The empty opportunity : local control of secondary schools and student achievement in the Philippines

- ---- / i ' t ~* , ! X Po0Iby Research WORKING PAPERS 9/ S Education and Employment Population and Human Resources Department The World Bank January 1992 WPS 825 The Empty Opportunity Local Control of Secondary Schools and Student Achievement in the Philippines Marlaine E. Lockheed and Qinghua Zhao Decentralization alone does not produce local control of schools. Schools must also be given resources. motivated students, edu- cated and experienced teachers, and control over teachers and school management. The Policy Research Workung Papets disseminate the findings of work in progress and encourage the exchange of ideas among Bank staff and all Dothers interested in development issues. These papers. disinbuted by the Research Advisory S,aff, carry thc namcs of the authors. reflector,lytheirviews,andshouldbeused and cited accordingly The findings,interercations, and conclusio s arethe authors'own 'Ihey should not be attnbuted to the World Bank. its Board of Directors, its m-inagement. or any of its member counuies  Policy Repyrch Education and Employment WPS 825 This paper -a product of the Education and Employment Division, Population and Human Resources Department-is part of a larger effort in the Department to understand the education sector, with particular reference to improving school effectiveness. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Darielle Eugene, room S6-224, extension 33678 (37 pages). January 1992. Lockheed and Zhao use a multilevel model to schools. Students in private schools outper- examine: formed students in govemment schools (0.88 points higher in mathematics). These differences - Differences in achievement and attitudes were attributable largely to the effects of student among grade 9 mathematics and science students selection. in 213 national govemment, private, and local schools in the Philippines. Lockheed and Zhao found that policies for centrally planned decentralization do not neces- Differences among these types of schools in sarily change what goes on in schools. Local social composition, available resources, class- schools were not managed as private schools. room orderliness, academic emphasis, and Local schools were given an ernp., opportunity: school decisionmaking. there was nothing for local control to control. Local schools had few resources - fewer of * Possible reasons for differences in achieve- them had laboratories and their teachers were ment. less educated and experienced than those in private schools. They found that - holding constant for age, gender, and socioeconomic status - students By contrast, managers of private schools had attending the three types of schools differed significant resources over which to exercise significantly. control. Teachers were better educated and experienced, and planned their instruction. Students in local schools scored lower in Students were motivated and completed their achievement (1.25 points lower in science and homework and assignments. And managers of 1.61 points lower in mathematics) and had less private schools exercised significant control over positive attitudes than students in government teaching and school management. Thc Policy Research Working Paper Series disseminates the findings of A kok under Ax ay in the Bank. An objcctive of the scries is to get these findings out quickly, c.en if presentations are less than fully po ished. Thc findings, interpretations, and conclusions in these papers do not necessa.ily represent official Bank policy. I'roduced by the Policy Research D)issemination Center  Introduction ............................................................ 1 Secondary Schooling in the Philippines .......................... 2 Comparative Effectiveness of Government and Private Schools .....2 What Accotvat.; for the Greater Effectiveness of Private Schools? ..................................................5 Method ............................................................... 10 Background ..................................................... 10 Models ......................................................... 12 Sample and Data ................................................ 13 Student Variables .............................................. 14 School Variables ............................................... 15 Results .18 Achievement Differences Between Schools ..... .................. 18 Other Differences Between Schools ................. 19 Explaining Differences Between Schools ................. 21 Summary and Conclusion ................. 31 Annex A ................. 34 References ................. 35 The World Bank does not accept responsibility for the views expressed herein, which are those of the authors and should not be attributed to the World Bank or to its affiliated organizations. The findings, interpretations, and conclusions are the results of research or analysis supported by the Bank; they do not necessarily represent official policy of the Bank. The comments of Linda Dove, Paul Glewwe, Stephen Heyneman, William Loxley, Neville Postlethwaite and the contribution of the IEA in making the data available to us are gratefully acknowledged.  - I - INTRODUCTION 1. Decentralization policies are at the heart of education reform effoits in many countries internationally. Two important types of policies are those that: (a) remove barriers to private education and (b) devolve authority and responsibility for schools from central level administrations to intermediate level organizations and ultimately to schools, relying more on local communities for school financing, with an overall goal of improving school effectiveness. While both types of policies are largely uninformed by empirical evidence regarding their iwpact on such education outcomes as student learning, in developing countries the evidence regarding the effects of local control is much weaker than that regarding private schools. This paper extends the literature on the impact of private education on achievement, while providing the first evidence on the impact of local control on achievement in a developing county. It analyzes data from 214 secondary schools in the Philippines to answer questions regarding (a) the relative effectiveness of local, government and private secondary schools, and (b) the factors that account for observed differences.  -2- Secondary Schooling in the PhiliRpines 2. Only about 40 percent of secondary school age youth are enrolled in secondary schools in developing countries. The vast majority of these students attend schools operated by national authorities (World Bank 1990). In the Philippines, secondary education covers 65 percent of the age cohort and is provided by three types of schools: private, national government public and local public, including village or baranguav, schools (Tan and Mingat 1989). Private schools are schools financed and managed non-governmentally; national government schools are publicly financed and managed schools identified as "national", "provincial" or "city" schools; local schools are baranguav or municipal schools. Laya (1987) reports that, in 1985, private, national government and local schools accounted for 42, 21, and 37 percent of all secondary enrollments, respectively. In the early 1980s, baranguav schools were more common in rural regions than in urban ones; Tan (1991) notes that they accounted for fewer than 4 percent of all secondary schools in the Metro Manila region, but over one half of all secondary schools in the Southern Tagalog. Baranguay schools. were originally set up as community self-help schools maintained by villages through communit' -ntributions in money and kind. The result was that the cost per student was significantly lower in all types of local schools (400 pesos in 1985) than in government schools (1570 pesos); per student costs in baranguav schools were lower than the average for local schools (Laya 1987). As these resources proved inadequate, baranguay schools have been recently nationalized. Comparative Effectiveness of Government and Private Schools 3. Expanding the provision of secondary education to a larger proportion of  -3- youth without lowering school quality or significantly increasing national education budgets present: a serious challenge to developing countries. Policy alternatives to nationally funded secondary schools may be necessary; two such alternatives are (a) relying on private schools to deliver secondary education and (b) devolving responsibility for education finance to local communities. Devolution of financial responsibility often carries with it an implicit expectatio(L that educational responsibilities will also be devolved. Locally controlled schools should mirror private schools in their finance, management and educational effectiveness. 