Student learning in Uganda : textbook availability and other factors

cREP 63 World Bank Reprint Series: Number 163 June 1980 Stephen P. Heyneman and Dean T. Jamison S Student Learing in Uganda: Textbook Availability and Other Factors r, t0 Reprinted with permission from the Comparative Education Review, vol. 24, no. 2 (June 1980), pp. 206-20.   Student Learning in Uganda: Textbook Availability and Other Factors STEPHEN P. HEYNEMAN AND DEAN T.JAMISON Student Learning: The General Problem Only recently has it been recognized that the average amount learned at the same age level in science, mathematics, and other universal subjects is substantially less in countries of low per capita incomes,' and that there has been a gradual but definitive shift in priorities in many low-income countries away from the need to place higher proportions of children in school and toward an investment in how much they learn after entry has been obtained. In Africa 14 out of 37 countries nlow enroll more than 70 percent of their children at the first level of schooling; in Asia and the Middle East it is 14 out of 22; in Central America it is 6 out of 10; in South America it is 9 out of 10. Nevertheless, even though significant advances have been made in the availability of classrooms, gross differences remain between poor coun- tries and rich countries, and within poor countries, in what pupils can expect to find in classrooms when they enter. The value of pedagogical equipment, furniture, and supplies at the fourth-grade level in Bolivia is 1 percent of what it is in Maryland.2 The OECD countries in 1975 in- vested 33 times more per primary school student than did countries with per capita incomes of less than US$265/year. Moreover, this gap has been widening. As a result of the demand that primary schools transfer more information more efficiently, per pupil investments are rising faster in rich countries than in poor countries.3 This shift in investment need is reflected to some degree in the lend- ing program of international organizations such as the World Bank. As These data were collected in 1972 under the sponsorship of the Ugandan government and the National Institute of Education, Makerere University. This analysis has been undertaken as part of the World Bank's research project on "Textbook Availability and Educational Quality," RP0671-60. However, the views and interpretations are those solely of the authors and, in particular, do not necessarily reflect those of the World Bank. ' Alex Inkeles, "The International Evaluation of Educational Achievement," Proceedings of the National Academy of Education 4 (1977): 139-200, particularly pp. 157-73. 2 Stephen P. Heyneman, "Primary Education in Bolivia: What's Wrong?" mimeographed (Washington, D.C.: Education Department, World Bank, 1979). 3 World Bank, Education Sector Policy Paper (Washington, D.C.. International Bank for Recon- struction and Development, 1979). If one were to attempt, however, to adjust for international variations in purchasing power that are not captured in computation of exchange rates, the ratio of spending in industrialized to nonindustrialized societies would decline. For example, at the official exchange rate, the 1975 per capita income of the United States was approximately 53 times that of India; after adjusting for differences in purchasing power, however, this ratio declines to 14 (see World Development Report, 1979 [Washington, D.C.: World Bank, 1979], p. 177). (C) 1980 by the Comparative and International Education Society. 0010-4086/80/2421 -0004$01 .30 206 June 1980  S-UDENT LEARNING far as we can predict, the bulk of the Bank Group's capital investment in education will continue to assist the expansion of specific educational institutions-96 percent of the resources dispersed between fiscal years 1970 and 1974, and 93 percent between fiscal years 1975 and 1978. But this proportion is expected to decrease. By fiscal year 1983 it is expected to decline to less than 85 percent. The remainder is accounted for by increases in lending not to specific institutions: for curriculum develop- ment, radio, TV, administration, and particularly for increases in the development, production, and distribution of learning materials. By fiscal year 1983 the Bank Group's Education Sector is expected to be investing US$51 million/year in classroom "software" alone-up from US$1.6 million/year a decade earlier. For example, 5 percent of the education projects contained a textbook component in fiscal year 1975; 10 percent in 1976; 25 percent in 1977. Though the rationale for this shift in emphasis is strong, it could be undertaken more surely were there sufficient evidence on which quality- improving investments were the most effective; but that is hardly the case. There were 16 times more studies published on the cognitive impact of teacher training in the United States in 1 year than over the previous 2 decades in low-income countries.4 Similar imbalances exist with respect to the evidence on textbooks, duplicating machines, radio, television, pupil health, maps, sound equipment, and educational magazines-in fact, with