Subscribe Today
Home Articles Reader Opinion Editorial Book Reviews Discussion Writers Guide About TCRecord
transparent 13

The Dropout Process in Life Course Perspective: Early Risk Factors at Home and School

by Karl L. Alexander, Doris R. Entwisle & Nader Kabbani - 2001

From a life course perspective, high school dropout culminates a long-term process of disengagement from school. The present paper uses data from a representative panel of Baltimore school children to describe this unfolding process. Over 40% of the study group left school at some point without a degree, but this high overall rate of dropout masks large differences across sociodemographic lines as well as differences involving academic, parental, and personal resources. A sociodemographic profile of dropout for the study group shows how dropout rates vary across different configurations of background risk factors including family socioeconomic status (SES), family type, and family stress level. Dropout risk factors and resources in support of children's schooling then are examined at four schooling benchmarks: the 1st grade, the rest of elementary school (years 25), the middle school (years 68), and year 9 (the 1st year of high school for those promoted each year). Academic, parental, and personal resources condition dropout prospects at each time point, with resources measured early in children's schooling forecasting dropout almost as well as those from later in children's schooling. Additionally, evidence is presented that resources add on to one another in moderating dropout risk, including risk associated with family SES. These patterns are discussed in terms of a life course view of the dropout process.

From a life course perspective, high school dropout culminates a long-term process of disengagement from school. The present paper uses data from a representative panel of Baltimore school children to describe this unfolding process. Over 40% of the study group left school at some point without a degree, but this high overall rate of dropout masks large differences across sociodemographic lines as well as differences involving academic, parental, and personal resources. A sociodemographic profile of dropout for the study group shows how dropout rates vary across different configurations of background risk factors including family socioeconomic status (SES), family type, and family stress level. Dropout risk factors and resources in support of children's schooling then are examined at four schooling benchmarks: the 1st grade, the rest of elementary school (years 2-5), the middle school (years 6-8), and year 9 (the 1st year of high school for those promoted each year). Academic, parental, and personal resources condition dropout prospects at each time point, with resources measured early in children's schooling forecasting dropout almost as well as those from later in children's schooling. Additionally, evidence is presented that resources add on to one another in moderating dropout risk, including risk associated with family SES. These patterns are discussed in terms of a life course view of the dropout process.


The economic engine that drives the U.S. economy generously rewards highly skilled, highly credentialed labor; but it can be as hard on those who don't fit in as it is kind to those who do. Under such circumstances, the social and personal costs of high school dropout can be high. In the United States, about half of all welfare recipients1 and half the prison population lack high school degrees (Educational Testing Service, 1995; National Research Council, 1993); and dropouts' earnings lag far behind those of degree holders, even when they work full time (Mishel & Bernstein, 1994; U.S. Department of Education, 1999) and even after earnings are adjusted for differences in school achievement and other factors that distinguish drop-outs from graduates (McDill, Natriello & Pallas, 1986).

Between 1972 and 1983 high school completion among 18-24 year-olds averaged around 83.5%, with annual figures varying by less than 1%. Since then, the average has been around 85.5%, again varying little year-to-year. Progress toward the national goal of 90% high school completion thus has been painfully slow, with no progress registered during the decade of the 1990s (NEGP, 1999: 17). And the distance yet to be traveled is greater for some segments of the population than for others. In central cities, for example, the rate of high school completion for 16-24 year-olds (a slightly different frame of reference) averaged 84.7% from 1990 through 1996 versus 90.3% in the suburbs and 88.5% outside metropolitan areas.2 Such census estimates are useful for comparative purposes, but the problem is greater still in high poverty cities where dropout rates typically hover in the vicinity of 30%-50% (Council of Great City Schools, 1994; Education Week, 1998).

Baltimore, Maryland, is one of those cities. Hard hit by the deindustrialization and suburbanization that has decimated the older cities of the northeast and midwest (e.g. Wilson, 1987; 1996), Baltimore has one of the highest dropout rates in the country (Annie E. Casey Foundation, 1997; Bomster, 1992). According to Census figures for 1989, over a fourth of Baltimoreans age 25-29 were out of school and without degrees (this includes GEDsU.S. Bureau of the Census, 1992). In raw figures, this amounts to over 17,000 young adults trying to establish a toehold in an economy that has little place for the degreeless. The experience of the Baltimore study group discussed in this paper parallels that city-wide: 42% of the panel left school without degrees, which drops to 24% when subsequent degree completion and GED certification are factored in.

Baltimore's high level of high school noncompletion is both symptom and cause of a city in distress. In 1990, its childhood poverty rate was 32.5% for children 18 and under (39.1% among African-Americans) as against 15.2% for the nation's 200 largest cities (Children's Defense Fund, 1992) That same year Baltimore ranked 11th among the nation's 100 largest cities in terms of poverty concentration (Kasarda, 1993: Table A.1). In 1997 it placed second highest among U.S. cities in the percentage of births to teen moms, it ranked near the top in percentage of low birth weight babies and infant mortality, and its juvenile arrest rate led the nation (Annie E. Casey Foundation, 1997). These conditions put many of Baltimore's children "at risk" of school failure in the near term and of a lifetime of struggle over the longer term.

Baltimore's problems are not all that different from those of other high poverty cities, but the situation there frames the life experience of the children of the Beginning School Study (BSS) and so is the backdrop to the research reported in this paper. A long-term panel study, the BSS has been monitoring the educational and life progress of a representative sample of Baltimore children since fall 1982, at which time members of the study group were beginning 1st grade in 20 of the city's public schools. The BSS is ongoing still, which means the project spans virtually the entirety of the panel's schooling. An appendix describes the project's research design and the measures used in the present analyses.

There is a pressing need to understand what distinguishes youth in our large cities who leave school before graduating from those who stay in schoolto understand what factors place them at risk of dropping out and to identify resources potentially within their reach that might mitigate this risk. The BSS has been looking on as the young people in our study group came to maturity and moved through the educational system. We have been witness to their successes and to their failures. Both contain valuable lessons. The project's long time frame and broad purview enable us to explore, from a life course perspective, sources of risk and resilience in these young people's lives. Our perspective on the dropout dynamic is developed in the next section.


Perhaps 16- and 17-year-olds are mature enough to do a proper cost-accounting of the pros and cons of the decision to leave schoolthe case at least is arguable.3 But before they leave for good, many fade out through chronic truancy; and before then many disengage from academics in other wayse.g., by getting in to trouble at school or putting little effort into their work. The roots of dropout extend deep and broad. The situation of the 16-year-old suspended three times for fighting, held back twice, reading at a 5th-grade level, and coming to school two days out of five may seem different from that of the 8-year-old who is unruly in class, reading one year behind grade level, and absent 4-5 days per month, but is it really? Not if the 8-year-old is in the process of becoming the 16-year-old. Future dropouts in the BSS panel averaged 16.4 absences as 1st graders, future graduates averaged 10.2 absences (a highly significant difference). A 6 day difference in absences at the start of children's schooling might seem a small matter, but not if it foreshadows even larger differences later: over their middle school years (years 6-8), dropouts-to-be registered 27.6 absences annually versus 11.8 for future graduates; in their 9th year of schoolthe 1st year of high school for those promoted each yearthe two groups averaged 46.8 and 13.5 absences, (all figures are from Baltimore City Public School (BCPS) records). With 47 recorded absences, these future dropouts were missing about one day out of every four"fade out" segueing to dropout.

From a life course perspective (Elder, 1994; Pallas, 1993), dropout is not so much an "event" as a "process," a process of progressive academic disengagement that often traces back to children's earliest experiences at school (e.g., Brooks-Gunn, Guo, & Furstenberg, 1993; Ensminger & Slusarcick, 1992; Gamier, Stein, & Jacobs, 1997). The habits of conduct and thought that prompt some children to leave school and others to stay take shape in a social matrix of overlapping spheres of influence that frame their academic development (Epstein, 1990; 1992; Jessor, 1993). These spheres of influence can be placed in the "ecological contexts" (Bronfenbrenner, 1979) of school, home, and community; and at their center is the developing child. To a substantial degree, influences issuing from these contexts "work through" young people's personal agency.4 At the most obvious level, deciding whether to remain in school or leave ultimately is a personal matter but of course the issue runs much deeper. How children come to think of themselves in the student role (attitude) and how they enact that role (behavior) are key mechanisms through which the press of the broader institutional context gets routed.

J. D. Finn's (1989) two social-psychological perspectives on the dropout dynamic illustrate how these multiple strands of development intersect over the life course, in this instance the life course of early schooling. Under Finn's "frustration-self-esteem" model, a record of poor performance causes children to question their competence and weakens their attachment to school. Dropout under such circumstances is a means of escape from an environment that is psychologically punishing. Finn's second dynamic, his "participation-identification" model, stresses positive experiences that encourage a sense of belonging (e.g., participation in the extracurriculum; recognition for accomplishment) and so strengthen attachment to school (e.g., Wehlage, Rutter, Smith, Lesko & Fernandez, 1989).

Finn discusses these two perspectives in parallel, but at a higher level of abstraction their distinctiveness blurs. Affective detachment from school is the immediate impetus to dropout, but whether children's school attachment is strong or weak develops over time as a result of their cumulative experience there: Are they fitting in comfortably and realizing success or are they struggling and not measuring up academically?

Experiences at home also influence children's academic commitment. Children respond positively when their parents are optimistic and hold high expectations for their school performance (Alexander, Entwisle, & Bedinger, 1994; Ensminger & Slusarcick, 1992; Entwisle, Alexander, &; Olson, 1997; Seginer, 1983). They achieve at a higher level than would be expected otherwise, and the positive feedback this brings about encourages greater academic commitment. This, in turn, promotes higher levels of academic achievement; and certainly marks and test scores predict dropout (Ensminger & Slusarcick, 1992). But parental support influences the dropout process in other ways also. By setting rules, helping with homework, monitoring school progress and the like (Astone &: McLanahan, 1991; Ensminger & Slusarcick, 1992; Rumberger, 1995), the parenting strategies adopted by low-income parents can insulate their children by promoting positive school adaptations and/or preventing anti-school adaptations (e.g., Elder, Eccles, Ardelt, & Lord, 1995).

Parental behaviors and support, successes and failures in the student role, children's emergent sense of an academic self-identity, and children's school engagement all are integral to the dropout process; but "process" connotes a "dynamic" and to understand how youth arrive at the brink of dropout in the upper grades requires a long time frame. Almost 30 years ago, Jerald Bachman (1972) observed that dropout". . . is the end result or symptom of other problems originating much earlier in life" (p. 27). Today the same point probably would be couched in life course language and imagery, but the essential insight would be the same: What happens in high school often is rooted in formative experiences that predate high school.

Most children, including so-called "at risk" youth, begin school enthusiastic, optimistic, and eager to learn. Unfortunately, for many this frame of mind is short-lived: Enjoyment of school, compliance with school routines, and academic self-image all tend to spiral downward the longer children are in school (Anderman & Maehr, 1994; Stipek & Maclver, 1989). Why some children follow this downward path and others manage to avoid it still is not fully understood, but the path itself we know can span many years. Erisminger and Slusarcick (1992), for example, report elevated dropout rates among children who were rated as highly aggressive by their 1st-grade teachers. Weak school performance, grade retention, and stressful family change in the early grades likewise forecast later dropout (e.g., Barrington & Hendricks, 1989; Brooks-Gunn, Guo, & Furstenberg, 1993; Cairns & Cairns, 1994; Ensminger & Slusarcick, 1992; Gamier, Stein, & Jacobs, 1997; Haveman & Wolfe, 1994; Roderick, 1994; Rumberger, 1995); and 1st grade measures of many of these same factors have proved predictive of dropout in the BSS as well (e.g., Alexander, Entwisle, & Horsey, 1997; Entwisle et al., 1997). The present paper adds to this view of dropout as a long-term developmental process by evaluating relevant risks and resources in rough parallel from 1st grade into high school.


The conceptual backdrop to the present analysis is depicted in Figure 1. Sociodemographic factors, shown in the left-most box, are treated as exogenous-family socioeconomic level (SES), race/ethnicity (African American; White), sex, family stress in 1st grade, mother's age and employment, and family type. These family conditions and personal characteristics situate the family and its members in socioeconomic space; but because background risk factors tend to be moderately to highly intercorrelated, the redundancy built into simple bivariate comparisons can cloud interpretation. Children in single parent households and those in lower SES households, for example, both drop out at higher than expected rates; but if they are substantially the same children in the two instances it is hard to know which aspect of their family context is more relevant. Controlled comparisons cannot resolve all such ambiguities, but it nonetheless is useful to know which differences prove robust and which fade when risk factors overlap.


School experiences, parental resources, and personal resources (as reflected in engagement behaviors and engagement attitudes) are introduced in the middle panels of Figure 1. The particular risk/resource mix we will be evaluating is not all that distinctive-at the level of issues, it is hard to imagine any framework that presumes broad coverage of the family/school context of dropout not allowing space for parents' attitudes, children's attitudes and behaviors, etc. (e.g., Astone & McLanahan, 1991; Brooks-Gunn et al., 1993; Chen, Kaufman, & Frase, 1997; Gamier et al., 1997; Rumberger, 1995)but the schedule of BSS data gathering we believe is distinctive, as is the opportunity afforded by that time line to evaluate conceptually relevant resources in parallel across the span of children's early schooling.

Members of the BSS panel were individually interviewed in their 1st, 2nd, 4th, 6th, 7th, 8th, and 9th school years (often twice annually); parents and teachers were queried too, and on much the same schedule. This schedule allows us to investigate the developmental backdrop to dropout at home and at school as a process that plays out over time.

Accordingly, the central objectives of the present analysis are to chart the time line and routing of early social and personal influences that condition dropout prospects guided by the imagery in Figure 1. Are there critical periods or turning points in the developmental course of dropout? Is there reason to think early influences are distinctively important? Lagged effects of early experiences and resources that add on to those of later experiences and resources would suggest so. And what of later experiences and resources? Do they merely transmit earlier influences or do the same kinds of resources and experiences become more important with age, as children become more self-aware and self-directed?

A related concern is whether personal and familial resources and experiences account for differences in dropout risk associated with the socio-demographic profile of dropout. Sixty percent of BSS children in families classified as lower SES (half the sample) leave school without degrees as against 15% of those of higher SES (top quarter of the sample). What lies behind this huge disparity in dropout risk across social lines? Is it because lower SES parents don't provide the kind of encouragement that keeps their children focused, because their children sour on school, or because of academic difficulties? These are all plausible candidates, but is the accounting they provide partial or whole? And what of their timing? Do they come into play early or late in children's schooling or does their importance increase with children's, time in school? These and related questions can be informed by monitoring the behavior of sociodemographic factors in relation to dropout when risk/resource measures from different stages of schooling are controlled.