4. Private school effectiveness. Research on private education in both developed ana developing countries indicates that, on average private schools are more effective and efficient than public schools. In North America, both private sectorian (Catholic) and elite non-sectarian private schools are more effective than public (government) schools in raising student achievement (Chubb and Moe 1989; Coleman, Hoffer and Kilgore 1982; Coleman and Hoffer 1987; Cookson and Percell 1985; Lee and Bryk 1989). Catholic private schools are also effective in enhar-ing equality, by reducing the gap in achievement between white and black students (Lee and Bryk 1989). Similar achievement effects have been reported for private schools in other developed countries (for Australia, Williams and Carpenter 1991; for the Netherlands, vanLaarhoven et al 1987) 5. In developing countries, less research on the comparative effectiveness of public (government) and private schools has been conducted, but the few available studies indicate that private schools are more effective than public schools in the third world as well (Jimenez, Lockheed and Paqueo 1991). For example,  - 4 - Jimenez, Lockheed and Wattanawaha (1987) found that, after controlling for 1.:evious achievemen , socioeconomic background and systematic selection by school type, students who were enrolled in private schools in Thailand significantly outperformed those enrolled in public schools; the difference amounted to 1.5 standard deviations. In the Dominican Republic, the advantage of private education was observed even for non-elite private schools (Jiminez et al. 1991). In Chile, students enrolled in private schools that were not subsidized by the government performed nearly twice as well on tests of reading and mathematics as did students in public schools (Rodriguez 1986). 6. Local school effectiveness. Two types of public schools are common in developing countries: national government schools and local community schools. Despite repeated calls for decentralization, little research on the comparative effectiveness of local versus national public schools has been carried out in developing countries. Yet in many countries, expansion of secondary education has depended upon such local schools. Examples include harambee schools in Kenya, local "district council" schools in Zimbabwe, and baranguay schools in the Philippines. These schools have expanded in numbers dramatically. In Kenya, communities were encouraged to build secondary schools on a "self-help" (harambee) basis; the demand for secondary education was so great that the number of harambee secondary schools increased from 557 in 1975 to nearly 1500 in 1985 (Eshiwani undated). 7. A similar expansion of secondary education was observed in Zimbabwe following independence; the number of secondary schools increased from 197 in 1980 to 1502 in 1989 (Ministry of Education, Zimbabwe 1990). Of these, 87  -5- percent were non-government schools and two thirds were local district council schools. District council e.chools, which typically enroll students who were unable to obtain a place in a narional secondary school, are managed by a local autnority, rather than by a national authority. The building site and construction of "district council" schools is contributed by the local community, but teacher salaries and some recurrent costs are financed nationally. 8. Characteristics of these local schools are simnilar to those of private schools: in comparison with public schools, the schools are often smaller, the teachers are less weii trained and paid, a-id the parents are more highly motivated to support the school through local contributions or school fees. Unlike elite private schools, however, which are highly selective and enroll students from comparatively advantaged backgrounds, local public schools are not selective and chey often are found in disadvantaged areas. Recent research suggests that achievement gain for students in local district council schools is not dissimi'lar to that of students in other types of secondary schools in Zimbabwe, although their initial level of achievement is lower (Riddell and Nyagura 1991). Local financing was also found to be related to increased efficiency for schools in the Philippines (Jimenez, Paqueo and de Vera 1988). What Accounts for the Greater Effectiveness of Private Schools? 9. Explanations for the observed difference in achievement between public and private schools are of two types: (a) those that have implications for improving (local) public secondary schools, and (b) those that do not.' Three lAlthough there is still considerable methodological debate about whether private schools are indeed more effective than public schools (Mhurnane 1984), this debate has not affected the proliferation of explanations for the differences.  explar.ations for private schools' apparent superiority that hold little promise for improving local public schools are: selectivity on the part of .ihools and parents in choosing the students and sc-ools, peer effects associated with this selectivity, and an historical stock of material and nonmaterial resources that are too expensive to replicate in local schools. Three explanations for the apparent superiority of private schools that may have implications for improving public education are: their emphasis on academic achievement, their more orderly environment, and their school-level control over decision making. This section reviews all six explanations. 10, Selectivity. The most important explanation for differences between the comparative effectiveness of public and private schools (and, parenthetically, between local and national government schools) is the difference in the composition of their student bodies. Generally speaking, students in private schools come from more advantaged backgrounds than do students in public schools, although there are some countries in which national public schools "cream" the better students (e.g. Tanzania, Cox and Jimenez 1990). It is therefore difficult to attribute differences in students' achievements to school characteristics alone, because a variety of nonschool factors also affect achievement. These factors include students' socioeconomic background, innate ability and individual motivation. Thus, unless nonschool factors are controlled appropriately, estimates of school effects will be contaminated by selectivity bias. Recent research has sought to control for selection effects through the use of modern statistical techniques; such studies with specific controls for selectivity have continued to show an advantage to private education, although peer effects have been pronounced (Jimenez, Lockheed and Paqueo 1991).  -7- 11. Peer effects. Peer effects have been widely recognized as contributing to differences in levels of achieve.oent between public and private schools. The average social class background of students in the school has been found to affect the average achievement of students in public and private schools in the United States (Lee and Bryk 1989), Thailand (Jimenez, Lockheed and Wattanawaha 1988) and the Dominican Republic (Jimenez et al. 1991). 12. Material and non-material resources. Material and non-material inputs are positively and significantly related to student achievement in developing countries (Heyneman and Loxley 1979; Fuller 1987; Lockheed and Verspoor 1991). Of particular importance are the availability and use of textbooks, the quantity of instructional time, formal educational attainment of teachers, and -- in some cases -- teachers' experience. Expenditures per student, which are unrelated to student achievement in developed countries (Hanushek, 1986), are also important. 1V. Available research on differences Letween public and private schools does not show consistent greeter resource availability for private schools, however. First, unit costs for students in private schools are substantially lower than those in public schools (Jimenez, Lockheed and Paqueo 1991). Second, even specific inputs do not ne:essarily favor private schools. For example, in Thailand, although private school teachers were more experienced and twice as likely to have received some type of inservice training than public school teachers, fewer private school teachers were formally certified to teach mathematics (Jimenez, Lockheed and Wattanawaha 1988). In the Dominican Republic, although more than twice as many students in both types of private schools had textbooks in comparison with students in public schools, teachers in non-elite  - 8 - private schools were less educated and experienced than teachers in public schools (Jimenez et al. 1991). Because of their apparent greater effectiveness, working with fewer resources, some analynts have concluded that private schools are more efficient chan public ones (Jimenez, Lockheed and Paqueo, 1991; Chubb and Moe, 1989). 