
Social Capital and Dropping Out of High School: Benefits to AtRisk Students of Teachers' Support and Guidanceby Robert G. Croninger & Valerie E. Lee  2001 Do teachers provide students with valuable forms of social capital? Do these forms of social capital increase the likelihood that students complete high school, particularly students who are at risk of failure? Using data from the National Educational Longitudinal Study (NELS:88), we address these questions and examine whether social capital reduces the likelihood of dropping out between the 10th and 12th grades for a cohort of 11,000 adolescents who attended more than 1,000 public and private high schools between 1990 and 1992. We measure social capital in two ways: (a) students’ beliefs about how much their 10thgrade teachers support their efforts to succeed in school and (b) teachers’ reports about whether individual 10thgrade students receive guidance from them about school or personal matters. We find that teachers are an important source of social capital for students. These teacherbased forms of social capital reduce the probability of dropping out by nearly half. However, students who come from socially disadvantaged backgrounds and who have had academic difficulties in the past find guidance and assistance from teachers especially helpful. We discuss the implications of these findings for investigations of dropping out, risk, and social capital. Do teachers provide students with valuable forms of social capital? Do these forms of social capital increase the likelihood that students complete high school particularly students who are at risk of failure? Using data from the National Educational Longitudinal Study (NELS:88), we address these questions and examine whether social capital reduces the likelihood of dropping out between the 10th and 12th grades for a cohort of 11,000 adolescents who attended more than 1,000 public and private high schools between 1990 and 1992. We measure social capital in two ways: (a) students' beliefs about how much their 10thgrade teachers support their efforts to succeed in school and (b) teachers' reports about whether individual 10thgrade students receive guidance from them about school or personal matters. We find that teachers are an important source of social capital for students. These teacherbased forms of social capital reduce the probability of dropping out by nearly half. However, students who come from socially disadvantaged backgrounds and who have had academic difficulties in the past find guidance and assistance from teachers especially helpful. We discuss the implications of these findings for investigations of dropping out, risky and social capital. INTRODUCTION Underachievement, substance abuse, delinquency, and poverty plague the lives of many young people (Carnegie Council on Adolescent Development [CCAD], 1993; National Research Council [NRC], 1993). Premature parenting, homelessness, and youth violence also threaten the life chances of adolescents (Grubb Be Lazerson, 1988). These features of adolescents' lives bring with them personal tragedies that compromise a young person's successful entry into adulthood. They also represent social problems that reflect broader structural changes that increasingly threaten the social fabric of neighborhoods and communities throughout the United States (Coleman, 1990). What are the causes of these personal tragedies and communal difficulties? One answer is a decline in the effectiveness of the social institutions that young people rely on for support and guidance. These institutions, which include families, religious associations, community groups, and educational organizations, provide young people with valuable forms of interpersonal assistance—what many observers refer to as social capital (e.g., Coleman, 1988; StantonSalazar, 1997). Although each of these institutions plays an important role in preparing young people for adulthood, schools are central to this developmental process and are an essential source of social capital for adolescents (Coleman, 1990). Young people who face economic and social hardships at home are especially dependent on schools for support and guidance if they cannot find these forms of social capital elsewhere in their lives (Furstenberg & Hughes, 1995; StantonSalazar, 1997). The importance of schooling in the transition to adulthood makes dropping out particularly harmful for adolescents, as it cuts students off from potentially valuable information, developmental opportunities, and personal assistance before they have established linkages to other important social institutions (Natriello, 1986). The dropout rate has declined substantially since the early 1940s, when less than half of all persons between the ages of 25 and 29 completed high school (Rumberger, 1987). Current estimates of the proportion of adolescents who drop out nationwide range from 7% to 16% for highschool age students (Ekstrom, Goertz, Pollock, 8c Rock, 1986; Kaufman, McMillen, Be Sweet, 1996; National Center for Education Statistics [NCES], 1992; Rumberger, 1987). Dropout rates in urban areas, however, are considerably higher (Council of Great City Schools [CGCS], 1994; Rumberger & Thomas, 2000). The consequences of not completing high school have become increasingly serious for young people. Students who drop out face substantially higher unemployment rates, lower lifelong earnings, higher incidence of criminal activity, and a greater likelihood of health problems than students who complete high school or go on to college (Education Testing Service [ETS], 1995; Natriello, 1986; Pallas, 1986, 1995; Rumberger, 1987). Nationwide increases in educational attainment, as well as historic changes in labor markets, have reduced a dropout's access to employment and such employmentbased benefits as health care. These trends dramatically reduce the life chances of adolescents with low levels of education (Pallas, 1995). In this study, we consider whether students' access to social capital from high school teachers reduces the risk that students will drop out of school. Students’ relationships with their teachers represent a potentially valuable resource that can help students resolve problems and succeed at school (Furstenberg Be Hughes, 1995; StantonSalazar, 1997). Teachers can provide students with emotional support and encouragement, information and guidance about personal or academic decisions, and additional assistance with schoolwork. We argue that one explanation for why some students complete high school rather than others is that they have more resources to draw upon based on their network of relationships with adults. Although we examine whether teacherbased forms of social capital reduce the probability of dropping out for all students, we also explore whether these forms of social capital have special benefits for students at risk of educational failure. We consider three types of atrisk students: (a) those who are members of socially disadvantaged groups (Natriello, McDill, & Pallas, 1990), (b) those who experience schoolrelated or academic difficulties prior to entering high school (Catterall, 1998), and (c) those who fall into both categories. We hypothesize that atrisk students have the most to gain from social capital from teachers but also the most to lose if they are without it. To test this hypothesis, we examine whether qualities of students' relationships with their teachers increase the probability that atrisk students complete high school. Before describing our study, we review the research on dropping out, alternative conceptualizations of risk, and reasons why teacherbased social capital may encourage high school completion. After providing this background, we present the research questions that guide our study, describe our methods, and present the results of .our investigation. In the final section, we discuss the implications of these findings for future investigations of dropping out, risk, and social capital. BACKGROUND DROPPING OUT Research on dropping out must confront a multitude of theoretical and methodological, problems. Central among these problems is disagreement about how to define and measure dropping out (for overviews, see Pallas, 1986, 1989; Rumberger, 1987). National longitudinal studies provide one set of estimates of the prevalence of dropping out. Data from the National Educational Longitudinal Study (1995), which followed a cohort of 8th graders in 1988 through high school, indicate that roughly 6% of students drop out during the first 2 years of high school (NCES, 1992; Rumberger, 1995), whereas 11% leave high school between the 10th and 12th grades (Kaufman et al., 1996). Data from High School and Beyond (1995), which followed a cohort of 10th graders 8 years earlier, suggest a slightly higher dropout rate of 14% for students enrolled in the last 2 years of high school (Ekstrom et al., 1986). Dropout rates vary widely between high schools (Pallas, 1986; Rumberger, 1995; Rumberger Sc Thomas, 2000) and between student populations within a school (Rumberger, 1987, 1995; Rumberger & Thomas, 2000). Large comprehensive high schools, particularly in urban settings, report the highest dropout rates (Bryk & Thum, 1989), with rates over 50% of the 9thgrade cohort in some urban high schools (CGCS, 1994). Dropout rates are also higher for racial and ethnic minority students when compared to white students, higher for students from lowincome households, singleparent households, and families in which one or more of a student's parents also failed to graduate from high school (Rumberger, 1987, 1995; Natriello et al., 1990). Several studies have also found male students to be more likely to leave school than female students (Pallas, 1986; Rumberger, 1987). Most dropouts leave school sometime between the 10th and 12th grades (Frase, 1989; Rumberger, 1995), when the legal age for doing so is 16 in most states. Longterm longitudinal studies, however, indicate that dropping out can be a cumulative process of disengagement that begins as early as the first grade (Alexander, Entwisle, & Horsey, 1997). Students who eventually drop out of school often have a history of absenteeism (Lee & Burkam, 1992), academic trouble (Bryk & Thum, 1989), and other forms of disengagement from school life (Finn, 1989; McNeal, 1995). When these problems reach a critical threshold, some students attempt to "resolve" them by leaving school—putting past failures behind them. Both ethnographic studies and surveybased studies indicate that students who leave high school prior to graduation often cite a lack of social and academic support as one reason for doing so. They feel disconnected from teachers, despite selfdescribed efforts to gain assistance from school personnel. Dropouts frequently complain that their teachers do not care about them, are not interested in how