The childrennow grown upwho have been participating in the BSS these many years were typical urban youngsters whose schooling overlapped the decades of the 1980s and the 1990s, a difficult period for many of our nation's large cities and their public school systems. Reflecting this, their dropout rate was high overall; and they began leaving school early, with many off-time when they left.

Questions about dropout were first posed of BSS students in the spring of 1991 (9th grade for students not held back) and repeated annually thereafter through the fall of 1999 (5 years after the group's expected high school graduation in spring 1994). As used here, "dropout" means leaving school at least once for an extended period of time prior to graduation for reasons other than illness (e.g., Morrow, 1986). Weaving together sources, dropout status could be determined for 92.3% of the original cohort (729 of 790). According to these self-reports, 41.6% of the group (N = 303) dropped out at some pointa figure in line with other estimates for Baltimore (Bomster, 1992).

The timing of dropout in the BSS is organized in Table 1 according to year and grade of dropout (based on N = 301; the timing of dropout could not be determined in two instances). The two distributions differ and how they intersect is revealing.

About 35% of BSS dropouts, amounting to 14% of the entire cohort, left school having attained less than a 10th-grade education. This is the standard in the literature for identifying "early leavers" (e.g., Rumberger, 1995; Schneider, Stevenson, & Link, 1994), and experience in the BSS aligns with national estimates (in this instance for 16-24 year-olds; U.S. Dept. of Education, 1997, p. 15). However, the mismatch in Table 1 between "grade attained at dropout" and "year left school" suggests that most of these early leavers repeated one or more grades before withdrawing: "During Year 11" (N = 95) is the modal year of dropout; 10th grade (N = 88) is the modal grade attained by dropouts. In fact, of the 95 youngsters who left during year 11, more left as 9th (N = 36) and as 10th (N = 37) graders than as "on-time" 11th graders (N = 22).

Sixteen is the earliest age at which students may leave school legally in Maryland. This age also corresponds with the timing of the transition from middle school to high school for youngsters who are a year or two behind. For these students, the stresses of being off-time relative to grade level expectation add to the pressures surrounding school transitions that challenge all children (e.g., Eccles, Lord, & Midgley, 1991). This confluence of forces apparently drives up dropout rates (e.g., Roderick 1995), especially when academic performance and school attachment are marginal, as is true for many repeaters (e.g., Alexander, Entwisle, & Dauber, 1994; Alexander, Entwisle, & Kabbani, 1999).


However, the hazard of dropping out remains high around the time of the next school transition alsohigh school graduation. It might seem odd to see so many students leave as 12th graders (N = 41, about 14% of the total), with the finish line seemingly so near; but this assumes the line is fixed, which may not be the case. Eighteen of the 41 12th-grade dropouts were in their 12th year of school at the time; others had been in school IS, 14, even 15 years, and we know from talking with them that many saw ahead not caps and gowns, but yet another year in 12th gradesurely a discouraging prospect. In fact, about 16% of BSS dropouts (N = 49) left school after completing 12 years.

These late dropouts demonstrate impressive commitment to school; but having already repeated at least one grade, they apparently decided it was time to try another course. Whether dropouts later return is a separate issue. The BSS panel's record of reenrollments is incomplete at this writing, but through Fall '99 (age 22/23) 9.2% of dropouts had returned to school and completed regular high school degrees. Another 30.4% obtained the GED, increasing the group's high school completion to 39.6%.5 This is close to the 44% return figure registered by dropouts nationally (U.S. Department of Education, 1998, p. 10) despite very different baseline levels of dropout in the two instances (41.6% in the BSS versus 21% nationally). Factoring in GED's yields a 24% level of high school noncompletion for the panel as young adults, very close to the level indicated in Census data for the city as a whole (U.S. Bureau of the Census, 1992).

Dropout incurs costs even when dropouts later achieve high school certification (e.g., Cameron & Heckman, 1993; Murnane, Willett, & Boudett, 1995), and clearly it would be preferable for more young people to realize success at school the first time through. One step toward this end is to understand better the circumstances that put so many low income urban youth at elevated risk of dropout. That is the agenda for the remainder of this paper. The next section describes risk factors at home and at school that distinguish dropouts (all dropouts, regardless of whether they later returned to school) from those who graduated without interruption.



The sociodemographic profile of BSS dropouts looks much like that seen nationally (Natriello, McDill, & Pallas, 1990; Rumberger, 1983; U.S. Department of Education, 1997). Socioeconomic status, race, and gender are key dimensions of social stratification that structure life opportunity in many ways including the likelihood of dropout (Table 2). African Americans are the only minority group with a large presence in Baltimore's public schools, so minority-majority comparisons in the BSS are limited to African Americans against Whites.6

Sixty percent of lower SES youth in the BSS leave school without degrees, a level four times that of upper SES youth. This is the largest comparative differential by far in Table 2; and with family SES skewed low in the BSS, the elevated risk of dropout associated with disadvantaged family standing applies to about half the sample.

Other family conditions weigh in too, however. The ones covered in Table 2 reflect home conditions during 1st grade. The transition from "home child" to "school child" often is a difficult adjustment for children, and how well it is managed has implications for the course of their later schooling (Belsky & MacKinnon, 1994; Entwisle & Alexander, 1989; 1993). A smooth transition generally augers well for the future, but the likelihood of having a smooth transition experience depends, among other things, on the psychological and material conditions in support of children's schooling at home. Here we see that dropout risk is elevated among 1st graders born to teenage mothers, among those in single parent households, and among those in homes characterized by higher than average levels of stressful change (e.g., death, divorce, family moves). These patterns accord with the literature generally, but are usually documented among older children (e.g., Gamier et al, 1997; Haveman & Wolfe, 1994; Rumberger, 1995), The dropout risks associated with these family conditions are of roughly the same magnitude but much lower than the "SES risk."


Also as found nationally (e.g., Haveman & Wolfe, 1994; Haveman, Wolfe, & Spaulding, 1991), dropout is less common among BSS children of working mothers, but in this instance the overall comparison obscures variation across subgroups. At the lowest SES level, 47% of children drop out when the mother works outside the home versus 65% when she does not, whereas at the middle SES level the figures are 22% versus 38% (both differences significant at the .05 level). The pattern reverses among upper SES youth, however, who register 16% dropout when the mother is employed versus 14% when she is not (the difference is not significant). This is consistent with broader studies of how maternal employment affects children's development (Zaslow & Hayes, 1986), and to see the difference across SES lines attenuate with diminishing financial pressure supports Haveman and Wolfe's interpretation that the income generated by maternal employment overrides consequences (presumably adverse) of a working mother's need to relinquish part of her role as caregiver. The BSS adds that a mother's income matters the most where the need is greatest, which seems plausible. These more detailed comparisons are not shown.

Boys are more prone to leave school without degrees than are girls among both African Americans and Whites, whereas African American and White dropout rates are close for both sexes (not significantly different). The Whites who attend public schools in Baltimore (and presumably in other places like Baltimore) are mostly low income; and the comparison in Table 2 suggests that their graduation prospects are not much different from those of their Black schoolmates who also are mostly low income. However, Black and White dropout rates also have been converging nationally (e.g., Day & Curry, 1998), so the parity seen here could well be characteristic of disadvantaged urban populations generally (e.g., McDonald & LaVeist, 1999).

Table 2's checklist approach to cataloging risk factors is useful as a point of departure; but in the reality of children's experience risks come "bundled," not separate, and their impact depends on exactly how they are configured. Multiple risk exposure almost certainly magnifies difficulties (e.g., Catterall, 1998; Chen, Kaufman, & Frase, 1997), whereas exemption from risk in one area might confer protection against the consequences of risk exposure in other areas (an "inoculation" effect).

The consequences of multiple risks aligning and crosscutting are explored in Chart 1, which maps contingent dropout risk incrementally at the intersection of family type, family stress conditions, race, and sexfour risk dimensions simultaneously, five if family socioeconomic level is counted (Chart 1 focuses on lower SES youth, 60% of whom leave school without degrees). The top-most branch distinguishes lower SES 1st graders in one-parent households from those in two-parent households. Then, for the two family types separately, the next branch separates children in households experiencing high versus low levels of stressful change, again referenced to conditions in 1st grade. With this divide, we can examine whether level of stressful change modulates risk for lower SES children in single parent and two parent households. Does family stability, for example, cushion children against the risk otherwise associated with residence in a lower SES, single parent household?


The last two branches of Chart 1 divide risk groups by race/ethnicity and sex, showing how potentially alterable conditions of family life affect graduation prospects across ascriptive lines. Ns are small at this level of detail, however.

Fifty-five percent of children in lower SES two-parent families drop out against 65% of children in lower SES one-parent families; and when these two-parent households are relatively free of stressful change, the risk falls to 50%. In comparison, dropout stands at 69% in lower SES, single-parent households characterized by (relatively) high levels of stressful change. This 19 percentage point spread (69% versus 50%) across family conditions for children in lower SES households exceeds the overall spread between middle SES and higher SES families (Table 2), and other third and fourth order differences in Chart 1 are larger still. For example, the 41% dropout figure for lower SES African Americans in stable two-parent households (a small group) approaches the BSS percentage dropout overall (40%), whereas the experience of boys in such households (an even smaller group) is more favorable still. At the other extreme, dropout registers over 80% among the handful of Whites who as 1st graders were in lower SES, highly stressed, single-parent households. Sample sizes are small in these instances and so probably not too much should be made of the exact figures, but the general pattern is informative nonetheless: Multiple problem conditions in the family add onto the burdens associated with low socioeconomic standing, whereas favorable conditions in lower SES families cushion those burdens somewhat.

To recapitulate the main points of this sectionfor children in places like Baltimore, high school dropout rates remain unacceptably high. Indeed, even the most favorably situated of Baltimore's school children fall short of the national target of 90% high school completion (National Educational Goals Panel, 1999).7 Still, not all young people are at similar risk of dropout, and the sociodemographic profile of dropout is useful in a diagnostic sense. Being male, for example, increases dropout risk but being African American apparently no longer does, at least not relative to Whites who attend Baltimore public schools. Although family socioeconomic status looms especially large in these comparisons, at least three other family factors also elevate risk: residing in a single-parent household, having a teenage mother, and being in a family with high levels of stressful change.

Such background risks often are treated separately in the literature (e.g., McLanahan & Sandefur, 1994; Zill, 1996); but their impact in fact is determined by how they intersect in children's lives, and their intersection varies by SES level. Half the BSS youngsters at the high end of the SES distribution registered none of these additional risks and not a single higher SES child had to contend with all three. This contrasts sharply with the situation in lower SES households, where all three additional risks were present in as many families as had none of them (16% in both instances).

That risk factors for dropout are more prevalent in lower SES households than in higher SES ones compounds the risk associated with low SES (Chart 1), but what of lower SES youth who overcome the odds and stay in school? Are their family circumstances also different? In some respects yes: Dropout risk moderates when such youngsters have two parents at home (in 1st grade), when their mothers defer childbearing until their 20's, and in the absence of stressful family change (again in 1st grade). Such patterns may be a source of leverage. Minimizing disruptive change in single parent households, for example, seemingly offsets some of the risk associated with that family type without changing family structure; and since 54% of lower SES households in the BSS have just one parent versus 28% of higher SES ones, any such gains would accrue disproportionately to lower SES children.

This section has profiled dropout in terms of objective family conditions and sociodemographic factors. These issues are compelling because they constitute the social backdrop to dropout; but although these background factors are associated with elevated risk, with the exception possibly of stressful family change, they themselves are not proximal causes of dropout. Rather, at the person-level the impetus traces to children's experiences in school and in the family that have bearing on their attachment to school. Risk factors suggested by this perspective are taken up in the next sections.


This section examines academic risk factors (e.g., Catterall, 19918). According to J. D. Finn (1989), many children respond to discomforting academic feedback by disengaging from school. Report cards signal competence judgments explicitly; educational track placements do so more subtly. Repeating a grade, receiving Special Education services, taking low level courses in middle school and high school, and pursuing a noncollege program in high school all imply low standing in the school's academic hierarchy and are potentially stigmatizing (Alexander, Entwisle & Legters, 1998; Entwisle & Alexander, 1993). These facets of school experience, along with children's ranking on standardized testsa less public criterion, but one no less relevant to producing discomforting academic feedbackare examined in relation to dropout at four time points: 1st grade, the rest of the elementary years (years 2-5), the middle school years (years 6-8), and year 9the 1st year of high school for those promoted regularly.

These educational benchmarks are the academic backdrop to dropout in the sense that few children leave school before 9 years (see Table 1). First grade is of particular interest to us as a developmental milestone because it defines the transition to full-time schooling, but these 9 years together span three educational transitionsinto 1st grade, from elementary to middle school, and into high school. After that point, attention would shift to a more expansive set of near-term correlates of dropout, including those that can be thought of as framing the transition to adulthood, e.g., the desire to work, home responsibilities, and pregnancy (e.g., Entwisle, Alexander, & Olson, 2000; Pallas, 1987; Rumberger, 1987; Wagenaar, 1987).

Test scores, marks, grade retention, and receipt of Special Education services, all obtained from school records, are monitored period-by-period. For test scores and marks, the values reported for years 2-5 and years 6-8 are the average of annual measures; for retention and Special Education, the criterion is any retention or any receipt of services during the years at issue. Course-level placements in the main academic subjects are referenced to the 1st year of middle school (6th grade, which could be year 7 or 8 for repeaters), when such curricular distinctions are first introduced, and then again in the 9th grade. These data too are from school records. For students attending city-wide magnet schools, initial high school program (college preparatory versus others) is deduced from the character of the school; for others, self-reports are used.8

In Table 3 test scores and report card marks from 1st grade forecast dropout risk with considerable accuracy, a result seen in other data also (e.g., Ensminger & Slusarcick, 1992). For example, comparing children whose 1st grade mark averages are in the "A-B" range against those with marks in the "D-.F" range, the percentage dropout differs by 41 points. By year 9, the difference is 52 points; so discrimination increases over the intervening years by about a quarter (i.e., 52/41 = 1.27), but from a high initial base. Also in accordance with the literature generally (see Jimerson, Anderson, & Whipple, 2000, for overview) repeating a grade is associated with elevated dropout risk. Although being on time in school hardly guarantees success for this mainly low income group, with dropout averaging 71% among repeaters across the four benchmark periods covered in Table 3 (including 1st grade), there can be little doubt that grade retention takes many youth off the path to graduation. Indeed, among multiple repeaters (36% of all repeaters in the BSS), dropout approaches a certainty: 80% overall; 94% for those retained in elementary and middle school (not shown in tables).

