Texas Students’ College Expectations: Does High School Racial Composition Matter? Michelle Bellessa Frost Princeton University This article explores the association between school racial composition and students’ expectations to graduate from a four-year college. In addition to the individual characteristics of students that have been repeatedly shown to influence educational goals, the results indicate that both school socioeconomic level and achievement composition are related to expectations. The results also suggest the counterintuitive finding that in similar schools, students in schools with greater concentrations of minority students are more likely to expect to attain a four-year college degree than are students in schools with lower proportions of minority students.

A

lthough more than 50 years have passed since the landmark Brown v. Board of Education case, public schools remain largely segregated by race and ethnicity. Furthermore, the recent removal of court mandates to desegregate schools and challenges to other race-based assignment policies have led to an increase in segregation (Orfield 2001). Across the nation, 66 percent of blacks and 73 percent of Hispanics attend schools in which at least half the students are not white (Orfield and Yun 1999). In Texas, a state with a large and rapidly growing minority school-age population, school segregation is widespread. For example, in 2002–03, twothirds of Hispanic and half of black students attended schools with at least 70 percent minority students (Texas Education Agency, 2003).1 Thus, it is vital to examine the association between various forms of segregation and students’ educational outcomes. Among these important educational outcomes, the goal to attend and complete college is an essential first step toward the eventual attainment of a college degree (Hossler, Braxton, and Coppersmith 1989). Early work in the status attainment tradition established the importance of educational expectations for

academic attainment and other school outcomes, both within and across racial groups (Buchmann and Dalton 2002; Campbell 1983; Hanson 1994; Hao and Bonstead-Bruns 1998; Sewell, Haller, and Ohlendorf 1970; Sewell, Haller, and Portes 1969). Expanding on studies that investigated only the effects of individuallevel attributes on the formation of expectations, other sociologists turned to contextual models to search for school influences on individual students’ educational ambitions, including an examination of how desegregation efforts were affecting black students’ outcomes (Alexander and Eckland 1975; Alwin and Otto 1977; Hauser, Sewell, and Alwin 1976; Meyer 1970; St. John 1975). However, by the late 1970s, this research was largely abandoned because of the lack of results showing any real effects of schools, as well as a changing political climate (Thrupp 1997). The importance of understanding the impacts of continued school segregation in a state with widely heterogeneous schools, together with recent advances in multilevel statistical models and computing ability, merits a renewed examination of the association between the racial composition of schools and Texas high school students’ educational expectations.

Sociology of Education 2007, Vol. 80 (January): 43–66

43

44 Specifically, three possible scenarios linking school racial composition to educational expectations are considered. First, school racial composition could act merely as a proxy for individual characteristics that directly influence expectations to complete college, such as academic achievement. This scenario suggests that there are no school effects on expectations and is the default position of most previous research. Second, it is possible that school effects are indeed present, but that the distribution of students by race/ethnicity within a school reflects other school characteristics, such as the school’s socioeconomic composition, that are related to the level of educational expectations within the school. Finally, it is possible that something fundamental about the racial composition of schools is directly related to the educational goals of the schools’ students. To analyze the relative explanatory power of these three alternatives, I used multilevel models to answer the following research questions: Does the level of educational expectations vary among Texas high schools? What is the relationship between a school’s racial composition and educational expectations? Do students’ characteristics and other school characteristics explain why school racial composition is related to educational expectations? Is the relationship between school racial composition and educational expectations different for students of various races? To investigate these questions, the article proceeds as follows. Grounded in a literature review, three hypotheses are suggested that provide possible explanations for how and why school racial composition is associated with students’ expectations of graduating from college. Following a discussion of a survey suited to test these propositions and the unique research context of Texas, I outline an analytic plan. The results of the hierarchical logistic regression models suggest the counterintuitive finding that among schools with some similar characteristics, greater concentrations of minority students are associated with the increased likelihood that students will expect to attain a four-year college degree. The final section discusses the implications of these findings.

Frost

SCHOOL RACIAL COMPOSITION AND EDUCATIONAL GOALS A central hypothesis of this research is that school characteristics, specifically a school’s racial/ethnic composition, are independently associated with students’ educational expectations. A simple way to conceptualize school effects is to envision a single student attending a given school and then alternatively attending a different school—that is, one with a different racial composition. My analysis addresses how this hypothetical student’s educational expectations would change by attending a substitute school in the absence of other changes in the student’s personal and family circumstances. In this section, I examine evidence to formulate three separate but interrelated hypotheses on the association between the racial and ethnic makeup of a school and students’ educational expectations.

Relationship to Student Characteristics First, school racial composition could act, in part, as a proxy for the aggregation of students’ individual characteristics. Thus, differences in average educational expectations across schools may reflect compositional differences in student bodies that are associated with racial composition. For example, a high minority school often includes significant shares of students from poor families and those with prior educational deficiencies. On the whole, these students have lower levels of expectation to complete college, but this is partly a function of individual students’ characteristics that are associated with the racial composition of the school, not of any school characteristic in and of itself. In the strongest version of this hypothesis, educational expectations for two statistically identical students who attend vastly different schools would be similar. In other words, this hypothesis suggests that apparent school effects are attributable, in part, to correlated individual characteristics of students. To improve the chances of isolating actual school effects, then, individual-level correlates of educational expectations must first be accounted for. In

Texas Students’ College Expectations fact, most research on differences in students’ educational expectations has focused on individual-level explanations. The remainder of this section thus briefly reviews the literature on student factors that influence educational goals. Starting with the Wisconsin status attainment model, Sewell and others elaborated on Blau and Duncan’s (1967) basic stratification model by incorporating social psychological processes, including educational expectations, to explain educational and occupational attainment (Sewell, Haller, and Ohlendorf 1970; Sewell et al. 1969; Sewell and Shah 1968). Initially, they focused on how the encouragement and expectations of people who are important to students, including parents, teachers, and friends, explained variation in students’ educational expectations and aspirations (Sewell and Hauser 1992).2 In addition, Sewell et al. (1969) theorized about the influence of both academic performance and social origins on students’ educational expectations. In extensions of Sewell and his colleagues’ original model of educational expectations (Sewell et al. 1970; Sewell et al. 1969; Sewell and Shah 1968), others have focused more exclusively on how socioeconomic factors— including the direct and opportunity costs of attaining a higher education—shape educational ambitions (Jencks, Crouse, and Mueser 1983; Morgan 1998). For example, Morgan’s (1998) rational-choice framework concentrates on how differences in family resources to pay for education and the expected benefits of educational credentials shape educational expectations. Research on the influences of race and ethnicity on educational ambitions has found that minority students, especially blacks, have higher educational aspirations and expectations than do whites of comparable socioeconomic status (SES) and educational characteristics (Hirschman, Lee, and Emeka 2004; Kao and Tienda 1998; Qian and Blair 1999). Goldsmith (2004) linked higher minority educational aspirations to other ambitious goals as a representation of a generalized pattern of optimism among minority youths. Examinations of immigrants’ expectations have further confirmed that first- and second-generation immi-

45 grants are highly optimistic about their children’s futures, often expressing college ambitions for their offspring (Cheng and Starks 2002; Kao and Tienda 1995). Thus, at the student level, previous studies in the Wisconsin status attainment tradition have identified three types of covariates that explain differences in educational expectations: the influence of significant others, students’ socioeconomic background, and students’ academic ability and experiences. Furthermore, other lines of research have suggested that minority and immigrant status are associated with college expectations. These student characteristics are not randomly distributed across schools—students are often clustered in schools according to these factors. Therefore, the association of racial composition with college expectations is partly derived from differences in the composition of student bodies within schools. Specifically, Hypothesis 1 states that once relevant characteristics of students are accounted for, between-school differences in educational expectations will shrink. Furthermore, an initial negative relationship between the proportion of minority students and the average level of educational expectations in a school will lessen, weakening the link between the two.