14. Emphasis on academic achievement. One explanation for the higher achievement in Catholic private schools versus public schools in the United States is that they place greater emphasis on engagement in academic activities, including higher rates of enrollment in academic courses. This, in turn, translates into such differences in student behavior as spending more time on homework (Coleman, Hoffer and Kilgore 1982). In developing countries, curricula are typically set nationally, and students have little choice over course selection. However, differences in the emphasis placed on academic achievement may vaty between schools, and this may translate into differences between public and private schools in the level of effort spent bv students on academi( ac:tivities. 15. Orderly environment. The effective schools literature notes repeatedly that schools with orderly environments have higher achievement (Purkey and Smith 1983). aiools in developing countries seldom suffer from the types of discipline problems that characterize many poor performing schools in developed countries They do suffer from teachers whose lessons are unplanned and who do not m,nitor or evaluate their students' progress. Private schools in developing countries appear to provide a more orderly environment for learning than do public schools. For example, in Thailand, priv- e school teachers reported  -9- apending more time maintaining order in their classrooms and more time testing their students than did teachers in public schools (Jimenez, Lockheed and Wattanawaha 1988) 16. Local control. Public and private schools differ significantly in terms of their management organization. In developing countries, seventy percent of secondary educatian is publicly provided, with schoois financed and managed by the central government. Teachers are hired and deployec by a central agency, curriculum is set nationally, and admission to secondary school is often controlled by national examinatior. with students placed in schools through central agencies. As a result, neither the local community nor the schcol principal exercises much control over key decisions, and inefficiencies are observed. Unlike centrally controlled public schools, private schools in both developed and de-eloping countries exercise managerial control over a wide range of decisions. For example, research has found that in U.S. Catholic private schoois, principals, teachers and parents have significantly greater control over decisions about the curriculum, instructional methods, allocating funds, hiring teachers, dismissing teachers, and discipline policies than do their counterparts in public schools (Hannaway 1991). Hannaway concludes that "there is something about public educational institutions that restricts their adaptation to local conditions" (Hannaway 1991, p. 122). 17. If locally controlled schools could adopt the management practices of private schools, they might be able to provide secondary education to students in developing countries with greater effectiveness and efficiency than is presently the case. At present, however, there is no available research that  - 10 - addresses the three questions posed in this paper: (a) how does the achievement of students in local secondary schools compare with that of students in either government or private secondary schools, and (b) how do these types of schools compare in terms of inputs and management, and (c) what characteristics of the schools account for any observed differences in achievement? 18. This paper contributes to the literature in three ways: (a) by exploring a larger variety of school types (national government public, local public and private schools) than previously examined in either developed or developing countries, (b) by extending the range of outc1me variables examined (achievement and attitudes), and (c) by using an appropriate multi-level model for examining school effects. It uses a hierarchical linear model (HLM) to estimate the effects of priveate, local public and national government public schools on student science achievement in the Philippines (see Raudenbush and Willms, 1990, for discussion of multi-level modelling). METHOD Background 19. Although school effects research requires the use of multi-level methods (see Aitken and Longford, 1986; Raudenbush, 1988 for reviews), most previous research on private school effects has been conducted with single level models (e.g. Coleman, Hoffer and Kilgore 1982; Willms 1985; Alexander and Pallas 1985). Even efforts that have used sophisticatee statistical techniques to adjust for sel! ctivity have employed single-level models (e.g. Jimenez, Lockheed and Wattanawaha 1988; Jimenez et al. 1991). The situation, however, is changing.  - 11 - 20. Recently, Lee and Bryk (1989) employed a multilevel model to estimate effects of Catholic schools on average secondary school achievement in the United States, and found significant sector effects for both average achievement and the achievement gap between black and white students. Lockheed and Bruns (1990) also employed a multi-level model to estimate school type effects on achievement in secondary schools in Brazil; private schools were more effective than other types of schools with respect to mathematics achievement. In Zimbabwe, Riddell and Nyagura (1991) examined school type effects on secondary achievement gain (from Form 2 to Form 4) and found significant positive effects for private schools. No research has examined differences in effectiveness of local versus national public schools, however. 21. The analysis in this paper seeks to determine the extent of differences in achievement and attitudes between students in government, private and local (largely baranguay) schools in the Philippines, and the possible causes of these differences. To do this, a multi-level modelling package, HLM, is used (Bryk, Raudenbush, Seltzer and Congdon, 1988). One advantage of the HLM procedure over ordinary least squares (OLS) is that it correctly estimates the standard errors for the school-level coefficients, so that the statistical significance of school-level variables is correctly estimated. A second advantage of multi-level modelling is that it models within-school relationships, such as within school correlations between social class and student achievement. It is therefore possible to examine the extent to which school characteristics aggravate or diminish within-school social class differences, should they exist.  12 - Models 22. We model two elements of achievement within schools: student outcomes and socio-economic status (SES) differentiation. The grade 9 student outcomes considered in these analyses are science and mathematics test scores, positive and negative attitudes toward science, and positive and negative attitudes toward school. For each of these outcomes, we examine its within-school correlation with SES. 23. The within-school model holds constant sex and age, and regresses science and mathematics achievement for student i within school j as a function of socio- economic status: (1) ACHj - 6jo + Ojl SES + eij (2) ATTij - o + Pjl SES + eij where ACH refers to science and mathematics achievement, and ATT refers to positive and negative attitudes towards science and school. 24. The achievement (attitudes) in each school is characterized in terms of two parameters: an intercept and one regression slope. Achievement (attitude) scores are continuous variables centered around their school means. The two parameters may be interpreted as follows: Pjo - Mean achievement (attitudes) for students in school j.  - 13 - Aj1 - The degree to which SES differences among students relate to subsequent achievenment (attitudes). 25. Effective schools would be characterized as simultaneously having a high average level of achievement, fji and a weak differentiation effect with regard to SES (i.e., a small value for Pjl). These effects are hypothesized to vary across schools as a function of sector (national public, local public and private) and school-level differences in social composition, material and non- material resources, emphasis on academic achievement (student motivation and effort), orderly environment and local control. Sample and Data 26. The data come from the IEA Second International Science Study, conducted in the Philippines in 1983. A two-stage stratified sampling design was used. Stratification was based on thirteen geographical regions oL the Philippines2. Within each region, schools were classified into public (government-supported) and private (supported by private funds) schools. Within the public sector, the schools were classified into two further strata: Barangay/Municipal High Schools (referred to in this paper as "local schools") and National, Provincial and City High Schools (referred to in this paper as "government schools"). This gave a total of 39 strata. At the first stage, schools were selected with probability proportional to the number of classes; at the second stage, one intact grade 9 class was chosen by simple random sampling. A total of 269 schools and approximately 10,000 students participated in the study. After cleaning, 2,locos Region, Cagayan Valley, Central Luzon, Southern Tagalog, Bicol, Western Visayas, Central Visayas, Eastern Visayas, Western Mindanao, Southern Mindanao, Southwestern Mindanao, Manila  - 14 - acceptable data for analysis were obtained from 214 schools and 8736 students.3 Student Variables 27. Dependent Variables. This study examines school effects on two student achievement and four student attitude outcome variables. The achievement variables are the science and mathematics core test scores, unadjusted for guessing, from the IEA Second International Science Study. The attitude scales were constructed from factor analyses of the student attitude survey from the same study. For this study, attitude items were recoded (1 - agree, 0 - uncertain, and -1 - disagree) and factor scores were constructed from these recoded items using data from the total student sample. The minimum number of cases for the attitude factors was 10,222 students; for this sample all factor scores have a mean of 0 and a standard deviation of 1. Positive attitude toward science includes eight items of the type "Science is an enjoyable school subject". Negative attitude toward science is comprised of four items of the type "Scientific discoveries do more harm than good." Positive attitude toward school is comprised of four items of the type "I enjoy everything about school", and negative attitude toward school is comprised of four items of the type "School is not very enjoyable." 