In contrast, the organizational fast track to graduation apparently is to take high level, college oriented courses as soon as the curriculum permits. In the middle grades, the BCPS distinguishes among Remedial, Regular, Honors, and Advanced Academic courses in the main academic subjects (our coding combines the last two). High school offers even more subject matter specializationvocational or business courses versus academic, for example. Both kinds of placements, we see, tie in with dropout: Children who take advanced courses in middle school and high school and those who start high school in a college-bound program have relatively low dropout rates, generally in the vicinity of youth with high test scores and marks.


Academic performance and educational track placements thus constitute critical filters for dropout, but do these contingencies associated with school program placement and academic differences in dropout risk also account for "family risk" as structurally defined? School performance generally follows social lines (e.g, White, 1982), so the question is not so much "whether" but "how much"? The relevant comparisons in Table 4 show that higher test scores and marks are associated with lower dropout rates for children at all SES levels. Good school performance in this sense is a rising tide that carries all ships, but does it carry them all to the same level? Does good school performance override social disadvantage?


From Table 4 it appears that social risk is only partly a matter of academic risk. Consider the pattern for 1st grade achievement scores. Overall, there is a 45% difference in dropout risk across the SES extremes (Table 2). In Table 4, for children with scores in the bottom third of the test distribution the difference is 48% (i.e., 69%-21%, amounting to a 3.2-fold difference); in the middle range of the distribution it is 42% (a 3.2-fold difference); and in the top third of the distribution it is 29% (a 3.6-fold difference). The absolute risk associated with SES level thus moderates when children test well in 1st grade, but even with test scores high there is more than a three-fold difference in dropout levels across the family SES extremes; and the picture is much the same for 1st grade marks and middle school test scores: Good school performance offsets SES disadvantage somewhat, but academically successful lower SES youth remain highly vulnerable.

Grade retention in relation to dropout mirrors this pattern in that repeaters, even those held back as early as 1st grade, are much more likely to drop out than are promoted youth. Seventy-five percent of the higher SES children held back in middle school drop outthe highest level by far for this relatively privileged groupas against just 11% of the never retained. But in absolute terms, the 85%-91% levels among repeaters from middle SES and lower SES households are higher still. Grade retention in both elementary and middle school thus elevates dropout risk "across the board," although the comparisons in Table 4 suggest the relationship strengthens for retentions closer to the legal age of dropout.


This section examines personal resources that bear on dropoutparents' psychological support for their children's schooling and children's own attitudes and behaviors. Children who work hard, play by the rules (of the school), are confident of their abilities, and hold personal goals that align with institutional ones tend to do better in school (Alexander, Entwisle, &: Dauber, 1993; Farkas, 1996; Keogh, 1986). And it helps if their parents also value school success, expect their children to do well at their schoolwork, and offer encouragement toward that end (Alexander et al., 1994; Entwisle, Alexander, & Olson, 1997; Seginer, 1983). This is not to say the dropout problem reduces to a dearth of psychosocial capitalthe social patterning of dropout and overall rates hovering in the vicinity of 30%-50% in many high poverty cities contradict this kind of oversimplification. Still, some personal and parental qualities promote children's success at school more so than others, and knowing which of these qualities matter the most, exactly when they begin to matter, and whether they compensate for a shortfall of more material resources may give leverage for offsetting structurally induced risk. Toward this end, Table 5 summarizes dropout rates in terms of parents' attitudes vis-a-vis their children's schooling, children's self-attitudes, and children's school engagement, as reflected in both attitude and behavior (e.g., Mosher & MacGowan 1985).

As with academic resources, here too each area is measured in rough parallel from 1st grade into children's 9th year of school (measurement details are provided in the appendix). On various occasions over the years, parents were queried about their children's ability to do schoolwork, how far they expected their children to go through school, and their mark expectations for upcoming report cards. Standardized "parental attitude" scales were constructed from these items year-by-year. The 1st grade and year 9 measures are used individually; the others are averaged across years to derive measures for years 2-5 and years 6-8. In Table 5, scale scores above the sample-wide average are classified as "favorable"; those below the average as "unfavorable." Though a crude distinction, this simple "high"-"low" divide nevertheless proves highly discriminating. Low parental support is associated with far higher dropout risk (over 100% higher for all years except 1st grade): regardless of when parents' attitudes are assessed, roughly 56% of children drop out when parental support is (relatively) low versus 27% when parental support is (relatively) high. First grade parent interviews were completed before issuance of 1st quarter marks, which means that parents' sense of their children's academic prospects before a single report card had been received is almost as discriminating as is their thinking at the start of high school.


Parents are active agents in shaping their children's futures. Here we have measured parents' attitudes; but attitudes predict behavior, and it is through both thought and deed that parents guide their children's development. For example, parents who expect their 1st-grade children to do well in school visit the library more often than do parents with lower expectations (Entwisle, Alexander, & Olson, 1997). In instrumental terms, supportive parental attitudes help move children along the path to school completion, and from Table 5 we can conclude that this is true even for beliefs held by parents at the very start of their children's formal schooling.

Parents presumably understand, if only imperfectly, how opportunities are structured in school and later in life. For them to sense how the present anticipates the future is not surprising. This same awareness might not be expected of their children, though, especially when they are young. Nevertheless, children's "thinking" and children's "doing" also forecast dropout in Table 5, although they do so on somewhat different timetables. Behavioral engagement, as measured in 1st grade, separates future dropouts from future graduates almost as effectively as behaviors 9 years later. Attitudes, on the other hand, become more discriminating with time, such that in year 9 patterns involving children's attitudes look much like the pattern among parents: The "favorable"-"unfavorable" and "high"-"low" differences both are in the vicinity of 30%, with 27% of children on the positive side of the attitudinal divide dropping out against 55%-59% of children on the negative side. But among parents, the figures are much the same all along, even back to 1st grade. Not so for children, whose early attitudes do not forecast dropout especially well. This makes sense from a developmental perspective, as children become more self-aware and self-reflective with age (e.g., Bandura, 1981; Stipek & Maclver, 1989). Indeed, it is impressive that self-doubts and low psychological engagement with school at age 6 link to elevated dropout risk at all (i.e. 47% dropout, as compared to .38% among those rated high in self-confidence and psychological engagement).9

With an attitude-to-later-behavior connection already discernable in 1st grade, it should come as no surprise to also see an early-behavior-to-later-behavior connection. Compared to those above the mean on a scale that includes work habits, ratings from report cards, and teachers' ratings of children's classroom deportment, twice as many children whose 1st grade engagement behavior scores fall below the mean eventually leave school early (61% versus 29%). And this difference does not change much until high school, at which point it increases to 71% versus 22%. Life course studies establish that transition points are difficult for children, so a jump of this sort around the time of the middle school to high school transition seems reasonable. Life course studies also emphasize continuities in development, so a strong connection back to behavioral disengagement when children's school careers are just commencing also seems reasonable. Engagement behaviors measured in years 1 and 9, for example, correlate .34, as against an attitude-to-attitude correlation over the same 9 year interval of just .06. And too, different strands of development often come into alignment as children maturein the present instance an attitude-to-behaviors correlation of .23 in 1st grade increases to .52 in year 9. The more typical "risk factor" approach has little to say about "timing" and "turning points," but such ideas are integral to a life course view of dropout as culminating a long-term, progressive disengagement from academic commitment.

Another question raised by taking a life course perspective on development is how multiple contexts are joined: Do personal resources alter the dropout risk posed by low family SES? This issue is addressed first in Table 6, and then in more detail in Charts 2 and 3. Table 5 established that dropout is higher when attitudes and behaviors are unfavorable and lower when attitudes and behaviors are favorable. Table 6 shows that this holds for families at all SES levels. For children in lower SES households, for example, dropout risk as assessed in 1st grade moderates to 50% when parents hold favorable attitudes and to 47% when students themselves are positively engaged behaviorally with school (60% is the "expected" level of dropout for such youngstersTable 2) By the same token, dropout rates increase when upper SES youth lack such resources. Unfavorable parental attitudes in 1st grade are a particular liability: Thirty-one percent of upper SES children in this category eventually drop out against 15% of upper SES youngsters overall and 10% when parental attitudes are favorable.




Whether personal resources are favorable or not thus shifts dropout prospects, but does this "shift" explain the social patterning of dropout? Children from lower SES families who have parents with favorable attitudes at 1st grade are 25% less likely to drop out (versus children from lower SES families who have parents with unfavorable attitudes). Children from higher SES families who have parents with favorable attitudes are 68% less likely to drop out. The benefits associated with favorable parental attitudes and the risks associated with unfavorable parental attitudes thus clearly play a larger role in the lives of children from higher SES families. Indeed, the same can be said for all personal resources identified in Table 6, both in 1st grade and in elementary school. Personal resources also play a more important role in the lives of children from mid-level SES families and especially in 1st grade.

This is not to say that lower SES children do not benefit from favorable parental attitudes and a positive sense of self. Supportive personal resources reduce the likelihood of dropout across the board, although the most significant reductions in dropout for lower SES children (in both 1st grade and middle school) are associated with engagement behaviors, not attitudes.

But whether personal resources moderate SES differences is not the only concern. Such resources play a positive role apart from any tie-in with background risk, and in combination their "insulation" value can be considerable. Chart 2 shows this for 1st grade; Chart 3 for the middle school years.

Comparisons in the top portion of the two charts essentially rearrange information from Table 6.10 For children at all SES levels, dropout risk declines when engagement attitudes are positive and increases when they are negative, with contrasts across the "favorable"-"unfavorable" divide generally sharper in the middle grades than in 1st grade. For instance, comparing lower SES youth with favorable and unfavorable engagement attitudes, a 4% 1st grade difference (Chart 2) increases to 22% over the middle school years (Chart 3).

The new insights in these displays is what happens when engagement behaviors are overlaid on engagement attitudes at the "base" of Charts 2 and 3. Lower SES youth whose 1st grade engagement attitudes and engagement behaviors both are negative drop out at a rate of 72% versus a dropout rate of 11 % among upper SES youth whose engagement attitudes and engagement behaviors both are positive. The corresponding figures for attitudes and behaviors measured during the middle school years are 80% and 4%. The "insulation" value of attitudes and behaviors thus add onto one another, such that prospects move toward the extremes when they align. However, personal resources do not always mirror the family's material resources; and when resources "misalign" this too has consequences. Having favorable engagement attitudes and behaviors in 1st grade reduces dropout among lower SES youth to 51%; in middle school, the reduction is to 41%. Forty-one percent dropout is still much too high, but with 60% dropout the norm for disadvantaged children in Baltimore (Charts 2 and 3), this represents a sizeable one-third reduction in risk.

For middle SES children the comparisons are even more compelling. Favorable middle school attitudes and behaviors cut their dropout rate almost in half (from 30% to 16%), and it increases by half (to 45%) when they are unfavorable. When positively engaged such mi4dle SES youngsters are more likely to finish high school without interruption than are all upper SES youth save those with a similar engagement profile; when they are disengaged upper SES children leave school at a higher rate than lower SES youth who evidence strong attachment to school. Comparisons among higher SES youth likewise evidence substantial personal resource contingency; whether in 1st grade or the middle grades, higher SES children who score low on engagement attitudes and behaviors are about twice as likely to leave school without degrees as are their same class peers generally (24% versus 15% in the first instance; 37% versus 17% in the second).

To sum up, to this point we have described risk factors for dropout involving family, school, and personal resources over the first 9 years of children's schooling, with intersecting and offsetting risks examined in combination. As expected, prospects for remaining in school improve when parents provide emotional support for their children's schooling and when children themselves are self-confident and engaged with school. However, not all "early precursors" follow the same developmental timetable: Although behaviors are important all along the way, children's attitudes become more important over time. These patterns thus bear out another tenet of life course ideas: Even very young children are very much "producers of their own development" (e.g., Lerner & Busch-Rossnagel, 1981), albeit here more so in terms of their actions than their attitudes.

We also have explored whether differences across SES lines in such resources account for SES differences in dropout risk. The issue obviously needs more study; but even when personal resources are equivalent, lower SES youth have much higher dropout rates than do upper SES youth. "Positive" personal and family resources do offset material disadvantage to some extent, but not enough to counteract fully the drag of low family SES.

Scrutinizing how dropout risk is patterned at this level of detail affords a great many insights, but such an approach is limiting as well. For example, case coverage at the terminal nodes of our branching diagrams was quite small with just four or five risk/resource factorstoo small for us to add additional branches. And too, simple percentage comparisonseven detailed percentaging of the kind presentedare not a proper basis for drawing conclusions about the independent role of various risk factors. Accordingly, as a complement to the descriptive materials just reviewed, a more broad brush but also more encompassing stock-taking is undertaken in the next sections. These report the results of logistic regression analyses, predicting dropout risk under the life course conceptual framework sketched in Figure 1.



We begin by reviewing odds ratios from univariate logistic regression analyses in which dropout is predicted from selected measures from the four risk/resource variable clusters used in the descriptive profile just reviewed: a) family background characteristics, b) academic performance and experiences, c) parents' attitudes, and, d) children's engagement attitudes and behaviors. However, collinearity and excessive attrition prompted us to winnow the predictor set for the multivariate regression analyses toward which the univariate regressions reported in this section build. Our objectives were to retain issue coverage, to reduce clutter, to maximize the size of the analysis sample, and to retain sample representativeness. For example, we dropped resource measures that evidenced significant associations with dropout individually but that fell to nonsignificance when other predictors from the same resource cluster were controlled: family type and maternal employment (both in 1st grade), special education services at the four stages of schooling, academic course level placements in the middle grades and grade 9 (e.g., honors level, regular, remedial), and curriculum concentration in grade 9 (college preparatory versus general and vocational). These school experiences and family conditions no doubt have bearing on dropout risk, but none stands out as distinctively important. Also, preliminary analyses distinguished between two measures of school performance (achievement test scores and report card marks) and two student attitude measures (engagement attitudes and academic self-image). At each stage of schooling all four measures were significantly associated with dropout at the zero-order level and the associations within pairings were close in size; however, collinearity was a concern and we elected to combine measures. One composite averages marks and test scores and another averages self-attitudes and engagement attitudes.

Two sets of univariate regressions are reported in Table 7 for this scaled down set of resource measures. The first entries are the coefficients obtained when dropout risk is predicted from each of the explanatory measures individually. Sample sizes, reported in parentheses, vary owing to uneven data coverage across predictor variables (case coverage on the [dependent] dropout variable is 729 throughout, representing 92% of the original panel) The second entries also are zero-order odds ratios, but calculated on the group with complete data coverage across all predictors (N = 364), the realized longitudinal sample.


This degree of sample attrition deserves careful scrutiny. Data coverage in general is reasonably goodthe case base for the individual measures in Table 7 averages 669, or almost 92% of the maximum possible (729). However, the cumulative effect of source-specific missing data is quite substantial, as upwards of 50 discrete data sources are being interwoven, including student surveys, parent surveys, teacher surveys, and school records. This obliges us to ask whether the realized analysis sample is faithful to those characteristics of the original group that bear upon the dropout process. Is the panel reasonably representative; can we generalize results based on it?