Relationship to Other School Characteristics Racial composition may be correlated with other school characteristics that directly shape students’ educational expectations. I examine two possibilities: school socioeconomic status and average school achievement level. First, schools with high proportions of minorities are likely to have a high concentration of economically disadvantaged children. Thus, with a high correlation between racial composition and SES, the average educational expectations of high minority schools may be directly explained by schools’ socioeconomic composition. Findings in the 1960s and early 1970s, emanating from a flurry of research on the relationship between school context and educational expectations, showed that students from high-SES schools had higher aspirations than did those from low-SES schools, even

46 after their individual class status was controlled (Meyer 1970; Michael 1961; Turner 1964). Researchers theorized that the aspirations and attainment of a working-class student might be raised by informal contact with wealthier peers in middle-class schools because of exposure to middle-class values, aspirations, and knowledge (Thrupp 1997). Presumably, students in nonpoor schools hold cultural norms that value educational attainment and identify college attendance as desirable and even inevitable (Jencks and Mayer 1990). The Coleman report (Coleman et al. 1966) also advanced this idea, suggesting that the transmission of middle-class values from socioeconomically privileged white students to poor minority children was an important benefit of desegregation. The normative role of peers in providing and reinforcing standards is central to this perspective. These findings suggest that aside from a student’s own family SES, there are accrued benefits from attending a nonpoor school; thus, the school socioeconomic composition of peers can also serve to influence students’ educational goals. They also imply that the average educational expectations in schools with large shares of minorities are depressed because high-minority schools are also schools with low SES. By comparing schools with similar shares of poor students, the overall relationship between school minority composition and the average level of educational expectations should become less negative. In other words, when schools are equalized by school SES, the adjusted level of educational expectations in high-minority schools should increase. The second school characteristic of likely importance that I considered also emerged from early research. High-minority schools are also more likely than are predominately white schools to have lower levels of overall achievement. Perhaps, then, school racial composition is partly an indicator of scholastic attainment. As Marsh (1987) discussed, there are several plausible influences of average school ability on educational goals. For example, being an average-ability student in a high-achieving school may affect academic self-concept (1) positively because of widely held normative expectations within these

Frost schools to attend college; (2) negatively because the frame of reference for personal comparison is based on the performance of high-achieving students; or (3) not at all because either the achievement context does not influence students’ expectations or the two previous effects are offsetting, leaving no net effect. In addition to considering the socioeconomic composition of a school, Davis (1966) examined the role of the college achievement context on students’ career aspirations. He found that equally able students had higher career aspirations when they attended colleges in which the average ability level was lower and called this the frog-pond effect. In a scholastically challenging setting, with academic rewards distributed primarily within schools, students suffer by comparison to more able students. Thus, grades and class rank, which are normed to the school population, sort students and provide a relative measure of academic potential. Davis suggested that peers supply a frame of reference for academic self-examination and comparison, with a high-ability student body providing a more competitive scholastic environment. Furthermore, students in high-ability schools stand out less in comparison to their bright counterparts and receive less attention and encouragement to attend college from teachers and counselors (Meyer 1970). Marsh (1984, 1987, 1991) subsequently elaborated on this theory by conducting a large amount of research on what he termed “the bigfish–little-pond effect,” finding that students of comparable academic ability have lower academic self-concepts and educational expectations in higher-ability than in lowerability schools. In a related line of inquiry on how racial composition interacts with the achievement context, St. John (1975) reported that most studies she reviewed found an association between black students’ movement to desegregated white schools and a corresponding drop in these students’ educational ambitions. Marsh (1987) also found that black students in segregated schools had higher academic self-concepts than expected, given their academic performance, which he attributed to their lower-achieving frames of reference.

Texas Students’ College Expectations These findings are consistent with Davis’s (1966) frog-pond model: As black students entered desegregated schools with more skilled and competitive student bodies, not only did their academic self-perceptions drop, but so did their expectations of attending college, mainly because “realistic possibilities” (Falk 1978)3 were reconceptualized. Similarly, Portes and Hao (2004) and Portes and MacLeod (1996) reported a deleterious effect of highly competitive student bodies on disadvantaged students. These findings suggest that an association between school achievement and educational expectations differs by students’ race, a proposition that I tested directly in my study. The frog-pond view implies a negative relationship between average levels of school achievement and educational expectations and suggests that predominately minority schools have higher average expectations owing to the lower average achievement levels of their student bodies. Thus, the theorized effects of school SES and school achievement on educational expectations are in opposite directions. Equalizing schools by achievement composition should decrease the adjusted average level of expectation in high-minority schools if Davis’s (1966) frogpond theory is correct. In summary, Hypothesis 2, with its two variants, suggests that the association between school racial composition and educational expectations can be explained partly by the relationship between school racial composition and school SES and achievement composition.

Effect of Racial Composition As an alternative to both Hypotheses 1 and 2, it is possible that something fundamental about racial composition directly shapes school-level educational expectations. Supporters of affirmative action in higher education have based recent legal arguments on this notion: Racial and ethnic diversity are essential in higher education because of the important and socially desirable educational outcomes that they foster (Gurin et al. 2002). This view presumes that there is some optimal racial and ethnic composition of student bodies that enhances educational outcomes

47 for students of all racial and ethnic groups. Likewise, it is possible that the racial composition of a high school could independently influence high school students’ educational expectations, with similar effects for students of all racial groups. Goldsmith (2004) offered a possible explanation for an independent effect of racial composition. He found that black and Hispanic students have higher educational expectations when they attend schools with greater concentrations of minority students. For white students, he found a positive association between high educational expectations (graduate and professional school) and school racial composition only in the schools with high proportions of both minority students and minority teachers. Because black and Hispanic students have more positive beliefs about their future prospects and more pro-school attitudes, Goldsmith suggested, their concentration in schools improves the normative climate. It is likely that such an effect could influence all students in a school, regardless of their race. Other theories have suggested a differential impact of racial composition on minority students. Ogbu (1991) and Fordham and Ogbu (1986) theorized the development of a minority oppositional culture whose main characteristic is direct opposition to the majority culture. Presumably, black or Hispanic students who desire to excel academically and pursue higher education are betraying their group identities by “acting white” and hence are marginalized from their groups. Fordham and Ogbu (1986) and Ogbu (1995) did not explicitly discuss educational goals, but they implied that aspiring to graduate from college falls within the norms of the white dominant society: “At the social level, peer groups discourage their members . . . from adopting the attitudes . . . that enhance academic success. They oppose adopting appropriate academic attitudes and behaviors because they are considered ‘white’” (Fordham and Ogbu 1986:183). Furthermore, these authors claimed that oppositional attitudes and behaviors are more prevalent in schools with high proportions of minority students. Thus, their premise, although not directly stated, was that there is a statistical

48 interaction between students’ individual ethnicity and the racial composition of their schools. Specifically, when black and Hispanic students attend schools with critical masses of their own racial groups, the development of an oppositional culture leads to a devaluation of academic goals and lowers educational ambitions. Similarly, Farkas, Lleras, and Maczuga (2002) argued that fewer blacks profess optimistic attitudes in minority-segregated schools. In concrete terms, then, higher proportions of minority students should be associated with lower average expectations, but only for these minority students. Fryer and Torelli (2005) focused on the concept of acting white by analyzing racial differences in the relationship between high school grades and popularity. Whereas white students’ social status and popularity rise monotonically with increasing grades, the social status and popularity of high-achieving black and moderately high- and high-achieving Hispanic students decline. Most relevant for this study, Fryer and Torelli examined whether school racial composition changes this relationship between achievement and popularity. They found that the social costs for minority students’ achievement are the highest in schools with few minority students. Thus, in schools with substantial shares of minority students, these students feel less social pressure to curtail their academic achievement. This finding, suggesting that acting white is more common in schools with few minority students, contradicts the thesis that larger groups of minority students foster an oppositional culture (Fordham and Ogbu 1986). Fryer and Torelli’s findings imply that higher levels of expectation to graduate from college may be associated with high-minority schools because the social costs to minority students holding them are lower than for minority students in white schools. In a similar vein, other studies have indicated that the self-esteem and academic selfperception of minority students are lower in racially balanced schools than in high-minority schools because of unfavorable comparisons to white students and social dissonance among racial groups (Drury 1980; Gray-Little and Carels 1997; St. John 1975). Thus, if academic self-image and self-esteem partly

Frost shape educational expectations, then minority students who attend schools in which their own race is more highly represented will have higher expectations of attending and graduating from college. This hypothesis is reminiscent of Davis’s (1966) frog-pond hypothesis. Thus, it suggests that minority students may have more ambitious educational goals when they are clustered together. Although I could not directly test these theories, I shed some light on their validity by an examination of the empirical evidence of an independent influence of racial composition on expectations, as well a differential impact of racial composition on students of different races.

