Urban-Rural Literacy Gaps in Sub-Saharan Africa: The Roles of Socioeconomic Status and School Quality YANHONG ZHANG

During the 1990s, sub-Saharan African countries experienced notable progress in universalizing primary education. Despite this progress, the region as a whole is far from attaining Education for All (EFA) and the Millennium Development Goals (MDG).1 In 2001, over 45 million children of primary school age in the region, almost 42 percent, were not enrolled in school.2 Only a handful of small countries in the region had reached a net enrollment rate (NER) greater than 90 percent.3 The NER of some countries was under 70 percent, indicating that these countries needed to expand their primary school system capacity in order to enroll all children. In addition, delayed enrollment, dropping out, and grade repetition are widespread. In 2001, 20–40 percent of children in first grade were at least 2 years above the official ages for that grade. In half the countries, less than 67 percent of children who started the first grade were able to continue beyond the fifth grade. In about a dozen countries, more than one-quarter of all primary school students repeated a grade. As a consequence of low participation, a child in sub-Saharan Africa could expect to receive, on average, 7 years of education in 2001—6–9 years less than in Western Europe and the Americas.4 The challenge to expand children’s participation in primary education in sub-Saharan Africa is exacerbated by low levels of academic achievement. In seven southern African countries participating in a regional study of achievement from 1995 to 1998, between 1 percent and 37 percent of grade 6 students reached the “desirable” level in reading, while 22–65 percent were 1

For details of the EFA and MDG, see UNESCO, Education for All: The Darkar Framework for Action (Paris: UNESCO, 2000); and United Nations, UN Millennium Declaration (New York: United Nations, 2000). For details on the monitoring of the achieving of the EFA goals, see UNESCO, EFA Monitoring Report 2002/2003: Education for All; Is the World On Track? (Paris: UNESCO, 2002), EFA Monitoring Report 2003/2004: Gender and Education for All; The Leap to Equality (Paris: UNESCO, 2003), and EFA Monitoring Report 2004/2005: Education for All; The Quality Imperative (Paris: UNESCO, 2004). 2 For improvement in achieving universal primary education, see UNESCO, EFA Monitoring Report 2004/2005. The figures on out-of-school children are based on UNESCO Institute for Statistics, Children Out of School: Measuring Exclusion from Primary Education (Montreal: UNESCO Institute for Statistics, 2005). 3 The NER is a ratio of the number of students enrolled in primary schools to the total number of children of primary school age. 4 UNESCO, EFA Monitoring Report 2004/2005. Comparative Education Review, vol. 50, no. 4. 䉷 2006 by the Comparative and International Education Society. All rights reserved. 0010-4086/2006/5004-0002$05.00 Comparative Education Review

581

ZHANG

at “minimum” level. In six French-speaking African countries, 14–43 percent of grade 5 students had “low” achievement in either French or mathematics.5 Rural education in many less developed countries is often synonymous with disadvantages for learning. Indeed, the available evidence suggests that, in the latter half of the 1990s, primary school students in rural areas consistently underperformed their urban counterparts by substantial margins in sub-Saharan Africa (though there was no clear pattern of gender disparities).6 Since most children in sub-Saharan Africa still reside in rural areas, improving school participation and raising the learning levels of rural children must be at the forefront in policies aimed at achieving EFA goals in these countries. In this article, I document the learning disadvantage of rural primary school students in sub-Saharan Africa. I then attempt to identify the factors underlying such disadvantages. Analyzing data from 14 school systems participating in the second study of the Southern and Western Africa Consortium for Monitoring Educational Quality (SACMEQ II), which was carried out between 2000 and 2002, this article addresses the following questions: How do the reading literacy scores of rural primary school students differ from their urban counterparts in each of the school systems in the beginning of the twentyfirst century? To what extent is there a pattern of rural disadvantage in reading literacy scores across these countries? In other words, are the rural-urban gaps consistent across all the school systems, or are they greater in some countries than in others? Finally, can the urban-rural gap in reading achievement be explained away by differences in student and school characteristics? Rural Disadvantages in Achievement

In sub-Saharan Africa, systematic and cross-nationally comparable evidence on the rural-urban disparities in achievement come from two recent regional studies. The first study is the aforementioned SACMEQ , which carried out two cycles of data collection in a group of anglophone countries during 1995–98 and 2000–2002, respectively. Analyses of the first cycle of 5

Ibid. Dhurumbeer Kulpoo, The Quality of Education: Some Policy Suggestions Based on a Survey of Schools; Mauritius, SACMEQ Policy Research Report no. 1 (Paris: International Institute for Educational Planning, 1998); Friedhelm Voigts, The Quality of Education: Some Policy Suggestions Based on a Survey of Schools; Namibia, SACMEQ Policy Research no. 2 (Paris: International Institute for Educational Planning, 1998); Thomas Machingaidze, Patrick Pfukani, and Sibangani Shumba, The Quality of Education: Some Policy Suggestions Based on a Survey of Schools; Zimbabwe, SACMEQ Policy Research no. 3 (Paris: International Institute for Educational Planning, 1998); Sebtuu Nassor and Khadija Ali Mohammed, The Quality of Education: Some Policy Suggestions Based on a Survey of Schools; Zanzibar, SACMEQ Policy Research no. 4 (Paris: International Institute for Educational Planning, 1998); Manasseh Nkamba and Joe Kanyika, The Quality of Education: Some Policy Suggestions Based on a Survey of Schools; Zambia, SACMEQ Policy Research Report no. 5 (Paris: International Institute for Educational Planning, 1998); Juliana Nzomo, Mary Kariuki, and Lilian Guantai, The Quality of Education: Some Policy Suggestions Based on a Survey of Schools; Kenya, SACMEQ Policy Research Report no. 6 (Paris: International Institute for Educational Planning, 2001); M. Michaelowa, Quality and Equity of Learning Outcomes in Francophone Africa (Montreal: UNESCO Institute for Statistics, 2004). 6