respect to all school resources. Since choices have to be made in an environment of scarcity, our evidence on factors contributing to the ef- ficiency in learning will have to be more abundant and of higher quality. This paper reports on the findings of part of a broader World Bank r esearch program that is designed to provide that evidence. Student Learning in Uganda Previous efforts to estimate the achievement impact of school re- sources in Uganda have utilized either zero-order correlations5 or regres- sions of school effects blocked in the aggregate..6 From this experience it is The ERIC system lists 388 titles published on this subject in the United States in 1977. A recent review of the evidence from low-income countries could locate 23 published between 1963 and 1977 (see Torsten Husen, Lawrence J. Saha, and Richard Noonon, "Teacher Training and Student Achievement in Less Developed Countries," World Bank Staff Working Paper no. 310, December 1978). A subsequent review, which made a specific effort to locate studies published in non-European languages, could locate a slightly higher number (see Beatrice Avalos and Wadi Haddad, "A Review of Teacher Effectiveness Research in Africa, India, Latin America, Middle East, Malaysia, Philip- pines and Thailand: Synthesis of Results," mimeographed [Ottawa: International Development Re- search Center,June 19791). Stephen 'P. Heyneman, "Differences in Construction, Facilities, Equipment and Academic Achievement among Ugandan Primary Schools," International Review of Education 23, no. 1 (1977): 35-45. 6Stephen P. Heyneman, "Influences on Academic Achievement: A Comparison of Results from Uganda and More Industrial Societies," Sociology of Education 49 (July 1976): 200-211. Comparative Education Review 207  H EYN EMhAN AND JAMISON evident that school facilities strongly influence achievement, and that school facilities are statistically more able to predict achievement in Uganda than they are in industrial societies. Still no one has yet at- tempted to answer the question of which school resources account for the impact of schools on learning. This paper is in response to that question. We will first discuss how the basic variables were measured and the methodological procedures we intend to utilize. Then we report on the impact of school resources on student learning, taking the school as the unit of analysis. Next we will take the pupil as the unit of analysis and attempt to estimate the impact both of pupil characteristics and of school characteristics on student learning. In assessing the impact of pupil char- acteristics we control for all school effects by using indicator variables, identifying which school a child is in, to capture aggregate school effects. Finally we summarize what we have learned. Data and Methods Sample These data were collected by Heyneman in 1972 from 61 Ugandan primary schools from five districts (North and South Karamoja, West Buganda, Bugisu, and Toro), and from all three urban areas (Kampala/ Entebbe, Mbale/Tororo, and Jinja). Within each locality schools possess- ing a seventh grade were identified and a minimum of 10 percent selected randomly. The final sample contained 10.7 percent of the schools, 13.1 percent of the grade 7 pupils, and 12.9 percent of teachers within the selected areas. The sample schools were situated in varied local settings-for example, isolated but economically developed areas, iso- lated but economically poor areas, plantation and peasant agricultural areas, urban areas (some with manufacturing and commerce), and areas of relative isolation from all modern stimuli. Political and economic con- siderations prevented a truly national sample, but there is reason to be- lieve that the major socioeconomic factors associated with Primary School Leaving Examination performance in Uganda are adequately rep- resented. The data are derived from four sources: separate question- naires for pupils and staff, an inventory of school physical facilities, and the pupil performance on the Primary Leaving Examination taken 8 months after the questionnaires were administered. Socioeconomic Status In our analysis, a pupil's socioeconomic status is indicated by a sum- mary measure of mother's education, father's education, father's occupa- tion, and consumer possessions in the home. The four component mea- sures have been explored elsewhere.' They are, of course, highly inter- correlated; none has an impact on achievement in a manner divergent 208 June 1980  S'l UDENT LEARNING from the others. In previous explorations their relationships with achievement have been found to be weak. However, this is the first oc- casion in which their impact has been calculated after having controlled for pupil intelligence. Teacher Language Ability Each teacher was asked to respond to six multiple-choice questions of English usage.8 Scores for all teachers in a school were then added to- gether and averaged by school. Like Hanushek,9 we have chosen to aver- age teacher scores because grade 7 students, if affected at all, are affected by more than the teacher in grade 7; they are affected by all their