DATA The data for this study were taken primarily from the Texas Higher Education Opportunity Project (THEOP),4 an ongoing longitudinal study designed to understand the consequences of Texas’s replacement of a race-sensitive college admissions regime with a percentage plan for minority students’ college enrollment.5 The sampling design of this study made the study of school effects possible. The survey was based on a stratified random sample of 108 Texas public schools with a student body consisting of at least 10 enrolled seniors and was further stratified on the basis of metropolitan-area status and the racial/ethnic composition of schools. Of the eligible schools that were selected, 93 percent participated in the study, for a total of 98 Texas high schools. The sample is representative of the student enrollment of Texas public high schools in spring 2002. During spring 2002, baseline data were collected in the sampled high schools from 19,969 sophomores and 13,803 seniors using a paper-and-pencil survey. A random sample of the original senior cohort is being followed as these students continue from high school to college and other post-high school activities. The first follow-up of the senior cohort took place during spring and summer 2003, and the second follow-up took place during spring and summer 2006. In addition, the sophomore cohort were reinterviewed during their senior year, in spring 2004. For the pur-

Texas Students’ College Expectations poses of this study, I used only the baseline data from the senior cohort. The survey asked the respondents about their grades and class rank; course taking; extracurricular activities; and knowledge and perceptions of college admissions, including the top 10-percent law (see note 5). The seniors were also asked about their future plans, college applications, and preferences for universities. Essential for the purposes of this study, the students were asked how much education they would like to attain; a series of questions about their educational experiences; and how much encouragement they received from teachers, counselors, and parents regarding college attendance, as well as standard background questions. To assess the contextual determinants of educational expectations, I merged data from the U.S. Department of Education and the Texas Education Agency (TEA) with the individual-level data files using a school identifier. The Common Core of Data (CCD), a program of the U.S. Department of Education, is an annual collection of school- and district-level information that I used to obtain information on school racial composition, the key independent variable, as well as on school poverty, measuring the proportion of students who are eligible for free and reduced-price lunches. The TEA provides school-level information about a variety of characteristics, including high school achievement and test scores and the size of schools. Finally, by aggregating survey data by school, I generated an additional contextual variable, the educational level of parents in a school. Although I used only the survey data from the seniors for statistical analyses, I aggregated senior and sophomore data by school to obtain greater precision in this contextual measure. I imposed two constraints on the study sample. First, I omitted all cases that lacked valid responses in the main dependent variable, educational expectations. Doing so excluded 11 percent of the sample cases. Second, because of the individual variable measuring racial status, I included only students who identified themselves as white, black, Hispanic, or Asian. Other ethnic groups had small sample sizes, so I omitted all students who reported that they were Native

49 American, “other,” or multiracial. Only a small proportion of the overall sample (1.7 percent ) was dropped for this reason. Thus, the final study sample consisted of 12,071 high school seniors who were clustered in 96 schools (8 to 672 students per school). To address the other individual missing data in the independent variables (67 percent of the students had no missing data, and 26 percent were missing only two pieces of data), I used predictive mean matching, a form of hot-deck imputation, to impute an observed value that is closest to the predicted value (Landerman, Land, and Pieper 1997; Little 1988).6 In general, hot-deck imputation uses a matching algorithm to replace missing values from a “nearest-neighbor” donor with the values of nonmissing variables. This strategy assumes that the data are missing at random, in that missing cases do not depend on the outcomes of interest or unobserved covariates, but may depend on observed covariates (Little and Rubin 1987). This assumption, unfortunately, is not verifiable from the data. I used this imputation method to preserve enough students’ responses by school to enable a multilevel analysis. Further advantages of this method include the fact that only eligible values of the missing variable are imputed (so, for example, grade point average, GPA, would always be assigned a value of 0 to 4) and that it is less sensitive to model misspecification than are other methods (Little 1988).

Texas as a Research Context Texas, like other immigrant-receiving states, has witnessed recent rapid change in its racial and ethnic diversity, especially among the school-age population. This situation makes Texas a propitious location to study school effects. Between 1980 and 2000, as the total school-age population increased, the white school-age population declined as a proportion of the total from 56 percent to 43 percent , while the Hispanic share of the total rose from 28 percent to 40 percent (Murdock et al. 1997). This trend is expected to continue. If the school-age population was distributed equally throughout the state by race, this demography would not give one reason

50 to investigate the effects of school racial composition on educational outcomes. However, this is not the case: The state is racially segregated by region, within cities, and by school district. Figure 1 shows how the racial distribution of students within schools varies by race across Texas. The bar farthest on the left shows the 2002 Texas statewide average school racial composition. As a trait of individual students, these differences indicate that, on average, schools are composed of about 50 percent white students, 36 percent Hispanic students, 12 percent black students, and 2 percent Asian students. Each additional bar to the right represents the average school racial composition for each racial subgroup. Because of housing patterns, residential segregation, and geographic clustering, students of different races attend schools with different racial compositions. Thus, Figure 1 shows that Hispanic students attend, on average, schools that are more than 60 percent Hispanic, 29 percent white, 8 percent black, and 2 percent Asian. Yet, on average, white students attend schools with two-thirds white students and only 20 percent Hispanics. All in all, a clear pattern of racial clustering within

Frost schools emerges. Given this racial separation of students in Texas, it is important to understand how school racial composition is associated with educational outcomes and, specifically, educational expectations.

MEASURES AND DESCRIPTIVE RESULTS Educational expectations were obtained from a closed-choice question that asked students how much education they realistically expected to attain. The choices ranged from high school graduation to attaining a Ph.D., MD, or other professional degree. I converted these responses to measure whether the students expected to attain a four-year college degree. I used a dichotomous measure for two reasons. First, substantively, a four-year college degree is an important social and labor market differentiator and is linked with higher levels of employment and wages, as well as health and satisfaction benefits (Arkes 1999; Perna 2000; Topel 2004). For these reasons, I was most concerned with whether students aspire to complete a bachelor’s degree,

Figure 1. School Racial Composition, by Individual Students’ Race. Source: 2002 THEOP

Texas Students’ College Expectations as a precursor to their actual attainment, rather than any other (greater or lesser) amount of education. My second reason was methodological. A continuous measure of total years of expected education is an alternative to the dichotomous measure of expectations that I used and has been used in other research on educational expectations (Goyette and Xie 1999; Morgan 1996). But although it is relatively clear what a year of education signifies at the elementary or high school level, this distinction blurs at the college level. Ranging from part-time students, who take more than four years to complete their degrees, to students who enter college with multiple Advanced Placement (AP) credits, the time to obtain a bachelor’s degree varies widely. I was not concerned with the total years of school expected (or attained), but with whether the students aspired to complete a bachelor’s degree. An additional alternate specification of my dependent variable is an ordered categorical variable, analyzed with ordered logit models, which would produce unnecessarily complicated results. Since students’ aspiration of graduating with a bachelor’s degree is the most important distinction for the purposes of this article, I chose to dichotomize expectations for this analysis. This measurement of educational expectations has been used successfully by others (Hirschman et al. 2004; Kao and Tienda 1998). Table 1 shows that, on average, 69 percent of students expect to attain a four-year college degree. Overall the expectations are high, given current statistics suggesting that only 32 percent of recent high school graduates actually attained a four-year college degree (according to my calculations from the March 2002 Current Population Survey) . Because I focused on high school seniors, high school dropouts were excluded from the sample, which likely inflated somewhat the proportion of the senior cohort who expected to complete college. At the same time, other studies have reported similarly high levels of expectations among high school students to complete a bachelor’s degree (Hirschman et al. 2004; Kao and Tienda 1998; Qian and Blair 1999). Table 1 also presents the differences in average expectation

51 by the racial composition of schools. Here, low- and high-minority schools are defined as schools in the first and last quartile, respectively, of the school proportion of total nonAsian minority students. Thus, 60 percent of students in high-minority schools, compared to 70 percent of students in low-minority schools, expect to attain a college degree. The key independent variables in my analysis measure school racial/ethnic composition, indicated by the proportion of black students and the proportion of Hispanic students within each school.7 As I noted earlier, the average school in the sample contained about 36 percent Hispanic students, 12 percent black students, and 2 percent Asian students. In the schools with the lowest shares of minority students, almost 90 percent of the students, on average, were white, compared to only 8 percent in schools with high proportions of minorities. One measure of the school’s SES is the school average of attained parental education, which captures differences in both parents’ income and values about education. Presumably, a greater proportion of collegeeducated parents within a school will translate to strong school support for higher education. A critical mass of college-educated parents can influence the quality of instruction because these parents are more likely to be a collective agent for high standards and an academic curriculum in a school. Because I was interested in students’ college expectations, I measured parental education at the school level as the proportion of parents who obtained at least a four-year college degree. The school average for the sample was onethird. However, it varied dramatically between high- and low-minority schools. In low-minority schools, 44 percent of parents attained at least a four-year college degree, whereas in high-minority schools, only 21 percent did so. As an additional measure of school SES, I used the proportion of students qualifying for free and reduced-price lunches. Only 23 percent of students in low-minority schools versus 61 percent in high-minority schools were poor enough to qualify for federal lunch assistance, while the sample average was 37 percent. A comparison of high- and low-minori-

52

Frost

Table 1. Descriptive Statistics, by School Racial Composition (standard deviations in parentheses)

Variables

Average

Dependent Variable Expect to complete a four-year college degree

0.686

School-Level Variables Racial/ethnic composition Percentage of black students

0.614

[24]

[24]

12.4 (17.8) 32.0 (30.1) 53.1 (31.0)

2.9 (04.1) 7.9 (5.3) 86.9 (6.7)

18.6 (29.2) 72.2 (29.3) 8.0 (6.5)

34.3 (17.4) 36.6 (22.1)

43.9 (19.6) 22.7 (15.3)

21.4 (9.8) 61.0 (15.7)

51.7 (18.3) 1305 (1,123)

58.2 (19.7) 804 (1,093)