582

November 2006

URBAN-RURAL GAPS IN AFRICA

SACMEQ data revealed substantial disparities in the reading and mathematics achievement scores of primary school students by school location and family socioeconomic status (SES). These findings emerge within all participating countries, although there was effectively no difference between girls and boys. Generally in all six countries, higher proportions of sixth graders from large cities and small towns met the “minimum” and “desirable” mastery levels of proficiency than did students from rural and isolated areas. Students from higher SES families were much more likely to meet the two thresholds consistently than their counterparts from lower SES families.7 Such patterns were also found in the 10 Francophone countries that participated in the other major African study of reading and mathematics achievement by second and fifth graders, Programme d’Analyse des Syste`mes E´ducatifs de la CONFEMEN (PASEC), conducted between 1992 and 2001.8 While there was no clear evidence of gender difference in reading and mathematics scores, students from rural areas on average had lower scores in the two subjects than did their counterparts from urban areas in all but one country as early as in the second grade. These disparities persisted until the fifth grade.9 Taking into consideration that academic underachievement is widespread in less developed countries, these findings suggest that rural students tend to be particularly vulnerable to educational disadvantage.10 Of course, rural disadvantages are not uniquely found in less developed countries. In an analysis of 24 industrialized countries participating in the Organization for Economic Cooperation and Development study of the literacy performance of 15-year-old students, James Williams found that rural students scored considerably lower in mathematics than did their counterparts in urban and medium-sized communities in 14 countries.11 Even though rural-urban disparities are an important form of educational inequality in less developed countries, the causes for such disparities in student learning have rarely been examined systematically using cross-national data. One starting point is to look at differences between rural and urban students in terms of their individual and school characteristics. These are the two primary sources of influence on student learning, which are the focus 7

See the six publications on SACMEQ previously cited in n. 6; Mioko Saito, “Gender vs. Socioeconomic Status and School Location Differences in Grade 6 Reading Literacy in Five African Countries,” Studies in Educational Evaluation 24, no. 3 (1998): 249–61. 8 The Program of the Analysis of Educational Systems of CONFEMEN, where CONFEMEN stands for Confe´rence des Ministres de l’E´ducation des Pays Ayant le franc¸ais en Partage (Conference of Education Ministers of Francophone Countries across the World). 9 Jean-Marc Bernard, Ele´ments d’appre´ciation de la qualite´ de l’enseignement primaire en Afrique francophone (2003), Le poˆle de Dakar (Dakar, Senegal), http://www.poledakar.org/article.php?id_articlep119 (last acccessed August 2, 2006); Katharina Michaelowa, Quality and Equity of Learning Outcomes in Francophone Africa (Montreal: UNESCO Institute for Statistics, 2004). 10 UNESCO, EFA Monitoring Report 2004/2005. 11 James H. Williams, “Cross-National Variations in Rural Mathematics Achievement: A Descriptive Overview,” Journal of Research in Rural Education 20, no. 5 (2005), http://www.umaine.edu/jrre/20-5.htm (last accessed May 22, 2006). Comparative Education Review

583

ZHANG

of numerous comparative studies of achievement. Using recently available data from SACMEQ II, Kenneth Ross and Linda Zuze have shown the relationship between family background and achievement in 14 school systems of southern and western Africa between 2000 and 2002. Their research confirms the findings from the previous analysis of the SACMEQ I: sixth graders from well-to-do backgrounds consistently outperformed their counterparts from disadvantaged family backgrounds. However, Ross and Zuze also pointed out that there was wide variation in the SES-related gaps in reading literacy scores across the 14 school systems. Such gaps tended to be larger in lowincome countries than in middle-income countries, though they were quite large in two middle-income countries in the study.12 Findings from the analysis of PASEC data suggest that, in addition to students’ family wealth, such indicators of inputs of school quality as the availability of textbooks play an important role in explaining the differences in reading and mathematics scores both within and between schools.13 Research into the effects of family versus school quality on learning has stoked controversy. Almost 3 decades ago, Philip Foster argued that in less developed countries educational inequality arises more from regional disparity than it does from individuals’ characteristics such as social class and ethnicity. Thus, observed Foster, research methods developed for studying social stratification in rich countries might illuminate the wrong problems in poor countries.14 This thesis was consistent with the empirical work of Stephen Heyneman and William Loxley, who, after analyzing a combination of cross-national and national data sets in the 1970s, concluded that schools tended to play a greater role in determining students’ learning achievement in poorer countries than in wealthy countries. The explanation was that schools in poorer countries vary more widely than those in wealthier countries in terms of their quality, in their use of trained teachers, and in materials. In poorer countries, therefore, which school a child attends makes a greater difference in how much he or she learns then it would in a rich country.15 12 Kenneth Ross and Linda Zuze, “Traditional and Alternative Views of School System Performance,” IIEP Newsletter 22, no. 4 (2004): 8–9. 13 Michaelowa, Quality and Equity of Learning Outcomes in Francophone Africa. 14 Philip Foster, “Educational and Social Differentiation in Less Developed Countries,” Comparative Education Review 21, nos. 2–3 (1977): 211–29. 15 Stephen Heyneman and William Loxley, “The Distribution of Primary School Quality within High- and Low-Income Countries,” Comparative Education Review 27, no. 1 (1983): 108–18, and “The Effect of Primary School Quality on Academic Achievement across Twenty-nine High- and Low-Income Countries,” American Journal of Sociology 88, no. 6 (1983): 1162–94. The same view was delineated in other works before and after these two references, i.e., Stephen Heyneman, “Influences on Academic Achievement: A Comparison of Results from Uganda and More Industrialized Countries,” Sociology of Education 49 ( July 1976): 200–211, “Differences between Developed and Developing Countries: Comments on Simmons and Alexander’s ‘Determinants of School Achievement,’” Economic Development and Cultural Change 28, no. 2 (1980): 403–6, and “The Search for School Effects in Developing Countries: 1966–1986,” Seminar Paper no. 33 (International Bank for Reconstruction and Development [IBRD], Washington, DC, 1986); Stephen Heyneman and William Loxley, “Influences on Academic Achievement across High- and Low-Income Countries: A Reanalysis of IEA Data,” Sociology of Education 55, no. 1 (1982): 13–21.