teachers. Other teachers have taught them in previous years, or have guided students outside the classroom environment. The degree of con- tact is particularly high in isolated schools.10 Textbook Availability Each book available in grade 1 and grade 7 classrooms was counted and the sum divided by the total number of pupils in those two grades. The resulting ratios of books/child in grade 1 and grade 7 were then com- bined to constitute total books/child. Total books/child is a measure of reading material of all kinds: textbooks, readers, pamphlets, workbooks, and library books. No distinction was made between old, inappropriate, or damaged books, or between books in a vernacular, in English, or whether in science, mathematics, social studies, or other subjects. Total books/child is a measure of any reading material available, at the lowest and highest grade levels, and is a reasonably accurate portrayal of read- ing material available for the school as a whole. Pupil Health This was calculated by having pupils respond to five questions: (i) whether they had suffered from malnutrition when they were young, (ii) whether they had ever seen blood with their stool, (iii) how often they suffered from chills and fever, (iv) how often they had been in a clinic or I Ibid. Stephen P. Heyneman, "A Brief Note on the Relationship between Socio-economic Status and Test Performance among Ugandan Primary School Children," Comparative Education Review 20 (February 1976): 42-47; "Why Impoverished Children Do Well in Ugandan Schools," Comparative Education 15 (June 1979): 175- 85; and "Differences between Developed and Developing Countries: A Comment on Simmons and Alexander's 'Determinants of School Achievement,'" Eco- nomic Development and Cultural Change 28 (January 1980): 403- 6. 8 Though vernaculars were permitted or sometimes encouraged in grades 1- 3, all grades were supposed to have English lessons, and grades 4- 6 were supposed to be taught exclusively in English. 9 Eric A. Hanushek, Education and Race: An Analysis of the Educational Production Process (Lexington, Mass.: Lexington, 1972). is Forty percent of the teachers had taught for more than 10 years; 30 percent had taught for more than 5 years in the same school (see Stephen P. Heyneman, "Relationships between Teachers' Characteristics and Differences in Academic Achievement among Ugandan Primary Schools," Edu- cation in Eastern Africa 6, no. 1 [19761:41-51). Comparative Education Review 209  HEYNEMANXN4D1JAMISON hospital overnight, and (v) how many diseases they had suffered from, that they knew about." Each problem was explained to them in English and in one or more of six other languages. Personal assistance was given to each pupil who was unsure. If after receiving assistance a pupil was still uncertain, he was asked to check "No." Thirty-seven percent reported that they had been confined to a hospital at least once for "more than a few days." Seventy-five percent claimed tu have had malaria; 12 percent had seen blood with their stool. Ten percent had had trachoma; 8 per- cent had had hookworm; 5 percent claimed to have been treated for malnutrition when they were toddlers. 12 Those within this 5 percent "malnourished when young" performed 25 percent of a standard devia- tion below the mean in academic achievement. Those reporting that they had hookworm also performed significantly worse. Raven's Progressive Matrices (RPM) Intelligence was measured by having each pupil respond to the 36 problems posed in the Raven's Progressive Matrices (RPM) test of percep- tual and spatial abilities, colored version. The test is entirely nonverbal. It contains 36 pictures with differing shapes, colors, and patterns. At the bottom of the page, six alternative pieces are given. The task is to choose which of the six alternative pieces best fits the missing space in the shape at the top of the page. Because it is easily administered and entirely nonverbal, the RPM is a commonly used instrument in nonindustrialized societies. 13 Percentage of Chiklren in School In communities where educational opportunities are limited, children who enter and who remain in school are distinct in one way or another by " On the list were malaria, hookworm, trachoma, tuberculosis, and other diseases that were commonly recognized and had specific vernacular words to describe them. 12 In a separate question children were asked what they ate before coming to school on the day of the visit to the school; 19 percent had had nothing to eat or to drink. 