37.0 (12.8) 1818 (977)

Percentage of white students

Student-Level Variables Race/ethnicity and immigration White Black Hispanic Asian Foreign born

0.701

[96]

Percentage of Hispanic students

Socioeconomic status Percentage of parents with four-year college degree Percentage qualifying for free or reduced-priced lunches Achievement Percentage of students meeting state standards, all tests School size

Low-Minority High-Minority Schoolsa Schoolsb

[12,071]

Parental socioeconomic status Parent a college graduate Parents own their home Educational experiences and background Enrolled in a college-prep track GPA, most recent grading period Attitudes toward education Number of Advanced Placement courses Significant others' influence Parents encourage college attendance Teachers encourage college attendance

[2,790]

Source of Data

Survey

CCD CCD CCD

Survey CCD

TEA TEA

[2,907]

0.518 0.108 0.334 0.040 0.115

0.789 0.047 0.114 0.050 0.083

0.061 0.149 0.780 0.010 0.200

Survey Survey Survey Survey Survey

0.404 0.828

0.528 0.862

0.202 0.782

Survey Survey

0.645 3.19 (.017) 2.71 (.011) 1.1 (.055)

0.604 3.31 (.036) 2.66 (.016) 1.18 (.124)

0.675 3.03 (.034) 2.86 (.017) 1.11 (.088)

Survey Survey Survey

0.943 0.866

0.950 0.862

0.929 0.897

Survey Survey

Survey

Note: School variables are measured as percentages for ease of interpretation at the multivariate stage of analysis and are labeled as such in this table. Other variables are measured as proportions. Student-level statistics are weighted. a Low-minority schools are schools in the first quartile of the distribution of total black and Hispanic students. b High-minority schools are schools in the last quartile of the distribution of total black and Hispanic students.

Texas Students’ College Expectations ty schools showed that schools with many minority students were, as expected, poorer. I measured school achievement using the proportion of the schools’ 10th graders who met state standards in all required state testing areas on the Texas Assessment of Knowledge and Skills (which includes English language arts, math, science, and social studies) reported by the TEA. Because of state requirements, seniors are not tested, and data are available only for 10th graders. Although my sample consisted of 12th graders, this measure reflects the achievement composition of the schools they attended. Because lower-ability students are more likely to withdraw from school between the 10th and 12th grades, measurement of academic achievement in the 10th grade provides a better index of school achievement than does a comparable measure of 12th graders. On average, about 50 percent of the sampled students passed the Texas state tests in all required testing areas. As expected, the lowminority schools had higher levels of achievement than did the high-minority schools, with 58 percent compared with 37 percent, respectively, of the students meeting the state’s standards in all tested areas. To isolate the impacts of school-level characteristics on educational expectations, I included several categories of individual-level covariates in the multilevel analysis that are well established correlates of educational expectations, including students’ socioeconomic background, students’ educational background and experience, influence of significant others, and students’ racial and immigration status.8 All the student characteristics were taken from the THEOP 2000 Wave 1 survey of seniors. Table 1 shows that approximately half the Texas high school students are white and one-third are Hispanic, consistent with state reports. The remainder of the student population is divided between blacks (about 11 percent) and Asians (about 4 percent). Thus, the school-age population in Texas is about equally split between white and minority students. Table 1 also shows how the other independent variables vary according to the racial composition of schools. Higher shares of parents of students in low-minority schools com-

53 pleted college. For the most part, students from higher-minority schools are disadvantaged with respect to their educational experiences and background, but there are some notable exceptions. Students in high-minority schools are more likely to be enrolled in a college preparatory curriculum, according to their self-reports.9 These students also report more positive attitudes toward school than do their counterparts in schools with few minority students. As anticipated, higher proportions of students in high-minority schools are immigrants.

ESTIMATING SCHOOL EFFECTS ON EDUCATIONAL GOALS I used a multilevel modeling strategy to estimate how school racial composition and other school characteristics are related to educational expectations. Multilevel regression has increasingly been used to estimate class, school, and other contextual effects within multiple-level data structures, especially when individual observations are clustered within these higher-level units. Because students clustered within schools are not statistically independent observations, traditional linear and binary regression models produce downwardly biased standard errors, which can lead to incorrect inferences about the statistical and substantive importance of schoolcontext variables (Guo and Zhao 2000). However, multilevel models explicitly adjust for the nonindependence of sample members who share a context, such as a school (Raudenbush and Bryk 2002). Two general types of multilevel models are used to estimate school effects on a specific student-level outcome: random-slope and random-intercept models. With random-slope models, the effects of individual students’ attributes are allowed to vary across schools; thus, these models explicitly test for differences in slopes across schools. Random-intercept models examine questions about differences in the level of a student outcome across schools. When the outcome variable differs across schools, they can be used to identify which school characteristics are responsible for differ-

54 ent outcomes. Because my research focused on the extent to which school racial composition and other explanatory variables explain the variance in educational expectations across schools, I fixed all the slopes10 and estimated random-intercept models to answer the following questions: Does the level of educational expectations vary across schools? Do student and school characteristics explain the relationship between school racial composition and educational expectations? Do student characteristics, school racial composition, and other school-level characteristics account for variation in educational expectations among schools? Specifically, I used HLM, a hierarchical linear modeling statistical computing program, to estimate hierarchical logistic random-intercept models.11 To begin, I first examine a simple histogram of the proportion of students by school who expect to complete a four-year college degree. This is the school-level distribution of the dependent variable and provides a visual description of how much variation in educational expectations exists between schools. Then I formally test whether there is significant between-school variation in the expectation of completing a college education by estimating a simple unconditional model without any predictors. This model provides a baseline estimate of τ00, the between-school variance, and its standard error, information to analyze the underlying premise of this article that educational expectations do, in fact, vary among schools. The initial estimate of τ00 serves as a point of comparison for subsequent models. Thus, I examine how much τ00 is reduced—or how much of the between-school variance in educational expectations is explained—by the inclusion of specific groups of variables. Subsequently, I model the effect of school racial/ethnic composition on educational expectations before any controls are included, providing an estimate of the observed relationship between them. Because of the continuous measure of school racial composition, coefficients represent the estimated change in the log odds of expecting a fouryear college degree that are associated with a unit change in the percentage of Hispanic or black students in a school.

Frost Next, I introduce student characteristics into my models to examine how much of the between-school variance in educational expectations is due to school differences in student composition and how the association between school racial composition and educational expectations changes with these additions, as detailed in Hypothesis 1. I also examine in brief how students’ individuallevel characteristics affect educational expectations. In these regressions, I centered the values of each student variable on the grand mean by subtracting the state mean from each one. Although grand mean centering does not change the estimated values of the regression slopes, it does shift the value of the intercept. With this transformation, the intercept provides an estimate of a school’s mean level of college expectation with the statewide average on all included variables. Because grand mean centering provides more stable estimates (Raudenbush and Bryk 2002), I used this transformation throughout the analysis for all the independent variables (including the interaction terms). In the next step of analysis, I test Hypothesis 2 by adding the other school characteristics to a model that includes both student racial composition and individual student characteristics. Because a school’s SES and achievement could be related to school racial composition and students’ expectation of completing a college degree, I use this model to examine how the estimates of the school racial composition variables change in the presence of these additional school characteristics and to observe if the other school variables exert an independent effect on educational expectations. I also include school size as a control variable. With this model, I can also evaluate Hypothesis 3 by looking for evidence of independent effects of the school’s minority composition on seniors’ college expectations after all relevant student and school characteristics are considered. I also include interaction effects between the races of individual students and their schools’ racial composition to ascertain whether the effect of racial composition on educational expectations varies by students’ race.

