esteem		(.012)				(.012)
Lot of energy—strongly agree		.065				.063
Lot of energy—strongly agree		(.043)				(.043)
Lot of energy—agree		.086***				.086***
Lot of energy—agree		(.033)				(.032)
I try very hard to do my best		−.081*				−.082*
I try very hard to do my best		(.045)				(.043)
I try hard enough, but not as hard as I could		−.050				−.056
I try hard enough, but not as hard as I could		(.034)				(.035)
Never skipped school		−.052				−.039
Never skipped school		(.043)				(.044)
Skipped school without excuse once		−.031				−.014
Skipped school without excuse once		(.057)				(.057)
Ever suspended		−.084**				−.084**
Ever suspended		(.038)				(.037)
Extraversion			.045***			.024*
Extraversion			(.012)			(.013)
Neuroticism			−.037**			−.013
Neuroticism			(.016)			(.019)
Agreeableness			−.042***			−.044***
Agreeableness			(.014)			(.014)
Conscientiousness			.032**			.010
Conscientiousness			(.013)			(.013)
Openness			−.069***			−.082***
Openness			(.016)			(.017)
Anxious personality scale				−.044***		−.012
Anxious personality scale				(.015)		(.015)
Optimistic personality scale				.073***		.040**
Optimistic personality scale				(.015)		(.015)
Anger/hostility personality scale				.048***		.045**
Anger/hostility personality scale				(.018)		(.020)
Internal locus of control					.108***	.106***
Internal locus of control					(.015)	(.018)
Observations	9,519	9,519	9,519	9,519	9,519	9,519
R ²	.206	.209	.218	.217	.218	.233

Note. All models include (1) base covariates spanning age, race, maternal education, and parental marital status; (2) education covariates, including PVT, GPA, and educational attainment; (3) grooming and physical and personality attractiveness measures; and (4) daily activities. All weighted least squares specifications include grade and school fixed effects. Clustered standard errors are at the school level. PVT = Picture Vocabulary Test; GPA = grade point average.

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

A Closer Look at Labor Market Mechanisms

Our models thus far establish that popularity comprises something distinct from existing measures of socioemotional skills. Further evidence on its precise contents comes from a closer examination of job attributes. As earnings depend on variables such as occupation, skills required, hours, and job tenure, we can evaluate how each dimension affects the popularity premium. Table 5 controls for weekly hours worked in the respondent’s most recent or current job during Wave 4. We assume a quadratic relationship between hours and earnings.¹⁴ Earnings are a concave function of hours, and increases in time worked until approximately 90 hours. Popular individuals work more, and accounting for this intensive margin reduces the popularity premium to 1.2 percent. Next, we include workplace disruptions into the model. Popular students report fewer disruptions, although the popularity premium remains unaffected.

Table 5.

Labor Market Mechanisms.

Measure	(1)	(2)	(3)	(4)	(5)	(6)
In-degree	.014***	.012***	.014***	.015***	.013***	.012***
In-degree	(.004)	(.004)	(.004)	(.004)	(.003)	(.004)
Out-degree	−.009*	−.008*	−.009**	−.010**	−.008**	−.008*
Out-degree	(.004)	(.004)	(.004)	(.004)	(.004)	(.004)
Analytical					.242***	.220***
Analytical					(.036)	(.043)
Cooperation					−.087	−.012
Cooperation					(.064)	(.080)
Detail					.312***	.281***
Detail					(.053)	(.070)
Leadership					.221***	.135***
Leadership					(.032)	(.039)
Social					−.098**	.010
Social					(.042)	(.049)
Concern					−.065	−.117**
Concern					(.043)	(.046)
Control					.110*	.074
Control					(.063)	(.078)
Dependability					−.212***	−.174*
Dependability					(.076)	(.092)
Effort					−.021	−.044
Effort					(.055)	(.082)
Independence					−.107**	−.077
Independence					(.043)	(.047)
Initiative					−.036	−.036
Initiative					(.068)	(.091)
Innovation					−.135***	−.119***
Innovation					(.038)	(.044)
Integrity					−.106**	−.057
Integrity					(.049)	(.068)
Persistence					.066	.038
Persistence					(.060)	(.067)
Stress					.017	−.042
Stress					(.057)	(.072)
Weekly hours		.043***				.043***
Weekly hours		(.004)				(.004)
Weekly hours²		−.000***				−.000***
Weekly hours²		(.000)				(.000)
# times fired, let go, or laid off			−.057**			−.052**
# times fired, let go, or laid off			(.023)			(.021)
Observations	9,519	9,519	9,519	9,519	9,519	9,519
R ²	.206	.274	.213	.234	.245	.304

Note. All models include (1) base covariates spanning age, race, maternal education, and parental marital status; (2) education covariates including PVT, GPA, and educational attainment; (3) grooming and physical and personality attractiveness measures; and (4) daily activities. All weighted least squares specifications include grade and school fixed effects. Clustered standard errors are at the school level. PVT = Picture Vocabulary Test; GPA = grade point average.