584

November 2006

URBAN-RURAL GAPS IN AFRICA

The same conclusions emerged from Heyneman’s study of Ugandan primary schools in the 1970s and Loxley’s study of Egyptian primary schools in 1980.16 The evidence from these studies, as well as the conclusions based on it, was subsequently challenged on methodological grounds. The gist of the criticism concerns the use of the ordinary least squares regression method, by which single-level regression models were superimposed on data that were hierarchical and had multiple levels by the very nature of education systems. The misapplication of the single-level models, as was the case with the Heyneman-Loxley studies and almost all of the other studies on school effects prior to the 1980s, generated erroneous findings. By fitting a multilevel regression model on data from secondary schools in Zimbabwe, Abby Rubin Riddell demonstrated that including an index of students’ home backgrounds reduced in-school- and class-level variances in student achievement after controlling for a measure of intake ability. This finding cast doubt on the conclusions of Heyneman and Loxley.17 David Baker, Brian Goesling, and Gerald LeTendre similarly employed multilevel regression methods to study variations in the relative effects of SES and school quality across 40 high- and middle-income nations (but no lowincome nations). By analyzing data on mathematics and science achievement data from 1995, they noted a consistent finding in all countries that school quality had less impact on inequality than did SES differences among families. The authors concluded that the Heyneman-Loxley effect had vanished and attributed the phenomenon to global mass schooling, which had taken place over the 3 decades since the 1970s. The authors speculated that the expansion of mass schooling allowed many educational systems to reach a resource threshold in quality within national systems of education. This may have reduced the variation in school quality and thus eliminated any difference that once may have existed between wealthier and poorer nations with regard to the influences of school quality.18 The common finding from recent investigations, showing the dependence of achievement on SES, does not necessarily contradict Foster’s emphasis on regional inequality. While the aforementioned African studies document the generality of the relation between inequalities in SES and inequalities in achievement, Foster’s contribution requires us to look beyond individual students and schools and to examine inequalities from a regional perspective. A prominent feature of schooling in much of the world is the rural-urban 16 See Heyneman, “Influences on Academic Achievement,” and “A Brief Note on the Relationship between Socioeconomic Status and Test Performance among Ugandan Primary School Children,” Comparative Education Review 20, no. 1 (1976): 42–47; William Loxley, “The Impact of Primary School Quality on Learning in Egypt,” International Journal of Educational Development 3, no. 1 (1983): 33–45. 17 Abby Rubin Riddell, “An Alternative Approach to the Study of School Effectiveness in Third World Countries,” Comparative Education Review 33, no. 4 (1989): 481–97. 18 David P. Baker, Brian Goesling, and Gerald K. Letendre, “Socioeconomic Status, School Quality, and National Economic Development: A Cross-National Analysis of the ‘Heyneman-Loxley Effect’ on Mathematics and Science Achievement,” Comparative Education Review 46, no. 3 (2002): 291–312.

Comparative Education Review

585

ZHANG

division in a host of factors important for student learning, such as individual, family, and school characteristics. Thus, because rural families have fewer resources than do urban families, especially in less developed countries, they tend to lag behind urban families in achievement even when they attend similar schools. Because rural children in less developed countries generally attend schools having fewer resources, however, rural students are subjected to a double jeopardy in their learning opportunities. Methods, Data, and Variables

Using data from 14 sub-Saharan school systems participating in SACMEQ II between 2000 and 2002, this article provides a further consideration of rural-urban gaps in achievement for primary school students. I build on previous studies by using a combination of variables, particularly those reflecting SES and school quality, to explain the generally observed rural disadvantage. Four sets of research questions guide my investigation: 1. In each nation’s school system that is examined, how did the sixth graders attending rural schools compare with their urban counterparts in terms of their reading literacy scores? Was there a pattern of the ruralurban gaps across countries? Were the rural disadvantages in learning outcomes larger in some countries than in others? 2. How different were schools in each country? How different were each nation’s rural schools from their urban schools? 3. Where there existed disparities in reading literacy scores between rural and urban students, to what extent were such disparities related to the differences in their individual SES and demographic characteristics? 4. Were rural-urban gaps in reading literacy attributable to differences in school resources and processes apart from students’ individual characteristics? Was there a pattern across the school systems, with respect to the role of school resources and processes, that could account for the observed rural-urban gaps in students’ literacy scores? The data used for this investigation came from SACMEQ II, which gauged the literacy and numeracy as well as the school conditions of sixth graders. Fourteen school systems in southern and western Africa participated in the study. These systems were located in Botswana, Kenya, Lesotho, Malawi, Mauritius, Mozambique, Namibia, Seychelles, South Africa, Swaziland, mainland Tanzania, Zanzibar of Tanzania, Uganda, and Zambia. The data collection was carried out in 2000 for all participating countries except for Mauritius in 2001 and Malawi in 2002. One unique feature of the SACMEQ study is that it was designed to address explicitly five policy concerns in participating counties. These concerns included inequalities in students’ characteristics and learning environments, teachers’ characteristics and viewpoints, school heads’ characteristics and viewpoints, equity in the allocation of human and material 586

November 2006

URBAN-RURAL GAPS IN AFRICA

resources, and the reading and mathematics achievement levels of both students and their teachers. The SACMEQ study adopted a two-stage sampling procedure by which schools were first selected and then 20 grade 6 students were randomly selected from each sampled school. In addition to a paper-andpencil test on reading literacy and numeracy for students and six teachers randomly selected from the same school, separate background questionnaires were also completed by the students, teacher, and school heads.19 The selection and organization of variables was based on an input-processoutput framework.20 This framework groups determinants of schooling outcomes into input and process variables at student, classroom, school, district, and community levels. The input variables are generally regarded as exogenous—meaning that input variables at each level are out of the control of the people at the corresponding level—while the process variables are endogenous. For instance, at the student level, such variables as gender, prior achievement, cognitive ability, and family SES are treated as input variables, and the quality of students’ school life, their sense of efficacy, and their attitudes toward schools are treated as process variables. Similarly, the class, school, district, and community levels include the size of the class, school, or district; intake composition; resources; and staffing characteristics. These factors are generally beyond the control at each level and so are treated as input variables. Although there is no consensus on what variables best measure school processes, most researchers would include variables representing the contexts and setting as well as the internal workings of schools, such as disciplinary climate, academic press (emphasis on academic achievement and expectations for student success), and teacher commitment and morale.21 A particular advantage of grouping variables using this framework is that it highlights the actions that teachers, school administrators, as well as students can take in influencing students’ learning outcomes. Outcome Variable

To address the research questions posed above, I focus on reading literacy scores. These scores reflect students’ performance on a paper-and-pencil test aiming to measure their ability to understand and use the forms of written language that are required by society and valued by individuals. Students were asked to complete a total of 83 multiple-choice questions of varying levels of difficulty after reading a variety of continuous written texts and forms. While easier items required matching words to pictures and involved concrete con19

Kenneth Ross, Mioko Saito, Stephanie Dolata, Miyako Ikeda, and Linda Zuze, Data Archive for the SACMEQ I and SACMEQ II Projects (Paris: IIEP-UNESCO, 2004). 20 For details of the input-process-output approach, see Stephen Raudenbush and J. Douglas Willms, “The Estimation of School Effects,” Journal of Educational and Behavioral Statistics 20, no. 4 (1995): 307–35. See also J. Douglas Willms, Monitoring School Performance: A Guide for Educators (London: Falmer, 1992). 21 Jaap Scheerens, Improving School Effectiveness, Fundamentals of Educational Planning Series (UNESCO: International Institute for Educational Planning, 2002). Comparative Education Review

