The Role of Large Data Sets in Educational Research

The Macroecology of Educational Outcomes

Ecological Embeddedness of Educational Processes

Early Labor Market Experience of Non-College Youth

The first large data sets probably date back to the first U.S. Census, conducted in 1790. For nearly two centuries, the decennial census has monitored the growth and development of the country, yielding valuable information about who we are, what we do, and how we live. Indeed, there is a considerable amount of data in current censuses that is of interest to educational researchers, specifically regarding school expenditures, enrollment, dropout, programs for the handicapped, and private schooling. Although new variables contributing to the timeliness of the data (e.g., computer usage) are added continually, past censuses are valuable for the historical analyses they make possible.

However, while they provide general pictures of national trends, census data often lack sufficient detail. Other large data sets must be considered. Depending on the area of research, different data sets can best serve the needs of the interested researchers. Economists might use the Census of Business and Industry; sociologists, the General Social Survey which annually collects data on a variety of public opinion topics; and educational researchers, the various data sets created by the National Center for Educational Statistics (NCES).

In this issue, we consider three studies, each of which uses a different type of data set. Overall, the studies clearly demonstrate that the analysis of large data sets compliments other more personal types of data collection, such as observational studies, by providing a basis of comparison sometimes lacking in such specific settings as well as a basis for more in-depth studies.

In the first study, David Bartelt examines the possibility that the changing composition of cities, rather than what is or is not happening in the schools, accounts for the perceived failure of the educational system. Using data from 58 U.S. cities, he constructs a unique data set aggregated to the central cities and their metropolitan areas. Historical in nature, the economic information includes data from 1920 through 1987 and population figures from 1930 through 1990, with specific 1986-87 school district information from the 1987 Census of Government.

Bartelt systematically examines levels of the consequences of economic transition, national migration, and urban decentralization as they relate to educational failure, defined as the proportion of individuals aged 16-19 who are either not in school or have not received a high school diploma. His findings will be published in "The Macroecology of Educational Outcomes," to be included in School Community Connections: Exploring Issues for Research and Practice (Rigsby, Reynolds, & Wang, in press).

William Yancey and Salvatore Saporito take a different approach. They collected data from three very dissimilar sources: first, police, social welfare, and health data were extracted from the U.S. Census at the census-tract level; second, information was gleaned from the Philadelphia Pupil Directory File, a database of all students enrolled in the city's public schools; third, school data were gathered from annual reports of the Philadelphia School District's Management Information Center for 1985 through 1991. In their paper, "Ecological Embeddedness of Educational Processes and Outcomes," also to be published in Rigsby et al., Yancey and Saporito explore the relationship between the educational character of inner-city schools and their neighborhoods, particularly policies of neighborhood schooling, busing, and desegregation.

Finally, William Stull and Michael Goetz use the High School & Beyond (HS&B) data set (produced by NCES) to explore the school-to-work transition of inner-city studentsQboth high school graduates and dropoutsQnot entering college. HS&B is particularly well suited for this analysis because of its longitudinal nature. In the base year (1980), extensive personal, educational, and occupational information was collected from 58,000 high school students (30,000 seniors and 28,000 sophomores). Subsequent surveys of the same individuals were conducted in 1982, 1984, and 1986. Additional information was obtained from high school transcripts.

Beside HS&B, NCES has produced a multitude of data sets, among them: NLS-72, the National Longitudinal Study of the High School Class of 1972; and NELS:88, the National Educational Longitudinal Study of 1988. These data sets are very large, both in topics included and students involved. Further, they share some common questions, thus lengthening the potential study time frame beyond their respective data collection periods.

Large data sets are versatile. Researchers can utilize these databases to describe an educational situation at one point in time or to develop explanatory models and predict outcomes. In addition, large data set analyses can aid the qualitative researcher in identifying both the unique and generalizable in his or her studies. Conversely, qualitative studies can help the quantitative researcher understand the process being modeled. The two methods are complimentary, not competitive.

After three decades of debate over the causes of educational deficits in urban schools, research suggests that the changing makeup of cities, not necessarily the schools themselves, accounts for much of the failure of our educational system. The movement of resources, jobs, and people from central city to suburb has created a hostile environment for communities and their institutions within the inner city. The task at hand is to examine how the forces of change constrain the effective operation of schools. The first step is to establish the empirical relationship between these macrosocial forces and educational accomplishment. A further implication is that the educational system- including its students-is part of a "macroecology" of urban relationships.