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

The next set of results concerns job content. Salaries and wages vary by required skills and workplace attributes. Individuals with socioemotional skills that promote friendship ties may be sorting into jobs that place a high premium on these skills. We begin by looking within occupations, which we use as a crude proxy for job skills and attributes. Adding occupational fixed effects to the model slightly increases the popularity premium (Column 4). The existence of substantial within-occupation variation in earnings suggests that occupational sorting plays an insignificant role. This does not rule out the possibility that individuals are sorting on dimensions of job content that are disguised by occupational categories.

To evaluate these dimensions, we use occupational “Work Styles” that characterize the suitability of different personality styles for the job. Column 5 controls for a diverse set of qualities, including the importance of analytical thinking, leadership, cooperation, and social orientation for the job. Collectively, the 15 work style covariates reduce the popularity coefficient to 1.3 percent. Simultaneously, accounting for hours worked, job disruptions, occupation, and skills required yields the same popularity premium of 1.2 percent (Column 6) as the specification that only considers weekly hours worked (Column 2). The evidence so far supports friendship nominations as exerting an effect on earnings that is distinct from the labor market mechanisms summarized here.

Conclusion

We present robust evidence that popularity heralds long-term economic success for a representative sample of U.S. adolescents. In doing so, we rule out potentially confounding roles played by several mechanisms, ranging from academic attainment to beauty. Net of these effects, a 1 standard deviation increase in popularity still raises the earnings premium by 5.3 percent almost 15 years into the future. The majority of variance in eventual earnings is explained by popularity among same-sex friends, suggesting that the attributes promoting nominations among same-sex peers are distinct from those valued by opposite-sex peers.

These results suggest that an inherently social feature, popularity, captures a persistent attribute that generates success later in life. Popularity predicts income above and beyond conventional personality measures such as the Big Five and self-esteem, and works independently of standard institutional features such as race, class, or gender. We speculate that this is a “likability” effect—a socioemotive skill or behavior trait that may be difficult to self-report but that others recognize and respond via friendship nominations. Our measure of popularity aggregates actual friendships, rather than aspirational notions of popularity. As such, we believe that this measure likely captures the practice of being a better friend rather than social prominence, implying that we cannot directly compare results with prior work on the lasting reputation effects associated with ascriptions of popularity or peer rejection (Gest et al. 2006).

There are three basic ways through which such an effect could manifest. First, we cannot rule out a consistent and latent individual trait observed by other adolescents that simultaneously generates social recognition and adult success. This would render popularity’s association with income spurious on the unobserved feature. On the contrary, adolescents might directly develop this skill by virtue of being popular. To the extent that popularity requires interaction with more diverse types of people, popular adolescents can learn how to navigate complex social situations better than unpopular ones. Alternatively, the effect might be to enhance one’s sense of self, generating a positive social self-confidence that then allows them to move assuredly and successfully through workplaces. We cannot know from these data which mechanism (or others) might generate the observed association. Rather, we see this as a key question for future research to help identify the specific behavioral mechanism that peers are responding to as well as how such mechanisms differ across social groups.

Finally, we note that there are multiple other dimensions of success that extend beyond the simple economic measure of income. Conceptually analogous extensions of this work could identify features of job outcomes that hinge on sociability, such as performance ratings, to help disentangle human capital from socioemotional skills. The same set of skills that matter for economic prosperity may also influence psychological and emotional well-being and thus affect multiple life course outcomes, including family formation, marriage stability, or even community and neighborhood engagement. More work is necessary to explore the multifaceted relationship between popularity and long-term success.

Footnotes

Acknowledgements

The authors would like to thank participants at the Duke Network Analysis Center (DNAC) seminar for their constructive and thoughtful feedback on earlier drafts of this article.

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 the research, authorship, and/or publication of this article: This work has been support by Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Grant R01 HD075712, a Horowitz Foundation for Social Policy grant, and a James S. McDonnel Foundation Complexity Scholars award.

Supplemental Material

The online supplementary material is available at .

Notes

References

Adecco Staffing US. 2013. “Lack of Soft Skills Negatively Impacts Today’s U.S. Workforce.” Available at: http://pages.adeccousa.com/rs/adeccousa/images/2014-mind-the-skills-gap.pdf

Almlund

Mathilde

Duckworth

Angela Lee

Heckman

James J.

Kautz

Tim D.

2011. Personality Psychology and Economics. Cambridge, MA: National Bureau of Economic Research.

Arnett

Jeffrey J.