Texas Students’ College Expectations

RESULTS The central question of my analysis is how school racial composition is related to high school students’ expectations of completing a four-year college degree. Beginning my analysis, I examine between-school differences in educational expectations with a histogram shown in Figure 2. These data are reported at the school level with a total sample of 96, so the x-axis measures the school average of the proportion of students who expect to attain a four-year college degree. For example, there are 23 schools in the sample in which 46 percent to 55 percent of the students expect to complete college, as shown by the fourth bar. For seniors, the school average proportion of students who expect to complete a four-year college degree ranges from 20 percent to 90 percent, with a clustering of schools falling between 50 percent and 80 percent. This figure demonstrates substantial heterogeneity of educational expectations at the school level across Texas high schools. Substantively, a school in which only 30 percent of the students expect to complete college is different from one in which 80 percent expect to complete college. To test formally whether there is significant

55 variation in the level of educational expectations among schools, as suggested in Figure 2, I test an unconditional model, which estimates only τ00, the between-school variance component, and an intercept describing the mean level of educational expectations. These results are presented in Model 1 of Table 2. Although hierarchical logistic models do not allow comparisons of the between- and within-school variance components of dependent variables, the initial estimate of the betweenschool variance component, or τ00, provides a benchmark for examining the sources of reduction in the between-school variance. Because the estimates of τ00 and its standard error confirm that the level of expectation of completing a four-year college degree varies significantly among schools,12 I next turn to a multivariate analysis to examine the sources of this variation and analyze the association between school racial composition and educational expectations. Model 2 in Table 2 shows the effects of the percentage of black and Hispanic students on plans to complete college before any other covariates are added. Although the distribution of black students is not related to educational expectations, the proportion of Hispanic students in a school is negatively associated with the proportion of students in

Figure 2. School Average Proportion of Seniors Who Expect to Graduate from a Four-Year College

56

Frost

Table 2. Hierarchical Logistic Regressions of College Expectations (Laplace estimation; log odds in parentheses) Model Intercept School Level Percentage of black students

1 0.740*** (0.071)

Percentage of Hispanic students School proportion of students who qualify for free or reduced-price lunches School proportion of parents with a college degree

2

3

4

5

6

0.759*** (0.071)

0.935*** (0.076)

0.855*** (0.056)

0.856*** (0.071)

0.853*** (0.058)

-0.005 (0.005) -0.005* (0.002)

-0.003 (0.005) 0.000 (0.003)

0.007+ (0.004) 0.012*** (0.003) -0.003 (0.005) 0.026*** (0.005) 0.010* (0.004) 0.006 (0.007)

0.005 (0.004) 0.012** (0.004) -0.002 (0.005) 0.026*** (0.006) 0.010** (0.004) 0.007 (0.006)

0.005 (0.004) 0.012*** (0.003) -0.003

(0.005)

Percentage of students meeting state standards, all tests School size

0.025*** (0.005) 0.010** (0.004) 0.007 (0.007)

Interaction Terms Percentage Black*Black 0.007 (0.007) -0.005 (0.007) 0.006 (0.005) -0.001 (0.003)

Percentage Black*Hispanic Percentage Hispanic*Black Percentage Hispanic*Hispanic Percentage of students meeting state standards *Black Percentage students meeting state standards *Hispanic

-0.012* (0.006) 0.002 (0.006)

Individual Level Black Hispanic Asian Foreign born Parental Socioeconomic Status Parent a college graduate Home ownership Educational Experiences and Background Enrolled in college prep track GPA, most recent grading period

Number of AP courses Attitudes about education Significant Others' Influence Parents encourage college attendance Teachers encourage college attendance τ00

0.341 (0.067)

0.300 (0.061)

0.143* (0.062) -0.313*** (0.073) 0.411** (0.178) -0.359*** (0.093)

0.131+ (0.068) -0.321*** (0.078) 0.381** (0.145) -0.361*** (0.092)

0.016 (0.098) -0.259** (0.098) 0.372* (0.152) -0.351*** (0.088)

0.131+ (0.074) -0.3 (0.084)*** 0.387 (0.144)** -0.357 (0.091)***

0.660*** (0.060) 0.207* (0.086)

0.626*** (0.063) 0.209* (0.089)

0.627*** (0.064) 0.217* (0.093)

0.629 (0.064)*** 0.211 (0.091)*

0.791*** (0.073) 0.596*** (0.041)

0.776*** (0.076) 0.603*** (0.044)

0.779*** (0.084) 0.602*** (0.047)

0.776 (0.076)*** 0.604 (0.046)***

0.245*** (0.018) 0.341*** (0.045)

0.240*** (0.019) 0.346*** (0.047)

0.239*** (0.019) 0.347*** (0.050)

0.239 (0.019)*** 0.346 (0.049)***

0.988*** (0.113) 0.432*** (0.069)

0.984*** (0.113) 0.435*** (0.076)

0.982*** (0.120) 0.437*** (0.077)

0.985 (0.116)*** 0.434 (0.079)***

0.233 (0.066)

0.084 (0.029)

0.083 (0.030)

0.083 (0.029)

Source: 2002 Texas Higher Educational Opportunity Study, 12,526 students clustered in 96 schools. *p < .05, **p < .01, ***p < .001.

Texas Students’ College Expectations a school who expect to complete a four-year college degree. In substantive terms, a 10 percent increase in the share of Hispanic students lowers the school odds of expecting a four-year college degree by 5 percent (1-exp [-.005*10]). Model 3 in Table 2 shows results from a model testing Hypothesis 1 to determine if racial composition acts as a proxy for the school composition of students’ characteristics. After individual SES, educational experiences and achievement, significant others’ influence, race, and immigration status are adjusted for, the negative coefficient of the proportion of Hispanic students on educational expectations disappears, suggesting that students in schools with increased proportions of Hispanic students have, on average, fewer of the resources and educational experiences that are associated with expectations of graduating from college. Furthermore, the 22 percent reduction in τ00 from Model 2 to Model 3 suggests that the aggregation of student-level characteristics related to educational expectations at the school level is responsible for a substantial portion of the between-school variance in educational expectations. The weakened relationship between school racial composition and educational goals lends some support to Hypothesis 1, which proposes that school racial composition acts as an indicator of individual students’ characteristics that influence students’ educational expectations. However, an examination of τ00 and its standard error for Model 3 shows that the between-school variance in educational expectations is still significant and not entirely accounted for by differences in school composition. For the most part, the student-level results from Model 3 are consistent with previous research and in the expected directions. Parental SES, students’ educational background and ability, and significant others’ influences are strongly and positively associated with students’ educational expectations.13 Individual immigration status is associated with lower expectations, even when SES and educational background are adjusted for. Of all the racial groups, Asian students have the highest likelihood of expecting to complete a bachelor’s degree after SES and educational

57 background are adjusted for. Black students are also more likely to have college expectations than are similar white students, in line with other research findings (Goldsmith 2004; Hirschman et al. 2004; Qian and Blair 1999). To evaluate Hypothesis 2, that racial composition is a proxy for other school characteristics that are associated with educational expectations, I examine a model that adds school characteristics to Model 3. These results are presented in column 4 of Table 2. This model shows that the presence of more college-educated parents and higher average achievement in high schools are strongly associated with students’ college expectations. Specifically, when the percentage of parents with a four-year college degree increases by 5 percent, the odds of expecting a college degree increase by 15 percent (exp[5*.026]). This effect is in addition to the significant, positive effect at the student level of parental college completion on expectations. Furthermore, the effect on college expectations of the school average pass rate on the state standardized tests is positive and significant: A 10 percent increase in a school’s pass rate is associated with an 11 percent increase in the odds of students expecting to graduate with a four-year college degree. Unfortunately, comparable data on achievement are not available at the student level, and although I adjusted for other student factors that are associated with achievement, including students’ GPA, I cannot eliminate the possibility that this school variable simply reflects variation in individual achievement. At best, it is likely that the effect of school achievement that is shown here is overstated. Although the lack of data on achievement at the student level limits the conclusions I can draw from the finding, the positive effect of achievement on expectations is opposite the prediction of Davis’s (1966) frog-pond model. In fact, as school average scholastic achievement increases, individual college expectations increase as well. Once schools are equalized by their socioeconomic and achievement composition, the proportion of both black and Hispanic students becomes positively and significantly

58 related to college expectations, although the coefficient of the proportion of black students is borderline significant. Substantively, among schools of comparable SES and achievement, a 10 percent increase in Hispanic students is associated with a 13 percent increase in the odds of expecting to complete a four-year college degree, compared to a 7 percent increase in the odds for a 10 percent increase in black students. As predicted by the first part of Hypothesis 2, higher levels of expectation to graduate with a bachelor’s degree are found in schools with both higher SES and lower proportions of minority students, so adjustment for the socioeconomic composition of schools results in a positive association between the proportion of minority students and overall levels of expectations. However, contrary to the predictions of the second part of Hypothesis 2, college expectations are positively associated with higher levels of school achievement. Since high-minority schools are more likely to have low average levels of scholastic achievement, higher proportions of minority students result in increased levels of expectations once adjustment is made for the achievement composition of schools. The statistical significance of the variables measuring the school racial composition after student and school-level characteristics are adjusted for suggests an independent effect of school racial composition, consistent with Hypothesis 3. The between-school variance for the full model of .084 represents a 64 percent decrease in τ00 from the previous model that controlled only for individual-level characteristics and school racial composition.14 Thus, a substantial proportion of the schoollevel differences in educational expectations is accounted for by the school variables included in my analysis. In another analysis not reported here, I found that this reduction is due largely to the addition of the school composition of college-educated parents. Thus, a primary reason for the betweenschool heterogeneity of educational expectations is derived from school differences in the average level of parents’ education. However, a comparison of τ00 and its standard error in Model 4 still shows that the variation in educational expectations between schools remains statistically significant, indicating