13 Edgar A. F. Bowden, "Perceptual Abilities of African and European Children Educated To- gether," Jounial of Social Psychology 72, no. 2 (1969): 149- 54; Ernest L. Klingelhofer, "Performance of Tanzanian Secondary School Pupils in the Raven's Standard Matrices Test," Journal of Social Psychology 72 (August 1967): 204- 15; Mallory Wober, "The Meaning and Stability of Raven's Prog- ressive Matrices Test among Africans," InternationalJournal of Psychology 4 (1969): 229- 35; Philip E. Vernon, "Administration of Group Intelligence Tests to East African Pupils," Bnttsh Journal of Educational Psychology 31 (1967): 282-91, and "Abilities and Educational Attainment in an East African Environment," Journal of Special Education I (Winter 1976): 335-45; Uma Sinha, "The Use of the Raven's Progressive Matrices in India," Indian Educational Research 3 (1968): 75-88; A. Om- bredane, "Principes pour une etude psychologique des noires due Congo Belge," Annee Psychologie 50 (1951): 521-47; L. Berlioz, "Etudes des 'progressive matrices' faites sur les Africains de Douala," Bulletin Centre de Recherces Psychotechnique 4 (1955): 33-44; S. Biesheuvel, "Psychological Tests and Their Application to Non-European Peoples," in Cross-cultural Studies, ed. Douglas R. Price-Williams (Baltimore: Penguin, 1970); Michael A. Durojaiye, "Is the Concept of African Intelligence Mean- ingful," East Afnca 7 (1971): 4-13; and Sydney H. Irvine, "Figural Tests of Reasoning in Africa: Studies in the Use of the Raven's Progressive Matrices across Cultures," InternationalJournal of Psy- chology 4 (1969): 217- 28. 210 June 1980  STUDENT LEARNING comparison to the general age cohort in the community.'4 Put another way, in communities with less educational opportunity, seventh-grade students are less representative of the general age cohort than are those pupils who are in grade 7 in areas where school attendance is universal. The percentage of the 5-14-year-old age cohort in primary school has been calculated for each sample school.'5 These ranged between 6 and 76 percent. There is a marked tendency for average academic performance to be lower in areas with higher percentages of the age cohort attending school (r = -.251, P < .05). The question posed on this occasion is whether the influence of this preselectivity is as pronounced as the influence of characteristics internal to classrooms. School Facilities In addition to the availability of reading materials, the school inven- tory tallied the presence of a duplicating machine, farm, staff room, electricity, boarding facilities, football or hockey field, and whether or not window frames (present in all schools) were filled with glass. Common to each of these elements is the fact that, for better or worse, they were determined by authorities particular to the school, that is to say, by the teachers, the headmaster, and the parent's committee.16 Elsewhere it has been argued that the presence of these physical facilities was an indica- tion of initiative, specific to the school.'7 The fact that the presence of one item was, without exception, intercorrelated with the presence of each of the others, and with achievement, gives us further reason to consider these items together as an indicator of general facilities rather than sepa- rate variables. Each is therefore included in a summary scale, which has a range of 0- 7. Pupil School Affiliation Among our interests is whether pupils, independent of ability and social background, are advantaged by being able to attend specific schools. One way to estimate the degree of this possible advantage is to estimate the statistical impact on achievement of students being affiliated with a particular school. Thus whether or not a student is in school number 12, or number 13, etc., is added to the regression equation as an 4 Stephen P. Heyneman, "Relationships between the Primary School Community and Academic Achievement in Uganda,"JournalofDevelopingAreas 11 (January 1977): 245-59. IS In West Buganda, Toro, and Karamoja the data reflect the proportion of the age group of the local county in school; in Bugisu, Kampala, Jinja, and Mbale the figures reflect district-level propor- tions. ' This stands in contrast to textbooks and high-quality teachers, which were distributed through central authorities (see Stephen P. Heyneman, "Changes in Efficiency and Equity Accruing from Government Involvement in Ugandan Primary Education," African Studies Review [April 1975], pp. 51-60) 7 Stephen P. Heyneman, "Differences in Construction, Facilities, Equipment and Academic Achievement among Ugandan Primary Schools." Comparative Education Review 211  HEVNEMAN,D JAMISON indicator variable for each school. (If a child attends school 12 the indi- cator variable for school 12 takes on the value 1 for that child; it takes on the value 0 otherwise.) Unit of Analysis: Individual or Aggregate? Divergent opinion exists on how to analyze academic achievement data-whether to use the pupil, the classroom, or the school as the unit of analysis. There are advantages and drawbacks to each. If pupils are cho- sen, then it is normal to assign levels of school or classroom quality to each individual pupil. This raises the statistical significance of school resouLces because of the increment in the units of observation. By contrast, if the classroom or the school is chosen as the unit of analysis, it is normal to average the characteristics of individual pupil achievement, intelligence, health, socioeconomic status, and attitudes-variables for which there is more variance within classrooms or schools than among them. If aver- aged, pupil characteristics lose significance because of the artificial at- tenuation of variance. Moreover, aggregating individual pupil data