28. Student-level predictors. In this paper, we analyze three student-level variables: gender, age and socio-economic background (SES). SES is a factor score with a mean of 0 and a standard deviation of 1 and is based on father's 3The original school data file contained 269 records for which data from the teacher questionnaire, the school questionnaire and mean values from student records could be matched. Thirty-four schools were deleted due to missing data on five or more veriables; 17 schools were deleted for lack of within school variation; end four schools were deleted for other miscellaneous reasons.  - 15 - occupation, mother's occupation, father's education and mother's education. Additional individual-level variables -- family size, home language, availability of reading materials in the home -- were too highly correlated with SES to be analyzed, although we report summary statistics on these variables as well. School Variables 29. Given the available data, measures were developed to indicate features of the school that have been found to be related to average achievement and within- school achievement differentiation. The variables have been grouped into five categories: the social composition of the school, material and non-material inputs, academic emphasis, orderly environment and local control. 30. School social composition. Six social composition variables were created. Average family size (the percentage of families in the school with more than 5 children), dialect as home language (percentage of students from dialect-speaking family), English as home language (percentage of students from English-speaking family), average age (mean age of the class), average availability of reading materials in home (mean number of books in the home), and average SES. 31. Material and non-material inputs. The IEA data set includes a wealth of teacher and teaching variables, but relative few variables that measure actual inputs. Variables selected for analysis in this paper represent only a few of those available. They are: teacher education (whether or not the teacher has studied post-secondary science), teacher experience (number of years teaching experience), class size (number of students in science class), student time on "experiments or field work" (whether or not students spend more than half their  - 16 - time in science on these activities), laboratory use, (whether or not students are taught science in a laboratory more than 60% of the time), frequent use of textbooks (whether or not teacher uses textbooks frequently), and use of small groups for instruction (whether or not teacher uses small groups). 32. Orderly environment. Two indicators of an orderly environment for teaching are analyzed: frequent testing (whether or not teacher uses teacher made tests frequently) and instructional planning (mean factor score of student report of teachers teaching style which emphasizes advance organizers, summaries and demonstrations). 33. Academic emphasis. Variables indicating an academic emphasis in the school are derived from student reports regarding their level of effort and autonomous study. Student motivation is a factor score derived from responses to four items about frequency of checking homework, trying hard on assignments, doing homework and handing it in on time. For this scale only, a high score represents less effort. Student active learning is a factor score derived from responses to five items about the extent to which students choose topics for study, make up problems, consult reference materials, and influence the topic of lessons. 34. School decision making. Three variables related to school-level decision- making were constructed; for each, a higher score indicates greater localization and less centralization of decision making. Areas covered include local control over teaching, factor score indicating degree of local control over curriculum and instruction (range and type of subjects taught, course content choice of textbooks); local control over management, a factor score indicating degree of  - 17 - local control over expenditures and teacher selection' and local control over students, a factor score indicating degree of local control over selection and management of students (selecting students, determining fees, making rules for students). Table 1: Description of Variables used in HLM Analysis, Philippines 1983 Student-level Dependent Variables Science test score: Science achievement score (range - 0 - 30) Mathematics test score: Mathematics achievement score (range - 0 - 20) Positive science attitude: Factor score based on 8 positive statements about science Negative science attitude: Factor score based on 4 negative statements about science Positive school attitude: Factor score based on 4 positive statements about school Negative school attitude: Factor score based on 4 negative statements about school Student-level Predictors Male: A dummy variable (1 - male; 0 - female) Age: Age in months SES: Factor score based on father's occupation, mother's occupation, father's education and mother's education English: A dummy variable (1 - English spoken at home; 0 - other) Pilipino: A dummy variable (1 - Pilipino spoken at home; 0 - other) Dialect: A dummy va_iable (1 - dialect spoken at home; 0 - other) Books: Number of books in home (1 - Large family: Number of children in family (1 - 5 + children in family; 0 other) School-level Predictors 1. School social composition Large families: I families with 5 or more children Dialect average: 2 students from dialect-speaking family English average: 2 students from English-speaking family Average age: Average age of the class Average books: Average number of books in the home Average SES: Average SES of the class 2. Material and non-material inputs Teacher post-secondary science: A dummy variable (1 - teacher studied post-secondary science) Teacher experience: Average number of years teachers have taught Class size: Number of students in class Student practice : A dummy variable (1 - students spend > 502 of time on experiments or fieldwork; 0 - other) Laboratories: A dummy variable (1 - teacher teaches * 602 of time in laboratory; 0 - other) Textbooks: A dummy variable (1 - teacher uses textbook; 0 - other) Groups: A dummy variable (1 - teacher uses small groups for instruction; 0 - other) 3. Orderly environment Frequent tests: A dummy variable (1 - teacher makes test; 0 - other) Instructional Planning: Average factor score of student report of teachers teaching style 4. Academic emphasis Student motivation: Average factor score of students' responsibility regarding homework and assignments (reverse) Student active learning: Average factor score of students' active learning 5. School decision-making Local teaching: A factor score indicating degree of local control over curriculum and instruction Local management: A factor score indicating degree of local control over school management Local student control: A factor score indicating degree of local control over selection and management of students  - 18 - RESULTS 35. In this section we discuss, first, the observed achievement differences between private, government and local secondary schools. Second, we discuss the observed differences between the three types of schools in terms of their available resources and social context. Then we present the results of the HLIF analyses, which explore reasons for the achievement differences. Achievement Differences Between Schools 36. Private, government and local secondary schools in the Philippines differ in terms of their average science and mathematics achievement and in terms of the- average attitudes of their students. 