Frost that there are other undetermined factors that influence differences between schools in students’ educational expectations. How robust is the finding that students who attend schools with large proportions of minority students are more likely to expect to graduate from a four-year college? The previously reported results suggest a monotonic relationship between the proportion of minority students and educational expectations. As an alternative, it is possible that the results are dependent on a critical mass of minority students and that schools (adjusted for student and school characteristics) with any greater proportion of black and Hispanic students do not have higher levels of educational expectations. I tested various specifications of the variables measuring school racial composition to examine whether the same results occurred in predominately white schools, mixed racial schools, and predominately minority schools. These analyses (based on dichotomous measures to represent segments of the school minority composition) reaffirmed my basic result: All other things being equal, college expectations are higher among students who attend schools with greater shares of minority students. These supplemental analyses,15 however, shed some additional light on the relationship between school racial composition and educational expectations. The coefficient for the proportion of black students is borderline significant, as shown in Table 2, Model 4, and I examined if there was a particular segment of the black distribution responsible for this finding. The results indicated that increased odds of expecting to graduate with a four-year college degree are found only in the 10 percent of schools with the highest proportion of black students—at least 40 percent of the total school population. This finding contrasts with the findings for the Hispanic distribution of students in high schools. Once a school’s population is one-fifth Hispanic, there is a monotonically increasing relationship between the proportion of Hispanic students and the odds of expecting to obtain a four-year college degree. The combination of the supplemental analysis with the results reported in Table 2 shows that the relatively few schools with the

Texas Students’ College Expectations most black students lead to the positive association between educational expectations and the proportion of black students. The results linking the proportion of Hispanic students and educational expectations are stronger and more consistent across a larger portion of the distribution of Hispanic students. Large proportions of Hispanic students in schools are associated with a higher-thanexpected school level of students’ expectations of graduating from a four-year college, given student and other school characteristics. I also considered whether there is an optimal mix of students associated with higher educational expectations and that any segregated school—either highly white or highly minority—would not exhibit the same level of expectations by including higher-order racial composition terms. But these nonsignificant results further support the positive monotonic influence of schools’ minority composition on educational expectations.16 The last two models I consider include interactions between student and school-level variables. Model 5 examines if the association between school racial composition and educational expectations differs by students’ race. The interaction terms are statistically insignificant, suggesting that, on average, all students in schools with greater proportions of minority students are more likely to expect to graduate with a bachelor’s degree. Thus, white students attending a school with 50 percent black and Hispanic students are more likely, on average, to expect to obtain a bachelor’s degree than are similar white students in a school with fewer minority students. This finding provides support for the argument that minority students improve the normative culture of the school. Finally, to test the idea that minority students’ expectations are lower in schools with high levels of achievement (Portes and Hao 2004; St. John 1975), I interact individual students’ race with the school achievement composition and present these results in Column 6 of Table 2. While the relationship between Hispanic students’ status and educational expectations does not change depending on the level of school achievement, this is not the case for black students. In schools

59 with higher overall achievement, black students have a lower likelihood of expecting to attain a bachelor’s degree than do similar black students in schools with lower achievement.

DISCUSSION Unlike most studies that have considered school effects on educational outcomes, this study found several school characteristics that are associated with higher educational expectations. My results show that more students expect to graduate with a bachelor’s degree in schools in which greater proportions of parents have college degrees. This could be partly a socioeconomic effect, as has been found in other studies (Jencks and Mayer 1990; Meyer 1970; Thrupp 1997). Furthermore, it is likely that higher concentrations of college-educated parents influence the normative expectations that students have of post-high school activity. Contrary to the findings of Davis (1966) and Marsh (1984, 1987, 1991), I also found that school achievement is positively related to educational expectations as well. In high-achieving contexts, it is likely that many students, teachers, and parents expect that most students in these schools will attend and complete college. More research is needed to clarify the reason for this finding and why it contradicts the findings of other research. For black students, I found that as the school level of achievement increases, the likelihood of expecting to graduate with a bachelor’s degree declines, similar to others’ findings (Marsh 1987; Portes and Hao 2004; Portes and MacLeod 1996; St. John 1975). This finding suggests that a high-achieving environment is not as beneficial to black students as it is to other students. The central finding of this article is that when similar schools are compared, greater proportions of minority students are associated with higher levels of students’ expectation to graduate with a four-year college degree. Specifically, the observed negative relationship between the proportion of minority students and educational expectations is reversed when schools with similar kinds of

60 students, socioeconomic levels, and scholastic achievement are compared. This is especially the case for schools with many Hispanic students. Moreover, the analysis suggests that attending schools with greater proportions of minority students positively influences students of all racial backgrounds. Whether higher proportions of minority students actually raise students’ goals to complete a bachelor’s degree or whether more segregated schools concentrate students with more ambitious educational goals is unclear. Because of the cross-sectional nature of the survey data that I used and the nonrandom assignment of students to schools, this question of causality cannot be definitively answered. It is possible that students differ across schools in ways that are related to their educational expectations but were not accounted for in this analysis. Nevertheless, this study showed that net of individual and school differences, students who attend more minority-segregated schools are more likely to have ambitious educational goals to complete college. What are the possible reasons for the counterintuitive finding that in similar school contexts, students in schools with greater shares of minority students are more likely to expect to graduate from a four-year college? The evidence from this analysis does not square completely with either the oppositional culture or the self-esteem hypothesis. Oppositional culture theory (Fordham and Ogbu 1986; Ogbu 1995; Ogbu and Simons 1994) implies that concentrated shares of minority students result in attitudes and behaviors that are at variance with mainstream ideals. This perspective suggests that black and Hispanic students in schools with large shares of minority students would be less likely to expect to graduate from college with a four-year degree. My findings refute this claim: I found that increased proportions of minority students are associated with higher educational expectations for all students. Instead, my results are closer to those proposed by the self-esteem hypothesis, which suggests that segregated schools benefit minority students’ academic self-image, leading to higher educational goals. Contrary to my findings, this effect is hypothesized only

Frost for minority students. Perhaps, as is suggested by the self-esteem hypothesis, though, black and Hispanic students have higher educational expectations because of the greater academic self-esteem in segregated schools that is due to the shielding of students from the complete spectrum of competition. My finding that black students’ expectations suffer in high-achieving contexts supports the self-esteem hypothesis somewhat. However, I found that white students in schools with greater concentrations of minority students have the same level of expectations as do minority students, whereas the self-esteem hypothesis suggests a differential effect by individual race. Of additional interest, Fryer and Torelli (2005) found that the social costs of minority students’ achievement are the highest in schools with few minority students. My basic findings do not contradict this finding: It is possible that minority students’ aspirations are curtailed in schools with more white students for this reason. When Fryer and Torelli’s finding is combined with the self-esteem hypothesis, it is possible that a critical mass of minority students is a necessary factor in explaining their self-esteem, ability to pursue academic success, and subsequent educational goals. If the concentration of minority students in a school influences the school normative climate, it is feasible that white students could also benefit with higher educational goals. Thus, the mechanisms explaining minority and white students’ educational goals could be different. Without direct measures of the social costs of educational aspirations and self-esteem, I cannot examine the merits of this possibility analytically. In a variation of this theme, Goldsmith (2004) reported that minority students attending segregated minority schools with many minority teachers are more optimistic and express more pro-school attitudes than do similar minority students attending white schools. He suggested that the racial composition of the teaching faculty is key to understanding this effect and that minority teachers are better able to encourage and influence minority students. Despite differences in our data (Goldsmith used 1998 national data from the National Education Longitudinal

Texas Students’ College Expectations Study), I attempted to replicated his results, but found no evidence that the racial composition of teachers in schools influenced students’ educational expectations. Goldsmith’s study focused on the high educational goals (graduate or professional school) of younger, eighth-grade students, whose expectations are not fully formed and who may have more unrealistic educational goals. Perhaps these younger, more impressionable, students are more influenced by their teachers than are older students who are more seasoned by life’s reality checks. Whatever the reasons for this difference, I found no evidence that the proportion of minority teachers in schools with large proportions of minority students is associated with high educational expectations. Goldsmith (2004) also claimed that the concentration of black and Hispanic students in schools improves the schools’ normative climate because these students have more optimistic and pro-school attitudes, which could lead, in turn, to more ambitious educational goals. He noted that “concentrating students with high beliefs generally will raise all students’ beliefs” (p. 141). In a subsequent analysis,17 I found that black and Hispanic students do have more pro-school and positive attitudes than do white students, and this seems to be one likely explanation, which would account for the reason why even for white students, schools with more minority students are associated with higher educational expectations. My findings also square with research by Ainsworth-Darnell and Downey (1998), who found that black students report more pro-school attitudes than do whites, which enhances their academic success. However, in further analyses, I found that the overall attitudinal climate of a school is unrelated to educational expectations and does not mediate the association between school racial composition and educational expectations. It is possible that these measures do not completely capture optimism and pro-school attitudes within a school. More research is needed to examine the association between school racial composition and school optimism. However, the influence of minority students’ optimism on school climate remains a distinct possibility.