may overlook important differences in the way school resources are utilized within schools on the basis of sex, ethnic, or SES groups."8 We have chosen to discuss the detei-minents of achievement at both the pupil and the school levels of aggregation. Achievement aggre- gated to the school level is indeed appropriate for Uganda, perhlaps even more so than for industrial societies. Pedagogical treatment of indi- vidual pupils does not differ dramatically within Ugandan classrooms, or between classrooms in the same school. There are no curriculum tracks, specialist personnel, or remedial reading teachers; there is no special equipment which can be assigned to specific individuals. These are op- tions (and problems) of wealthier school systems. In Uganda the cur- riculum is universal for all children, and the assignment of teachers, equipment, and materials is based on national standards and effected by national authorities. Differences do exist, but they are differences between schools, not within them. Thus for our purposes we have chosen to analyze the achievement of primary school classrooms as well as the achievement of individual primary school pupils. Student Achievement Learning, in this instance, is defined as the achievement performance on the academic selection examination which every seventh-grade pupil must take. The test is rigorous. Because there are secondary school 18 Larry J. Griffin and Karl L. Alexander, "Schooling and Socio-economic Attainments: High School and College Influences," AmenrcanJournalof Sociology 84 (September 1978): 319-48; and Karl Alexander, Martha Cook, and Edward McDill, "Curriculum Tracking and Educational Stratification: Some Further Evidence," Amencan Sociological Review 43 (February 1978): 47- 67 212 June 1980  STUDENr LEARNING places for only one child in 10, tension -and motivation run high. The test is administered simultaneously over a 3-day period in 2,500 locations. Grading is by computer. Children are identified by numbers; security is tight, and accusations of fraud are surprisingly few. Separate content is administered each day of the test: English language (reading comprehen- sion and grammatical usage) may come on the first day, mathematics on the second, the general paper (science, geography, and social studies) on the third, etc. We will attempt to predict performance on each section separately and also together as a total. Lea rmng Net of Ability to Learn More than anything else, current levels of achievement can be ex- pected to follow from past levels of achievement, or from the level of natural capacity. It is not always specified whether survey research is accounting for the variance in the amount learned, in the amount of change in learning, or in the capacity to learn. Four techniques exist for disentangling these various elements, as follows: (a) subtracting a pretest achievement score from a posttest achievement score; (b) entering a pre- test achievement score as an independent variable in a regression predict- ing the posttest score; (c) using a measure of natural capacity in place of a pretest and subtracting it from the posttest achievement score; or (d) entering the measure of natural capacity as an independent variable. We will not discuss type a or type b because we could acquire academic achievement data at one point in time only. However, we did acquire each pupil's performance on the Raven's Progressive Matrices (RPM) test, and we use their score on the RPM to control for natural ability in a fashion identical to controls which are placed on other independent variables in a regression equation (type d). In rejecting type c we are mindful of the work done by Cronbach and Furby.'9 They have argued that differences in initial and final scores regress toward the mean; that is, there will be a negative correlation between initial score and achievement change.20 By contrast, entering RPM scores as an independent variable is relatively "problem free."21 1I LeeJ. Cronbach and Lita Furby, "How Should We Measure 'Change'-or Should We?" Psycho- logical Bulletin 74, no. I (1970): 68-80. 20 Cited by Peter R. Moock and David Rhodes, "Achievement in the Philadelphia Public School Systerm," IRCD Bulletin (Teachers College), vol. 13 (Summer 1978). 2' Ibid. This has been the preferred procedure of our previous work and of others. See Hanushek (n. 9 above); Dean T. Jamison, "Radio Education and Student Repetition in Nicaragua," in The Nicaragua Radio Mathematics Project, 1976-1977, ed. Patrick Suppes, Barbara Searle. and Jamesine Friend (Stanford, Calif.: Institute for Mathematical Studies in the Social Sciences, 1979); Richard J. Murnane, The Impact of School Resources on the Learning of Inner City Children (Cambridge, Mass.. Ballinger, 1975); and Donald R. Winkler, "Educational Attainment and School Peer Group Composition,"JournalofHuman Resources 10 (1975): 189-204. Comparative Education Review 213  HEYNEMAN ANDJAMISON Determinents of Student Learning in Uganda Analysis Using School-Average Data We have taken the school averages of seven potential influences on academic achievement: pupil SES, pupil RPM, pupil