Selective attrition and its possible distorting influence have been assessed in various ways, two of which we report herethe side-by-side comparison of zero-order odds ratios presented in Table 7 and a similar alignment of variable means reported in Table 8. These checks generally are reassuring. We begin with the odds ratio estimates.

Most of the zero-order odds ratios are significant in both sets of results in Table 7 and many are quite substantial. An odds ratio of 1.0 is equivalent to "no difference." For example, the panel odds ratio of 1.09 associated with race implies that the odds of dropping out among African Americans are about the same as the odds among Whites. Entries greater than 1.0 signify that relative risk goes up as values of the predictor variable increase (e.g., stressful family change increases dropout risk); whereas values less than 1.0 signify that relative risk goes down as values of the predictor variable increase. High family SES, for example, is associated with decreased dropout risk. The family SES odds ratio of .219 in Table 7 implies an almost five-fold difference in dropout risk for children one unit apart (about a standard deviation) on the SES scale.

Other differences in Table 7 are also large. Those involving school performance and engagement behaviors, for example, and grade retention. Repeating 1st grade is associated with more than a three-fold increase in dropout risk, whereas middle school repeaters are vastly more likely to eventually leave school as are nonrepeaters. And it is intriguing that engagement attitudes and behaviors appear to become more discriminating with age (remember, odds ratios closer to LOO signify weaker associations), a reasonable developmental progression. We should note too that mother's age bears a nonlinear relation to dropout, the only nonlinearity detected in preliminary analyses: Dropout risk decreases with mother's age except among the oldest BSS mothersthose over 35. Their children evidence high levels of dropout.11

Much could be said about these and other substantive details of Table 7, but that discussion is best deferred until the multivariate analyses guided by Figure 1 are presented in the next section. For now it is sufficient that we see significant and sizeable connections to eventual dropout for a whole host of risks and resources over children's formative years.


Returning to the issue of sample attrition, comparing across the two columns of Table 7, the odds ratios in most instances are very similar, perhaps remarkably so given the extent of sample attrition. There are no sign reversals, only one comparison crosses the significant-nonsignificant divide (i.e., the estimates for sex, both of which show boys at elevated risk of dropout relative to girls), and the coefficients themselves generally are quite close: The average panel estimate is within about 12% of its corresponding full-sample counterpart. Moreover, the larger discrepancies, those in the 20%-25% range, are scattered throughout the table and evidence no particular patterning. Parents' attitude is the one possible exception in that three of the four panel estimates of its effect are larger than the corresponding full sample estimates. In the multivariate results, however, parents' attitudes make only a weak showing (see below). Hence, even if associations involving parents' attitudes are somewhat inflated owing to selective attrition, the consequences of this for our main results appear minimal.

Much the same can be said of the means comparisons reported in Table 8. The biggest difference involves the panel's racial-ethnic makeup: African Americans constitute 64% of the panel as against 55% of the original sample. This difference corresponds to about .18 full-sample standard deviations. Three other differences in Table 8 exceed .10 SDs (engagement behaviors in elementary school; school performance and engagement behaviors in middle school), but in general the panel and full sample averages are quite close: Across all comparisons, the discrepancies average about .057 standard deviations.

We have performed other attrition checks as well (e.g., screening separately on the variable clusters defined around the several panels of Figure 1), and those results too indicate little attrition bias. Fifty percent sample shrinkage certainly is higher than we would like, but we have checked its implications carefully and so far as we can determine distortion owing to selective attrition is minor. That being the case, the panel estimation, of Figure 1 is taken up in the next section.


The analysis reported in Table 9 evaluates determinants of dropout through the conceptual lens of Figure 1. Entries in column 7 are what economists and path analysts call "structural estimates"all predictors, exogenous and endogenous, are in the equation together. These results are our best evidence of how the various pieces fit together in terms of direct effects on dropout and we will turn to them shortly. But when assessing the role of many correlated predictors there is a risk of overcontrolling and that risk is magnified when some of the predictors, as in the present instance, are lagged versions of others. So, for example, the lasting or direct influence of 1st grade risk/resource measures in column 7 is estimated controlling for counterpart measures from the remaining years of elementary school, the middle school years, and children's 9th year of school (the first year of high school for nonrepeaters). "Best evidence," under such circumstances, might not be especially good evidence. One way to guard against missing something important is to build up gradually from simple to complex. That is the strategy used in the present section, with the life course chronology of Figure 1 as scaffolding.


The first four columns of Table 9 report odds ratios for equations defined around the four stages of schooling in Figure 1. For each stage, effects of resources are estimated net of others from the same stage and a second set of estimations adds controls for background factors. This is what distinguishes the paired entries in columns 1-4. In other words, the 1st grade cluster evaluates 1st grade risk/resource measures together and also adjusts their effects for differences associated with race/ethnicity, sex, family SES level, family stress, and mother's age. The second column reports the same two estimations for resource measures from the remaining elementary years, and so forth.

When controls are phased in this way, many of the relationships that were significant at the zero-order level drop to nonsignificance. This is to be expected. Also, explanatory power generally increases across levels of schooling, from a pseudo-r2 estimate of .321 in 1st grade (with background factors included) to a pseudo-r2 estimate of .498 for year 9 predictors.12 This too might have been expected: Effects of risk factors and resources become more definitive as children mature, accumulate more experience, and the opportunity for action nears.

The effect estimates themselves indicate both continuity and change. Engagement behaviors and school performance, for example, stand out as distinctively important at all stages of schooling, The odds ratios estimated for both are significant in almost every instance and-their discriminatory power increases over time.

On the size of these effects, consider the 1st grade estimates. Without background controls, the odds ratios for engagement behaviors and academic performance in relation to dropout are quite close.604 and .653 respectively.13 Hence, comparing children alike in other respects, a unit increase in either (both are scaled as standard scores) is associated with about a 35%-40% reduction in dropout risk. With background controls added, the behavior effect retains its significance; and though the effect of school performance fades at that point, in the results for all three later periods both resource measures retain their significance net of background factors. Additionally, over time their associated effects get much larger. For behavioral engagement the increase is from .652 in 1st grade with background controlled to .322 in year 9; for school performance, it is from .921 (not significant) in 1st grade to .333 in year 9.

The engagement behavior scale combines work habit ratings from report cards and teacher ratings of children's classroom deportment (e.g., externalizing behaviorfights, teases, fidgets; adaptabilityenthusiasm, creativity). Comportment that aligns with the values and expectations of the school thus seems to insulate children against dropout risk. The relevance of behavioral engagement in itself is none too surprising, as such alignment embodies the "person" side of person-environment fit (Keogh, 1986; Lerner, Lerner, & Zabski, 1985) and is important for a range of academic outcomes, both in the BSS (Alexander, Entwisle, & Dauber, 1993) and in other studies (e.g., Farkas, 1996). Its relevance for dropout might not have been expected to extend to children's behavior as 1st graders though (except for a similar result in Ensminger and Slusarcick, 1992), and that it emerges as singularly important so early in the process certainly would have been hard to foresee.

These results, reflect continuity across levels of schooling, but what of change? One difference is evident in column 3, the middle grades results. Grade retention first appears on the list of significant predictors at that point, and that is the only time it attains significance in these stage-by-stage estimations (see columns 1-4). The bivariate associations between retention and dropout are significant at every stage of schooling (see Table 7); but with other risks and resources controlled, it is only in the middle grades that retention stands out as distinctively important. And its effect in the middle grades is large: Repeating a grade then is associated with a sevenfold increase in dropout risk, and this is with concurrent measures of school performance, children's attitudinal and behavioral school engagement, and parents' psychological supports all controlled.

Grade retention as a risk factor for dropout is well established in the literature (e.g., Grissom & Shepard, 1989; Jimerson, 1999; Rumberger, 1995; Rumberger & Larson, 1998; Temple, Reynolds, & Miedel, 1998), and the association itself is hardly in dispute. Less certain, though, is whether retention causes dropout. The uncertainty inheres in the so-called "endogeneity problem." A shared history of poor school adjustment and of low levels of school achievement often is the backdrop to both retention and dropout. BSS research on determinants of grade retention shows this for repeaters (Alexander, Entwisle, & Dauber, 1994; Dauber, Alexander, & Entwisle, 1993), and the descriptive comparisons reviewed previously establish this as well for BSS dropouts. Under such circumstances, the seeming effect of grade retention on dropout could be spurious. As a government report (Goal 2 Work Group, 1993, p. 18) puts it: "were these students who would have been more likely to dropout even if they had not been retained?"

To achieve clarity, the standard remedy is to adjust statistically for the presumed common causes. The present embodies the past, and the concurrent measures of school performance, behavioral engagement, and so forth controlled in column 3 reflect not just children's middle school standing but also in some measure their relevant developmental histories. Viewed this way, the set of controls used here is both high quality and reasonably comprehensive (at least of person-level candidates).

The effect of these adjustments is to reduce the middle grades retention effect from 23.16 at the zero-order level (Table 7) to roughly the 7.1-7.2 range in column 3 of Table 9. This comparison contains two lessons: First, the simple unadjusted association indeed exaggerates the importance of grade retention as an impetus to dropout; second, the adjusted estimates of retention's influence remain large, suggesting that something about the retention experience indeed makes repeaters more likely than nonrepeaters to leave school without degrees.

The particulars of that "something" are not revealed in these analyses, but the fact that grade retention as a distinctive force first emerges in the middle grades may be a clue. For one thing, when middle grades repeaters are designated for retention they are not as far behind their promoted classmates academically as, say, 1st grade repeaters are at the time of their retentions (Alexander, Entwisle, & Dauber, 1994). If grade retention were simply a proxy for relevant academic difficulties, then 1st grade retention would be more problematic than middle grades retention, but this isn't the case even in the zero-order associations.

And the fact that grade retention stands out with school performance and deportment controlled suggests that something outside the child is at issue. The social side of schooling and repeater's off-time status seem likely candidates.

Grade retention takes children off the normal timetable of grade progressions, making repeaters conspicuous and possibly complicating their" social integration with classmates. This can cause problems at any age, but conditions peculiar to the middle grades may well heighten them. For one thing, the adolescent middle grades years (typically age 12-14) are a time of heightened self-consciousness when "fitting in" is paramount. But middle grades repeaters may find it hard to "fit in," having just recently been separated from their friends and thrust in to the company of younger children at a time in life when a year or two difference in age counts for a great deal (physically, emotionally, and socially). And since repeating is less common in the middle grades than in the early elementary years (Shepard & Smith, 1989), these children stand out even more.

Moreover, challenges on the social side of schooling still are fresh when the time comes to negotiate the middle school to high school transition. Educational transitions are hard on children generally (Belsky & Mackinnon, 1994; Dunn, 1988; Eccles, Lord, & Midgley, 1991; Entwisle & Alexander, 1989; Simmons & Blyth, 1987), and the high school transition is no exception. Relative to middle schools, high schools typically are larger, more bureaucratic, more impersonal, and more academically challenging. Under such circumstances, even high achieving, well integrated students can experience difficulty; and repeaters, being doubly disadvantaged in these respects, can't count on either to help ease their adjustment. Roderick (1995), for example, documents a precipitous drop in (eventual) dropouts' marks at school transitions, whereas in the BSS (Alexander, Entwisle, & Kabbani, 1999) marks and the scheduling of retention are closely aligned, with middle grades repeaters' marks falling off sharply in middle school.

And owing to repeaters' off-time status, all of this is happening around the time they are near the legal age of dropout (16 in Maryland). This puts them in a vastly different situation than elementary school repeaters within a year or two of their retentions. By virtue of their age, when middle grades repeaters get to high school they have an option for escaping a situation that many no doubt find punishing; and the elevated risk of dropout associated with middle grades retention in column 3 of Table 9 suggests that a good many in fact avail themselves of that option (see Grissom & Shepard, 1989, and Roderick, 1994, for similar interpretation).

We might also mention a procedural matter at this point that bears on the results presented for year 9, where the retention effect is not significant. For children promoted each year, year 9 would be the 1st year of high school; but owing to retention only 63% of the panel is at (or ahead of) grade level in year 9. The "stage of schooling" overlay thus is not altogether clean here, but the checking we have done suggests that the results presented in column 4 of Table 9 are not misleading on that account. We say this because we have tested for interactions involving the year 9 predictors in Table 9 and children's on-time/off-time status in year 9. None is significant.

In these checks, off-time status in year 9 itself is associated with elevated dropout risk, as would be expected from the significant middle school retention effects already discussed: children who are behind at the time of the middle grades to high school transition owing to earlier retentions have odds of dropping out roughly five times those of on-time 9th graders (not shown in tables). That is not our immediate concern, however. Rather, the question is whether ignoring children's grade-level standing in year 9 yields misleading results for the other year 9 predictors in Table 9. Do engagement behaviors, for example, relate differently to dropout among on-time 9th graders than among off-time middle schoolers; or perhaps parental supports play a different role at the two stages of schooling? More generally, are effects conditional on stage of schooling? When interaction tests were performed to detect such conditionality, none involving year 9 predictors was significant. It thus appears that the "cluster" results for year 9 apply equally to repeaters and to on-time youth.

To this point we have been commenting on significant coefficients in Table 9, but some of the nonsignificant coefficients also merit mention: Parents' attitudes are non-significant throughout and children's engagement attitudes do not emerge as significant until year 9 and then only marginally so (.10 level). We find this perplexing, as both were significantly related to dropout at the zero-order level at each stage; and we had expected both would be of consequence in the controlled estimations also, at least in the upper grades.

Our perspective holds that academic engagement with school manifests itself in both affect and behavior. Children's school conduct is important as far back as 1st grade, but their attitudes seemingly are irrelevant. Young children's attitudes toward self and school are rather amorphous, but as children mature their thinking crystallizes (e.g., Alsaker & Olweus, 1992). They become more self-directed, which suggests their ideas ought to play a stronger role in guiding their behavior, including dropout.

Likewise, the role of parental counsel and support ought to be more in evidence when children are older and dropout more salient as an issue. With the academic record and other considerations comparable and dropout a real option, the child whose parent is a supportive presence we expected would be less likely to leave schoolfor a young person teetering at the edge, parental "press" conceivably could mean the difference between staying and leaving.