well they do in school, and are unwilling to help with problems (Fine, 1986; MacLeod, 1987). In exit interviews, roughly half of all dropouts say that they left school because they were failing or just didn't like Being at school, whereas onethird say they quit because they didn't get along with their teachers or other students (Catterall, 1998). By their own accounts, many dropouts have fewer positive social interactions and less access to assistance from teachers than their more successful peers. Positive social relationships can create powerful incentives to attend school, even when schoolwork is difficult and classroom expectations are troublesome (LeCompte & Dworkin, 1991; Wehlage, Rutter, Smith, Lesko, & Fernandez, 1989). Such relationships may serve as a safety valve for adolescents, providing them with emotional support, encouragement, and actual assistance when academic or personal problems threaten to overwhelm them. During these difficult times, teachers and other adults may play an important role in preventing negative developmental outcomes from occurring. By being a reliable source of emotional support, guidance, and assistance to adolescents, teachers may bolster students' confidence and strengthen their ability to acquire a high school education, especially when students experience difficulties at scho6l or elsewhere in their lives (Luthar & Zigler, 1991; Wehlage et al, 1989). RISK STATUS The construct of risk is important in studies of dropping out (Natriello, 1986; Natriello et al., 1990; Pallas, 1989). Because dropping out is the ultimate form of educational withdrawal, correlates of dropping out often serve as risk factors in studies that seek to identify groups of students who are likely to have trouble in school. As a construct, risk indicates the probability of future difficulties and not an explanation for why difficulties occur. Ambiguity regarding the causes of risk have led some observers to question the use of the term, suggesting that it unfairly labels students and creates stereotypes that lower expectations for achievement (Swadener & Lubeck, 1995). Others argue that the construct of risk should encompass not demographic but actual attitudinal and performance differences between individuals, as the latter rely on individual rather than group differences to predict outcomes (Catterall, 1998). In general, two broad categories of risk factors—social and academic—can be identified in the literature. Social risk refers to demographic factors correlated with a higher likelihood of school difficulties (Pallas, 1989). Race, languageminority status, gender, family income, parents' education, and family structure have all been used as social factors to characterize students' risk of experiencing schoolrelated problems (Natriello et al., 1990). The greater the accumulation of social disadvantage associated with factors, the greater the presumed risk of failure. Thus, students who come from poor, singleparent households, where parents did not graduate from high school, are thought to be more at risk than students who possess only one or two of these socialrisk factors (Luthar & Zigler, 1991; Schorr, 1988). Academic risk refers to students' school performance at a specific point in time (Catterall, 1998). Unlike social risk, academic risk highlights the actual manifestation of schoolrelated problems. These early problems predict future difficulties in schools, such as absenteeism and skipping classes (Bryk & Thum, 1989; Lee &: Burkam, 1992), disengagement from school activities (Finn, 1989; McNeal, 1995), and, of course, dropping out (Alexander et al., 1997). Such factors as low grades, low educational expectations, earlygrade retention, and discipline problems gauge how much trouble students have had in school. If students fail to resolve these difficulties positively, they can find themselves in a downward spiral that leads to other adolescent and schoolrelated troubles (Alexander et al., 1997; Schorr, 1988). Such a depiction of risk fits well with studies that conceptualize dropping out as a cumulative process of withdrawal from school. Although constructs of social and academic risk are obviously related, we think that it is important that they be distinguished conceptually. Some adolescents may find it difficult to break a cycle of failure in school even though they come from privileged social backgrounds, whereas students from socially disadvantaged backgrounds^{i} may experience academic difficulties even when they have positive academic histories. We argue that as students enter high school they can be classified into four broad categories of risk: (a) students with no indication of risk, (b) students socially at risk, (c) students academically at risk, and (d) students both socially and academically at risk of failure. Most studies of risk and dropping out compare probabilities for students with no risk to students with one or both types of risk; far less is known about how the process of dropping out may differ for students with different configurations of risk status. SOCIAL CAPITAL Social capital highlights ways in which social organization—in the form of small networks of relationships and broad societal patterns of interactions— enhances the productive capacity of individuals and groups (Coleman, 1988, 1990). Such relationships are thought to be especially important to adolescents, who often require adult guidance and assistance in performing important developmental tasks (StantonSalazar, 1997). Although a number of studies demonstrate an empirical link between dropping out and social capital, most of these studies equate social capital with social control. In these studies, structural features of children's families, neighborhoods, or schools are thought to be indicators of the strength of normative ties (or adult expectations) that govern adolescent behaviors (Coleman, 1988; Hofferth, Boisjoly, & Duncan, 1998; Lichter, Cornwell, & Eggebeen, 1993; Smith, Beaulieu, & Israel, 1992). Studies that do focus on studentteacher relationships often use theories of alienation to explain the impact of such relationships on dropping out (LeCompte & Dworkin, 1991; Newmann, 1981; Wehlage et al., 1989) Although alienation theory goes beyond a narrow focus on social control as an explanation for why some students drop out of school rather than others, such theories shed little light on the ways in which students and teachers negotiate access to valued social resources. As social network theories point out, social relationships represent important distribution points for socially valued resources and opportunities (Wellman, 1983). Access to such ties and the social capital that flows through them provides individuals with important advantages. Whereas alienation theory tends to focus on the absence of positive ties between students and teachers in explaining student outcomes, we believe that it would be more useful to focus on the nature of social ties that different groups of students have with their teachers. From the perspective of social capital, differences in the probability of dropping out can be explained by differences in the quality of the social networks that comprise a student's interactions with teachers. Such networks can provide students with valuable resources including emotional support, information, guidance, or assistance in accomplishing school tasks (StantonSalazar, 1997). When students have trouble in school or when schoolwork becomes difficult for them, students may especially need these types of resources to be successful. Whether students gain access to such resources depends on the structural characteristics of their social networks (e.g., the extent to which adults permeate such networks or the strength of the ties that comprise a network; Coleman, 1988; Wellman, 1983), the institutional utility of the resources that networks make available to adolescent members (Bourdieu, 1985; Fortes, 1998), and the normative expectations that govern the distribution of support among adolescents in a school (Barrera & Baca, 1990; Fortes, 1998; Wellman, 1983). We recognize that students may benefit from social capital other than from their teachers. Administrators, guidance counselors, teacher aides, coaches, clerical workers, or even custodial staff may provide students with advice, guidance, and support. Peers may also provide valuable forms of assistance and serve as confidants and mentors to students. Nonetheless, we contend that, among the adults that students interact with at school, students' relationships with their teachers are most important. Studies of dropping out consistently correlate beliefs about teachers with early school departures (Catterall, 1998; Fine, 1986; Rumberger, 1987, 1995). Teachers seem to represent an especially important source of social capital available to adolescents in considering whether to stay or leave. RESEARCH QUESTIONS Based on these theoretical perspectives, we pose three sets of questions; all focus on whether attributes of students’ relationships with their teachers—what we call social capital—reduce the probability that they will leave school before graduation. We are also interested in determining whether the impact of social capital on dropping out depends on students' risk status when they enter high school. For each question, we provide specific hypotheses based on our reading of the literature. 1. Risk status and dropping out. What is the relationship between social risk, academic risk, and the probability that students drop out of high school? Does the accumulation of socialrisk factors increase the probability of dropping out, independent of academic risk? We hypothesize that the accumulation of socialrisk factors increases the probability of dropping out, as does the combination of academic and social risk. We also pose that each form of risk—social and academic—has an independent effect on the likelihood that students do and do not complete high school. 2. Social capital from teachers and dropping out. Do forms of teacherbased social capital influence the likelihood that students drop out of high school? Does access encourage high school completion even when students' risk status on entering high school is taken into account? We believe that social capital through positive and supportive relationships with teachers has beneficial effects for all high school students, including students who enter high school at risk of failure. The more social capital students have from teachers, the more likely they are to complete high school. 