587

ZHANG

cepts and everyday objects, successful completion of the more difficult items required students to use text structure and organization to identify an author’s assumptions and purposes. These questions linked a text to establish multiple meanings, including analogy and allegory. Students’ responses were scaled using Item Response Theory methods and then standardized so as to have a mean of 500 and a standard deviation of 100 for all students in the study. Individual Inputs

Three variables representing the sixth graders’ individual characteristics were examined in the context of their relationship with the students’ literacy scores. These variables included each student’s age in months, whether the student was a girl, and an index of the family SES. The SES index was constructed using information about the highest level of schooling completed by the parents, household possessions, sources of lighting, as well as the material of the household floor, wall, and roof. The value of the index ranges from 1 to 15, with greater values indicating higher levels of family SES. Individual Process Variables

I examined two indicators of individual processes, each representing aspects of learning that are alterable at the student level. The first was a dummy variable indicating whether or not the student had ever repeated a grade. To the extent that a student sometimes is told to repeat a grade by the school because of undesirable academic performance, grade repetition is not within the realm of decision making by the student or the parents. However, I categorized repetition as an individual process because it is an individuallevel variable, and it is difficult to tell from the data whether grade repetition, when it occurred, was a decision made by the student, the parents, or the school. The second individual process variable was a composite of home interest in the student’s academic work. This composite was constructed on the basis of students’ responses to questions about the frequency that somebody other than the teacher did the following: (1) made sure that the student did the homework, (2) usually helped the student with homework, (3) asked the student to do work on school subjects, and (4) asked the student questions about what she or he did in school subjects. The composite was constructed in such a way that the greater value indicates a higher level of home interest in the student’s academic work and vice versa. School Input Variables

My analyses considered six variables, each representing aspects of schools that are not easily alterable from the viewpoint of teachers and school principals and, thus, can be labeled as “school inputs.” The first variable indicated whether the school was located in a rural or remote area (one if “yes” and zero if small towns and large cities). The second variable reflected the overall family background of the student intake, which was the school average of 588

November 2006

URBAN-RURAL GAPS IN AFRICA

the student SES index. These two variables are sometimes called “school context” indicators. The next three variables reflected the level of physical resources, with one variable indicating whether or not the school building was in good condition, another variable indicating the quantity of school facilities and equipment, and the third physical resource variable reflecting the number of instructional resources available to reading teachers. One important school input is the quality of teachers, and SACMEQ II administered a reading and numeracy test to six teachers from each sampled school. The common items on the test of students and teachers allowed the test results of teachers to be put on the same scale as that of the students. Thus, the final school input variable was the average reading scores of the teachers of each school. School Process Variables

Six variables representing school processes were selected based on the literature and on the availability of information in the background questionnaires. Previous research suggests that effective schools generally share the following characteristics: homework and feedback to student learning, parental involvement in the child’s school work, monitoring, and evaluation. Based on the information available from the background questionnaires, four variables were used to represent these aspects of school processes. The first was the student report of whether or not the teacher assigned reading homework and, if homework was indeed assigned, whether or not the teacher corrected it. The variable was aggregated at the school level. The second variable indicated whether or not the reading teacher met frequently with parents to discuss the student’s academic progress. The other two variables were also based on teacher reports. One was a composite of the extent to which the visiting inspectors provided positive feedback to the teacher, and the other was the number of visits by inspectors between 1998 and 2000. Two additional composite variables representing school processes were constructed on the basis of reports by the school head. The first such school process variable reflected the prevalence of student behavioral problems (arriving late to school, absenteeism, skipping classes, as well as alcohol abuse/ possession and fights among students). The other variable reflected the prevalence of teacher behavioral problems (teachers arriving late at school, teacher absenteeism, the intimidation or bullying of students, and teacher health problems). Methods of Data Analysis

My choice of analytic strategy was guided by each of my four sets of individual research questions. In order to address the first two sets of research questions, that is, whether and to what extent differences existed between rural and urban students and between the rural and urban schools they Comparative Education Review

589

ZHANG

attended, I conducted bivariate analyses. To explore the other research questions, I estimated a series of two-level hierarchical linear models of the following form: Yij p b 0j ⫹ b1j X 1ij ⫹ … ⫹ bq j X qij ⫹ rij b 0j p g00 ⫹ g01W1j ⫹ … ⫹ g0sWsj ⫹ u 0j b1j p g10 . . . bq j p gq0 , where Yij was the reading literacy score for student i in school j; b1j–bqj was a vector coefficients reflecting the estimated relationships between reading literacy and a set of individual-level variables, Xq (q p 1, . . . Q ); b0j was the mean literacy score for school j; r01–r0S was the estimated relationship between the school averages of reading literacy scores and a set of school-level variables, Ws (s p 1, . . . S); and g00 was the grand mean literacy score adjusted for the individual- and school-level characteristics. The term rij captured the difference between the score of student i and the mean score for school j, and uoj was the deviation of the mean score for school j from the grand mean. The residuals, rij and uoj, are assumed to be normally distributed, with a mean of zero, and variances of j2 and t00. Initially, a model without any predictors, that is, a “null” model, was fitted to the data for each country. In this case, g00 was the grand mean literacy score adjusted for design effects, and j2 and t00 measured the variances in reading literacy scores that existed between and within schools. I then added the dummy variable representing whether or not a school was located in a rural area. This allowed me to estimate the extent of disadvantage experienced by students because they attended rural schools, g01. Following this, I added the individual- and school-level input and process variables to the models in order to detect the changes in g01, the rural advantage in literacy scores, which is the focus of the last two sets of research questions. The multilevel data analysis was carried out using hierarchical linear modeling software.22 Rural-Urban Disparities in Reading Literacy Scores

Table 1 provides the mean and standard deviation of all the student-level variables used in the analysis, as well as the rural-urban differences. As can be seen, the average reading scores ranged from 428 points in Malawi to 582 22

Anthony Bryk and Stephen W. Raudenbush, Hierarchical Linear Models: Applications and Data Analysis Methods, 2nd ed. (Newbury Park, CA: Sage, 2002). 590

November 2006

TABLE 1 Means, Standard Deviations, and Rural-Urban Differences in Student Characteristics

Reading Scores Country Botswana Kenya Lesotho Malawi Mauritius Mozambique Namibia Seychelles South Africa Swaziland Mainland Tanzania Uganda Zambia Zanzibar Tanzania Total

Age in Months

% Girls

Index of Students’ SES

% Repeaters

Index of Home Interest in Students’ Academic Work

n

Mean

SD

Diff.