health, teacher lan- guage ability, school physical facilities, textbook availability, and the local enrollment ratio of each school in the sample. These wve have regressed against average school performance on math, English, general paper, and total achievement tests, and also on RPM.22 Means, standard devia- tions, and regression results can be found in table 1. In table 1 we have aggregated all potential influences to the level of the school since our data on school variables (e.g., textbook availability) already are school-level aggregates. We do not view this level of data aggregation necessarily as a drawback. Because of financial constraints, policy changes have to be made at the level of the school; that is, teacher English quality and other school facilities, if altered up or down, would have to be altered for all children in a given school together. We are therefore interested in what effect these alterations might have. Several observations seem appropriate for table 1. The first is that all three school variables (teachers' English, textbooks, and physical facil- ities) have positive and consistent (though rarely statistically significant) effects on all cognitive tests. Raising the average English-language ability of school teachers by I point, for example, would raise average school English-language scores-by 1.3 points, general paper scores by 1.5 points, and total achievement points by 4.1 points. Similar gains to be expected from improving the availability of textbooks and school physical facili- ties can be estimated from the appropriate regression coefficients in table 1. Among the three school characteristics it is difficult to isolate one or another as being predominant. The most noticeable categories of influence might be textbook availability on the English-language test and total achievement scores, and school physical facilities on RPM perform- ances.23 However, if one averages the size of the significance measures (t-values) across cognitive tests, as we have done in the right-hand col- 22 We have entered RPM both as dependent and independent variables because we are aware that no test of natural ability is uninfluenced by environment, particularly school environment. Results here, particularly with regard to school physical facilities, bear us out. 23 In addition to examining the impact of textbook availability on student achievement in gen- eral, we also used the school-level data to examine whether book availability was differentially effec- tive for high- or low-SES students and high- or low-ability students. We approached this question both by adding textbook by SES or RPM interactions into the regressions and by stratifying the sample into various SES and RPM groups. Our sample sizes were small and regression coefficients often unstable and of low,significance. Our results do suggest, somewhat unfortunately, that the effect of textbook availability is more pronounced for higher-SES and higher-ability schools. How- ever, this is not the case for teacher English, which has the most pronounced impact on low-SES schools, whether or not they averaged high on the RPM measure of intelligence. 214 June 1980  0 TABLE I l NPFIUENCES oN LEARNIN;: THE SCHOOi,-LFEI. UNS-I ANDARDI ZED RECRESSION COEFFICIENTS(N = 61) 0 General Total Average Variable Mean SD RPM English Paper Mathematics Achievement I-Value Constan1t - 19.047 29.326 39.779 27.800 96.905 ... SES 9.664 2.791 .084 -.235 -.807** -.093 -1.135 *- (.484) (.906) (2.042) (.265) (1.253) (.99) RPM 23 113 3.237 -- .539"* .546 .473 1.558** (2.648) (1.760) (I 709) (2.190) (2.11) Local enrollment ratio 41.951 17.468 .(11 -.033 -.044 -.065 - 142 (.446) (.914) (.789) (1.312) (1.116) (.92) Z Ptipilhealth 3.005 .500 -.121 1.681 .305 1.186 3.172 (.152) (1.416) (.167) (736) (.765) (.65) Teacher'sEnglisIqUalitY 4.087 .895 .149 1.327* 1.479 1.273 4.080* (.329) (1.958) (1.434) (1.383) (1.724) (1.4) z TexIbook availability 6 530 6.318 .081 .194* .072 .020 .286 (1.075) (1.707) (417) (.126) (.720) (.81) Schoolphysicallacilities 3.219 1.669 .681" * .536 .887 .352 1.775 (2.556) (1.268) (1.379) (613) (1.203) (1.4) R2 (adjusted fot df) 146 .327 .137 .070 .191 ... NOTE. -t-values are in parentheses. English mean = 51.467 (SD = 5.458); general paper mean= 52.843 (SD = 7.340); math mean = 95.056 (SD = 6.306); total achievement mean = 149.366 (SD = 17.393). *P<.1. *P <.05.  