3. Risk status, social capital, and dropping out. Do different dimensions of social capital affect the likelihood of high school completion equally? Do students with specific combinations of social and academic risk experience additional benefits from having these dimensions of social capital available to them? We hypothesize that different dimensions of social capital will have different effects on dropping out. Moreover, we contend that the benefits of having social capital, as well as the negative consequences of not having it, will be greatest for students at risk of failure. METHOD DATA To address these questions, we use data from the National Educational Longitudinal Study (NELS), a biennial generalpurpose survey sponsored by the National Center for Education Statistics (NCES). The purpose of NELS is to "study the educational, vocational, and personal development of students at various grade levels, and the personal, familial, social, institutional, and cultural factors that may affect that development" (Ingels, Dowd, Baldridge, et al., 1994, p. 1). NELS surveyed a large and nationally representative sample of 8thgrade students attending public and private schools in 1988. Survey staff then followed the development of these students through high school into postsecondary institutions and the work force, resurveying them every 2 years (e.g., 1990 and 1992). We use data from the base year, first, and second followup surveys, when students were 8th, 10th, and 12th graders. The analytic sample for this study is a large subset of the 8th through 12thgrade longitudinal panel, restricted to students with full data on key variables. We structured the sample to investigate dropping out during the last 2 years of high school, when the majority of dropouts depart. The analytic sample includes 10,979 students who attended 1,063 public and private high schools between 1988 and 1992, about twothirds of the 8thgrade students who participated in the first and second followup surveys.^{ii} MEASURES Dropout Status Our dependent variable, dropout status, measures whether students left school between the 10th and 12th grades. The 12thgrade NELS student data set included an item that indicated students' dropout status at the end of the 199192 school year (Ingels, Dowd, Stipe, et al, 1994). Using this item, we created a dummycoded variable for whether students dropped out during the last two years of school (1 = yes, 0 = no). Eleven percent of the students in the sample fell into this category. Because we had no measure of students' social capital in the 9th grade, we did not include in our sample students who left school prior to the 10th grade.^{iii} Risk Status An important construct for our study is students' risk status when they enter high school. We constructed a series of measures of students' risk status from items included in the base year student survey. The first set indicates whether students fall in one or more of the following socialrisk categories: (a) live in household where family income is at or below the 1988 poverty threshold; (b) belong to a languageminority group; (c) belong to a disadvantagedminority group (i.e., Black, Hispanic, or American Indian); (d) live in a singleparent household; or (e) have a mother who failed to complete high school (or a father in the case of a maleheaded, singleparent household). Each of these measures is significantly associated with dropping out in our sample and has been used to characterize students as socially at risk in other studies (Natriello, 1986; Natriello et al., 1990; Rumberger, 1987).^{iv} Fortyfour percent of students in the sample possessed one or more of these socialrisk factors. We constructed three measures of socialrisk status based on the abovementioned measures: (a) students with one or more socialrisk factors, (b) two or more socialrisk factors, and (c) three or more socialrisk factors. Fewer than 3% of students possessed more than three factors. We coded each variable to indicate the incremental effect of additional socialrisk factors on dropping out. When all three variables are in a model, the first measure estimates the probability of dropping out for students with 0 versus 1 factor, the second measure the probability of dropping out for students with 1 versus 2 factors, and the third measure the probability of dropping out for students with 2 versus 3 or more factors.^{v} The second form of risk, academicrisk status, indicates whether students experienced one or more of the following difficulties before high school: (a) gradepoint average of less than a C during middleschool years, (b) held back between the 2nd and 8th grades, (c) no expectation of education beyond high school, (d) sent to the office more than once during the 1st semester of their 8thgrade year, or (e) parents notified more than once about schoolrelated problems during the same time period. Each of these factors is also significantly associated with dropping out for our data, and these or similar factors have been used to describe students as academically at risk in other studies (Catterall, 1998; Rumberger, 1987).^{vi} We use a dummycoded version of this measure in the study (1 = academically at risk, 0 = not).^{vii} Slightly more than onethird of students in the sample possess one or more of these academicrisk factors. Nearly half of the students in our sample who are socially at risk entered high school also academically at risk. Social Capital We use two measures of social capital in this study: studentteacher relations and studentteacher talks outside the classroom. The first taps the notion of students' trust in their teachers, whereas the second is an indication of potentially helpful exchanges between students and one or more of their teachers outside of the classroom. Student teacher relation is a composite made up of six descriptions of teachers made by 10^{th} grade students (Cronbach's alpha .79). Items include whether these adolescents believe teachers (a) are interested in them, (b) value what they say, (c) are good at teaching, (d) care about them and whether they succeed in school, (e) recognize and praise them when they work hard, and (f) put them down in the classroom (reversed). We used principal components factor analysis to weight items. We transformed the composite into zscores (M = 0, SD = 1). Along with student data, NELS includes teacher reports about students in the sample (Ingels, Dowd, Baldridge, et al., 1994). On average, two 10^{th}grade teachers who taught students in one of four core subjects (mathematics, science, social studies, or English) answered questions about their classrooms, schools, colleagues, and students. A single item in the 10thgrade teacher questionnaire asked teachers whether each student talked with them about schoolwork, academic decisions, or personal matters outside of class. Using this item, we created a dummycoded variable to indicate whether or not at least one of the two teachers said yes (1 = yes, 0 = no). Teachers identified roughly half of the students as receiving some form of assistance outside of class. The correlation between this measure and student teacher relation is positive but low (r = .10).^{viii} Measures of social capital used in other studies are more likely to tap an aspect of social structure, parent values, or behaviors than relationships between students and adults in school (e.g., Coleman, 1988; Lichter et al., 1993; Smith et al., 1992). Surprisingly few of these studies even consider teachers as sources of social capital for young people, even though adolescents interact with their teachers daily over extended periods of time (see StantonSalazar & Dornbusch, 1995, for an exception). Nonetheless, we argue that teachers provide students with the most direct source of assistance. Although students may not have equal access to support from their teachers, we believe that those who do are more likely to succeed in school than those who do not. Control Variables Numerous studies indicate that teachers respond more positively to highachieving, motivated students (Lee, Dedrick, & Smith, 1991) and that such students are also more likely to graduate from high school (Rumberger, 1987, 1995). Studies also suggest that there are important differences in how specific demographic groups solicit or react to aid from others (Barrera Be Baca, 1990). To control for these possible confounding effects, we use four different control variables in this study: student gender, 8thgrade achievement, 8thgrade academic behaviors, and 10thgrade academic behaviors. Student gender is a dummycoded variable indicating whether a student is female or male (1 = female, 0 = male). Nearly half of the students in the sample are female. Eighthgrade achievement is the average of students' reading comprehension and mathematics' scores on the NELS' achievement tests; each of these tests was administered when students were in the 8th grade. We use the z score of this measure in our study (M = 0, SD =1). Because we want to control for students' academic behaviors before they enter high school, as well as investigate its effects while students are in high school, we developed measures for these behaviors in the 8th and 10th grades. The base year and first followup surveys included identical items about students' academic behaviors. We used principal component factor analysis to create two composites from students' answers to these survey items, one for students' 8thgrade academic behaviors (Gronbach's alpha, 0.52) and one for students' 10thgrade academic behaviors (Cronbach's alpha, 0.56). The higher the value, the more students said that they (a) attend lasses regularly, (b) spend time doing homework, (c) go to class prepared, and (d) generally act in ways associated with being a good student. We zscored each composite for use in our study (M = 0, SD = 1). The correlation between the 8thgrade and 10thgrade measure is positive and modestly high (r = .51). (See the appendix for a detailed description of all variables used in this study.) ANALYTIC MODEL We rely on logistic regression to examine the effects of social capital on dropping out, examining models in order of the simplest to the most complex (Cohen Be Cohen, 1983).^{ix} We enter gender and measures of risk status in the first step, the academic background controls in the second step, the two measures of social capital in the third step, and the measure of high school behaviors in the final step. This strategy permits us to examine reductions in the magnitude of coefficients due to the entry of conceptually grouped measures. We are especially interested in considering how access to social capital may reduce the risk of dropping out for specific risk populations (Step 3) and whether high school behaviors reduce the effects of social capital on dropping out (Step 4). We also consider possible interaction effects between each measure of social capital and our measures of risk status, using models that do not include students' 10thgrade behaviors.