Our reasoning is reflected to some extent in how the measures of engagement behaviors and attitudes track over time. For children these are correlated .37 in 1st grade, but by year 9 their relationship increases to .64a much tighter "bundling." Perhaps, then, attitudes are working through behavior, and the role of affective engagement is muted when concurrent behavioral engagement is controlled. This seems a reasonable possibility, and there is some support for it: When the measure of behavioral engagement is deleted from the analysis, the estimate for children's attitudinal engagement in year 9 increases (from .644 to .456) and becomes fully significant (at the .01 level); additionally, in the middle school years it becomes marginally significant before background measures are controlled. And although this exclusion has no effect on the results for parental supports, when the equations from Table 9 are estimated on a nonpanel basis so as to maximize sample size occasion-by-occasion, the effect of children's attitude is marginally significant in the middle grades results (N = 465) and that for parents' attitude is marginally significant in the year 9 results (N = 506).

In sum, when effects of resources are evaluated stage-wise, parents' psychological capital as a resource in support of children's schooling apparently does not add onto the effects of the other resources included in our analysis framework and only a modest role is indicated for children's own dispositions. These conclusions are provisional, however, as the picture could change when resources are phased in across stages so as to mimic children's cumulative history of schooling. Columns 5-7 of Table 9 adopt this perspective, but before turning to those results we need to comment on one further feature of the comparisons afforded by columns 1-4.


As mentioned, two sets of estimates are reported in each column of Table 9the first evaluates resources at a particular stage of schooling without background measures in the equation and the second adds background factors. The second set of estimates is to explore whether resources at different school stages account for sociodemographic differences in dropout risk. At issue is how the odds ratios for background factors change when experience and resource measures are controlled. They will attenuate when background factors and resources overlap (i.e., covary); and according to the interpretative logic of Figure 1, such overlap comes about because background effects work through the variables responsible for the attenuation. Said differently, resources mediate background effects.

The panel coefficients associated with race/ethnicity and gender fall short of significance at both the zero-order level (Table 7) and in the controlled results, leaving children's vulnerability associated with low family SES as a candidate for mediation, along with mother's age and family stress in 1st grade. In Table 9 the adjusted SES coefficient with 1st grade resources controlled is .302, not much different from its zero-order value of .219.

First grade resources thus account for a small portion of lower SES youths' elevated dropout risk, but the "accounting" increases after 1st grade. The family SES coefficient attenuates to .377 when resources from the upper elementary years are controlled in column 2; and at higher levels of schooling the "net" SES effect is smaller still.488 in the middle school years and .417 in year 9. Over time, then, school behavior and school performancethe strongest and most consistent predictors from the resource setbecome increasingly important to the dropout disparity across social lines. An initial five-fold difference in the odds of dropout comparing children a unit apart on the SES gradient reduces to a roughly two-fold "net" difference at the upper grades.

A doubling of dropout risk at low SES levels still leaves much to be learned, but it is telling nonetheless that children's academic standing and school behavior from before high school anticipate to such an extent the later socioeconomic patterning of dropout. Effects of family stress (in 1st grade) and mother's age, on the other hand, apparently have little to do with the many resource measures tested in Table 9. Their odds ratios in columns 1-4 are not much different from their corresponding zero-order values.


We began our evaluation of Figure 1 by examining effects of resources at each stage of schooling separately, the motivation being to guard against missing something important in a glut of intercorrelated explanatory variables. That approach proved informative. We saw, for example, that explanatory power for the resource set increased at each stage of schooling, that some resources were important at all stages of schooling (engagement behaviors and school performance), and that some resources were important at particular stages of schooling (retention in the middle grades and possibly children's engagement attitudes in year 9). Now we are ready to take a broader view that encompasses children's entire history of schooling, but one that increments the information base gradually so as to preserve potentially important detail. Accordingly, columns 5-7 add onto the 1st-grade predictors in column 1 sequentially, following the time line of Figure 1: the remaining years of elementary school first, the middle school years next, then year 9. In this way, the analysis traces the BSS panel's journey through school, with later experiences building upon the history that precedes them.

According to the pseudo-r2 figures, explanatory power increases steadily when later resource measures are added to earlier ones, from .321 when just 1st grade resources are evaluated (column 1) to .594 when all four stages of schooling are considered together (column 7). The latter figure also reflects a sizeable increase in explanatory power over that obtained when just resources from year 9 are used (.498 in column 4). This speaks to one of the issues raised in the introduction: that later resources add onto the influence of earlier ones. For the other issues, the coefficient estimates need to be examined: Are early resources distinctively important or is their influence routed through their developmental sequelae; are effects at certain stages of schooling so prominent as to suggest critical periods or turning points?

When 1st grade measures and measures from the remaining years of elementary school are evaluated together, school performance and behaviors from the later period retain their significance (school performance at the .10 level of significance). However, two measures from 1st grade also register significant effects in column 5school performance, which reverses sign from when the 1st grade cluster was evaluated alone, and grade retention, which was not significant previously (see results in column 1).

The 1st grade performance coefficient in column 5 associates better performance with elevated dropout risk. This anomalous result fuels suspicion that collinearity remains a problem. The equation in column 5 includes two measures of school performance and two measures of grade retention. All four are moderately intercorrelated, and the pattern of results in column 5 seems suspectfor example, the performance coefficient from 1st grade is negative whereas that from the remaining elementary years is positive, and both significant. Unstable coefficient estimates and sign reversals are classic signs of collinearity, but other conclusions from this analysis seem reliable in that they hold up when 1st grade performance is excluded from the analysis. These include the significant lagged effect of 1st grade retention on later dropout and the strong effect of engagement behaviors over the upper elementary grades, an effect that apparently carries along the earlier importance of 1st grade engagement behaviors (i.e., the engagement behavior coefficient was significant in column 1, but drops to non-significance when its counterpart measure from the upper elementary years is controlled).

The addition of middle school resources in column 6 of Table 9 yields a particularly interesting pattern. The analysis at this point controls for children's cumulative resources and experiences at home and at school over the primary grades, and so the middle grades resources build on that history. School performance from the middle grades is important in these results, whereas effects of performance from the elementary years fade. A similar pattern is obtained for engagement behaviors, but only with a bit of detective work: The coefficient for middle grades behavioral engagement falls short of significance in Table 9; but when the same equation is estimated on a nonpanel basis (N = 410), it reaches significance at the .10 level, and again the effects associated with earlier measures of school engagement fade.

This presumably reflects developmental continuity over the course of children's schooling; and much the same happens when year 9 measures are added to the predictor set in column 7, at least for engagement behaviors. Its effect in year 9 again emerges as fully significant, and the effect of behavioral engagement from the middle grades drops to nonsignificance. And although children's school performance does not stand out in a distinctive way in the year 9 results, children's engagement attitudes do. These first emerged as significant in the middle grades results (column 6) and their effect is significant in year 9 also, at which point the middle grades effect fadesanother manifestation, presumably, of developmental continuity.

In the previous section we acknowledged surprise that children's attitudes were not more prominent in the results, especially for older children whose sense of self ought to be better defined. Here it appears that attitudes indeed do matter "at the margin" when children reach their teen years, but only when evaluated in the context of their relevant developmental history. With performance trajectories similar, with the history of grade retentions and promotions alike, and with behavioral engagement equated (both concurrently and over time), attitudes come into play: A positive sense of self in the student role and a favorable view of the academic enterprise mitigate dropout risk.

The results discussed thus far all are consistent with the idea that resources in support of children's schooling cumulate over the course of their schooling: Later resources mediate the effects of their earlier counterparts (no lagged effects) and their effects strengthen (increased explanatory power). But Table 9 also includes a conspicuous exception to this pattern involving grade retention. In the stage-by-stage assessment, only retention in the middle grades increased the likelihood of dropout. Now, though, with school performance, patterns of behavioral and attitudinal engagement, and parental support equated all the way back to 1st grade, retentions at every stage of schooling prior to high school are associated with elevated dropout risk. The pattern is quite striking: The coefficient for 1st grade retention is significant in column 5 with elementary resources controlled; in column 6 with elementary and middle school resources controlled; and in column 7 with elementary, middle grades, and year 9 resources controlled. The coefficient for later elementary retention is significant in column 6 with elementary and middle grades resources controlled and in column 7 with elementary, middle grades, and year 9 resources controlled. The coefficient for middle grades retention is significant in column 6 with elementary and middle grades resources controlled and in column 7 with elementary, middle grades, and year 9 resources controlled.

We should note too that these effects are additive, implying that multiple retentions increase the hazard of dropout over the hazard associated with a single detention. Given that 36% of BSS repeaters experience two or more retentions, this is of considerable consequencerecall that 80% of multiple repeaters in the BSS leave school without degrees, including 94% of those held back in both elementary school and the middle grades.

All the significant retention effects in columns 5-7 precede high school, so all have the effect of putting children off-time when they make the transition from middle school to high school. Retention in year 9, on the other hand, is not associated with elevated dropout risk in the context of the full model. Though significant at the zero-order level, that association in the controlled comparisons drops to nonsignificant.

We previously directed attention to the social pressures that arise when repeaters are made "deviant" owing to rigid age-grading at school (e.g., Tyack & Cuban, 1995; Tyack & Tobin, 1994). That discussion was directed at middle grades retentions; and retention during the middle grades still appears to magnify dropout risk more than does retention during the primary grades, but the odds of dropout more than double even for distant retentions.14 The off-time problem, if that is the issue, seemingly attaches to all retentions that predate high school, an insight that comes to light only when the dropout dynamic is viewed in Me course perspective.


In addition to the main effects analyses reported in Tables 7 and 9, we also inspected for interactions. The objective of this exploratory exercise was to identify resources that stand out as especially important for children whose family circumstances and personal characteristics put them at elevated risk of dropoutlow family SES, for example, or being male. However, we also inspected for interactions within the resource setdoes positively adaptive behavior, for example, cushion adverse consequences of poor school performance in the early grades?

Many tests were run, but only a handful of interactions were significant and they followed no particular pattern. We conclude on this basis that risks and resources over the years before high school operate pretty much "across the board"they neither favor nor penalize particular kinds of children and their effects add onto one another. Providing high-risk children access to protective resources can offset effects of forces that put them at riskthat much seems clearbut the analysis has not identified resources that would profitably be targeted at high-risk youth specifically or exclusively.

There is some sentiment in the literature on resilient children that to qualify as "protective" an intervention or experience ought to be distinctively important for high-risk groups. McCord (1994, p. 122), for example, holds that "Preventive factors must be potent in the presence of risk. They need not be potent when risk factors are absent" (see also Rutter, 1987). We are not of such a mind; but by that standard, the results of these analyses are disappointing: The qualities and experiences that have insulation value are the same regardless of children's risk status, and that holds whether "risk" is defined in terms of background characteristics (e.g., Kaufman, Bradby, & Owings, 1992) or academic performance (e.g., Catterall, 1998).


This paper has provided a fine-grained look the high school graduation prospects of a high-risk sample of Baltimore youth in relation to resources and experiences over the pre-high school years. An accumulating literature traces the developmental roots of dropout to children's formative experiences at home and school (e.g., Brooks-Gunn, Guo, & Furstenberg, 1993; Cairns & Cairns, 1994; Ensminger & Slusarcick, 1992; Gamier, Stein, & Jacobs, 1997). The present inquiry adds additional evidence that this "trace" extends at least back to 1st grade.

A life course perspective on dropout directs attention to how multiple strands of human development (e.g., cognitive, affective) intersect and are shaped over time in response to children's experiences in the principal socializing contexts of their upbringingthe home and the school. But it is a process in which children play an active role. Through actions taken (or not) and through the views that guide those actions, they direct their own development. One question, then, is when such self-direction first becomes apparent. In our results, behaviors proved consequential as early hi the process as we were able to measure them (i.e., the fall of 1st grade); whereas, attitudes did not weigh in as distinctively important until much later.

This reverses how children are treated in much of the literature on schooling and social inequality, where attitudes and like constructs (e.g., aspirations and expectations) have commanded much attention. Although children's actions, apart perhaps from those implicit in measures of school performance, until recently (e.g., Cairns & Cairns, 1994; Ensminger & Slusarcick, 1992) have been virtually absent from the scene, children's actions have consequences; and this is true even if the principals themselves are not attuned to them. It is a stretch to think that 1st graders will self-regulate their classroom behavior in light of considerations so far removed as their later graduation prospects, but others who care about them and can be more forward looking need to be aware of these connections.

The experiences at home and at school that shape children's development influence not just how they think but also what they do. Habits of conduct, once established, tend to persist, as do reputations grounded in those habits. Teachers, we know, rate children on the basis of their classroom deportment (Natriello & Dornbusch 1983). We also know that children's work habits, classroom engagement, and compliance with school routines carry considerable weight in determining achievement levels (e.g., Alexander, Entwisle, & Dauber, 1993) and are implicated in achievement differences across social lines (e.g., Farkas, 1996). In the present results, engagement behaviors at school rival test scores and report card marks in forecasting eventual dropoutand this holds all along the way, including 1st grade.

The socializing influences of home and school themselves are situated within a broader context of social-structural constraints that influence, if not determine, the quality and character of what is experienced in those settings and the social lenses through which children filter their experience. This is evident in how socioeconomic differentials in dropout risk attenuate when allowance is made for children's school deportment and performance. But structural constraints are relevant for other reasons as wellthey often determine opportunity, for example, and by doing so channel children along different developmental paths quite apart from their role as venues for socialization (e.g., Kerckhoff, 1976; 1993). Family socioeconomic standing evidently is one such constraining influence, which implicates the family in both socialization and allocation: With parental attitudes, children's attitudes, and children's school behavior all equated statistically, upper and lower SES youth still evidence quite different dropout probabilities.

At issue here is the large "unexplained variance" component of the SES-dropout connection. More study is needed to ferret out the nature of that connection. Community conditions (e.g., Crane, 1991; Elliott, et al., 1996) and school organizational conditions (e.g., Bryk & Thum, 1989; Entwisle, Alexander, & Olson, 1997; Rumberger, 1995; Wehlage & Rutter, 1986) beyond the person-level performances and placements that have been our present focus no doubt are relevant; and we suspect aspects of family circumstance not well reflected in standard socioeconomic measures (e.g., the duration and depth of poverty) also play a role, along with "nearer term" life-stage considerations involving the transition to adulthood. Intensive work commitments (i.e., more than 15-20 hours weekly during the school year), for example, apparently weaken commitment to school (Committee on the Health and Safety Implications of Child Labor, 1998),15 as does becoming a parent, at least among girls (e.g., Anderson, 1993). Problem encounters with the law and delinquent involvement also can interfere with school continuation, a "cause" of dropout that has received surprisingly little attention (e.g., Elliott, 1965; Kaplan, Peck, & Kaplan, 1997; Mensch & Kandel, 1988; Stroup & Robins, 1972. But see also Fagan & Pabon, 1990).

These sundry experiences all involve young people in adult-type roles, possibly before they are ready to assume adult-type responsibilities. In life course parlance, they are "accelerated role transitions" (e.g., Pallas, 1984; 1993; Russell, 1980). Such early onset gives young people a taste of adulthood, a heady experience that can make the dependent student role seem awfully confining. This adds on to other obvious considerations. Dropouts, by and large, have not found much comfort at school and for that reason their attachment to the institution is fragile. When faced with conflicting demands on their time and attractive opportunities elsewhere, school may well have the weakest claim.