2000. “Emerging Adulthood: A Theory of Development from the Late Teens through the Twenties.” American Psychologist 55(5):469–80.

Aucejo

Esteban

. 2015. Explaining Cross-racial Differences in the Educational Gender Gap. London, England: Centre for Economic Performance, London School of Economics and Political Science.

Averett

Susan L.

Korenman

Sanders

. 1998. The Economic Reality of the Beauty Myth. Rochester, NY: Social Science Research Network.

Bagwell

Catherine L.

Newcomb

Andrew F.

Bukowski

William M.

1998. “Preadolescent Friendship and Peer Rejection as Predictors of Adult Adjustment.” Child Development 69(1):140–53.

Baker

Paula C.

Keck

Canada

Mott

Frank

Quinlan

Stephen

. 1993. NLSY Child Handbook: A Guide to the 1986-1990 National Longitudinal Survey of Youth Child Data. Rev. ed. Columbus, OH: The Ohio State University.

Baldasaro

Ruth E.

Shanahan

Michael J.

Bauer

Daniel J.

2013. “Psychometric Properties of the Mini-IPIP in a Large, Nationally Representative Sample of Young Adults.” Journal of Personality Assessment 95(1):74–84.

Biddle

Jeff E.

Hamermesh

Daniel S.

1998. “Beauty, Productivity, and Discrimination: Lawyers’ Looks and Lucre.” Journal of Labor Economics 16(1):172–201.

10.

Bowles

Samuel

Gintis

Herbert

. 1976. Schooling in Capitalist America: Educational Reform and the Contradictions of Economic Life. New York, NY: Routledge.

11.

Bowles

Samuel

Gintis

Herbert

Osborne

Melissa

. 2001. “The Determinants of Earnings: A Behavioral Approach.” Journal of Economic Literature 39:1137–76.

12.

Brown

B. Bradford

. 2004. “Adolescents’ Relationships with Peers.” Pp 363-394 in: Lerner

R. M.

Steinberg

(Eds.) Handbook of Adolescent Psychology. 2nd ed. Hoboken, NJ: John Wiley and Sons.

13.

Cairns

Robert D.

Cairns

Beverly D.

1995. Lifelines and Risks: Pathways of Youth in Our Time. Cambridge, England: Cambridge University Press.

14.

Case

Anne

Paxson

Christina

. 2008. “Stature and Status: Height, Ability, and Labor Market Outcomes.” Journal of Political Economy 116(3):499–532.

15.

Cawley

John

Heckman

James

Vytlacil

Edward

. 2001. “Three Observations on Wages and Measured Cognitive Ability.” Labour Economics 8(4):419–42.

16.

Christakis

Nicholas A.

Fowler

James H.

2014. “Friendship and Natural Selection.” Proceedings of the National Academy of Sciences 111(3):10796–10801.

17.

Coleman

J. Michael

Fults

Betty A.

1982. “Self-Concept and the Gifted Classroom: The Role of Social Comparisons.” Gifted Child Quarterly 26:116–20.

18.

Conti

Gabriella

Galeotti

Andrea

Müller

Gerrit

Pudney

Stephen

. 2013. “Popularity.” Journal of Human Resources 48(4):1072–94.

19.

Cunha

Flavio

Heckman

James J.

Lochner

Lance

Masterov

Dimitriy V.

2006. In: Hanushek

E. A.

Welch

(Eds.) “Chapter 12 Interpreting the Evidence on Life Cycle Skill Formation.” Pp. 697–812 in Handbook of the Economics of Education, Vol. 1. Amsterdam: North Holland.

20.

Davis

James A.

1966. “The Campus as a Fish Pond: An Application of the Theory of Relative Deprivation to Career Decisions of College Men.” American Journal of Sociology 72:17–31.

21.

Duckworth

Angela L.

Seligman

Martin E. P.

2005. “Self-Discipline Outdoes IQ in Predicting Academic Performance of Adolescents.” Psychological Science 16(12):939–44.

22.

Erikson

Erik H.

1968. Identity: Youth and Crisis. New York: W.W. Norton.

23.

Farkas

George

. 2003. “Cognitive Skills and Noncognitive Traits and Behaviors in Stratification Processes.” Annual Review of Sociology 29:541–62.

24.

Festinger

Leon

. 1954. “A Theory of Social Comparison Processes.” Human Relations 7(3):117–40.

25.

Fletcher

Jason

. 2014. “Friends or Family? Revisiting the Effects of High School Popularity on Adult Earnings.” Applied Economics 46(20):2408–17.

26.

Fortin

Nicole M.

2008. “The Gender Wage Gap among Young Adults in the United States: The Importance of Money versus People.” Journal of Human Resources 43(4):884–918.

27.

Gest

Scott D.

Sesma

Arturo

Masten

Ann S.