^{x} Because the effects of risk are central to these analyses, it is possible that the use of 8thgrade achievement and behaviors as baseline controls "overcontrols" for students' academic background. Both social risk and academic risk are associated with students' 8thgrade achievement and behaviors. The relationships are strongest between our dummycoded measure of academicrisk status and the baseline controls, as these measures are both empirically and conceptually linked (correlations are r = .38 for academic behaviors and r = .39 for achievement). To minimize multicollinearity in these analyses, and to facilitate the interpretation of results, we stratify our investigation by students' academicrisk status. That is, we estimate the probability of dropping out for two groups of students: (a) students who enter high school with no academicrisk factors (n = 7,513 students) and (b) students for whom one or more academicrisk factors are present (n = 3,466 students). Although this strategy makes a minimal sacrifice to statistical power, it preserves the importance of controlling for students' academic background and facilitates the interpretation of results. We examine identical models that include measures of social risk for each population. This strategy permits an estimation of the effects of social capital for four categories of students: (a) students with no risk, (b) socially atrisk students with no history of trouble in school, (c) academically atrisk students from socially advantaged homes, and (d) academically atrisk students from socially disadvantaged homes. Using these estimates, we compare the relative benefits of social capital on the probability of dropping out for students with these different combinations of risk status. RESULTS BIVARIATE RESULTS We describe the characteristics of students who complete and drop out of high school in Table 1. With the exception of gender, students who complete high school are significantly different from students who drop out on all variables (p < .001). These differences are substantial in the case of risk, academic background, and high school behaviors. Risk Status There are large differences in the risk status of students who complete and those who drop out of high school. Dropout students are twice as likely as graduates to be academically at risk. Dropout students are also more likely to be socially at risk. Dropout students are nearly three times as likely as graduates to possess three or more socialrisk factors and are roughly twice as likely to possess two factors. Dropout students are also more likely than students who complete high school to have a single risk factor, but not to the same degree as multiple factors. These associations, though unadjusted for other factors that might explain dropping out, confirm the empirical link between the measures of risk used in this study and a failure to complete high school. Academic Background Academic background on entering high schools is strongly related to dropping out.^{xi} Students who drop out between the 10th and 12th grades enter high school with much lower levels of achievement and less desirable academic behaviors. Academic background differs between students who do and don't complete high school by .8 SD for 8thgrade achievement and .6 SD for academic behaviors—very large differences. These results are consistent with studies that show dropouts have weaker academic backgrounds and often a history of schoolrelated trouble when they enter high school (Alexander et al., 1997; Bryk & Thum, 1989; Catterall, 1998; ETS, 1995; Lee & Burkam, 1992). TeacherBased Social Capital Dropouts have less of the two forms of social capital that we examine in this study than students who complete high school. Dropouts characterize their relationships with teachers less positively than graduates (a difference of .4 SD), and teachers report a smaller proportion of dropouts than eventual graduates receive advice outside of class (12% less). These results, coupled with differences in 8thgrade achievement and academic behaviors, portray dropouts as students with greater educational needs but fewer teacherbased resources to draw upon in school. Not only do dropouts enter high school with substantially lower achievement and less positive behaviors, they also have less support and guidance from teachers to address these difficulties. High School Behaviors Students who do and do not graduate also differ substantially in academic behaviors during high school. Graduates are more likely than dropouts to spend time studying after school, come to class prepared, and believe others see them as good students. This difference is even greater than the difference between eventual graduates and dropouts in the 8th grade (.9 v. .6 SD). Clearly, students' academic behaviors during high school are more strongly related to high school completion than students’ academic behaviors before high school. MULTIVARIATE RESULTS We use logistic regression to test whether the relationship between social capital and dropping out persists, once we take account of differences between students when they enter high school. Table 2 presents the logistic regression coefficients for the effects of gender, academic background, socialrisk status, social capital, and 10thgrade behaviors on the probability of dropping out. Because we divided the sample by academic risk, there are two sets of effects with each model. The first column presents results for students who enter high school without any indication of academic difficulty; the second, results for students who enter high school academically at risk of failure. Coefficients are in log odds. The Chisquare statistics at the bottom of the columns indicate that each model predicts dropping out better than would be expected by chance. All continuous variables in the models are in a zscore metric based on the entire population of students. We centered gender, however, on the population mean for each sample (or column). Recall that social risk is coded to estimate the incremental effect of additional factors (i.e., 0 v. 1 factor, 1 v. 2 factors, and 2 v. 3 or more factors). Negative coefficients reduce the probability of dropping out; positive coefficients increase the probability. Does Risk Matter? The constants for each analysis represent the log odds of dropping out for students with average characteristics who enter high school without (column 1) and with (column 2) academic risk. Academic risk increases the probability of dropping out, as shown by the difference in constants (1.70 v. 3.43). Although both log odds are negative (students are more likely to graduate than not), the log odds, as represented by the constants, are greater for students with academic risk than students without academic risk (the value is closer to 0). The difference in likelihood is 1.73 in model 1 and decreases to 1.09 in subsequent models where the effects of academic background, social capital, and high school behaviors are included. Beyond academic risk, social risk also contributes to dropping out. For both students with and without academic risk, a single socialrisk factor increases the logs odds of dropping out (.72 and .65, respectively). An additional two factors (i.e., 35 factors) increases the log odds from .65 to 1.62 in model 1 (.65 + .97) for students with no academic risk, but additional factors make no difference for students who entered high school academically at risk. Although accumulation of socialrisk factors accentuates the probability of dropping out, it does so only for those students who begin high school without any academicrelated problems. Such a pattern of effects suggests that a disturbing proportion of atrisk students fail to complete high school for nonacademic reasons. Social advantage or disadvantage becomes less important in predicting dropping out once students experience academic difficulties in school. Although gender is not the focus of this study, its effects are sufficiently surprising to warrant attention. Female students are more likely than male students to drop out of high school in models 24, once risk and social capital are taken into account. In model 1, female students are more likely to drop out than male students who have no history of trouble in school. This result is different than the percentages displayed in Table 1, which shows no difference between males and females in the likelihood of dropping out of high school (and different from what has been reported elsewhere [see Pallas, 1986; Rumberger, 1987]). A plausible explanation is that once academic achievement and behaviors are taken into account, female students are more likely than male students to drop out for nonschoolrelated reasons. For example, female students may be especially vulnerable to disruptive life events, such as premature parenting or requests to assist parents with childcare, that compromise their ability graduate from high school. These effects indicate important differences in the likelihood of dropping out for students, even after controlling for differences in academic background (model 2) and 10thgrade behaviors (model 4). Both socially and academically atrisk students are more likely to drop out than their nonrisk counterparts, as are female students compared to male students. Additional socialrisk factors increase the likelihood of dropping out for students who enter high school with no risk, though not for students who already have low educational expectations and a history of schoolrelated difficulties at the end of the 8th grade. Even after controlling for the effects of academic background, both socially and academically atrisk students are still more likely to drop out of high school than students without these characteristics.