Mean

SD

Diff.

%

Diff.

Mean

SD

Diff.

%

Diff.

Mean

SD

Diff.

3,322 3,299 3,155 2,333 2,945 3,177 5,048 1,484 3,163 3,139 2,854 2,642 2,611 2,514

521.1 546.5 451.2 428.9 536.4 516.7 448.8 582.0 492.3 529.6 545.9 482.4 440.1 478.2 496.8

88.3 88.9 57.9 49.9 121.6 64.7 86.6 124.2 122.4 68.2 90.1 91.2 84.7 70.4 97.9

⫺36.7 ⫺48.1 ⫺28.2 ⫺16.4 ⫺9.8 ⫺18.5 ⫺85.4 ⫺7.0 ⫺118.7 ⫺40.1 ⫺72.8 ⫺36.4 ⫺57.0 ⫺23.9 ⫺50.5

157.8 168.4 169.6 174.0 135.8 176.7 166.4 138.8 156.9 166.4 180.4 171.4 166.7 179.1 165.5

13.7 18.9 22.1 26.0 5.8 22.9 22.4 4.8 19.1 19.8 18.0 22.0 20.4 19.0 22.9

4.5 6.2 2.4 12.4 .0 8.3 10.9 .4 6.7 6.9 11.8 5.3 15.3 7.1 9.9

51.0 50.3 55.6 47.8 48.1 40.3 51.9 50.1 52.5 51.6 52.2 44.5 48.4 51.7 49.9

⫺1.9 ⫺2.6 3.6 ⫺4.9 ⫺.4 ⫺12.8 1.4 ⫺3.0 ⫺.4 .0 .2 ⫺7.7 ⫺8.5 ⫺1.7 ⫺1.1

6.9 6.1 5.8 5.2 10.6 5.4 6.3 11.0 8.5 7.5 5.2 4.9 6.3 5.5 6.7

3.2 2.6 2.3 2.6 1.8 2.8 3.3 1.8 3.4 2.9 2.5 2.4 2.9 2.9 3.3

⫺2.4 ⫺1.9 ⫺1.3 ⫺2.7 ⫺.4 ⫺1.7 ⫺3.8 ⫺.3 ⫺3.6 ⫺2.9 ⫺3.1 ⫺1.4 ⫺2.7 ⫺3.2 ⫺2.7

31.4 64.1 60.8 66.1 18.7 78.2 54.1 10.3 42.3 59.3 23.3 52.9 51.5 27.6 47.6

4.2 9.9 2.9 5.0 .6 -8.2 11.8 -2.4 19.6 12.1 5.5 2.2 15.5 6.4 8.3

10.8 10.9 10.5 9.1 11.1 10.9 11.0 11.2 11.1 10.2 11.5 10.5 10.4 9.7 10.7

2.2 1.8 2.1 2.2 1.9 1.9 2.2 1.9 2.1 2.0 2.7 2.2 2.1 1.9 2.2

⫺.8 ⫺.7 ⫺.3 ⫺.6 ⫺.1 ⫺.1 ⫺.6 ⫺.6 ⫺.6 ⫺.5 ⫺1.3 ⫺.7 ⫺.6 ⫺.6 ⫺.6

Note.—Diff. p difference.

ZHANG

in Seychelles, a difference of more than a standard deviation on the international scale. Countries differed not only in their average reading scores but also in the variability of test scores as indicated by the standard deviation of mean scores. Countries with higher levels of average reading scores had greater variability among their students. Exceptions include South Africa, whose mean score was below the average but who had the second highest level of variability, and Mozambique and Swaziland, both of which had relatively high scores, on average, but lower levels of variability. Rural-urban gaps in literacy varied tremendously across the 14 school systems. The average rural sixth graders in all countries scored 50 points lower on the reading test than their urban counterparts. But the gap was 118 points in South Africa, more than a standard deviation on the international scale and much larger than the differences between South Africa’s overall reading score and the scores of other countries in this study. The other school systems with sizable rural disadvantages included Kenya (48), Namibia (85), the mainland of Tanzania (72), and Zambia (57). At the other extreme were Mauritius and Seychelles, where, on average, rural students scored only 10 and 7 points, respectively, below their urban counterparts (differences not statistically different from zero). While there was no obvious relationship between the size of the ruralurban gap and the national average reading scores, the gaps were wider in systems with greater overall variability in their reading scores—in all school systems except in Mauritius and Seychelles. For example, South Africa had the largest gap and the second greatest variability. Gaps in such school systems as Kenya, Swaziland, and Botswana were moderate, as was the variability in the reading scores. Similarly, Lesotho, Malawi, and Mozambique had relatively smaller gaps and smaller variability. It is beyond the scope of this article to identify the factors underlying the total variability in reading scores. However, the results suggest that rural-urban disparities might be an important source of variability in most of the 14 school systems. Differences in Student and School Characteristics

In addition to literacy scores, rural students also differed from their urban counterparts in their average age. For instance, sixth graders in Mauritius and Seychelles, on average, were about 11 years of age, much younger than the average of 15 years in Tanzania. The older age in Tanzania implied later entry into primary school, repetition, or both, for a large number of students. Other countries where the sixth-grade student population was older than 14 years included Kenya, Lesotho, Malawi, Mozambique, and Uganda. The wide standard deviation in students’ ages in all of the countries except Mauritius and Seychelles suggested a greater heterogeneity in students in these school systems. Older students may outperform their younger peers simply because of their relative maturity. However, if a child is older than his classmates 592