61 Finally, the finding that schools with high concentrations of Hispanic students have higher-than-expected levels of students’ expectations of graduating with a bachelor’s degree suggests that something specifically related to Hispanics may be relevant. Kao and Tienda (1995) found that the immigration status of parents is more important in explaining the success of immigrant youths than is the youths’ individual immigrant status. They suggested that immigrant parents have ambitious goals for and optimistic views of their children’s future. Furthermore, second-generation youths have the academic advantage of English language proficiency compared with non-native-born students. Thus, when immigrant parents’ optimism is paired with second-generation language fluency, students are particularly likely to succeed in school. These findings suggest that the composition of second-generation youths may be related to the school level of educational expectations. Substantial concentrations of secondgeneration youths, which I could not measure, may also influence the normative climate of the school and influence students’ educational expectations. Future research can shed light on which, if any, of these explanations is accurate. Although school racial composition, the focus of this article, is associated with college expectations, it is noteworthy that other school factors, especially the parental education composition of schools, are more important in explaining why schools differ in their levels of students’ educational expectations. Thus, although students in high-minority schools do have the advantages of higherthan-expected educational goals, they are also in school contexts that are disadvantaged by low SES and low achievement, in addition to greater levels of individual socioeconomic and educational disadvantage. Furthermore, educational expectations are only one of many important academic outcomes that accrue to students while they are in high school and only one of the first of many steps to eventually attaining a college degree. Further work is needed to consider how school racial composition is associated with other stages of the college preparation and attainment process, including academic

62 preparation, application, and initial and continued enrollment in college.

NOTES 1. This figure was obtained by my analysis of 2003 TEA data. 2. Significant others’ expectations can by conveyed indirectly through modeling, as in the case of admired peers or educated adults, or by specific encouragement of college attendance, often given by influential adults, such as parents and teachers (Haller 1982; Sewell et al. 1969). 3. Although Falk theorized that black students’ expectations would drop with their integration into white schools, he did not find this effect. 4. The project’s primary investigator is Marta Tienda of Princeton University. More information about the data, as well as ongoing research, can be found at the project’s web site: http://www.texastop10.princeton.edu. 5. In the 1996 Fifth Circuit Court decision Hopwood v. University of Texas Law School, race-sensitive affirmative action for college admissions was banned. In its place, the Texas legislature passed HB 558, popularly known as the top-10 percent law, which allows the top 10 percent of students from any Texas high school to attend any public university in Texas. 6. In order of the variable with the most missing data to that with the least missing, I regressed each variable with missing values on all the other individual-level variables that were used in the analyses and sorted the data on the basis of predicted values for the variable of interest. I then divided my sample into bins of 50 respondents each to locate donors for missing values. Within each bin, I randomly selected a nonmissing value to impute a value for missing cases. I repeated this process for each variable with missing data and flagged all instances in which the data were imputed. 7. The limited variance in the school proportion of Asian students led to unstable estimates when this variable was included in the model. In addition, because of the small proportion of Asian students in schools, there was little difference between the proportion

Frost white and the proportion white and Asian. Thus, the reference category is the percentage of white/Asian students. Furthermore, although others have combined the proportion of black and Hispanic students to form one measure of school racial composition, because of unique effects shown later in the article, I used these two separate measures of black and Hispanic composition. 8. Unfortunately, immigration data are available only for the students, not for their parents. 9. Without more information, it is impossible to determine the reason for this finding. I had no ability to control for the quality of the courses the students had taken, so it is possible that students in high-minority schools might have taken lower-quality courses, relative to those found in lower-minority schools, even though these courses were more likely to be labeled “college prep.” 10. Random-slope models could be used to explore whether individual effects vary by school. Such an analysis, which is beyond the scope of this article, could be fruitfully used in future research. 11. These models are estimated with a sixthorder approximation of the likelihood based on Laplace transform approximations for Bernoulli models. This approximation provides the most accurate estimates of effects for a multilevel model with a dichotomous outcome. See Rodríguez and Goldman (1995). 12. HLM does not provide significance tests of the Laplace transform approximations because of disagreement over exactly how to estimate p-values, but using a rough rule of thumb, by which parameter estimates larger than 1.96 times their standard error are significant at the .05 level, the estimate of τ00 here clearly meets this criterion. 13. Self-reported GPA could be biased if students are more likely to report higher grades than they actually earned. However, this does not present significant difficulties for my analysis for several reasons. First, Cassady (2001) found that GPA is accurately reported, more so than self-reported SAT scores. In addition, the GPA measure used here is based on the most recent grading period, making memory lapses less problematic. Finally, the misreporting of this relatively minor variable,

Texas Students’ College Expectations although an important issue, is less relevant for an article that focuses on school context. 14. Because of the sequential ordering of my models, it is impossible to compare τ00 from a model with only student characteristics to one with both student and school characteristics. Another analysis, not shown here, estimated τ00 to be .242 when only student characteristics were included, a 29 percent reduction from the unconditional model. Adding school characteristics to this model results in a τ00 of .082, leading to the conclusion that school characteristics are responsible for a reduction of 66 percent in the between-school variance that is not accounted for by student characteristics. 15. These supplemental analyses adjust for all student and school-level characteristics found in Model 4, Table 2. The results are available on request. 16. An anonymous reviewer pointed out a potential methodological explanation for this finding. Perhaps higher drop-out rates in high-minority schools have left a selective population of seniors. Further research could investigate this possibility by examining younger students’ educational expectations. 17. The THEOP data include a group of questions regarding students’ educational attitudes. A simple cross tabulation by racial subgroup showed that, on average, black and Hispanic students exhibit more positive attitudes toward school.

REFERENCES Ainsworth-Darnell, James W., and Douglas B. Downey. 1998. “Assessing the Oppositional Culture Explanation for Racial/Ethnic Differences in School Performance.” American Sociological Review 63:536–53. Alexander, Karl, and Bruce K. Eckland. 1975. “Contextual Effects in the High School Attainment Process.” American Sociological Review 40:402–16. Alwin, Duane F., and Luther B. Otto. 1977. “High School Context Effects on Aspirations.” Sociology of Education 50:259–73. Arkes, Jeremy. 1999. “What Do Educational Credentials Signal and Why Do Employers Value Credentials?” Economics of Education Review 18:133–41.

63 Blau, Peter M., and Otis Dudley Duncan. 1967. The American Occupational Structure. New York: John Wiley & Sons. Buchmann, Claudia, and Ben Dalton. 2002. “Interpersonal Influences and Educational Aspirations in 12 Countries: The Importance of Institutional Context.” Sociology of Education 75:99–122. Campbell, Richard T. 1983. “Status Attainment Research: End of the Beginning or Beginning of the End?” Sociology of Education 56:47–62. Cassady, Jerrell C. 2001. “Self-Reported GPA and SAT: A Methodological Note.” Practical Assessment, Research, and Evaluation 7 (12). Available at http://PAREonline.net/getvn. asp?v=7&n=12 Cheng, Simon, and Brian Starks. 2002. “Racial Differences in the Effects of Significant Others on Students' Educational Expectations.” Sociology of Education 75:306–27. Coleman, James, Ernest Q. Campbell, Carl F. Hobson, James McPartland, Alexander Mood, F. D. Weinfeld, and R. L. York. 1966. Equality of Educational Opportunity. Washington, DC: U.S. Government Printing Office. Davis, James A. 1966. “The Campus as a Frog Pond: An Application of the Theory of Relative Deprivation to Career Decisions of College Men.” American Journal of Sociology 72:17–31. Drury, Darrel W. 1980. “Black Self-Esteem and Desegregated Schools.” Sociology of Education 53:88–103. Falk, William W. 1978. “School Desegregation and the Educational Attainment Process: Some Results from Rural Texas Schools.” Sociology of Education 51:282–88. Farkas, George, Christy Lleras, and Steve Maczuga. 2002. “Does Oppositional Culture Exist in Minority and Poverty Peer Groups?” American Sociological Review 67:148–55. Fordham, Signithia, and John U. Ogbu. 1986. “Black Students' School Success: Coping with the Burden of 'Acting White.'” Urban Review 18:176–206. Fryer, Roland, and Paul Torelli. 2005. “An Empirical Analysis of 'Acting White.'” NBER Working Paper No. W11334. Available at http:// ssrn.com/abstract=723303 Goldsmith, Pat Antonio. 2004. “Schools' Racial Mix, Students' Optimism, and the Black-White and Latino-White Achievement Gaps.” Sociology of Education 77:121–47. Goyette, Kimberly, and Yu Xie. 1999. “Educational Expectations of Asian American Youths: Determinants and Ethnic Differences.” Sociology of Education 72:22–36.