^{xii} Does Social Capital From Teachers Matter? The results in models 3 and 4 of Table 2 indicate that social capital increases the likelihood that students will complete high school. Among students who enter high school with no history of academic difficulties, studentteacher relations significantly reduces the log odds of dropping out (b = .33), though studentteacher talks does not (b = .05). However, judging from the coefficients in Table 2, academically atrisk students benefit more from access to social capital than students with no history of difficulty in school. Among academically atrisk students, positive relations with teachers reduce the log odds of dropping out (b = .21), as do informal interactions with teachers outside of the classroom (b = .41). These informal opportunities for gaining assistance from teachers appear to be especially important for students who have a history of difficulties at school. When students' high school academic behaviors are considered (model 4), this, pattern of effects persists, although the magnitude of effect for studentteacher relations is roughly halved (.33 and .21 in model 3 compared .16 and .08 in model 4). The magnitude of the coefficient for studentteacher talks, however, is unchanged by taking students' 10thgrade behaviors into account. The reduction in effect for studentteacher relations may indicate that this dimension of social capital operates indirectly through students' academic behaviors. By gaining students' trust, teachers may be better able to encourage students to engage in positive academic behaviors; these behaviors in turn may prevent students from becoming mired in problems that prompt many students to leave school before graduating. Our models, however, do not permit us to test this hypothesis fully. Do the Benefits Depend on Risk Status? We used a simplified version of model 3 to investigate whether the benefits of social capital depend on students' risk status when they enter high school. Table 3 presents these results as a twostep logistic regression. In the first model, we estimate effects of access to social capital controlling for gender, social risk factors, and students' academic background. We recoded the measures of social risk to reflect statistically significant increments in model 3 of Table 2 (i.e., 0 v. 12 and 12 v. 35 factors for students with no academic risk, and 0 v. 15 factors for students with academic risk). As before, we estimate effects for students with and without academic risk separately. In model 2 of Table 3, we test for interactions between measures of social capital and social risk. For students with no academic risk, the benefits of social capital do not depend on students' socialrisk status—all four interaction terms are nonsignificant in model 2, column 1. However, for students who are academically at risk when they enter high school, effects do differ by socialrisk status. For socially atrisk students, benefits are almost exclusively linked to a single form of social capital—studentteacher talks—whereas students from socially advantaged backgrounds benefit most from studentteacher relations. Adjusting for these interactions, the effects for socially atrisk students are b = .07 for studentteacher relations (.42 + .35) and b = .47 for student teacher talks (.20 + .27); effects for students with no socialrisk factors are b = .42 for studentteacher relations and nonsignificant for studentteacher talks. Because results in Table 3 are in log odds, not probabilities, estimating the magnitude of the combined effects of social capital on dropping out is not straightforward. To facilitate comparisons, we present these results as probabilities in Figure 1. We use the coefficients in Table 3 to calculate the probability of dropping out for students who enter high school (a) at no risk, (b) socially at risk, (c) academically at risk, and (d) both socially and academically at risk of failure. Within each risk population, we calculate the probability of dropping out for students with high and low values for our two measures of teacherbased social capital.13 We consider students to have a high level of social capital if they talk to their teachers about schoolrelated or personal matters (studentteacher talks = 1) and have a value of + 1 SD on our measure of studentteacher relations; we consider students to have a low level of social capital if they do not talk to teachers outside the classroom (studentteacher talks = 0) and have a value of 1 SD on the measure of studentteacher relations. In Figure 1, shaded bars give the probabilities of dropping out for students with low levels of social capital; solid black bars display the probabilities for students with high levels. The first three sets of bars display results for students who possess different numbers of socialrisk factors (0, 12, or 35) but no history of academic difficulties; the second three sets of bars indicate probabilities for the same groupings of students who also have a history of schoolrelated trouble. The reference for each set of bars is a student who enters high school with an average academic background but varying risk status and levels of social capital. Estimates of the probability of dropping out range from a low of .03 to a high of .24. Students with low social capital have a higher probability of dropping out than students with high social capital, regardless of their risk status (i.e., each shaded bar is substantially higher than the adjacent black bar). Across the populations that we examine, the combined effects of high levels of social capital (studentteacher relations and talks) reduce the probability of dropping out by nearly half (e.g., from .05 to .03 for students with no risk and from .15 to .07 for students with academic but no social risk). The greater students' access to teacherbased social capital, the greater the probability that they will complete high school. Students who enter high school with academic trouble have the highest probability of dropping out of school, which is somewhat exacerbated if these students also come from socially disadvantaged backgrounds. Nonetheless, when academically atrisk students have more social capital, the probability of dropping out is reduced considerably. For students from socially disadvantaged households, the probability is still slightly higher than average (.14 v. .11), whereas for students from socially advantaged families, probabilities are less than the unadjusted average for this sample (.07 v. .11). DISCUSSION AND CONCLUSIONS TEACHERS AS A SOURCE OF SOCIAL CAPITAL Do students benefit from being able to draw on social capital through their relationships with teachers? Our findings suggest that students do. When adolescents trust their teachers and informally receive guidance from teachers, they are more likely to persist through graduation. Although teacherbased forms of social capital are generally beneficial for all students, those who benefit most are students most at risk of dropping out of high school. This is especially true for socially at risk students who enter high school with low educational expectations and a history of schoolrelated problems. We focus on only two dimensions of teacher support and guidance, which we describe as social capital—an attitudinal dimension that includes students' characterization of their social ties to teachers and a behavioral dimension based on teachers' reports about their informal exchanges with specific students. We also provide some indirect evidence that some benefits of social capital may operate through improved academic behaviors in high school, in that these behaviors explain roughly half of the positive effect associated with students' characterizations of their relationships with teachers. Although our measures of social capital are rather "blunt"—they do not detail the scope of students' interactions with teachers, the frequency of interactions, or the quality of support and guidance provided to them— the effects that we found are substantial. We suspect that with more detailed measures of social capital, the effects of teacherbased social capital would be even greater. Based on these findings, we argue that it is important to investigate the social dimensions of schooling in understanding not only dropping out but also how it and other schoolrelated difficulties might be prevented. Breakdowns in social control and alienation are prominent explanations for early school withdrawal. However, these theories say little about how adolescents make use of social ties to enhance the likelihood that they will succeed in school or what structural factors influence adolescents' access to social capital that flows through social networks. From the theoretical vantage point of social capital, an absence of positive social relationships and contacts with teachers denies students resources that help them develop positively. From the perspective of risk, access to social capital represents an important resource for students who enter high school with academic difficulties. These students represent a sizable proportion of the high school population (roughly onethird using our measures), and they are at considerable risk of dropping out of high school (roughly three times as likely as students who have had no academic difficulties). Supportive relationships and guidance from teachers increases the likelihood that socially and academically atrisk students complete high school. For those students most at risk of dropping out, students who are both socially and academically at risk of failure, informal exchanges with teachers outside of class are especially beneficial. Such contacts with teachers considerably boost their chances for graduation. These results confirm the fundamental proposition that we put forth at the beginning of this paper—teachers are an especially important source of social capital for adolescents at risk of educational failure. Such a finding is consistent with a growing recognition that the quality of students' relationships with teachers is an important predictor of educational success (see, e.g., McDermott, 1977; Raywid, 1995; Sizer, 1984). Socially disadvantaged students often have less access to all forms of capital in their lives, including financial and human capital (CGAD, 1993). To help compensate for the absence of social and academic resources in other areas of students' lives, teachers can provide tutoring, academic counseling, and guidance about educational decisions. Moreover, teachers, through their daily interactions with students, can provide young people with emotional support and encouragement, particularly when schoolrelated difficulties undermine their confidence in themselves as learners. As our findings indicate, even when students enter high school with a history of academic difficulties, direct guidance and support from teachers can make an important difference in their willingness to persist through graduation. INVESTIGATIONS OF RISK Another important component of our study is an investigation of risk and its consequences for dropping out. We identify two different types of risksocial and academic—and we argue that these types of risk can be used to construct four broad risk categories. We recognize that there are other ways of configuring risk, as well as objections to even using the construct of risk at all in studies. Nonetheless, we contend that our findings underscore the importance of conceptualizing risk both demographically and behaviorally. Our results identify negative consequences associated with each. Although academic risk has the largest impact on the probability that students drop out, social risk is also associated with elevated probabilities of school failure. These results are disturbing. Although roughly half of socially atrisk students enter high school without prior academic problems and expectations for postsecondary education, they are only half as likely to complete high school as students from more socially advantaged backgrounds. Support and guidance from teachers increase the likelihood that socially disadvantaged student complete