November 2006

URBAN-RURAL GAPS IN AFRICA

because she or he repeated grades, then older ages could indicate learning difficulties. Hence, the direction of the relationship between child’s age and reading scores is ambiguous. In all the school systems except in Mauritius, students attending rural schools on average were older than their urban counterparts. The gap ranged from less than 1 month in Seychelles to about 1 year or more in Malawi, in the mainland of Tanzania, and in Zambia. The fourth column of table 1 highlights the challenge that school systems in southern and western Africa faced in achieving gender parity in school participation. Especially in Mozambique and Uganda, girls were disadvantaged compared in school access with boys. By contrast, in Lesotho, boys were at a relative disadvantage. It is noteworthy that in Mozambique, Uganda, and Zambia, girls from rural areas were at a particular disadvantage in school access. The variable representing student sex was included because it has been shown to be an important predictor of a student’s reading ability.23 Children from rural areas of less developed countries often suffer a socioeconomic disadvantage. This certainly seems to be the case for countries participating in the SACMEQ study. The fifth column of table 1 shows that, on average, students attending rural schools came from families with lower levels of SES in every school system. Table 1 also indicates that families of sixth graders in Mauritius and Seychelles were, on average, much better off than those in other countries, especially Malawi, Mozambique, Uganda, and both the Tanzanian mainland and Zanzibar. While students’ families in Mauritius and Seychelles were better off, the gaps between the well-off and not so well-off were much smaller in these countries than was the case in Botswana, Namibia, and South Africa, as can be seen by the standard deviations in each country. In terms of schooling, sixth graders in rural areas were more likely than urban students to have experienced grade repetition. Rural students also received less academic support at home from a family member discussing school work with them. The last two columns on the right side of table 1 indicate that in eight out of the 14 school systems more than half of the sixth graders had repeated a grade at least once. Such students made up two-thirds or more of the sixth-grade population in Malawi and Mozambique. Except in Mozambique and Seychelles, grade repetition was more prevalent among rural students in all the other countries. Table 2 provides the summary statistics for the school-level variables used for this analysis. The SACMEQ study was designed to have a representative sample of sixth-grade students in each of the participating school systems, not to have a representative sample of schools. As such, these statistics should 23 Ina V. S. Mullis, Michael O. Martin, Eugene J. Gonzalez, and Anne M. Kennedy, PIRLS 2001 International Report: IEA’s Study of Reading Literacy Achievement in Primary Schools in 35 Countries (Chestnut Hill, MA: International Study Center, Boston College, 2003); UNESCO/UIS-OECD, Literacy Skills for the World of Tomorrow (Montreal: UNESCO Institute for Statistics, 2003).

Comparative Education Review

593

TABLE 2 Means and Rural-Urban Differences in School Characteristics 594

Country Botswana Kenya Lesotho Malawi Mauritius Mozambique Namibia Seychelles South Africa Swaziland Mainland Tanzania Uganda Zambia Zanzibar Tanzania Total

n 170 185 177 140 153 176 270 24 169 168 181 163 173 145

School Mean of Students’ SES

School Building in Relatively Good Condition

Index of School Material Resources

Index of Reading Instructional Resources

Average Reading Scores of Teachers

Mean

Diff.

Mean

Diff.

Mean

Diff.

Mean

Diff.

Mean

Diff.

7.0 6.3 5.6 5.1 10.5 5.1 6.8 11.0 8.4 7.4 4.9 5.0 6.0 5.0 6.4

⫺2.7 ⫺1.9 ⫺1.4 ⫺2.7 ⫺.5 ⫺1.7 ⫺3.8 ⫺.4 ⫺3.2 ⫺2.7 ⫺2.6 ⫺1.5 ⫺2.8 ⫺3.3 ⫺2.6

.61 .67 .33 .41 .83 .55 .54 .63 .54 .51 .47 .26 .49 .55 .52

⫺.02 ⫺.17 ⫺.14 ⫺.03 ⫺.03 ⫺.03 ⫺.20 ⫺.03 ⫺.38 ⫺.14 ⫺.10 ⫺.31 ⫺.32 ⫺.04 ⫺.18

9.8 7.9 6.3 4.2 14.5 6.7 10.9 16.6 11.6 8.4 5.3 7.4 6.2 5.8 8.3

⫺2.4 ⫺2.7 ⫺1.9 ⫺1.5 ⫺.6 ⫺2.3 ⫺7.9 .5 ⫺9.1 ⫺3.6 ⫺1.2 ⫺3.4 ⫺3.7 ⫺2.2 ⫺4.1

11.0 8.5 10.3 7.3 9.4 6.3 8.9 12.2 9.5 10.0 7.0 7.8 8.2 6.1 8.6

⫺.6 ⫺1.0 ⫺.1 ⫺.3 .4 ⫺.4 ⫺1.7 .0 ⫺1.5 ⫺1.1 .5 ⫺1.3 ⫺.1 ⫺.3 ⫺.7

758.1 791.5 724.0 712.0 . . . 716.3 730.5 820.0 . . . 747.8 705.5 701.2 761.6 649.4 730.2

⫺7.9 .2 ⫺13.2 ⫺10.9 . . . ⫺4.9 ⫺59.4 ⫺4.4 . . . ⫺.4 ⫺9.1 ⫺26.2 2.8 ⫺12.0 ⫺22.9

595

Teacher Assigns and Corrects Reading Homework

Teacher Frequently Meets with Parents to Discuss Students’ Progress

Index of Positive Feedback by Visiting Inspectors

Mean

Diff.

Mean

Diff.

Mean

Diff.

Mean

Diff.

Mean

.80 .94 .90 .70 .91 .88 .85 .83 .79 .83 .75 .79 .62 .72 .81

⫺.01 ⫺.02 .02 ⫺.02 .08 ⫺.01 ⫺.02 ⫺.06 .03 .00 ⫺.06 .04 ⫺.05 ⫺.03 ⫺.02

.81 .89 .79 .89 .74 .87 .68 .97 .80 .65 .73 .79 .83 .69 .78

⫺.16 ⫺.05 ⫺.02 .07 .11 ⫺.01 ⫺.11 .04 ⫺.19 ⫺.02 .18 .00 ⫺.11 ⫺.18 ⫺.06

6.46 6.77 5.40 5.33 5.41 4.73 3.13 . . . 2.69 2.40 7.33 6.49 5.38 5.74 5.09

.60 .99 .94 .32 1.13 .63 1.85 . . . .19 ⫺.72 .34 ⫺.12 ⫺.93 ⫺1.05 .64

2.27 6.94 4.31 4.49 . . . 3.47 .06 . . . 1.84 . . . 3.62 5.88 2.37 5.03 3.44

.09 .29 1.25 ⫺.03 . . . ⫺.09 ⫺.02 . . . .53 . . . ⫺.98 .54 ⫺.75 ⫺.31 .46

2.72 1.49 2.36 2.36 1.25 1.36 2.21 2.00 2.62 2.95 4.83 4.63 2.67 1.85 2.56

Botswana Kenya Lesotho Malawi Mauritius Mozambique Namibia Seychelles South Africa Swaziland Mainland Tanzania Uganda Zambia Zanzibar Tanzania Total Note.—Diff. p difference.