For these reasons and others, many poor performing students find life outside school more to their liking than life inside school. It is an implicit pain-gain calculation in which the school oft-times will lose. The decision is not so much irrational as shortsighted, a consequence of the "pseudo-maturity" that comes with premature role transitions (e.g., Shanahan, Elder, Burchinal, & Conger, 1995).

A comprehensive life course perspective would integrate the longer-term developmental influences from the family-school socialization arena that has framed the present inquiry with nearer-term constraints of the sort mentioned. In fact, the role of grade retention documented in the present analysis is precisely such a bridge, involving, as it does, school-based structural constraints.

Other structural aspects of school organization also may be relevant to dropout, but the evidence for them is weaker than that for grade retention. In preliminary analyses a whole host of school stratification measures had significant, and in some instances sizeable, zero-order associations with dropout, including receipt of special education services, academic course level placements in middle school and 9th grade, and curriculum concentration in 9th grade. None, though, retained its significance in the full model. But even with school performance and behavior controlled, grade retention was found to elevate dropout risk. These controls eliminate the most plausible person-level candidates for explaining the link and so cast suspicion on organizational processes. In this vein, we directed attention to how rigid age-grading burdens children who fall off the expected timetable of age-grade progressions by complicating their social integration at school.

The timing of retention's effects in our results certainly accords with this interpretation. Grade repetition just before the transition to high school elevates dropout risk especially; but earlier retentions also boost the odds of dropout, in most instances by several-fold. In fact, when the analysis was structured so as to follow the time line of children's schooling, all retentions prior to year 9that is to say, all retentions that have the effect of delaying high school entrywere associated with increased dropout risk.

In light of recent calls to end social promotion (American Federation of Teachers, 1997; U.S. Department of Education, 1999), there is a particular need to understand the likely consequences of different promotion/retention practices. Elsewhere (Alexander, Entwisle, & Kabbani, 1999) we have advocated "third way" alternatives to the two traditional remedies when children fall short of reasonable promotion standards. Moving them ahead ill-prepared for what awaits them is not good educational practice, nor is wholesale "recycling" through grade repetition. However, we also have argued (Alexander, Entwisle, & Dauber, 1994) that commentary on grade retention has tended to exaggerate its adverse effects (e.g., House, 1989; Smith & Shepard, 1987). In our earlier results, repeaters suffered no emotional scars from the experience and often their school performance improved. That earlier work did not examine high school dropout however, and the evidence presented here implicates a side of grade retention that has little to do with those other lines of debatethat is, does retention help or hurt academically; is it stigmatizing?

Grade retention apparently takes many youth off the path to school completion, and this happens quite apart from any consequences for self-image and/or school performance, whether favorable or not. The problem seemingly is an organizational one and, if that is the case, then fixing it will require an organizational solution (e.g., to somehow relax the overly tight link between age and grade). In the meantime, the well-being of many young people hangs in the balance.

There is considerable enthusiasm at present for "get tough" policies to hold back poor-performing children. Juxtaposing this with the robust dropout risk that attaches to grade retention leads to a discomforting conclusion: If those policies do not also effectively address children's learning needs and social integration at school, over the long haul they risk doing more harm than good. In light of the present results, and similar evidence reported by others (e.g., Grissom & Shepard, 1989; Roderick, 1993; Rumberger, 1995; Temple, Reynolds, & Miedel, 1998), it would not surprise us in the least to see dropout rates escalate when these "toughened up" children begin to come of age (e.g., McDill, Natriello, & Pallas, 1986; Pallas, Natriello, & McDill, 1987).

And what of the long-term process of school disengagementthe perspective that has motivated this inquiry? That imagery seems quite appropriate when applied to high school dropout. About 60% of BSS children in lower SES families drop out of school versus 40% overall and 15% of those in higher SES families. Such a huge disparity signals a problem of immense proportions. Of the background factors considered, family socioeconomic level bears the strongest relation to dropout by farstrong enough to say that the dropout problem in Baltimore, at its core, is a problem of economic and social disadvantage. But family structure, mother's age, family stress, and maternal employment also were shown to enhance or reduce dropout risk, along with the other academic and personal resources we have examined. A stable family, for example, and good academic performance at the start of school improve the graduation prospects of disadvantaged youth.

But many dropout-prone youth do not get off to a good start at school and over time their problems mount. Recovery from a shaky beginning at school is always possible; but by the time dropout-prone youths get to high school, the battle for many effectively has been lost. How does one "reengage" children who exit the primary grades plagued by self-doubt, alienated from things academic, over-age for grade, prone to "problem behaviors," and with weak academic skills? Imaginative approaches to school reform at the upper grades surely need to be encouraged (e.g., Wehlage et al., 1989), but against such an accumulated history the prognosis is not good.

This does not, of course, give license to ignore what happens later. To the contrary, behavior patterns, performance measures, and other considerations from middle school and the 1st year of high school add explanatory power over their elementary school counterparts and mediate much of the influence of those earlier measures. From a developmental perspective both roles are importantthe first reflects accentuation of prior patterns, the second continuity. They identify possible points of inroad all along the way, and those too are important. But on the whole and in general, prevention is to be preferred over remediation. Once children have fallen behind, to catch up requires that they make greater than average strides. That's asking a great deal of children who have experienced only failure and frustration at school; and it is asking a great deal too of those responsible for helping them. How children acclimate to school initially, we see, lays the foundation for what follows later, even many years later.

One prominent critic (Fine, 1990) has charged that these kind of conclusions amount to "blaming the victim." We do not view an early risk factor approach in that light; nor do we agree with Fine that our approach precludes a structural critique of the arrangements in school and in the broader society that contribute to the disadvantage of the disadvantaged. Indeed, by documenting huge differences in graduation prospects across SES lines beyond those anticipated by a host of personal, family, and school resources over the first 9 years of children's schooling, the present analysis effectively refutes the notion that school failure is reducible to personal failing. But in recognizing this, it would be equally wrong to strip lower SES families and their children of personal agencyparents who are materially poor can and do act in ways that support their children's schooling, and their children too play a role in directing their own academic development. Our risk factor approach identifies sources of leverage that are potentially within the reach of such families (and of those committed to helping them), and the life course overlay suggests strategic times for their deploymentearlier, as a rule, is better than later.


Children who were beginning 1st grade in Baltimore City Public Schools (BCPS) in fall 1982 were selected for study through a two-stage process. First, 20 schools were chosen randomly from within strata defined by racial mix (6 African American; 6 White; 8 integrated) and by socioeconomic status (14 inner city or working class; 6 middle class); then students were randomly sampled from 1st grade classrooms using kindergarten rosters from the previous school year, supplemented by class rosters after school began in the fall. Three percent of parents refused to have their children participate.

The final sample of 790 nonrepeating 1st graders was 55% African American, 45% White (versus about 23% White enrollment system-wide in the BCPS at the timethe BSS deliberately oversampled Whites in order to sustain comparisons by race). Parents' educational levels ranged from less than 8th grade to graduate and professional degrees, averaging 11.9 years of schooling. According to school records, 67% of BSS families received free or reduced-price school meals (about the same percentage as system-wide); with 56% of the study youngsters were living in two-parent households at the beginning of 1st grade.

Data were collected from students through face-to-face interviews beginning in the fall of 1982 and continuing twice yearly (fall and spring) for most of the nine years covered in the present analysis (project years 3 and 5 are exceptions-no pupil interviews were done during those years). In years 1, 2, 4, 6, 7, 8, and 9, teachers responded to self-administered questionnaires. Parents were surveyed on much the same schedule as teachers, but also in year 3. Parent surveys almost always were completed before the end of the first marking period in the fall. In the early years, parent surveys were self-administered, later they were done mainly by phone, but the occasional request to be interviewed personally was accommodated. School records were reviewed for data on marks, test scores, retentions, and the like.

Initially, only students who remained in the original 20 schools were followed. Beginning in the 4th project year, tracking was extended to public schools throughout the city, with data from school records backfilled when possible for students who had changed schools before year 4. In the fall of project year 6 (1987), 490 of the original 790 students (62%) remained in city schools and were still participating in the study. The next year (project year 7), tracking was extended outside the BCPS and a major effort was launched to reenroll those who had earlier left the studymainly children who had transferred out of the BCPS. At the end of project year 13 (spring 1995), participation stood at 84% (N = 663) of the original group.

The still active sample compares well with the original group. Means and standard deviations for key variables (e.g., test scores, marks, family context measures) from 1st grade are quite similar, for example. However, children lost from the sample resemble dropouts more so than graduates on most measures (see Table Al at the end of this appendix).



"Dropout" was self-defined by panel members in annual interviews from year 9 of the study (age 14) through age 22/23. A typical dropout question (from year 12) asked: Are you currently attending high school? Response options were (caps in original):

1. Yes. I was in school for the whole year.

2. No. I attended some of this year, but dropped out DURING the 1993-1994 school year.

3. No. I received a high school diploma early (NOT A GED) DURING the 1993-1994 school year.

4. No. I dropped out BEFORE the start of the 1993-994 school year.

5. No. I received a high school diploma (NOT A GED) BEFORE the start of the 1993-1994 school year.

Other questions covered the exact timing and grade when the first dropout occurred. When individual records of school attendance were incomplete or inconsistent, a judgement was made on the basis of the entire record as to whether (and when) the panel member had dropped out.


Sex is coded 1 for girls, 0 for boys.

Race is coded 1 for African Americans, 0 for all others, most of whom are White.

Family SES is constructed as the average of five items, after conversion to Z scores:

1. & 2. Mother's and Father's Years of Education, from parent reports pooled over years 1-8.

3. & 4. Mother's and Father's Occupational Status, coded in the SEI metric (Featherman & Stevens, 1982) from parent reports pooled over years 1-9.

5. Family Income (low versus not low), as indicated by participation in the reduced price school meal program over years 3-5 (from school records).

Family SES scale scores are available for 787 of 790 panel members with just under 70% calculated on 4 or 5 indicators and 5.4% on a single indicator. Alpha reliability calculated on all 5 items (N = 386) is .86. For descriptive purposes, families also are ranked according to socioeconomic level: "lower," "mid-range," and "higher." With the cutting points selected, mother's education averages 10.0 years for the lower SES group, 12.0 years for the middle group, and 14.6 years for the upper group; the respective percentages participating in the meal subsidy program for low income families are 95.1, 53.4, and 13.1. There are few genuinely wealthy households in the BSS, and it should be understood that these descriptors-"lower" and "higher/upper"are relative to the sample's makeup. In fact, half the cohort is located in the lower SES category, a reflection of the study group's low socioeconomic standing overall.

Family Type during 1st grade distinguishes two-parent families (natural or step, coded 1) from all other arrangements (coded 0) (from parent reports).

Maternal Age at the birth of the study child was determined from self-reported birth dates.

Mother Born Before 1941 is a generational variable determined from self-reported dates of birth (mother born before 1941 = 1; otherwise = 0). Mother's age evidences a nonlinear relationship with the risk of dropout. The relationship appears quadratic and changes sign when mother's age reaches around 36 years (i.e., for mothers born before 1941).

Teen Mother distinguishes mothers who were under age 20 at the birth of the study child (coded 1) from mothers 20 and over (coded 0), as determined from mother and child birth dates (self-reported).

Mother Employed during 1st grade is determined from parent reports: 1 = yes; 0 = no.

Family Change during 1st grade is the sum of up to seven events: divorce, marriage, a family move, illness, death, adults leaving the household, adults entering the household. It is measured retrospectively from parent reports over project years 6-9.


California Achievement Test (GAT) Scores are the average of fall and spring reading comprehension and math concepts/reasoning subtests after converting scale scores to standard scores. Averages are calculated within years (as reported for 1st grade and year 9), and then averaged across years to determine children's standing over years 2-5 (the remainder of elementary school) and over years 6-8 (the middle school years). The BCPS stopped fall testing after year 6, so for years 7 and 8 only spring scores are available. After year 8, the BCPS discontinued use of the CAT battery; but beginning in the spring of year 9 and continuing for roughly 18 months, the BSS did its own testing. These scores are referenced back to spring of year 9 using a linear interpolation.

Report card marks, from school records, are measured each year as the average of reading and math marks from all four quarters (1= failing; 13 = A+). Beginning in year 6, English and science marks also are included.

Receipt of Special Education Services is from BCPS records: 1 = received some services during the period at issue; 0 = no services during the period at issue. In descriptive analyses, separate classes ("high") are distinguished from pull-out programs ("low"). These data are not available for children who transferred out of the BCPS.

Grade retention information is mainly from school records, with retrospective student and parent reports from years 10-12 used to identify retentions outside the BCPS and to correct some mistakes in school records. 1= retained during the period at issue; 0 = not retained. Total retentions period-by-period also are available.

Course level placements in 6th and 9th grade are determined from school records. High level English, math, and science courses are those classified as Advanced Academic or Honors. Low level courses are those classified as remedial (which includes reading in 6th grade). Foreign language courses at any level are considered high in both grades. We experimented with various organizations of these data (e.g., total number of high-level courses; total number of low-level courses), and the program "mix" constructions we settled on provide about the same degree of discrimination as the others. The "mostly high" and "mostly low" designations capture the preponderance of course taking across the four subjects after "netting out" courses at other levels (valuing regular level courses at .5 and high- and low-level courses at 1.0 for this purposefor example, a "mostly high" program could include two regular-level courses or one low-level course). Likewise a "regular" program can include a net total of one or no high-level or low-level courses (i.e., after subtracting the number of high from the number low).

High school program in 9th grade (1 = college preparatory; 0 = other) is determined, in the first instance, by the character of the student's high school (e.g., city-wide academic, city-wide voc-tech). For those attending comprehensive zoned schools, self-reports are used.


Parent Attitudes are constructed from four items:

1. The average of mark expectations for the upcoming report card in math and reading, and, beginning in middle school, English and science also (coded from 1 to 13, with 1 = failing; 13 = A+).

2. Conduct expectations for the upcoming report card (usually dichotomized, with 1 = satisfactory; 0 = needs improvementthe distinctions used in the BCPS during the primary grades).

3. Parents' sense of their children's "ability to do school work" relative to others in the school (coded from 1 to 5, with 5 = much better; 1 = much worse).

4. Parents' expectation for their children's eventual level of school attainment (1 = not finish high school; 6 = more than college).