Tellegen

Auke

. 2006. “Childhood Peer Reputation as a Predictor of Competence and Symptoms 10 Years Later.” Journal of Abnormal Child Psychology 34:509–26.

28.

Giordano

Peggy C.

2003. “Relationships in Adolescence.” Annual Review of Sociology 29(1):257–81.

29.

Glaeser

Edward L.

Laibson

David

Sacerdote

Bruce

. 2002. “An Economic Approach to Social Capital.” The Economic Journal 112(483):F437–58.

30.

Hamermesh

Daniel S.

Biddle

Jeff E.

1994. “Beauty and the Labor Market.” The American Economic Review 84(5):1174–94.

31.

Hansell

Stephen

. 1981. “Ego Development and Peer Friendship Networks.” Sociology of Education 54:51–63.

32.

Hartup

Willard W.

1996. “The Company They Keep: Friendships and Their Developmental Significance.” Child Development 67(1):1–13.

33.

Heckman

James J.

Rubinstein

Yona

. 2001. “The Importance of Noncognitive Skills: Lessons from the GED Testing Program.” The American Economic Review 91(2):145–49.

34.

Heckman

James J.

Stixrud

Jora

Urzua

Sergio

. 2006. “The Effects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior.” Journal of Labor Economics 24(3):411–82.

35.

Herrnstein

Richard J.

Murray

Charles

. 1996. Bell Curve: Intelligence and Class Structure in American Life. 1st ed. New York: Free Press.

36.

Jencks

Christopher

. 1979. Who Gets Ahead? The Determinants of Economic Success in America. New York, NY: Basic Books.

37.

Jensen

Arthur R.

1998. The G Factor: The Science of Mental Ability. 1st ed. Westport, CT: Praeger.

38.

Kuhn

Peter

Weinberger

Catherine

. 2005. “Leadership Skills and Wages.” Journal of Labor Economics 23(3):395–436.

39.

Ladd

Gary W.

2006. “Peer Rejection, Aggressive or Withdrawn Behavior, and Psychological Maladjustment from Ages 5 to 12: An Examination of Four Predictive Models.” Child Development 77(4):822–46.

40.

Lin

Nan

. 2002. Social Capital: A Theory of Social Structure and Action. Cambridge, England: Cambridge University Press.

41.

Lundberg

Shelly

. 2014. Skill Disparities and Unequal Family Outcomes. IZA Discussion Paper No. 8346.

42.

Marsh

Herbert W.

Parker

John W.

1984. “Determinants of Student Self-Concept: Is It Better to Be a Relatively Large Fish in a Small Pond Even If You Don’t Learn to Swim as Well?” Journal of Personality and Social Psychology 47:213–31.

43.

McFarland

Daniel A.

Moody

James

Diehl

David

Smith

Jeffrey A.

Thomas

Reuben J.

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

44.

McPherson

Miller

Smith-Lovin

Lynn

Cook

James M.

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

45.

Moody

James

. 2004. “The Structure of a Social Science Collaboration Network: Disciplinary Cohesion from 1963 to 1999.” American Sociological Review 69(2):213–38.

46.

Mueller

Gerrit

Plug

Erik

. 2006. “Estimating the Effect of Personality on Male and Female Earnings.” Industrial and Labor Relations Review 60:3–22.

47.

Persico

Nicola

Postlewaite

Andrew

Silverman

Dan

. 2004. “The Effect of Adolescent Experience on Labor Market Outcomes: The Case of Height.” Journal of Political Economy 112(5):1019–1053.

48.

Rotter

Julian B.

1966. “Generalized Expectancies for Internal versus External Control of Reinforcement.” Psychological Monographs: General and Applied 80(1):1–28.

49.

Schulenberg

John E.

Sameroff

Arnold J.

Cicchetti

Dante

. 2004. “The Transition to Adulthood as a Critical Juncture in the Course of Psychopathology and Mental Health.” Development and Psychopathology 16:799–806.

50.

Sullivan

Harry S.

1953. The Interpersonal Theory of Psychiatry. New York: W.W. Norton.

51.

Tuma

Nancy Brandon

Hallinan

Maureen T.

1979. “The Effects of Sex, Race, and Achievement on Schoolchildren’s Friendships.” Social Forces 57(4):1265–85.

52.

Turner

Victor

. 1967. “Betwixt and between: The Liminal Period in Rites of Passage.” Pp 93-111 in The Forest of Symbols. Ithaca, NY: Cornell University Press.

53.

Wasserman

Stanley

Faust

Katherine

. 1994. Social Network Analysis: Methods and Applications. Cambridge, England: Cambridge University Press.

54.

Wolfe

Raymond N.

Johnson

Scott D.

1995. “Personality as a Predictor of College Performance.” Educational and Psychological Measurement 55(2):177–85.