No. of Visits by Inspectors, 1998–2000

Index of Problems of Student Behavior Diff. .00 ⫺.53 .15 1.07 ⫺.10 .10 .08 ⫺1.52 .71 ⫺.75 ⫺.38 1.17 .72 .04 .47

Index of Problems of Teacher Behavior Mean 3.55 3.64 3.75 4.13 2.69 4.20 3.79 3.79 3.84 3.86 4.15 4.88 4.49 3.81 3.90

Diff. .06 ⫺.08 ⫺.19 ⫺.13 .00 ⫺.36 .21 ⫺1.25 .50 ⫺.52 ⫺.05 .81 ⫺.46 ⫺.32 .03

ZHANG

not and cannot be used to draw firm conclusions about the schools of the sixth graders. Nevertheless, these summary statistics provide a rough indication of the school characteristics as well as rural or urban advantages. Table 2 suggests that even though school systems varied in terms of the overall quality of physical and human resources, rural schools everywhere were at a disadvantage. For instance, while fewer than half of the sampled schools in Lesotho, Malawi, mainland Tanzania, Uganda, and Zambia reported that their school buildings were in relatively good condition, in all countries rural schools were less likely to report so. In addition, rural schools overall had fewer facilities or equipment. Teachers in rural schools generally had access to fewer instructional resources. In most of these countries, rural sixth-grade teachers seemed to lag behind their urban counterparts with respect to reading abilities. While the data suggest that rural schools were disadvantaged in the quality of their physical and human resources, there was no clear pattern of rural-urban differences in terms of some of the major school processes. In summary, a preliminary look at the data shows that rural students not only lagged behind their urban counterparts in reading ability but also compared unfavorably in the school conditions that are important to academic success in general. The SES levels of the families of rural students were lower, and the rural students tended to have less home support for their academic work. In addition, rural students tended to be older than their urban counterparts, a result of late entry into the school system, a higher incidence of grade repetition, or a combination of both. Even though many schools in the SACMEQ countries might benefit from a boost in physical and human resources, this was especially true in rural areas, where more school buildings needed major repairs, where teachers had fewer instructional resources, where there were fewer facilities, and where teachers had lower reading scores. The Role of Demographic Factors and SES in Creating Rural-Urban Disparities

The third and fourth sets of research questions focus on the extent to which the differences in reading literacy scores between rural and urban students were attributable to differences in individual and school characteristics. Table 3 presents the results obtained by fitting a series of multilevel models to address these questions. To save space, the table shows only the coefficients of the dichotomous variable indicating that a school was located in a rural community for each model that was fitted (described in the previous section). These results are also displayed in figure 1. The first column of table 3 shows the rural disadvantage in reading literacy scores for each country by fitting the “null” model to the data. These estimates are slightly different from the rural-urban differences shown in table 1 be596

November 2006

TABLE 3 What Accounts for Rural Disadvantages in Literacy Scores?

Country Botswana Kenya Lesotho Malawi Mauritius Mozambique Namibia Seychelles South Africa Swaziland Mainland Tanzania Uganda Zambia Zanzibar Tanzania Note.—Diff. p difference.

Adjusted for Individual SES, Age, and Sex

Further Adjusted for School Context and Resources

Further Adjusted for School Processes

Further Adjusted for Individual Processes

Raw Diff.

SE

Diff.

SE

Diff.

SE

Diff.

SE

Diff.

SE

⫺37.72 ⫺49.15 ⫺26.97 ⫺17.93 ⫺8.82 ⫺17.20 ⫺87.53 ⫺4.05 ⫺106.11 ⫺38.03 ⫺66.14 ⫺35.96 ⫺54.44 ⫺26.52

(6.78) (9.00) (6.34) (5.59) (1.78) (6.69) (8.11) (14.64) (11.9) (7.78) (6.75) (13.68) (6.86) (6.13)

⫺14.53 ⫺27.18 ⫺23.43 ⫺5.59 .96 ⫺11.67 ⫺63.19 1.54 ⫺83.58 ⫺22.78 ⫺40.73 ⫺26.72 ⫺26.50 ⫺10.01

(4.91) (7.80) (5.92) (5.25) (8.70) (6.68) (6.18) (12.22) (10.24) (6.43) (6.73) (12.89) (5.44) (6.66)

⫺.90 ⫺3.68 ⫺9.57 ⫺2.48 6.97 ⫺5.28 ⫺3.31 . . . 18.14 6.95 ⫺24.21 18.01 ⫺12.65 ⫺.60

(5.53) (7.76) (6.50) (6.16) (9.32) (7.14) (6.51) . . . (12.07) (7.14) (6.33) (11.94) (5.97) (8.52)

⫺3.22 ⫺3.62 ⫺9.93 ⫺.86 8.52 ⫺5.42 ⫺1.45 . . . 17.46 4.34 ⫺21.99 16.86 ⫺12.15 ⫺2.38

(5.46) (7.59) (6.34) (6.21) (9.85) (7.46) (6.51) . . . (11.42) (6.89) (6.81) (11.13) (6.18) (9.19)

⫺4.16 ⫺2.34 ⫺9.93 ⫺1.65 9.16 ⫺5.37 ⫺1.75

(5.44) (7.65) (6.28) (6.21) (9.61) (7.45) (6.39)

16.99 5.40 ⫺20.54 17.50 ⫺12.49 ⫺.06

(11.30) (6.94) (6.89) (11.01) (6.20) (9.04)

Fig. 1.—What accounts for urban-rural differences in literacy?

URBAN-RURAL GAPS IN AFRICA

cause the former were adjusted for the design effects.24 They nevertheless paint a similar picture for all the school systems. Students attending rural schools in Mauritius and Seychelles met their counterparts’ performance in urban schools. Sixth graders attending schools in urban communities had an advantage of about 17 points in Malawi and Mozambique compared with more than 100 points in South Africa. Overall, the rural-urban gap was striking in Kenya, Namibia, South Africa, mainland Tanzania, and Zambia, where rural students lagged behind their urban counterparts in reading literacy scores by about half a standard deviation or more in the distribution of scores among all countries. Rural students, as described above, suffered from inferior home and school circumstances. The second column of table 3 presents the estimated rural disadvantage in reading scores after adjusting for students’ SES, their ages, and their sex. As expected, the rural disadvantage was reduced in all school systems, though the size of the reduction varied. Given the same SES as urban students, rural sixth graders in Mauritius and Seychelles could even outperform their counterparts in urban schools. The rural gaps narrowed by more than half in Botswana, Malawi, Zanzibar, and Zambia. By contrast, the reduction in the rural-urban gap in reading literacy scores was less than one-third in Namibia and South Africa, two countries showing large gaps without adjusting for SES and demographic characteristics. The reduction was also modest in countries such as Lesotho, Namibia, and Uganda, where the rural disadvantage was relatively small. These results suggest that a large part of the relatively inferior performance in the reading literacy test of the rural students was attributable to their home circumstances and their own characteristics. The Role of School Contexts and Processes in Urban-Rural Inequality in Literacy