Toward an Ecology of Education Analyses focusing on city change rather than educational failure necessitates a paradigmatic shift in focus, from schools to urban processes. Thus, the focus of concern becomes the network of institutions and processes that affect the schools. The main organizational feature of American education is that it is both embedded in national and multinational institutions and localized in its point of delivery: Basically, schools are firmly planted in a network of social processes. The ecological model suggests that it is possible to distinguish the salient characteristics of these social arrangements as a means of better understanding the outcomes of the educational process. It also suggests that we can identify the support services which may need integration into and coordination with the educational process.

Cities, which form the largest embedding context, include both a network of communities and an institutional matrix of private and public organizations addressing various political, economic, and social activities. "Inner cities" and "inner-city schools" imply areas stratified by race and class. Much of the literature on the educational deficits of urban schooling ties educational outcomes to distributional problems/inequities in educational financing and "family" variables such as income level (Kennedy, Jung, & Orland, 1986). In both instances, racial isolation and poverty combine to negatively affect the likelihood of educational success.

Our research is unique in that it systematically examines the consequences of economic transition, national migration, and urban decentralization on a major indicator of educational successQthe proportion of students aged 16-19 who are either not in school or have not earned a diploma, using a data set from 58 cities.

The Macroecology of Postwar American Cities Three intersecting macrosocial forces have dominated American urban life since World War II: The explosive growth of suburbs, combined with the persistent flight of manufacturing, has created a new, decentralized form of urban life. This spatially segregates, by poverty and race, a smaller, less economically viable, and more African-American city from its more prosperous suburbs (Wilson, 1987). Further, the growth of a "postindustrial" economy, dominated by the service sector, has impacted on the economic externalities of education, affecting both employment possibilities (Bluestone & Harrison, 1982) and the fiscal health of cities (O'Connor, 1973).

Thus, inner-city schools are increasingly the schools of remnant populations and communities trapped by their economic irrelevance or their links to diminished labor markets. Further, they are increasingly dependent on an overloaded and endangered fiscal base.

Additionally, the economic activity which undergirds urban life has changed. From the factories and railroads of the late 19th and early 20th centuries, America now has a service sector economy, and expectations have grown regarding the nature of job skills. Thus, a city's capacity to address changing educational needs and the extraeducational influences upon schools are contingent on the city's success in dealing with these changes in urban social structure. The fact is, the inducement for succeeding within the public school system has typically been economicQa job. The System of Cities and Its Educational Indicators To empirically examine the ways in which the systemic, macroecological character of inner city manifests itself, educational variables should be set in the context of the factors driving the national system of cities. Such an examination should include a recognition of changes in both the types and locations of jobs in metropolitan areas, as well as corresponding shifts in population.

The data set used is a synthesis of several economic, population, and government census data sets, aggregated to the level of central cities and their metropolitan areas. The database covers economic census materials from 1929-1987, and population data from 1930-1990. The basic variable is the manufacturing ratioQthe proportion of manufacturing jobs compared to wholesale, retail, service, and manufacturing jobs combined. In general, the literature on urban change suggests that urban fiscal problems and problems associated with the thinning out of cities are associated with persistent attachment to a manufacturing base.

African-American Urban Isolation The basic shift from a manufacturing to a nonmanufacturing economy has been substantial; the pattern is one of decentralization and migration. Older manufacturing centers have lower population increases and, in the most heavily industrial cities, clear losses.

African Americans migrated to older centers of manufacturing industry while whites were beginning to move to the emerging economic centers of the Sun Belt (Long, 1988). As the character of migration changed, lower income and less-educated households, particularly African-American ones, remained behind. In a larger context, as population and capital relocated after World War II, they tended to leave cities where African Americans had resettled during the pre-War and immediate post-War period.

The Costs of Inner-City Education While the economic base of cities has certainly changed, the accompanying decentralization appears to have had the strongest direct relationship with inner-city educational issues (Katznelson & Weir, 1985; Kantor & Brenzel, 1993). In particular, the spatial deconcentra- tion of population in urban areas interacts with economic dislocation to decrease the access of inner-city households to jobs and wages (Noyelle, 1987). It appears that suburbanization is as much an economic as a social response to a changed urban economic order. Inner-city schools are increasingly the schools of remnant populations and communities trapped by their economic irrelevance or their links to diminished labor markets. The data demonstrate that the more decentralized a city is, the higher the level of instructional expenditure, and the heavier the economic burden on the taxpayer. Decentralization is a significant correlate, possibly causally linked, with fiscal distress (Rusk, 1993). Essentially, these data suggest that schooling in decentralized cities is more expensive, and places a heavier burden on the taxpayer. Thus, cities experiencing the greatest population losses and, by extension, a more diluted tax base, must simultaneously carry an increasingly costly educational system.