HEYNEMAN ANDJAMISON umn, school physical facilities and teacher English appear to be the more significant: 1.4 as opposed to 0.81. It is surprising here that the influence of SES on all (academic) achievement tests is negative. From previous analyses we would have expected it to be low and/or insignificant; but in these analyses the influence of SES was not explored after having controlled for ability. That it is negative is of interest, though any interpretation must be made with caution because, when analysis is undertaken at the pupil-specific level, the apparent impact of SES can alter. Analysis Using Pupil-specific Data To calculate the learning influences on individual pupils we ran the same regression,24 displayed in table 1, on individual pupils; this time we assigned characteristics of school and teacher quality identically to each pupil in the same school. The results are displayed in table 2. Alterations do occur in the size of the regression coefficients, but they are not major. The influence of textbook availability on the RPM is .081 at the classroom level and .047 at the pupil level; on English-language achievement the difference is .194 versus .223; on general paper .072 versus -.034; on mathematics achievement .020 versus .104; and on total achievement .286 versus .311.25 The most important difference between table 1 and table 2 can be found in the degree of statistical significance, measured in t-values. For each variable, the degree of statistical significance increases markedly in table 2. The average t-value (for the five dependent variables) is in- creased from 0.81 to 2.4 in the case of textbook availability; from 1.4 to 3.3 for school physical facilities; from 1.4 to 4.5 for teacher's English. There is one exception to the rule that in countries with severe finan- cial constraints, like Uganda, educational interventions cannot be ad- ministered to pupils individually. The exception is with health and nutri- tion. As a variable, pupil health appears to have about the same average degree of statistical significance as textbook availability, though with less variation in significance by subject. However, unlike school facilities (or the quality of a teacher's English, or textbook availability), vitamin sup- plements, parasitic treatments, and antibiotics can be administered by teachers to those pupils whose ailments are more obvious.26 What this implies is that an intervention in health and nutrition, because it could be 24 Sex has been added as an individual characteristic. 15 From the evidence to date, textbook availability is the most consistently positive predictor of school achievement in less industrialized societies, substantially more consistent, e.g., than length of teacher training (see Stephen P. Heyneman, Joseph P. Farrell, and Manual A. Sepulveda-Stuardo, "Textbooks and Achievement: What We Know," World Bank Staff Working Paper no. 298, October, 1978). 2a School lunch programs, an equally important device, would have to be administered on a school-by-school rather than a pupil-by-pupil basis. 216 June 1980  n TABLE 2 INFLUENCESON LEARNING: TiE INDIVII)UALPUPIL-LEVEL UNSTANDARDIZED REGRESSION COEFFICIENTS (N= 1,907) General Total Average 0 Variable Meani SD RPM English Paper Mathemaucs Achievement t-Value Constant ... ... 18.473 33.922 37.556 25.089 97 061 e- Sex 359 .480 -2.989** -2 208- 7 669** -5.483** - 15420' (7.471) (4.409) (12.761) (7.528) (9.787) (8.4) SES 9 916 4.340 .277** 098 -.123 .031 -.020 (4.439) (1.544) (1.615) (.335) (.100) (1.6) RPM 24.072 8.468 459 .488"* .710*" 1 649"- (16.203) (14.370) (17.248) (18.519) (13.3) L.ocalenrollmentratio 41.951 17.468 .002 -.065** -.075** -.094** -.240" (.197) (4 757) (4.621) (4.739) (5.622) (4.0) Pupilhealth 3.022 1.760 165 .300* .450** .437' 1.207' (1.551) (2284) (2.853) (2.284) (2.918) (2.4) Teacher'senglish 388 1 471"* 1.928** 1.620*" 5.101" C** (1.726) (5.307) (5.794) (4.016) (5.848) (4.5) Textbook availability ... .047 233- -.034 .104- .311* (1.375) (5.290) (.677) (2.426) (2.342) (2.4) School physical facillites ... .606- .537"* .749"* .283 1.501' (4.755) (3 398) (3 950) (1.230) (3.018) (3.3) R2 (adjusted fot df) ... ... .068 .229 .231 .209 .267 NO1E.-t-values are in parentheses. English mean = 52.473 (SD = 11.480): general paper mean = 53.495 (SD = 13.800); mathematics mean = 46.014 (SD = 16.495); totalachlevelnentmeani = 152.024 (SI) = 37.05). * P <.05. "P <.01.  HEYNEMAN ANDJAMISON administered to specific children who were not already healthy, holds out the possibility of a more targeted gain in learning than, for example, increasing the availability of textbooks. The Influence of Pupil Affiliation with Particular Schools Counting the availability of educational equipment and the abilities of teachers is an imperfect way to estimate the sum total of school effects. However necessary for calculating which variable might make the best educational investment, inevitably something is not counted and ther-e- fore its effect may be overlooked. We have attempted to get around this problem by noting simply the affiliation of a student with a particular school and its achievement im- pact, separate from those characteristics about a student which are de- termined by influences outside the school: sex, health, SES, intelligence, and the like. Thus, for table 3, we have calculated 65 coefficients: four for student-specific characteristics and 61 corresponding to the indicator variables denoting school affiliation. To be frank, we have had some difficulty deciding how to display the list of 65 regression coefficients in a concise fashion. To have such a lonig list is hardly customary, and the importance of particular affiliations cannot be apparent to a reader unfamiliar with the name of each school. What we have done in table 3 is to average the absolute value of the coefficients of the indicator variables denoting school