37. Achievement. Average science achievement in the Philippines was the lowest of all countries that participated in the IEA study, with an average for all students of 11.5 points on the 30-point core test (IEA 1988). The science scores of students in both government public and private schools in this analysis are slightly higher than the national average reported in the IEA report. However, science achievement in local public schools is 1.5 points lower than the national average, and nearly two points -- approximately one-half standard deviation -- lower than in private and government schools. The differential in mathematics achievement is even greater, with students in local schools scoring nearly a full standard deviation below those in private schools. 38. Attitudes. Student attitudes towards science and school are also less  - 19 - positive in local schools, and these differences range from one- fourth to one-half of a standard deviation. The direction of the signs of the average attitude factor scores is important. For all four attitudes, students in local public schools hold fewer positive attitudes and more negative attitudes, while the reverse is the case for students in government public and private schools. Other Differences Between Schools 39. Government, local and private schools also differ in the types of students who attend them and their available resources and social context. In general, students in local secondary schools are disadvantaged in comparison with students who attend private or government secondary schools (See Table 2). 40. Student characteristics. Students in local schools are more likely to come from more disadvantaged backgrounds than students in either private or government schools. In terms of socio-economic status, the SES background of local school students is more than three-quarters of a standard deviation lower than that of students in private schools and two-thirds lower than that of students in government schools. Their homes have fewer resources that are supportive of school. Students in local schools are less likely to speak English (the language of instruction and in which they were tested) at home and more likely to speak a local dialect. They report having fewer books in their homes; they report having more sibs; they are four to six months older, on average, than those in either private or government schools, which can be the consequence of either starting school late or of repeating a grade. Approximately 40% of all students in all schools are boys.  - 20 - Table 2: Means and standard deviations of student-level variables used in HLM analysis for private, government and local schools in the Philippines, 1983 Private Government Local (N - 2960) (N - 3470) (N - 2306) Variables Mean S.D. Mean S.D. Mean S.D. Student-level Dependent Variables Science test score 11.86 4.30 11.88 4.80 10.11 4.10 Mathematics test score 11.37 3.66 10.55 3.39 8.65 3.58 Positive science attitude 0.12 0.91 0.07 0.97 -0.30 1.10 Negative science attitude -0.12 0.89 -0.13 0.92 0 .3 1.14 Positive school attitude 0.07 0.94 0.04 0.97 -0.15 1.08 Negative school attitude -0.13 0.96 -0.02 1.01 0.14 0.99 Student-level Predictors Male (X) 41.96 49.36 40.49 49.09 41.07 49.21 Age in months 188.91 23.11 190.60 22.56 195.18 34.11 SES factor score 0.20 1.00 0.07 0.98 -0.58 0.70 English spoken at home (x) 0,41 6.36 0.35 5.87 0.22 4.65 Philipino spoken at home (Z) 36.01 48.01 30.55 46.07 21.34 40.98 Dialect spoken at home (2) 38.61 48.69 44.12 49.66 54.03 49.85 Books in the home 2.71 1.29 2.51 1.23 1.99 1.12 Large family 63.16 48.25 67.41 46.88 73,45 44.17 41. Social Composition. As a result of these family background differences, the social composition of private, government and local schools also differ from one another (Table 3). Local schools have a higher proportion of students that come from large families and that speak a local dialect; they have a lower proportion of students that speak English and have more than two books in the home. More of their classmates come from lower SES backgrounds. 42. Inputs. The three types of schools differ in a number of other respects as well, with local schools consistently disadvantaged. Teachers in local schools have fewer years of post-secondary science education and teaching experience; they teach less in laboratories and less frequently use small groups for instruction; their students report that their teachers are less likely to use a teaching style that emphasizes advance organizers, summaries and demonstrations.  - 21 - However, students spend more of their science class on practice, have smaller science classes and use textbooks more frequently. 43. Possibly as a consequence of differences in family background and school resources, student motivation4 is much lower in local schools than in private and government schools, and students in local schools are less responsible about completing thAir homework. However, students in local schools report being more actively engaged in their learning than are students in private or government schools. 44. With respect to school decision-making, both local and government schools report less local control over the curriculum and school management than do private schools. Local schools exercise slightly more control over the selection and management of students than do public or government schools, however. ExDlaining Differences Between Schools 45. In this section, we address four questions: (a) how much of the observed differences in student achievement and attitudes is attributable to student background and how much to characteristics of their schools? (b) do the average differences between the three types of schools remain after taking into account the family background differences of the students that attend them? (c) do they remain after taking into account peer effects (the contextual effects model), and (d) what other school characteristics may account for average differences in achievement? 4A high score represents less effort  - 22 - Table 3: Means and standard deviationc of school-level variables used in HLM analysis, for private, government and local schools in the Philippines, 1983 Private Government Local (N - 70) (N 83) (N - 61) Variables Mean S.D. Mean S.D. Mean S.D. 1. School social composition Large families 0.64 0.16 0.68 0.12 0.74 0.11 Dialect average 0.40 0.36 0.45 0.34 0.55 0.32 English average 0.11 0.11 0.10 0.09 0.05 0.08 Avorage age 189.64 4.91 191.22 5.42 197.05 7.00 Average books 2.68 0.63 2.47 0.48 1.97 0.32 Average SES 0.18 0.53 0.02 0.51 -0.61 0.32 2. Material and non-material inouts Teacher post-secondary science 0.73 0.45 0.73 0.44 0.49 0.50 Teaching experience 10.60 6.86 10.84 5.97 8.26 4.88 Class size 42.03 8.93 41.49 7.89 37.25 8.75 Student practice 0.61 0.49 0.66 0.48 0.69 0.47 Laboratories 0.70 0.46 0.63 0.49 0.48 0.50 Textbooks 0.31 0.47 0.22 0.41 0.43 0.50 Groups 0.33 0.47 0.37 0.49 0.16 0.37 3. Orderly environment Frequent testing 0.77 0.42 0.77 0.42 0.74 0.44 Instructional planning 0.03 0.44 0.04 0.39 -0.11 0.54 4. Academic emphasis Student motivation (reverse) -0.08 0.23 -0.07 0.27 0.25 0.42 Student active learning -0.13 0.43 -0.02 0.37 0.29 0.33 5. School decision-making Teaching control 0.51 0.98 -0.38 0.80 -0.16 0.94 Management control 0.24 0.84 -0.03 1.00 -0.19 1.05 Students control -0.11 0.93 -0.14 0.84 0.00 0.97 46. The Unconditional Model. The first step in the HIM estimation process involves fitting an unconditional, or random regression model. We do this in two stages. In the first stage, we partition the variance in the achievement and attitude scores into their between unit (school) and within unit (individual) components. The results are presented in Table 4; they show that school level factors account for approximately half of the variance in both science and mathematics achievement, but very little -- only 10-20 percent -- of the variance in attitudes.  * 23 - Table 4: Results of HLM variance component analysis: Philippines secondary, 1983 (percent of total variance accounted for) Score School Individual Science teat score 43 57 Math*matics test score 52 48 Positive attitude toward science 21 79 Negative attitude toward science 17 83 Positive attitude toward school 10 90 Negative attitudw toward school 11 89 47. In the second stage, for each of the six outcome measures, we fit an unconditional model that includes one random student level variable (SES) and two fLxed student level characteristics (sex and age)5. Table 5 summarizes the results from the six models. All three student-level characteristics were significant predictors of science and mathematics achievement; male students scored significantly higher on both tests than did female students, although the male advantage in science was only about 15% of a standard deviation and in mathematics only about 10% of a standard deviation, neither of which is considered meaningful (Cohen 1969). Male students also held less positive attitudes toward both science and school than did female students. Students from higher SES backgrounds had higher achievement and more positive attitudes than did students from lower SES backgrounds, ceteris paribus. Although the residual variance of SES was allowed to vary among schools, the SES differentiation estimates were not reliable (reliabilities range from .01 to .12). Therefore, in the remaining analyses, SES is treated as a fixed variable (residual variance set at zero). Significant school-level random effects exist even after controlling for individual level student characteristics, and the achievement and attitude estimates are quite reliable (.62 to .94). These differences between 5The other student background characteristics were too highly correlated with SES to be included as independent variables.  - 24 - schools indicate that we can proceed to the second step in the HLM analysis (the Chi-square chart indicates that all estimated parameter variances are significantly different from zero; see Annex A). Table 5: Parameter estimates from six HLM unconditional models, Philippines secondary, 1983 (t-statistics in parentheses) Positive Negative Positive Negative Science Mathematics attitude attitude attitude attitude test test toward toward toward toward Independent variable score score science science school school Individual level effect Male (fixed) 0.79 0.34*** -0.10 0.23*** -0.18 0.24*** (10.95) (5.89) (5.26) (11.96) (8.53) (11.65) Age (fixed) -0.01*** -0.01*** -0.00 0.00*** -0.00 0.00 (4.51) (5.43) (0.33) (3.68) (0.26) (1.69) SES (random) 0.54*** 0.26*** 0.12*** -0.06*** 0.07*** -0.00 (10.63) (6.78) (9.67) (4.96) (5.52) (0.10) School level effects Mean 12.82*** 11.78*** 0.06 -0,47*** 0.09 0.29** (29.19) (33.23) (0.52) (4.46) (0.83) (2.65) Percent between-school 9.75 8.02 20.0 12.5 10.0 0.0 variance axplained *** p c .001; ** p C .01 48. School Type Effects Model. In the second step, we address the major purpose of this study: to explore differences in the achievement and attitudes of students attending different types of secondary schools. Comparisons were made, therefore, between the achievement and attitudes of those attending national public, local public and private schools. Two variables, "Local public schools" and "Private Schools", were added to each of the two between-school equations for both achievement and attitudes; national public schools are the omitted category. Table 6 summarizes the results of these analyses.  - 25 - Table 6: Parameter estimates from six HLM school type effects models Philippines secondary, 1983 (t-statistics in parentheses) Science Mathematics Positive Negative Positive Negative test test towasd toward toward toward Independent variable score score scietace science school school Individual level fixed effects Male U.79*** 34*** -0.10*** 0.23*** -0.18*** 0.24*** (10.91) (5.85) (5.18) (11.95) (8.51) (11.67) Age -0.01*** -0.01*** -0.00 0.00*** -0.00 0.00 (4.38) (5.36) (0.34) (3.55) (0.14) (1.53) SES 0.55*** 0.26*** 0.11*** -0.05*** 0,06*** 0,01 (12.23) (7.17) (9.08) (3.95) (4.94) (0.48) School level effects Mean 13.15*** 11.93*** 0.11 -0.56*** 0.11 -0.27* (26.03) (29.52. (1.01) (5.12) (0.95) (2.34) Local public -1.25* -1.61*** -0 29*** 0,40*** -0.16** 0.13* (2.48) (3.92) (4.04) (3.30) (2.74) (2.12) Private -0.26 0.88* 0.04 0.01 0.04 -0.13* (0.06) (2.22) (0.60) (0.10) (0.72) (2.37) Percent between-school 11.7 20.3 20.0 25.0 10.0 9.0 variance explained *** p C .001; ** p < .01; *p < .05 49. The signs of the effects for school type on average achievement and attitudes are generally in the expected direction. In local public schools, average science and mathematics achievement scores are lower than in national public schools, ceteris paribus. With student background held constant, students in local public schools score 1.25 points (about 25 percent of a standard deviation) lower in science and 1.61 points (about 50 percent of a standard deviation) lower in mathematics than students in national public schools. Effects of these magnitudes are both statistically significant and meaningful, according to Cohen (1969). Student attitudes are less positive in local schools as well. Students in local schools report less favorable attitudes towards both science and school in general than students in national public schools; the effects are statistically significant for all four attitudes. 50. With background effects held constant, students in private and national  - 26 - public schools perform comparably in science, but private school students outperform students in national public schools by .88 points in mathematics (about 25 percent of a standard deviation). Stud_nts in private schools also have somewhat -ore favorable attitudes towards science and school, but these effects are statistically significant only for one variable, negative attitudes towards school. 51. Contextual-Effects Model. According to Lee and Bryk (1989), a contextual effect in HLM is represented by including the school aggregate of a student-level variable in the between-school model; in this case, we include the class average SES score in the between-school models for average achievement and average attitudes. This .epresents the composition of students in each school with respect to their SES and approximates possible initial differences in achievement and selection effects. The results of these analyses are summarized in Table 7.  - 27 - Table 7: Parameter estimates from six HLM contextual effects models, Philippines secondary, 1983 (t-statistice in parentheses) Positive Negative Positive Negative Science Mathematics attitude attitude attitude attitude test test toward toward toward toward Independent variable score score science science school school Individual level fixed effects Male 0.79*** 0.34*** -0.10*** 0.23*** -0.18*** 0.24*** (10.92) (5.87) (5.15) (11.93) (8.50) (11.68) Age -0.01*** -0.01*** -0.00 0.00** -0.00 0.00 (4.31) (5.28) (0.19) (3.41) (0.08) (1.54) SES 0.54*** 0.24*** 0.10*** -0.04** 0.06*** 0.01 (11.79) (6.66) (8.04) (3.01) (4.48) (0.41) School level effects Mean 13.09*** 11.87*** 0.90 -0.54*** 0.10 -0.27* (26.31) (30.18) (0.82) (4.96) (0.88) (2.35) Local public -0.13 -0.43 -0.09 0.25*** -0.12 0.13* (0.23) (0.98) (1.24) (3,70) (1.80) (1.97) Private -0.28 0.61 0.00 0.04 0.03 -0.13* (0.60) (1.63) (0.05) (0.66) 0.55) (2.37) Average SES 1.81*** 1.90*** 0.32*** -0.24*** 0.07 0.01 (4.23) (5.62) (5.33) (4.42) (1.35) (0.22) Percent between-school 18.5 30.7 30.0 31.3 10.0 9.0 variance explained ***D < .001; **p <.01; *p <.05 52. Peer effects are powerful predictors of achievement in the Philippines. Average SES scores are significantly related to both mathematics and science test scores, and the observed differences in achievement among local, government and private schools are nearly entirely accountable to differences in the average SES of students in these schools. That is, the size and statistical significance of coefficients for "Local public school" and "Private School" disappear with the inclusion of "Average SES" in the between-school model. 53. Average SES scores are also significantly related to attitudes towards science, but they have no relationship to attitudes toward school in general. The difference in average SES between schools, however, does not account for the more negative attitudes towards science and schools held by students in local  - 28 - schools, and the more positive attitudes toward schools held by students in private schools. 54. Other exRlanations for achievement differences. Even though the school type effect was nearly entirely explained by the social composition of the school, there remains significant between-school variance in achievement to be explained. Some of the differences in achievement may be due to other measured differences among the schools. The last step in our analysis involves exploring how differences among schools with respect to material and non-material inputs, orderly environment, academic emphasis, and school-level decision-making affect the average achievement and attitudes of students. 55. For this, we employ a feature of the HLM package called "exploratory analysis", which estimates slopes and standard errors of variables, presently not included in a model, as if they were included. We do this for each variable separately, and we report in Table 8 all variables for which the estimated t- statistic is greater than 1.65 (2-tailed p < .10). The model onto which these variables were added is the "contextual effects" model.  - 29 - TablA 8: Estimated slopes (gama) and their standard *rrors obtained by regressing estimated Bayes residuals from contextuol effects model on between-unit variables Variables Gamsm Standard Gamma Standard Error Error 1. Related to achievement. Scienco Mathematics Teacher post-secondary science 0.86 0.40 n.s. n.s. Laboratories 0.73 0.39 0.90 0.30 Instructional planning 1.27 0.41 1.27 0.32 Student motivation (reverse) -1.57 0.55 -1.30 0.43 Student active learning -1.15 0.46 -1.11 0.36 Local studant control -0.44 0.21 -0.41 0.16 2. Related to positive attitudes Science School Teacher post-secondary science n.s. n.s. -0.07 0.04 Laboratories 0.10 0.05 n.s. n.s. Student practice n.s. n.s. 0.08 0.04 Instructional planning 0.35 0.05 0.29 0.03 Student motivation (reverse) -0.24 0.07 -0.20 0.05 Textbooks 0.12 0.05 0.07 0.04 Groups 0.14 0.05 0.14 0.04 3. Related to nexative attitudes Science School Instructional planning -0.25 0.04 -0.22 0.04 Student motivation (reverse) 0.35 0.06 0.15 0.06 Student active learning 0.17 0.05 n.s. n.s. Groups -0.13 0.04 -0.11 0.04 n.s. not significant 56. The exploratory analyses yield both anticipa-ed and unanticipated results. A few material and non-material inputs were important in boosting achievement. Science achievement was higher in schools with more scientifically educated teachers (students whose teachers studied post-secondary science scored nearly one point higher on the science test than students whose teachers lacked post- secondary science training), and both science and mathematics achievement were higher in schools where teachers had access to science laboratories (also, nearly one point higher). However, most of these variables -- class size, the teacher's experience, use of textbooks, use of small groups, student time on experiments -  - 30 - - were unrelated to student achievement in either subject. One measure of an orderly environment -- the students' report that the teacher plans his or her instruction (teacher explains work at the outset, summarizes work at the end, does demonstrations and explains relevance of work to students) was significantly related to achievement in both science and mathematics; each standard deviation of instructional planning was associated with a 1.27 point gain on both the science and the mathematics test. Achievement was also higher in schools with greater academic emphasis, in which students were more motivated and reported expending more effort on homework and in-class assignments; here, each standard deviation of student motivation was associated with a 1.57 point gain on the science test and a 1.30 point gain on the mathematics test. Unexpectedly, achievement was unrelated to greater local control over decision-making over teaching and school management and was negatively related to local control over student admission and regulation and to more active student learning. 57. Attitudes towards science were more positive for students whose teachers planned their instruction, used textbooks, taught in laboratories and placed students in small groups for work. With the exception of use of laboratories, these same teacher practices also affected students' attitudes towards school in general. Surprisingly, student attitudes were more negative about science when students were involved in more active learning and more negative about school when there was more local control over student admission and regulation. Variables unrelated to residu.l variance in student attitudes were: teacher experience, class size, teacher testing, student active learning and all other variables related to the local control of schools.  - 31 - SUMMARY AND CONCLUSION 58. This paper has examined the relative effectiveness of 214 national public (government) schools, local public schools and private schools on secondary science achievement and attitudes in the Philippines. Significant differences were found among these three types of schools in terms of student achievement and available resources; peer effects and the context for learning were examined for their mediating effects on school type differences. 59. First, with SES, age and gender held constant, significant differences were observed in the achievement and attitudes of students in the three types of schools: * Students in local schools had lower achievement (1.25 points lower in science and 1.61 points lower in mathematics) and less positive attitudes than their counterparts in government schools. * Students in private schools outperformed students in government schools (.88 points higher in mathematics). 60. Second, local public schools had fewer resources than either national public or private schools. In particular, fewer local public schools had laboratories and teachers in local schools were less educated and experienced.  - 32 - 61. Third, the paper explores reasons for these differences, starting with differences in the social composition of the school. Peer effects (average SES) were powerful predictors of achievement and attitudes towards science, although they had little effect on attitudes toward school. With contextual variables included in the models, virtually all school type effects disappeared. That is, no differences between local public, private and government schools with respect to either science or mathematics achievement were observed, once the average SES of students in the school had been entered into the model. However, significant residual variance between schools remained. 62. Correlates of the residual variance included several variables previously identified as significant determinants of achievement: the student's teacher studied science at the post-secondary level, planned his or her instruction, and used a science laboratory; the students were more motivated, and e,pended more effort on homework and assignments. However, local control over teaching, school management and students were unrelated or even negatively related to achievement. 63. At the outset, we hypothesized that if local public schools were able to harness the managerial strategies of private schools, they might be able to raise the achievement of their students. The results of this study indicate that centrally planned decentralization does not necessarily produce local level control. Local schools were not managed as private schools; they reported little local control over either teaching or school management - - much less than private schools reported. Student motivation was lower in local schools than in either private or government schools. Teachers engaged in less instructional planning than in either private or government schools. Thus, the opportunities presented  - 33 - by decentralization were not employed to improve student achievement in local schools. 64. Local schools were provided an empty opportunity, with nothing for local control to control. Laya (1987) notes that the per-student expenditures in local schools are significantly lower than those in government schools; this suggests that fewer resources were available about which to make local decisions. Data from the present study indicates that the resources available to school managers in local schools were less abundant than those in either private or government schools. Managers of under-supplied schools, such as the local schools in the Philippines it' the early 1980s, cannot easily compensate for absences of material and non-material inputs by managerial sleights-of-hand. Thay need the basic inputs with which to manage. 65. By comparison, managers of private schools had significant resources over which to exercise control. Teachers were educated and experienced, and they planned their instruction; students were motivated and completed their homework and assignments. Managers of private schools exercised significant control over decisions regarding teaching and school management. These results suggest that policies for decentralization alone do not necessarily change what goes on in schools.  - 34 - Annex A: Additional statistics Chi Square Table Statisticdl Reliability of significance school-level Parameter Estimated value Chi square (p-value) random effects Table 5 (213 degrees of freedom) Mean Achievement Science 8.90 4250.5 .000 .930 Mathematics 6.65 5352.0 .000 .940 Sciatt+ 0.16 1343.3 .000 .775 Sciatt- 0.14 1199.5 .000 .740 Schatt+ 0.09 753.3 .000 .626 Schatt- 0.11 871.7 .000 .664 SES Differentiation Science 0.10 243.9 072 .124 Mathematics 0.04 219.6 .363 .075 Sciatt+ 0.00 196.4 <.500 .052 Sciatt- 0.00 206.6 <.500 .056 Schatt+ 0.00 188.4 <.500 .011 Schatt- 0.00 173.9 <.500 .031 Table 6 (211 degrees of freedom) Mean achievement Science 8.70 6885.3 .000 .969 Mathematics 5.76 7388.4 .000 .971 Sciatt++ 0.16 1893.1 .000 ego Sciatt- 0.12 1482.6 .000 .858 Schatt4 0.08 1049.5 .000 .799 Schatt- 0.10 1199.3 .000 .817 Table 7 (210 degrees of freedom) Mean achievement Science 8.03 6195.8 .000 .967 Mathematics 5.01 6380.7 .000 .967 Sciatt+ 0.14 1644.6 .000 .875 Sciatt- 0.11 1355.7 .000 .846 Schatt+ 0.09 1040.6 .000 .798 Schatt- 0.10 1199.6 .000 .818  - 35 - References Aitkin, M., & N. Longford (1986). Statistical modelling issues in school effectiveness studies (with discussion). Journal of the Royal Statistical Society, Series B, 149:1-43. Alexander, K.L. & A.M. Pallas (1985). School sector and cognitive performance: When is a little a little? Sociology of Education, 58, 115-127. Arcelo, A.A. (1990). Financing education in the Philippines and its implications on planning and reform. Paper prepared for its International Conference on Education Finance, Beijing, Chira, March (canceled). Bryk, A.S., S. W. Raudenbush, M. Seltzer, & R. T. Congdon, Jr. (1986). An Introduction to HLM: Computer Program and User's Guide. Chicago: University of Chicago (processed). Chubb, J. & T. Moe (1985). Politics, markets and the