64 Gray-Little, Bernadette, and Robert A. Carels. 1997. “The Effect of Racial Dissonance on Academic Self-Esteem and Achievement in Elementary, Junior High, and High School Students.” Journal of Research on Adolescence 7:109–31. Guo, Guang, and Hongxin X. Zhao. 2000. “Multilevel Modeling for Binary Data.” Annual Review of Sociology 26:441–62. Gurin, Patricia, Eric L. Dey, Sylvia Hurtado, and Gerald Gurin. 2002. “Diversity and Higher Education: Theory and Impact on Educational Outcomes.” Harvard Educational Review 72:330–66. Haller, A. O. 1982. “Reflections on the Social Psychology of Status Attainment.” Pp. 3–28 in Social Structure and Behavior: Essays in Honor of William H. Sewell, edited by Robert M. Hauser, David Mechanic, Archibald O. Haller, and Taissa S. Hauser. New York: Academic Press. Hanson, Sandra L. 1994. “Lost Talent: Unrealized Educational Aspirations and Expectations Among U.S. Youths.” Sociology of Education 67:159–83. Hao, Lingxin, and Melissa Bonstead-Bruns. 1998. “Parent-Child Differences in Educational Expectations and the Academic Achievement of Immigrant and Native Students.” Sociology of Education 71:175–98. Hauser, Robert M., William H. Sewell, and Duane F. Alwin. 1976. “High School Effects on Achievement.” Pp. 309–41 in Schooling and Achievement in American Society, edited by David L. Featherman. New York: Academic Press. Hirschman, Charles, Jennifer C. Lee, and Amon S. Emeka. 2004. “Explaining Race and Ethnic Disparities in Educational Ambitions.” Paper presented at the annual meeting of the Population Association of America, Boston. Hossler, Don, John Braxton, and Georgia Coppersmith. 1989. “Understanding Student College Choice.” Pp. 231–88 in Higher Education: Handbook of Theory and Research Volume 5, edited by John C. Smart. New York: Agathon Press. Jencks, Christopher, James Crouse, and Peter Mueser. 1983. “The Wisconsin Model of Status Attainment: A National Replication with Improved Measures of Ability and Aspiration.” Sociology of Education 56:3–19. Jencks, Christopher, and Susan E. Mayer. 1990. “The Social Consequences of Growing Up in a Poor Neighborhood.” Pp. 111–86 in Inner City Poverty in the United States, edited by Laurence Linn and Michael McGeary. Washington, DC: National Academy Press. Kao, Grace, and Marta Tienda. 1995. “Optimism and Achievement: The Educational Performance of

Frost Immigrant Youth.” Social Science Quarterly 76:1–19. ——. 1998. “Educational Aspirations of Minority Youth.” American Journal of Education 106:349–84. Landerman, Lawrence R., Kenneth C. Land, and Carl F. Pieper. 1997. “An Empirical Evaluation of the Predictive Mean Matching Method for Imputing Missing Values.” Sociological Methods and Research 26:3–33. Little, Roderick J. 1988. “Missing-Data Adjustment in Large Surveys (with Discussion).” Journal of Business and Economic Statistics 6:287–301. Little, Roderick J., and Donald B. Rubin. 1987. Statistical Analysis with Missing Data. New York: John Wiley and Sons. Marsh, Herbert W. 1984. “Self Concept: The Application of a Frame of Reference Model to Explain Paradoxical Results.” Australian Journal of Education 28:165–81. ——. 1987. “The Big-Fish–Little Pond Effect on Academic Self-Concept.” Journal of Educational Psychology 79:280–95. ——. 1991. “Failure of High-Ability High Schools to Deliver Academic Benefits Commensurate with Their Students' Ability Levels.” American Educational Research Journal 28:445–80. Meyer, John W. 1970. “High School Effects on College Intentions.” American Journal of Sociology 76:59–70. Michael, John A. 1961. “High School Climates and Plans for Entering College.” Public Opinion Quarterly 25:585–95. Morgan, Stephen L. 1996. “Trends in Black-White Differences in Educational Expectations: 1980–1982.” Sociology of Education 69:308– 19. ——. 1998. “Adolescent Educational Expectations: Rationalized, Fantasized, or Both?” Rationality and Society 10:131–62. Murdock, Steve H., Nazrul Hoque, Martha Michael, Steve White, and Beverly Pecotte. 1997. The Texas Challenge: Population Change and the Future of Texas. College Station: Texas A&M University Press. Ogbu, John U. 1991. “Low School Performance as an Adaptation: The Case of Blacks in Stockton, California.” Pp. 249–85 in Minority Status and Schooling: A Comparative Study of Immigrant and Involuntary Minorities, edited by John U. Ogbu. New York: Garland. ——. 1995. “Cultural Problems in Minority Education: Their Interpretations and Consequences—Part One: Theoretical Background.” Urban Review 27:189–205. Ogbu, John U., and Herbert D. Simons. 1994. “Cultural Models of School Achievement: A

Texas Students’ College Expectations Quantitative Test of Ogbu's Theory. (Cultural Models of Literacy: A Comparative Study, Project 12).” National Center for the Study of Writing and Literacy, Berkeley, CA. Orfield, Gary. 2001. “Schools More Separate: Consequences of a Decade of Resegregation.” The Civil Rights Project, Harvard University, Cambridge, MA. Available at http://www.civilrightsproject.harvard.edu/research/deseg/Sch ools_More_Separate.pdf Orfield, Gary, and John T. Yun. 1999. “Resegregation in American Schools.” The Civil Rights Project, Harvard University, Cambridge, MA. Available at http://www.civilrightsproject.harvard.edu/research/deseg/Resegregation_American_Schools99.pdf Perna, Laura Walter. 2000. “Differences in the Decision to Attend College Among African Americans, Hispanics, and Whites.” Journal of Higher Education 71:117–41. Portes, Alejandro, and Lingxin Hao. 2004. “The Schooling of Children of Immigrants: Contextual Effects on the Educational Attainment of the Second Generation.” Proceedings of the Natural Academy of Sciences 101:11920–27. Portes, Alejandro, and Dag MacLeod. 1996. “Educational Progress of Children of Immigrants: The Roles of Class, Ethnicity, and School Context.” Sociology of Education 69:255–75. Qian, Zhenchao, and Sampson Lee Blair. 1999. “Racial/Ethnic Differences in Educational Aspirations of High School Seniors.” Sociological Perspectives 42:605–25. Raudenbush, Stephen W., and Anthony S. Bryk. 2002. Hierarchical Linear Models: Applications and Data Analysis Methods (2nd ed.). Thousand Oaks, CA: Sage. Rodríguez, Germán, and Noreen Goldman. 1995. An Assessment of Estimation Procedures for Multilevel Models with Binary Responses.”

65 Journal of the Royal Statistical Society, Series A 158:73–90. Sewell, William H., Archibald O. Haller, and George W. Ohlendorf. 1970. “The Educational and Early Occupational Attainment Process: Replication and Revision.” American Sociological Review 35:1014–27. Sewell, William H., Archibald O. Haller, and Alejandro Portes. 1969. “The Educational and Early Occupational Attainment Process.” American Sociological Review 34:82–92. Sewell, William H., and Robert M. Hauser. 1992. “A Review of the Wisconsin Longitudinal Study of Social and Psychological Factors in Aspirations and Achievements 1963–1992.” Center for Demography and Ecology University of Wisconsin, Madison. Sewell, William H., and Vimal P. Shah. 1968. “Social Class, Parental Encouragement, and Educational Aspirations.” American Journal of Sociology 73:559–72. St. John, Nancy H. 1975. School Desegregation Outcomes for Children. New York: John Wiley and Sons. Texas Education Agency. 2003. “Academic Excellence Indicator System Report, 2002–2003.” Austin: Author. Thrupp, Martin. 1997, March 24–28. “The School Mix Effect: How the Social Class Composition of School Intakes Shapes School Processes and Student Achievement.” Paper presented at the annual meeting of the American Educational Research Association, Chicago. Topel, Robert. 2004. “The Private and Social Values of Education.” Pp. 47–57 in Education and Economic Development, A Federal Reserve Bank of Cleveland Research Conference. Cleveland, Ohio: Federal Reserve Bank of Cleveland. Turner, Ralph H. 1964. The Social Context of Ambition. San Francisco: Chandler.

Michelle Bellessa Frost, Ph.D., is a recent graduate of the Office of Population Research, Princeton University. Her main fields of interest are education and stratification. Her current research focus is on high school students’ access to guidance counseling. This research was supported by grants from the Ford, Mellon, Hewlett, and Spencer Foundations and Grant SES-0350990 from the National Science Foundation. I gratefully acknowledge institutional support from the Office of Population Research (NICHD Grant R24 H0047879). Special thanks go to Marta Tienda, Mario Small, Germán Rodriguez, and Meredith Kleykamp for their helpful comments and feedback at various stages in this research. A version of this article was presented at the Notestein Seminar Series, Princeton University, May 2004. Direct correspondence to Michelle Bellessa Frost, Office of Population Research, Princeton University, Princeton, NJ, 08544; email: [email protected].