These items are available for at least some parents for years 1, 2, 3, 4, 6, 7, 8, 9, and 10. Parent attitude scores are constructed each year as the average of the measures available, after conversion to standard scores. Alpha reliabilities calculated year-by-year on the four items range between .61 and .71, except in 1st grade (.43). These annual constructions then are averaged across years to derive measures that span several years. To fill out coverage in year 9, when just 463 parents were surveyed (as against 651 in year 10), data from year 10 are used for parents not interviewed the previous year. However, to avoid problems of endogeneity, no substitution is done for children who dropped out between the two surveys. This approach increases year 9 case coverage on the parent attitude scale from 441 to 613 (relative to a maximum of 729the number of children whose graduation status is known).

Children's Self-Attitudes are constructed from four sources:

1. The average of mark expectations for the upcoming report card in math and reading, and, beginning in middle school, English and science also {coded from 1 to 13, with 1 = failing; 13 = A+).

2. Conduct expectations for the upcoming report card (usually dichotomized, with 1 = satisfactory; 0 = needs improvementthe distinctions used in the BCPS in the primary grades).

3. Self expectation for eventual level of school attainment (1 = not finish high school; 6 = more than college).

4. A measure of academic self-image, itself having three components:

a. Children's self-rating of their "smartness" relative to others in the school (4 = one of the smartest; 1 = not as smart as most kids).

b. A five item scale that rates self-ability in different academic areas (e.g "How good are you at. . .": math, reading, being a good student, learning new things quickly, and writing/handwriting, with response options ranging from 5 = very good to 1 = very bad).

c. A six item self-rating of academic competence developed by Harter (1982; e.g., "Some kids feel they are very good at their schoolwork, others feel. . . Is this very true for you or sort of true for you?").

Educational expectations, the Harter scale, and the "smartness" item were not elicited in 1st grade, nor were educational expectations asked in year 2; the "How good are you at . . ." questions were not asked in year 9. Otherwise, all sources are available each year students were interviewed (Years 1, 2, 4, 6, 7, 8, 9), with questions sometimes repeated fall and spring. Student self-attitude scores are constructed each year as the average of the measures available, after conversion to standard scores. These then are averaged across years to derive measures that span several years. Alpha reliabilities for the academic self-image component of the self-attitudes composite average .70 and range from .60 in first grade to .84 in year 9. Reliabilities for the self-attitudes constructions themselves (based on 3 or 4 "items") are low in years 1, 2, and 4 (averaging .44), but average .65 over years 6-9 (alphas would be higher if calculated using the elements of the self-image scale separately).

The measures used to construct engagement attitude and engagement behavior scales change over time. The final constructions were guided, on an occasion-by-occasion basis, by extensive psychometric analyses that most years screened out numerous candidate items. Alpha reliabilities for the final engagement attitude scales range from .58 in 1st grade to .76 in year 7 (averaging .67); for engagement behaviors the range is from .74 in year 8 to .86 in year 6 (averaging .81).

Engagement behavior for the elementary years is constructed from math and English/reading teachers' ratings (on a scale of 1 to 6) of externalizing behaviors (7 items, with scores reflected, e.g., teases; fights) and of classroom adaptability (4 items, e.g., is enthusiastic; creative), report card ratings of student work habits (5 items, e.g., works independently; completes assignments), and report card ratings of conduct. The measure for year 6 combines report card information on absences, work habits, and conduct (the last only for children still in elementary school in year 6), student reports of time spent on homework, and teacher ratings of externalizing and adaptability behaviors. In years 8 and 9, questions asked of students and parents about problems at school (6 of students, 3 of parents; e.g., "I was sent to the office because I was misbehaving") and of students about cutting school and cutting class are added to the year 6 measures, but work habit ratings are dropped (these are not included on report cards in the middle grades) and conduct ratings are reported by teachers rather than taken from report cards. Items are standardized and averaged within years. Measures that span years are constructed as the average of these averages.

Engagement attitudes for years 1 and 2 are the average of four school satisfaction items, repeated fall and spring: "Would you say schoolwork is usually pretty dull or pretty interesting?" (Interesting = 1; Dull = 0); "If you could go to any school you wanted to, would you pick this school or some other school?" (this school = 1; other school = 0); "Do you like school a lot (coded 2), think it's just OK (coded 1), or not like it much at all (coded 0)?"; "Do you like [your teacher] a lot (coded 2); think he/she is just OK (coded 1), or not like him/her much at all (coded 0)?" In year 4, responses to six items asked of students about why they study (e.g., because it is interesting, because you like to solve hard problems) and three items about motivation for doing schoolwork (to get good grades, to please parents) are combined with the four satisfaction questions. After year 4, the teacher question from the satisfaction set and the study/ schoolwork questions were dropped, but other items were added. In years 6, 7, and 8 these include questions asked of teachers and parents about whether the study child "hates going to school." In year 9, seven self-ratings of the importance of school-related issues (e.g., finishing high school; doing well in school; being good at math) also are included. Response options that reflect favorably on school commitment are scored high throughout. To rate engagement attitudes, items are standardized and averaged within years. Measures that span years are constructed as the average of these averages.

Summary statistics for all measures are provided in Table Al for drop-outs and graduates separately. For comparison, information also is provided for members of the cohort whose graduation status could not be determined (N = 61). This last group, which in most respects resembles dropouts more than graduates, is excluded from all the comparisons presented in this paper.



This research was supported by Spencer Foundation Grant Nos. B-1517 ("Disengagement and Dropout: A Study of the Long-Term Process that Leads to Early Withdrawal from School") and 199800106 ("The Transition to Adulthood Among Urban Youth"), by the Office of Educational Research and Improvement Grant Nos. R306F970128 ("The Transition Out-of-School Among Urban Youth") and R117D4005 to the Johns Hopkins University Center for Research on the Education of Students Placed at Risk. We thank Carrie Horsey and Christine McRae for their able research assistance in earlier stages of this project, Jennifer Johnson for editorial suggestions, and Philip Kaufman for help deciphering census data on high school dropout.


Alexander, K. L, Entwisle, D. R., & Bedinger, S. D. (1994). When expectations work: Race and socioeconomic differences in school performance. Social Psychology Quarterly, 57(4), 283-299.

Alexander, K. L., Entwisle, D. R., & Dauber, S. L. (1993). First grade classroom behavior: Its short- and long-term consequences for school performance. Child Development, 64 (3), 801-814.

Alexander, K. L., Entwisle, D. R. & Dauber, S. L. (1994). On the success of failure: A reassessment of the effects of retention in the primary grades. Cambridge, MA: Cambridge University Press.

Alexander, K. L., Entwisle, D. R., & Horsey, C. (1997). From first grade forward: Early foundations of high school dropout. Sociology of Education, 70 (2), 87-107.

Alexander, K. L., Entwisle, D. R., & Kabbani, N. (1999, November 29-December 1). Grade retention, social promotion and "third way" alternatives. National Invitational Conference on Early Childhood Learning: Programs for a New Age. Alexandria, VA.

Alexander, K. L., Entwisle, D. R., & Legters, N. (1998, August). On the multiple faces of first grade tracking. Paper presented at the annual meeting of the American Sociological Association, San Francisco, GA.

Alsaker, F. D., & Olweus, D. (1992). Stability of global self-evaluations in early adolescence: A cohort longitudinal study. Journal of Research on Adolescence, 2 (2), 123-145.

American Federation of Teachers. (1997). Passing on failure: District promotion policies and practices. Washington, D.C.: Author.

Anderman, E. M., & Maehr, M. L. (1994). Motivation and schooling in the middle grades. Review of Educational Research, 64 (2), 287-309.

Anderson, D. K. (1993). Adolescent mothers drop out. American Sociological Review, 58 (5), 735-738

Annie E. Casey Foundation, (1997). Kids count data book: State profiles of child well-being. Baltimore, MD: Author.

Astone, N. M., & McLanahan, S. S. (1991). Family structure, parental practices and high school completion. American Sociological Review, 56 (3), 309-320.

Bachman, J. G. (1972). Dropouts are losers, says who? Today's Education, 61 (4), 27-31

Bandura, A. (1981). Self-referent thought: A developmental analysis of self-efficacy. In J. H. Flavell & L. Ross (Eds.), Social cognitive development: Frontiers and possible futures (pp. 200-239). Cambridge: Cambridge University Press.

Barrington, B. L., & Hendricks, B. (1989). Differentiating characteristics of high school graduates, dropouts, and nongraduates. Journal of Educational Research, 82 (6), 309-319.

Belsky, J., & MacKinnon, C. (1994). Transition to school: Developmental trajectories and school experiences. Early Education and Development, 5 (2), 106-119. Bomster, M. (1992, 18 September). City's dropout rate ranked 9th-worst in nation in 1990. The Baltimore Sun, pp. C1-4.

Bronfenbrenner, U. (1979). The ecology of human development: Experiments by nature and design. Cambridge, MA: Harvard University Press.

Brooks-Gunn, J., Guo, G., & Furstenberg Jr., F. F. (1993). Who drops out and who continues beyond high school? A 20 year follow-up of Black urban youth. Journal of Research on Adolescence, 3(3), 271-294

Bryk, A. S., & Thum, Y. M. (1989). The effects of high school organization on dropping out: An exploratory investigation. American Educational Research Journal, 26(3), 353-383.

Cairns, R, B., & Cairns, B. D, (1994). Lifelines and risks: Pathways of youth in our time. Cambridge, G.B.: Cambridge University Press.

Cameron, S. V., & Heckman, J. J. (1993). The nonequivalence of high school equivalents. Journal of Labor Economics, 11 (1), 1-47.

Catterall, J. S. (1998, Feb.), Risk and resilience in student transitions to high school. American Journal of Education, 106, 302-333.

Chen, X., Kaufman, P., & Frase, M (1997). Risk and resilience: The effects on dropping out of school. Paper presented at the annual meeting. American Educational Research Association, Chicago, IL.

Children's Defense Fund. (1992). City poverty data from 1990 census. Washington, DC: Author.

Committee on the Health and Safety Implications of Child Labor. (1998). Protecting youth at work. Washington, DC: National Academy Press.

Council of Great City Schools. (1994). National urban education goals: 1992-93 indicators report. Washington, DC: Author.

Crane, J. (1991). The epidemic theory of ghettos and neighborhood effects on dropping out and teenage childbearing. American Journal of Sociology, 96(5), 1226-1259.

Dauber, S. L., Alexander, K. L, & Entwisle, D. R. (1993). Characteristics of retainees and early precursors of retention in grade: Who is held back? Merrill-Palmer Quarterly, 59(3), 326-343.

Day, J., & Curry, A. (1998). Educational attainment in the United States: March 1997 [#P20-505]. Current Population Reports: Population Characteristics. Washington, DC: U.S. Bureau of the Census.

Dunn, J. (1988). Normative life events as risk factors in childhood. In M. Rutter (Ed.), Studies of psychosocial risk: The power of longitudinal data (pp. 227-244). New York: Cambridge University Press.

Eccles, J. S., Lord, S., & Midgley, C. (1991). What are we doing to early adolescents? The impact of educational contexts on early adolescents. American Journal of Education, 99 (4), 521-542.

Education Week. (1998, January). Quality counts, '98: An Education Week Pew Charitable Trust report on education in the 50 states. Education Week, 77(17).

Educational Testing Service. (1995). Dreams deferred: High school dropouts in the United States. Princeton, NJ: Educational Testing Service, Policy Information Service.

Elder Jr., G. H. (1994). Time, human agency, and social change: Perspectives on the life course. Social Psychology Quarterly, 57 (1), 4-14.

Elder Jr., G. H., Eccles, J. S., Ardelt, M., & Lord, S. (1995). Inner-city parents under economic pressure: Perspectives on the strategies of parenting. Journal of Marriage and the Family, 57 (3), 771-784.

Elliott, D, S. (1965). Delinquency, school attendance and dropout. Social Problems, 13(3), 307-314.

Elliott, D. S., Wilson, W. J., Huizinga, D., Sampson, R. J., Elliott, A., & Rankin, B. (1996). The effects of neighborhood disadvantage on adolescent development. Journal of Research in Crime and Delinquency, 33 (4), 389-426.

Ensminger, M. E., & Slusarcick, A. L. (1992, April). Paths to high school graduation or dropout: A longitudinal study of a first-grade cohort. Sociology of Education, 65 (2), 95-113.

Entwisle, D. R., &: Alexander, K. L. (1989). Early schooling as a "critical period" phenomenon. In K. Namboodiri & R. Corwin (Eds.), Sociology of education and socialization (pp. 27-55). Greenwich, CT: JAI Press.

Entwisle, D. R., & Alexander, K. L. (1993). Entry into schools: The beginning school transition and educational stratification in the United States. In Annual Review of Sociology (Vol. 19, pp. 401-423). Palo Alto, CA: Annual Reviews, Inc.

Entwisle, D. R., Alexander, K. L., & Olson, L. S. (1997). Children, schools and inequality. Boulder, CO: Westview Press.

Entwisle, D. R., Alexander, K. L., & Olson, L. S. (2000, August). Urban teenagers: Work and dropout. Paper presented at the annual meeting of the American Sociological Association, Washington, D.C.

Epstein, J. L. (1990). School and family connections: Theory, research, and implications for integrating sociologies of education and family. In D. linger & M. B. Sussman (Eds.), Families in community settings: Interdisciplinary perspectives (pp. 99-126). New York: Haworth Press.

Epstein, J. L. (1992). School and family partnerships. In M. Alkin (Ed.), Encyclopedia of educational research, sixth edition (pp. 11391151). New York: MacMillan.

Fagan, J., & Pabon, E. (1990). Contributions of delinquency and substance use to school dropout among inner-city youths. Youth and Society, 21(3), 306-354

Farkas, G. (1996). Human capital or cultural capital: Ethnicity and poverty groups in an urban school district. New York: Gruyter.

Featherman, D. L., & Stevens, G. (1982). A revised socioeconornic index of occupational status: Application in analysis of sex differences in attainment. In R. M. Mauser, D. Mechanic, A. O. Haller, & T. Hauser (Eds.), Social structure and behavior: Essays in honor of William Hamilton Sewell (pp. 141-182). New York: Academic Press.

Fine, M. (1990). Making controversy: Who's "at risk?". Journal of Urban and Cultural Studies, 1 (1), 55-68.

Finn Jr. C. E. (1987, Spring). The high school dropout puzzle. The Public Interest, 87, 3-22.

Finn Jr., C. E, (1989, Summer). Dropouts and grownups. The Public Interest, 96, 131-134.

Finn, J. D. (1989). Withdrawing from school. Review of Educational Research, 59 (2), 117-142.

Garnier, H. E., Stein, J. A., & Jacobs, J. K. (1997). The process of dropping out of high school: A 19-year perspective. American Educational Research Journal, 34(2), 395-419.

Goal 2 Work Group. (1993). Reaching the goals: Goal 2, high school completion. Washington, D.C. Office of Research and Improvement, U.S. Department of Education.