The third column of table 3 displays the estimated rural disadvantage in reading literacy scores after further taking into account the average SES levels of the schools, the condition of the school buildings, the number of school facilities and equipment, and the average reading scores of the teachers. These and the rest of the expanded multilevel models, including school-level variables, were not fitted to the data from Seychelles since the number of sampled schools, 24, was too few to obtain reliable parameter estimates for school-level variables. In addition, an earlier analysis of the data indicates that rural-urban differences, when they existed, were much less pronounced in Seychelles than in most of the other SACMEQ school systems and thus might be less of a policy concern in Seychelles. In all of the other school systems except for mainland Tanzania and Zambia, the rural advantage virtually disappeared after further considering 24

Bryk and Raudenbush, Hierarchical Linear Models.

Comparative Education Review

599

ZHANG

differences in school context and physical and human resources. Even in mainland Tanzania and Zambia, the rural disadvantage was reduced to 24 and 13 points, respectively, about one-third or less than the gap before making this adjustment. The changes were most dramatic for Swaziland, South Africa, and Uganda, where rural students would be expected to score higher than the urban students. In South Africa, it is worth recalling, the gap in average reading scores of rural and urban students was reduced by only about onefifth after controlling for individual SES, age, and sex, which was quite modest relative to other countries. The virtual reversal of the direction of the ruralurban gap after further controlling for school context and resource variables shows South Africa’s sharp urban-rural differences in SES and school resources. While the data do not allow us to establish any causal link between rural-urban differences in the SES intake of students and school resources, on the one hand, and students’ learning achievement, on the other hand, South African interventions aiming to close the rural-urban achievement gap may benefit from enabling schools in rural areas to obtain equal and adequate levels of resources.25 The last two columns of table 3 present the estimated rural disadvantage in reading literacy scores after further controlling for the six variables representing school processes and two variables representing individual learning processes. As can be seen, the coefficients representing rural-urban differences in reading scores basically remained the same. This might be a result of a lack of differences between rural and urban schools in the values of the process variables, as shown in table 1. In the case of the individual learning processes, which did exhibit differences in average values for rural and urban schools, this might be a result of the correlation between the two process variables and the other variables already included in the model. The advantages of urban over rural students in reading scores are presented graphically in figure 1. The figure presents urban advantages both in terms of unadjusted, raw differences with rural areas and also controlling for the factors included in the second and third models that were seen in the second and third columns of table 3. In summary, the analysis reveals that in 12 out of the 14 school systems in southern and western Africa, sixth graders in rural areas underperformed their urban counterparts in reading literacy. A large part of this disadvantage, however, could be accounted for by differences in such individual characteristics as family SES, age, and sex. It is noteworthy that in all but two of the school systems, the disadvantage of rural students virtually disappeared after considering differences in these individual students’ characteristics, school contexts, and resources between rural and urban areas. 25

For details on the evolution of funding mechanisms in South African schools, see Andrew Reschovsky, “Financing Schools in the New South Africa,” Comparative Education Review 50, no. 1 (2006): 21–45. 600

November 2006

URBAN-RURAL GAPS IN AFRICA

Summary and Conclusion

Improving schooling in rural areas will pose a serious challenge in the less developed countries. But doing so also presents a valuable opportunity for national and international initiatives to widen educational access and provide better learning environments for all children. The challenge comes because of the distance that needs to be traveled by many of the world’s less developed countries in order to meet the educational goals of EFA and MDG. The reality of rural schooling in most of these countries is even further away from the targets. Since rural areas contain substantial majorities of the populations of many less developed countries, attending to the learning needs of rural children improves the chances of achieving the overall goals of EFA and MDG. The analysis of the SACMEQ data, as reported here, can contribute to the attainment of these goals by illuminating disparities in rural-urban reading achievement. While drastic improvements are needed in some countries in order to raise the overall level of learning outcomes for students, the evidence presented in this article suggests that rural-urban gaps in literacy are even larger than the differences between most countries. Thus, improving the learning of rural students must be central in any plans to raise the overall levels of learning of less developed countries. The analysis reported here suggests places where policies and interventions can begin in order to close the rural-urban gap. First, the fact that rural students underperformed their urban counterparts by large margins in most countries should not be surprising, since the former generally experienced inferior learning conditions. Compared with their urban counterparts, rural students had lower levels of family SES, were older in age, were more likely to have repeated a school grade, and had less home support for their academic work. The importance of these differences in students’ individual characteristics was evident from the fact that that they accounted for sizable proportions of the rural-urban gaps in reading literacy scores across the 14 school systems. Thus, policies and interventions aimed at closing the ruralurban learning gaps will need to take such differences into account. Second, even though no clear pattern of rural-urban differences was found in terms of school processes, the fact is that rural schools had fewer and lower-quality resources than did urban schools in almost all cases. These school processes included how reading teachers assigned and corrected student homework, how frequently reading teachers met with students’ parents to discuss the children’s academic progress, as well as how frequently inspectors visited the teachers and to what extent the visiting inspectors provided positive and useful feedback to the teachers. The resources that were examined included the conditions of school buildings, the number of school facilities and equipment, the number of instructional resources available to reading teachers, and the teachers’ reading proficiency. The rural-urban gaps in students’ reading litComparative Education Review

601

ZHANG

eracy scores effectively disappeared in most countries after taking into account the context and the resources of rural and urban schools. Consistent with the conclusions of studies using different methods in different countries, my research finds that school resources are important determinants of the rural-urban gaps in students’ literacy. Because students from higher SES backgrounds are likely to attend better-resourced schools, the estimated effects of school resources on the reduction in rural-urban gaps could be conservative and understate the true effect of resource inequality. From analysis of SACMEQ data, one cannot say that rural schools will definitely perform better if they are provided with more quality physical and human resources. However, doing so is a necessary condition for reducing and eliminating disparities in learning outcomes and raising learning levels for all students. The results of this study illustrate that rural-urban differences in learning outcomes are partly due to differences in the ways that students approach learning. My study suggests that improving school processes and strengthening home support for children’s academic work are both indispensable for eliminating between- and within-school inequities in students’ learning outcomes. Raising the levels of learning outcomes for all students requires an integrated rather than a piecemeal approach, one taking into account all sources of inequality.

602

November 2006