School Dropout Discussion of school dropout is usually considered within the framework of schools failing students (Natriello, 1987). However, students in many communities are leaving school and opting toward other choices, such as jobs (Pallas, 1980). The result is that dropout rates may be high in cities experiencing economic and educational deprivation, but also high where economic opportunities are sufficiently present to lure students from schools into the labor force.

The analysis indicates that city dropout rates appear to be a function of both opportunities and constraints. Dropout rates tend to be higher in cities which retain a high manufacturing base, and in which the African-American population is large. These are cities in which the revenue load is high, but the instructional expense somewhat lower, and where the educational needs of students tend to be higher due to linguistic isolation. Interestingly, the dropout rate is also affected by the growth rate, indicating that there are conditions in which potential increased opportunities are associated with dropout.

Discussion and Implications The educational literature suggests that differences in race, class, segregation, and family resources (both social and economic) are reflected in differential educational attainment, and that they codetermine, with educational credentials, the economic attainment of students.

In sum, this research suggests that: (a) there is a relationship between city development trajectories and patterns of educational funding; (b) there is a distinct pattern to the levels of racial segregation found in cities, strongly linked to their economic history; and (c) a strong ecological relationship exists between the economic development of cities, opportunity structures, funding levels, and degree of dropout. Finally, educational outcomes are highly correlated with other products of the city system, such as economic opportunities and the relative presence of single-parent households.

The continued failure of inner-city schools despite 25 years of federal, state, and local efforts to improve urban education may be understood in terms of the fundamental changes that have taken place, not in schools, but in the social ecology of cities and the structure of urban labor markets. The larger socioeconomic context of schooling, the ecologies of the communities in which schools are embedded, is a principal determinant of the success of schools (Kantor & Brenzel, 1993).

The differentiation and segregation of urban neighborhoods by race, ethnicity, and socioeconomic status (Massey, 1993), coupled with the fact that schools generally draw students from relatively small geographic areas, leads us to hypothesize that differences in social, political, and economic characteristics observable between cities and suburbs may also be observed between local communities and schools within cities such as Philadelphia. Thus, our research explores the relationship between the educational character of inner-city schools and the mosaic of differentiated communities in which they are embedded.

Methods Our approach is based on the premise that the areas in which students actually live more accurately reflect a school's community than does the school's immediate neighborhood. Consequently, it was necessary to determine where each student in Philadelphia lives, to obtain information describing these areas, and to summarize this information for each school.

Using a computer matching program, data describing each student's census tract were attached to his or her student record. These data were then aggregated for each school by calculating the average value of a given characteristic. After aggregating neighborhood data, we merged them with school characteristic data, including average standardized reading test scores; rates of average daily attendance and pupil turnover; and the percentages of students participating in busing, transportation assistance, and free or reduced-price lunch programs.

Schools, Communities, and School Compositions Despite the expected parallels between the racial/ethnic and socioeconomic character of the city's neighborhoods and schools, the two are not mirror images. Two factors distort this reflection, and both vary across the city.

First, only 71% of Philadelphia's students attended public schools. White students and students from families of relatively high socioeconomic status were much more likely to attend private or parochial schools. The average population percentages of African Americans and Latinos for all tracts in the city were 44.7% and 7.1%, respectively; the averages for these two groups for tracts represented by public elementary school students were 53.6% and 7.6%. Public elementary school students were also more heavily drawn from tracts with higher proportions of poverty.

Second, schools vary in the degree to which they draw students from relatively small geographic areas. Students may be bused to schools via magnet or other special programs (many of which are part of school district efforts to desegregate or relieve overcrowding) which are outside theeir neighborhoods.