Grissom, J., & Shepard, L. A. (1989). Repeating and dropping out of school. In L. A. Shepard & M. L. Smith (Eds.), flunking grades: Research and policies on. retention, (pp. 34-63). London: Falmer Press.

Harter, S. (1982). The perceived competence scale for children. Child Development, 53 (1), 8797.

Haveman, R. H., & Wolfe, B. L. 1994). Succeeding generations: On the effects of investments in children. New York: Russell Sage Foundation.

Haveman, R., Wolfe, B., & Spaulding, J. (1991). Childhood events and circumstances influencing high school completion. Demography, 28 (1), 133-157.

House, E. (1989). Policy implications of retention research. In L. A. Shepard & M. L. Smith (Eds.), flunking grades: Research and policies on retention (pp. 202-213). London: Palmer Press.

Jessor, R. (1993). Successful adolescent development among youth in high-risk settings. American Psychologist, 48 (2), 117-126.

Jimerson, S. R. (1999). On the failure of failure: Examining the association between early grade retention and education and employment outcomes during late adolescence. Journal of School Psychology, 37(3), 243-272.

Jimerson, S. R., Anderson, G. E., & Whipple, A. D. (2000). Winning the battle and losing the war: The association of grade retention and dropping out of high school. Unpublished manuscript, University of California, Santa Barbara.

Kaplan, D. S., Peck, B. M., & Kaplan, H. B. (1997,July/August). Decomposing the academic failure-dropout relationship: A longitudinal analysis. Journal of Educational Research, 90, 331343.

Kasarda, J. D. (1993). Inner-city concentrated poverty and neighborhood distress: 1970 to 1990. Housing Policy Debate, 4 (3), 253-302.

Kaufman, P., Bradby, D., & Owings, J. (1992). National Education Longitudinal Study of 1988: Characteristics of at-risk students in NELS:88 [Contractor Report, NCES 92-042]. Washington, D.C.: U.S. Department of Education, Office of Educational Research and Improvement, National Center for Education Statistics.

Keogh, B. K. (1986). Temperament and schooling: Meaning of goodness of fit. In J. V. Lerner & R. M. Lerner (Eds.), Temperament and social interaction during infancy and childhood (pp. 89-108). San Francisco: Jossey-Bass.

Kerckhoff, A. C. (1976). The status attainment process: Socialization or allocation? Social Forces, 55(2), 368-381.

Kerckhoff, A. C. (1993). Diverging pathways: Social structure and career deflections. New York: Cambridge Press.

Lerner, J. V., Lerner, R. M., & Zabski, S. (1985). Temperament and elementary school children's actual and rated academic performance: A test of a "goodness-of-fit" model. Journal of Child Psychology and Psychiatry, 26 (1), 125-136.

Lerner, R. M., & Busch-Rossnagel, N. (Eds.). (1981). Individuals as producers of their development: A life-span perspective. New York: Academic Press.

Maryland State Department of Education. (1994). The fact book 1993-94: A statistical handbook. Baltimore, MD: Author.

McCord, J. (1994). Resilience as a dispositional qualify: Some methodological points. In M. C. Wang & E. W. Gordon (Eds.), Educational resilience in inner-city America: Challenges and prospects (pp. 109-118). Hillsdale, NJ: Lawrence Erlbaum Associates.

McDill, E. L., Natriello, G., & Pallas, A. M. (1986). A population at risk: Potential consequences of tougher school standards for student dropouts. American Journal of Education, 94 (2), 135-181.

McDonald, K. B., & LaVeist, T. A. (1999). Black advantage among similarly-situated urban poor. Unpublished manuscript, Department of Sociology, The Johns Hopkins University.

McLanahan, S. S., & Sandefur, G. (1994). Growing up with a single parent: What hurts, what helps. Cambridge, MA: Harvard.

Mensch, B. S., & Kandel, D. B. (1988, April). Dropping out of high school and drug involvement. Sociology of Education, 61, 95-113.

Mishel, L., & Bernstein, J. (1994). The state of working America. New York: Economic Policy Institute.

Morrow, G. (1986). Standardizing practices in the analysis of school dropouts. Teachers College Record, 87 (3), 342-355.

Mosher, R., & MacGowan, B. (1985). Assessing student engagement in secondary schools: Alternative conceptions, strategies of assessing, and instruments [Eric #ED272812]. University of Wisconsin.

Murnane, R. J., Willett, J. B., & Boudett, K. P. (1995). Do high school dropouts benefit from obtaining a GED? Educational Evaluation and Policy Analysis, 17 (2), 133-147.

Nagelkerke, N. (1991). A note on a general definition of the coefficient of determination. Biowtrika, 78 (3), 691-692.

National Education Goals Panel (NEGP). (1999). The National Education Goals report: Building a nation of learners, 1999, Washington, DC: U.S. Government Printing Office.

National Research Council. (1993). Losing generations: Adolescents in high-risk settings. Washington, DC: National Academy Press.

Natriello, G., & Dornbusch, S. M. (1983). Bringing behavior back in: The effects of student characteristics and behavior on the classroom behavior of teachers. American Educational Research Journal, 20 (1), 29-43.

Natriello, G., McDill, E. L., & Pallas, A. M. (1990). Schooling disadvantaged children: Racing against catastrophe. New York: Teachers College Press.

Pallas, A. M. (1984). The determinants of high school dropouts. Unpublished Ph.D. dissertation, Baltimore, MD: The Johns Hopkins University.

Pallas, A. M. (1987). School dropouts in the United States [Issue paper CS87-426, Center for Education Statistics, Office of Educational Research and Improvement]. U.S. Department of Education.

Pallas, A. M. (1993, Winter). Schooling in the course of human lives: The social context of education and the transition to adulthood in industrial society. Review of Educational Research, 63, 409-447.

Pallas, A. M., Natriello, G., & McDill, E. L. (1987). The high costs of high standards: School reform and dropouts. Urban Education, 22, 103-114.

Roderick, M. (1993). The path to dropping out: Evidence for intervention. Westport, CT: Auburn House.

Roderick, M. (1994). Grade retention and school dropout: Investigating the association. American Educational Research Journal, 31 (4), 729-759.

Roderick, M. (1995). School transitions and school dropout. Advances in Educational Policy, 1, 135-185.

Rumberger, R. W. (1983). Dropping out of high school: The influence of race, sex and family background. American Educational Research Journal, 20 (2), 199-220.

Rumberger, R. W. (1987). High school dropouts: A review of issues and evidence. Review of Educational Research, 57 (2), 101-121.

Rumberger, R. W. (1995). Dropping out of middle school: A multilevel analysis of students and schools. American Educational Research Journal, 32 (3), 583625.

Rumberger, R, W., & Larson, K. A. (1998, November). Student mobility and the increased risk of high school dropout. American Journal of Education, 107, 1-35.

Russell, C. S, (1980). Unscheduled parenthood: Transition to "parent" for the teenager. Journal of Social Issues, 36 (1), 45-63.

Rutter, M. (1987). Psychosocial resilience and protective mechanisms. American Journal of Ortho-psychiatry, 57 (3), 316-331.

Schneider, B., Stevenson, D., & Link, J. (1994, April). Social and cultural capital: Differences between students who leave school at different periods in their school careers. Paper presented at the annual meeting. American Educational Research Association, New Orleans.

Seginer, R. (1983). Parents' expectations and children's academic achievements: A literature review. Merrill-Palmer Quarterly, 29 (1), 1-23.

Shanahan, M. J., Elder Jr., G. H., Burchinal, M., & Conger, R. D. (1995, August). Adolescent earnings and relationships with parents: The work-family nexus in urban and rural ecologies. Paper presented at the annual meeting. American Sociological Association, Washington, D.C.

Shepard, L. A., & Smith, M. L. (1989). Introduction and overview. In L. A. Shepard & M. L. Smith (Eds.), Flunking grades: Research and policies on retention. London: Palmer Press.

Simmons, R. G., & Blyth, D. A. (1987). Moving into adolescence: The impact of pubertal change and school context. Hawthorn, NY: Aldine de Gruyter.

Smith, M. L., & Shepard, L. A. (1987). What doesn't work: Explaining policies of retention in the early grades. Phi Delta Kappan, 66 (2), 129-134.

Stipek, D., & Maclver, D. (1989). Developmental changes in children's assessment of intellectual competence. Child Development, 60 (6), 521-538.

Stroup, A. L., & Robins, L. N. (1972). Elementary school predictors of high school dropout among Black males. Sociology of Education, 45 (2), 212-222.

Temple, J. A., Reynolds, A. J., & Miedel, W. T. (1998). Can early intervention prevent high school dropout? Evidence from the Chicago Child-Parent Centers [Discussion Paper No. 1180-98]. Institute for Research on Poverty. Madison, WI: University of Wisconsin.

Toby, J. (1989a, Spring). Of dropouts and stay-ins: The Gershwin approach. The Public Interest, 96, 3-13.

Toby, J. (1989b, Summer). Coercion or choice. The Public Interest, 96, 134-136.

Toby, J., & Armor, D. J. (1992, Winter). Carrots or sticks for high school dropouts. The Public Interest, 106, 76-90.

Tyack, D., & Cuban, L. (1995). Tinkering toward Utopia: A century of public school reform. Cambridge, MA: Harvard University Press.

Tyack, D., & Tobin, W. (1994, Fall). The grammar of schooling: Why has it been so hard to change? American Educational Research Journal, 31, 453-479.

U. S. Bureau of the Census. (1992), Census of population and housing, 1990: 5% public use microdata samples U.S., Washington, D.C.: Author.

U.S. Department of Education. (1997). Dropout rates in the United States: 1995. NCES 97-473, by Marilyn M. McMillen, Phillip Kaufman, and Steve Klein. Washington, DC: U.S. Department of Education, National Center for Education Statistics.

U.S. Department of Education. (1998). Subsequent educational attainment of high school dropouts [NCES 98-085, by Jennifer Berktold, Sonya Geis, and Phillip Kaufman]. Washington, DC: U.S. Department of Education, National Center for Education Statistics.

U.S. Department of Education. (1999). Taking responsibility for ending social promotion. Washington, DC: Author.

Wagenaar, T. C. (1987). What do we know about dropping out of high school? In A. C. Kerckhoff (Ed.), Research on sociology of education and socialization, Vol. 7 (pp. 161-190). Greenwich, CT: JAI Press.

Wehlage, G. G., & Rutter, R. A. (1986). Dropping out: How much do schools contribute to the problem? Teachers College Record, 87 (3), 374-392.

Wehlage, G. G., Rutter, R. A., Smith, G., Lesko, N., & Fernandez, R. (1989). Reducing the risk: Schools as communities of support. Philadelphia: Falmer Press.

White, K. R. (1982). The relation between socioeconomic status and academic achievement. Psychological Bulletin, 91(3), 461-481.

Wilson W J. (1987). The truly disadvantaged: The inner city, the underclass, and public policy. Chicago: University of Chicago Press.

Wilson, W. J. (1996). When work disappears: The world of the new urban poor. New York: Knopf.

Zaslow, M. J., & Hayes, C. D. (1986). Sex differences hi children's responses to psychological stress: Toward a cross-context analysis. In M. E. Lamb, N. L. Brown, & B. Rogoff (Eds.), Advances in developmental psychology, vol 4 (pp. 285-337). Hillsdale, NJ: Erlbaum.

Zedlewski, S. R., & Alderson, D. (2001). Do families on welfare in the post-TANF era differ from their pre-TANF counterparts? [Discussion Paper 01-03]. Washington, D.C.: Urban Institute.

Zill, N. (1996). Family change and student achievement: What we have learned and what it means for schools. In A. Booth & J. F. Dunn (Eds.), Family-school Links: How do they affect educational outcomes? (pp. 139-174). Mahwah, NJ: Erlbaum.

KARL L. ALEXANDER is the John Dewey Professor of Sociology, The Johns Hopkins University. His areas of interest include sociology of education, social stratification, and patterns of personal and academic development over the life course. With Doris Entwisle and Susan Dauber, he presently is working on an update of their 1994 volume On the Success of Failure: A Reassessment of the Effects of Retention in the Primary Grades, to be re-issued by Cambridge University Press in 2002. Another on-going project, also using data from the Beginning School Study, is examining the cohort's educational pathways from a life course perspective. The present article is one product of that project.

DORIS R. ENTWISLE is Professor Emerita of Sociology, The Johns Hopkins University. Her main area of interest is the sociology of human development over the life course, with an emphasis on issues of inequality. With Karl Alexander and Linda Olson, her most recent book is Children, Schools, and Inequality, Westview Press, 1997. A former Guggenheim Fellow, in 1997 she received the Society of Research in Child Development Award for Distinguished Scientific Contributions to Child Development.

NADER S. KABBANI is an Economist with the Economic Research Service of the U.S. Department of Agriculture. He works on issues related to rural workforce development and food assistance/food security. He recently completed a Ph.D. in Economics at The Johns Hopkins University, where he wrote his dissertation on the effects of public sector training programs on the employment and earnings of non-participant workers.

Cite This Article as: Teachers College Record Volume 103 Number 5, 2001, p. 760-822
https://www.tcrecord.org ID Number: 10825, Date Accessed: 1/20/2022 5:35:54 AM

Purchase Reprint Rights for this article or review
Article Tools
Related Articles

Related Discussion
Post a Comment | Read All

About the Author
  • Karl Alexander
    Johns Hopkins University
    E-mail Author
    KARL L. ALEXANDER is the John Dewey Professor of Sociology, The Johns Hopkins University. His areas of interest include sociology of education, social stratification, and patterns of personal and academic development over the life course. With Doris Entwisle and Susan Dauber, he presently is working on an update of their 1994 volume On the Success of Failure: A Reassessment of the Effects of Retention in the Primary Grades, to be re-issued by Cambridge University Press in 2000. Another on-going project, also using data from the Beginning School Study, is examining the cohort's educational pathways from a life course perspective. The present article is one product of that project.
  • Doris Entwisle
    Johns Hopkins University
    E-mail Author
    DORIS R. ENTWISLE is Professor Emerita of Sociology, The Johns Hopkins University. Her main area of interest is the sociology of human development over the life course, with an emphasis on issues of inequality. With Karl Alexander and Linda Olson, her most recent book is Children, Schools, and Inequality Westview Press, 1997. A former Guggenheim Fellow, in 1997 she received the Society of Research in Child Development Award for Distinguished Scientific Contributions to Child Development.
  • Nader Kabbani
    Economic Research Service, USDA
    NADER S. KABBANI is an Economist with the Economic Research Service of the U.S. Department of Agriculture. He works on issues related to rural workforce development and food assistance/food security. He recently completed a Ph.D. in Economics at The Johns Hopkins University, where he wrote his dissertation on the effects of public sector training programs on the employment and earnings of non-participant workers.
Member Center
In Print
This Month's Issue