affiliation. The ef- fect is pronounced. On an average, being in a good or bad school can affect the student's total achievement score by 15 points. Approximately half of the school-affiliation coefficients are significant at the P < .05 level (on total achievement score) and ultimately can explain 39 percent of the explained variance.27 Perhaps most interesting is the ability of Ugandan schools to affect performance on an intelligence test. Though less than half of the vari- ance is explained at all, 73 percent of what is explained can be attributed to school affiliation.28 This lends further evidence to the theory that performance on culture-free tests such as the RPM is hardly free of environmental effects, even though we are able to explain much less variance in RPM than in the achievement-test scores. 27 For purposes of comparability with the IEA (science) results, previous estimates attributable to school effects did not control for intelligence. 28 In the United States, James Coleman and others find that school characteristics explain more variance in intellectual aptitude than academic achievement. They explain it as follows: "Achieve- ment scores cover material that is nearly the same in all school curriculums toward which all schools teach alike, while the ability tests cover material that the school teaches more incidentally and thus with more differential success. Consequently, student bodies that differ at the beginning of school become slightly more alike wiih respect to skills most directly r elated to a standard curriculum, but do not with the skills in which the curriculum is less standard" (see James S. Coleman, Ernest Q. Campbell, Carol J. Hobsen, James McPartland, Alexander M. Mood, Frederic Weinfeld, and Robert L. York, The Equality of Educational Opportunity [Washington, D.C.: Department of Health, Education, and Welfare, 1966]). 218 June 1980  P 10 TABLE 3 THIE INFIUENCE OF SCHOOIL AFFILIATION ON LEARNINC NET OF INDIVIDUAL PUIII. CHARACTERISTICS (N = 2,009) RPM Total Variable (N = 2,019) English General Mathemnalics Achievement m Constant 21.41 41.40 44.00 28.45 114.17 StUdenlt caractelristics: I. Sex -2.35'* -1.92** -7.20** -5.06'' -14.30* (5 94) (4.01) (13.22) (7.10) (4.62) 2. Raven's Progressive Natrices (RPM) - .46** .52** .73'- 1.70'* C (16.77) (16 90) (18.03) (20.14) 3. Health .13 .23 .40- .37* 1.04** (1.29) (1.84) (2.86) (2.01) (2.73) 4. SocioeconomI1icstatus(SES) .18*- .27' .21*' .02 .48'* > (3 04) (3.91) (2.59) (.19) (2 20) z 5. School affiliation: O Meanabsolute valueofcoefficient 2.94 4.92 7.05 4.95 15.17 %ofcoefficientssignifical1tatP> 05 22.00 41.00 56.00 19.00 48.00 %ofcoefficientSsignificantatP>.0l 1100 27.00 41.00 10.00 27.00 R2(adjusled for df) .15 .34 .41 .28 .38 %ofR2attributabletoschoOlaffiliationa 73 50 53 32 39 This is defined to equal 100 limes the ratio of the R2 of the regression containing both the student and school-affiliation variables, less the R20ofthe regressioncoIntainingoinl) StUdentvariablestothe R2 of the regression coItainingboth. * P <.05. t P <.01.  HEYENEMAN ANDJAMISON How much additional impact emerges from this affiliation method of estimating school effects? To answer this we turned school affiliation into a dependent variable and predicted it on the basis of textbook availability, teacher's English quality, and school physical facilities. These characteristics can predict a maximum of only 40 percent of the school- affiliation influence. Thus the impact of schools on pupil achievement is at least 60 percent over and above the combined explanatory power of our three strongest school variables. This difference between the school effects which we can specify through ouI measurements of class- room quality and the school effects which we can quantify but cannot disaggregate is pronounced; and it is important enough to warrant fur- ther research. Summary We have learned that differences among schools are extremely pow- erful determinants of school achievement in Uganda. Yet there is much about school impact that we have left uncounted, and therefore much we have yet to learn about how to measure school quality. However much we have left to learn though, the search for methodological improvement should not delay intelligent investments in environments where the needs are genuine. And in Uganda, improving the quality of a teacher's lan- guage ability, the availability of textbooks, school physical facilities, and the pupil level of health and nutrition would be intelligent investments. 220 June 1980    THE WORLD BANK Headquarters: 1818 H Street, N.W. H Washington, D.C. 20433, U.S.A. European Office: 66, avenue d'Iena 75116 Paris, France Tokyo Office: Kokusai Building, 1-1 Marunouchi 3-chome Chiyoda-ku, Tokyo 100, Japan The full range of World Bank publications, both free and for sale, is described in the World Bank Catalog of Publications, and of the continuing research program of the World Bank, in World Bank Research Program: Ab- stracts of Current Studies. The most recent edition of each is available with- out charge from: PUBLICATIONS UNIT THE WORLD BANK 1818 H STREET, N.W. WASHINGTON, D.C. 20433 U.S.A.