The outcome of these distortions is that Latinos, African Americans, and children from low-income families have become concentrated in Philadelphia's "local" public schools. To systematically explore this issue, we conducted multiple regression analyses in which the characteristics of the communities represented in each school were regressed against the proportions of the schools' African-American, Latino, and low-income student populations. For Latinos, we found that the only factor strongly associated with school populations is the concentration of Latinos in feeder areas. Few children from Latino neighborhoods, which are uniformly poor and segregated, go to private schools. This matter is more complex for African Americans. Schools that draw students from areas with higher rates of private school attendance or that are racially mixed but internally segregated contain higher proportions of African-American students than would be expected given the racial composition of their surrounding communities. Although these areas are racially mixed and have the demographic potential for the racial integration of schools, they are instead associated with interracial conflict (Goldstein & Yancey, 1988). This may partially explain the relatively high rates of private school attendance by white children.

The association between rates of poverty in communities and rates of low-income students in schools is not as strong. In effect, two additional community characteristics were found to be related to increased concentrations of low-income students: the degree of local racial/ethnic segregation and the economic heterogeneity of the schools' communities. Schools that draw students from racially mixed but internally segregated communities have higher proportions of low-income students, suggesting that in these communities higher income students frequently attend private schools or public schools outside the local community. The economic heterogeneity of these schools' communities is evident in the fact that schools drawing students from diverse census tracts (magnet schools, for the most part) have lower concentrations of low-income students than would be expected.

The systemic consequences of these school- and community-level ecological processes appear when we examine the levels of racial/ethnic and economic segregation in Philadelphia's public schools. Despite federal and local resolve, the racial/ethnic segregation in the city's schools has declined little since 1950. Our analysis also suggests that high levels of economic segregation in the schools are partly a consequence of the voluntary busing program which, while reducing racial segregation, has drawn students from higher income families away from their neighborhood schools, thereby decreasing the schools' economic heterogeneity. In short, efforts to decrease racial segregation have increased economic segregation.

Types of School-Community Ecologies To further explore the relationships between neighborhood and school characteristics, we conducted a cluster analysis in which a series of community characteristics represented in schools ("school- community characteristics") were used as criteria for classifying the city's public elementary schools into relatively homogeneous groups ("school-community clusters"). These characteristics included the percentages of census tract populations that were African American, Latino, poor, or attending private schools; the number of infants born to teenage mothers and to mothers who received inadequate prenatal care; and the standard deviation of the average rent across census tracts.

Seven school-community clusters emerged from this analysis: "White Middle Class," "Magnet Schools," "African-American Middle Class," "White Working Class," "African-American Poor," "African-American Working Class," and "Latino Poor." Our analyses indicated that some of the school-community clusters are characteristically quite similar, particularly the African-American Working Class and African-American Poor clusters. By contrast, the White Middle Class and Latino Poor clusters were the most dissimilar. Community Ecologies, Academic Climates, and Outcomes To explore the relationships between the characteristics of neighborhoods, their schools' academic climates, and the educational success of students, we considered the manner in which four variablesQ the percentage of students bused to a given school, the rate of student turnover during the academic year, average daily attendance, and students' average scores on standardized reading testsQdiffered across the school-community clusters delineated above. It is vital for urban education systems to recognize that communities and schools differ in their needs and circumstances. And these differences must be taken into account if schools are to be effective. Our analyses revealed that rates of turnover, the percentage of students bused, and average daily attendance (which together provide an indicator of the quality of overall academic climate) are directly related to reading test scores. Examination of the clusters from poorest (Latino Poor) to most economically advantaged (White Middle Class) reveals a relatively clear pattern of decreasing rates of student turnover and increasing rates of daily attendance. Reading scores also mirror this pattern.

As with the general characteristics of the clusters themselves, we analyzed similarities among the academic climates of the schools in each cluster. Overall, the academic climates of the schools in the Magnet Schools and White Middle Class clusters are most similar while, not surprisingly, schools in the Latino Poor cluster exhibit academic climates that are significantly dissimilar to those of the other six clusters.

Separate, systematic comparisons of (a) school-community clusters in general and (b) the academic climates of the schools in each cluster, provide a particularly useful means of examining the relationship between the characteristics of the communities in which schools are embedded and the quality of their academic climates. With few exceptions, we observed significant correlations between dissimilarities in clusters overall and dissimilarities in academic climates. This indicates that differences in the nature of the communities in which schools are embedded are directly related to differences in their academic climates and outcomes.

Conclusions In addition to demonstrating the important links between schooling success and the communities in which schools are embedded, the methodologies of this analysis also provide a means to identify schools whose communities and families do not have adequate access to needed services. For example, school-community populations with high rates of children born with inadequate prenatal care could be targeted for additional health seervices. Information regarding rates of teenage motherhood could help school districts identify schools and students who would benefit from additional family planning, sex education, birth control, and special programs for teenage mothers.