organization of schools. Paper presented at the Annual Meeting of the American Political Science Association, New Orleans, August. Cohen, J. (1969). Statistical Power Analysis for the Behavioral Sciences. New York: Academic Press. Coleman, J., T. Hoffer, & S. Kilgore (1982). High School Achievement: Public. Catholic and Private Schools ComRared. New York: Basic. Coleman, J. & T. Hoffer (1987). Public and Private High Schools: The Impact of Communities. New York: Basic Cookson, P. W. Jr. & C. H. Persell (1985). Preparing for Rower: America's elite boardina schools. New York: Basic Cox, D. & E. Jimenez (1990). The relative effectiveness of private and public schools: Evidence from two developing countries. Journal of Development Economics, 34, 99-121. Eshiwani, G.S. (undated). Private secondary schools in Kenya: A study of some aspects of quality education. Nairobi, Kenya: Kenyatte University Bureau of Educational Research. Fuller, B. (1987). Raising school quality in developing countries: What investments boost learning. Review of Educational Research, 57, 255-292. Hannaway, J. (1991). The organization and management of public and Catholic schools: Looking inside the 'black box.' International Journal of Educational Research, 15, 463-482 Hanushek, E. (1986). The economics of schooling: Production and efficiency in public schools. Journal of Economic Literature, 23, 1141-1177.  - 36 - Heyneman, S. & W. Loxley (1983). The effect of primary school quality on academic achievement across twenty-nine high and low-income countries. American Journal of Sociolo-,y, 88, 1161-1194. IEA (1988). Science Achievement in 17 Countries. Oxford: Pergamon Press. Jimenez, E., M.E. Lockheed & N. Wattanawaha (1988). The relative efficiency of private and public schools: The case of Thailand. World Bank Economic Review, 2_, 139-164. Jimenez, E., M. E. Lockheed & V. Paqueo (1991). The relative efficiency of private and public schools in developing countries. World Bank Research Observer, 6, 205-218. Jimenez, E., M.E. Lockheed, E. Luna & V. Paqueo (1991). School effects and costs for private and public schools in the Dominican Republic. International Journal of Educational Research, 15, 393-410. Laya, J. and Associates (1987). Philivnine secondary education financing and rate of return, Bureau of Secondary Education and Secondary Education Development Budget, Manila, February (processed). Lee, V.E. & A. S. Bryk (1988). CurricultL tracking as mediating the socia. distribution of high school achievement. Sociology of Education, 61, 78- 94. Lee, V. E. & A. S. Bryk (1989). A multi-level model of the social distribution of high school achievement. Sociology of Education, 62, 193-207. Lockheed, M.E. & B. Bruns (1990). School effects on achievement in secondary mathematics and Portuguese in Brazil. Working Paper WPS 525. Washington, DC: World Bank, Population and Human Resources Department (processed). Lockheed, M.E. & N. T. Longford (1989). A multilevel model of school effectiveness in a develoRing country. World Bank Discussion Paper No.69, Washington, D.C.: World Bank. Lockheed, M.E. & A.M. Verspoor (1991). Improving Primary Education in Developing Countries. New York: Oxford University Press. Murnane, R. (1984). Comparisons of public and private schools: Lessons from the uproar. Journal of Human Resources, 20, 263-267. Purkey, S.C. & M.S. Smith (1983). Effective schools: A review. Elementary School Journal, 84, 427-452. Raudenbush, S.W. & A. S. Bryk (1986). A hierarchical model for studying school effects. Sociology of Education, 59, 1-17. Raudenbush, S.W. and J.D. Willms (1990). Schools. Classrooms and Pupils: International Studies of Schooling from a Multilevel PersRective. New York: Academic Press.  - 37 - Raudenbush, S.W. (1988). Educational applications of hierarchical linear models: A review. Journal of Educational Statistics, 13, 85-116. Riddell, A.R. (1989). An alternative approach to the study of school effectiveness in third world countries. Comparative Education Review, 33, 481-497. Riddell, A.R. & Nyagura, L. (1991). What causes differences an achievement in Zimbabwe's secondary schools? Working paper WPS 705. Washington, DC: World Bank, Population Human Resources Department (processed). Robitaille, D. F. & R. A. Garden (1989). The IEA study of mathematics II: Contexts and outcomes of school mathematics. Oxford: Pergamon Press. Rodriguez, J. (1986). School achievement and decentralization policy: The Chilean Case. Washington, DC: World Bank (processed). Tan, Edita A. (1991). Structure and performance of Philippine secondary schools. Tan, Jee-Peng and A. Mingat (1989). Educational Development in Asia. Washington, DC: The World Bank. vanLaarhoven, P., B. Bakker, J. Dron.ers & H. Schijf (1987). Achievement in public and private secondary education in the Netherlands. Paper presented at the Annual Meeting of the American Sociological Association, Chicago IL, August 17-21. Williams, T. & P. Carpenter (1991). Private schooling and public achievement in Australia. International Journal of Educational Research, 15, 411-432. Willms, J.D. (1985). Catholic-school effects on academic achievement: New evidence from the High School and Beyond Follow-Up study, Sociology of Education, 58, 98-114. World Bank (1990). World Development Report. Washington, DC: Author.  Policy Research Working Paper Series Contact 2D AUthob 1L for D2a.r WPS800 The Legal Framework for Private Cheryl W. Gray November 1991 CECSE Sector Development in a Transitional Rebecca J. Hanson 37188 Economy: The Case of Poland Michael A. Heller Peter lanachokov Youssef Djehane WPS801 Unraveling the Mysteries of China's Arvind Panagariya November 1991 D. Ballantyne Foieign Trade Regime: A View from 37947 Jiangsu Province WPS802 Strengthening the Bank's Population Steven W. Sinding November 1991 0. Nadora Work in the Nineties 31091 WPS803 Financial Regulation: Changing the Millard Long November 1991 W. Pitayatonakarn Rules of the Game Dimitri Vittas 37666 WPS804 Global Trends in Raw Materials Boum-Jong Choe November 1991 S. Lipscomb Consumption 33718 WPS805 Privatization in the Soviet Union: Sergei Shatalov November 1991 CECSE The Beginnings of a Transition 37188 WPS806 Measuring Commercial Bank Dimitri Vittas November 1991 W. Pitayatonakarn Efficiency: Use and Misuse of Bank 37666 Operating Ratios WPS807 Moderate Inflation Rudiger Dornbusch November 1991 S. Moussa Stanley Fischer 33490 WPS808 The New Trade Protection: Price Ann Harrison November 1991 D. Ballantyne Effects of Antidumping and 37947 Countervailing Measures in the United States WPS809 Openness and Growth: A Time Ann Harrison November 1991 WDR Office Series, Cross-Country Analysis for 31393 Developing Countries WPS810 Poverty and Income Distribution Francois Bourguignon November 1991 D. Ballantyne during Adjustment: Issues and Jaime de Melo 37947 Evidence from the OECD Project Christian Morrisson WPS811 Comparative Resource Allocations Peter T. Knight December 1991 D. Atzal to Human Resource Development Sulaiman S. Wasty 36335 WPS812 Alternative Forms of External Stijn Claessens December 1991 S. King-Watson Finance: A Survey 31047 WPS813 Price Stabilization for Raw Jute Takamasa Akiyama December 1991 D. Gustafson in Bangladesh Panos Varangis 33714  Policy Research Working Paper Series Contact IitL Athor forpAuthor t rpe WPS814 Finance, Growth, and Public Policy Mark Gertler December 1991 W. Pitayatonakarn Andrew Rose 37666 WPS815 Governance and Economy: A Review Deborah Brautigam December 1991 Z. Kranzer 37494 WPS816 Economic Consequences of German Gerhard Pohl December 1991 CECSE Reunification: 12 Months After the Big 37188 Bang WPS817 How Does Brady-Type Commercial Mohua Mukherjee December 1991 Y. Arellano Debt Restructuring Work? 31379 WPS818 Do Rules Control Power9 GATT J. Michael Finger January 1992 N. Artis Articles and Arrangements in the Sumana Dhar 37947 Uruguay Round WPS819 Financial Indicators and Growth in a Robert G. King January 1992 W. Pitayatonakarn Cross Section of Countries Ross Levine 37666 WPS820 Taxation in Decentralizing Socialist Christopher Heady December 1991 Economies: The Case of China Pradeep K. Mitra WPS821 Wages and Unemployment in Poland: Fabrizio Coricelli January 1992 V. Berthelmes Recent Developments and Policy Ana Revenga 39175 Issues WPS822 Paternalism and the Alleviation of Nancy Jesurun-Clements January 1992 F. Betancourt Poverty 18-126 WPS823 How Private Enterprise Organized Steven M. Jaffee January 1992 C. Spooner Agricultural Markets in Kenya 30464 WPS824 Back-of-the-Envelope Estimates Sergio Margulis January 1992 J. Arevalo of Environmental Damage Costs in 30745 Mexico WPS825 The Empty Opportunity: Local Control Marlaine E. Lockheed January 1992 D. Eugene of Secondary Schools and Student Oinghua Zhao 33678 Achievement in the Philippines WPS826 Do Workers in the Informal Sector Ariel Fiszbein January 1992 N. Perez Benefit from Cuts in the Minimum 31947 Wage? WPS827 Free Trade Agreements with the Refik Erzan January 1992 J. Jacobson United States: What's in It for Latin Alexander Yeats 33710 America?