high school, but they do not eliminate the negative consequences associated with poverty, low educational attainment by parents, minority status, or family composition. Any one of these factors, on average, still reduces the likelihood of high school completion. Why? Nothing in our results (or in our thinking) points to students themselves as an explanation. To the contrary, these are students with positive school trajectories whom one would expect to have an equally good chance of completing high school. However, we suspect that these results point to another important difference in the resources available to adolescents—the quality of the schools and neighborhoods in which many socially disadvantaged students learn and live. Even when students have access to caring and supportive teachers, the environments in which they live may still overwhelm them and perhaps even the adults upon whom they rely for support. Inequities in the scope of all forms of capital available to students and teachers may partially account for the difference in the high school completion rates of socially advantaged and disadvantaged students. We consider equity an important topic for educational research and policy. The problem of inequity has been historically framed in terms of educational outcomes for specific social populations less well served by public institutions (Howe, 1997). Considering social and academic factors in investigations of risk makes it possible to distinguish the consequences of different forms of risk—specifically those that involve demographic differences from those that involve performance differences. In our opinion, the former highlights important issues of equity and social justice that would be overlooked by relying exclusively on behavioral factors, as some have argued (e.g., Catterall, 1998), Failing to identify these issues is a disservice to socially disadvantaged students who face obstacles to success despite their positive school attitudes, behavior, and performance. SOCIAL CAPITAL AS A CONSTRUCT We have presented social capital as an alternative explanation to social control and alienation theories for why social relationships play a role in students' decisions to drop out or persist through graduation. As a construct, social capital gives greater prominence to student agency than does alienation theory, a perspective consistent with ethnographic studies that examine how students gauge their teachers' intentions, their own prospects for success at school, and the relationship between school and subsequent life chances (MacLeod, 1987). By examining two different aspects of teacherbased social capital, we further sought—within the confines of generalpurpose survey data—to identify if (and how) social capital influences high school completion. Our evidence indicates that it does. One mechanism appears to be positive academic behaviors engendered by students' trust in their teachers, the second actual guidance or assistance from teachers. We purposefully focused on the quality of students' relationships with teachers, for we felt it was important to examine directly the potential benefit of such relationships and exchanges to students. Prior studies that utilize social capital theory to examine dropping out have focused primarily on structural qualities that surround students' relationships with adults in and outside of school (Coleman, 1988; Furstenberg & Hughes, 1995; Hofferth, Boisjoly, & Duncan, 1998; Lichter, Gornwell, & Eggebeen, 1993; Smith, Beaulieu, & Israel, 1992). In our opinion, these studies overlook schoolrelated adults (in this case, teachers) as important sources of social capital, and fail to link the construct of social capital to any differences in students' access to the social resources that adults can provide to them. Moreover, such studies rarely highlight the manner in which adolescents themselves purposefully seek out and use teacherbased resources to enhance their school efforts. Without such a focus, the interpersonal and social mechanisms through which social capital influences educational outcomes remain largely hidden. Although we find social capital to be a useful construct in explaining why some students do and others do not drop out, our study does not capture its full explanatory potential. As a theoretical explanation for adolescents' academic development, social capital remains largely undeveloped. Demonstrating the positive effects of two dimensions of teacherbased social capital with NELS* data represents only a start; additional work is required to understand more fully the nature of the resources that students can acquire from their social networks, the effects of these resources on adolescent development (negative and positive), incentives for accessing these resources for different populations of students, and the factors that encourage teachers to actually provide support and assistance to students when doing so may require some personal sacrifice. Such work faces the challenge of drawing on related theoretical perspectives, while distinguishing social capital from other sociological constructs (Epstein, 1996; Fortes, 1998). Social capital is an intuitively appealing construct (Schneider, 1996). Even when shrouded by conceptual ambiguity, it focuses attention on the role of social relationships in promoting desirable outcomes for children—in the case of this study, on how social relationships between teachers and students play a critical role in whether students at risk of failure graduate from high school. Understanding the quality of these relationships is fundamental to understanding how students actually obtain human capital, as well as how schools can facilitate the process more successfully and fairly. As a construct, social capital highlights the deeply social nature of schools and holds out the promise of creating more supportive social institutions for adolescents and adults alike. Such a promise is well worth pursuing, not only for atrisk adolescents but children in general. APPENDIX: DESCRIPTION OF VARIABLES USED IN THE STUDY DEMOGRAPHIC CONTROL Gender This dummycoded measure comes from the NELS base year student survey (BYSEX). A value of 1 indicates a student is female (1 = Female), and a value of 0 indicates a student is male (0 =Male). Nearly half (49.4%) of the sample is female. The unweighted n is 10,979. ACADEMIC BACKGROUND Average 8thGrade Achievement We converted NELS 8thgrade reading comprehension (BY2XRIRR) and mathematics (BY2XMIRR) test scores to z scores. Then we averaged these values to create a measure of students' academic achievement before entering high school. The measure is standardized (M = 0, SD = 1), and the distribution is near normal. The unweighted n is 10,944. EighthGrade Academic Behaviors This composite has four dimensions: attendance (BYS75, BYS76, BYS77), daily preparation for class (BYS78A, BBYS78B, BYS78C), homework behaviors (BYHOMEWK), and a general rating of whether or not behaviors are thought to be characteristic of a "good student" (BYS56C). All items come from the 8thgrade student survey. We used principal component factor analysis to obtain factor scores for each dimension. The composite is the weighted sum of the four dimensions. We squared values to obtain a near normal distribution and then standardized the measure (M = 0, SD = 1). The unweighted n is 10,979. Cronbach's alpha for the composite is .53. SOCIAL RISK Poor Household This dummycoded measure indicates whether a student's family reported a household income below the federal poverty guidelines ($11,650 for a family of four in 1988). We used BYFAMINC and BYFAMSIZ, both variables on the NELS base year student data file, to determine poverty status (1 = Poor, 0 = Not). Nearly 15% of the students in the analytic sample come from a poor household. The unweighted n is 10,039. Disadvantaged Minority We created a dummycoded measure of whether students identified themselves as African American, Hispanic, or American Indian (1 = Disadvantaged Minority). Otherwise the value is 0 (0 = Not). We constructed the measure from BYRAGE, a composite included in the base year student file. Sixteen percent of the students in the sample belong to a disadvantaged minority group using this definition. The unweighted n is 10,904. Language Minority This variable denotes whether students come from a home in which a language other than English is usually spoken (1 = Yes, 0 = No). This variable relies on teacher and student reports and is a composite included in the NELS base year student file (BYLM). Eleven percent of students in the sample come from a languageminority household. The unweighted n for this variable is 10,978. Mother Dropped Out This measure indicates whether or not a student's mother—or father, in the case of a singleparent, maleheaded household—failed to graduate from high school (1 = Yes, 0 = No). We constructed the measure using student reports (BYS34A, BYSMB) and parent reports (BYP30, BYP31) taken from the base year NELS survey. Fourteen percent of the students in the sample have a parent who failed to graduate from high school. The unweighted n is 10,456. SingleParent Household This dummycoded measure indicates whether students come from a singleparent household (1 = Yes, 0 = No). We used BYGOMP, a composite measure in the base year NELS student file, to construct the variable. Sixteen percent of students are in singleparent households. The unweighted n is 10,908. One or More SocialRisk Factors Students with one or more socialrisk factors have a value of 1 on this dummycoded variable (1 = One or More); students with no socialrisk factors have a value of 0 (0 = None). Fortyfour percent of the students possessed one or more socialrisk factors. The unweighted n is 10,908. Two or More SocialRisk Factors Students with two or more socialrisk factors have a value of 1 on this measure (1 = Two or More); students with one or no factors have a value of 0 (0 = One or None). When used with one or more socialrisk factors in a model, the coefficient is equal to the difference in the outcome for students with one and students with two or more factors. Nearly 21% of the students in the sample have two or more socialrisk factors present. The unweighted n is 10,908. Three or More SocialRisk Factors This measure, like the other dummycoded variables, indicates whether students possess multiple risk factors (1 = Three or More, 0 = Two or Fewer). When used with two or more risk factors, the coefficient is equal to the difference in the outcome for students with two and students with three or more factors. Nine percent of the students have three or more factors present. The unweighted n is 10,908. ACADEMIC RISK Low Educational Expectations This dummycoded measure indicates whether students said that they do not expect to continue their education after graduating from high school (1 = Yes, 0 = No). We constructed the variable using BYS45, a measure on the base year student file. Ten percent of the students said that they do not anticipate continuing their education after high school. The unweighted n is 10,979. Held Back One or More Grades We created a dummycoded measure of whether students were held back between 2nd and 8th grades (1 = Yes, 0 = No). We used data from the base year parent survey (BYP46CH) and student survey (BYS47CI) to do so. Ten percent of the students in the sample were held back at least once since the second grade. The unweighted n is 9,891. Sent to Office Two or More Times This variable denotes whether students said that they were sent to the office two or more times for schoolwork and/or behavioral difficulties during the first semester of their 8thgrade year (1 = Yes, 0 = No). We used two items from the base year student survey to construct the measure (BY355A, BYS55B). Ten percent of the students said that they were sent to the office more than once during this period. The unweighted n for this variable is 10,883. Parents Warned Two or More Times This measure indicates whether students said that their parents were notified two or more times about low grades, attendance problems, and/or behavioral difficulties during the first semester of their 8thgrade year (1 = Yes, 0 = No). All items come from the base year student survey (BYS55C, BYS55D, BYS55E). Eighteen percent of the students said that their parents were notified more than once during this period. The unweighted n is 10,877. Low Grades This variable indicates whether students have a GPA lower than 2.0 or a "C" average (1 = Yes, 0 = No). We used BYGRADS, a composite measure in the NILS base year student file, to construct this variable. The composite is based on student selfreport of grades in basic subjects for the 6th, 7th, and 8th grades. Eight percent of students have less than a "CM average just before entering high school. The unweighted n is 10,979. One or More AcademicRisk Factors Students with one or more academicrisk factors have a value of 1 on this dummycoded variable (1 = One); students with no academicrisk factors have a value of 0 (0 = None). Roughly onethird (34%) of the students in the sample have one or more academicrisk factors. The unweighted n is 10,979. SOCIAL CAPITAL StudentTeacher Relations We used student characterizations of their teachers to create a measure of studentteacher relations. The composite includes six items, each of which comes from the NELS 10thgrade student survey (F1S7H, F1S7L, F1S7G, F1S66G, F1S7I, F1S7J). We used principal component factor analysis to obtain factor scores for each item. The composite is the weighted sum of the items. The measure is standardized (M = 0, SD = 1). Cronbach's alpha is .79. StudentTeacher Talks We constructed this dummycoded variable using items on the NELS first followup teacher survey (F1T1_5, F1T5_3), which asked whether a student (by name) talked with a teacher about schoolwork, academic decisions, or personal matters outside of class. Either one or two teachers in core subject areas (English, Mathematics, Science, or Social Studies) answered this item for each student. If at least one teacher said yes, we coded the value as 1 (1 = Yes); if no teacher said yes, we coded the value as 0 (0 = No). About half of the students (50.9%) solicited advice or talked with at least one of their 10thgrade teachers about school and personal matters. The unweighted n is 10,979. HIGHSCHOOL BEHAVIORS TenthGrade Academic Behaviors Like 8thgrade academic behaviors, this composite has four dimensions: attendance (F1S13, F1S10A, F1S10B), daily preparation for class (F1S40A, F1S40B, F1S40C), homework behaviors (F1HOMEWK), and a general rating (F1S67D), All items for the composite come from the NILS 10thgrade student survey. We used the same factor scores used for the 8^{th} grade composite to weight items. As with the 8thgrade composite, we squared the values to achieve a near normal distribution. The unweighted n is 10,979. Gronbach's alpha for the composite is .56. HIGHSCHOOL COMPLETION Dropout Status This dummycoded variable indicates whether students dropped out of high school (1 = Yes, 0 = No). We used a composite on the NELS 12thgrade student file to construct the variable (F2EVDOST), Eleven percent of the students in the sample dropped out of high school between the 10th and 12th grades. The unweighted n is 10,979. The research for this paper was supported by a grant from the National Institute on the Education of atRisk Students ($R306F60096), Office of Educational Research and Improvement, U.S. Department of Education. The institute's support does not imply its agreement with the opinions and statements expressed herein, and responsibility for any errors of fact or interpretation rests solely with the authors. Address all correspondence to Robert G. Croninger, University of Maryland, Department of Education Policy and Leadership, 2110 Benjamin Building, College Park, MD, 20742 or by email, rcl64@umail.umd.edu. REFERENCES Alexander, K. L., Entwisle, D. R., & Horsey, C. S. (1997). From first grade forward: Early foundations of high school dropout Sociology of Education, 70, 87107. Barrera, M., & Baca, L. M. (1990) Recipient reactions to social support: Contributions of enacted support, conflicted support, and network orientation Journal of Social and Personal Relationships, 7, 541551. Bourdieu, P. (1985). The forms of social capital In J. G. Richardson (Ed.) Handbook of theory and research for the Sociology of Education (pp. 241258) New York: Greenwood, Bryk, A. S., & Thum, Y. M. (1989). The effects of high school organization on dropping out: An exploratory investigation American Educational Research Journal 26, 353384. Carnegie Council on Adolescent Development [CCAD] (1993) A matter of time: Bisk and opportunity in nonschool hours. New York: Author. Catterall, J. S. (1998). Risk and resilience in student transition to high schools American Journal of Education, 106, 302333. Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences, 2nd ed. Hillsdale, NJ: Lawrence Erlbaum. Coleman, J. S. (1988). Social capital in the creation of human capital [Special issue] American Journal of Sociology 94, S95S120. Coleman, J. S. (1990). Foundations of social theory Cambridge, MA: Harvard University Press. Council of Great City Schools (GGCS) (1994) National urban education goals: 199293 indicators report. Washington, DC: Author. Croninger, Robert G. (1997). Does social capital influence students' academic development Implications for students at risk of educational failure Unpublished dissertation University of Michigan, Ann Arbor. Educational Testing Service (ETS) (1995) Dreams deferred: High school dropouts in the United States. Princeton, NJ: Author. Ekstrom, R. B., Goertz, M. E., Pollock, J. M., & Rock, D. A. (1986). Who drops out of high school and why? Findings from a national study Teachers College Record, 87, 356—373. Epstein, J. L. (1996). Federal education legislation: Implications for school, family, and community connections. In K. M. Borman, P. W. Gookson, Jr., A. R. Sadovnik, & J. Z. Spade (Eds.), Implementing educational reform: Sociological perspectives on educational policy (pp. 215244). Norwood, NJ: Ablex. Fine, M. (1986). Why urban adolescents drop into and out of public high schools. In G. Natriello (Ed.), School dropouts: Patterns and policies (pp. 89105). New York: Teachers College Press. Finn, J. D. (1989). Withdrawing from school Review of Educational Research, 59, 117143. Frase, M. (1989). Dropout rates in the United States: 1988. Washington, DC: U.S. Department of Education, Office of Educational Research and Improvement. Furstenberg, F. F., Jr., & Hughes, M. E. (1995). Social capital and successful development among at risk youth Journal of Marriage and the Family, 57, 580592. Grubb, W. N., & Lazerson, M. (1988). Broken promises How Americans fail their children. Chicago: University of Chicago Press. High School and Beyond: 19801992 (HSB) [Electronic datafile]. (1995). Washington, DC: National Center for Education Statistics [Producer and Distributor]. Hoffeitli, S. L., Boisjoly, J., & Duncan, G. J. (1998). Parent's extrafamilial resources and children's school attainment Sociology of Education, 71, 246268. Howe, K. R (1997), Understanding equal educational opportunity. Social justice, democracy, and schooling New York: Teachers College Press. Ingels, S. J., Dowd, K. L., Baldridge, J. D., Stipe, J. L., Bartot, V. H., & Frankel, M. R. (1994, September). Second followup: Student component data file user's manual Washington, DC: U.S. Department of Education, Office of Educational Research and Improvement. Ingels, S. J., Dowd, K. L., Stipe, J. L., Baldridge, J. D., Bartot, V. H., & Frankel, M. R. (1994 October). Second followup: Dropout component data file user's manual Washington, DC: U.S. Department of Education, Office of Educational Research and Improvement. Kaufman, P., McMillen, M., & Sweet, D. (1996) A comparison of high school dropout rates in 1982 and 1992 (NOES 96893). Washington, DC: U.S. Department of Education, Office of Educational Research and Improvement. LeCompte, M. D., Be Dworkin, A. G. (1991). Giving up on school Student dropouts and teacher burnouts. Newbury Park, CA: Corwin Press. Lee, V. E., & Burkam, D. T. (1992). Transferring high schools: An alternative to dropping out? American Journal of Education 100, 420453. Lee, V. E., Dedrick, R. E, & Smith, J. B. (1991) The effects of social organization of schools on teachers' efficacy and satisfaction Sociology of Education, 64, 190208. Lichter, D. T, Corawell, G. T, & Eggebeen, D. J. (1993). Harvesting human capital: Family structure and education among rural youth Rural Sociology, 58, 5375. Luthar, S. S., & Zigler, E. (1991). Vulnerability and competence: A review of research on resilience in children American Journal of Orthopsychiatry, 61, 622. MacLeod, J. (1987). Ain't no makin' it. Leveled aspirations in a lowincome neighborhood Boulder, CO: Westview Press. McDermott, R. P. (1977). Social relations as contexts for learning in school Harvard Educational Review, 47, 198213. McNeal, R. B. (1995). Extracurricular activities and high school dropouts Sociology of Education 68, 6280 National Center for Education Statistics [NCES] (1992) Transitions experienced by 1988 eighth graders. Washington, DC: U. S. Department of Education, Office of Educational Research and Improvement. National Educational Longitudinal Study of 1988 (NELS:88) [Electronic datafile]. (1995). Washington, DC: National Center for Education Statistics [Producer and Distributor]. National Research Council [NRC] (1993) Losing generations: Adolescents in highrisk settings Washington, DC: National Academy Press. Natriello, G. (Ed.) (1986). School dropouts: Patterns and policies. New York: Teachers College