We do not maintain that city schools are helpless institutions that passively adapt to external processes. Rather, we feel that it is vital for urban education systems to recognize that communities and schools differ in their needs and circumstances, and that these differences must be taken into account if schools are to be effective.

The majority of American youth do not attend a post-secondary institution of higher learning after leaving high school. Instead, they enter the labor force as part-time or full-time workers, join the military, become homemakers, or pursue other activities. Until recently, the social science research community largely overlooked the members of this group, except for those who engaged in antisocial behavior. This neglect resulted partly from a lack of data describing the group's activities, and partly from a belief, usually implicit, that college-educated youth were more important to national economic well-being than their non-college counterparts.

These circumstances have changed dramatically over the past twenty years. First, a number of large national data sets describing the secondary and post-secondary behavior of young people over time have become available. These include the National Longitudinal Study of the High School Class of 1972 (NLS-72), High School and Beyond (HS&B), and the National Educational Longitudinal Study of 1988 (NELS:88).

Second, the rapid economic growth of our closest international competitors has increasingly been attributed to the high skill level of their non-college-educated populations. The successes of these countries, particularly Japan and Germany, have caused many researchers and policymakers to question the American system of educating and training its non-college-educated work force.

At CEIC, my colleagues and I are currently focusing our research on the early labor market experience of young people, both high school dropouts and graduates, who are not in college or the military but are working part time or full time for pay. Our central concern is the relationship between a student's academic and nonacademic experiences in high school and his or her success in the labor market after leaving school.

The data come from the sophomore cohort of the HS&B longitudinal survey, which provides the most recent national data on the post-secondary work experience of non-college youth. In 1980, approximately 15,000 high school sophomores were chosen to participate in the four waves of the survey carried out in 1980, 1982, 1984, and 1986. Of these, roughly 12,000 completed questionnaires in all four years. We refer to this latter group as the full-participation cohort. Some of our analysis focuses on this group and some on a subcohort comprised of at-risk youth attending urban high schools (various definitions of "at risk" were used, with little effect on the basic results).

Our first round of analysis examines the wages earned in 1984, two years after graduation. The literature on the determinants of wages for non-college youth is small, but there seems to be some consensus that grades and other indicators of academic performance in high school do not greatly influence earnings. Thus, many analysts conclude that employers of non-college youth value "socialization" skills above academic success when hiring and promoting. That is, they are more interested in reliability, respect for authority, work ethic, and human relations skills than in knowledge of mathematics, English, and history.

Our preliminary results partially confirm and partially conflict with this conventional wisdom. For both the full- participation cohort and its urban at-risk subcohort the standard regression measures for goodness of fit are low. Most variations in wage rates across individuals cannot be explained by our models, suggesting that the initial process of matching young people to jobs after they leave school is highly random. To simplify slightly, job seekers appear to accept first offers and employers to fill openings with the first applicants.

For the full participation cohort, however, a number of variables had statistically significant coefficients. For example, youth who did more homework, participated in athletics, avoided getting in trouble with the law or school authorities, and were in nonacademic tracks (i.e., vocational or general) received higher wages than other students (all else being equal). However, the coefficients of the grades and test score variables were not significant, nor, surprisingly, was the variable indicating possession of a high school diploma. Our results, thus, tend to confirm the conjectures of earlier researchers that nonacademic skills are more important than academic skills to employers of non-college youth.

In addition to these high school variables, we found that sex, race (in part), family background, marital status (particularly for males), work experience, and local labor market conditions had an effect on the wages received by this group.

In the case of urban at-risk youth, no high school variables and very few control variables had a statistically significant influence on wage rates. A tentative conclusion is that the job matching process for this group is especially random and, therefore, no relationship exists between an urban at-risk student's high school performance and the wage he or she receives after leaving school. Given the relatively short time horizon of many young people living in urban areas, this finding may explain why so many are hostile to the academic and disciplinary requirements of secondary education.

Our findings suggest a number of policy options involving intervention on either the supply or demand side of the labor market. On the supply side, schools could do more to prepare students for work and, more importantly, help place them in jobs after graduation by developing long-term relationships with local employers. On the demand side, firms could be encouraged to more thoroughly evaluate applicants and reward those with good academic and nonacademic skills with higher wages. This could promote good discipline and academic achievement, particularly in urban areas where many students are at risk of educational failure.

*****