Press. Natriello, G., McDill, E. L., & Pallas, A, M. (1990). Schooling disadvantaged children. Racing against catastrophe. New York: Teachers College Press. Newmann, F. M. (1981) Reducing student alienation in high schools: Implications of theory Harvard Educational Review, 51, 546564. Pallas, A. (1986). The condition of education, 1986 Washington, DC: U.S. Department of Education, Office of Educational Research and Improvement, Pallas, A. (1989). Conceptual and measurement problems in the study of school dropouts In R, G. Corwin Be K. Namboodiri (Eds.), Research in sociology of education and socialization (Vol. 8., pp. 87116); Greenwich, CT: JA1 Press. Pallas, A. (1995). Schooling, achievement, and mobility In P. W. Cookson, Jr. & B. Schneider (Eds.), Transforming schools (pp. 1133) New York: Garland Portes, A. (1998). Social capital: Its origins and application in modern sociology Annual Review of Sociology, 24, 124. Raywid, M. A. (1995). Alternatives and marginal students In M. C. Wang & M. C. Reynolds (Eds.), Making a difference for students at risk: Trends and alternatives (pp. 119155). Thousand Oaks, CA: Corwin Press. Rosenthal, R., &: Rosnow, R. L. (1984). Essentials of behavioral research: Methods and data analysis. New York: McGraw Hill. Rumberger, R. W. (1987). High school dropouts: A review of issues and evidence Review of Educational Research, 57, 101112. Rumberger, R. W. (1995). Dropping out of middle school: A multilevel analysis of students and schools American Educational Research Journal, 32, 583625. Rumberger, R. W., & Thomas, S. L. (2000), The distribution of dropout and turnover rates among urban and suburban high schools Sociology of Education, 73, 3967. Schneider, B. (1996). School, parent, and community involvement: The Federal government invests in social capital. In K. M. Borman, P. W. Cookson, Jr., A. R. Sadovnik, & J. Z. Spade (Eds.), Implementing educational reform: Sociological perspectives on educational policy (pp. 193214). Norwood, NJ: Ablex. Schorr, L. B. (1988). Within our reach Breaking the cycle of disadvantage New York: Anchor Books. Sizer, T. R. (1984). Horace's compromise: The dilemma of the American high school Boston: Houghton Mifflin. Smith, M. H., Beaulieu, L. J., Sc Israel, G. D. (1992). Effects of human capital and social capital on dropping out of high school in the south Journal of Research in Rural Education, 8, 7587. StantonSalazar, R. D. (1997). A social capital framework for understanding the socialization of racial minority children and youth. Harvard Educational Review, 67, 140. StantonSalazar, R, D., & Dornbusch, S. M. (1995). Social capital and the reproduction of inequality: Information networks among Mexicanorigin high school students Sociology of Education, 68, 116135. Swadener, B. B., & Lubeck, S. (Eds.) (1995). Children and families "at promise." Deconstructing the discourse of risk. Albany, NY: SUNY. Wehlage, G. G., Rutter, R. A., Smith, G. A., Lesko, N., & Fernandez, R, R. (1989). Reducing the risk. Schools as communities of support London, UK: Falmer Press. Wellman, B. (1983). Network analysis: Some basic principles. In R. Collins (Ed.), Sociological theory (pp. 155200). San Francisco: JosseyBass. ROBERT G. CRONINGER is an assistant professor in the Department of Education Policy and Leadership at the University of Maryland. His research focuses on how education policies and practices affect the educational opportunities of students from different social and economic backgrounds. He teaches courses in education policy, the sociology of education, and quantitative research methods. VALERIE E. LEE is a professor of education at the University of Michigan. Her research interests focus on issues of educational equity. Her recent research has focused on (a) high schools divided into schoolswithinschools, (b) instructional effects on learning in Chicago elementary schools, and (c) early childhood educational contexts that are especially effective for children in poverty. She teaches courses in the sociology of education and advanced quantitative research methods. Endnotes^{i} We use the terms "social" and "academic disadvantage" broadly and interchangeably with "social" and "academic risk." ^{ii} We employed three data filters to select the analytic sample. First, we restricted the sample to those students who participated in the first three data collection waves of NELS (i.e., surveys were available for students at 8th, 10th, and 12th grade). Next, we restricted the sample to those students who did not have missing data on key variables, particularly the measures of social capital and dropout status as of the 1992 followup survey. A third filter limited the sample to students who attended a public, Catholic, or National Association of Independent School (NAIS) high school. This filter eliminated private high schools representative of small populations (e.g., high schools operated by religious organizations) and most high schools with unusual organizational structures (e.g., military or reservation schools). This sample is slightly more advantaged than the full 8th through 12thgrade student panel. The analytic sample has somewhat fewer students from socially disadvantaged backgrounds and fewer students who reported having academic trouble prior to entering high school. ^{iii} The NELS 10thgrade survey provides the first indication of students' relationships and interactions with high school teachers, at a point when early dropouts would have already left school. Students were asked about their interactions with middle or junior high school teachers in the base year survey, but these teachers would not have been accessible to students once they entered high school. ^{iv} On average, students with a socialrisk factor are twice as likely as students without a socialrisk factor to drop out of high school. Students from poor households are 2.6 times as likely to drop out as students from more advantaged households, whereas language minority students are 1.5 times as likely to drop out, disadvantaged minority students and students from singleparent households are 1.7 times as likely, and students with mothers who failed to complete high school are 2.7 times as likely to drop out. ^{v} We performed a variety of analyses to determine empirically the implications of different measurement models for both social and academic risk status. The majority of students possess none of these characteristics, so neither social nor academic risk status is suitable as a continuous variable. Students who possess multiple factors are at the greatest risk of failure, but our analysis indicated that the possession of any single social or academic risk factor significantly enhances a student's risk of low achievement, negative academic behaviors, and dropping out, even when controlling for additional factors. (See Croninger, 1997, for the full results of these analyses.) ^{vi} On average, students with an academic risk factor are three times as likely as students without an academic risk factor to drop out of high school. Students who have been held back are 4,9 times as likely to drop out as students who have not been held back, whereas student with low educational expectations and two or more disciplinary referrals to the office are 3.2 times as likely to drop out, students with low gradepoint averages are 2.7 times as likely, and students whose parents received two or more warnings are 3.0 times as likely to drop out. ^{vii} We did not examine the cumulative effect of academic risk in this study. Although the factors that we use to tap academic risk have been used in other studies, our factors are not as theoretically grounded as those that we use to tap social risk. Moreover, the introduction of another set of incrementally coded measures of risk introduced a level of complexity into the analysis that we determined to be unwarranted. Thus, we limited our analysis to examining the cumulative effects of social risk for students with and without any indication of academic risk when they entered high school. ^{viii} This correlation is surprisingly small, though it is undoubtedly due to the relatively weak psychometric properties of the measure of actual interactions between specific students and teachers. Unfortunately, NELS data do not provide information about the specific content of each of these interactions or about how often students seek guidance and assistance from teachers. Moreover, even if students do not seek guidance from teachers surveyed by NELS, it is possible that they solicit help from other teachers or adults at school, such as coaches, guidance counselors, or support staff, Although this variable is a "blunt" measure of students' access to social capital, our analysis shows it captures potentially important effects. ^{ix} We decided not to use multilevel modeling for this study, although the newest version of HLM permits the modeling of binary outcomes. Our focus here is on effects manifested through individual interactions and relationships with students, not on effects attributable to school organization and structure. The use of HLM would also limit the complexity of our models at the individual level of analysis. ^{x} We do not use students' high school behaviors in our model to test interactions with risk status because we consider these behaviors to be potentially the effect of students' access to social capital. Including 10thgrade behaviors in model 4, obscures our estimates of the effects of teacherbased social capital on the probability that students fail to finish high school. All statistically significant interactions reported in model 4 are also statistically significant, though smaller, when 10thgrade behavior is included. ^{xi} We report results for continuous measures in standard deviation units. Rosenthal and Rosnow (1984) provide guidelines for characterizing the substantive meaning of different magnitudes of effects based on standard deviation units, or, in this case, mean differences between students who drop out and complete high school. Less than .10 SD is a very small or trivial effect, .10 to .30 SD is a small but nontrivial effect, .30 to .50 SD is a moderate effect, and effects greater than .50 SD are large. ^{xii} Because continuous variables are coded as zscores on the entire population of students, the effect of academic risk in model 4 is for students with average 8thgrade behaviors, 8thgrade achievement, studentteacher relations, and 10thgrade behaviors. The use of a common reference point for the constants facilitates the comparison of effects across samples. 13 Recall that continuous variables are centered on the grand mean for the entire population of students. Gender is centered on the respective population mean for students with and without one or more academic risk factors. This facilitates the comparison of effects across models.


